Modelo Regressivo e Inteligência Artificial para Previsão de Carga no Longo Prazo: Um Estudo de Caso

Abstract

The load forecasting for an electrical system is the basis for expansion projections in generation, transmission and distribution sectors, and also for planning, operation and control. The errors minimization of load forecasting is brought with the reduction of costs and improvement of the general performance of the sectors involved, in such a way as to guarantee the quality of the services provided by the system, and to keep the operation safe and reliable. In this sense, the proposed study consists on carrying out the load forecasting of the Brazil’s southern subsystem using three regression models (neutral, optimistic and pessimistic), evaluating the performance of each model considering their trend curve and deviations (maximum and minimum) compared with load data samples. In addition, the load forecasting will be performed from a Multi-Layer Perceptron (MLP) artificial neural network considering the same load data, in order to evaluate the accuracy of the model. Afterwards, the regression model which presents the best performance will be combined with the MLP, extending the forecasting the future load with a long-term horizon, based on the current conjuncture of the Brazilian electricity sector.