Heuristic optimization applied to ANN training for predicting renewable energy sources production

Predicting complex systems like Renewable Energy Sources depending nonlinearly on several physical parameters can be approached by a phenomenological method based on Artificial Neural Networks. This contribute shows the application of heuristic optimization algorithms to the training phase of ANN whose aim is to predict renewable power production as function of environmental variables like solar irradiance and temperature. The training problem is cast as the minimisation of a cost function whose degrees of freedom are the parameters of the neural network. A Differential Evolution algorithm is compared with the more usual gradient-based minimization procedure and the comparison of their performances is presented. In all the cases tested, Differential Evolution training process is showing results in line with those of the differential-based ones, with comparable convergence speed, always showing a more reliable behaviour, avoiding problems of premature convergence and overfitting affecting in some applications 20% of gradient-based procedures.