Performance evaluation of Box-Jenkins and linear-regressions methods versus the study-period´s variations: Tunisian grid case

In the aim of forecasting the electricity demand in Tunisia, we have attempted in this paper to evaluate the performances of two traditional methods namely the univariate Box-Jenkins analysis (ARIMA models) and the multiple linear regressions based on economic and demographic variables (gross domestic product per capita and population). Forecasting algorithms are based on historical data period and provide results for a given future period. The evaluation of forecasting errors is calculated regarding the variation of historical data period and the future one. Forecasted results are calculated by means of the ARIMA (Autoregressive Integrated Moving Average) univariate models and their performances are compared to those of regressions models. The influence of historical data and prediction periods on forecasting performance is investigated to evaluate the minimum period that gives acceptable forecasting errors.