Dynamically Optimizing Parameters in Support Vector Regression: An Application of Electricity Load Forecasting

Abstract

This study develops a novel model, GA-SVR, for parameters optimization in support vector regression and implements this new model in a problem forecasting maximum electrical daily load. The real-valued genetic algorithm (RGA) was adapted to search the optimal parameters of support vector regression (SVR) to increase the accuracy of SVR. The proposed model was tested on a complicated electricity load forecasting competition announced on the EUNITE network. The results illustrated that the new GA-SVR model outperformed previous models. Specifically, the new GA-SVR model can successfully identify the optimal values of parameters of SVR with the lowest prediction error values, MAPE and maximum error, in electricity load forecasting.