Predictive and Multigranularity Resilience Assessment of Urban Transportation Based on Neural Controlled Differential Equation

Crafting a dynamic and accurate resilience assessment method for urban transportation, marked by complex road networks and frequent disturbances, poses a significant challenge. Existing work mainly focuses on statically assessing historical traffic resilience and cannot dynamically divide spatial regions according to disturbance scales. In this article, we propose a predictive and multigranularity assessment method. First, we develop an attention-based spatial-temporal hypergraph neural controlled differential equation model, which can accurately predict traffic conditions under disturbances. Second, we construct a multigranularity disturbance propagation model that adaptively divides a traffic network into multiple granularities according to disturbance scales. Then, we design a real-time resilience assessment algorithm capable of quantifying spatial-temporal dynamic resilience indicators for each granularity area. Extensive experiments on urban transportation in California during heavy rainfall reveal an inverse relationship between California's resilience and rainfall intensity. In addition, its downtown exhibits strong resilience, while coastal and interior areas show relatively weaker resilience, with some interior areas experiencing prolonged recovery times.