103

3604 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 59, NO. 9, SEPTEMBER 2012

Control Design and Implementation for HighPerformance Shunt Active Filters in

Aircraft Power GridsJunyi Liu, Pericle Zanchetta, Member, IEEE, Marco Degano, Member, IEEE, and Elisabetta Lavopa

Abstract—This paper presents the design and implementationof a Shunt Active Filter (SAF) for aircraft power networks usingan accurate wide-band current control method based on IterativeLearning Control (ILC). The SAF control system is designed tocompensate harmonic currents, with a 400 Hz supply voltage.This work introduces useful design strategies to increase the er-ror-decay speed and improve the robustness of the SAF controlsystem by using a hybrid P-type ILC controller. Detailed designof the hybrid P-type ILC controller and simulation results arepresented. The overall system implementation is demonstratedthrough experimental results on a laboratory prototype.

Index Terms—Active filters, harmonic distortion, iterativelearning control (ILC), power quality.

I. INTRODUCTION

IN THE PAST two decades the increasing intensive use ofnonlinear loads has resulted in a substantial reduction of

power quality in electric power systems. Current harmonicsproduced by nonlinear loads, such as power electronic convert-ers and electrical drives cause supply voltage harmonics and anumber of related problems in power distribution networks. Inmore recent years this problem has affected also smaller distri-bution grids like for example in ships and aircrafts. The “moreelectric aircraft” trend, consisting in the replacement of mostof hydraulic/pneumatic actuators with electronically controlledelectromechanical devices, is gaining interest in the aerospaceindustry. In fact, it is expected to provide significant benefitsin terms of actuation accuracy, employ flexibility, system de-pendability, energy efficiency and overall lifecycle cost thanksto the reduced maintenance requirements [1]. Since aircraftelectric power systems are relatively small with a rough balanceof rated power of loads and generators, power quality issuesgrow with the number and size of the loads driven by staticconverters. This is leading to a growing interest in active mainsinterfaces and active filters suited to operate in aerospace ambit.In particular, shunt active filters (SAFs) are generally used toturn unbalanced, non-resistive and distorting loads into equiva-lent balanced resistive linear loads. SAFs have been intensively

Manuscript received November 5, 2010; revised May 5, 2011; acceptedJune 26, 2011. Date of publication August 18, 2011; date of current versionApril 13, 2012.

The authors are with the Department of Electrical and ElectronicEngineering of the University of Nottingham, NG7 2RD Nottingham,U.K. (e-mail: [email protected]; [email protected];[email protected]; [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TIE.2011.2165454

studied in the literature for 50/60 Hz grids applications andmany suitable control strategies are developed and proposedin the literature [2]–[9]. In [2] a control technique with two-step prediction has been developed. In [3] a stability analysis ofthe common dead-beat control is presented and a technique toincrease its robustness is proposed. In [4] a selective harmoniccompensation technique is developed, based on synchronousframe controllers. In [5] a resonant controller optimized withgenetic algorithms is proposed. In [7] the authors present arepetitive control scheme based on a finite-impulse responsedigital filter. The application of these techniques in the avionicambit presents specific issues mainly related to the higherrated supply frequency, which deserves further investigation.In [10] a complete model of an aircraft electric power systemis developed using an accurate harmonic cancellation method.A multi-level converter solution is instead proposed in [11] inorder to obtain a good reference tracking with limited switchingfrequency. In [12] an improved deadbeat digital controller ispresented for shunt active filters used for compensation of loadharmonics in aircraft power systems. As mentioned earlier, theaerospace ambit poses specific challenges for both the powerand control parts of SAFs due to the much higher frequency(400 Hz in spite of 50/60 Hz standard industrial applications).Such aspect will be exacerbated in the future when the plannedvariable-frequency-and-voltage operation will be adopted [10].In fact, good compensation performances require a suitablylarge bandwidth of control loops in comparison to the basefrequency. This would in turn require, in a 400 Hz powersupply, expensive high specs power semiconductor devices andmicrocontrollers/dsp technology.

This paper will try to address these issues and propose asimple and efficient control solution based on the use of aHybrid P-type Iterative learning control (ILC) together withuseful guidelines for an optimized design. This is a fully digitalcontrol solution that can be implemented using current standardtechnology. Simulation and experimental results confirm theeffectiveness and the reliability of the proposed strategy.

