Knowledge-graph-based risk propagation modeling and optimization for cascading fault mitigation in power systems

Modern converter-dominated power systems exhibit tightly coupled electrical, protection, and control interactions that significantly amplify cascading fault vulnerability. Traditional contingency analysis and corrective control strategies primarily focus on local overload mitigation and treat fault propagation as an exogenous or threshold-triggered process, thereby limiting their ability to proactively suppress multi-layer cascade escalation. This paper proposes a unified knowledge-graph-driven optimization framework for endogenous cascading risk propagation modeling and proactive containment. A multi-relational knowledge graph is constructed to represent heterogeneous system dependencies, including physical electrical connectivity, protection coordination relationships, and control coupling interactions. Within this structure, cascading exposure is modeled as a dynamic state variable that evolves over time according to graph-induced propagation dynamics and operational stress conditions. The mitigation problem is formulated as a multi-period mixed-integer nonlinear optimization that jointly determines topology reconfiguration, converter control parameter adjustment, protection setting modification, and selective load shedding actions subject to AC power flow feasibility, stability margins, and intervention resource constraints. By embedding the risk propagation equations directly into the optimization, cascading containment becomes a proactive decision-making process rather than a reactive response. A Jacobian-based sensitivity characterization is further derived to identify hyper-coupled nodes and dominant amplification pathways, enabling targeted intervention prioritization. Case studies conducted on a modified IEEE 118-bus system with 48% inverter-based penetration demonstrate that the proposed framework reduces peak cumulative exposure by approximately 70%, confines component outages to fewer than 10 units under critical disturbance scenarios, and increases critical clearing time from below 100 ms to above 250 ms through coordinated synthetic inertia and damping tuning. The results confirm that multi-layer structural awareness and endogenous risk modeling substantially improve cascading resilience compared with conventional reactive strategies. The proposed framework provides a systematic and computationally tractable approach for cascading fault containment in modern cyber-physical power grids. Comparative experiments against security-constrained OPF, overload-based load shedding, topology-only reconfiguration, and converter-control-only mitigation are reported to separate the benefit of the integrated multi-relational graph model from individual intervention mechanisms. The study also specifies reproducibility details, including variable definitions, graph edge-construction rules, solver settings, convergence tolerances, computation time, and code availability.