Estimation of GM(1,1) Model Parameters Based on LS-SVM Algorithm and Application in Load Forecasting

In order to overcome the defects of traditional parameters estimation method in GM(1,1) model by means of least square procedure and enhance the forecasting accuracy of GM(1,1) in medium and long-term load forecasting precision, an improvement GM(1,1) model based on LS-SVM algorithm is presented. This method constructs the grey LS-SVM with background value and raw data series as the training sample according to the character of grey difference equation, converts the GM(1,1) model parameters estimation problem into a grey LS-SVM parameters estimation problem, then the regression parameters in the grey LS-SVM are solved based on the LS-SVM algorithm and the GM(1,1) model parameters estimation are also obtained. Using this method in this paper to estimate the GM (1,1) model parameters, the method follows structural risk minimization principles and is suitable for GM (1,1) modelling of small samples. This method is applied to long-term load forecasting, compared with forecasting effect analysis of traditional GM (1,1) model to prove the validity and the superiority of the model.