Abstract: With the power grid becomes more and more complex, the amount of features of the power flow is also getting huge. There is some limitation in the information of operation laws contained in the features of key tidal current sections selected according to experience, and it is needed to have more feature selections and safety assessments to supplement the transient stability information. Therefore, this paper proposes a feature selection WCMI based on the weighted conditional mutual information. Based on the evaluation criteria, the feature set with the highest prediction accuracy is selected to form an optimized feature set. A model interpretation based on the SHAP value is suggested to interpret the machine learning model globally and locally. Finally, the simulation analysis is carried out on the IEEE-39 bus system, and the results demonstrate the effectiveness of the proposed method.