An Application of Load Forecasting Based on ARIMA Models and Particle Swarm Optimization

This paper presents a novel application to predict the electric load one day ahead with online implementation and self-control. Methodologically, is based on three different stages. The first one involves the selection of the best ARIMA models for each hour of the day, using a particle swarm optimization algorithm. Before this, the removal of the daily and weekly seasonality of the time series is required, as well as the outliers. The second stage is precisely the forecasting activity, and in the last one it is proposed the mechanism of control to get a more automated process. To validate the application, real data sets of two Cuban distribution networks are used. The paper also includes a detailed analysis of the load behavior to justify each step. Finally, based on the high precision of the model obtained during the validation period and the graphic results, the effectiveness of this application and its great significance for a better performance of the distribution system is demonstrated.