Research on Short-term Power Load Forecasting Based on APSO-SVR
Haoxiang Zhao, Ruichen Tang, Zongyi He · 2024
Since its introduction, the Support Vector Machine (SVM) model has been widely applied in various aspects of society. In the field of prediction, many scholars and experts have proposed the Support Vector Regression (SVR) model based on SVM specifically for regression problems, but the original SVR model has certain deficiencies. As a result, numerous experts and scholars have proposed various optimization methods to address the shortcomings of the original SVR model, achieving good applications. The SVR model often leads to overfitting or underfitting due to improper selection of penalty parameters and kernel parameters. Based on this, this paper optimizes the selection of penalty parameters and kernel parameters by using the Adaptive Particle Swarm Optimization (APSO) algorithm to obtain optimal parameters, thereby improving the SVR model. A new APSO-SVR model is proposed and applied to a case study of Romanian power output prediction. The results indicate that the combined optimized PSO-SVR has higher fitting accuracy and better prediction performance, validating the effectiveness and practicality of the PSO-SVR model and expanding the application scope of SVR.