A Probabilistic Graphical Model for Predicting Cascade Failures of Electric Vehicle Charging Networks Caused by Hurricanes

Abstract

Electric vehicle charging networks (EVCNs) are vulnerable to power outages. The resilience of critical charging infrastructures is vital for providing uninterrupted service. When a hurricane occurs, the offline charging stations (CSs) may cause the redistribution of EV traffic flow, resulting in the congestion at the healthy CSs and potentially leading to a cascade failure. This paper presents a probabilistic graphical modeling (PGM) framework for predicting the hurricane-induced cascading failures of EVCNs. A Bayesian network is proposed to capture the relationships between a hurricane, the power and traffic systems coupled by an EVCN. An AC-based Cascading Failure model (ACCF) is established to simulate the line overloading and load shedding. EV load redistribution is formulated as a traffic assignment problem considering the road impedance between the nearby CSs. The proposed PGM is trained by the Expectation-Maximization (EM) algorithm considering the traffic flow as the hidden variable. Clique tree method is used to infer the propagation probability of cascading failure. The case study demonstrates the causal relationship between hazards and cascading failures of EVCNs under different hurricane scenarios. Results show that the proposed method can predict the propagation of cascading failures in EVCNs by identifying and quantifying the correlation among the susceptible CSs.