The grey composite prediction based on support vector regression

The prediction effect of GM(l,n) model is not always satisfied. The known correction methods of residual errors either need preprocess the error data to satisfy specific conditions such as non-negative, quasi-exponential law or require much more data to the train sample. Firstly, the paper improves the traditional accumulated generating operation and provides a kind of Increase accumulated generating operation (IAGO) which generates the required data sequence without high order AGO. Then, the paper proposes a kind of grey composite prediction method based on SVR where GM(1,1) model is used to predict and SVR makes the correction for the GM(l,l)´s prediction results. This method synthetically utilizes the merits of the grey system theory and SVR and thus has higher prediction precision. Especially, the paper provides a heuristic arithmetic of how to ascertain the increase coefficients and obtain the prediction values. Finally, the method is used for the medium-term or long-term forecast of regional economy and displays good application effect.