Research Article Windowed Least Square Algorithm Based PMSM … · 2019. 7. 31. · PMSM Parameters Estimation of Windowed Least Square Algorithm. When , ,and are xedvalue,thePMSM
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2013 Article ID 131268 11 pageshttpdxdoiorg1011552013131268
Research ArticleWindowed Least Square Algorithm Based PMSMParameters Estimation
Song Wang
School of Mechanical Electrical amp Information Engineering Shandong University Weihai China
Correspondence should be addressed to Song Wang wangsong sdu163com
Received 4 June 2013 Revised 26 July 2013 Accepted 26 July 2013
Academic Editor Juan J Nieto
Copyright copy 2013 Song Wang This is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
Stator resistance and inductances in 119889-axis and 119902-axis of permanent magnet synchronous motors (PMSMs) are importantparameters Acquiring these accurate parameters is usually the fundamental part in driving and controlling system design toguarantee the performance of driver and controller In this paper we adopt a novel windowed least algorithm (WLS) to estimatethe parameters with fixed value or the parameter with time varying characteristic The simulation results indicate that the WLSalgorithm has a better performance in fixed parameters estimation and parameters with time varying characteristic identificationthan the recursive least square (RLS) and extended Kalman filter (EKF) It is suitable for engineering realization in embeddedsystem due to its rapidity less system resource possession less computation and flexibility to adjust the window size according tothe practical applications
1 Introduction
The high-field-strength neodymium-iron-boron (NdFeB)magnets have become commercially available with afford-able prices so the permanent magnet synchronous motor(PMSM) is receiving increasing attention due to its highspeed high power density and high efficiency It is very suit-able for some high-performance requirement applicationsfor example robotics aerospace electric ship propulsionsystems and wind power generation systems [1ndash3] It hasbeen shown that PMSM can provide significant performanceimprovement in many variable speed applications [4] Thecommonly used control method in motor control is vectorcontrol The method has a requirement of obtaining relatedparameters of the motor Therefore acquiring accurateparameters of the motor is usually the fundamental partin driving system design We cannot measure the motorparameters with normal no-load test and locked rotor test inthe work site Moreover with the increasing working time ofthe motor and the surrounding environment changes someparameters of the motor will be changed Therefore servodrivers usually have the function of parameters identificationand self-tuning [3]
Stator resistance and inductances in 119889-axis and 119902-axis areimportant parameters of motor model which are consideredas constants usually However these parameters vary withdifferent operation conditions when motor is running [5]The study object of this paper was permanent magnet ser-vomotor produced by Huada Company in Wuhan of ChinaThe experiment data showed that stator resistance valueranged from 119877
119904to 13119877
119904and that inductance value ranged
from 119871119889to 1004119871
119889when the temperature ranged from 20
degrees to 80 degrees Thus the temperature of motor had agreat influence on stator resistance and inductance in 119889-axisand 119902-axis When these parameters are treated as constantsthe stability and control performance of the system will beaffectedTherefore the realization of parameters (119877
119904 119871119889 and
119871119902) identification is essential for motion control of PMSMModel identification and parameter estimation tech-
niques have becomemature after years of development Fromthe least square estimation theory [6ndash14] and its variousimproved algorithm [15ndash23] to Kalman filter algorithm [24ndash27] neural network [28 29] genetic algorithm [30ndash32]and so forth they can serve as parameter estimation toolsHowever these methods have their own characteristics andapplicability
2 Mathematical Problems in Engineering
The least square estimation is one of the most simple andmost mature parameter estimation methods However theamount of calculation of the traditional least square methodwill increase with time sequence increase It is hard to realizein embedded chip due to the large amount of calculationand there is a problem of data saturation Kalman filteringalgorithm is put forward for system identification by Kalmanin 1960 and there is a wide range of use However it issensitive to the initial conditions [26] and its performanceis poor for time varying parameter identification [33 34]With the development of artificial intelligence technologyneural network [28 29] and genetic algorithm [30ndash32] areused to the parameter estimation These intelligent methodscan get the identification results with high accuracyHoweverit is hard to apply them in practical parameter estimationdue to the large amount of calculation and complexity ofthe algorithm Therefore least square algorithm is also acommonly parameter estimationmethod Kinds of improvedalgorithms are proposed to promote the identification per-formance of traditional least square algorithm [18ndash22 35ndash41] For example the forgetting factor is introduced in therecursive least square estimation and the past time of datawill be forgotten by index rate [42 43] However it stillcannot discard the past time of data [16] but just weakensthe impact of the past time of data for the current parameterestimation Another method is to use window method forthe time series data [16 23 44] This method can discard thepast time of data flexibly and eliminate the impact of the pasttime of data for future parameter estimationThewindow sizecan be set flexible according to practical application In thispaper we adopt windowed least square algorithm for statorresistance 119877
119904 119871119889 and 119871
119902inductance estimation and make a
comparison with recursive least square and extended kalmanFilter (EKF) From the simulation result we can see thatwindowed least square algorithm has a better performancein convergence speed and identification precision for fixedparameters and parameters with time varying characteristicsFrom the view of algorithm complexity the windowed leastsquare algorithm is suitable for engineering realization inembedded chip such as DSP and ARM
This paper is consisted of the five sections Section 2describes the principle of least square theory and the recur-sive least square algorithm Section 3 illustrates thewindowedleast square algorithm Section 4 does some simulations forPMSM parameter estimation Section 5 analyses the simula-tion results and shows some conclusions
2 Least Square Estimation and Recursive LeastSquare Estimation
21 The Principle of Least Square Estimation The earlieststimulus for the development of the least square estimationtheory was apparently provided by astronomical studiesin which planet and comet motions were studied usingtelescopic measurement data The principle of the parameterestimation is simple and does not need any statistical charac-teristics of the variables It is used in system identification andparameter estimation widelyThe least square estimation still
can provide an accurate solution when other identificationmethods lose efficacy
Supposing 119910(119894) and 1199091(119894) 1199092(119894) sdot sdot sdot 119909
119899(119894) are the observa-
tion sequences of 119910 and 119909 at 1199051 1199052sdot sdot sdot 119905119898 The relationship of 119910
and 119909 is expressed
[[[[
[
119910 (1)
119910 (2)
119910 (119898)
]]]]
]
=
[[[[
[
1199091(1) sdot sdot sdot 119909
119899(1)
1199091(2) sdot sdot sdot 119909
119899(2)
1199091(119898) sdot sdot sdot 119909
119899(119898)
]]]]
]
[[[[
[
1205791
1205792
120579119899
]]]]
]
(1)
where 120579 = (1205791 1205792 120579119899) is the measured parameter set
and 119899 is the number of parameters We hope to estimatetheir values by the observation value of 119910 and 119909 at differenttime sequences 119898 is the time sequences to estimate the 119899
parameters 120579119894119898 ge 119899 is required and if119898 = 119899 we can get the
single solution from (1) as (2)
120579 = 119883minus1
119910 (2)
where 120579 is the estimation value of 120579 and inverse matrix 119883minus1
of119883 is required
120576 = 119910 minus 119883120579 (3)
where 120576 = (1205761 1205762sdot sdot sdot 120576119898)119879 is the error vector
22 Recursive Least Square Estimation In practical parame-ter estimation the data is always constantly to be refreshedTherefore we can further deduce (2) to a recursion algorithmThis algorithm does not need to compute the inverse matrixcalculation repeatedly and reduces the time-consuming andsystem resources occupation
The least square estimation using 119898 groups of data isshown as follows
120579 (119898) = (119883119879
119898119883119898)minus1
119883119879
119898119884119898 (7)
The119898+1moment data is (119909(119898+ 1) 119910(119898+ 1)) and then
119884119898+1
= 119883119898+1
120579 (119898 + 1) (8)
Mathematical Problems in Engineering 3
where
119884119898+1
=
[[[[[[
[
119910 (1)
119910 (119898)
119910 (119898 + 1)
]]]]]]
]
= [
[
119884119898
119910 (119898 + 1)
]
]
119883119898+1
=
[[[[[[
[
1199091(1) 119909
119899(1)
1199091(119898) 119909
119899(119898)
1199091(119898 + 1) 119909
119899(119898)
]]]]]]
]
= [
[
119883119898
119909119879
(119898 + 1)
]
]
(9)
The new least estimation equation is shown as follows
120579 (119898 + 1) = (119883119879
119898+1119883119898+1
)minus1
119883119879
119898+1119884119898+1
(10)
In order to get out of inverse matrix calculation of119883119879119898+1
Equation (11) is the recursive least square (RLS) estima-tion Recursive least square algorithm is called the general-ization Kalman filter algorithm [45 46] It is the engineeringrealization method of the least square estimation theory [17]
From the calculation of least square estimation andrecursive least square estimation we can see that the past timeof data has a big effect to future parameter estimation and alarge number of data calculations have occupied the systemresources seriously Therefore it is difficult to be realized inembedded systems
3 Windowed Least Square Estimation
Recursive least square algorithm can be used to real-timeparameter estimation However the algorithm uses the pasttime of data and the past time of data has the same impor-tance as the present data in the algorithm It weakened theimportance of current data caused a lot of system resourcespossession and affected the estimation speed and precision[44] In order to guarantee the instantaneity of parameterestimation the paper adopts windowed least squares (WLS)to estimate the parameters of PMSMThealgorithm simulatesthe window processing function of communication signalThe time series data used for parameter estimation are addedwindow handle to reduce the calculation of estimation andsystem resources possession making the algorithm easy forengineering realization
Suppose that 119875(119898) = (119883119879
119898119883119898)minus1 the parameter estima-
tion by (119909(119896) 119910(119896)) 119896 = 1 2 119898 is as follows
120579 (119898) = 119875 (119898)119883119879
119898119884119898 (12)
where
119883119898
=
[[[[
[
119909119879
(1)
119909119879
(2)
119909119879
(119898)
]]]]
]
119884119898
=
[[[[
[
119910 (1)
119910 (2)
119910 (119898)
]]]]
]
(13)
Consider increasing a group of data (119909(119898+ 1) 119910(119898+ 1))and then the parameter estimation is shown as follows [16 2344]
The window size is adjustable according to actual needsbased on the data length of regulation This can guaranteethe speed of calculation and can reduce the system resourcespossession too At this time the parameter estimation isrelated to the current 119898 data sample the past time of datahas no effect on parameter estimation and this can ensurethe instantaneity and accuracy of the parameter estimation
4 Simulations
41 PMSM Model The voltage equations flux linkage equa-tions and electromagnetic torque equations of PMSM in 119889 119902frames are as follows [47 48]
where119906119902and119906119889are voltages in 119902-axis and119889-axis respectively
119894119902and 119894
119889are currents in 119902-axis and 119889-axis 119877
119904is phase
resistance of stator 119871119889and 119871
119902are inductances in 119889-axis and
119902-axis 120596119903is rotor velocity 120595
119891is flux linkage established by
magnets and 119901 is the differential operatorThe mathematical model of PMSM is discretized to
estimate parameters (119877119904 119871119889 and 119871
119902) The discrete model of
PMSM is as follows
119906119902(119896) = 119877
119904119894119902(119896) + 119901120595
119902(119896) + 120596
119903120595119889(119896)
119906119889(119896) = 119877
119904119894119889(119896) + 119901120595
119889(119896) + 120596
119903120595119902(119896)
(25)
where 120595119889and 120595
119902are as follows
120595119889(119896) = 119871
119889119894119889(119896) + 120595
119891
120595119902(119896) = 119871
119902119894119902(119896)
(26)
The PMSM simulation model is established by MAT-LABSIMULINK and the PMSM running data is obtainedby the model The simulation model is shown in Figure 1
In MATLABSIMULINK the 119877119904 119871119889 and 119871
119902parameters
of PMSM are fixed in simulation We cannot simulate thetime varying characteristic of 119877
119904 119871119889 and 119871
119902 Therefore we
design a motor simulation model according to the require-ment The 119877
119904 119871119889 and 119871
119902can be changed flexibly in the
simulation
42 PMSM Parameters Estimation of Windowed Least SquareAlgorithm When 119877
119904 119871119889 and 119871
119902are fixed value the PMSM
simulation data is obtained by MATLAB The windowedleast square algorithm is used to identify the parametersThe algorithm with different window sizes is used for 119877
119904
119871119889 and 119871
119902identification From the identification result
(Figures 2 3 and 4) we can see that bigger window sizehas a better identification result However the window sizedoes not obviously have an effect on the promotion ofparameter identification precision when 119877
119904 119871119889 and 119871
119902are
fixedDifferent window sizes have big effect on the results of
parameter estimation when the estimated parameters havetime varying characteristicThemotor parameters119877
119904119871119889 and
119871119902are measured at 234∘C 30∘C 40∘C 50∘C 60∘C 70∘C and
80∘C Using piecewise linear method to simulate the timevarying of the three parameters the motor running data isobtained by MATLAB
The windowed least square is used to identify the 119877119904 119871119889
and 119871119902 The identification result is shown in Figures 5 6 and
7 when the 119877119904 119871119889and 119871
119902are changed at the same time From
the Figures 5ndash7 we can see that shorter window size has lowereffect on the promotion of identification precision Howeverthe window size is too big to improve the identificationprecisionThe identification result is better when the windowsize is 300ndash400 In the motor model 120595
Figure 5 Estimation result of 119877119904of WLS with different window
sizes when the parameters have time varying characteristic
method It is proposed in 1960 by Kalman [49] The theoryis applied to practical engineering immediately when it isput forward The Apollo program and C-5 plane navigationsystem design are the most successful application examplesExtended Kalman filter (EKF) is an improved model of theKalman filter which is one of the most widely applied innonlinear system filter
Mathematical Problems in Engineering 7
Discrete system state equation of EKF is
X (119896) = A (119896 minus 1)X (119896 minus 1) + B (119896 minus 1)U (119896 minus 1)
+ C (119896 minus 1) + w (119896 minus 1)
Z (119896) = H (119896 minus 1)X (119896) + k (119896)
where X(119896) is the system state vector U(119896) is the systeminput vector Z(119896) is the system observation vector w(119896) isthe system random noise vector k(119896) is the system randomobservation noise vector w(119896) and k(119896) are noise sequenceswith zero mean and the covariance matrices are Q(119896) andR(119896)
119879119904is the sampling period the discrete linear state space
equation (29) of PMSM is established by discretization andlinearization of the model (30)
Figure 6 Estimation result of 119871119889of WLS with different window
sizes when the parameters have time varying characteristic
Using the simulation model and getting the motor output 119894119889
119894119902 119906119889 119906119902120596119890 the initial values of119875119876 and119877 in EKF algorithm
are
119875 = diag ([01 01 00004 0002 002])
119876 = diag ([30 15 0005 003 003])
119877 = diag ([01 002])
(32)
When119877119904 119871119889 and 119871
119902of motor are fixed the identification
result table of EKF recursive least square and windowed leastsquare algorithm is shown in Table 1
From Table 1 we can see that identification result ofEKF algorithm is as good as the windowed least squarealgorithm when 119877
119904 119871119889 and 119871
119902are fixed The comparison
diagrams of identification result are shown in Figure 8 whenthe parameters are fixed
EKF and recursive least square algorithms cannot achievereasonable result when the parameters have time varyingcharacteristic or have a drastic change However windowedleast square algorithm can achieve good identification resultwhen 119877
119904 119871119889 and 119871
119902have time varying characteristic at the
same time (Figures 5ndash7)
5 Analysis and Conclusion
Through the previous different PMSM parameters identifica-tion experiments we can see the following
(1) When the parameters of PMSM have no time vary-ing characteristic three methods can achieve better
Figure 7 Estimation result of 119871119902of WLS with different window
sizes when the parameters have time varying characteristic
identification result in precision and accuracy Inthe calculation and instantaneity of identificationrecursive least square algorithm has a fatal flaw ofdata saturation so the precision and accuracy of thealgorithm are hard to guarantee It is difficult torealize in embedded system the real-time parametersidentification due to the amount of calculations andsystem resources possession of EKF [28]Thewindowsize of windowed least square algorithm is flexibleso we can choose the collected data according tothe changes of the parameters It will reduce theinfluence of the past time of data to the currentparameter identification guarantee the accuracy andinstantaneity of identification and reduce the systemresources possession at the same time
(2) When the parameters of PMSM have strong timevarying characteristic the EKF and recursive leastsquare algorithms cannot guarantee the precisionand accuracy of identification However windowedleast square algorithm can get better identificationresultTherefore EKF and recursive least square algo-rithms are suitable for fixed parameters estimationor parameters with weak time varying characteristicidentification Windowed least square algorithm canget a good result both for fixed parameters and fortime varying parameters identification
Embedded technology is widely used in the motor driverand controller at present However the embedded chip(MCU DSP ARM etc) has certain restriction in computingspeed and storage space Therefore windowed least square
Mathematical Problems in Engineering 9
Table 1 Estimation result comparison of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
Temperature (∘C) 234 30 40 50 60 70 80
119877119904
Actual data (Ω) 09664 10008 10373 10770 11245 11592 11751EKF
Figure 8 Comparison diagram of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
10 Mathematical Problems in Engineering
Figure 9 The PMSM experiment system
Figure 10 The prototype DSP-based PMSM driver
algorithm is a better choice for PMSM parameters identi-fication of motor driver and controller This paper is thebeginning of work There are a lot of work to do such astransplant the algorithm to practical controller and controlsystem (in Figures 9 and 10) which is designed to control thePMSM in practical application
Acknowledgment
This paper is supported by the Shandong Province Scienceand Technology Development Plan of China (Grant no2011GGE27053)
References
[1] M A Rahman and P Zhou ldquoAnalysis of brushless permanentmagnet synchronous motorsrdquo IEEE Transactions on IndustrialElectronics vol 43 no 2 pp 256ndash267 1996
[2] MOoshimaA Chiba A Rahman andT Fukao ldquoAn improvedcontrol method of buried-type IPM bearingless motors consid-ering magnetic saturation and magnetic pull variationrdquo IEEETransactions on Energy Conversion vol 19 no 3 pp 569ndash5752004
[3] K Liu Z Q Zhu Q Zhang and J Zhang ldquoInfluence ofnonideal voltage measurement on parameter estimation inpermanent-magnet synchronous machinesrdquo IEEE Transactionson Industrial Electronics vol 59 no 6 pp 2438ndash2447 2012
[4] F Caricchi F Crescimbini and O Honorati ldquoLow-cost com-pact permanent magnet machine for adjustable-speed pumpapplicationrdquo IEEETransactions on IndustryApplications vol 34no 1 pp 109ndash116 1998
[5] P Milanfar and J H Lang ldquoMonitoring the thermal conditionof permanent-magnet synchronous motorsrdquo IEEE Transactions
onAerospace and Electronic Systems vol 32 no 4 pp 1421ndash14291996
[6] T Kailath ldquoAn innovations approach to least-squares estima-tionmdashpart I linear filtering in additive white noiserdquo IEEETransactions on Automatic Control vol 13 pp 646ndash655 1968
[7] D G Robertson and J H Lee ldquoA least squares formulation forstate estimationrdquo Journal of Process Control vol 5 no 4 pp291ndash299 1995
[8] J S Gibson G H Lee and C F Wu ldquoLeast-squares estimationof inputoutput models for distributed linear systems in thepresence of noiserdquo Automatica vol 36 no 10 pp 1427ndash14422000
[9] S Tunali and I Batmaz ldquoDealing with the least squares regres-sion assumptions in simulation metamodelingrdquo Computers ampIndustrial Engineering vol 38 no 2 pp 307ndash320 2000
[10] R M Fernandez-Alcala J Navarro-Moreno and J C Ruiz-Molina ldquoLinear least-square estimation algorithms involvingcorrelated signal and noiserdquo IEEE Transactions on Signal Pro-cessing vol 53 no 11 pp 4227ndash4235 2005
[11] V Kratschmer ldquoLeast-squares estimation in linear regressionmodels with vague conceptsrdquo Fuzzy Sets and Systems vol 157no 19 pp 2579ndash2592 2006