II. SAF SYSTEM OVERVIEW

The SAF for this design uses the conventional structureshown in Fig. 1. A nonlinear load, represented by a diodebridge rectifier, is connected to the power network to generateharmonics and emulate a distorted current in the aircraft powernetwork. A Voltage Source Converter (VSC) is connected to the

0278-0046/$26.00 © 2011 IEEE

LIU et al.: CONTROL DESIGN AND IMPLEMENTATION FOR HIGH PERFORMANCE SHUNT ACTIVE FILTERS IN AIRCRAFT POWER GRIDS 3605

Fig. 1. Standard scheme of a shunt active filter.

Fig. 2. Structure of SAF control system.

power network at the Point of Common Coupling (PCC) andinjects the current if to compensate for the current harmonicsin il. The SAF control system is a cascade control loop, whichincludes an outer voltage loop and an inner current loop asshown in Fig. 2.

The outer voltage loop controls the energy balance of thecapacitor in the VSC to maintain a constant DC-link voltage(Vdc). The inner current loop takes the sum of demand supplycurrent (Is∗) and demand current from the outer voltage loop(Idc∗) as the current reference amplitude. This is then multi-plied by a sinusoidal three-phase template synchronous with thesupply voltage to generate the current reference.

The actual supply current is can be derived by the measure-ment of the load current il and the SAF output current if . Thecurrent loop then produces a demand voltage to control the SAFcurrent if . This demand voltage needs to be subtracted to themeasured voltage at the PCC to produce the reference voltagefor the PWM modulator.

According to the scheme depicted in Fig. 1, the SAF experi-mental prototype used includes a standard 3-legs IGBT basedVSC whose leg rated current is 15 A whereas the designedDC bus voltage is 400 V. The DC terminals of the inverterare connected to a capacitors bank featuring a 2200 µF ca-pacity, whereas the AC terminals are connected to the PCC viathree filtering inductors featuring equivalent series parameters

Fig. 3. Concept of P-type ILC control.

L = 1 mH, R = 0.15 Ω. The supply voltage is 115 Vrms phaseto neutral at 400 Hz. The switching and sampling frequenciesare set at 14.4 kHz which gives 36 samples per cycle.

III. PRINCIPLES OF THE P-TYPE ILC

ILC is a linear control technique suitable for systems whichpresent a repetitive behavior; it is therefore a promising solutionfor SAF applications as it provides a very accurate steady statecurrent regulation to cancel harmonic currents in the powernetwork. ILC is based on the internal modeling principle: thecontrol loop iteratively adjusts the output signal of the con-troller by learning the error in the previous repetition (cycle),thus the tracking error of the controller can be iterativelyreduced. Theoretically, within a finite time, the control systemcan achieve zero tracking error [13]–[19].

The P-type ILC is an intelligent control structure which canbe applied for the regulation of systems operating under arepeated reference signal (see Fig. 3) [20]. The P-type ILClearns the tracking error from the previous repetition ek−1(z)and uses a learning update algorithm to adjust the control signaluk(z) in the current repetition to reduce the tracking error.

Let’s consider a single-input single-output system. ILC canbe used to limit the tracking error caused by a periodicaldisturbance if the system transfer function Gp(z) satisfies thefollowing conditions [21]:

1) reference signal is repetitive;2) system has the same initial condition in each repetition;3) measurement noise of the output is small;4) system dynamics is invariant.

The simple learning update algorithm used in the P-type ILCcontroller can be represented by

uk = uk−1(z) + L(z)ek−1(z) (1)

where uk(z) is the z-domain transform of the control signaldriving the system plant at the kth repetition, L(z) is calledlearning factor and the ek−1(z) is the z-domain transform of thetracking error at the k-1th repetition. The error ek(z) betweenthe actual current and the reference, and the output yk(z) of theplant are given, respectively, by

ek(z) = yd(z) − yk(z) (2)

yk(z) = Gp(z)uk (3)

3606 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 59, NO. 9, SEPTEMBER 2012

where yd(z) is the reference signal. If the conditions 1–4 aresatisfied, substituting (1) and (3) into (2) yields

ek(z) = [1 − L(z)Gp(z)] ek−1(z) (4)

where [1 − L(z)Gp(z)] is the transfer function between thetracking errors in the current and next repetitions. By settingz = ejωT , where T is the sampling time, the tracking error willdecay to zero after a finite number of repetitions if

∣∣1 − L(ejωT )Gp(ejωT )∣∣ < 1, ωT ∈ [−π, π]. (5)

