Intelligent Enhancement of Branch Transient Transmission Capacity Index Criterion for Transient Stability Discrimination: Robustness Augmentation and Critical Threshold Modification

Transient stability discrimination (TSD) plays an important role in preventing large-scale cascading failures and ensuring the post-fault security of power systems. Existing machine learning (ML)-based TSD methods are always hindered by the lack of interpretability, which restricts their implementation in actual power grids. This work proposes a novel TSD scheme that integrates mechanism-induced criterion with ML techniques to address this shortcoming. Firstly, the branch transient transmission capacity index (BI) is enhanced by incorporating a high order connectivity measure, thereby improving its resilience to the missing of transient observations from phasor measurement units (PMUs). Afterwards, an adaptive localized generalization error estimation (ALGEE) algorithm is developed to evaluate the variance upper bound of the minimal BI value prediction error. Hence the real-time probabilistic distribution of BI can be generated. Finally, we utilize the risk penalty function, as well as the corresponding extreme value theorem and numerical solution supported by solid mathematical derivation and proof, to obtain the dynamic optimal critical threshold through balancing TSD accuracy and speed. The proposed TSD scheme is tested on a simplified provincial power system by which the efficacy, scalability and interpretability are demonstrated even considering incomplete PMU observations.