Using multilayer Neural Network to increase the Prediction Accuracy: application in the Taïba Ndiaye Power Plant

The Taïba plant is a wind power plant interconnected to the SENELEC network, and its production fills a gap in electricity consumption. Indeed, it represents 157 MW of installed capacity. Like all intermittent power plants, production depends on the environmental parameters of the area where it is located. Bad weather can cause instability in the electricity grid. It is necessary to use methods to forecast its production. This will facilitate the decision making on the amount of energy to be produced to meet the demand. In this sense, this paper aims to predict wind generation by dividing the prediction data into 80% for training our model and 20% for testing the prediction efficiency of the model in order to quantify the energy produced and to allow an optimal transition between intermittent and fossil energy sources. The proposed neural network model was combined with the Levenberg-Marquardt algorithm while varying the number of hidden layers to evaluate the impact of these layers on the prediction efficiency. The rapid convergence of this algorithm and the increase in hidden layers enhanced the prediction performances. The results obtained show that varying the number of hidden layers increases the performance of neural models applied to intermittent energy prediction. The proposed approach gives its best prediction accuracy of 94.57% for the 100 hidden layer network.