Relation (5) represents the error-decay condition of the P-typeILC controller; the quantity |1 − L(ejωT)Gp(ejωT)| is callederror-decay factor. L(ejωT) is the learning factor and includestwo components: learning gain (L) and phase shift. The P-typeILC has to satisfy the error-decay condition for all frequenciesfrom zero to the Nyquist frequency; failing this, the trackingerror may be amplified at certain frequencies. In addition,a smaller error-decay factor determines a faster error-decayspeed. Hence, a good design of the P-type ILC controller needsto determine a learning gain which satisfies the error-decaycondition and generates minimized values of the error-decayfactor at the frequencies of interest [20].

IV. SAF HYBRID P-TYPE ILC CURRENT

CONTROL DESIGN

The P-type ILC controller, in which the control action isbased only on the previous cycle tracking error, provides a veryeffective steady state harmonic compensation due to its repeti-tive action, but does not produce a good transient performance.In particular, it is very sensitive to the variations in the referencesignal between each nearby repetition. In particular, for thecurrent control in SAF applications, a non-periodical demandcurrent from the voltage control loop is added to the demandsupply current, during transient conditions, to generate thereference of the inner current control loop (Fig. 2). The demandcurrent from the outer voltage loop may cause a large variationof the reference signal, above all during initial repetitions andload variations. Given the poor dynamic response of a P-typeILC controller, a hybrid P-type ILC controller is preferred forthis specific application.

The hybrid P-type ILC controller is shown in Fig. 4, where aPI controller has been added in parallel to a normal P-type ILCsystem. The PI controller is usually designed by ignoring theP-type ILC term, and by using classical methods like forexample the root locus. Regarding the determination of thedelay values of the memories and the learning factor, furtherconsiderations need to be made.

First of all it should be pointed out that, since the P-type ILCcontroller is combined with a PI controller in this application,the Gp(z) in (4) does not correspond to the current controlplant transfer function G(z). If P(z) is the PI controller transferfunction, from the system in Fig. 4, the following equation canbe obtained:

ek(z) =(

1 − L(z)G(z)z−1

1 + G(z)z−1P (z)

)ek−1(z). (6)

Fig. 4. Structure of the hybrid P-type ILC SAF current control.

Comparing (6) with (4), the Gp(z) can be determined as

Gp(z) =G(z)z−1

1 + G(z)z−1P (z). (7)

The learning factor design procedure is based on the use ofthe error-decay condition. This condition can be represented ina Nyquist diagram as a unit circle with central point at (1, 0).The error-decay factor at a certain frequency is represented bythe distance between the central point and the point on the locusof L(ejωT)Gp(ejωT) at the same frequency. The learning factorL(ejωT) is selected to ensure that the locus lies inside the unitcircle to satisfy the error-decay condition and to minimize thevalue of error-decay factor at the frequencies of interest. Inorder to do so, the learning factor phase shift component is usedto push the locus of Gp(ejωT) into the first and fourth quadrantsin a Nyquist diagram, while the learning gain (L) adjusts thegain of the locus to provide small values of error-decay factorat the frequencies of interest for the control (fundamental andharmonic frequencies in SAF applications).

In this application, the learning factor phase shift componentis a time-advance unit zm. As shown in Fig. 5, by selectingthis phase shift component as z2, the locus of L(ejωT)Gp(ejωT)lies in the first and fourth quadrant; the learning gain has beenchosen as 0.813 to keep the locus inside the unit circle andto minimize the values of the error-decay factor at the mainharmonic frequencies. The memories (Mem1 and Mem2) inFig. 4 are the discrete delays used to delay the tracking error andcontrol signals for an entire repetition. As discussed earlier, thelearning factor phase shift component is a time advance unit zm,hence the discrete delay for the error signal (Mem 2) becomesz−(N−m) where N (=36 in this case) is the number of samplesin one period of the fundamental, while the delay for the controlsignal remains z−N. The Bode diagram of the magnitude ofthe closed loop current control using the direct and the hybridP-type ILC, respectively is presented in Fig. 6.

Fig. 6(a) shows the whole frequency band up to the Nyquistfrequency, while Fig. 6(b) shows an expanded view of theharmonic frequencies. In Fig. 6(a) it can be seen that the hybridP-type ILC approach shows a higher closed loop bandwidthwhich provides a faster dynamic response in the proposed cur-rent control loop. Fig. 6(b) shows that the hybrid ILC provideshigher selective effect at each harmonic order.

