Abstract: The stochastic volatility of wind resources causes phase lag in numerical weather prediction （NWP） wind speed data, which in turn leads to poor accuracy of wind power forecast. Accurate wind speed prediction can improve the utilization of renewable energy and grid-connected power quality. Aiming at this problem, a combined NWP wind speed correction with nonparametric kernel density estimation and a novel variable modal decomposition-bidirectional gated recurrent unit （VMD-BGRU） combined model for wind power prediction is proposed. Firstly, the NWP wind speed error sub-dataset is corrected by non-parametric kernel density estimation and BGRU network. Secondly, the improved optimized whale algorithm is used to identify BGRU parameters, as well as decompose the wind power with rime optimized VMD. Then, the corrected wind speed and modal components are used to forecast wind power. Finally, the results of point prediction and interval prediction of the method are verified in different wind farms. It shows that prediction accuracy of the model is higher than others. Moreover, after using speed correction method, the R2 is improved by 12.56%. For interval prediction of wind power in different seasons, when the confidence level is between 95% and 75%, the PICP index is above 0.9254 and the PINAW is below 0.1068. Therefore, the model can provide more accurate confidence intervals, which can provide a reliable basis for future high-precision power distribution.