Temporal Difference Learning Based Critical Component Identifying Method with Cascading Failure Data in Power Systems

摘要

In order to mitigate the risks of cascading failures in power systems, the critical faults that lead the system to severe load losses should be identified. Since piles of fault data containing the failure features have been stored during system operations, it is drawing increasing attention to leverage the data for criticalness analyses in recent years. This paper proposes a Temporal Difference learning (TD-learning) based method to identify critical component faults by exploring cascading failure fault chain data. The TD-learning method can allocate the final load losses to faults at different stages so that the relevance of the faults to severe blackouts can be derived. Eventually, the faults allocated larger losses are suggested to be more relevant to severe blackouts, i.e., these faults are more critical. At the same time, the precedent faults putting the critical components in jeopardy can be easily found, which uncovered the preconditions of component criticalness. The method is validated on the IEEE 14-bus system, and a considerable decrease of blackout risks is achieved by enhancing the critical components identified by this method.