Models LSSVR and PLSSVR With Heteroscedastic Gaussian Noise Characteristics and Its Application for Short-Term Wind-Speed Forecasting

Abstract

Proximal least squares support vector regression is a new regression machine designed by using regularization principle technology and least squares support vector regression. In this paper, we use the above models framework to build a new regression model, called the Proximal least squares support vector regression model with Heteroscedastic Gaussian noise (PLSSVR-HGN). Based on the Heteroscedastic noise characteristics in the application field, the least square method is introduced and the regularization terms are added respectively. PLSSVR-HGN is a regression model with equality constraints based on Heteroscedasticity, which not only improves training speed and generalization ability, but also effectively improves prediction accuracy. In order to solve the parameter selection of models LSSVR-HGN and PLSSVR-HGN, the Particle swarm optimization algorithm with fast convergence speed and good robustness is selected to optimize its parameters. In order to verify the forecasting performance of LSSVR-HGN and PLSSVR-HGN, it is compared with the classical regression models on the UCI data-set and wind-speed data-set. Experimental results indicate that the proposed models not only inherit most of the merits of the original LSSVR, but also has more stable and reliable generalization performance and more accurate prediction results. These applications demonstrate the correctness and effectiveness of the proposed models.

Keywords

• Heteroscedastic Gaussian noise
• proximal least squares support vector
• regression model
• short-term wind-speed forecasting