An integration modeling framework for individual-scale daily mobility estimation

This paper presents a novel microsimulation framework (BayABM) that integrates population synthesis, Bayesian Networks (BNs), and activity-based models (activity-BM) to model individual heterogeneity in mobility. The population synthesis simulates a synthetic population with socioeconomic and household-level attributes. The Bayesian Network is adopted to learn and compute the probability distributions of the mobility attributes and infer the mobility that most likely arises at a given time. A hierarchical location model with a random forest is developed to refine the destination choice based on the probability obtained from BN. A total of six mobility attributes are simulated, i.e., travel frequency, travel purpose, travel destination location, travel mode, activity duration, and travel duration. The strengths of the proposed framework are: (1) it can model the individual mobility dynamics; (2) it can generate individual mobility for a large-scale population; and (3) it has broader coverage with better representativeness and less strict data requirement. The proposed model is tested with a Miami-Dade County, Florida case study. A total of 7,253,672 travels for 2,680,607 persons and 860,380 households across the county are generated from a public travel diary survey. Results show that individuals at 20-50 age and 0-20 age travel most frequently, with a total of 3,413,710 and 2,074,677 activities per day. The travel peak of a day occurs at 11:00 and 18:00. The results demonstrate the efficacy of the enhanced model, especially in large-scale applications.