An improved SVM-based model with peak recognition for electricity demand forecasting

For the characteristics of multi-factor and the small sample size of long-term electricity demand forecasting, a modified SVM-based model with peak recognition is considered, which can increase the weight of peak error in the loss function of structural risk minimization to adjust the penalty function. To build modified SVM model, the selection method of model parameters is analyzed and sample data is trained by SMO, so that prediction accuracy and generalized performance are greatly improved. According to the electricity demand forecasting results of Wuxi city, the analysis results show that the prediction method is reasonable and effective, with promotion application value.