Towards Universal Urban Patterns-of-Life Simulation

TL;DR

The study tackles the challenge of modeling urban mobility in a universal, scalable manner. It introduces SimPOL, an agent-based framework where daily schedules arise from mandatory activities and evolving needs, validated against the 2017 NHTS and scalable to over $2\times 10^{7}$ agents. A normalized similarity metric demonstrates that a single parameterization reproduces key mobility patterns across multiple cities with typical similarity above 0.80, and calibration can yield further gains in flow, activity, and trip accuracy. The approach, built on Census/ACS, OpenStreetMap, and a parallelized Repast4Py implementation, supports city-scale scenario analysis for infrastructure stress testing, disaster recovery, innovation diffusion, and disease spread in urban systems.

Abstract

Understanding urban mobility requires models that capture how people interact with and navigate the built environment. We present a scalable, generalizable agent-based framework in which daily schedules emerge from the interplay between mandatory (e.g., work, school) and flexible (e.g., errands, food, leisure) activities, driven by evolving individual needs. The results of our model are validated against empirical patterns from the 2017 U.S. National Household Travel Survey, including activity distributions, origin-destination flows, and trip-chain length distributions. We introduce a normalized similarity metric to quantify agreement between simulated and empirical patterns. Most cities achieve scores above 0.80, demonstrating strong alignment without the need for city-specific calibration. The model scales efficiently to over 20 million agents, enabling full-population simulations of large metropolitan areas. This combination of universality and scalability enables scenario analysis for infrastructure stress testing, disaster recovery, innovation diffusion, and disease spread in urban systems.

Figures (4)

• Figure 1: Conceptual diagram of the agent-based model proposed here. The model’s input (a) consists of the spatial distribution of infrastructure, population composition, sociodemographic attributes, and social networks. Infrastructure includes residential buildings, where agents live and spend the night, and non-residential buildings, where agents go to work, school, restaurants, recreational places, and run errands. The model (b) assigns both mandatory and needs-driven activities to agents. Mandatory activities depend on agent type: workers, students, and homemakers. Workers must go to work, students attend school, and homemakers do not engage in either work or school activities. The model’s output (c) is a sequence of trips starting and ending at home, representing agents’ daily patterns of life. In this example, the agent leaves home for work ($T_1$), goes to a restaurant for lunch ($T_2$), returns to work ($T_3$), then after work runs an errand ($T_4$), goes to a restaurant ($T_5$), visits a recreational place ($T_6$), and finally returns home ($T_7$). Each agent is characterized by three parameters that represent the intensity of growth of their food, social, and errand needs (d). When a need intensity exceeds a threshold (set here at $1$), a trip is triggered to a building associated with that activity. Buildings are chosen from a set (e) according to a probability distribution (f) that reflects the agent’s preference for each building. In other words, each agent is assigned a predefined list of potential destinations together with a ranked preference ordering. By combining these components, the model generates detailed activity schedules for all agents, reflecting realistic patterns of life.
• Figure 2: Same set of parameters may lead to different activity patterns across cities. Simulation results showing the probability distribution of activity destinations, home, work, school, restaurant, recreation, and errands for two metropolitan areas, Minneapolis Riverside, using the same set of model parameters. Despite identical behavioral rules and parameter values in the agent-based model, the resulting activity patterns differ across cities, driven by differences in population composition (e.g., higher proportions of workers in the Minneapolis, and higher proportion of homemakers in the Riverside).
• Figure 3: Patterns of life are constrained by infrastructure configuration. The cities shown here, Hartford, Milwaukee, and San Jose, have similar population compositions in terms of the proportions of workers, students, and homemakers. This similarity was expected to produce comparable simulation results under the same set of parameters. However, the spatial layout of San Jose leads to longer home-to-work distances, which limits the number of trips agents can complete in a day and reduces the frequency of flexible activities such as recreation and errands. When a single parameter (travel speed) is adjusted, agents in San Jose are able to fit more trips into their daily schedules, thereby reproducing the expected activity frequency distribution (indicated by *).
• Figure 4: Empirical trends approximated with simulation standard parameters are refined by adjustment. Panel (a) shows the spatial distribution of the population in the Cleveland metropolitan area, where darker colors indicate higher density. Panel (b) presents the frequency of visits to different activity destinations. Results are compared across the standardized simulation, the empirical distributions from the NHTS, and an adjusted version of the simulation. Panel (c) depicts the distribution of the number of daily trips per person, again contrasting standardized, empirical, and adjusted cases. Panels (d)–(f) show trip flow diagrams that capture the movement of agents between activity types. Panel (d) corresponds to the standardized simulation, illustrating baseline patterns. Panel (e) represents empirical flows derived from the NHTS, providing real-world reference. Panel (f) shows the adjusted simulation, which better aligns with empirical observations by calibrating model parameters. Together, these panels illustrate how model calibration improves the agreement between simulated and observed mobility patterns across both activity frequencies and trip flows.