Abstract: To improve the accuracy of ultrashort-term wind power prediction, a data-physical hybrid-driven ultrashort-term wind power prediction method is proposed. First, the ultrashort-term WPP model with bidirectional recurrent residual net-work is constructed, and the prediction results in the test set are used as the prediction template. Then, a polynomial-linear regression model is utilized to fit the wind speed-power curve of the wind farm, and the wind-power curve (WPC) is used to predict at the high fluctuation points. Finally, a dynamic switching mechanism between different models is established according to the wind speed fluctuation threshold, the template prediction value is modified according to the switching time point, and the prediction value is set to zero for the samples that the corrected wind speed is less than the cut-in wind speed. Experimental validation is carried out with data provided by a wind farm with an installed capacity of 400.5 MW in Jilin province of China, the average normalized root mean square error predicted in step 16 of the test set is 0.158 9, and the favorable switchover accounts for 90.86% of all the switches, which verify the validity and applicability of the proposed ultra-short- term wind power prediction model.