Application of Neural Networks and Genetic Algorithms for Predicting the Optimal Sizing Coefficient of Photovoltaic Supply (PVS) Systems

Abstract

In literature several methodologies based on artificial intelligence techniques (neural networks, genetic algorithms and fuzzy-logic) have been proposed as alternatives to conventional techniques to solve a wide range of problems in various domains. The purpose of this work is to use neural networks and genetic algorithms for the prediction of the optimal sizing coefficient of Photovoltaic Supply (PVS) systems in remote areas when the total solar radiation data are not available. A database of total solar radiation data for 40 sites corresponding to 40 locations in Algeria, have been used to determine the iso-reliability curves of a PVS system (CA, CS) for each site. Initially, the genetic algorithm (GA) is used for determining the optimal coefficient (CAop, CSop) for each site by minimizing the optimal cost (objective function). These coefficients allow the determination of the number of PV modules and the capacity of the battery. Subsequently, a feed-forward neural network (NN) is used for the prediction of the optimal coefficient in remote areas based only on geographical coordinates; for this, 36 couples of CAop and CSop have been used for the training of the network and 4 couples have been used for testing and validation of the model. The simulation results have been analyzed and compared with classical models in order to show the importance of this methodology. The Matlab (R) Ver. 7 has been used for this simulation.