A. Mellit1, M. Benghanem2, A. Hadj Arab3 and A. Guessoum4
(1,"> University Center of Medea, Institute of Engineering Sciences, Ain Dahab
26000,Algeria. Tel/Fax: (213) 25 5812 53 e-mail mellit_adel2001@yahoo. fr

(2) University of Sciences Technology Houari Boumediene (USTHB), Faculty of Electrical

Engineering, El-Alia, P. O.Box 32. Algiers 16111, Algeria
Phone/Fax: (213) 21 21 82 14, e-mail m. benghanem@caramail. com

(3) Development Center of Renewable Energy (CDER), P. O.Box 62, Bouzareah, Algiers

16000, Algeria Tel: (213) 21 91 15 03 e-mail: hadjarab@hotmail. com

(4) Ministry for the Higher Education and Scientific Research, Algiers, Algeria
Tel: (213) 21 91 11 04, e-mail: guessouma@hotmail. com

Introduction

Many authors use the feed-forward neural network networks for modeling and forecasting time series [1]. Modeling time series include the area of stochastic prediction. The optimal prediction of a signal sample (in a minimum mean square sense), give a finite number of past samples, is its conditional expectation [2], but the computation of the conditional expectation requires the knowledge of the joint probability of the current sample and the past simple, which is generally not known. Because of this the conditional expectation is in general non-linear, funding the solution is mathematically intractable signal. Therefore, the methods for designing the non-linear signal predictors are sub-optimal, and they can only attempt to approximate the conditional expectation of the current sample. These sub-optimal methods, like Markov chains [3] autoregressive (AR) [4], these methods based on simplifying statistical assumption about the measured data, which are not always satisfied. Others method based on fractal dimension give acceptable results [5]. Neural networks are widely accepted as a technology offering an alternative way to tackle complex and ill-defined problems. They can be trained to predict results from examples, and are fault tolerant in the sense that they are able to handle noisy and incomplete data. They are able to deal with non-linear problems, and once trained can perform prediction at very high speed. The neural network approach provides a good solution to such a problem, because its design is based on training and, therefore, no statistical assumption (modeling) is needed the source data. This method has been successfully applied in several areas like speech processing; image coding; forecasting sunspots prediction solar radiation data and sizing PV systems parameters [6,7]. The aim of this study is to present a new model for prediction and modeling the different signals coming from the monitoring of stand­alone PV system using the adaptive network combining between RBF network and IIR filter. Estimated signals allow to analyzing the performance of PV systems and the sizing of PV systems. Possible applications can be found in: Analyzing the performance of stand-alone PV systems; State prediction of stand-alone PV systems (1 day, 2 days,..); Sizing of stand-alone PV systems; Study of storage energy cumulated; Control of maximum power point tracker (MPPT)

()Corresponding author