Intelligent Assessment and Enhancement of Urban Flood Resilience from the Comprehensive Perspective of Driving Force-Pressure-State-Response

Existing research has predominantly examined discrete phases of flood disasters in isolation, it has largely overlooked the interdependencies among key flood risk indicators and consequently failed to integrate these indicators holistically into flood resilience assessments. To address this limitation, this study develops a process-oriented flood resilience assessment framework that synergistically integrates natural-economic-social-hazard (NESH) model with driving force–pressure–state–response (DPSR) model, explicitly aligning both frameworks across full temporal sequence of disaster cycle. Index grid data are processed for weight prediction using multi-scale convolutional neural network (CNN), multi-head spatial attention mechanism (MHA), gated feature fusion network (GFN), and deep weight prediction head (DWPH). To reconcile subjective expert judgments with objective data-driven weights, a game-theoretic method (GTM) is employed to derive regionally optimized integrated weights. This study systematically assesses spatial distribution patterns of urban flood resilience across multiple rainfall intensity scenarios using improved multi-factor comprehensive evaluation model coupled with spatial autocorrelation analysis. Results reveal state resilience displays a distinct ring-shaped spatial pattern. In contrast, response resilience exhibits a core-periphery gradient, peaking in the city center and diminishing monotonically with distance from the core. Concurrently, the mean resilience index declines steadily—from 0.592 under low-intensity rainfall to 0.538 under high-intensity rainfall—revealing a robust attenuation trend with increasing hydrological stress. Pipe network dredging and enhanced vegetation coverage substantially reduce inundation depth and spatial extent of 5-year return period flood events; however, their risk-mitigation efficacy declines progressively under higher-intensity rainfall scenarios—particularly for events exceeding the 10-year return period. This study advances an actionable foundation for flood risk governance.