[12] M J Garcıa-Ligero A Hermoso-Carazo and J Linares-PerezldquoLeast-squares linear estimation of signals from observationswith Markovian delaysrdquo Journal of Computational and AppliedMathematics vol 236 no 2 pp 234ndash242 2011
[13] S Ma C Quan R Zhu C J Tay L Chen and Z GaoldquoApplication of least-square estimation in white-light scanninginterferometryrdquo Optics and Lasers in Engineering vol 49 no 7pp 1012ndash1018 2011
[14] Q Wang and L Zhang ldquoLeast squares online linear discrimi-nant analysisrdquo Expert Systems with Applications vol 39 no 1pp 1510ndash1517 2012
[15] C J Demeure and L L Scharf ldquoSliding windows and latticealgorithms for computing QR factors in the least squares theoryof linear predictionrdquo IEEE Transactions on Acoustics Speechand Signal Processing vol 38 no 4 pp 721ndash725 1990
[16] K Zhao L Fuyun H Lev-Ari and J G Proakis ldquoSlidingwindow order-recursive least-squares algorithmsrdquo IEEE Trans-actions on Signal Processing vol 42 no 8 pp 1961ndash1972 1994
[17] H Liu and Z He ldquoA sliding-exponential window RLS adaptivefiltering algorithm properties and applicationsrdquo Signal Process-ing vol 45 no 3 pp 357ndash368 1995
[18] K Yoo and H Park ldquoFast residual computation for slidingwindow recursive least squares methodsrdquo Signal Processing vol45 no 1 pp 85ndash95 1995
[19] Y Xia M S Kamel and H Leung ldquoA fast algorithm for ARparameter estimation using a novel noise-constrained least-squares methodrdquo Neural Networks vol 23 no 3 pp 396ndash4052010
[20] A Aknouche E M Al-Eid and A M Hmeid ldquoOffline andonline weighted least squares estimation of nonstationarypower119860119877119862119867 processesrdquo Statistics amp Probability Letters vol 81no 10 pp 1535ndash1540 2011
[21] L Xie H Yang and F Ding ldquoRecursive least squares parameterestimation for non-uniformly sampled systems based on thedata filteringrdquo Mathematical and Computer Modelling vol 54no 1-2 pp 315ndash324 2011
[22] J Oliver R Aravind and K M M Prabhu ldquoImproved leastsquares channel estimation for orthogonal frequency divisionmultiplexingrdquo IET Signal Processing vol 6 no 1 pp 45ndash532012
Mathematical Problems in Engineering 11
[23] T Sadiki M Triki and D T M Slock ldquoWindow optimizationissues in recursive least-squares adaptive filtering and trackingrdquoin Proceedings of the 38th IEEE Annual Asilomar Conferenceon Signals Systems and Computers pp 940ndash944 Pacific GroveCalif USA November 2004
[24] G Welch and G Bishop ldquoAn introduction to the Kalman filterrdquo1997
[25] P J Hargrave ldquoA tutorial introduction to Kalman filteringrdquo inProceedings of the IEE Colloquium on Kalman Filters Introduc-tion Applications and Future Developments pp 11ndash16 1989
[26] M Gautier and P Poignet ldquoExtended Kalman filtering andweighted least squares dynamic identification of robotrdquo ControlEngineering Practice vol 9 no 12 pp 1361ndash1372 2001
[27] H M Al-Hamadi and S A Soliman ldquoKalman filter foridentification of power system fuzzy harmonic componentsrdquoElectric Power Systems Research vol 62 no 3 pp 241ndash248 2002
[28] T Boileau N Leboeuf B Nahid-Mobarakeh and F Meibody-Tabar ldquoOnline identification of PMSM parameters parameteridentifiability and estimator comparative studyrdquo IEEE Transac-tions on Industry Applications vol 47 no 4 pp 1944ndash1957 2011
[29] A Bechouche H Sediki D O Abdeslam and S HaddadldquoIdentification of induction motor at standstill using artificialneural networkrdquo in Proceedings of the 36th Annual Conferenceon IEEE Industrial Electronics Society (IECON rsquo10) pp 2908ndash2913 Glendale Ariz USA 2010
[30] F Alonge F DrsquoIppolito and FM Raimondi ldquoLeast squares andgenetic algorithms for parameter identification of inductionmotorsrdquo Control Engineering Practice vol 9 no 6 pp 647ndash6572001
[31] S Mishra ldquoA hybrid least square-fuzzy bacterial foragingstrategy for harmonic estimationrdquo IEEE Transactions on Evo-lutionary Computation vol 9 no 1 pp 61ndash73 2005
[32] R Liao H Zheng S Grzybowski and L Yang ldquoParticleswarm optimization-least squares support vector regressionbased forecasting model on dissolved gases in oil-filled powertransformersrdquo Electric Power Systems Research vol 81 no 12pp 2074ndash2080 2011
[33] R A Zadeh A Ghosh and G Ledwich ldquoCombination ofKalman filter and least-error square techniques in powersystemrdquo IEEE Transactions on Power Delivery vol 25 no 4 pp2868ndash2880 2010
[34] S Bolognani R Oboe andM Zigliotto ldquoSensorless full-digitalPMSM drive with EKF estimation of speed and rotor positionrdquoIEEE Transactions on Industrial Electronics vol 46 no 1 pp184ndash191 1999
[35] M Haardt ldquoStructured least squares to improve the per-formance of ESPRIT-Type algorithmsrdquo IEEE Transactions onSignal Processing vol 45 no 3 pp 792ndash799 1997
[36] M Ghogho A Swami and A K Nandi ldquoNon-linear leastsquares estimation for harmonics in multiplicative and additivenoiserdquo Signal Processing vol 78 no 1 pp 43ndash60 1999
[37] J Angeby ldquoEstimating signal parameters using the nonlinearinstantaneous least squares approachrdquo IEEE Transactions onSignal Processing vol 48 no 10 pp 2721ndash2732 2000
[38] J F Weng and S H Leung ldquoNonlinear RLS algorithm foramplitude estimation in class a noiserdquo IEE ProceedingsmdashCommunications vol 147 no 2 pp 81ndash86 2000
[39] D Zachariah M Sundin M Jansson and S ChatterjeeldquoAlternating least-squares for low-rank matrix reconstructionrdquoIEEE Signal Processing Letters vol 19 no 4 pp 231ndash234 2012
[40] R Montoliu and F Pla ldquoGeneralized least squares-based para-metric motion estimationrdquo Computer Vision and Image Under-standing vol 113 no 7 pp 790ndash801 2009
[41] Z Yingjie and G Liling ldquoImproved moving least squares algo-rithm for directed projecting onto point cloudsrdquoMeasurementvol 44 no 10 pp 2008ndash2019 2011
[42] S Seongwook J-S Lim S J Baek and K-M Sung ldquoVariableforgetting factor linear least squares algorithm for frequencyselective fading channel estimationrdquo IEEE Transactions onVehicular Technology vol 51 no 3 pp 613ndash616 2002
[43] S MorimotoM Sanada and Y Takeda ldquoMechanical sensorlessdrives of IPMSM with online parameter identificationrdquo IEEETransactions on Industry Applications vol 42 no 5 pp 1241ndash1248 2006
[44] H Sakai and H Nakaoka ldquoFast sliding window QRD-RLSalgorithmrdquo Signal Processing vol 78 no 3 pp 309ndash319 1999
[45] S Reece and S Roberts ldquoAn introduction to Gaussian processesfor the Kalman filter expertrdquo in Proceedings of the 13th Confer-ence on Information Fusion (FUSION rsquo10) pp 1ndash9 2010
[46] A Giorgano F M Hsu and J Wiley ldquoBook review least-square estimationwith applications to digital signal processingrdquoIEE Proceedings FmdashCommunications Radar and Signal Process-ing vol 132 no 7 1985
[47] S Wang R Zhao W Chen G Li and C Liu ldquoParameter iden-tification of PMSMbased on windowed least square algorithmrdquoin Proceedings of the Manufacturing Science and Technology(ICMST rsquo11) pp 5940ndash5944 Singapore 2012
[48] S Wang S Shi C Chen G Yang and Z Qu ldquoIdentificationof PMSM based on EKF and elman neural networkrdquo inProceedings of the IEEE International Conference on Automationand Logistics (ICAL rsquo09) pp 1459ndash1463 Shenyang China 2009
[49] R E Kalman ldquoA new approach to linear filtering and predictionproblemsrdquo Transactions of the ASMEmdashJournal of Basic Engi-neering D vol 82 pp 35ndash45 1960
The least square estimation is one of the most simple andmost mature parameter estimation methods However theamount of calculation of the traditional least square methodwill increase with time sequence increase It is hard to realizein embedded chip due to the large amount of calculationand there is a problem of data saturation Kalman filteringalgorithm is put forward for system identification by Kalmanin 1960 and there is a wide range of use However it issensitive to the initial conditions [26] and its performanceis poor for time varying parameter identification [33 34]With the development of artificial intelligence technologyneural network [28 29] and genetic algorithm [30ndash32] areused to the parameter estimation These intelligent methodscan get the identification results with high accuracyHoweverit is hard to apply them in practical parameter estimationdue to the large amount of calculation and complexity ofthe algorithm Therefore least square algorithm is also acommonly parameter estimationmethod Kinds of improvedalgorithms are proposed to promote the identification per-formance of traditional least square algorithm [18ndash22 35ndash41] For example the forgetting factor is introduced in therecursive least square estimation and the past time of datawill be forgotten by index rate [42 43] However it stillcannot discard the past time of data [16] but just weakensthe impact of the past time of data for the current parameterestimation Another method is to use window method forthe time series data [16 23 44] This method can discard thepast time of data flexibly and eliminate the impact of the pasttime of data for future parameter estimationThewindow sizecan be set flexible according to practical application In thispaper we adopt windowed least square algorithm for statorresistance 119877
119904 119871119889 and 119871
119902inductance estimation and make a
comparison with recursive least square and extended kalmanFilter (EKF) From the simulation result we can see thatwindowed least square algorithm has a better performancein convergence speed and identification precision for fixedparameters and parameters with time varying characteristicsFrom the view of algorithm complexity the windowed leastsquare algorithm is suitable for engineering realization inembedded chip such as DSP and ARM
This paper is consisted of the five sections Section 2describes the principle of least square theory and the recur-sive least square algorithm Section 3 illustrates thewindowedleast square algorithm Section 4 does some simulations forPMSM parameter estimation Section 5 analyses the simula-tion results and shows some conclusions
2 Least Square Estimation and Recursive LeastSquare Estimation
21 The Principle of Least Square Estimation The earlieststimulus for the development of the least square estimationtheory was apparently provided by astronomical studiesin which planet and comet motions were studied usingtelescopic measurement data The principle of the parameterestimation is simple and does not need any statistical charac-teristics of the variables It is used in system identification andparameter estimation widelyThe least square estimation still
can provide an accurate solution when other identificationmethods lose efficacy
Supposing 119910(119894) and 1199091(119894) 1199092(119894) sdot sdot sdot 119909
119899(119894) are the observa-
tion sequences of 119910 and 119909 at 1199051 1199052sdot sdot sdot 119905119898 The relationship of 119910
and 119909 is expressed
[[[[
[
119910 (1)
119910 (2)
119910 (119898)
]]]]
]
=
[[[[
[
1199091(1) sdot sdot sdot 119909
119899(1)
1199091(2) sdot sdot sdot 119909
119899(2)
1199091(119898) sdot sdot sdot 119909
119899(119898)
]]]]
]
[[[[
[
1205791
1205792
120579119899
]]]]
]
(1)
where 120579 = (1205791 1205792 120579119899) is the measured parameter set
and 119899 is the number of parameters We hope to estimatetheir values by the observation value of 119910 and 119909 at differenttime sequences 119898 is the time sequences to estimate the 119899
parameters 120579119894119898 ge 119899 is required and if119898 = 119899 we can get the
single solution from (1) as (2)
120579 = 119883minus1
119910 (2)
where 120579 is the estimation value of 120579 and inverse matrix 119883minus1
of119883 is required
120576 = 119910 minus 119883120579 (3)
where 120576 = (1205761 1205762sdot sdot sdot 120576119898)119879 is the error vector
22 Recursive Least Square Estimation In practical parame-ter estimation the data is always constantly to be refreshedTherefore we can further deduce (2) to a recursion algorithmThis algorithm does not need to compute the inverse matrixcalculation repeatedly and reduces the time-consuming andsystem resources occupation
The least square estimation using 119898 groups of data isshown as follows
120579 (119898) = (119883119879
119898119883119898)minus1
119883119879
119898119884119898 (7)
The119898+1moment data is (119909(119898+ 1) 119910(119898+ 1)) and then
119884119898+1
= 119883119898+1
120579 (119898 + 1) (8)
Mathematical Problems in Engineering 3
where
119884119898+1
=
[[[[[[
[
119910 (1)
119910 (119898)
119910 (119898 + 1)
]]]]]]
]
= [
[
119884119898
119910 (119898 + 1)
]
]
119883119898+1
=
[[[[[[
[
1199091(1) 119909
119899(1)
1199091(119898) 119909
119899(119898)
1199091(119898 + 1) 119909
119899(119898)
]]]]]]
]
= [
[
119883119898
119909119879
(119898 + 1)
]
]
(9)
The new least estimation equation is shown as follows
120579 (119898 + 1) = (119883119879
119898+1119883119898+1
)minus1
119883119879
119898+1119884119898+1
(10)
In order to get out of inverse matrix calculation of119883119879119898+1
Equation (11) is the recursive least square (RLS) estima-tion Recursive least square algorithm is called the general-ization Kalman filter algorithm [45 46] It is the engineeringrealization method of the least square estimation theory [17]
From the calculation of least square estimation andrecursive least square estimation we can see that the past timeof data has a big effect to future parameter estimation and alarge number of data calculations have occupied the systemresources seriously Therefore it is difficult to be realized inembedded systems
3 Windowed Least Square Estimation
Recursive least square algorithm can be used to real-timeparameter estimation However the algorithm uses the pasttime of data and the past time of data has the same impor-tance as the present data in the algorithm It weakened theimportance of current data caused a lot of system resourcespossession and affected the estimation speed and precision[44] In order to guarantee the instantaneity of parameterestimation the paper adopts windowed least squares (WLS)to estimate the parameters of PMSMThealgorithm simulatesthe window processing function of communication signalThe time series data used for parameter estimation are addedwindow handle to reduce the calculation of estimation andsystem resources possession making the algorithm easy forengineering realization
Suppose that 119875(119898) = (119883119879
119898119883119898)minus1 the parameter estima-
tion by (119909(119896) 119910(119896)) 119896 = 1 2 119898 is as follows
120579 (119898) = 119875 (119898)119883119879
119898119884119898 (12)
where
119883119898
=
[[[[
[
119909119879
(1)
119909119879
(2)
119909119879
(119898)
]]]]
]
119884119898
=
[[[[
[
119910 (1)
119910 (2)
119910 (119898)
]]]]
]
(13)
Consider increasing a group of data (119909(119898+ 1) 119910(119898+ 1))and then the parameter estimation is shown as follows [16 2344]
The window size is adjustable according to actual needsbased on the data length of regulation This can guaranteethe speed of calculation and can reduce the system resourcespossession too At this time the parameter estimation isrelated to the current 119898 data sample the past time of datahas no effect on parameter estimation and this can ensurethe instantaneity and accuracy of the parameter estimation
4 Simulations
41 PMSM Model The voltage equations flux linkage equa-tions and electromagnetic torque equations of PMSM in 119889 119902frames are as follows [47 48]
where119906119902and119906119889are voltages in 119902-axis and119889-axis respectively
119894119902and 119894
119889are currents in 119902-axis and 119889-axis 119877
119904is phase
resistance of stator 119871119889and 119871
119902are inductances in 119889-axis and
119902-axis 120596119903is rotor velocity 120595
119891is flux linkage established by
magnets and 119901 is the differential operatorThe mathematical model of PMSM is discretized to
estimate parameters (119877119904 119871119889 and 119871
119902) The discrete model of
PMSM is as follows
119906119902(119896) = 119877
119904119894119902(119896) + 119901120595
119902(119896) + 120596
119903120595119889(119896)
119906119889(119896) = 119877
119904119894119889(119896) + 119901120595
119889(119896) + 120596
119903120595119902(119896)
(25)
where 120595119889and 120595
119902are as follows
120595119889(119896) = 119871
119889119894119889(119896) + 120595
119891
120595119902(119896) = 119871
119902119894119902(119896)
(26)
The PMSM simulation model is established by MAT-LABSIMULINK and the PMSM running data is obtainedby the model The simulation model is shown in Figure 1
In MATLABSIMULINK the 119877119904 119871119889 and 119871
119902parameters
of PMSM are fixed in simulation We cannot simulate thetime varying characteristic of 119877
119904 119871119889 and 119871
119902 Therefore we
design a motor simulation model according to the require-ment The 119877
119904 119871119889 and 119871
119902can be changed flexibly in the
simulation
42 PMSM Parameters Estimation of Windowed Least SquareAlgorithm When 119877
119904 119871119889 and 119871
119902are fixed value the PMSM
simulation data is obtained by MATLAB The windowedleast square algorithm is used to identify the parametersThe algorithm with different window sizes is used for 119877
119904
119871119889 and 119871
119902identification From the identification result
(Figures 2 3 and 4) we can see that bigger window sizehas a better identification result However the window sizedoes not obviously have an effect on the promotion ofparameter identification precision when 119877
119904 119871119889 and 119871
119902are
fixedDifferent window sizes have big effect on the results of
parameter estimation when the estimated parameters havetime varying characteristicThemotor parameters119877
119904119871119889 and
119871119902are measured at 234∘C 30∘C 40∘C 50∘C 60∘C 70∘C and
80∘C Using piecewise linear method to simulate the timevarying of the three parameters the motor running data isobtained by MATLAB
The windowed least square is used to identify the 119877119904 119871119889
and 119871119902 The identification result is shown in Figures 5 6 and
7 when the 119877119904 119871119889and 119871
119902are changed at the same time From
the Figures 5ndash7 we can see that shorter window size has lowereffect on the promotion of identification precision Howeverthe window size is too big to improve the identificationprecisionThe identification result is better when the windowsize is 300ndash400 In the motor model 120595
Figure 5 Estimation result of 119877119904of WLS with different window
sizes when the parameters have time varying characteristic
method It is proposed in 1960 by Kalman [49] The theoryis applied to practical engineering immediately when it isput forward The Apollo program and C-5 plane navigationsystem design are the most successful application examplesExtended Kalman filter (EKF) is an improved model of theKalman filter which is one of the most widely applied innonlinear system filter
Mathematical Problems in Engineering 7
Discrete system state equation of EKF is
X (119896) = A (119896 minus 1)X (119896 minus 1) + B (119896 minus 1)U (119896 minus 1)
+ C (119896 minus 1) + w (119896 minus 1)
Z (119896) = H (119896 minus 1)X (119896) + k (119896)
where X(119896) is the system state vector U(119896) is the systeminput vector Z(119896) is the system observation vector w(119896) isthe system random noise vector k(119896) is the system randomobservation noise vector w(119896) and k(119896) are noise sequenceswith zero mean and the covariance matrices are Q(119896) andR(119896)
119879119904is the sampling period the discrete linear state space
equation (29) of PMSM is established by discretization andlinearization of the model (30)
Figure 6 Estimation result of 119871119889of WLS with different window
sizes when the parameters have time varying characteristic
Using the simulation model and getting the motor output 119894119889
119894119902 119906119889 119906119902120596119890 the initial values of119875119876 and119877 in EKF algorithm
are
119875 = diag ([01 01 00004 0002 002])
119876 = diag ([30 15 0005 003 003])
119877 = diag ([01 002])
(32)
When119877119904 119871119889 and 119871
119902of motor are fixed the identification
result table of EKF recursive least square and windowed leastsquare algorithm is shown in Table 1
From Table 1 we can see that identification result ofEKF algorithm is as good as the windowed least squarealgorithm when 119877
119904 119871119889 and 119871
119902are fixed The comparison
diagrams of identification result are shown in Figure 8 whenthe parameters are fixed
EKF and recursive least square algorithms cannot achievereasonable result when the parameters have time varyingcharacteristic or have a drastic change However windowedleast square algorithm can achieve good identification resultwhen 119877
119904 119871119889 and 119871
119902have time varying characteristic at the
same time (Figures 5ndash7)
5 Analysis and Conclusion
Through the previous different PMSM parameters identifica-tion experiments we can see the following
(1) When the parameters of PMSM have no time vary-ing characteristic three methods can achieve better
Figure 7 Estimation result of 119871119902of WLS with different window
sizes when the parameters have time varying characteristic
identification result in precision and accuracy Inthe calculation and instantaneity of identificationrecursive least square algorithm has a fatal flaw ofdata saturation so the precision and accuracy of thealgorithm are hard to guarantee It is difficult torealize in embedded system the real-time parametersidentification due to the amount of calculations andsystem resources possession of EKF [28]Thewindowsize of windowed least square algorithm is flexibleso we can choose the collected data according tothe changes of the parameters It will reduce theinfluence of the past time of data to the currentparameter identification guarantee the accuracy andinstantaneity of identification and reduce the systemresources possession at the same time
(2) When the parameters of PMSM have strong timevarying characteristic the EKF and recursive leastsquare algorithms cannot guarantee the precisionand accuracy of identification However windowedleast square algorithm can get better identificationresultTherefore EKF and recursive least square algo-rithms are suitable for fixed parameters estimationor parameters with weak time varying characteristicidentification Windowed least square algorithm canget a good result both for fixed parameters and fortime varying parameters identification
