Grey model using dimensional information update technology always contains the latest information from sample data, it guarantees the accuracy of mid-long term load forecasting. Semi-parametric regression model which combines the advantages of parametric model and nonparametric model, it can fully reflect the complexity and uncertainties of the load change. This paper put forward an improved method which combines grey model with semi-parametric regression model by time-varying weight for load forecasting, the proposed method would make full use of data information and consider its inherent regularity completely, which makes prediction more realistic. At last, a comparison of the error has been made between the single model and the combination model. The test example results show that this method has higher precision.