Abstract: Rapid identification and protection of critical sections in cascading failures caused by high-risk N-k contingencies are vital for preventing large-scale blackouts. To address the issues arising from the diversity of N-k cascading fault source scenarios, the complexity of fault propagation paths, and the difficulty in defining protection strategy targets, this study proposes a critical section identification and protection model that combines Extreme Gradient Boosting (XGBoost) with Bayesian hyperparameter optimization. First a high-risk N-k fault set is constructed and the propagation of cascading faults under load rates ranging from 0.1 to 10 is simulated randomly. A dataset is generated with line load rates as inputs and residual load as the output. Subsequently, Bayesian optimization is applied to fine-tunes the hyperparameters of XGBoost model. Finlly, protection resource configuration strategies for high-risk N-k fault scenarios are identified. Simulations on the IEEE 39-bus system demonstrate that for 88% of scenarios, adjusting the power flow capacity of three critical lines can maintain residual load above 80%.

Key words: Cascading faults, Section identification, Bayesian hyperparameter optimization, Data-driven, XGBoost algorithm, Machine learning