Ensembles of Selected and Evolved Predictors using Genetic Algorithms for Time Series Prediction

This work proposes the use of Neural Networks Ensembles to predict future values of an electrical load time series. At first, to generate these ensembles it is necessary to make several predictions of the same time series using various different networks in which every single one alone is sufficiently competent to predict the above mentioned time series. Therefore, we applied Genetic Algorithms to evolve the parameters of four types of networks: MLPs Neural Networks, Recurrent Neural Networks, Radial Basis Neural Networks and Neuro-fuzzy Networks. As a result, we came up with a set of genetically evolved networks as possible candidates to compose the final ensemble. Finally, in order to achieve a better model, selections (using Genetic Algorithms) of the most suitable networks were made to compose the final ensembles.