V. CONTROL DESIGN ENHANCEMENT

There are a few drawbacks in the previous design procedure,which can be summarized in the following three points.

LIU et al.: CONTROL DESIGN AND IMPLEMENTATION FOR HIGH PERFORMANCE SHUNT ACTIVE FILTERS IN AIRCRAFT POWER GRIDS 3607

Fig. 5. Locus of L(ejωT )Gp(ejωT ) − L = 0.813, m = 2.

Fig. 6. Bode plot of the magnitude of the SAF current control loop with directand hybrid P-type ILC controller (traditional design) , respectively. (a) wholespectrum. (b) expanded view of harmonics.

1) If the PI controller in the hybrid P-type ILC system is de-signed ignoring the presence of the P-type ILC controller,the error-decay speed of the current control system will belimited.

2) The non-periodic demand current (I∗dc) from the voltagecontrol loop causes a variation of the current reference ateach repetition during transient conditions.

Fig. 7. Overall, SAF control scheme.

Even if the hybrid P-type ILC system performs muchbetter than the P-type in this respect, it still produces aninaccurate current tracking when the reference variationis large.

3) The hybrid P-type ILC controller presents poor robust-ness also against other non-periodical disturbances suchas measurement noise.

This work proposes optimized design solutions in order toincrease the error-decay speed and the ILC robustness againstreference variations and measurement noise. The introductionof a variable learning gain L and of a forgetting factor α(Fig. 7) is accompanied by modifications in the design strategyof current loop PI controller.

A. PI controller Design Procedure in Hybrid P-Type ILC

Equation (7) shows that the design choice of P(z) can modifymagnitude and phase of the frequency response of Gp(z).

Thus an optimized design of the PI controller needs toaddress two main issues: 1) produce a satisfying dynamicresponse of the current control and 2) produce a magnitudeincrease and phase shift decrease of Gp(z) at the frequenciesof interest for the control, so that the error-decay factor atthose frequencies can be reduced. This can be achieved eitherthrough an iterative trial and error procedure or, for a moreaccurate design, using software optimization routines (for ex-ample based on Genetic Algorithms). A trial and error tuningcan be carried out from (7), setting the desired magnitude andphase of the frequency response Gp(ejωT) for the particularharmonic frequency to be compensated. However, if the aim isto compensate multiple harmonic components, an optimizationtechnique, like a GA, is a more suitable approach, in order tofind the controller that gives the best performance overall. Inthe work presented in this paper, the authors have implementedGenetic Algorithms for the optimization of the PI controller.The GA has been implemented in order to obtain a magnitudefrequency response as close as possible to one (0 dB) in therange of frequencies corresponding to the harmonic compo-nents to be compensated, and a phase shift as close as possibleto zero in the same range of frequencies. These conditionscorrespond to an operation point positioned as close as possibleto the point with coordinates (1,0) in the complex plane, whichguarantees the fastest error decay, as shown in Fig. 5. Thedesign of the traditional PI controller is performed ignoringthe ILC dynamics, adopting as a design criterion the speed ofresponse (the controller is designed setting the fastest speed

3608 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 59, NO. 9, SEPTEMBER 2012

Fig. 8. Frequency response of Gp(z).

Fig. 9. Bode plot of the SAF current control loop with traditional andoptimized design hybrid P-type ILC controllers.

of response possible). On the other hand, the optimized PI isdesigned with the criteria mentioned above, in order to satisfythe error decay condition. After that, a value for the learninggain is chosen in order to improve the error decay condition,given the optimized PI. The learning gain will have the effectof a scaling factor for the magnitude of Gp(z), shifting up theresponse to larger values, which makes it easier to improve theresponse at all frequencies by means of a constant value of L.Fig. 8 shows the effect of the suggested design approach to thefrequency response of Gp(z), which improves the error decayfactor. Fig. 9 shows an expanded view of the Bode diagram ofthe magnitude of the current control loop using the traditionaland optimized hybrid P-type ILC at the harmonic frequencies.It can be seen that the optimized hybrid P-type ILC provideslarger gains at the higher harmonic frequencies (up to thesystem Nyquist frequency), hence those harmonic componentscan be more efficiently compensated in the SAF current control,if the optimized PI design is chosen.

