The forecasting model based on modified SVRM and PSO penalizing Gaussian noise

摘要

The ε-insensitive loss function has no penalizing capability for white (Gaussian) noise from training series in support vector regression machine (SVRM). To overcome the disadvantage, the relation between Gaussian noise model and loss function of SVRM is studied. And then, a new loss function is proposed to penalize the Gaussian noise in this paper. Based on the proposed loss function, a new ν-SVRM, which is called g-SVRM, is put forward to deal with training set. To seek the optimal parameters of g-SVRM, an improved particle swarm optimization is also proposed. The results of application in car sale forecasts show that the forecasting approach based on the g-SVRM model is effective and feasible, the comparison between the method proposed in this paper and other ones is also given, which proves this method is better than ν-SVRM and other traditional methods.