Enhancing neural network based load forecasting via preprocessing.

Abstract

The importance of Short-Term Load Forecasting (STLF) has increased, lately. With deregulation and competition, energy price forecasting has become a big business. Load bus forecasting is essential for feeding the analytical methods used for determining energy prices. The variability and nonstationarity of loads are getting worse due to the dynamics of energy tariffs. Besides, the number of nodal loads to be predicted does not allow frequent interventions from load forecasting specialists. More autonomous load predictors are needed in the new competitive scenario. Despite the success of neural network based STLF, techniques for preprocessing the load data have been overlooked. In this paper, different techniques for preprocessing a load series have been investigated. The main goal is to induce stationarity and to emphasize the relevant features of the series in order to produce more robust load forecasters. One year of load data from a Brazilian electric utility has been used to validate the proposed