Summary: Understanding human mobility patterns is vital for the coordinated development of cities in urban agglomerations. Existing mobility models can capture single-scale travel behavior within or between cities, but the unified modeling of multi-scale human mobility in urban agglomerations is still analytically and computationally intractable. In this study, by simulating people’s mental representations of physical space, we decompose and model the human travel choice process as a cascaded multi-class classification problem. Our multi-scale unified model, built upon cascaded deep neural networks, can predict human mobility in world-class urban agglomerations with thousands of regions. By incorporating individual memory features and population attractiveness features extracted by a graph generative adversarial network, our model can simultaneously predict multi-scale individual and population mobility patterns within urban agglomerations. Our model serves as an exemplar framework for reproducing universal-scale laws of human mobility across various spatial scales, providing vital decision support for urban settings of urban agglomerations. The bigger picture: As urban areas develop, neighboring cities gradually converge to form highly integrated urban spatial forms through a process known as urban agglomeration. Within urban agglomerations, travel occurs at different spatial scales, for example, within local neighborhoods, within cities, or between cities. This makes understanding human mobility in urban agglomerations inherently complex. While various models have been developed in the past to describe human mobility, they generally cannot model and predict the complex multi-scale travel that occurs within urban agglomerations. Methods that can better model mobility in complex urban agglomerations could have significant practical implications for topics such as urban resource management, disease control, and transportation hub optimization.
Full Text Finder