Abstract: In order to improve the accuracy of wind speed forecast in new energy power forecasting, based on the meteorological numerical forecast model, sensitivity tests were carried out using four background field data and eight parameterization schemes. Then, based on the wind speed observation data, the wind speed forecast effect of the set forecasting members at different layers height is evaluated. Based on the results, this paper establishes a multi-mode fusion model through Bayesian model averaging method and the correlation coefficient sliding average method, and obtains more accurate deterministic wind speed forecast results. The results show that the correlation coefficient of predicted wind speed can be improved by the ensemble members which join the EC background field for fusion, thereby providing a strong support for improving the forecast accuracy of the new energy power.