Spatiotemporal causal modeling of cascading failures in power systems
提出一个时空因果模型，从历史数据中学习故障传播的因果机制，实现前向预测、反向根因分析，并指导电网加固，比传统方法更高效。
• A spatiotemporal causal model is proposed for cascading failure analysis. • A novel counterfactual method performs robust root cause analysis of cascades. • Causal metrics guide more efficient grid hardening than traditional methods. • Causal metrics identify critical components for robust system monitoring. The interconnected nature of modern power grids makes them susceptible to cascading failures, complex events that can trigger widespread blackouts. Data-driven analysis of these events is often limited by an undue on statistical correlation, which hinders model explainability and prevents true root cause analysis. This paper proposes a spatiotemporal causal inference framework that learns the underlying cause-and-effect mechanisms of failure propagation from historical data. The proposed Structural Causal Model enables forward-looking prediction, backward-looking root cause analysis via counterfactuals, and guides grid reinforcement through causal metrics. Results on standard literature test cases validate the framework’s effectiveness. The model achieves high prediction accuracy (84.29%), outperforming several benchmarks, and correctly identifies the initial failure in over 70% of scenarios. Furthermore, a hardening strategy guided by causal metrics is shown to be significantly more efficient than traditional approaches, requiring less than half the number of reinforced components to achieve a comparable level of system reliability. The proposed framework offers a robust, explainable, and actionable methodology for understanding and mitigating cascading failures.