Combination forecasting method based on the fractal dimension weight
Ji-Ran Zhu, Yuancan Xu, Hua Leng, Hai-Guo Tang, Hanyang Gong, Zhidan Zhang, Pei Ao · 2016
In order to improve the prediction accuracy, a combined forecasting method based on fractal dimension weight is proposed in this paper.Firstly, since the amount of the original data will affect the accuracy of forecasting, the three spline interpolation method is used to increase the amount of data.Secondly, historical data fitting values is obtained by the unbiased grey forecasting model, the SVM regression forecasting model and the BP neural network model.According to these fitting values, the box dimension of every single forecasting model is calculated.The box dimension normalization results are taken as the weights of single forecasting model.Finally, the results of single forecasting models are combined by using the weighted average method.Verified by an example, the proposed combined forecasting method has higher accuracy than the single forecasting models.