Electricity plays a critical role in energy sustainability. Accurate electricity demand forecasting supports achieving energy sustainability in Sri Lanka by enabling more effective planning and management of both renewable and non-renewable energy sources, which are required to generate the electricity. Thus, this study determines the best statistical and deep learning models for short-term electricity demand forecasting using a 4-year time series dataset, provided by the Ceylon Electricity Board of Sri Lanka (Jan 2020–Mar 2024). Initially, the Linear Regression, Polynomial Regression, and Fast Fourier Transform methods used to develop a baseline model by comparing the error rates of their predictions across different sequence lengths. Subsequently, the study proposes the use of Multilayer Perceptron and Long Short-Term Memory (LSTM) as deep learning methods to develop better predictive models for next-day electricity demand. The prediction accuracy of these two models was assessed using key performance metrics, including Mean Absolute Percentage Error, Mean Absolute Error, and Root Mean Square Error. Finally, the performance metrics of each deep learning model were compared against those of the baseline model. The findings show that the LSTM method is very effective for predicting electricity demand. It works well with the dataset and gives the lowest error values for all performance metrics. The final demand forecasting model contributes to smarter grid development, enhances renewable energy integration, and supports energy sustainability by enabling a more energy-efficient future.
Shiwakoti et al. (Wed,) studied this question.