Due to serious wind power network security such as the randomness and suddenness of wind speed, and the current wind speed forecasting cannot meet the industrial demand, it is urgent to realize the influence of wind speed forecasting. In this article, the short-term forecast of wind speed is revised. The structure and prediction method, combined with historical NWP (numerical weather forecast) wind speed and historical measured wind speed and other related data, established an online modification model for short-term wind speed forecasting. The model uses historical NWP wind speed as model input data, and historical measured wind speed as output data. First, train the model in an offline environment to test the effect of the model; secondly, train the model in an online environment and modify the model dynamically; finally, get the optimal short-term NWP wind speed forecast modification model. Using the established modification model, the historical NWP data of a wind spot in North China was modified. Compared with the original NWP data, the accuracy was improved by 0.861 m/s, which proved the effectiveness of the NWP modification method. At the same time, it proves the effectiveness of the online model and reduces the model’s dependence on historical data.