Short-Term Load Forecasting Based on Bayesian Ridge Regression Coupled with an Optimal Feature Selection Technique

Abstract

Load forecasting has been an important aspect in power system operations. The increase in the integration of different renewable energy resources during past decades has made it even more crucial as an accurate load forecast can be highly beneficial for the energy market as well as the ongoing economic dispatch and unit commitment problem. The increased influence of artificial intelligence and machine learning in electrical engineering has also caused an improvement in load forecasts immensely. This study presents a short-term load forecasting methodology using Bayesian Ridge Regression paired up with an optimal feature selection technique which is a combination of Coyote Optimization Algorithm and Quadratic Discriminant Analysis. The test systems used in the study are based on the historical load data obtained from publicly available API offered by PJM data miner 2 and the weather data obtained using Visual Crossing. Before the application of the feature selection technique, the features were engineered by lagging the weather and the load data. The results of this method are compared with multiple state-of-the-art load forecasting methods including Gradient Boosting Regressor, Random Forest Regressor, Ensemble from ElasticNet, and Bagging with Decision Tree. The proposed method proved to be superior as it showed a noticeable decrease in mean absolute percentage error and root-mean-square error.