Micro-mobility behavior resilience analysis in extreme weather events based on a knowledge-informed machine learning approach

Micro-mobility plays an increasingly vital role in urban transportation, yet research on the micro-mobility behavior resilience remains limited. This study employs shared bicycle data from Shenzhen before, during, and after extreme weather events to develop a novel framework for assessing the micro-mobility behavior resilience. First, the Soft DTW-based K-medoids time series clustering method was employed to explore dynamic spatiotemporal patterns of community response and recovery within sub-districts, traffic analysis zones (TAZs), and cell grids. This analysis revealed the emergence of three distinct patterns across various community spatial and temporal scales. Then, resilience was quantified using the resilience triangle principle, with the impact of the built environment analyzed through the GW-lightGBM model. The model using TAZs as the research unit performed the best. Finally, the study analyzed the spatially heterogeneous effects and nonlinear associations of the built environment on micro-mobility behavior resilience within the top-performing model. Socioeconomic attributes were found to have the highest average contribution to micro-mobility resilience, with road density and rooftop density also playing significant roles. The findings offer valuable insights for data-driven approaches to quantify and analyze disaster behaviors, establish resilience measurement standards, and devise planning strategies to optimize local resource allocation and cultivate highly resilient cities.