Abstract:

A multi-stage gray support vector machine ensemble prediction model (Dm_GM(1,1)SVM) is presented by analyzing the gray model GM (1,1) and the support vector machine model (SVM).The prediction accuracy of gray model is improved through multistage buffer operators to predict various relevant indicators.Meanwhile, the particle swarm optimization algorithm is used to find the optimal parameters of the support vector machine model, which include RBF kernel parameters and penalty parameters and the optimal pair is (c, g).Thus,the optimal support vector machine regression model is determined. Finally the final output value is predicted by inputting the predictive value of each indicator to support the vector machine model.The results show that Dm_GM (1,1)SVM has a higher prediction accuracy compared with the gray prediction model and the BP neural network prediction model in this case,and that multi-stage gray forecasting model combined with support vector machine model has a practical value in solving practical prediction problems.

Key words: multi-stage gray prediction model;support vector machine;integrated forecasting;buffer operator;particle swarm optimization