Abstract:For the traditional combination forecasting model that determines the invariable weight coefficient, the characteristics of the weighting changes of each time point are not taken into account, the method for determining the weight coefficient of the variable weight combination forecasting model is introduced. However, the weight coefficient of non-negative variable weight combination forecast model at each time could be viewed as composition data. There are some papers that apply composition data to combination forecast model. However, the existed literatures merely discussed the situation that the weight is not zero. Therefore, this paper proposes a weighting method about non-negative variable weight combination forecast model that is combined with the spherical coordinate transformation method based on composition data. Finally, in order to verify the feasibility and rationality of the model, an example about the exchange rate between America and China in the period of 2017-07-03 to 2018-05-11 is given. The numerical result shows that the method can effectively improve the prediction accuracy.