The transfer functions of the traditional and the optimized PIcontroller utilized for the application presented in this paper are

P (z) =4.1 · (z − 0.979)

z − 1(8)

P (z) =9.39 · (z − 0.989)

z − 1. (9)

B. Variable Learning Gain (L)

From a robustness analysis of the control system, it has beenobserved that the P-type ILC control with a smaller learninggain (L) has a faster error-decay speed when a large disturbanceoccurs in the SAF current control loop. However, when adisturbance with lower intensity occurs, a smaller value of Lcan reduce the error-decay speed for the P-type ILC controller.Therefore, if the value of the learning gain (L) is made variableaccording to the intensity of the disturbance, then the P-typeILC controller can achieve optimized error-decay speed in eachdifferent condition. The idea is to implement a learning gainthat reduces the effect of the reference variation. The ILC showsits best performance when the reference is constant, because thealgorithm is based on the error at the previous repetition, hence,when the error variation is large, a small learning gain reducesthe effect of the variation, improving the performance of theILC. When the variation is small, a larger value of the learninggain can be used. A rigorous demonstration of this concept isgiven in [22].

When the SAF is enabled to full load, the ideal solution is toset a small value of L, as in that condition the variation in thecurrent reference is large. The value of L is then increased ateach repetition, as a function of the attenuation of this referencevariation, finally reaching a constant value when the referencevariation is negligible.

The approach used in this method first finds the most suitablevalue of L which can provide the fastest error-decay within thefirst repetition. Then using the value of L generated from theprevious repetition as a reference point, it derives the valueof L for the next repetition using the same method; this isiterated for the subsequent repetitions until the disturbancebecomes negligible. The optimization of the variable learninggain has been carried out by means of Genetic Algorithms(GA). The GA is programmed in order to minimize the fitnessfunction represented by the control error. On the basis of thisminimization criterion, it finds the optimal learning gain value.It has been found that the learning gain obtained as a result ofthe GA optimization can be expressed by the following relation,where it can be expressed as a linear function of the errorvariation, as in (10):

L(z) = −0.3395 · [ek(z) − ek−1(z)] + 3.2238. (10)

Fig. 10(a) shows the case where the ASF is enabled to fullload at 0.1 s. The current reference starts from zero, increasingwith a constant high slope during the time interval 0.1–0.106 s.The variable learning gain is equal to 0.4 during this interval (ithas been found that for very high variations of the reference it isbest to saturate L to a minimum value—0.4 in this application).After 0.106 s the reference variation decreases, so L linearlyincreases to 3.2238. From 0.15 s to 0.16 s the reference signalis constant, so its variation is zero, and the learning gain is set toa constant value equal to 3.2238. Fig. 10(b) shows the variablelearning gain implementation in the transient case when, att = 0.3 s, the load is switched from full to half. The currentreference is constant in the time interval 0.29–0.3 s. Duringthis interval the reference variation is zero, so L is equal to3.2238. When the transient occurs, the reference decreases with

LIU et al.: CONTROL DESIGN AND IMPLEMENTATION FOR HIGH PERFORMANCE SHUNT ACTIVE FILTERS IN AIRCRAFT POWER GRIDS 3609

Fig. 10. Concept of variable learning gain.

a high slope, so the learning gain, after a few repetitions, goesdown to a low value (0.4) and it remains constant for thetime interval 0.3–0.305 s. From 0.305 s the reference variationstarts decreasing and L increases as a ramp up to 3.2238, andit remains constantly equal to that value when the referencevariation becomes zero, from 0.32 s onwards.

C. Forgetting Factor (α)

The forgetting factor is a gain introduced in the controlaction signal feedback (Fig. 7), which has been successfullyapplied to improve the P-type ILC system robustness againstmeasurement white noise in robotic motion application [23].

As shown in Fig. 7, the forgetting factor is used in thisimplementation to reduce the value of the control action signaluk. It therefore reduces the impact of random noise on thelearning process, allowing the controller to focus its action onthe repetitive component of the tracking error in the followingrepetitions. This solution however will produce a lower errordecay speed.