Embedded technology is widely used in the motor driverand controller at present However the embedded chip(MCU DSP ARM etc) has certain restriction in computingspeed and storage space Therefore windowed least square
Mathematical Problems in Engineering 9
Table 1 Estimation result comparison of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
Temperature (∘C) 234 30 40 50 60 70 80
119877119904
Actual data (Ω) 09664 10008 10373 10770 11245 11592 11751EKF
Figure 8 Comparison diagram of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
10 Mathematical Problems in Engineering
Figure 9 The PMSM experiment system
Figure 10 The prototype DSP-based PMSM driver
algorithm is a better choice for PMSM parameters identi-fication of motor driver and controller This paper is thebeginning of work There are a lot of work to do such astransplant the algorithm to practical controller and controlsystem (in Figures 9 and 10) which is designed to control thePMSM in practical application
Acknowledgment
This paper is supported by the Shandong Province Scienceand Technology Development Plan of China (Grant no2011GGE27053)
References
[1] M A Rahman and P Zhou ldquoAnalysis of brushless permanentmagnet synchronous motorsrdquo IEEE Transactions on IndustrialElectronics vol 43 no 2 pp 256ndash267 1996
[2] MOoshimaA Chiba A Rahman andT Fukao ldquoAn improvedcontrol method of buried-type IPM bearingless motors consid-ering magnetic saturation and magnetic pull variationrdquo IEEETransactions on Energy Conversion vol 19 no 3 pp 569ndash5752004
[3] K Liu Z Q Zhu Q Zhang and J Zhang ldquoInfluence ofnonideal voltage measurement on parameter estimation inpermanent-magnet synchronous machinesrdquo IEEE Transactionson Industrial Electronics vol 59 no 6 pp 2438ndash2447 2012
[4] F Caricchi F Crescimbini and O Honorati ldquoLow-cost com-pact permanent magnet machine for adjustable-speed pumpapplicationrdquo IEEETransactions on IndustryApplications vol 34no 1 pp 109ndash116 1998
[5] P Milanfar and J H Lang ldquoMonitoring the thermal conditionof permanent-magnet synchronous motorsrdquo IEEE Transactions
onAerospace and Electronic Systems vol 32 no 4 pp 1421ndash14291996
[6] T Kailath ldquoAn innovations approach to least-squares estima-tionmdashpart I linear filtering in additive white noiserdquo IEEETransactions on Automatic Control vol 13 pp 646ndash655 1968
[7] D G Robertson and J H Lee ldquoA least squares formulation forstate estimationrdquo Journal of Process Control vol 5 no 4 pp291ndash299 1995
[8] J S Gibson G H Lee and C F Wu ldquoLeast-squares estimationof inputoutput models for distributed linear systems in thepresence of noiserdquo Automatica vol 36 no 10 pp 1427ndash14422000
[9] S Tunali and I Batmaz ldquoDealing with the least squares regres-sion assumptions in simulation metamodelingrdquo Computers ampIndustrial Engineering vol 38 no 2 pp 307ndash320 2000
[10] R M Fernandez-Alcala J Navarro-Moreno and J C Ruiz-Molina ldquoLinear least-square estimation algorithms involvingcorrelated signal and noiserdquo IEEE Transactions on Signal Pro-cessing vol 53 no 11 pp 4227ndash4235 2005
[11] V Kratschmer ldquoLeast-squares estimation in linear regressionmodels with vague conceptsrdquo Fuzzy Sets and Systems vol 157no 19 pp 2579ndash2592 2006
[12] M J Garcıa-Ligero A Hermoso-Carazo and J Linares-PerezldquoLeast-squares linear estimation of signals from observationswith Markovian delaysrdquo Journal of Computational and AppliedMathematics vol 236 no 2 pp 234ndash242 2011
[13] S Ma C Quan R Zhu C J Tay L Chen and Z GaoldquoApplication of least-square estimation in white-light scanninginterferometryrdquo Optics and Lasers in Engineering vol 49 no 7pp 1012ndash1018 2011
[14] Q Wang and L Zhang ldquoLeast squares online linear discrimi-nant analysisrdquo Expert Systems with Applications vol 39 no 1pp 1510ndash1517 2012
[15] C J Demeure and L L Scharf ldquoSliding windows and latticealgorithms for computing QR factors in the least squares theoryof linear predictionrdquo IEEE Transactions on Acoustics Speechand Signal Processing vol 38 no 4 pp 721ndash725 1990
[16] K Zhao L Fuyun H Lev-Ari and J G Proakis ldquoSlidingwindow order-recursive least-squares algorithmsrdquo IEEE Trans-actions on Signal Processing vol 42 no 8 pp 1961ndash1972 1994
[17] H Liu and Z He ldquoA sliding-exponential window RLS adaptivefiltering algorithm properties and applicationsrdquo Signal Process-ing vol 45 no 3 pp 357ndash368 1995
[18] K Yoo and H Park ldquoFast residual computation for slidingwindow recursive least squares methodsrdquo Signal Processing vol45 no 1 pp 85ndash95 1995
[19] Y Xia M S Kamel and H Leung ldquoA fast algorithm for ARparameter estimation using a novel noise-constrained least-squares methodrdquo Neural Networks vol 23 no 3 pp 396ndash4052010
[20] A Aknouche E M Al-Eid and A M Hmeid ldquoOffline andonline weighted least squares estimation of nonstationarypower119860119877119862119867 processesrdquo Statistics amp Probability Letters vol 81no 10 pp 1535ndash1540 2011
[21] L Xie H Yang and F Ding ldquoRecursive least squares parameterestimation for non-uniformly sampled systems based on thedata filteringrdquo Mathematical and Computer Modelling vol 54no 1-2 pp 315ndash324 2011
[22] J Oliver R Aravind and K M M Prabhu ldquoImproved leastsquares channel estimation for orthogonal frequency divisionmultiplexingrdquo IET Signal Processing vol 6 no 1 pp 45ndash532012
Mathematical Problems in Engineering 11
[23] T Sadiki M Triki and D T M Slock ldquoWindow optimizationissues in recursive least-squares adaptive filtering and trackingrdquoin Proceedings of the 38th IEEE Annual Asilomar Conferenceon Signals Systems and Computers pp 940ndash944 Pacific GroveCalif USA November 2004
[24] G Welch and G Bishop ldquoAn introduction to the Kalman filterrdquo1997
[25] P J Hargrave ldquoA tutorial introduction to Kalman filteringrdquo inProceedings of the IEE Colloquium on Kalman Filters Introduc-tion Applications and Future Developments pp 11ndash16 1989
[26] M Gautier and P Poignet ldquoExtended Kalman filtering andweighted least squares dynamic identification of robotrdquo ControlEngineering Practice vol 9 no 12 pp 1361ndash1372 2001
[27] H M Al-Hamadi and S A Soliman ldquoKalman filter foridentification of power system fuzzy harmonic componentsrdquoElectric Power Systems Research vol 62 no 3 pp 241ndash248 2002
[28] T Boileau N Leboeuf B Nahid-Mobarakeh and F Meibody-Tabar ldquoOnline identification of PMSM parameters parameteridentifiability and estimator comparative studyrdquo IEEE Transac-tions on Industry Applications vol 47 no 4 pp 1944ndash1957 2011
[29] A Bechouche H Sediki D O Abdeslam and S HaddadldquoIdentification of induction motor at standstill using artificialneural networkrdquo in Proceedings of the 36th Annual Conferenceon IEEE Industrial Electronics Society (IECON rsquo10) pp 2908ndash2913 Glendale Ariz USA 2010
[30] F Alonge F DrsquoIppolito and FM Raimondi ldquoLeast squares andgenetic algorithms for parameter identification of inductionmotorsrdquo Control Engineering Practice vol 9 no 6 pp 647ndash6572001
[31] S Mishra ldquoA hybrid least square-fuzzy bacterial foragingstrategy for harmonic estimationrdquo IEEE Transactions on Evo-lutionary Computation vol 9 no 1 pp 61ndash73 2005
[32] R Liao H Zheng S Grzybowski and L Yang ldquoParticleswarm optimization-least squares support vector regressionbased forecasting model on dissolved gases in oil-filled powertransformersrdquo Electric Power Systems Research vol 81 no 12pp 2074ndash2080 2011
[33] R A Zadeh A Ghosh and G Ledwich ldquoCombination ofKalman filter and least-error square techniques in powersystemrdquo IEEE Transactions on Power Delivery vol 25 no 4 pp2868ndash2880 2010
[34] S Bolognani R Oboe andM Zigliotto ldquoSensorless full-digitalPMSM drive with EKF estimation of speed and rotor positionrdquoIEEE Transactions on Industrial Electronics vol 46 no 1 pp184ndash191 1999
[35] M Haardt ldquoStructured least squares to improve the per-formance of ESPRIT-Type algorithmsrdquo IEEE Transactions onSignal Processing vol 45 no 3 pp 792ndash799 1997
[36] M Ghogho A Swami and A K Nandi ldquoNon-linear leastsquares estimation for harmonics in multiplicative and additivenoiserdquo Signal Processing vol 78 no 1 pp 43ndash60 1999
[37] J Angeby ldquoEstimating signal parameters using the nonlinearinstantaneous least squares approachrdquo IEEE Transactions onSignal Processing vol 48 no 10 pp 2721ndash2732 2000
[38] J F Weng and S H Leung ldquoNonlinear RLS algorithm foramplitude estimation in class a noiserdquo IEE ProceedingsmdashCommunications vol 147 no 2 pp 81ndash86 2000
[39] D Zachariah M Sundin M Jansson and S ChatterjeeldquoAlternating least-squares for low-rank matrix reconstructionrdquoIEEE Signal Processing Letters vol 19 no 4 pp 231ndash234 2012
[40] R Montoliu and F Pla ldquoGeneralized least squares-based para-metric motion estimationrdquo Computer Vision and Image Under-standing vol 113 no 7 pp 790ndash801 2009
[41] Z Yingjie and G Liling ldquoImproved moving least squares algo-rithm for directed projecting onto point cloudsrdquoMeasurementvol 44 no 10 pp 2008ndash2019 2011
[42] S Seongwook J-S Lim S J Baek and K-M Sung ldquoVariableforgetting factor linear least squares algorithm for frequencyselective fading channel estimationrdquo IEEE Transactions onVehicular Technology vol 51 no 3 pp 613ndash616 2002
[43] S MorimotoM Sanada and Y Takeda ldquoMechanical sensorlessdrives of IPMSM with online parameter identificationrdquo IEEETransactions on Industry Applications vol 42 no 5 pp 1241ndash1248 2006
[44] H Sakai and H Nakaoka ldquoFast sliding window QRD-RLSalgorithmrdquo Signal Processing vol 78 no 3 pp 309ndash319 1999
[45] S Reece and S Roberts ldquoAn introduction to Gaussian processesfor the Kalman filter expertrdquo in Proceedings of the 13th Confer-ence on Information Fusion (FUSION rsquo10) pp 1ndash9 2010
[46] A Giorgano F M Hsu and J Wiley ldquoBook review least-square estimationwith applications to digital signal processingrdquoIEE Proceedings FmdashCommunications Radar and Signal Process-ing vol 132 no 7 1985
[47] S Wang R Zhao W Chen G Li and C Liu ldquoParameter iden-tification of PMSMbased on windowed least square algorithmrdquoin Proceedings of the Manufacturing Science and Technology(ICMST rsquo11) pp 5940ndash5944 Singapore 2012
[48] S Wang S Shi C Chen G Yang and Z Qu ldquoIdentificationof PMSM based on EKF and elman neural networkrdquo inProceedings of the IEEE International Conference on Automationand Logistics (ICAL rsquo09) pp 1459ndash1463 Shenyang China 2009
[49] R E Kalman ldquoA new approach to linear filtering and predictionproblemsrdquo Transactions of the ASMEmdashJournal of Basic Engi-neering D vol 82 pp 35ndash45 1960
Equation (11) is the recursive least square (RLS) estima-tion Recursive least square algorithm is called the general-ization Kalman filter algorithm [45 46] It is the engineeringrealization method of the least square estimation theory [17]
From the calculation of least square estimation andrecursive least square estimation we can see that the past timeof data has a big effect to future parameter estimation and alarge number of data calculations have occupied the systemresources seriously Therefore it is difficult to be realized inembedded systems
3 Windowed Least Square Estimation
Recursive least square algorithm can be used to real-timeparameter estimation However the algorithm uses the pasttime of data and the past time of data has the same impor-tance as the present data in the algorithm It weakened theimportance of current data caused a lot of system resourcespossession and affected the estimation speed and precision[44] In order to guarantee the instantaneity of parameterestimation the paper adopts windowed least squares (WLS)to estimate the parameters of PMSMThealgorithm simulatesthe window processing function of communication signalThe time series data used for parameter estimation are addedwindow handle to reduce the calculation of estimation andsystem resources possession making the algorithm easy forengineering realization
Suppose that 119875(119898) = (119883119879
119898119883119898)minus1 the parameter estima-
tion by (119909(119896) 119910(119896)) 119896 = 1 2 119898 is as follows
120579 (119898) = 119875 (119898)119883119879
119898119884119898 (12)
where
119883119898
=
[[[[
[
119909119879
(1)
119909119879
(2)
119909119879
(119898)
]]]]
]
119884119898
=
[[[[
[
119910 (1)
119910 (2)
119910 (119898)
]]]]
]
(13)
Consider increasing a group of data (119909(119898+ 1) 119910(119898+ 1))and then the parameter estimation is shown as follows [16 2344]
The window size is adjustable according to actual needsbased on the data length of regulation This can guaranteethe speed of calculation and can reduce the system resourcespossession too At this time the parameter estimation isrelated to the current 119898 data sample the past time of datahas no effect on parameter estimation and this can ensurethe instantaneity and accuracy of the parameter estimation
4 Simulations
41 PMSM Model The voltage equations flux linkage equa-tions and electromagnetic torque equations of PMSM in 119889 119902frames are as follows [47 48]
where119906119902and119906119889are voltages in 119902-axis and119889-axis respectively
119894119902and 119894
119889are currents in 119902-axis and 119889-axis 119877
119904is phase
resistance of stator 119871119889and 119871
119902are inductances in 119889-axis and
119902-axis 120596119903is rotor velocity 120595
119891is flux linkage established by
magnets and 119901 is the differential operatorThe mathematical model of PMSM is discretized to
estimate parameters (119877119904 119871119889 and 119871
119902) The discrete model of
PMSM is as follows
119906119902(119896) = 119877
119904119894119902(119896) + 119901120595
119902(119896) + 120596
119903120595119889(119896)
119906119889(119896) = 119877
119904119894119889(119896) + 119901120595
119889(119896) + 120596
119903120595119902(119896)
(25)
where 120595119889and 120595
119902are as follows
120595119889(119896) = 119871
119889119894119889(119896) + 120595
119891
120595119902(119896) = 119871
119902119894119902(119896)
(26)
The PMSM simulation model is established by MAT-LABSIMULINK and the PMSM running data is obtainedby the model The simulation model is shown in Figure 1
In MATLABSIMULINK the 119877119904 119871119889 and 119871
119902parameters
of PMSM are fixed in simulation We cannot simulate thetime varying characteristic of 119877
119904 119871119889 and 119871
119902 Therefore we
design a motor simulation model according to the require-ment The 119877
119904 119871119889 and 119871
119902can be changed flexibly in the
simulation
42 PMSM Parameters Estimation of Windowed Least SquareAlgorithm When 119877
119904 119871119889 and 119871
119902are fixed value the PMSM
simulation data is obtained by MATLAB The windowedleast square algorithm is used to identify the parametersThe algorithm with different window sizes is used for 119877
119904
119871119889 and 119871
119902identification From the identification result
(Figures 2 3 and 4) we can see that bigger window sizehas a better identification result However the window sizedoes not obviously have an effect on the promotion ofparameter identification precision when 119877
119904 119871119889 and 119871
119902are
fixedDifferent window sizes have big effect on the results of
parameter estimation when the estimated parameters havetime varying characteristicThemotor parameters119877
119904119871119889 and
119871119902are measured at 234∘C 30∘C 40∘C 50∘C 60∘C 70∘C and
80∘C Using piecewise linear method to simulate the timevarying of the three parameters the motor running data isobtained by MATLAB
The windowed least square is used to identify the 119877119904 119871119889
and 119871119902 The identification result is shown in Figures 5 6 and
7 when the 119877119904 119871119889and 119871
119902are changed at the same time From
the Figures 5ndash7 we can see that shorter window size has lowereffect on the promotion of identification precision Howeverthe window size is too big to improve the identificationprecisionThe identification result is better when the windowsize is 300ndash400 In the motor model 120595
Figure 5 Estimation result of 119877119904of WLS with different window
sizes when the parameters have time varying characteristic
method It is proposed in 1960 by Kalman [49] The theoryis applied to practical engineering immediately when it isput forward The Apollo program and C-5 plane navigationsystem design are the most successful application examplesExtended Kalman filter (EKF) is an improved model of theKalman filter which is one of the most widely applied innonlinear system filter
Mathematical Problems in Engineering 7
Discrete system state equation of EKF is
X (119896) = A (119896 minus 1)X (119896 minus 1) + B (119896 minus 1)U (119896 minus 1)
+ C (119896 minus 1) + w (119896 minus 1)
Z (119896) = H (119896 minus 1)X (119896) + k (119896)
where X(119896) is the system state vector U(119896) is the systeminput vector Z(119896) is the system observation vector w(119896) isthe system random noise vector k(119896) is the system randomobservation noise vector w(119896) and k(119896) are noise sequenceswith zero mean and the covariance matrices are Q(119896) andR(119896)
119879119904is the sampling period the discrete linear state space
equation (29) of PMSM is established by discretization andlinearization of the model (30)
Figure 6 Estimation result of 119871119889of WLS with different window
sizes when the parameters have time varying characteristic
Using the simulation model and getting the motor output 119894119889
119894119902 119906119889 119906119902120596119890 the initial values of119875119876 and119877 in EKF algorithm
are
119875 = diag ([01 01 00004 0002 002])
119876 = diag ([30 15 0005 003 003])
119877 = diag ([01 002])
(32)
When119877119904 119871119889 and 119871
119902of motor are fixed the identification
result table of EKF recursive least square and windowed leastsquare algorithm is shown in Table 1
From Table 1 we can see that identification result ofEKF algorithm is as good as the windowed least squarealgorithm when 119877
119904 119871119889 and 119871
119902are fixed The comparison
diagrams of identification result are shown in Figure 8 whenthe parameters are fixed
EKF and recursive least square algorithms cannot achievereasonable result when the parameters have time varyingcharacteristic or have a drastic change However windowedleast square algorithm can achieve good identification resultwhen 119877
119904 119871119889 and 119871
119902have time varying characteristic at the
same time (Figures 5ndash7)
5 Analysis and Conclusion
Through the previous different PMSM parameters identifica-tion experiments we can see the following
(1) When the parameters of PMSM have no time vary-ing characteristic three methods can achieve better
Figure 7 Estimation result of 119871119902of WLS with different window
sizes when the parameters have time varying characteristic
identification result in precision and accuracy Inthe calculation and instantaneity of identificationrecursive least square algorithm has a fatal flaw ofdata saturation so the precision and accuracy of thealgorithm are hard to guarantee It is difficult torealize in embedded system the real-time parametersidentification due to the amount of calculations andsystem resources possession of EKF [28]Thewindowsize of windowed least square algorithm is flexibleso we can choose the collected data according tothe changes of the parameters It will reduce theinfluence of the past time of data to the currentparameter identification guarantee the accuracy andinstantaneity of identification and reduce the systemresources possession at the same time
(2) When the parameters of PMSM have strong timevarying characteristic the EKF and recursive leastsquare algorithms cannot guarantee the precisionand accuracy of identification However windowedleast square algorithm can get better identificationresultTherefore EKF and recursive least square algo-rithms are suitable for fixed parameters estimationor parameters with weak time varying characteristicidentification Windowed least square algorithm canget a good result both for fixed parameters and fortime varying parameters identification
Embedded technology is widely used in the motor driverand controller at present However the embedded chip(MCU DSP ARM etc) has certain restriction in computingspeed and storage space Therefore windowed least square
Mathematical Problems in Engineering 9
Table 1 Estimation result comparison of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
Temperature (∘C) 234 30 40 50 60 70 80
119877119904
Actual data (Ω) 09664 10008 10373 10770 11245 11592 11751EKF
Figure 8 Comparison diagram of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
10 Mathematical Problems in Engineering
Figure 9 The PMSM experiment system
Figure 10 The prototype DSP-based PMSM driver
algorithm is a better choice for PMSM parameters identi-fication of motor driver and controller This paper is thebeginning of work There are a lot of work to do such astransplant the algorithm to practical controller and controlsystem (in Figures 9 and 10) which is designed to control thePMSM in practical application
Acknowledgment
This paper is supported by the Shandong Province Scienceand Technology Development Plan of China (Grant no2011GGE27053)
References
[1] M A Rahman and P Zhou ldquoAnalysis of brushless permanentmagnet synchronous motorsrdquo IEEE Transactions on IndustrialElectronics vol 43 no 2 pp 256ndash267 1996
[2] MOoshimaA Chiba A Rahman andT Fukao ldquoAn improvedcontrol method of buried-type IPM bearingless motors consid-ering magnetic saturation and magnetic pull variationrdquo IEEETransactions on Energy Conversion vol 19 no 3 pp 569ndash5752004
[3] K Liu Z Q Zhu Q Zhang and J Zhang ldquoInfluence ofnonideal voltage measurement on parameter estimation inpermanent-magnet synchronous machinesrdquo IEEE Transactionson Industrial Electronics vol 59 no 6 pp 2438ndash2447 2012
[4] F Caricchi F Crescimbini and O Honorati ldquoLow-cost com-pact permanent magnet machine for adjustable-speed pumpapplicationrdquo IEEETransactions on IndustryApplications vol 34no 1 pp 109ndash116 1998
[5] P Milanfar and J H Lang ldquoMonitoring the thermal conditionof permanent-magnet synchronous motorsrdquo IEEE Transactions
onAerospace and Electronic Systems vol 32 no 4 pp 1421ndash14291996
