Short-Term Power Load Forecasting Based on Feature Selection and Improved Gaussian Process Regression

In order to overcome the influence of many factors on electricity load and to adapt to the complex power system environment of today, this paper proposes a short-term forecast of electricity load based on feature selection and improved Gaussian process regression. First, the original feature data and load data were input into the Max-Relevance and Min-Redundancy algorithm for feature selection to obtain the feature set with the greatest correlation with the load and the least redundancy among the features; then the GPR model optimized by the GWO algorithm was trained with this feature set, and the trained prediction model was obtained by iteratively updating the parameters. Finally, the effectiveness and accuracy of this method were verified through actual case simulation, and the evaluation indexes of other models were compared to verify that the feature selection and hyperparameter search part of this method is beneficial to improve the prediction accuracy.