Aplicación de La Red Neuronal Artificial Feedforward Backpropagation para la predicción de demanda de energía eléctrica en la Empresa Eléctrica Riobamba S.A.

Abstract

This research proposes a model based on the artificial neural network Feedforward Back propagation capable of predicting the demand of electric power with a percentage of absolute error lower than the one generated due to the methodology used by a distributor, with which it is intended to contribute to the planning of operation and maintenance of power plants and at the same time to serve as a model for other institutions with similar characteristics. Field observations was performed on the substations and measurers of the three outputs where 70128 observations were obtained, of which 61344 were used for the training of the network and 8784 for tests of the model. Through pre-processing of data it was detected and corrected 406 lost data and 320 atypical data, which the majority correspond to the year 2009 and 2014, it was determined that the percentage of mean absolute error (MAPE) of the prediction model of the demand electrical energy based on the neural network Feedforward Back propagation was 2.63%, while the one based on multiple linear regression was 4.56%. It is concluded that the prediction model of the demand for electrical energy based on the neural network FeedForward Backpropagation holds better prediction performance. Before designing a neural, it is recommended to perform the preprocessing of data to correct atypical data, lost and softened of the time series in order to obtain

satisfactory results.