[6] T Kailath ldquoAn innovations approach to least-squares estima-tionmdashpart I linear filtering in additive white noiserdquo IEEETransactions on Automatic Control vol 13 pp 646ndash655 1968
[7] D G Robertson and J H Lee ldquoA least squares formulation forstate estimationrdquo Journal of Process Control vol 5 no 4 pp291ndash299 1995
[8] J S Gibson G H Lee and C F Wu ldquoLeast-squares estimationof inputoutput models for distributed linear systems in thepresence of noiserdquo Automatica vol 36 no 10 pp 1427ndash14422000
[9] S Tunali and I Batmaz ldquoDealing with the least squares regres-sion assumptions in simulation metamodelingrdquo Computers ampIndustrial Engineering vol 38 no 2 pp 307ndash320 2000
[10] R M Fernandez-Alcala J Navarro-Moreno and J C Ruiz-Molina ldquoLinear least-square estimation algorithms involvingcorrelated signal and noiserdquo IEEE Transactions on Signal Pro-cessing vol 53 no 11 pp 4227ndash4235 2005
[11] V Kratschmer ldquoLeast-squares estimation in linear regressionmodels with vague conceptsrdquo Fuzzy Sets and Systems vol 157no 19 pp 2579ndash2592 2006
[12] M J Garcıa-Ligero A Hermoso-Carazo and J Linares-PerezldquoLeast-squares linear estimation of signals from observationswith Markovian delaysrdquo Journal of Computational and AppliedMathematics vol 236 no 2 pp 234ndash242 2011
[13] S Ma C Quan R Zhu C J Tay L Chen and Z GaoldquoApplication of least-square estimation in white-light scanninginterferometryrdquo Optics and Lasers in Engineering vol 49 no 7pp 1012ndash1018 2011
[14] Q Wang and L Zhang ldquoLeast squares online linear discrimi-nant analysisrdquo Expert Systems with Applications vol 39 no 1pp 1510ndash1517 2012
[15] C J Demeure and L L Scharf ldquoSliding windows and latticealgorithms for computing QR factors in the least squares theoryof linear predictionrdquo IEEE Transactions on Acoustics Speechand Signal Processing vol 38 no 4 pp 721ndash725 1990
[16] K Zhao L Fuyun H Lev-Ari and J G Proakis ldquoSlidingwindow order-recursive least-squares algorithmsrdquo IEEE Trans-actions on Signal Processing vol 42 no 8 pp 1961ndash1972 1994
[17] H Liu and Z He ldquoA sliding-exponential window RLS adaptivefiltering algorithm properties and applicationsrdquo Signal Process-ing vol 45 no 3 pp 357ndash368 1995
[18] K Yoo and H Park ldquoFast residual computation for slidingwindow recursive least squares methodsrdquo Signal Processing vol45 no 1 pp 85ndash95 1995
[19] Y Xia M S Kamel and H Leung ldquoA fast algorithm for ARparameter estimation using a novel noise-constrained least-squares methodrdquo Neural Networks vol 23 no 3 pp 396ndash4052010
[20] A Aknouche E M Al-Eid and A M Hmeid ldquoOffline andonline weighted least squares estimation of nonstationarypower119860119877119862119867 processesrdquo Statistics amp Probability Letters vol 81no 10 pp 1535ndash1540 2011
[21] L Xie H Yang and F Ding ldquoRecursive least squares parameterestimation for non-uniformly sampled systems based on thedata filteringrdquo Mathematical and Computer Modelling vol 54no 1-2 pp 315ndash324 2011
[22] J Oliver R Aravind and K M M Prabhu ldquoImproved leastsquares channel estimation for orthogonal frequency divisionmultiplexingrdquo IET Signal Processing vol 6 no 1 pp 45ndash532012
Mathematical Problems in Engineering 11
[23] T Sadiki M Triki and D T M Slock ldquoWindow optimizationissues in recursive least-squares adaptive filtering and trackingrdquoin Proceedings of the 38th IEEE Annual Asilomar Conferenceon Signals Systems and Computers pp 940ndash944 Pacific GroveCalif USA November 2004
[24] G Welch and G Bishop ldquoAn introduction to the Kalman filterrdquo1997
[25] P J Hargrave ldquoA tutorial introduction to Kalman filteringrdquo inProceedings of the IEE Colloquium on Kalman Filters Introduc-tion Applications and Future Developments pp 11ndash16 1989
[26] M Gautier and P Poignet ldquoExtended Kalman filtering andweighted least squares dynamic identification of robotrdquo ControlEngineering Practice vol 9 no 12 pp 1361ndash1372 2001
[27] H M Al-Hamadi and S A Soliman ldquoKalman filter foridentification of power system fuzzy harmonic componentsrdquoElectric Power Systems Research vol 62 no 3 pp 241ndash248 2002
[28] T Boileau N Leboeuf B Nahid-Mobarakeh and F Meibody-Tabar ldquoOnline identification of PMSM parameters parameteridentifiability and estimator comparative studyrdquo IEEE Transac-tions on Industry Applications vol 47 no 4 pp 1944ndash1957 2011
[29] A Bechouche H Sediki D O Abdeslam and S HaddadldquoIdentification of induction motor at standstill using artificialneural networkrdquo in Proceedings of the 36th Annual Conferenceon IEEE Industrial Electronics Society (IECON rsquo10) pp 2908ndash2913 Glendale Ariz USA 2010
[30] F Alonge F DrsquoIppolito and FM Raimondi ldquoLeast squares andgenetic algorithms for parameter identification of inductionmotorsrdquo Control Engineering Practice vol 9 no 6 pp 647ndash6572001
[31] S Mishra ldquoA hybrid least square-fuzzy bacterial foragingstrategy for harmonic estimationrdquo IEEE Transactions on Evo-lutionary Computation vol 9 no 1 pp 61ndash73 2005
[32] R Liao H Zheng S Grzybowski and L Yang ldquoParticleswarm optimization-least squares support vector regressionbased forecasting model on dissolved gases in oil-filled powertransformersrdquo Electric Power Systems Research vol 81 no 12pp 2074ndash2080 2011
[33] R A Zadeh A Ghosh and G Ledwich ldquoCombination ofKalman filter and least-error square techniques in powersystemrdquo IEEE Transactions on Power Delivery vol 25 no 4 pp2868ndash2880 2010
[34] S Bolognani R Oboe andM Zigliotto ldquoSensorless full-digitalPMSM drive with EKF estimation of speed and rotor positionrdquoIEEE Transactions on Industrial Electronics vol 46 no 1 pp184ndash191 1999
[35] M Haardt ldquoStructured least squares to improve the per-formance of ESPRIT-Type algorithmsrdquo IEEE Transactions onSignal Processing vol 45 no 3 pp 792ndash799 1997
[36] M Ghogho A Swami and A K Nandi ldquoNon-linear leastsquares estimation for harmonics in multiplicative and additivenoiserdquo Signal Processing vol 78 no 1 pp 43ndash60 1999
[37] J Angeby ldquoEstimating signal parameters using the nonlinearinstantaneous least squares approachrdquo IEEE Transactions onSignal Processing vol 48 no 10 pp 2721ndash2732 2000
[38] J F Weng and S H Leung ldquoNonlinear RLS algorithm foramplitude estimation in class a noiserdquo IEE ProceedingsmdashCommunications vol 147 no 2 pp 81ndash86 2000
[39] D Zachariah M Sundin M Jansson and S ChatterjeeldquoAlternating least-squares for low-rank matrix reconstructionrdquoIEEE Signal Processing Letters vol 19 no 4 pp 231ndash234 2012
[40] R Montoliu and F Pla ldquoGeneralized least squares-based para-metric motion estimationrdquo Computer Vision and Image Under-standing vol 113 no 7 pp 790ndash801 2009
[41] Z Yingjie and G Liling ldquoImproved moving least squares algo-rithm for directed projecting onto point cloudsrdquoMeasurementvol 44 no 10 pp 2008ndash2019 2011
[42] S Seongwook J-S Lim S J Baek and K-M Sung ldquoVariableforgetting factor linear least squares algorithm for frequencyselective fading channel estimationrdquo IEEE Transactions onVehicular Technology vol 51 no 3 pp 613ndash616 2002
[43] S MorimotoM Sanada and Y Takeda ldquoMechanical sensorlessdrives of IPMSM with online parameter identificationrdquo IEEETransactions on Industry Applications vol 42 no 5 pp 1241ndash1248 2006
[44] H Sakai and H Nakaoka ldquoFast sliding window QRD-RLSalgorithmrdquo Signal Processing vol 78 no 3 pp 309ndash319 1999
[45] S Reece and S Roberts ldquoAn introduction to Gaussian processesfor the Kalman filter expertrdquo in Proceedings of the 13th Confer-ence on Information Fusion (FUSION rsquo10) pp 1ndash9 2010
[46] A Giorgano F M Hsu and J Wiley ldquoBook review least-square estimationwith applications to digital signal processingrdquoIEE Proceedings FmdashCommunications Radar and Signal Process-ing vol 132 no 7 1985
[47] S Wang R Zhao W Chen G Li and C Liu ldquoParameter iden-tification of PMSMbased on windowed least square algorithmrdquoin Proceedings of the Manufacturing Science and Technology(ICMST rsquo11) pp 5940ndash5944 Singapore 2012
[48] S Wang S Shi C Chen G Yang and Z Qu ldquoIdentificationof PMSM based on EKF and elman neural networkrdquo inProceedings of the IEEE International Conference on Automationand Logistics (ICAL rsquo09) pp 1459ndash1463 Shenyang China 2009
[49] R E Kalman ldquoA new approach to linear filtering and predictionproblemsrdquo Transactions of the ASMEmdashJournal of Basic Engi-neering D vol 82 pp 35ndash45 1960
The window size is adjustable according to actual needsbased on the data length of regulation This can guaranteethe speed of calculation and can reduce the system resourcespossession too At this time the parameter estimation isrelated to the current 119898 data sample the past time of datahas no effect on parameter estimation and this can ensurethe instantaneity and accuracy of the parameter estimation
4 Simulations
41 PMSM Model The voltage equations flux linkage equa-tions and electromagnetic torque equations of PMSM in 119889 119902frames are as follows [47 48]
where119906119902and119906119889are voltages in 119902-axis and119889-axis respectively
119894119902and 119894
119889are currents in 119902-axis and 119889-axis 119877
119904is phase
resistance of stator 119871119889and 119871
119902are inductances in 119889-axis and
119902-axis 120596119903is rotor velocity 120595
119891is flux linkage established by
magnets and 119901 is the differential operatorThe mathematical model of PMSM is discretized to
estimate parameters (119877119904 119871119889 and 119871
119902) The discrete model of
PMSM is as follows
119906119902(119896) = 119877
119904119894119902(119896) + 119901120595
119902(119896) + 120596
119903120595119889(119896)
119906119889(119896) = 119877
119904119894119889(119896) + 119901120595
119889(119896) + 120596
119903120595119902(119896)
(25)
where 120595119889and 120595
119902are as follows
120595119889(119896) = 119871
119889119894119889(119896) + 120595
119891
120595119902(119896) = 119871
119902119894119902(119896)
(26)
The PMSM simulation model is established by MAT-LABSIMULINK and the PMSM running data is obtainedby the model The simulation model is shown in Figure 1
In MATLABSIMULINK the 119877119904 119871119889 and 119871
119902parameters
of PMSM are fixed in simulation We cannot simulate thetime varying characteristic of 119877
119904 119871119889 and 119871
119902 Therefore we
design a motor simulation model according to the require-ment The 119877
119904 119871119889 and 119871
119902can be changed flexibly in the
simulation
42 PMSM Parameters Estimation of Windowed Least SquareAlgorithm When 119877
119904 119871119889 and 119871
119902are fixed value the PMSM
simulation data is obtained by MATLAB The windowedleast square algorithm is used to identify the parametersThe algorithm with different window sizes is used for 119877
119904
119871119889 and 119871
119902identification From the identification result
(Figures 2 3 and 4) we can see that bigger window sizehas a better identification result However the window sizedoes not obviously have an effect on the promotion ofparameter identification precision when 119877
119904 119871119889 and 119871
119902are
fixedDifferent window sizes have big effect on the results of
parameter estimation when the estimated parameters havetime varying characteristicThemotor parameters119877
119904119871119889 and
119871119902are measured at 234∘C 30∘C 40∘C 50∘C 60∘C 70∘C and
80∘C Using piecewise linear method to simulate the timevarying of the three parameters the motor running data isobtained by MATLAB
The windowed least square is used to identify the 119877119904 119871119889
and 119871119902 The identification result is shown in Figures 5 6 and
7 when the 119877119904 119871119889and 119871
119902are changed at the same time From
the Figures 5ndash7 we can see that shorter window size has lowereffect on the promotion of identification precision Howeverthe window size is too big to improve the identificationprecisionThe identification result is better when the windowsize is 300ndash400 In the motor model 120595
Figure 5 Estimation result of 119877119904of WLS with different window
sizes when the parameters have time varying characteristic
method It is proposed in 1960 by Kalman [49] The theoryis applied to practical engineering immediately when it isput forward The Apollo program and C-5 plane navigationsystem design are the most successful application examplesExtended Kalman filter (EKF) is an improved model of theKalman filter which is one of the most widely applied innonlinear system filter
Mathematical Problems in Engineering 7
Discrete system state equation of EKF is
X (119896) = A (119896 minus 1)X (119896 minus 1) + B (119896 minus 1)U (119896 minus 1)
+ C (119896 minus 1) + w (119896 minus 1)
Z (119896) = H (119896 minus 1)X (119896) + k (119896)
where X(119896) is the system state vector U(119896) is the systeminput vector Z(119896) is the system observation vector w(119896) isthe system random noise vector k(119896) is the system randomobservation noise vector w(119896) and k(119896) are noise sequenceswith zero mean and the covariance matrices are Q(119896) andR(119896)
119879119904is the sampling period the discrete linear state space
equation (29) of PMSM is established by discretization andlinearization of the model (30)
Figure 6 Estimation result of 119871119889of WLS with different window
sizes when the parameters have time varying characteristic
Using the simulation model and getting the motor output 119894119889
119894119902 119906119889 119906119902120596119890 the initial values of119875119876 and119877 in EKF algorithm
are
119875 = diag ([01 01 00004 0002 002])
119876 = diag ([30 15 0005 003 003])
119877 = diag ([01 002])
(32)
When119877119904 119871119889 and 119871
119902of motor are fixed the identification
result table of EKF recursive least square and windowed leastsquare algorithm is shown in Table 1
From Table 1 we can see that identification result ofEKF algorithm is as good as the windowed least squarealgorithm when 119877
119904 119871119889 and 119871
119902are fixed The comparison
diagrams of identification result are shown in Figure 8 whenthe parameters are fixed
EKF and recursive least square algorithms cannot achievereasonable result when the parameters have time varyingcharacteristic or have a drastic change However windowedleast square algorithm can achieve good identification resultwhen 119877
119904 119871119889 and 119871
119902have time varying characteristic at the
same time (Figures 5ndash7)
5 Analysis and Conclusion
Through the previous different PMSM parameters identifica-tion experiments we can see the following
(1) When the parameters of PMSM have no time vary-ing characteristic three methods can achieve better
Figure 7 Estimation result of 119871119902of WLS with different window
sizes when the parameters have time varying characteristic
identification result in precision and accuracy Inthe calculation and instantaneity of identificationrecursive least square algorithm has a fatal flaw ofdata saturation so the precision and accuracy of thealgorithm are hard to guarantee It is difficult torealize in embedded system the real-time parametersidentification due to the amount of calculations andsystem resources possession of EKF [28]Thewindowsize of windowed least square algorithm is flexibleso we can choose the collected data according tothe changes of the parameters It will reduce theinfluence of the past time of data to the currentparameter identification guarantee the accuracy andinstantaneity of identification and reduce the systemresources possession at the same time
(2) When the parameters of PMSM have strong timevarying characteristic the EKF and recursive leastsquare algorithms cannot guarantee the precisionand accuracy of identification However windowedleast square algorithm can get better identificationresultTherefore EKF and recursive least square algo-rithms are suitable for fixed parameters estimationor parameters with weak time varying characteristicidentification Windowed least square algorithm canget a good result both for fixed parameters and fortime varying parameters identification
Embedded technology is widely used in the motor driverand controller at present However the embedded chip(MCU DSP ARM etc) has certain restriction in computingspeed and storage space Therefore windowed least square
Mathematical Problems in Engineering 9
Table 1 Estimation result comparison of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
Temperature (∘C) 234 30 40 50 60 70 80
119877119904
Actual data (Ω) 09664 10008 10373 10770 11245 11592 11751EKF
Figure 8 Comparison diagram of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
10 Mathematical Problems in Engineering
Figure 9 The PMSM experiment system
Figure 10 The prototype DSP-based PMSM driver
algorithm is a better choice for PMSM parameters identi-fication of motor driver and controller This paper is thebeginning of work There are a lot of work to do such astransplant the algorithm to practical controller and controlsystem (in Figures 9 and 10) which is designed to control thePMSM in practical application
Acknowledgment
This paper is supported by the Shandong Province Scienceand Technology Development Plan of China (Grant no2011GGE27053)
References
[1] M A Rahman and P Zhou ldquoAnalysis of brushless permanentmagnet synchronous motorsrdquo IEEE Transactions on IndustrialElectronics vol 43 no 2 pp 256ndash267 1996
[2] MOoshimaA Chiba A Rahman andT Fukao ldquoAn improvedcontrol method of buried-type IPM bearingless motors consid-ering magnetic saturation and magnetic pull variationrdquo IEEETransactions on Energy Conversion vol 19 no 3 pp 569ndash5752004
[3] K Liu Z Q Zhu Q Zhang and J Zhang ldquoInfluence ofnonideal voltage measurement on parameter estimation inpermanent-magnet synchronous machinesrdquo IEEE Transactionson Industrial Electronics vol 59 no 6 pp 2438ndash2447 2012
[4] F Caricchi F Crescimbini and O Honorati ldquoLow-cost com-pact permanent magnet machine for adjustable-speed pumpapplicationrdquo IEEETransactions on IndustryApplications vol 34no 1 pp 109ndash116 1998
[5] P Milanfar and J H Lang ldquoMonitoring the thermal conditionof permanent-magnet synchronous motorsrdquo IEEE Transactions
onAerospace and Electronic Systems vol 32 no 4 pp 1421ndash14291996
[6] T Kailath ldquoAn innovations approach to least-squares estima-tionmdashpart I linear filtering in additive white noiserdquo IEEETransactions on Automatic Control vol 13 pp 646ndash655 1968
[7] D G Robertson and J H Lee ldquoA least squares formulation forstate estimationrdquo Journal of Process Control vol 5 no 4 pp291ndash299 1995
[8] J S Gibson G H Lee and C F Wu ldquoLeast-squares estimationof inputoutput models for distributed linear systems in thepresence of noiserdquo Automatica vol 36 no 10 pp 1427ndash14422000
[9] S Tunali and I Batmaz ldquoDealing with the least squares regres-sion assumptions in simulation metamodelingrdquo Computers ampIndustrial Engineering vol 38 no 2 pp 307ndash320 2000
[10] R M Fernandez-Alcala J Navarro-Moreno and J C Ruiz-Molina ldquoLinear least-square estimation algorithms involvingcorrelated signal and noiserdquo IEEE Transactions on Signal Pro-cessing vol 53 no 11 pp 4227ndash4235 2005
[11] V Kratschmer ldquoLeast-squares estimation in linear regressionmodels with vague conceptsrdquo Fuzzy Sets and Systems vol 157no 19 pp 2579ndash2592 2006
[12] M J Garcıa-Ligero A Hermoso-Carazo and J Linares-PerezldquoLeast-squares linear estimation of signals from observationswith Markovian delaysrdquo Journal of Computational and AppliedMathematics vol 236 no 2 pp 234ndash242 2011
[13] S Ma C Quan R Zhu C J Tay L Chen and Z GaoldquoApplication of least-square estimation in white-light scanninginterferometryrdquo Optics and Lasers in Engineering vol 49 no 7pp 1012ndash1018 2011
[14] Q Wang and L Zhang ldquoLeast squares online linear discrimi-nant analysisrdquo Expert Systems with Applications vol 39 no 1pp 1510ndash1517 2012
[15] C J Demeure and L L Scharf ldquoSliding windows and latticealgorithms for computing QR factors in the least squares theoryof linear predictionrdquo IEEE Transactions on Acoustics Speechand Signal Processing vol 38 no 4 pp 721ndash725 1990
[16] K Zhao L Fuyun H Lev-Ari and J G Proakis ldquoSlidingwindow order-recursive least-squares algorithmsrdquo IEEE Trans-actions on Signal Processing vol 42 no 8 pp 1961ndash1972 1994
[17] H Liu and Z He ldquoA sliding-exponential window RLS adaptivefiltering algorithm properties and applicationsrdquo Signal Process-ing vol 45 no 3 pp 357ndash368 1995
[18] K Yoo and H Park ldquoFast residual computation for slidingwindow recursive least squares methodsrdquo Signal Processing vol45 no 1 pp 85ndash95 1995
[19] Y Xia M S Kamel and H Leung ldquoA fast algorithm for ARparameter estimation using a novel noise-constrained least-squares methodrdquo Neural Networks vol 23 no 3 pp 396ndash4052010
[20] A Aknouche E M Al-Eid and A M Hmeid ldquoOffline andonline weighted least squares estimation of nonstationarypower119860119877119862119867 processesrdquo Statistics amp Probability Letters vol 81no 10 pp 1535ndash1540 2011
[21] L Xie H Yang and F Ding ldquoRecursive least squares parameterestimation for non-uniformly sampled systems based on thedata filteringrdquo Mathematical and Computer Modelling vol 54no 1-2 pp 315ndash324 2011
[22] J Oliver R Aravind and K M M Prabhu ldquoImproved leastsquares channel estimation for orthogonal frequency divisionmultiplexingrdquo IET Signal Processing vol 6 no 1 pp 45ndash532012
Mathematical Problems in Engineering 11
[23] T Sadiki M Triki and D T M Slock ldquoWindow optimizationissues in recursive least-squares adaptive filtering and trackingrdquoin Proceedings of the 38th IEEE Annual Asilomar Conferenceon Signals Systems and Computers pp 940ndash944 Pacific GroveCalif USA November 2004
[24] G Welch and G Bishop ldquoAn introduction to the Kalman filterrdquo1997
[25] P J Hargrave ldquoA tutorial introduction to Kalman filteringrdquo inProceedings of the IEE Colloquium on Kalman Filters Introduc-tion Applications and Future Developments pp 11ndash16 1989
[26] M Gautier and P Poignet ldquoExtended Kalman filtering andweighted least squares dynamic identification of robotrdquo ControlEngineering Practice vol 9 no 12 pp 1361ndash1372 2001
[27] H M Al-Hamadi and S A Soliman ldquoKalman filter foridentification of power system fuzzy harmonic componentsrdquoElectric Power Systems Research vol 62 no 3 pp 241ndash248 2002