To demonstrate the effect of the forgetting factor, Fig. 11shows a zoomed view of the frequency responses of the openloop transfer function around the 5th harmonic frequency ofthe P-type ILC (designed in the previous sections), by usingdifferent forgetting factor values. It can be found that theforgetting factor narrows the system bandwidth at the harmonic

Fig. 11. Frequency response of the P-type ILC controller at 5th harmonic withdifferent forgetting factor value.

frequencies making the control action more selective, but itdecreases the closed loop gain at the same frequency. Thedegree of such effect is related to the value of α, i.e., a smallerα means more effectiveness in compensating the tracking error(higher gain). On the other hand, a larger α indicates highersteady-state tracking error (lower gain). Studies show that witha small value of α (set to 0.01), the P-type ILC controller ismore effective in compensating the SAF current tracking errorin the presence of white noise owing to a good compromisebetween gain and selectiveness at harmonic frequencies. Theoverall gain of the system is also reduced at high frequency.

VI. SIMULATION AND EXPERIMENTAL RESULTS

The SAF and the proposed control have been simulated inMatlab-Simulink environment using the circuit configurationand data as in Section II. The nonlinear load consists of adiode bridge rectifier equipped with input inductive series filter(L = 1 mH), supplying a 1 kW resistive load.

Fig. 12 shows the supply current before and after the SAFcompensation. Through the harmonic spectra shown in Fig. 13,it can be found that, the optimized hybrid P-type ILC systemreduces the THD from 26.5% to 1.72%. In the same transientconditions of Fig. 10, the simulation results in Figs. 14 and15 clearly show that the optimized hybrid P-type ILC canactually provide an overall better dynamic response over thedirect one for the SAF current control. Fig. 16 shows that theoptimization of the hybrid P-type ILC controller leads to a92.2% smaller initial error, and a 33% faster error-decay speedfor the SAF control system with an average steady state errorof 0.001 A. In order to investigate and compare the robustnessof the control system with and without the use of a forgettingfactor, measurement white noise with a value of 10% of thesupply current is added to the measured supply current signal.As shown in Fig. 17, the compensated supply current using acontrol system without the forgetting factor, presents a THDequal to 3.85%. Comparatively, the use of a forgetting factorequal to 0.01 successfully reduces the THD to 2.11%.

The experimental implementation uses the circuit configu-ration and data as in simulation. The control system is com-posed of a main board featuring a TMS320C6713 digital signal

3610 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 59, NO. 9, SEPTEMBER 2012

Fig. 12. Supply current without (upper) and with (lower) SAF compensation.

Fig. 13. Harmonic spectrum without (a) and with (b) SAF compensation.

Fig. 14. Supply current tracking using direct (a) and optimized hybrid(b) P-type ILC controlled SAF system; SAF is enabled to full load at 0.1 s.

Fig. 15. Supply current tracking using direct (a) and optimized hybrid(b) P-type ILC controlled SAF system; load is switched from full to half at0.3 s.

processor (DSP) and an auxiliary board equipped with a fieldprogrammable gate array (FPGA) used for data acquisition andpulse generation. Such boards may be noticed on the top inthe picture of Fig. 18, showing the prototype SAF except the

LIU et al.: CONTROL DESIGN AND IMPLEMENTATION FOR HIGH PERFORMANCE SHUNT ACTIVE FILTERS IN AIRCRAFT POWER GRIDS 3611

Fig. 16. Average tracking error in each repetition with the traditional andoptimized hybrid P-type ILC controller.

Fig. 17. Harmonic spectrum of the compensated supply current with SAFcontrol system with and without forgetting factor presence of measurementnoise.

Fig. 18. Top view of the experimental SAF prototype.

AC side inductors. The fixed frequency line-to-line voltage at400 Hz, 115 Vrms emulating an aircraft power supply wasprovided by a 12 kVA dedicated controlled static generator(Chroma 61705). Fig. 19 shows the experimental line currentand its harmonic spectrum before the compensation whileFig. 20 shows the same quantities after the compensation. Theharmonic content is noticeably improved with the current THD

Fig. 19. Supply line current (a) and its harmonic spectrum (b) without SAFcompensation (experimental).

which reduces from 27.6% to 3.32%. The effectiveness of theSAF compensation at high harmonic frequencies (almost upto the Nyquist frequency) is also evident from the spectra.Fig. 21 shows the experimental supply current compared withits reference, during a transient where the load is switched fromfull to half. It can be seen how, after the few initial cycles, thetracking is very accurate during the transient. Figs. 22 and 23show the oscilloscope screenshots with expanded views of thetransient and the steady-state. In these figures, the experimentalsupply voltage and supply current are presented, demonstratinghigh quality compensation and unity power factor operation.

