Comparative Analysis of Short-Term Load Forecasting Using Machine Learning Techniques

Short-Term Load Forecasting is essential in estimating future energy demand in power systems, energy utilities, and industrial settings. For energy suppliers and other players in the markets for electric energy generation, transmission, and distribution, load forecasting is a crucial instrument. Additionally, the prediction of load is essential for effectively planning and overseeing power system operations. Load forecasting has significant effects on a variety of power system applications, such as energy production, load shedding, contract analysis, and infrastructure construction. In this study, the authors compare and contrast three forecasting methods for short-term load forecasting namely, Gradient Boosting (referred to as GB), Random Forest (referred to as RF), and K-Nearest Neighbors (abbreviated as KNN). The historical load (annual) and New York calendar data are input parameters into the forecasting models for each of these strategies. The research assesses how well these methods perform in terms of Root Mean Square Error (RMSE), Mean Absolute Error, R2-score, and computing time. The comparative analysis finds that KNN is the most effective option, with an amazing R2-score of 98.43%, followed by RF at 97.12% and GB at 95.52%. Furthermore, KNN offers tremendous computing efficiency, emphasizing its suitability for real-time applications. This work lays the door for improved load forecasting in dynamic energy systems by using the capabilities of these models.