[28] T Boileau N Leboeuf B Nahid-Mobarakeh and F Meibody-Tabar ldquoOnline identification of PMSM parameters parameteridentifiability and estimator comparative studyrdquo IEEE Transac-tions on Industry Applications vol 47 no 4 pp 1944ndash1957 2011
[29] A Bechouche H Sediki D O Abdeslam and S HaddadldquoIdentification of induction motor at standstill using artificialneural networkrdquo in Proceedings of the 36th Annual Conferenceon IEEE Industrial Electronics Society (IECON rsquo10) pp 2908ndash2913 Glendale Ariz USA 2010
[30] F Alonge F DrsquoIppolito and FM Raimondi ldquoLeast squares andgenetic algorithms for parameter identification of inductionmotorsrdquo Control Engineering Practice vol 9 no 6 pp 647ndash6572001
[31] S Mishra ldquoA hybrid least square-fuzzy bacterial foragingstrategy for harmonic estimationrdquo IEEE Transactions on Evo-lutionary Computation vol 9 no 1 pp 61ndash73 2005
[32] R Liao H Zheng S Grzybowski and L Yang ldquoParticleswarm optimization-least squares support vector regressionbased forecasting model on dissolved gases in oil-filled powertransformersrdquo Electric Power Systems Research vol 81 no 12pp 2074ndash2080 2011
[33] R A Zadeh A Ghosh and G Ledwich ldquoCombination ofKalman filter and least-error square techniques in powersystemrdquo IEEE Transactions on Power Delivery vol 25 no 4 pp2868ndash2880 2010
[34] S Bolognani R Oboe andM Zigliotto ldquoSensorless full-digitalPMSM drive with EKF estimation of speed and rotor positionrdquoIEEE Transactions on Industrial Electronics vol 46 no 1 pp184ndash191 1999
[35] M Haardt ldquoStructured least squares to improve the per-formance of ESPRIT-Type algorithmsrdquo IEEE Transactions onSignal Processing vol 45 no 3 pp 792ndash799 1997
[36] M Ghogho A Swami and A K Nandi ldquoNon-linear leastsquares estimation for harmonics in multiplicative and additivenoiserdquo Signal Processing vol 78 no 1 pp 43ndash60 1999
[37] J Angeby ldquoEstimating signal parameters using the nonlinearinstantaneous least squares approachrdquo IEEE Transactions onSignal Processing vol 48 no 10 pp 2721ndash2732 2000
[38] J F Weng and S H Leung ldquoNonlinear RLS algorithm foramplitude estimation in class a noiserdquo IEE ProceedingsmdashCommunications vol 147 no 2 pp 81ndash86 2000
[39] D Zachariah M Sundin M Jansson and S ChatterjeeldquoAlternating least-squares for low-rank matrix reconstructionrdquoIEEE Signal Processing Letters vol 19 no 4 pp 231ndash234 2012
[40] R Montoliu and F Pla ldquoGeneralized least squares-based para-metric motion estimationrdquo Computer Vision and Image Under-standing vol 113 no 7 pp 790ndash801 2009
[41] Z Yingjie and G Liling ldquoImproved moving least squares algo-rithm for directed projecting onto point cloudsrdquoMeasurementvol 44 no 10 pp 2008ndash2019 2011
[42] S Seongwook J-S Lim S J Baek and K-M Sung ldquoVariableforgetting factor linear least squares algorithm for frequencyselective fading channel estimationrdquo IEEE Transactions onVehicular Technology vol 51 no 3 pp 613ndash616 2002
[43] S MorimotoM Sanada and Y Takeda ldquoMechanical sensorlessdrives of IPMSM with online parameter identificationrdquo IEEETransactions on Industry Applications vol 42 no 5 pp 1241ndash1248 2006
[44] H Sakai and H Nakaoka ldquoFast sliding window QRD-RLSalgorithmrdquo Signal Processing vol 78 no 3 pp 309ndash319 1999
[45] S Reece and S Roberts ldquoAn introduction to Gaussian processesfor the Kalman filter expertrdquo in Proceedings of the 13th Confer-ence on Information Fusion (FUSION rsquo10) pp 1ndash9 2010
[46] A Giorgano F M Hsu and J Wiley ldquoBook review least-square estimationwith applications to digital signal processingrdquoIEE Proceedings FmdashCommunications Radar and Signal Process-ing vol 132 no 7 1985
[47] S Wang R Zhao W Chen G Li and C Liu ldquoParameter iden-tification of PMSMbased on windowed least square algorithmrdquoin Proceedings of the Manufacturing Science and Technology(ICMST rsquo11) pp 5940ndash5944 Singapore 2012
[48] S Wang S Shi C Chen G Yang and Z Qu ldquoIdentificationof PMSM based on EKF and elman neural networkrdquo inProceedings of the IEEE International Conference on Automationand Logistics (ICAL rsquo09) pp 1459ndash1463 Shenyang China 2009
[49] R E Kalman ldquoA new approach to linear filtering and predictionproblemsrdquo Transactions of the ASMEmdashJournal of Basic Engi-neering D vol 82 pp 35ndash45 1960
where119906119902and119906119889are voltages in 119902-axis and119889-axis respectively
119894119902and 119894
119889are currents in 119902-axis and 119889-axis 119877
119904is phase
resistance of stator 119871119889and 119871
119902are inductances in 119889-axis and
119902-axis 120596119903is rotor velocity 120595
119891is flux linkage established by
magnets and 119901 is the differential operatorThe mathematical model of PMSM is discretized to
estimate parameters (119877119904 119871119889 and 119871
119902) The discrete model of
PMSM is as follows
119906119902(119896) = 119877
119904119894119902(119896) + 119901120595
119902(119896) + 120596
119903120595119889(119896)
119906119889(119896) = 119877
119904119894119889(119896) + 119901120595
119889(119896) + 120596
119903120595119902(119896)
(25)
where 120595119889and 120595
119902are as follows
120595119889(119896) = 119871
119889119894119889(119896) + 120595
119891
120595119902(119896) = 119871
119902119894119902(119896)
(26)
The PMSM simulation model is established by MAT-LABSIMULINK and the PMSM running data is obtainedby the model The simulation model is shown in Figure 1
In MATLABSIMULINK the 119877119904 119871119889 and 119871
119902parameters
of PMSM are fixed in simulation We cannot simulate thetime varying characteristic of 119877
119904 119871119889 and 119871
119902 Therefore we
design a motor simulation model according to the require-ment The 119877
119904 119871119889 and 119871
119902can be changed flexibly in the
simulation
42 PMSM Parameters Estimation of Windowed Least SquareAlgorithm When 119877
119904 119871119889 and 119871
119902are fixed value the PMSM
simulation data is obtained by MATLAB The windowedleast square algorithm is used to identify the parametersThe algorithm with different window sizes is used for 119877
119904
119871119889 and 119871
119902identification From the identification result
(Figures 2 3 and 4) we can see that bigger window sizehas a better identification result However the window sizedoes not obviously have an effect on the promotion ofparameter identification precision when 119877
119904 119871119889 and 119871
119902are
fixedDifferent window sizes have big effect on the results of
parameter estimation when the estimated parameters havetime varying characteristicThemotor parameters119877
119904119871119889 and
119871119902are measured at 234∘C 30∘C 40∘C 50∘C 60∘C 70∘C and
80∘C Using piecewise linear method to simulate the timevarying of the three parameters the motor running data isobtained by MATLAB
The windowed least square is used to identify the 119877119904 119871119889
and 119871119902 The identification result is shown in Figures 5 6 and
7 when the 119877119904 119871119889and 119871
119902are changed at the same time From
the Figures 5ndash7 we can see that shorter window size has lowereffect on the promotion of identification precision Howeverthe window size is too big to improve the identificationprecisionThe identification result is better when the windowsize is 300ndash400 In the motor model 120595
Figure 5 Estimation result of 119877119904of WLS with different window
sizes when the parameters have time varying characteristic
method It is proposed in 1960 by Kalman [49] The theoryis applied to practical engineering immediately when it isput forward The Apollo program and C-5 plane navigationsystem design are the most successful application examplesExtended Kalman filter (EKF) is an improved model of theKalman filter which is one of the most widely applied innonlinear system filter
Mathematical Problems in Engineering 7
Discrete system state equation of EKF is
X (119896) = A (119896 minus 1)X (119896 minus 1) + B (119896 minus 1)U (119896 minus 1)
+ C (119896 minus 1) + w (119896 minus 1)
Z (119896) = H (119896 minus 1)X (119896) + k (119896)
where X(119896) is the system state vector U(119896) is the systeminput vector Z(119896) is the system observation vector w(119896) isthe system random noise vector k(119896) is the system randomobservation noise vector w(119896) and k(119896) are noise sequenceswith zero mean and the covariance matrices are Q(119896) andR(119896)
119879119904is the sampling period the discrete linear state space
equation (29) of PMSM is established by discretization andlinearization of the model (30)
Figure 6 Estimation result of 119871119889of WLS with different window
sizes when the parameters have time varying characteristic
Using the simulation model and getting the motor output 119894119889
119894119902 119906119889 119906119902120596119890 the initial values of119875119876 and119877 in EKF algorithm
are
119875 = diag ([01 01 00004 0002 002])
119876 = diag ([30 15 0005 003 003])
119877 = diag ([01 002])
(32)
When119877119904 119871119889 and 119871
119902of motor are fixed the identification
result table of EKF recursive least square and windowed leastsquare algorithm is shown in Table 1
From Table 1 we can see that identification result ofEKF algorithm is as good as the windowed least squarealgorithm when 119877
119904 119871119889 and 119871
119902are fixed The comparison
diagrams of identification result are shown in Figure 8 whenthe parameters are fixed
EKF and recursive least square algorithms cannot achievereasonable result when the parameters have time varyingcharacteristic or have a drastic change However windowedleast square algorithm can achieve good identification resultwhen 119877
119904 119871119889 and 119871
119902have time varying characteristic at the
same time (Figures 5ndash7)
5 Analysis and Conclusion
Through the previous different PMSM parameters identifica-tion experiments we can see the following
(1) When the parameters of PMSM have no time vary-ing characteristic three methods can achieve better
Figure 7 Estimation result of 119871119902of WLS with different window
sizes when the parameters have time varying characteristic
identification result in precision and accuracy Inthe calculation and instantaneity of identificationrecursive least square algorithm has a fatal flaw ofdata saturation so the precision and accuracy of thealgorithm are hard to guarantee It is difficult torealize in embedded system the real-time parametersidentification due to the amount of calculations andsystem resources possession of EKF [28]Thewindowsize of windowed least square algorithm is flexibleso we can choose the collected data according tothe changes of the parameters It will reduce theinfluence of the past time of data to the currentparameter identification guarantee the accuracy andinstantaneity of identification and reduce the systemresources possession at the same time
(2) When the parameters of PMSM have strong timevarying characteristic the EKF and recursive leastsquare algorithms cannot guarantee the precisionand accuracy of identification However windowedleast square algorithm can get better identificationresultTherefore EKF and recursive least square algo-rithms are suitable for fixed parameters estimationor parameters with weak time varying characteristicidentification Windowed least square algorithm canget a good result both for fixed parameters and fortime varying parameters identification
Embedded technology is widely used in the motor driverand controller at present However the embedded chip(MCU DSP ARM etc) has certain restriction in computingspeed and storage space Therefore windowed least square
Mathematical Problems in Engineering 9
Table 1 Estimation result comparison of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
Temperature (∘C) 234 30 40 50 60 70 80
119877119904
Actual data (Ω) 09664 10008 10373 10770 11245 11592 11751EKF
Figure 8 Comparison diagram of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
10 Mathematical Problems in Engineering
Figure 9 The PMSM experiment system
Figure 10 The prototype DSP-based PMSM driver
algorithm is a better choice for PMSM parameters identi-fication of motor driver and controller This paper is thebeginning of work There are a lot of work to do such astransplant the algorithm to practical controller and controlsystem (in Figures 9 and 10) which is designed to control thePMSM in practical application
Acknowledgment
This paper is supported by the Shandong Province Scienceand Technology Development Plan of China (Grant no2011GGE27053)
References
[1] M A Rahman and P Zhou ldquoAnalysis of brushless permanentmagnet synchronous motorsrdquo IEEE Transactions on IndustrialElectronics vol 43 no 2 pp 256ndash267 1996
[2] MOoshimaA Chiba A Rahman andT Fukao ldquoAn improvedcontrol method of buried-type IPM bearingless motors consid-ering magnetic saturation and magnetic pull variationrdquo IEEETransactions on Energy Conversion vol 19 no 3 pp 569ndash5752004
[3] K Liu Z Q Zhu Q Zhang and J Zhang ldquoInfluence ofnonideal voltage measurement on parameter estimation inpermanent-magnet synchronous machinesrdquo IEEE Transactionson Industrial Electronics vol 59 no 6 pp 2438ndash2447 2012
[4] F Caricchi F Crescimbini and O Honorati ldquoLow-cost com-pact permanent magnet machine for adjustable-speed pumpapplicationrdquo IEEETransactions on IndustryApplications vol 34no 1 pp 109ndash116 1998
[5] P Milanfar and J H Lang ldquoMonitoring the thermal conditionof permanent-magnet synchronous motorsrdquo IEEE Transactions
onAerospace and Electronic Systems vol 32 no 4 pp 1421ndash14291996
[6] T Kailath ldquoAn innovations approach to least-squares estima-tionmdashpart I linear filtering in additive white noiserdquo IEEETransactions on Automatic Control vol 13 pp 646ndash655 1968
[7] D G Robertson and J H Lee ldquoA least squares formulation forstate estimationrdquo Journal of Process Control vol 5 no 4 pp291ndash299 1995
[8] J S Gibson G H Lee and C F Wu ldquoLeast-squares estimationof inputoutput models for distributed linear systems in thepresence of noiserdquo Automatica vol 36 no 10 pp 1427ndash14422000
[9] S Tunali and I Batmaz ldquoDealing with the least squares regres-sion assumptions in simulation metamodelingrdquo Computers ampIndustrial Engineering vol 38 no 2 pp 307ndash320 2000
[10] R M Fernandez-Alcala J Navarro-Moreno and J C Ruiz-Molina ldquoLinear least-square estimation algorithms involvingcorrelated signal and noiserdquo IEEE Transactions on Signal Pro-cessing vol 53 no 11 pp 4227ndash4235 2005
[11] V Kratschmer ldquoLeast-squares estimation in linear regressionmodels with vague conceptsrdquo Fuzzy Sets and Systems vol 157no 19 pp 2579ndash2592 2006
[12] M J Garcıa-Ligero A Hermoso-Carazo and J Linares-PerezldquoLeast-squares linear estimation of signals from observationswith Markovian delaysrdquo Journal of Computational and AppliedMathematics vol 236 no 2 pp 234ndash242 2011
[13] S Ma C Quan R Zhu C J Tay L Chen and Z GaoldquoApplication of least-square estimation in white-light scanninginterferometryrdquo Optics and Lasers in Engineering vol 49 no 7pp 1012ndash1018 2011
[14] Q Wang and L Zhang ldquoLeast squares online linear discrimi-nant analysisrdquo Expert Systems with Applications vol 39 no 1pp 1510ndash1517 2012
[15] C J Demeure and L L Scharf ldquoSliding windows and latticealgorithms for computing QR factors in the least squares theoryof linear predictionrdquo IEEE Transactions on Acoustics Speechand Signal Processing vol 38 no 4 pp 721ndash725 1990
[16] K Zhao L Fuyun H Lev-Ari and J G Proakis ldquoSlidingwindow order-recursive least-squares algorithmsrdquo IEEE Trans-actions on Signal Processing vol 42 no 8 pp 1961ndash1972 1994
[17] H Liu and Z He ldquoA sliding-exponential window RLS adaptivefiltering algorithm properties and applicationsrdquo Signal Process-ing vol 45 no 3 pp 357ndash368 1995
[18] K Yoo and H Park ldquoFast residual computation for slidingwindow recursive least squares methodsrdquo Signal Processing vol45 no 1 pp 85ndash95 1995
[19] Y Xia M S Kamel and H Leung ldquoA fast algorithm for ARparameter estimation using a novel noise-constrained least-squares methodrdquo Neural Networks vol 23 no 3 pp 396ndash4052010
[20] A Aknouche E M Al-Eid and A M Hmeid ldquoOffline andonline weighted least squares estimation of nonstationarypower119860119877119862119867 processesrdquo Statistics amp Probability Letters vol 81no 10 pp 1535ndash1540 2011
[21] L Xie H Yang and F Ding ldquoRecursive least squares parameterestimation for non-uniformly sampled systems based on thedata filteringrdquo Mathematical and Computer Modelling vol 54no 1-2 pp 315ndash324 2011
[22] J Oliver R Aravind and K M M Prabhu ldquoImproved leastsquares channel estimation for orthogonal frequency divisionmultiplexingrdquo IET Signal Processing vol 6 no 1 pp 45ndash532012
Mathematical Problems in Engineering 11
[23] T Sadiki M Triki and D T M Slock ldquoWindow optimizationissues in recursive least-squares adaptive filtering and trackingrdquoin Proceedings of the 38th IEEE Annual Asilomar Conferenceon Signals Systems and Computers pp 940ndash944 Pacific GroveCalif USA November 2004
[24] G Welch and G Bishop ldquoAn introduction to the Kalman filterrdquo1997
[25] P J Hargrave ldquoA tutorial introduction to Kalman filteringrdquo inProceedings of the IEE Colloquium on Kalman Filters Introduc-tion Applications and Future Developments pp 11ndash16 1989
[26] M Gautier and P Poignet ldquoExtended Kalman filtering andweighted least squares dynamic identification of robotrdquo ControlEngineering Practice vol 9 no 12 pp 1361ndash1372 2001
[27] H M Al-Hamadi and S A Soliman ldquoKalman filter foridentification of power system fuzzy harmonic componentsrdquoElectric Power Systems Research vol 62 no 3 pp 241ndash248 2002
[28] T Boileau N Leboeuf B Nahid-Mobarakeh and F Meibody-Tabar ldquoOnline identification of PMSM parameters parameteridentifiability and estimator comparative studyrdquo IEEE Transac-tions on Industry Applications vol 47 no 4 pp 1944ndash1957 2011
[29] A Bechouche H Sediki D O Abdeslam and S HaddadldquoIdentification of induction motor at standstill using artificialneural networkrdquo in Proceedings of the 36th Annual Conferenceon IEEE Industrial Electronics Society (IECON rsquo10) pp 2908ndash2913 Glendale Ariz USA 2010
[30] F Alonge F DrsquoIppolito and FM Raimondi ldquoLeast squares andgenetic algorithms for parameter identification of inductionmotorsrdquo Control Engineering Practice vol 9 no 6 pp 647ndash6572001
[31] S Mishra ldquoA hybrid least square-fuzzy bacterial foragingstrategy for harmonic estimationrdquo IEEE Transactions on Evo-lutionary Computation vol 9 no 1 pp 61ndash73 2005
[32] R Liao H Zheng S Grzybowski and L Yang ldquoParticleswarm optimization-least squares support vector regressionbased forecasting model on dissolved gases in oil-filled powertransformersrdquo Electric Power Systems Research vol 81 no 12pp 2074ndash2080 2011
[33] R A Zadeh A Ghosh and G Ledwich ldquoCombination ofKalman filter and least-error square techniques in powersystemrdquo IEEE Transactions on Power Delivery vol 25 no 4 pp2868ndash2880 2010
[34] S Bolognani R Oboe andM Zigliotto ldquoSensorless full-digitalPMSM drive with EKF estimation of speed and rotor positionrdquoIEEE Transactions on Industrial Electronics vol 46 no 1 pp184ndash191 1999
[35] M Haardt ldquoStructured least squares to improve the per-formance of ESPRIT-Type algorithmsrdquo IEEE Transactions onSignal Processing vol 45 no 3 pp 792ndash799 1997
[36] M Ghogho A Swami and A K Nandi ldquoNon-linear leastsquares estimation for harmonics in multiplicative and additivenoiserdquo Signal Processing vol 78 no 1 pp 43ndash60 1999
[37] J Angeby ldquoEstimating signal parameters using the nonlinearinstantaneous least squares approachrdquo IEEE Transactions onSignal Processing vol 48 no 10 pp 2721ndash2732 2000
[38] J F Weng and S H Leung ldquoNonlinear RLS algorithm foramplitude estimation in class a noiserdquo IEE ProceedingsmdashCommunications vol 147 no 2 pp 81ndash86 2000
[39] D Zachariah M Sundin M Jansson and S ChatterjeeldquoAlternating least-squares for low-rank matrix reconstructionrdquoIEEE Signal Processing Letters vol 19 no 4 pp 231ndash234 2012
[40] R Montoliu and F Pla ldquoGeneralized least squares-based para-metric motion estimationrdquo Computer Vision and Image Under-standing vol 113 no 7 pp 790ndash801 2009
[41] Z Yingjie and G Liling ldquoImproved moving least squares algo-rithm for directed projecting onto point cloudsrdquoMeasurementvol 44 no 10 pp 2008ndash2019 2011
[42] S Seongwook J-S Lim S J Baek and K-M Sung ldquoVariableforgetting factor linear least squares algorithm for frequencyselective fading channel estimationrdquo IEEE Transactions onVehicular Technology vol 51 no 3 pp 613ndash616 2002
[43] S MorimotoM Sanada and Y Takeda ldquoMechanical sensorlessdrives of IPMSM with online parameter identificationrdquo IEEETransactions on Industry Applications vol 42 no 5 pp 1241ndash1248 2006
[44] H Sakai and H Nakaoka ldquoFast sliding window QRD-RLSalgorithmrdquo Signal Processing vol 78 no 3 pp 309ndash319 1999
[45] S Reece and S Roberts ldquoAn introduction to Gaussian processesfor the Kalman filter expertrdquo in Proceedings of the 13th Confer-ence on Information Fusion (FUSION rsquo10) pp 1ndash9 2010
[46] A Giorgano F M Hsu and J Wiley ldquoBook review least-square estimationwith applications to digital signal processingrdquoIEE Proceedings FmdashCommunications Radar and Signal Process-ing vol 132 no 7 1985
[47] S Wang R Zhao W Chen G Li and C Liu ldquoParameter iden-tification of PMSMbased on windowed least square algorithmrdquoin Proceedings of the Manufacturing Science and Technology(ICMST rsquo11) pp 5940ndash5944 Singapore 2012
[48] S Wang S Shi C Chen G Yang and Z Qu ldquoIdentificationof PMSM based on EKF and elman neural networkrdquo inProceedings of the IEEE International Conference on Automationand Logistics (ICAL rsquo09) pp 1459ndash1463 Shenyang China 2009
[49] R E Kalman ldquoA new approach to linear filtering and predictionproblemsrdquo Transactions of the ASMEmdashJournal of Basic Engi-neering D vol 82 pp 35ndash45 1960
Figure 5 Estimation result of 119877119904of WLS with different window
sizes when the parameters have time varying characteristic
method It is proposed in 1960 by Kalman [49] The theoryis applied to practical engineering immediately when it isput forward The Apollo program and C-5 plane navigationsystem design are the most successful application examplesExtended Kalman filter (EKF) is an improved model of theKalman filter which is one of the most widely applied innonlinear system filter