VII. CONCLUSION

The more electric aircraft trend has led to the increasing useof power converters on board of modern aircrafts; thus the useof shunt active power filters becomes desirable for on-boardpower quality issues. The much higher rated supply frequencymakes the SAF design scenario more challenging comparedto standard 50/60 Hz industrial grid applications, due to theperformance limits of power and control devices. This paperinvestigated the development of an optimized hybrid P-typeILC controlled SAF for aerospace applications working in anaircraft power system with supply frequency of 400 Hz.

3612 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 59, NO. 9, SEPTEMBER 2012

Fig. 20. Supply line current (a) and its harmonic spectrum (b) with SAFcompensation (experimental).

Fig. 21. Supply current tracking; load switched from full to half at 0.3 s(experimental).

This system requires therefore the ability of compensatingharmonics at very high frequencies. Considering both simu-lation and experimental results, it can be concluded that theproposed SAF control has proved to be very effective foraccurate reduction of current harmonics on aircraft power grids,

Fig. 22. Oscilloscope screenshot showing the supply voltage and the supplycurrent during the transient (load switched from full to half).

Fig. 23. Oscilloscope screenshot showing the supply voltage and the supplycurrent during the steady-state (load switched from full to half).

using ordinary equipment and reasonable switching frequencyand also ensuring good dynamics in transient conditions.

REFERENCES

[1] J. A. Rosero, J. A. Ortega, E. Aldabas, and L. Romeral, “Moving towardsa more electric aircraft,” IEEE Aerosp. Electron. Syst. Mag., vol. 22, no. 3,pp. 3–9, Mar. 2007.

[2] M. Odavic, P. Zanchetta, M. Sumner, C. Ladisa, and Z. Jakopovic, “A“two ahead” predictive controller for active shunt power filters,” in Proc.IECON, 2006, pp. 4713–4718.

[3] L. Malesani, P. Mattavelli, and S. Buso, “Robust dead-beat current controlfor PWM rectifiers and active filters,” IEEE Trans. Ind. Appl., vol. 35,no. 3, pp. 613–620, May/Jun. 1999.

[4] P. Mattavelli, “A closed loop selective harmonic compensation for activefilters,” IEEE Trans. Ind. Appl., vol. 37, no. 1, pp. 81–89, Jan./Feb. 2001.

[5] W. Lenwari, M. Sumner, and P. Zanchetta, “The use of genetic algorithmsfor the design of resonant compensators for active filters,” IEEE Trans.Ind. Electron., vol. 56, no. 8, pp. 2852–2861, Aug. 2009.

[6] S. Bhattacharya, T. M. Frank, and D. M. Divan, “Active filter sys-tem implementation,” IEEE Ind. Appl. Mag., vol. 4, no. 5, pp. 47–63,Sep./Oct. 1998.

[7] P. Mattavelli and F. P. Marafao, “Repetitive-based control for selectiveharmonic compensation in active power filters,” IEEE Trans. Ind. Elec-tron., vol. 51, no. 5, pp. 1018–1024, Oct. 2004.

[8] Z. Shu, Y. Guo, and J. Lian, “Steady-state and dynamic study of activepower filter with efficient FPGA-based control algorithm,” IEEE Trans.Ind. Electron., vol. 55, no. 4, pp. 1527–1536, Apr. 2008.

[9] B. Singh and J. Solanki, “An implementation of an adaptive control al-gorithm for a three-phase shunt active filter,” IEEE Trans. Ind. Electron.,vol. 56, no. 8, pp. 2811–2820, Aug. 2009.

[10] A. Eid, H. El-Kishky, M. Abdel-Salam, and M. T. El-Mohandes, “Onpower quality of variable-speed constant-frequency aircraft electric powersystems,” IEEE Trans. Power Del., vol. 25, no. 1, pp. 55–65, Jan. 2010.

LIU et al.: CONTROL DESIGN AND IMPLEMENTATION FOR HIGH PERFORMANCE SHUNT ACTIVE FILTERS IN AIRCRAFT POWER GRIDS 3613

[11] M. Odavic, P. Zanchetta, and M. Sumner, “A low switching frequencyhigh bandwidth current control for active shunt power filter in aircraftspower networks,” in Proc. IECON, 2007, pp. 1863–1868.

[12] V. Biagini, M. Odavic, P. Zanchetta, M. Degano, and P. Bolognesi, “Im-proved dead beat control of a shunt active filter for aircraft power sys-tems,” in Proc. IEEE ISIE, Jul. 4–7, 2010, pp. 2702–2707.

