Long-Term Electric Demand Forecasting Using a Multiple Linear Regression Model

Resumen

The continuous increase in energy demand highlights the need for its forecasting to ensure energy supply. Various methods have been used to predict this demand; however, this study proposes an approach based on a multiple linear regression model to perform energy demand forecasting. Demand energy is considered the dependent variable, while the independent variables include socioeconomic factors such as population, gross domestic product (GDP), energy production and temperature. To validate the accuracy of the model, statistical indicators such as the coefficient of determination (R2) and percentage error metrics, including the mean absolute deviation (MAD) and the mean absolute percentage error (MAPE), are applied. The proposed method is applied to the energy demand of the Peruvian national case, using historical data collected from 1995 to 2021. The results show an error of 1.7% between the historical demand used for model training (70% of the data, corresponding to 1995–2013) and the forecasted demand for the validation period (30% of the historical data, corresponding to 2014–2021), indicating a good level of accuracy in the estimates. Based on these results, long-term forecasting is performed, projecting electric demand from 2022 to 2038. The proposed methodology is useful for energy demand forecasting studies carried out by operators of national electrical power systems, such as COES in the case of Peru.