Mathematical Problems in Engineering 7
Discrete system state equation of EKF is
X (119896) = A (119896 minus 1)X (119896 minus 1) + B (119896 minus 1)U (119896 minus 1)
+ C (119896 minus 1) + w (119896 minus 1)
Z (119896) = H (119896 minus 1)X (119896) + k (119896)
where X(119896) is the system state vector U(119896) is the systeminput vector Z(119896) is the system observation vector w(119896) isthe system random noise vector k(119896) is the system randomobservation noise vector w(119896) and k(119896) are noise sequenceswith zero mean and the covariance matrices are Q(119896) andR(119896)
119879119904is the sampling period the discrete linear state space
equation (29) of PMSM is established by discretization andlinearization of the model (30)
Figure 6 Estimation result of 119871119889of WLS with different window
sizes when the parameters have time varying characteristic
Using the simulation model and getting the motor output 119894119889
119894119902 119906119889 119906119902120596119890 the initial values of119875119876 and119877 in EKF algorithm
are
119875 = diag ([01 01 00004 0002 002])
119876 = diag ([30 15 0005 003 003])
119877 = diag ([01 002])
(32)
When119877119904 119871119889 and 119871
119902of motor are fixed the identification
result table of EKF recursive least square and windowed leastsquare algorithm is shown in Table 1
From Table 1 we can see that identification result ofEKF algorithm is as good as the windowed least squarealgorithm when 119877
119904 119871119889 and 119871
119902are fixed The comparison
diagrams of identification result are shown in Figure 8 whenthe parameters are fixed
EKF and recursive least square algorithms cannot achievereasonable result when the parameters have time varyingcharacteristic or have a drastic change However windowedleast square algorithm can achieve good identification resultwhen 119877
119904 119871119889 and 119871
119902have time varying characteristic at the
same time (Figures 5ndash7)
5 Analysis and Conclusion
Through the previous different PMSM parameters identifica-tion experiments we can see the following
(1) When the parameters of PMSM have no time vary-ing characteristic three methods can achieve better
Figure 7 Estimation result of 119871119902of WLS with different window
sizes when the parameters have time varying characteristic
identification result in precision and accuracy Inthe calculation and instantaneity of identificationrecursive least square algorithm has a fatal flaw ofdata saturation so the precision and accuracy of thealgorithm are hard to guarantee It is difficult torealize in embedded system the real-time parametersidentification due to the amount of calculations andsystem resources possession of EKF [28]Thewindowsize of windowed least square algorithm is flexibleso we can choose the collected data according tothe changes of the parameters It will reduce theinfluence of the past time of data to the currentparameter identification guarantee the accuracy andinstantaneity of identification and reduce the systemresources possession at the same time
(2) When the parameters of PMSM have strong timevarying characteristic the EKF and recursive leastsquare algorithms cannot guarantee the precisionand accuracy of identification However windowedleast square algorithm can get better identificationresultTherefore EKF and recursive least square algo-rithms are suitable for fixed parameters estimationor parameters with weak time varying characteristicidentification Windowed least square algorithm canget a good result both for fixed parameters and fortime varying parameters identification
Embedded technology is widely used in the motor driverand controller at present However the embedded chip(MCU DSP ARM etc) has certain restriction in computingspeed and storage space Therefore windowed least square
Mathematical Problems in Engineering 9
Table 1 Estimation result comparison of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
Temperature (∘C) 234 30 40 50 60 70 80
119877119904
Actual data (Ω) 09664 10008 10373 10770 11245 11592 11751EKF
Figure 8 Comparison diagram of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
10 Mathematical Problems in Engineering
Figure 9 The PMSM experiment system
Figure 10 The prototype DSP-based PMSM driver
algorithm is a better choice for PMSM parameters identi-fication of motor driver and controller This paper is thebeginning of work There are a lot of work to do such astransplant the algorithm to practical controller and controlsystem (in Figures 9 and 10) which is designed to control thePMSM in practical application
Acknowledgment
This paper is supported by the Shandong Province Scienceand Technology Development Plan of China (Grant no2011GGE27053)
References
[1] M A Rahman and P Zhou ldquoAnalysis of brushless permanentmagnet synchronous motorsrdquo IEEE Transactions on IndustrialElectronics vol 43 no 2 pp 256ndash267 1996
[2] MOoshimaA Chiba A Rahman andT Fukao ldquoAn improvedcontrol method of buried-type IPM bearingless motors consid-ering magnetic saturation and magnetic pull variationrdquo IEEETransactions on Energy Conversion vol 19 no 3 pp 569ndash5752004
[3] K Liu Z Q Zhu Q Zhang and J Zhang ldquoInfluence ofnonideal voltage measurement on parameter estimation inpermanent-magnet synchronous machinesrdquo IEEE Transactionson Industrial Electronics vol 59 no 6 pp 2438ndash2447 2012
[4] F Caricchi F Crescimbini and O Honorati ldquoLow-cost com-pact permanent magnet machine for adjustable-speed pumpapplicationrdquo IEEETransactions on IndustryApplications vol 34no 1 pp 109ndash116 1998
[5] P Milanfar and J H Lang ldquoMonitoring the thermal conditionof permanent-magnet synchronous motorsrdquo IEEE Transactions
onAerospace and Electronic Systems vol 32 no 4 pp 1421ndash14291996
[6] T Kailath ldquoAn innovations approach to least-squares estima-tionmdashpart I linear filtering in additive white noiserdquo IEEETransactions on Automatic Control vol 13 pp 646ndash655 1968
[7] D G Robertson and J H Lee ldquoA least squares formulation forstate estimationrdquo Journal of Process Control vol 5 no 4 pp291ndash299 1995
[8] J S Gibson G H Lee and C F Wu ldquoLeast-squares estimationof inputoutput models for distributed linear systems in thepresence of noiserdquo Automatica vol 36 no 10 pp 1427ndash14422000
[9] S Tunali and I Batmaz ldquoDealing with the least squares regres-sion assumptions in simulation metamodelingrdquo Computers ampIndustrial Engineering vol 38 no 2 pp 307ndash320 2000
[10] R M Fernandez-Alcala J Navarro-Moreno and J C Ruiz-Molina ldquoLinear least-square estimation algorithms involvingcorrelated signal and noiserdquo IEEE Transactions on Signal Pro-cessing vol 53 no 11 pp 4227ndash4235 2005
[11] V Kratschmer ldquoLeast-squares estimation in linear regressionmodels with vague conceptsrdquo Fuzzy Sets and Systems vol 157no 19 pp 2579ndash2592 2006
[12] M J Garcıa-Ligero A Hermoso-Carazo and J Linares-PerezldquoLeast-squares linear estimation of signals from observationswith Markovian delaysrdquo Journal of Computational and AppliedMathematics vol 236 no 2 pp 234ndash242 2011
[13] S Ma C Quan R Zhu C J Tay L Chen and Z GaoldquoApplication of least-square estimation in white-light scanninginterferometryrdquo Optics and Lasers in Engineering vol 49 no 7pp 1012ndash1018 2011
[14] Q Wang and L Zhang ldquoLeast squares online linear discrimi-nant analysisrdquo Expert Systems with Applications vol 39 no 1pp 1510ndash1517 2012
[15] C J Demeure and L L Scharf ldquoSliding windows and latticealgorithms for computing QR factors in the least squares theoryof linear predictionrdquo IEEE Transactions on Acoustics Speechand Signal Processing vol 38 no 4 pp 721ndash725 1990
[16] K Zhao L Fuyun H Lev-Ari and J G Proakis ldquoSlidingwindow order-recursive least-squares algorithmsrdquo IEEE Trans-actions on Signal Processing vol 42 no 8 pp 1961ndash1972 1994
[17] H Liu and Z He ldquoA sliding-exponential window RLS adaptivefiltering algorithm properties and applicationsrdquo Signal Process-ing vol 45 no 3 pp 357ndash368 1995
[18] K Yoo and H Park ldquoFast residual computation for slidingwindow recursive least squares methodsrdquo Signal Processing vol45 no 1 pp 85ndash95 1995
[19] Y Xia M S Kamel and H Leung ldquoA fast algorithm for ARparameter estimation using a novel noise-constrained least-squares methodrdquo Neural Networks vol 23 no 3 pp 396ndash4052010
[20] A Aknouche E M Al-Eid and A M Hmeid ldquoOffline andonline weighted least squares estimation of nonstationarypower119860119877119862119867 processesrdquo Statistics amp Probability Letters vol 81no 10 pp 1535ndash1540 2011
[21] L Xie H Yang and F Ding ldquoRecursive least squares parameterestimation for non-uniformly sampled systems based on thedata filteringrdquo Mathematical and Computer Modelling vol 54no 1-2 pp 315ndash324 2011
[22] J Oliver R Aravind and K M M Prabhu ldquoImproved leastsquares channel estimation for orthogonal frequency divisionmultiplexingrdquo IET Signal Processing vol 6 no 1 pp 45ndash532012
Mathematical Problems in Engineering 11
[23] T Sadiki M Triki and D T M Slock ldquoWindow optimizationissues in recursive least-squares adaptive filtering and trackingrdquoin Proceedings of the 38th IEEE Annual Asilomar Conferenceon Signals Systems and Computers pp 940ndash944 Pacific GroveCalif USA November 2004
[24] G Welch and G Bishop ldquoAn introduction to the Kalman filterrdquo1997
[25] P J Hargrave ldquoA tutorial introduction to Kalman filteringrdquo inProceedings of the IEE Colloquium on Kalman Filters Introduc-tion Applications and Future Developments pp 11ndash16 1989
[26] M Gautier and P Poignet ldquoExtended Kalman filtering andweighted least squares dynamic identification of robotrdquo ControlEngineering Practice vol 9 no 12 pp 1361ndash1372 2001
[27] H M Al-Hamadi and S A Soliman ldquoKalman filter foridentification of power system fuzzy harmonic componentsrdquoElectric Power Systems Research vol 62 no 3 pp 241ndash248 2002
[28] T Boileau N Leboeuf B Nahid-Mobarakeh and F Meibody-Tabar ldquoOnline identification of PMSM parameters parameteridentifiability and estimator comparative studyrdquo IEEE Transac-tions on Industry Applications vol 47 no 4 pp 1944ndash1957 2011
[29] A Bechouche H Sediki D O Abdeslam and S HaddadldquoIdentification of induction motor at standstill using artificialneural networkrdquo in Proceedings of the 36th Annual Conferenceon IEEE Industrial Electronics Society (IECON rsquo10) pp 2908ndash2913 Glendale Ariz USA 2010
[30] F Alonge F DrsquoIppolito and FM Raimondi ldquoLeast squares andgenetic algorithms for parameter identification of inductionmotorsrdquo Control Engineering Practice vol 9 no 6 pp 647ndash6572001
[31] S Mishra ldquoA hybrid least square-fuzzy bacterial foragingstrategy for harmonic estimationrdquo IEEE Transactions on Evo-lutionary Computation vol 9 no 1 pp 61ndash73 2005
[32] R Liao H Zheng S Grzybowski and L Yang ldquoParticleswarm optimization-least squares support vector regressionbased forecasting model on dissolved gases in oil-filled powertransformersrdquo Electric Power Systems Research vol 81 no 12pp 2074ndash2080 2011
[33] R A Zadeh A Ghosh and G Ledwich ldquoCombination ofKalman filter and least-error square techniques in powersystemrdquo IEEE Transactions on Power Delivery vol 25 no 4 pp2868ndash2880 2010
[34] S Bolognani R Oboe andM Zigliotto ldquoSensorless full-digitalPMSM drive with EKF estimation of speed and rotor positionrdquoIEEE Transactions on Industrial Electronics vol 46 no 1 pp184ndash191 1999
[35] M Haardt ldquoStructured least squares to improve the per-formance of ESPRIT-Type algorithmsrdquo IEEE Transactions onSignal Processing vol 45 no 3 pp 792ndash799 1997
[36] M Ghogho A Swami and A K Nandi ldquoNon-linear leastsquares estimation for harmonics in multiplicative and additivenoiserdquo Signal Processing vol 78 no 1 pp 43ndash60 1999
[37] J Angeby ldquoEstimating signal parameters using the nonlinearinstantaneous least squares approachrdquo IEEE Transactions onSignal Processing vol 48 no 10 pp 2721ndash2732 2000
[38] J F Weng and S H Leung ldquoNonlinear RLS algorithm foramplitude estimation in class a noiserdquo IEE ProceedingsmdashCommunications vol 147 no 2 pp 81ndash86 2000
[39] D Zachariah M Sundin M Jansson and S ChatterjeeldquoAlternating least-squares for low-rank matrix reconstructionrdquoIEEE Signal Processing Letters vol 19 no 4 pp 231ndash234 2012
[40] R Montoliu and F Pla ldquoGeneralized least squares-based para-metric motion estimationrdquo Computer Vision and Image Under-standing vol 113 no 7 pp 790ndash801 2009
[41] Z Yingjie and G Liling ldquoImproved moving least squares algo-rithm for directed projecting onto point cloudsrdquoMeasurementvol 44 no 10 pp 2008ndash2019 2011
[42] S Seongwook J-S Lim S J Baek and K-M Sung ldquoVariableforgetting factor linear least squares algorithm for frequencyselective fading channel estimationrdquo IEEE Transactions onVehicular Technology vol 51 no 3 pp 613ndash616 2002
[43] S MorimotoM Sanada and Y Takeda ldquoMechanical sensorlessdrives of IPMSM with online parameter identificationrdquo IEEETransactions on Industry Applications vol 42 no 5 pp 1241ndash1248 2006
[44] H Sakai and H Nakaoka ldquoFast sliding window QRD-RLSalgorithmrdquo Signal Processing vol 78 no 3 pp 309ndash319 1999
[45] S Reece and S Roberts ldquoAn introduction to Gaussian processesfor the Kalman filter expertrdquo in Proceedings of the 13th Confer-ence on Information Fusion (FUSION rsquo10) pp 1ndash9 2010
[46] A Giorgano F M Hsu and J Wiley ldquoBook review least-square estimationwith applications to digital signal processingrdquoIEE Proceedings FmdashCommunications Radar and Signal Process-ing vol 132 no 7 1985
[47] S Wang R Zhao W Chen G Li and C Liu ldquoParameter iden-tification of PMSMbased on windowed least square algorithmrdquoin Proceedings of the Manufacturing Science and Technology(ICMST rsquo11) pp 5940ndash5944 Singapore 2012
[48] S Wang S Shi C Chen G Yang and Z Qu ldquoIdentificationof PMSM based on EKF and elman neural networkrdquo inProceedings of the IEEE International Conference on Automationand Logistics (ICAL rsquo09) pp 1459ndash1463 Shenyang China 2009
[49] R E Kalman ldquoA new approach to linear filtering and predictionproblemsrdquo Transactions of the ASMEmdashJournal of Basic Engi-neering D vol 82 pp 35ndash45 1960
where X(119896) is the system state vector U(119896) is the systeminput vector Z(119896) is the system observation vector w(119896) isthe system random noise vector k(119896) is the system randomobservation noise vector w(119896) and k(119896) are noise sequenceswith zero mean and the covariance matrices are Q(119896) andR(119896)
119879119904is the sampling period the discrete linear state space
equation (29) of PMSM is established by discretization andlinearization of the model (30)
Figure 6 Estimation result of 119871119889of WLS with different window
sizes when the parameters have time varying characteristic
Using the simulation model and getting the motor output 119894119889
119894119902 119906119889 119906119902120596119890 the initial values of119875119876 and119877 in EKF algorithm
are
119875 = diag ([01 01 00004 0002 002])
119876 = diag ([30 15 0005 003 003])
119877 = diag ([01 002])
(32)
When119877119904 119871119889 and 119871
119902of motor are fixed the identification
result table of EKF recursive least square and windowed leastsquare algorithm is shown in Table 1
From Table 1 we can see that identification result ofEKF algorithm is as good as the windowed least squarealgorithm when 119877
119904 119871119889 and 119871
119902are fixed The comparison
diagrams of identification result are shown in Figure 8 whenthe parameters are fixed
EKF and recursive least square algorithms cannot achievereasonable result when the parameters have time varyingcharacteristic or have a drastic change However windowedleast square algorithm can achieve good identification resultwhen 119877
119904 119871119889 and 119871
119902have time varying characteristic at the
same time (Figures 5ndash7)
5 Analysis and Conclusion
Through the previous different PMSM parameters identifica-tion experiments we can see the following
(1) When the parameters of PMSM have no time vary-ing characteristic three methods can achieve better
Figure 7 Estimation result of 119871119902of WLS with different window
sizes when the parameters have time varying characteristic
identification result in precision and accuracy Inthe calculation and instantaneity of identificationrecursive least square algorithm has a fatal flaw ofdata saturation so the precision and accuracy of thealgorithm are hard to guarantee It is difficult torealize in embedded system the real-time parametersidentification due to the amount of calculations andsystem resources possession of EKF [28]Thewindowsize of windowed least square algorithm is flexibleso we can choose the collected data according tothe changes of the parameters It will reduce theinfluence of the past time of data to the currentparameter identification guarantee the accuracy andinstantaneity of identification and reduce the systemresources possession at the same time
(2) When the parameters of PMSM have strong timevarying characteristic the EKF and recursive leastsquare algorithms cannot guarantee the precisionand accuracy of identification However windowedleast square algorithm can get better identificationresultTherefore EKF and recursive least square algo-rithms are suitable for fixed parameters estimationor parameters with weak time varying characteristicidentification Windowed least square algorithm canget a good result both for fixed parameters and fortime varying parameters identification
Embedded technology is widely used in the motor driverand controller at present However the embedded chip(MCU DSP ARM etc) has certain restriction in computingspeed and storage space Therefore windowed least square
Mathematical Problems in Engineering 9
Table 1 Estimation result comparison of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
Temperature (∘C) 234 30 40 50 60 70 80
119877119904
Actual data (Ω) 09664 10008 10373 10770 11245 11592 11751EKF
Figure 8 Comparison diagram of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
10 Mathematical Problems in Engineering
Figure 9 The PMSM experiment system
Figure 10 The prototype DSP-based PMSM driver
algorithm is a better choice for PMSM parameters identi-fication of motor driver and controller This paper is thebeginning of work There are a lot of work to do such astransplant the algorithm to practical controller and controlsystem (in Figures 9 and 10) which is designed to control thePMSM in practical application
Acknowledgment
This paper is supported by the Shandong Province Scienceand Technology Development Plan of China (Grant no2011GGE27053)
References
[1] M A Rahman and P Zhou ldquoAnalysis of brushless permanentmagnet synchronous motorsrdquo IEEE Transactions on IndustrialElectronics vol 43 no 2 pp 256ndash267 1996
[2] MOoshimaA Chiba A Rahman andT Fukao ldquoAn improvedcontrol method of buried-type IPM bearingless motors consid-ering magnetic saturation and magnetic pull variationrdquo IEEETransactions on Energy Conversion vol 19 no 3 pp 569ndash5752004
[3] K Liu Z Q Zhu Q Zhang and J Zhang ldquoInfluence ofnonideal voltage measurement on parameter estimation inpermanent-magnet synchronous machinesrdquo IEEE Transactionson Industrial Electronics vol 59 no 6 pp 2438ndash2447 2012
[4] F Caricchi F Crescimbini and O Honorati ldquoLow-cost com-pact permanent magnet machine for adjustable-speed pumpapplicationrdquo IEEETransactions on IndustryApplications vol 34no 1 pp 109ndash116 1998
[5] P Milanfar and J H Lang ldquoMonitoring the thermal conditionof permanent-magnet synchronous motorsrdquo IEEE Transactions
onAerospace and Electronic Systems vol 32 no 4 pp 1421ndash14291996
[6] T Kailath ldquoAn innovations approach to least-squares estima-tionmdashpart I linear filtering in additive white noiserdquo IEEETransactions on Automatic Control vol 13 pp 646ndash655 1968
[7] D G Robertson and J H Lee ldquoA least squares formulation forstate estimationrdquo Journal of Process Control vol 5 no 4 pp291ndash299 1995
[8] J S Gibson G H Lee and C F Wu ldquoLeast-squares estimationof inputoutput models for distributed linear systems in thepresence of noiserdquo Automatica vol 36 no 10 pp 1427ndash14422000
[9] S Tunali and I Batmaz ldquoDealing with the least squares regres-sion assumptions in simulation metamodelingrdquo Computers ampIndustrial Engineering vol 38 no 2 pp 307ndash320 2000
[10] R M Fernandez-Alcala J Navarro-Moreno and J C Ruiz-Molina ldquoLinear least-square estimation algorithms involvingcorrelated signal and noiserdquo IEEE Transactions on Signal Pro-cessing vol 53 no 11 pp 4227ndash4235 2005
[11] V Kratschmer ldquoLeast-squares estimation in linear regressionmodels with vague conceptsrdquo Fuzzy Sets and Systems vol 157no 19 pp 2579ndash2592 2006
[12] M J Garcıa-Ligero A Hermoso-Carazo and J Linares-PerezldquoLeast-squares linear estimation of signals from observationswith Markovian delaysrdquo Journal of Computational and AppliedMathematics vol 236 no 2 pp 234ndash242 2011
[13] S Ma C Quan R Zhu C J Tay L Chen and Z GaoldquoApplication of least-square estimation in white-light scanninginterferometryrdquo Optics and Lasers in Engineering vol 49 no 7pp 1012ndash1018 2011
[14] Q Wang and L Zhang ldquoLeast squares online linear discrimi-nant analysisrdquo Expert Systems with Applications vol 39 no 1pp 1510ndash1517 2012
[15] C J Demeure and L L Scharf ldquoSliding windows and latticealgorithms for computing QR factors in the least squares theoryof linear predictionrdquo IEEE Transactions on Acoustics Speechand Signal Processing vol 38 no 4 pp 721ndash725 1990
[16] K Zhao L Fuyun H Lev-Ari and J G Proakis ldquoSlidingwindow order-recursive least-squares algorithmsrdquo IEEE Trans-actions on Signal Processing vol 42 no 8 pp 1961ndash1972 1994
[17] H Liu and Z He ldquoA sliding-exponential window RLS adaptivefiltering algorithm properties and applicationsrdquo Signal Process-ing vol 45 no 3 pp 357ndash368 1995
[18] K Yoo and H Park ldquoFast residual computation for slidingwindow recursive least squares methodsrdquo Signal Processing vol45 no 1 pp 85ndash95 1995
[19] Y Xia M S Kamel and H Leung ldquoA fast algorithm for ARparameter estimation using a novel noise-constrained least-squares methodrdquo Neural Networks vol 23 no 3 pp 396ndash4052010
