Bayesian Post-Fault Power System Dynamic Trajectory Prediction
Bendong Tan, Junbo Zhao · IEEE Transactions on Power Systems · 2025
Predicting post-fault dynamic trajectories is important for corrective control to ensure stable and secure power system operation. This article proposes a new Bayesian post-fault power system dynamic trajectory prediction method. Specifically, a novel Bayesian multi-output prediction model termed as stochastic variational deep kernel regressor (SVDKR), is proposed for predicting the trajectories of all generators using several time steps of measurements. An out-of-step generator detector is developed to identify in-synchronization and out-of-step generators. Subsequently, an ensemble prediction mechanism is proposed to utilize two SVDKRs for separately predicting stable and unstable scenarios. This allows correcting the predictions for in-synchronization generators within unstable scenarios, thereby achieving accurate trajectory predictions for generators. Numerical results on the IEEE 39-bus and the Illinois 200-bus power systems demonstrate the effectiveness and scalability of the proposed method.