Abstract:
"This study focuses on accurately forecasting daily peak electricity demand in Sri Lanka, a critical aspect of energy management. During peak hours, electricity generation relies heavily on thermal energy, which consumes substantial resources. Better forecasts allow operators to prepare power generation in advance, reducing resource wastage and operational costs.
Data from 2009 to 2018 was analyzed, incorporating external factors such as temperature, humidity, and holidays to enhance forecast accuracy. The preliminary analysis revealed that peak demand is lower on weekends and significantly reduced during New Year holidays, with holidays showing a strong correlation with lower electricity usage.
To identify the most accurate forecasting model, the study tested regression with ARIMA and SARIMA errors and LSTM recurrent neural networks. While LSTM and SARIMA with regression produced higher errors, the SARIMA (5,0,5) (2,0,1) [30] model was identified as the most effective, achieving a Mean Absolute Percentage Error (MAPE) of 1.72. Additionally, it was found that humidity has an insignificant impact on demand when temperature and calendar effects are considered.
This research demonstrates the value of incorporating external variables and advanced statistical models to enhance forecasting accuracy. The results provide actionable insights for improving resource allocation and optimizing power generation in Sri Lanka."