[20] A Aknouche E M Al-Eid and A M Hmeid ldquoOffline andonline weighted least squares estimation of nonstationarypower119860119877119862119867 processesrdquo Statistics amp Probability Letters vol 81no 10 pp 1535ndash1540 2011
[21] L Xie H Yang and F Ding ldquoRecursive least squares parameterestimation for non-uniformly sampled systems based on thedata filteringrdquo Mathematical and Computer Modelling vol 54no 1-2 pp 315ndash324 2011
[22] J Oliver R Aravind and K M M Prabhu ldquoImproved leastsquares channel estimation for orthogonal frequency divisionmultiplexingrdquo IET Signal Processing vol 6 no 1 pp 45ndash532012
Mathematical Problems in Engineering 11
[23] T Sadiki M Triki and D T M Slock ldquoWindow optimizationissues in recursive least-squares adaptive filtering and trackingrdquoin Proceedings of the 38th IEEE Annual Asilomar Conferenceon Signals Systems and Computers pp 940ndash944 Pacific GroveCalif USA November 2004
[24] G Welch and G Bishop ldquoAn introduction to the Kalman filterrdquo1997
[25] P J Hargrave ldquoA tutorial introduction to Kalman filteringrdquo inProceedings of the IEE Colloquium on Kalman Filters Introduc-tion Applications and Future Developments pp 11ndash16 1989
[26] M Gautier and P Poignet ldquoExtended Kalman filtering andweighted least squares dynamic identification of robotrdquo ControlEngineering Practice vol 9 no 12 pp 1361ndash1372 2001
[27] H M Al-Hamadi and S A Soliman ldquoKalman filter foridentification of power system fuzzy harmonic componentsrdquoElectric Power Systems Research vol 62 no 3 pp 241ndash248 2002
[28] T Boileau N Leboeuf B Nahid-Mobarakeh and F Meibody-Tabar ldquoOnline identification of PMSM parameters parameteridentifiability and estimator comparative studyrdquo IEEE Transac-tions on Industry Applications vol 47 no 4 pp 1944ndash1957 2011
[29] A Bechouche H Sediki D O Abdeslam and S HaddadldquoIdentification of induction motor at standstill using artificialneural networkrdquo in Proceedings of the 36th Annual Conferenceon IEEE Industrial Electronics Society (IECON rsquo10) pp 2908ndash2913 Glendale Ariz USA 2010
[30] F Alonge F DrsquoIppolito and FM Raimondi ldquoLeast squares andgenetic algorithms for parameter identification of inductionmotorsrdquo Control Engineering Practice vol 9 no 6 pp 647ndash6572001
[31] S Mishra ldquoA hybrid least square-fuzzy bacterial foragingstrategy for harmonic estimationrdquo IEEE Transactions on Evo-lutionary Computation vol 9 no 1 pp 61ndash73 2005
[32] R Liao H Zheng S Grzybowski and L Yang ldquoParticleswarm optimization-least squares support vector regressionbased forecasting model on dissolved gases in oil-filled powertransformersrdquo Electric Power Systems Research vol 81 no 12pp 2074ndash2080 2011
[33] R A Zadeh A Ghosh and G Ledwich ldquoCombination ofKalman filter and least-error square techniques in powersystemrdquo IEEE Transactions on Power Delivery vol 25 no 4 pp2868ndash2880 2010
[34] S Bolognani R Oboe andM Zigliotto ldquoSensorless full-digitalPMSM drive with EKF estimation of speed and rotor positionrdquoIEEE Transactions on Industrial Electronics vol 46 no 1 pp184ndash191 1999
[35] M Haardt ldquoStructured least squares to improve the per-formance of ESPRIT-Type algorithmsrdquo IEEE Transactions onSignal Processing vol 45 no 3 pp 792ndash799 1997
[36] M Ghogho A Swami and A K Nandi ldquoNon-linear leastsquares estimation for harmonics in multiplicative and additivenoiserdquo Signal Processing vol 78 no 1 pp 43ndash60 1999
[37] J Angeby ldquoEstimating signal parameters using the nonlinearinstantaneous least squares approachrdquo IEEE Transactions onSignal Processing vol 48 no 10 pp 2721ndash2732 2000
[38] J F Weng and S H Leung ldquoNonlinear RLS algorithm foramplitude estimation in class a noiserdquo IEE ProceedingsmdashCommunications vol 147 no 2 pp 81ndash86 2000
[39] D Zachariah M Sundin M Jansson and S ChatterjeeldquoAlternating least-squares for low-rank matrix reconstructionrdquoIEEE Signal Processing Letters vol 19 no 4 pp 231ndash234 2012
[40] R Montoliu and F Pla ldquoGeneralized least squares-based para-metric motion estimationrdquo Computer Vision and Image Under-standing vol 113 no 7 pp 790ndash801 2009
[41] Z Yingjie and G Liling ldquoImproved moving least squares algo-rithm for directed projecting onto point cloudsrdquoMeasurementvol 44 no 10 pp 2008ndash2019 2011
[42] S Seongwook J-S Lim S J Baek and K-M Sung ldquoVariableforgetting factor linear least squares algorithm for frequencyselective fading channel estimationrdquo IEEE Transactions onVehicular Technology vol 51 no 3 pp 613ndash616 2002
[43] S MorimotoM Sanada and Y Takeda ldquoMechanical sensorlessdrives of IPMSM with online parameter identificationrdquo IEEETransactions on Industry Applications vol 42 no 5 pp 1241ndash1248 2006
[44] H Sakai and H Nakaoka ldquoFast sliding window QRD-RLSalgorithmrdquo Signal Processing vol 78 no 3 pp 309ndash319 1999
[45] S Reece and S Roberts ldquoAn introduction to Gaussian processesfor the Kalman filter expertrdquo in Proceedings of the 13th Confer-ence on Information Fusion (FUSION rsquo10) pp 1ndash9 2010
[46] A Giorgano F M Hsu and J Wiley ldquoBook review least-square estimationwith applications to digital signal processingrdquoIEE Proceedings FmdashCommunications Radar and Signal Process-ing vol 132 no 7 1985
[47] S Wang R Zhao W Chen G Li and C Liu ldquoParameter iden-tification of PMSMbased on windowed least square algorithmrdquoin Proceedings of the Manufacturing Science and Technology(ICMST rsquo11) pp 5940ndash5944 Singapore 2012
[48] S Wang S Shi C Chen G Yang and Z Qu ldquoIdentificationof PMSM based on EKF and elman neural networkrdquo inProceedings of the IEEE International Conference on Automationand Logistics (ICAL rsquo09) pp 1459ndash1463 Shenyang China 2009
[49] R E Kalman ldquoA new approach to linear filtering and predictionproblemsrdquo Transactions of the ASMEmdashJournal of Basic Engi-neering D vol 82 pp 35ndash45 1960
Figure 6 Estimation result of 119871119889of WLS with different window
sizes when the parameters have time varying characteristic
Using the simulation model and getting the motor output 119894119889
119894119902 119906119889 119906119902120596119890 the initial values of119875119876 and119877 in EKF algorithm
are
119875 = diag ([01 01 00004 0002 002])
119876 = diag ([30 15 0005 003 003])
119877 = diag ([01 002])
(32)
When119877119904 119871119889 and 119871
119902of motor are fixed the identification
result table of EKF recursive least square and windowed leastsquare algorithm is shown in Table 1
From Table 1 we can see that identification result ofEKF algorithm is as good as the windowed least squarealgorithm when 119877
119904 119871119889 and 119871
119902are fixed The comparison
diagrams of identification result are shown in Figure 8 whenthe parameters are fixed
EKF and recursive least square algorithms cannot achievereasonable result when the parameters have time varyingcharacteristic or have a drastic change However windowedleast square algorithm can achieve good identification resultwhen 119877
119904 119871119889 and 119871
119902have time varying characteristic at the
same time (Figures 5ndash7)
5 Analysis and Conclusion
Through the previous different PMSM parameters identifica-tion experiments we can see the following
(1) When the parameters of PMSM have no time vary-ing characteristic three methods can achieve better
Figure 7 Estimation result of 119871119902of WLS with different window
sizes when the parameters have time varying characteristic
identification result in precision and accuracy Inthe calculation and instantaneity of identificationrecursive least square algorithm has a fatal flaw ofdata saturation so the precision and accuracy of thealgorithm are hard to guarantee It is difficult torealize in embedded system the real-time parametersidentification due to the amount of calculations andsystem resources possession of EKF [28]Thewindowsize of windowed least square algorithm is flexibleso we can choose the collected data according tothe changes of the parameters It will reduce theinfluence of the past time of data to the currentparameter identification guarantee the accuracy andinstantaneity of identification and reduce the systemresources possession at the same time
(2) When the parameters of PMSM have strong timevarying characteristic the EKF and recursive leastsquare algorithms cannot guarantee the precisionand accuracy of identification However windowedleast square algorithm can get better identificationresultTherefore EKF and recursive least square algo-rithms are suitable for fixed parameters estimationor parameters with weak time varying characteristicidentification Windowed least square algorithm canget a good result both for fixed parameters and fortime varying parameters identification
Embedded technology is widely used in the motor driverand controller at present However the embedded chip(MCU DSP ARM etc) has certain restriction in computingspeed and storage space Therefore windowed least square
Mathematical Problems in Engineering 9
Table 1 Estimation result comparison of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
Temperature (∘C) 234 30 40 50 60 70 80
119877119904
Actual data (Ω) 09664 10008 10373 10770 11245 11592 11751EKF
Figure 8 Comparison diagram of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
10 Mathematical Problems in Engineering
Figure 9 The PMSM experiment system
Figure 10 The prototype DSP-based PMSM driver
algorithm is a better choice for PMSM parameters identi-fication of motor driver and controller This paper is thebeginning of work There are a lot of work to do such astransplant the algorithm to practical controller and controlsystem (in Figures 9 and 10) which is designed to control thePMSM in practical application
Acknowledgment
This paper is supported by the Shandong Province Scienceand Technology Development Plan of China (Grant no2011GGE27053)
References
[1] M A Rahman and P Zhou ldquoAnalysis of brushless permanentmagnet synchronous motorsrdquo IEEE Transactions on IndustrialElectronics vol 43 no 2 pp 256ndash267 1996
[2] MOoshimaA Chiba A Rahman andT Fukao ldquoAn improvedcontrol method of buried-type IPM bearingless motors consid-ering magnetic saturation and magnetic pull variationrdquo IEEETransactions on Energy Conversion vol 19 no 3 pp 569ndash5752004
[3] K Liu Z Q Zhu Q Zhang and J Zhang ldquoInfluence ofnonideal voltage measurement on parameter estimation inpermanent-magnet synchronous machinesrdquo IEEE Transactionson Industrial Electronics vol 59 no 6 pp 2438ndash2447 2012
[4] F Caricchi F Crescimbini and O Honorati ldquoLow-cost com-pact permanent magnet machine for adjustable-speed pumpapplicationrdquo IEEETransactions on IndustryApplications vol 34no 1 pp 109ndash116 1998
[5] P Milanfar and J H Lang ldquoMonitoring the thermal conditionof permanent-magnet synchronous motorsrdquo IEEE Transactions
onAerospace and Electronic Systems vol 32 no 4 pp 1421ndash14291996
[6] T Kailath ldquoAn innovations approach to least-squares estima-tionmdashpart I linear filtering in additive white noiserdquo IEEETransactions on Automatic Control vol 13 pp 646ndash655 1968
[7] D G Robertson and J H Lee ldquoA least squares formulation forstate estimationrdquo Journal of Process Control vol 5 no 4 pp291ndash299 1995
[8] J S Gibson G H Lee and C F Wu ldquoLeast-squares estimationof inputoutput models for distributed linear systems in thepresence of noiserdquo Automatica vol 36 no 10 pp 1427ndash14422000
[9] S Tunali and I Batmaz ldquoDealing with the least squares regres-sion assumptions in simulation metamodelingrdquo Computers ampIndustrial Engineering vol 38 no 2 pp 307ndash320 2000
[10] R M Fernandez-Alcala J Navarro-Moreno and J C Ruiz-Molina ldquoLinear least-square estimation algorithms involvingcorrelated signal and noiserdquo IEEE Transactions on Signal Pro-cessing vol 53 no 11 pp 4227ndash4235 2005
[11] V Kratschmer ldquoLeast-squares estimation in linear regressionmodels with vague conceptsrdquo Fuzzy Sets and Systems vol 157no 19 pp 2579ndash2592 2006
[12] M J Garcıa-Ligero A Hermoso-Carazo and J Linares-PerezldquoLeast-squares linear estimation of signals from observationswith Markovian delaysrdquo Journal of Computational and AppliedMathematics vol 236 no 2 pp 234ndash242 2011
[13] S Ma C Quan R Zhu C J Tay L Chen and Z GaoldquoApplication of least-square estimation in white-light scanninginterferometryrdquo Optics and Lasers in Engineering vol 49 no 7pp 1012ndash1018 2011
[14] Q Wang and L Zhang ldquoLeast squares online linear discrimi-nant analysisrdquo Expert Systems with Applications vol 39 no 1pp 1510ndash1517 2012
[15] C J Demeure and L L Scharf ldquoSliding windows and latticealgorithms for computing QR factors in the least squares theoryof linear predictionrdquo IEEE Transactions on Acoustics Speechand Signal Processing vol 38 no 4 pp 721ndash725 1990
[16] K Zhao L Fuyun H Lev-Ari and J G Proakis ldquoSlidingwindow order-recursive least-squares algorithmsrdquo IEEE Trans-actions on Signal Processing vol 42 no 8 pp 1961ndash1972 1994
[17] H Liu and Z He ldquoA sliding-exponential window RLS adaptivefiltering algorithm properties and applicationsrdquo Signal Process-ing vol 45 no 3 pp 357ndash368 1995
[18] K Yoo and H Park ldquoFast residual computation for slidingwindow recursive least squares methodsrdquo Signal Processing vol45 no 1 pp 85ndash95 1995
[19] Y Xia M S Kamel and H Leung ldquoA fast algorithm for ARparameter estimation using a novel noise-constrained least-squares methodrdquo Neural Networks vol 23 no 3 pp 396ndash4052010
[20] A Aknouche E M Al-Eid and A M Hmeid ldquoOffline andonline weighted least squares estimation of nonstationarypower119860119877119862119867 processesrdquo Statistics amp Probability Letters vol 81no 10 pp 1535ndash1540 2011
[21] L Xie H Yang and F Ding ldquoRecursive least squares parameterestimation for non-uniformly sampled systems based on thedata filteringrdquo Mathematical and Computer Modelling vol 54no 1-2 pp 315ndash324 2011
[22] J Oliver R Aravind and K M M Prabhu ldquoImproved leastsquares channel estimation for orthogonal frequency divisionmultiplexingrdquo IET Signal Processing vol 6 no 1 pp 45ndash532012
Mathematical Problems in Engineering 11
[23] T Sadiki M Triki and D T M Slock ldquoWindow optimizationissues in recursive least-squares adaptive filtering and trackingrdquoin Proceedings of the 38th IEEE Annual Asilomar Conferenceon Signals Systems and Computers pp 940ndash944 Pacific GroveCalif USA November 2004
[24] G Welch and G Bishop ldquoAn introduction to the Kalman filterrdquo1997
[25] P J Hargrave ldquoA tutorial introduction to Kalman filteringrdquo inProceedings of the IEE Colloquium on Kalman Filters Introduc-tion Applications and Future Developments pp 11ndash16 1989
[26] M Gautier and P Poignet ldquoExtended Kalman filtering andweighted least squares dynamic identification of robotrdquo ControlEngineering Practice vol 9 no 12 pp 1361ndash1372 2001
[27] H M Al-Hamadi and S A Soliman ldquoKalman filter foridentification of power system fuzzy harmonic componentsrdquoElectric Power Systems Research vol 62 no 3 pp 241ndash248 2002
[28] T Boileau N Leboeuf B Nahid-Mobarakeh and F Meibody-Tabar ldquoOnline identification of PMSM parameters parameteridentifiability and estimator comparative studyrdquo IEEE Transac-tions on Industry Applications vol 47 no 4 pp 1944ndash1957 2011
[29] A Bechouche H Sediki D O Abdeslam and S HaddadldquoIdentification of induction motor at standstill using artificialneural networkrdquo in Proceedings of the 36th Annual Conferenceon IEEE Industrial Electronics Society (IECON rsquo10) pp 2908ndash2913 Glendale Ariz USA 2010
[30] F Alonge F DrsquoIppolito and FM Raimondi ldquoLeast squares andgenetic algorithms for parameter identification of inductionmotorsrdquo Control Engineering Practice vol 9 no 6 pp 647ndash6572001
[31] S Mishra ldquoA hybrid least square-fuzzy bacterial foragingstrategy for harmonic estimationrdquo IEEE Transactions on Evo-lutionary Computation vol 9 no 1 pp 61ndash73 2005
[32] R Liao H Zheng S Grzybowski and L Yang ldquoParticleswarm optimization-least squares support vector regressionbased forecasting model on dissolved gases in oil-filled powertransformersrdquo Electric Power Systems Research vol 81 no 12pp 2074ndash2080 2011
[33] R A Zadeh A Ghosh and G Ledwich ldquoCombination ofKalman filter and least-error square techniques in powersystemrdquo IEEE Transactions on Power Delivery vol 25 no 4 pp2868ndash2880 2010
[34] S Bolognani R Oboe andM Zigliotto ldquoSensorless full-digitalPMSM drive with EKF estimation of speed and rotor positionrdquoIEEE Transactions on Industrial Electronics vol 46 no 1 pp184ndash191 1999
[35] M Haardt ldquoStructured least squares to improve the per-formance of ESPRIT-Type algorithmsrdquo IEEE Transactions onSignal Processing vol 45 no 3 pp 792ndash799 1997
[36] M Ghogho A Swami and A K Nandi ldquoNon-linear leastsquares estimation for harmonics in multiplicative and additivenoiserdquo Signal Processing vol 78 no 1 pp 43ndash60 1999
[37] J Angeby ldquoEstimating signal parameters using the nonlinearinstantaneous least squares approachrdquo IEEE Transactions onSignal Processing vol 48 no 10 pp 2721ndash2732 2000
[38] J F Weng and S H Leung ldquoNonlinear RLS algorithm foramplitude estimation in class a noiserdquo IEE ProceedingsmdashCommunications vol 147 no 2 pp 81ndash86 2000
[39] D Zachariah M Sundin M Jansson and S ChatterjeeldquoAlternating least-squares for low-rank matrix reconstructionrdquoIEEE Signal Processing Letters vol 19 no 4 pp 231ndash234 2012
[40] R Montoliu and F Pla ldquoGeneralized least squares-based para-metric motion estimationrdquo Computer Vision and Image Under-standing vol 113 no 7 pp 790ndash801 2009
[41] Z Yingjie and G Liling ldquoImproved moving least squares algo-rithm for directed projecting onto point cloudsrdquoMeasurementvol 44 no 10 pp 2008ndash2019 2011
[42] S Seongwook J-S Lim S J Baek and K-M Sung ldquoVariableforgetting factor linear least squares algorithm for frequencyselective fading channel estimationrdquo IEEE Transactions onVehicular Technology vol 51 no 3 pp 613ndash616 2002
[43] S MorimotoM Sanada and Y Takeda ldquoMechanical sensorlessdrives of IPMSM with online parameter identificationrdquo IEEETransactions on Industry Applications vol 42 no 5 pp 1241ndash1248 2006
[44] H Sakai and H Nakaoka ldquoFast sliding window QRD-RLSalgorithmrdquo Signal Processing vol 78 no 3 pp 309ndash319 1999
[45] S Reece and S Roberts ldquoAn introduction to Gaussian processesfor the Kalman filter expertrdquo in Proceedings of the 13th Confer-ence on Information Fusion (FUSION rsquo10) pp 1ndash9 2010
[46] A Giorgano F M Hsu and J Wiley ldquoBook review least-square estimationwith applications to digital signal processingrdquoIEE Proceedings FmdashCommunications Radar and Signal Process-ing vol 132 no 7 1985
[47] S Wang R Zhao W Chen G Li and C Liu ldquoParameter iden-tification of PMSMbased on windowed least square algorithmrdquoin Proceedings of the Manufacturing Science and Technology(ICMST rsquo11) pp 5940ndash5944 Singapore 2012
[48] S Wang S Shi C Chen G Yang and Z Qu ldquoIdentificationof PMSM based on EKF and elman neural networkrdquo inProceedings of the IEEE International Conference on Automationand Logistics (ICAL rsquo09) pp 1459ndash1463 Shenyang China 2009
[49] R E Kalman ldquoA new approach to linear filtering and predictionproblemsrdquo Transactions of the ASMEmdashJournal of Basic Engi-neering D vol 82 pp 35ndash45 1960
Figure 8 Comparison diagram of EKF RLS and WLS when 119877119904 119871119889 and 119871
119902are fixed value
10 Mathematical Problems in Engineering
Figure 9 The PMSM experiment system
Figure 10 The prototype DSP-based PMSM driver
algorithm is a better choice for PMSM parameters identi-fication of motor driver and controller This paper is thebeginning of work There are a lot of work to do such astransplant the algorithm to practical controller and controlsystem (in Figures 9 and 10) which is designed to control thePMSM in practical application
Acknowledgment
This paper is supported by the Shandong Province Scienceand Technology Development Plan of China (Grant no2011GGE27053)
References
[1] M A Rahman and P Zhou ldquoAnalysis of brushless permanentmagnet synchronous motorsrdquo IEEE Transactions on IndustrialElectronics vol 43 no 2 pp 256ndash267 1996
[2] MOoshimaA Chiba A Rahman andT Fukao ldquoAn improvedcontrol method of buried-type IPM bearingless motors consid-ering magnetic saturation and magnetic pull variationrdquo IEEETransactions on Energy Conversion vol 19 no 3 pp 569ndash5752004
[3] K Liu Z Q Zhu Q Zhang and J Zhang ldquoInfluence ofnonideal voltage measurement on parameter estimation inpermanent-magnet synchronous machinesrdquo IEEE Transactionson Industrial Electronics vol 59 no 6 pp 2438ndash2447 2012
[4] F Caricchi F Crescimbini and O Honorati ldquoLow-cost com-pact permanent magnet machine for adjustable-speed pumpapplicationrdquo IEEETransactions on IndustryApplications vol 34no 1 pp 109ndash116 1998
[5] P Milanfar and J H Lang ldquoMonitoring the thermal conditionof permanent-magnet synchronous motorsrdquo IEEE Transactions
onAerospace and Electronic Systems vol 32 no 4 pp 1421ndash14291996
[6] T Kailath ldquoAn innovations approach to least-squares estima-tionmdashpart I linear filtering in additive white noiserdquo IEEETransactions on Automatic Control vol 13 pp 646ndash655 1968
[7] D G Robertson and J H Lee ldquoA least squares formulation forstate estimationrdquo Journal of Process Control vol 5 no 4 pp291ndash299 1995
[8] J S Gibson G H Lee and C F Wu ldquoLeast-squares estimationof inputoutput models for distributed linear systems in thepresence of noiserdquo Automatica vol 36 no 10 pp 1427ndash14422000
[9] S Tunali and I Batmaz ldquoDealing with the least squares regres-sion assumptions in simulation metamodelingrdquo Computers ampIndustrial Engineering vol 38 no 2 pp 307ndash320 2000
[10] R M Fernandez-Alcala J Navarro-Moreno and J C Ruiz-Molina ldquoLinear least-square estimation algorithms involvingcorrelated signal and noiserdquo IEEE Transactions on Signal Pro-cessing vol 53 no 11 pp 4227ndash4235 2005
[11] V Kratschmer ldquoLeast-squares estimation in linear regressionmodels with vague conceptsrdquo Fuzzy Sets and Systems vol 157no 19 pp 2579ndash2592 2006
[12] M J Garcıa-Ligero A Hermoso-Carazo and J Linares-PerezldquoLeast-squares linear estimation of signals from observationswith Markovian delaysrdquo Journal of Computational and AppliedMathematics vol 236 no 2 pp 234ndash242 2011
