An Improved XGBoost-based Transient Stability Assessment Method for Power System

This study introduces an improved XGBoost-based algorithm, which enhances the XGBoost algorithm by including the EFL loss function to handle the problem of sample imbalance of the transient stability assessment in power system. First, the XGBoost model is strengthened by the addition of the Enhanced Focal Loss (ELF) function, which significantly lessens the negative effects of the imbalance problem between stable and unstable samples on learning. The EFL-XGBoost is then trained utilizing the power system’s operational data during the elimination of the fault issue. The simulation results on the WEPRI-36 node system demonstrate that, in comparison to other machine learning algorithms, the EFL-XGBoost algorithm has better robustness in the case of sample imbalance and has a higher recognition rate for the unstable samples while maintaining a higher overall evaluation accuracy; compared with the Focal Loss function, Enhanced Focal Loss function more significantly improves the recall ratio of model prediction; and compared with the XGBoost algorithm, the EFL-XGBoost has significantly improved the ability of identifying difficult samples.