Energia natural afluente monthly forecasting using an Artificial Neural Networks MLPs

This work presents an Artificial Neural Network multilayer perceptron model for the prediction of Energia Natural Afluente, the model is based on a cross-validation and learning method, where groups of neurons compete when the network receives a new input. The knowledge of the future behavior of the inflow flows is of vital importance to optimize the planning of the operation in hydrothermal systems to know the possible states of the future inflows, as well as the intrinsic probabilities of event, even helps the generation, distribution and commercialization companies to position in the electricity market in order to optimize their financial returns, betting on unlikely events in the “less trained” eyes. The ability of the artificial neural network models to approximate is as good as the configurations of their training, being able to lead the network to an excessive adjustment of the data, losing the capacity of generalization and compromising the prediction errors, therefore this is an element in defining the best parameters until the network is validated. The performance and acuity of the model was tested against the historical database that was used in the training of the network. The results showed that it is possible to use Artificial Neural Networks to make predictions of future affluent natural energy, and may even be applied to portfolio optimization methods for companies that commercialize electric energy in the Brazilian electricity market.