[13] S Ma C Quan R Zhu C J Tay L Chen and Z GaoldquoApplication of least-square estimation in white-light scanninginterferometryrdquo Optics and Lasers in Engineering vol 49 no 7pp 1012ndash1018 2011
[14] Q Wang and L Zhang ldquoLeast squares online linear discrimi-nant analysisrdquo Expert Systems with Applications vol 39 no 1pp 1510ndash1517 2012
[15] C J Demeure and L L Scharf ldquoSliding windows and latticealgorithms for computing QR factors in the least squares theoryof linear predictionrdquo IEEE Transactions on Acoustics Speechand Signal Processing vol 38 no 4 pp 721ndash725 1990
[16] K Zhao L Fuyun H Lev-Ari and J G Proakis ldquoSlidingwindow order-recursive least-squares algorithmsrdquo IEEE Trans-actions on Signal Processing vol 42 no 8 pp 1961ndash1972 1994
[17] H Liu and Z He ldquoA sliding-exponential window RLS adaptivefiltering algorithm properties and applicationsrdquo Signal Process-ing vol 45 no 3 pp 357ndash368 1995
[18] K Yoo and H Park ldquoFast residual computation for slidingwindow recursive least squares methodsrdquo Signal Processing vol45 no 1 pp 85ndash95 1995
[19] Y Xia M S Kamel and H Leung ldquoA fast algorithm for ARparameter estimation using a novel noise-constrained least-squares methodrdquo Neural Networks vol 23 no 3 pp 396ndash4052010
[20] A Aknouche E M Al-Eid and A M Hmeid ldquoOffline andonline weighted least squares estimation of nonstationarypower119860119877119862119867 processesrdquo Statistics amp Probability Letters vol 81no 10 pp 1535ndash1540 2011
[21] L Xie H Yang and F Ding ldquoRecursive least squares parameterestimation for non-uniformly sampled systems based on thedata filteringrdquo Mathematical and Computer Modelling vol 54no 1-2 pp 315ndash324 2011
[22] J Oliver R Aravind and K M M Prabhu ldquoImproved leastsquares channel estimation for orthogonal frequency divisionmultiplexingrdquo IET Signal Processing vol 6 no 1 pp 45ndash532012
Mathematical Problems in Engineering 11
[23] T Sadiki M Triki and D T M Slock ldquoWindow optimizationissues in recursive least-squares adaptive filtering and trackingrdquoin Proceedings of the 38th IEEE Annual Asilomar Conferenceon Signals Systems and Computers pp 940ndash944 Pacific GroveCalif USA November 2004
[24] G Welch and G Bishop ldquoAn introduction to the Kalman filterrdquo1997
[25] P J Hargrave ldquoA tutorial introduction to Kalman filteringrdquo inProceedings of the IEE Colloquium on Kalman Filters Introduc-tion Applications and Future Developments pp 11ndash16 1989
[26] M Gautier and P Poignet ldquoExtended Kalman filtering andweighted least squares dynamic identification of robotrdquo ControlEngineering Practice vol 9 no 12 pp 1361ndash1372 2001
[27] H M Al-Hamadi and S A Soliman ldquoKalman filter foridentification of power system fuzzy harmonic componentsrdquoElectric Power Systems Research vol 62 no 3 pp 241ndash248 2002
[28] T Boileau N Leboeuf B Nahid-Mobarakeh and F Meibody-Tabar ldquoOnline identification of PMSM parameters parameteridentifiability and estimator comparative studyrdquo IEEE Transac-tions on Industry Applications vol 47 no 4 pp 1944ndash1957 2011
[29] A Bechouche H Sediki D O Abdeslam and S HaddadldquoIdentification of induction motor at standstill using artificialneural networkrdquo in Proceedings of the 36th Annual Conferenceon IEEE Industrial Electronics Society (IECON rsquo10) pp 2908ndash2913 Glendale Ariz USA 2010
[30] F Alonge F DrsquoIppolito and FM Raimondi ldquoLeast squares andgenetic algorithms for parameter identification of inductionmotorsrdquo Control Engineering Practice vol 9 no 6 pp 647ndash6572001
[31] S Mishra ldquoA hybrid least square-fuzzy bacterial foragingstrategy for harmonic estimationrdquo IEEE Transactions on Evo-lutionary Computation vol 9 no 1 pp 61ndash73 2005
[32] R Liao H Zheng S Grzybowski and L Yang ldquoParticleswarm optimization-least squares support vector regressionbased forecasting model on dissolved gases in oil-filled powertransformersrdquo Electric Power Systems Research vol 81 no 12pp 2074ndash2080 2011
[33] R A Zadeh A Ghosh and G Ledwich ldquoCombination ofKalman filter and least-error square techniques in powersystemrdquo IEEE Transactions on Power Delivery vol 25 no 4 pp2868ndash2880 2010
[34] S Bolognani R Oboe andM Zigliotto ldquoSensorless full-digitalPMSM drive with EKF estimation of speed and rotor positionrdquoIEEE Transactions on Industrial Electronics vol 46 no 1 pp184ndash191 1999
[35] M Haardt ldquoStructured least squares to improve the per-formance of ESPRIT-Type algorithmsrdquo IEEE Transactions onSignal Processing vol 45 no 3 pp 792ndash799 1997
[36] M Ghogho A Swami and A K Nandi ldquoNon-linear leastsquares estimation for harmonics in multiplicative and additivenoiserdquo Signal Processing vol 78 no 1 pp 43ndash60 1999
[37] J Angeby ldquoEstimating signal parameters using the nonlinearinstantaneous least squares approachrdquo IEEE Transactions onSignal Processing vol 48 no 10 pp 2721ndash2732 2000
[38] J F Weng and S H Leung ldquoNonlinear RLS algorithm foramplitude estimation in class a noiserdquo IEE ProceedingsmdashCommunications vol 147 no 2 pp 81ndash86 2000
[39] D Zachariah M Sundin M Jansson and S ChatterjeeldquoAlternating least-squares for low-rank matrix reconstructionrdquoIEEE Signal Processing Letters vol 19 no 4 pp 231ndash234 2012
[40] R Montoliu and F Pla ldquoGeneralized least squares-based para-metric motion estimationrdquo Computer Vision and Image Under-standing vol 113 no 7 pp 790ndash801 2009
[41] Z Yingjie and G Liling ldquoImproved moving least squares algo-rithm for directed projecting onto point cloudsrdquoMeasurementvol 44 no 10 pp 2008ndash2019 2011
[42] S Seongwook J-S Lim S J Baek and K-M Sung ldquoVariableforgetting factor linear least squares algorithm for frequencyselective fading channel estimationrdquo IEEE Transactions onVehicular Technology vol 51 no 3 pp 613ndash616 2002
[43] S MorimotoM Sanada and Y Takeda ldquoMechanical sensorlessdrives of IPMSM with online parameter identificationrdquo IEEETransactions on Industry Applications vol 42 no 5 pp 1241ndash1248 2006
[44] H Sakai and H Nakaoka ldquoFast sliding window QRD-RLSalgorithmrdquo Signal Processing vol 78 no 3 pp 309ndash319 1999
[45] S Reece and S Roberts ldquoAn introduction to Gaussian processesfor the Kalman filter expertrdquo in Proceedings of the 13th Confer-ence on Information Fusion (FUSION rsquo10) pp 1ndash9 2010
[46] A Giorgano F M Hsu and J Wiley ldquoBook review least-square estimationwith applications to digital signal processingrdquoIEE Proceedings FmdashCommunications Radar and Signal Process-ing vol 132 no 7 1985
[47] S Wang R Zhao W Chen G Li and C Liu ldquoParameter iden-tification of PMSMbased on windowed least square algorithmrdquoin Proceedings of the Manufacturing Science and Technology(ICMST rsquo11) pp 5940ndash5944 Singapore 2012
[48] S Wang S Shi C Chen G Yang and Z Qu ldquoIdentificationof PMSM based on EKF and elman neural networkrdquo inProceedings of the IEEE International Conference on Automationand Logistics (ICAL rsquo09) pp 1459ndash1463 Shenyang China 2009
[49] R E Kalman ldquoA new approach to linear filtering and predictionproblemsrdquo Transactions of the ASMEmdashJournal of Basic Engi-neering D vol 82 pp 35ndash45 1960
algorithm is a better choice for PMSM parameters identi-fication of motor driver and controller This paper is thebeginning of work There are a lot of work to do such astransplant the algorithm to practical controller and controlsystem (in Figures 9 and 10) which is designed to control thePMSM in practical application
Acknowledgment
This paper is supported by the Shandong Province Scienceand Technology Development Plan of China (Grant no2011GGE27053)
References
[1] M A Rahman and P Zhou ldquoAnalysis of brushless permanentmagnet synchronous motorsrdquo IEEE Transactions on IndustrialElectronics vol 43 no 2 pp 256ndash267 1996
[2] MOoshimaA Chiba A Rahman andT Fukao ldquoAn improvedcontrol method of buried-type IPM bearingless motors consid-ering magnetic saturation and magnetic pull variationrdquo IEEETransactions on Energy Conversion vol 19 no 3 pp 569ndash5752004
[3] K Liu Z Q Zhu Q Zhang and J Zhang ldquoInfluence ofnonideal voltage measurement on parameter estimation inpermanent-magnet synchronous machinesrdquo IEEE Transactionson Industrial Electronics vol 59 no 6 pp 2438ndash2447 2012
[4] F Caricchi F Crescimbini and O Honorati ldquoLow-cost com-pact permanent magnet machine for adjustable-speed pumpapplicationrdquo IEEETransactions on IndustryApplications vol 34no 1 pp 109ndash116 1998
[5] P Milanfar and J H Lang ldquoMonitoring the thermal conditionof permanent-magnet synchronous motorsrdquo IEEE Transactions
onAerospace and Electronic Systems vol 32 no 4 pp 1421ndash14291996
[6] T Kailath ldquoAn innovations approach to least-squares estima-tionmdashpart I linear filtering in additive white noiserdquo IEEETransactions on Automatic Control vol 13 pp 646ndash655 1968
[7] D G Robertson and J H Lee ldquoA least squares formulation forstate estimationrdquo Journal of Process Control vol 5 no 4 pp291ndash299 1995
[8] J S Gibson G H Lee and C F Wu ldquoLeast-squares estimationof inputoutput models for distributed linear systems in thepresence of noiserdquo Automatica vol 36 no 10 pp 1427ndash14422000
[9] S Tunali and I Batmaz ldquoDealing with the least squares regres-sion assumptions in simulation metamodelingrdquo Computers ampIndustrial Engineering vol 38 no 2 pp 307ndash320 2000
[10] R M Fernandez-Alcala J Navarro-Moreno and J C Ruiz-Molina ldquoLinear least-square estimation algorithms involvingcorrelated signal and noiserdquo IEEE Transactions on Signal Pro-cessing vol 53 no 11 pp 4227ndash4235 2005
[11] V Kratschmer ldquoLeast-squares estimation in linear regressionmodels with vague conceptsrdquo Fuzzy Sets and Systems vol 157no 19 pp 2579ndash2592 2006
[12] M J Garcıa-Ligero A Hermoso-Carazo and J Linares-PerezldquoLeast-squares linear estimation of signals from observationswith Markovian delaysrdquo Journal of Computational and AppliedMathematics vol 236 no 2 pp 234ndash242 2011
[13] S Ma C Quan R Zhu C J Tay L Chen and Z GaoldquoApplication of least-square estimation in white-light scanninginterferometryrdquo Optics and Lasers in Engineering vol 49 no 7pp 1012ndash1018 2011
[14] Q Wang and L Zhang ldquoLeast squares online linear discrimi-nant analysisrdquo Expert Systems with Applications vol 39 no 1pp 1510ndash1517 2012
[15] C J Demeure and L L Scharf ldquoSliding windows and latticealgorithms for computing QR factors in the least squares theoryof linear predictionrdquo IEEE Transactions on Acoustics Speechand Signal Processing vol 38 no 4 pp 721ndash725 1990
[16] K Zhao L Fuyun H Lev-Ari and J G Proakis ldquoSlidingwindow order-recursive least-squares algorithmsrdquo IEEE Trans-actions on Signal Processing vol 42 no 8 pp 1961ndash1972 1994
[17] H Liu and Z He ldquoA sliding-exponential window RLS adaptivefiltering algorithm properties and applicationsrdquo Signal Process-ing vol 45 no 3 pp 357ndash368 1995
[18] K Yoo and H Park ldquoFast residual computation for slidingwindow recursive least squares methodsrdquo Signal Processing vol45 no 1 pp 85ndash95 1995
[19] Y Xia M S Kamel and H Leung ldquoA fast algorithm for ARparameter estimation using a novel noise-constrained least-squares methodrdquo Neural Networks vol 23 no 3 pp 396ndash4052010
[20] A Aknouche E M Al-Eid and A M Hmeid ldquoOffline andonline weighted least squares estimation of nonstationarypower119860119877119862119867 processesrdquo Statistics amp Probability Letters vol 81no 10 pp 1535ndash1540 2011
[21] L Xie H Yang and F Ding ldquoRecursive least squares parameterestimation for non-uniformly sampled systems based on thedata filteringrdquo Mathematical and Computer Modelling vol 54no 1-2 pp 315ndash324 2011
[22] J Oliver R Aravind and K M M Prabhu ldquoImproved leastsquares channel estimation for orthogonal frequency divisionmultiplexingrdquo IET Signal Processing vol 6 no 1 pp 45ndash532012
Mathematical Problems in Engineering 11
[23] T Sadiki M Triki and D T M Slock ldquoWindow optimizationissues in recursive least-squares adaptive filtering and trackingrdquoin Proceedings of the 38th IEEE Annual Asilomar Conferenceon Signals Systems and Computers pp 940ndash944 Pacific GroveCalif USA November 2004
[24] G Welch and G Bishop ldquoAn introduction to the Kalman filterrdquo1997
[25] P J Hargrave ldquoA tutorial introduction to Kalman filteringrdquo inProceedings of the IEE Colloquium on Kalman Filters Introduc-tion Applications and Future Developments pp 11ndash16 1989
[26] M Gautier and P Poignet ldquoExtended Kalman filtering andweighted least squares dynamic identification of robotrdquo ControlEngineering Practice vol 9 no 12 pp 1361ndash1372 2001
[27] H M Al-Hamadi and S A Soliman ldquoKalman filter foridentification of power system fuzzy harmonic componentsrdquoElectric Power Systems Research vol 62 no 3 pp 241ndash248 2002
[28] T Boileau N Leboeuf B Nahid-Mobarakeh and F Meibody-Tabar ldquoOnline identification of PMSM parameters parameteridentifiability and estimator comparative studyrdquo IEEE Transac-tions on Industry Applications vol 47 no 4 pp 1944ndash1957 2011
[29] A Bechouche H Sediki D O Abdeslam and S HaddadldquoIdentification of induction motor at standstill using artificialneural networkrdquo in Proceedings of the 36th Annual Conferenceon IEEE Industrial Electronics Society (IECON rsquo10) pp 2908ndash2913 Glendale Ariz USA 2010
[30] F Alonge F DrsquoIppolito and FM Raimondi ldquoLeast squares andgenetic algorithms for parameter identification of inductionmotorsrdquo Control Engineering Practice vol 9 no 6 pp 647ndash6572001
[31] S Mishra ldquoA hybrid least square-fuzzy bacterial foragingstrategy for harmonic estimationrdquo IEEE Transactions on Evo-lutionary Computation vol 9 no 1 pp 61ndash73 2005
[32] R Liao H Zheng S Grzybowski and L Yang ldquoParticleswarm optimization-least squares support vector regressionbased forecasting model on dissolved gases in oil-filled powertransformersrdquo Electric Power Systems Research vol 81 no 12pp 2074ndash2080 2011
[33] R A Zadeh A Ghosh and G Ledwich ldquoCombination ofKalman filter and least-error square techniques in powersystemrdquo IEEE Transactions on Power Delivery vol 25 no 4 pp2868ndash2880 2010
[34] S Bolognani R Oboe andM Zigliotto ldquoSensorless full-digitalPMSM drive with EKF estimation of speed and rotor positionrdquoIEEE Transactions on Industrial Electronics vol 46 no 1 pp184ndash191 1999
[35] M Haardt ldquoStructured least squares to improve the per-formance of ESPRIT-Type algorithmsrdquo IEEE Transactions onSignal Processing vol 45 no 3 pp 792ndash799 1997
[36] M Ghogho A Swami and A K Nandi ldquoNon-linear leastsquares estimation for harmonics in multiplicative and additivenoiserdquo Signal Processing vol 78 no 1 pp 43ndash60 1999
[37] J Angeby ldquoEstimating signal parameters using the nonlinearinstantaneous least squares approachrdquo IEEE Transactions onSignal Processing vol 48 no 10 pp 2721ndash2732 2000
[38] J F Weng and S H Leung ldquoNonlinear RLS algorithm foramplitude estimation in class a noiserdquo IEE ProceedingsmdashCommunications vol 147 no 2 pp 81ndash86 2000
[39] D Zachariah M Sundin M Jansson and S ChatterjeeldquoAlternating least-squares for low-rank matrix reconstructionrdquoIEEE Signal Processing Letters vol 19 no 4 pp 231ndash234 2012
[40] R Montoliu and F Pla ldquoGeneralized least squares-based para-metric motion estimationrdquo Computer Vision and Image Under-standing vol 113 no 7 pp 790ndash801 2009
[41] Z Yingjie and G Liling ldquoImproved moving least squares algo-rithm for directed projecting onto point cloudsrdquoMeasurementvol 44 no 10 pp 2008ndash2019 2011
[42] S Seongwook J-S Lim S J Baek and K-M Sung ldquoVariableforgetting factor linear least squares algorithm for frequencyselective fading channel estimationrdquo IEEE Transactions onVehicular Technology vol 51 no 3 pp 613ndash616 2002
[43] S MorimotoM Sanada and Y Takeda ldquoMechanical sensorlessdrives of IPMSM with online parameter identificationrdquo IEEETransactions on Industry Applications vol 42 no 5 pp 1241ndash1248 2006
[44] H Sakai and H Nakaoka ldquoFast sliding window QRD-RLSalgorithmrdquo Signal Processing vol 78 no 3 pp 309ndash319 1999
[45] S Reece and S Roberts ldquoAn introduction to Gaussian processesfor the Kalman filter expertrdquo in Proceedings of the 13th Confer-ence on Information Fusion (FUSION rsquo10) pp 1ndash9 2010
[46] A Giorgano F M Hsu and J Wiley ldquoBook review least-square estimationwith applications to digital signal processingrdquoIEE Proceedings FmdashCommunications Radar and Signal Process-ing vol 132 no 7 1985
[47] S Wang R Zhao W Chen G Li and C Liu ldquoParameter iden-tification of PMSMbased on windowed least square algorithmrdquoin Proceedings of the Manufacturing Science and Technology(ICMST rsquo11) pp 5940ndash5944 Singapore 2012
[48] S Wang S Shi C Chen G Yang and Z Qu ldquoIdentificationof PMSM based on EKF and elman neural networkrdquo inProceedings of the IEEE International Conference on Automationand Logistics (ICAL rsquo09) pp 1459ndash1463 Shenyang China 2009
[49] R E Kalman ldquoA new approach to linear filtering and predictionproblemsrdquo Transactions of the ASMEmdashJournal of Basic Engi-neering D vol 82 pp 35ndash45 1960
[23] T Sadiki M Triki and D T M Slock ldquoWindow optimizationissues in recursive least-squares adaptive filtering and trackingrdquoin Proceedings of the 38th IEEE Annual Asilomar Conferenceon Signals Systems and Computers pp 940ndash944 Pacific GroveCalif USA November 2004
[24] G Welch and G Bishop ldquoAn introduction to the Kalman filterrdquo1997
[25] P J Hargrave ldquoA tutorial introduction to Kalman filteringrdquo inProceedings of the IEE Colloquium on Kalman Filters Introduc-tion Applications and Future Developments pp 11ndash16 1989
[26] M Gautier and P Poignet ldquoExtended Kalman filtering andweighted least squares dynamic identification of robotrdquo ControlEngineering Practice vol 9 no 12 pp 1361ndash1372 2001
[27] H M Al-Hamadi and S A Soliman ldquoKalman filter foridentification of power system fuzzy harmonic componentsrdquoElectric Power Systems Research vol 62 no 3 pp 241ndash248 2002
[28] T Boileau N Leboeuf B Nahid-Mobarakeh and F Meibody-Tabar ldquoOnline identification of PMSM parameters parameteridentifiability and estimator comparative studyrdquo IEEE Transac-tions on Industry Applications vol 47 no 4 pp 1944ndash1957 2011
[29] A Bechouche H Sediki D O Abdeslam and S HaddadldquoIdentification of induction motor at standstill using artificialneural networkrdquo in Proceedings of the 36th Annual Conferenceon IEEE Industrial Electronics Society (IECON rsquo10) pp 2908ndash2913 Glendale Ariz USA 2010
[30] F Alonge F DrsquoIppolito and FM Raimondi ldquoLeast squares andgenetic algorithms for parameter identification of inductionmotorsrdquo Control Engineering Practice vol 9 no 6 pp 647ndash6572001
[31] S Mishra ldquoA hybrid least square-fuzzy bacterial foragingstrategy for harmonic estimationrdquo IEEE Transactions on Evo-lutionary Computation vol 9 no 1 pp 61ndash73 2005
[32] R Liao H Zheng S Grzybowski and L Yang ldquoParticleswarm optimization-least squares support vector regressionbased forecasting model on dissolved gases in oil-filled powertransformersrdquo Electric Power Systems Research vol 81 no 12pp 2074ndash2080 2011
[33] R A Zadeh A Ghosh and G Ledwich ldquoCombination ofKalman filter and least-error square techniques in powersystemrdquo IEEE Transactions on Power Delivery vol 25 no 4 pp2868ndash2880 2010
[34] S Bolognani R Oboe andM Zigliotto ldquoSensorless full-digitalPMSM drive with EKF estimation of speed and rotor positionrdquoIEEE Transactions on Industrial Electronics vol 46 no 1 pp184ndash191 1999
[35] M Haardt ldquoStructured least squares to improve the per-formance of ESPRIT-Type algorithmsrdquo IEEE Transactions onSignal Processing vol 45 no 3 pp 792ndash799 1997
[36] M Ghogho A Swami and A K Nandi ldquoNon-linear leastsquares estimation for harmonics in multiplicative and additivenoiserdquo Signal Processing vol 78 no 1 pp 43ndash60 1999
[37] J Angeby ldquoEstimating signal parameters using the nonlinearinstantaneous least squares approachrdquo IEEE Transactions onSignal Processing vol 48 no 10 pp 2721ndash2732 2000
[38] J F Weng and S H Leung ldquoNonlinear RLS algorithm foramplitude estimation in class a noiserdquo IEE ProceedingsmdashCommunications vol 147 no 2 pp 81ndash86 2000
[39] D Zachariah M Sundin M Jansson and S ChatterjeeldquoAlternating least-squares for low-rank matrix reconstructionrdquoIEEE Signal Processing Letters vol 19 no 4 pp 231ndash234 2012
[40] R Montoliu and F Pla ldquoGeneralized least squares-based para-metric motion estimationrdquo Computer Vision and Image Under-standing vol 113 no 7 pp 790ndash801 2009
[41] Z Yingjie and G Liling ldquoImproved moving least squares algo-rithm for directed projecting onto point cloudsrdquoMeasurementvol 44 no 10 pp 2008ndash2019 2011
[42] S Seongwook J-S Lim S J Baek and K-M Sung ldquoVariableforgetting factor linear least squares algorithm for frequencyselective fading channel estimationrdquo IEEE Transactions onVehicular Technology vol 51 no 3 pp 613ndash616 2002
[43] S MorimotoM Sanada and Y Takeda ldquoMechanical sensorlessdrives of IPMSM with online parameter identificationrdquo IEEETransactions on Industry Applications vol 42 no 5 pp 1241ndash1248 2006
[44] H Sakai and H Nakaoka ldquoFast sliding window QRD-RLSalgorithmrdquo Signal Processing vol 78 no 3 pp 309ndash319 1999
[45] S Reece and S Roberts ldquoAn introduction to Gaussian processesfor the Kalman filter expertrdquo in Proceedings of the 13th Confer-ence on Information Fusion (FUSION rsquo10) pp 1ndash9 2010
[46] A Giorgano F M Hsu and J Wiley ldquoBook review least-square estimationwith applications to digital signal processingrdquoIEE Proceedings FmdashCommunications Radar and Signal Process-ing vol 132 no 7 1985
[47] S Wang R Zhao W Chen G Li and C Liu ldquoParameter iden-tification of PMSMbased on windowed least square algorithmrdquoin Proceedings of the Manufacturing Science and Technology(ICMST rsquo11) pp 5940ndash5944 Singapore 2012
[48] S Wang S Shi C Chen G Yang and Z Qu ldquoIdentificationof PMSM based on EKF and elman neural networkrdquo inProceedings of the IEEE International Conference on Automationand Logistics (ICAL rsquo09) pp 1459ndash1463 Shenyang China 2009
[49] R E Kalman ldquoA new approach to linear filtering and predictionproblemsrdquo Transactions of the ASMEmdashJournal of Basic Engi-neering D vol 82 pp 35ndash45 1960