Data-Driven Method for Rapid Prediction of Cascading Failures in Renewable-Integrated Power Systems

With the continuous development of renewable-integrated power systems, the probability of cascading failures in the power grid has significantly increased. Traditional cascading failure prediction methods struggle to handle the complexity and operating conditions of the new power grid in terms of timeliness and accuracy. Therefore, this paper proposes a Random Forest and SHAP analysis model based on real-time measurement data from the Wide Area Measurement System (WAMS). The model uses the operational state information of the power grid from cascading failure samples to quickly predict the system's cascading failure risk, demonstrating good prediction accuracy, speed, and generalization ability. Additionally, the SHAP analysis method is used to explain the prediction results, identifying which features are most important and how they influence the model's prediction performance. Test results on the IEEE 39-bus system case show that, for a sample set of 14,000, the Random Forest model achieves a prediction accuracy exceeding 90%, with a prediction time of only 0.12 seconds for the validation set. Moreover, the SHAP analysis provides a good explanation for the prediction results.