Optimization of distributed generation of renewable energy sources by intelligent techniques Marcello Pucci – Institute for Studies on Intelligent Systems for Automation (I.S.S.I.A),

Optimization of Distributed Generation of Renewable Energy Sources by Intelligent Techniques

Marcello Pucci

ISSIA (Istituto di Studi sui Sistemi Intelligenti per l’Automazione) – CNRVia Dante, 12 – 90100 Palermo

[email protected]

Outline:-Introduction

-Photovoltaic Plants-Structure of the system-Addressed issues-Fuzzy control of PV voltage-Neural MPPT of PV plants-Neural emulation of PV plants-K-means clustering of PV data-Partial shading effect in PV plants

-Wind Plants-Structure of the system-Control techniques of IM wind generators-Neural MPPT of IM generators

-Conclusions

Introduction:Among the main research topics there are:

- Control techniques of power converters for exploiting the best dynamic performance of the sources, according to their characteristics and nonlinearities.- Design of power converter topologies suited for a multi-source input from several renewable sources.- Experimental emulation of the renewable sources to reproduce all the operating working conditions, avoiding any potential damage or risk.-Optimal exploit of the input energy source by MPPT-like (Maximum Power Point Tracking) techniques. - Connection techniques of the converter to the electrical grid.- Effective estimation of the producible energy from a PV plant.- Study of the partial shading effects in PV plants.

Photo-voltaic plant – Structure of the systemIn multi-string technology, several strings of PV modules with separate maximum power point tracking (MPPT) systems, represented by DC/DC boost converters, are connected to a common grid connected inverter.

Since every string can be controlled individually, the overall efficiency of the PV plant is increased compared to former centralized technology. Moreover the enlargement and upgrade of the plant is possible simply by plugging new strings, with their own DC/DC converters, into the existing platform.

Photo-voltaic plant – Addressed issuesThe main issues which have been faced up to are:

-Intelligent voltage control of the DC-DC boost power converter to guarantee the stability of the system in all working conditions and correspondigly achieve the best dynamic performance.-Intelligent emulation of the PV panel characteristics, including their dependence on the solar irradiation and temperature.-Optimal exploitation of the solar energy by neural mapping of the PV chacteristic integrated with a perturbe&observe technique.-Use of statistical tools in order to obtain an effective estimation of the energy produced by a photovoltaic array. -Study of partial shading effects on I-V curves in large photovoltaic fields by a reduced number of irradiance sensors.

PV plant – Fuzzy control of PV voltageThe transfer function of the averaged small-signal model of the PV array can be written as:

H (s)K0

s2 2s 2

K0 V0

nLCPV

rd RL rdLCPV

rdRLCPV L2rdLCPV

Photo-voltaic plant – ResultsTransition of the operating point from the constant voltage region at G=950 W/m2 to the constant current region at G=550 W/m2

Photo-voltaic plant – K-means clustering

0 0.2 0.4 0.6 0.8 1

1

2

3

Silhouette Value

Clu

ster

The whole experimental data set (about 12000 couples of values of current and voltage, between 100 and 1100 W/m2 and 20 and 50°C)

225 230 235 240 245 2500

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

5.5

Grid Voltage [V]

Cur

rent

[A]

tIVE i

Clusteri

iG

1

)1(82421

11

Useful experimental data subset (8242 points) for accurate energy forecast

Silhouette plot for k=3

kWhtIVE idatasampledalli 755 kWhtIVE idatasampledalli 755

kWhtIVE iCluster

i 7451

kWhtIVE iCluster

i 7451

PV plant–Kriging estimation of the shadow

G rid

G rid

G rid

G rid

G rid

G rid

1

2

3

4

5

6

0 50 100 150 200 250 300 350 4000

2

4

6

Voltage [V]C

urre

nt [A

]

0 50 100 150 200 250 300 350 4000

2

4

6

Voltage [V]

Cur

rent

[A]

Kriging based

Measurement based

UPPER STRING

LOWER STRING

Kriging based

Measurement based

Spatial shadowing distribution (3D and 2D) on the PV field

Kriging map of the shadowing with one sensor per module.

Kriging map of the shadowing with the 25% of sensors.

I-V characteristics of the upper and lower strings in array 4

Photo-voltaic plant – Neural MPPT of PV plantsThe Growing Neural Gas (GNG) neural network has been used to obtain the estimation of the maximum power point on the basis of the instantaneous measurements of voltage and current supplied by the PV source; Starting from this point a variable step P&O algorithm searches for and locks the maximum power working point

Photo-voltaic plant – ResultsIrradiance transition from 850 W/m2 to 550 W/m2.

Photo-voltaic plant – Neural emulation of PVModel based emulators have the following drawbacks: 1) model parameters should be well known, 2) the array model is highly non-linear and complex, 3) the modeling cannot be extended to other kinds of sources (Fuel Cells for example). Another approach consists in determining a numerical model of the PV array, based on experimental data processed by AI.

Photo-voltaic plant – Results

Wind plants: Structure of the systemInduction machine wind generators with back-to-back inverter topology and vector control on both the machine side and grid side inverters seem to be a very good solution to achieve high performance in controlling the electromechanical power conversion with minimum impact on the grid.

Wind plants: Neural MPPT of IM generatorsThe Growing Neural Gas neural network implements the inverse turbine model; it outputs the estimated wind speed from the actual machine torque and speed. On this basis and the knowledge of the optimal tip speed ratio, the computation of the optimal power reference speed is computed.

Wind plants: Experimental set-up

Wind plants: ResultsStep variation of the wind speed from 4 m/s to 6 m/s

Conclusions:-Several applications of artificial intelligence to distributed generators have been done.-A fuzzified adaptive PI controller for the DC-DC boost converter of the PV plant has been devised which guarantees the best dynamic performance and the stability the system in all the working conditions.-A neural MPPT technique, based on the Growing Neural Gas, has been devised which permits the PV plant work always in the MPP point, according to any solar irradiance and environmental temperature variations. This method has been further integrated with a perturbe&observe technique.-A neural PV emulator, based on the Growing Neural Gas and a controlled DC-DC buck converter, has been devised permitting to simulate experimentally the behaviour of the PV array in all working conditions.

Conclusions:-Two intelligent techniques have been devised, respectively for clustering significant data for energy production and for energy forecast in partial shading conditions. -A neural MPPT technique, based on the Growing Neural Gas, has been devised which permits the Induction Generator wind plant to work always in the MPP point, according to any variations of the wind speed.

Thank you for your attention!