A Causal Framework For Predictive and Explainable Analysis of Power System Cascading Failures

Analyzing cascading failures in power systems is critical, yet data-driven methods often struggle with explainability due to the intrinsic complexity of their underlying algorithms and their reliance on statistical correlation rather than true causation. To bridge this gap, this work develops a spatiotemporal framework grounded in causal inference. We construct a Structural Causal Model (SCM) that learns the genuine cause-andeffect relationships governing failure propagation using historical observations of previous cascading events. The resulting SCM serves a dual purpose: it acts as a high-fidelity predictive tool to forecast the progression of cascading events and as an explanatory engine that uses causal metrics to uncover the grid’s underlying vulnerability structure. Validation on standard IEEE test systems confirms the model’s superior predictive power and scalability. The framework achieves an accuracy of 97.61% on the IEEE 14-bus system and 84.29% on the IEEE 123-bus case, outperforming established benchmarks. The framework’s explanatory value is demonstrated by its ability to identify and characterize key buses based on their causal roles as failure sources, sinks, or intermediaries. In summary, this research contributes a transparent and robust methodology that enhances both the prediction and comprehension of cascading failures, offering a more actionable tool for improving the resilience of modern power systems.