[13] S. S. Saab, “Optimality of first-order ILC among higher order ILC,” IEEETrans. Autom. Control, vol. 51, no. 8, pp. 1332–1336, Aug. 2006.

[14] R. Costa-Castello, “Demonstration of the internal model principle bydigital repetitive control of an educational laboratory plant,” IEEE Trans.Educ., vol. 48, no. 1, pp. 73–80, Feb. 2005.

[15] S. K. Sahoo and J.-X. Xu, “Indirect torque control of switched reluctancemotors using iterative learning control,” IEEE Trans. Power Electron.,vol. 20, no. 1, pp. 200–208, Jan. 2005.

[16] R.-E. Precup, S. Preitl, J. K. Tar, M. L. Tomescu, M. Takacs, P. Korondi,and P. Baranyi, “Fuzzy control system performance enhancement by itera-tive learning control,” IEEE Trans. Ind. Electron., vol. 55, no. 9, pp. 3461–3475, Sep. 2008.

[17] R. Marino and P. Tomei, “An iterative learning control for a class ofpartially feedback linearizable systems,” IEEE Trans. Autom. Control,vol. 54, no. 8, pp. 1991–1996, Aug. 2009.

[18] D. Meng, Y. Jia, J. Du, and S. Yuan, “Robust discrete-time iterativelearning control for nonlinear systems with varying initial state shifts,”IEEE Trans. Autom. Control, vol. 54, no. 11, pp. 2626–2631, Nov. 2009.

[19] K. Abidi and J.-X. Xu, “Iterative learning control for sampled-data sys-tems: Theory to practice,” IEEE Trans. Ind. Electron., vol. 58, no. 7,pp. 3002–3015, Jul. 2011, DOI: 10.1109/TIE.2010.2070774.

[20] S.-J. Yu and X.-D. Qi, The Theories and Applications of the IterativeLearning Control. Norwell, MA: Kluwer, 2005.

[21] Z. Bien and J.-X. Xu, Iterative Learning Control Analysis, Design, Inte-gration and Applications. Norwell, MA: Kluwer, 1998.

[22] T. Senping, C. Dingxi, and T. Huiping, “Algorithm and robustness ofiterative learning control for a class of singular system,” in Proc. 27thChin. Control Conf., 2008, pp. 106–109.

[23] C. J. Chien and J. S. Liu, “A P-type iterative learning controller forrobust output tracking of non-linear time-varying system,” Int. J. Control,vol. 64, no. 2, pp. 319–334, 1996.

Junyi Liu received the B.Eng. degree in electri-cal engineering from the University of Nottingham,Nottingham, U.K., in July 2006. He received thePh.D. degree from the University of Nottingham, inDecember 2010.

He is currently working as a research fellowin South University of Science and Technology ofChina. His main research interests are in the fieldof power quality, active filters, power converters inparticular for electrical systems on aircrafts.

Pericle Zanchetta (M’00) received the 5 years Lau-rea degree in electronic engineering from the Tech-nical University of Bari, Bari, Italy, in 1993 and thePh.D. degree in electrical engineering from the sameUniversity, in 1997.

In 1998, he became Assistant Professor of PowerElectronics and control at the Technical Universityof Bari. In 2001, he became lecturer in control ofpower electronics systems in the PEMC researchgroup at the University of Nottingham, Nottingham,U.K., where he is currently Associate Professor. He

has published over 120 papers in international Journals and conferences. Hismain research interests are in the field of power quality and harmonics, activepower filters, Repetitive and Model Predictive Control of power converters,Design and Identification using Heuristic optimization strategies.

Marco Degano received the 5 years Laurea degree inelectronic engineering from the University of Udine,Udine, Italy, in April 2004.

In February 2008, he joined the Power ElectronicsMachines and Control (PEMC) research group atthe University of Nottingham, Nottingham, U.K.,where he is currently working towards his Ph.D.His main research interests are in the field of activefilters, power converters and EMC filters for aircraftapplications.

Elisabetta Lavopa received the 5 years Laurea de-gree in electrical engineering from the TechnicalUniversity of Bari, Bari, Italy, in February 2005. Shereceived the Ph.D. degree from the University ofNottingham, Nottingham, U.K., in December 2010.

She is currently a Research Fellow in the PEMCresearch division at the University of Nottingham.Her main research interests are in the field of powerquality, active filters, power converters in particularfor electrical systems on aircrafts, algorithms forharmonic analysis.