OWA filters and forecasting models applied to electric power load time series

Abstract

In this paper we propose a novel approach for time series forecasting based on ordered weighted averaging operators (OWA) as linear filter and forecasting models. The OWA operators describe a family of averaging operators parameterized by the choice the weights or filter coefficients. Starting with a nonstationary time series of a given phenomenon, we evaluate the use of linear decaying and constant weights as filtering processes of time series. Moreover, we investigate the effectiveness of using exponential weighted moving average process as a filter linear. After the application of the linear operators, we formulate the best possible forecasting models, ARIMA and neural network models, for short-term forecasting for each of the new structured time series using the usual optimal procedures for a real load data from the Southest Brazilian Company. A residual analysis of these forecasting models is given. In addition, the classical ARIMA and neural network models are also developed for the subject data, and the results are compared with the proposed models that we have introduced. In all cases the new models give better short-term forecasting results than the classical models.