A Comprehensive Review of Various Machine Learning Techniques used in Load Forecasting

Background: Load forecasting is a crucial element in power utility business load forecasting and has influenced key decision-makers in the industry to predict future energy demand with a low error percentage to supply consumers with load-shedding-free and uninterruptible power. By applying the right technique, utility companies may save millions of dollars by using load prediction with a lower proportion of inaccuracy. Aims: This study paper aims to analyse the recently published papers (using the New York Independent System Operator's database) on load forecasting and find the most optimised forecasting method for electric load forecasting. Methods: An overview of existing electric load forecasting technology with a complete examination of multiple load forecasting models and an in-depth analysis of their MAPE benefits, challenges, and influencing factors is presented. The paper reviews hybrid models created by combining two or more predictive models, each offering better performance due to their algorithm's merits. Hybrid models outperform other machine learning (ML) approaches in accurately forecasting power demand. Results: Through the study, it is understood that hybrid methods show promising features. Deep learning algorithms were also studied for long-term forecasting. Conclusion: In the future, we can extend the study by extensively studying deep learning methods.