Short-term load forecasting using a two-stage sarimax model

The primary aim of this study is to develop a new forecasting system for hourly electricity load in six Italian macro-regions. The statistical methodology is centered around a dynamic regression model in which important external predictors are included in a seasonal autoregressive integrate moving average process (sarimax). Specifically, the external variables are lagged hourly loads and calendar effects. We first use backward stepwise regression to estimate regression parameters and obtain residual series. We then identify an optimal sarima process for the residuals and, finally, the parameters of the regression and of the time series models are jointly estimated using a sarimax process selected within a small set of variants of the sarima model found for the residuals. One-day and nine-day ahead prediction performance of the proposed methodology show that intelligent integration of linear regression, time series and computational resources into a unique approach may provide accurate predictions for short-term electric loads. (c) 2017 Elsevier Ltd. All rights reserved.