Power System On-line Transient Stability Prediction by Margin Indices and Random Forests

摘要

This paper addresses a novel approach for on-line transient stability prediction for power systems. In the proposed framework, the feasible instability classes (ICs) of a power system is first identified by off-line simulation considering the uncertainties of load and all potential contingencies. Accordingly, after contingencies, the stability margins (SMs) for each possible IC can be rapidly calculated using direct methods. These SMs are chosen as features for the prediction models trained by random forests, which further demonstrate a better prediction performance compared to other features used in previous machine learning based method. The proposed approach is validated on two IEEE test systems and compared to existing methods.