Machine Learning-Based Cascade Size Prediction Analysis in Power Systems

Cascading failures, although not very common, are remaining concerns in power systems, which can result in enormous electricity service interruption with massive costs for society. Extensive research has been conducted to understand these complex phenomena and mitigate their effects. With the availability of large energy data, data-driven and machine learning (ML)-based techniques have emerged to support the efforts towards cascade resilient power systems. Predicting the risk and scale of cascading failures is one of the key areas that can support other essential functions for controlling and mitigating such events. In this work, predicting cascade size after the initial triggers are modeled and analyzed through data-driven ML-based frameworks. The prediction performance of various ML techniques including Random Forest, Decision Tree, k-Nearest Neighbor, and Artificial Neural Network are evaluated at different stages of the cascade. It has been observed that the power flow data in the early stages of the cascade can be enough to achieve promising performance in predicting the risk of large cascades.