To improve the forecasting accuracy of wind power, a wind power forecasting (WPF) model based on feature cleaning combined with deep hybrid kernel extreme learning machine (DHKELM) is proposed. The Pearson correlation coefficient (PCC) method is used to optimize the features, eliminate irrelevant features, and the isolated forest (IF) algorithm is used to clean the anomalous data in the retained features, which enhances the correlation between the selected features and wind power, and then combines the DHKELM to predict the wind power. Finally, the proposed model is applied to predict the wind power of a 1MW wind turbine, and the results of the progressive forecasting of the proposed method and the forecasting results of different models are compared. The results show that the PCC-IF-DHKELM forecasting model has a root-mean-square error of 0.01928 MW, an average absolute error of 0.015254 MW, and a correlation coefficient of 0.99435, which is significantly better than the other forecasting models, proving the effectiveness of the PCC-IF-DHKELM forecasting method.