Time Series and Machine Learning Methods for Short-Term Load Forecasting in Modern Power Systems

Accurate load forecasting is crucial for modern power systems but faces challenges like data scarcity, tool limitations, and lack of institutional support. Inaccurate forecasts lead to resource inefficiency and inadequate electricity supply. This paper compares time series and machine learning methods for short-term load forecasting. Time series analysis uses statistical techniques to predict future values from past data, identifying trends and cycles. Machine learning algorithms, including neural networks, support vector machines, and decision trees, handle large, non-linear data relationships and improve with more data. The meta-analysis thoroughly assesses these approaches by examining multiple studies and considering aspects such as prediction performance, computational complexity, data demands, and flexibility. The detailed comparison will help the researchers understand each method's merits and demerits. Thus, knowing these details, the researchers can easily select the most appropriate forecasting model for the given problem, contributing to the enhancement of load forecasting in modern power systems.