Análise do efeito da temperatura na previsão de curto prazo da demanda de energia elétrica através de redes neurais

Abstract

Demand forecasting is a crucial and decisive activity for the operation of electrical power systems, thus estimating the adequate load to be supplied. In the short-term horizon, energy consumption forecasts occur from minutes, hours and even a week ahead and climatic factors that can influence demand are analyzed, thereby obtaining the best instant load profile. Several techniques are applied for load prediction, among, there are artificial neural networks, which are non-linear computational intelligence systems that operate like neural networks human brain and reproduce human characteristics, such generalization, association, learning and abstraction information. The feedforward architecture is characterized as the most appropriate model for modeling a system, among the main types of networks with this architecture is the multilayer perceptron with Backpropagation training algorithm, which has applicability in the prediction of time series and standards recognition. In this context, this work proposes a methodology to carry out demand forecast in the short-term using real energy, demand and temperature data, which demonstrate the characteristics of the demand curve profiles. Therefore, a model with artificial multilayer perceptron neural networks is being using the scaled conjugate gradient backpropagation (SCGB) training algorithm. The network is developed using MATLABR software, with Neural Network Toolbox tool (nntool) applied to create, train and present data for analysis 24 hours ahead. For the forecast system tested, the results showed the efficiency of the created ANN, and adequate forecast curves with satisfactory MSE and MAPE error values.