Comparing SARIMA and Prophet Models for Short-Term Electricity Load Forecasting

An accurate electricity load forecasting is very essential for people to maintain the grid stability and also to plan an efficient energy operation. This study compares two time series forecasting methods-Seasonal Autoregressive Integrated Moving Average (SARIMA) and Prophet-using hourly demand data from 2015 to mid-2020. The dataset was divided into training, validation, and testing periods to make sure robust evaluation. SARIMA is interpretability, so it served as a traditional benchmark, and Prophet is more flexible in modelling multiple seasonality and holiday effects. The results show that comparing to SARIMA, the Prophet has a low MAE, RMSE, and MAPE values, which indicated that Prophet has a higher predictive accuracy and stability. When adding the holiday regressors in Prophet, it produced negligible improvement, which suggesting that weekly cycles captured most of the variation. The residual diagnostics also confirmed that Prophet’s residuals were closer to white noise. These finding demonstrate that the Prophet’s superior adaptability to the real-world electricity load data. and also highlight that it is important to combine statistical interpretability with model flexibility for future forecasting research.