Forecasting Electricity Consumption in Malaysia using Holt-Winters’ Exponential Smoothing, SARIMA and FB Prophet

Abstract

Electricity consumption forecasting plays a vital role in supporting power system reliability, efficient energy planning and sustaining economic and country development. With rising demand, robust and accurate forecasting models are important to support development and decision making. This study aims to apply and compare different time series forecasting models to predict future electricity consumption in Malaysia across total, local domestics and local commercial sectors. Holt-Winters’ method, Box-Jenkins Seasonal ARIMA (SARIMA) model and FB Prophet model were used and forecast evaluation is conducted using Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The future electricity consumption was forecasted using the most accurate forecasting model. The dataset consist of 78 observations for each of the sectors and was obtained from January 2018 to June 2024 based on Department of Statistics Malaysia. (DOSM) website. During statistical analysis it revealed that each model’s outcomes are very similar to each other in terms of the fitted values are plotted closely with the actual values indicating all models able to effective capture the underlying trend and seasonality in the data. During the forecast evaluation comparison, it shows that for total electricity consumption, FB Prophet model outperforms Holt-Winters’ methods and SARIMA model with lowest MAPE, MAE and RMSE values. This outcome is also similar for the local domestic and local commercial data forecasting analysis. Hence, this model is selected as the best forecasting model with high accuracy. The use of larger datasets is recommended to further improve models’ understanding of historical patterns to deliver good and more accurate predictions.