Abstract: Wind power has strong volatility and randomness, and the changes in unit monitoring data are complex. In order to improve the accuracy of wind power prediction, a wind power prediction feature selection method based on multi-information metrics fusion (MIMF) is proposed. Based on the analysis of three typical feature selection methods, including decision tree, L₁-regularization, and recursive feature elimination, it was determined that while decision trees offer clarity in expressing feature importance, L₁-regularization mitigates overfitting, and recursive feature elimination accounts for inter-feature correlations. Leveraging the strengths of these methods, a composite feature selection framework was devised, wherein the union of features selected by each technique is further refined based on their respective intercorrelations. Subsequently, the constructed feature selection method fused with multi-information metrics is simulated and analyzed. The simulation results on the measured data of a wind farm show that compared with the single feature selection method, this method can effectively improve the prediction accuracy of wind power.

Key words: feature selection, information fusion, correlation, power prediction