Optimising routine road maintenance through spatio-temporal deep graph clustering of pavement distress

Routine road maintenance serves as a cost-effective strategy for addressing localised pavement distress in urban road networks, helping to prevent severe deterioration while minimising traffic disruptions. However, its operational efficiency is often limited by the traditional reliance on administrative boundaries for defining maintenance zones, resulting in frequent repairs and inefficient resource allocation. This study introduces a data-driven framework that optimises routine road maintenance zoning by capturing the spatio-temporal heterogeneity of pavement distress. A fine-grained dataset is constructed, integrating detailed distress records with relevant socio-economic and environmental variables. To extract temporal patterns, dynamic time warping (DTW) and time-lag cross-correlation (TLCC) are applied, while spatial characteristics are analysed using a parameter-optimised geodetector. Based on these spatio-temporal characteristics, reinforcement graph clustering (RGC) is employed for deep graph clustering of the pavement distress, followed by a secondary clustering to finalise maintenance zones. The proposed method is validated on urban road data from Xuhui District, Shanghai. Results demonstrated a significant increase in distress coverage per maintenance operation, along with reductions in monthly inspection frequency and overall financial costs, highlighting the effectiveness of this intensive maintenance strategy.