Framework for a Real-Time Autonomous Cascading Failure Prediction Model

Blackouts cause significant damage to both consumers and utilities. Since blackouts typically start as a cascading failure, the prediction of such a cascade can effectively prevent blackouts from propagating. The majority of the current cascading failure prediction models assume that the model only needs to be trained once, either when it is designed or when the system undergoes topology changes. However, this limits the efficacy and robustness of such models. Hence, this paper aims to design a framework for autonomous cascading failure prediction models that can self-improve while being connected to the grid in real-time. To successfully achieve this, importance sampling and case-based reasoning are used to optimize the amount of data and time needed to retrain the model in real-time. The results indicate that such an approach allows the models to naturally shift to a different model as the inputs change and significantly improves the accuracy of the model as more datapoints are obtained.