Stochastic Multimodal Network Modeling: Hidden Markov Model Based Synthetic Population Generation for Use in Microsimulation Models of Transit Systems

University Transportation Centers Program2018PDFTech ReportKonduri, Karthik CMondal, AmitUniversity of Connecticut. Department of Civil and Environmental EngineeringUniversity of North Carolina at Charlotte. Center for Advanced Multimodal Mobility Solutions and EducationUnited States. Department of Transportation. University Transportation Centers (UTC) ProgramUnited States. Department of Transportation. Office of the Assistant Secretary for Research and TechnologyUniversity of North Carolina at Charlotte. Center for Advanced Multimodal Mobility Solutions and EducationUnited StatesConnecticutCase studiesMicrosimulationPublic transitHidden Markov modelsMultimodal networksStochastic modelsSynthetic populationsProject ID: 2017 Project 0569A3551747133FinalIn order to apply microsimulation-based models of land use and travel demand, socio-economic and demographic attributes about all individuals in a region are required. This disaggregate level information is not readily available and people resort to population synthesis procedures. These procedures combine readily available information in the form of sample data and marginal distributions to generate the required inputs. In this study, a simulation-based technique for population synthesis using a Hidden Markov Model (HMM) framework is presented. An important feature of the proposed approach is the ability to generate more heterogeneous synthetic households and persons. The proposed simulation-based approach is demonstrated using a case study for Connecticut. Synthetic population is generated for two block groups in Connecticut under alternate configurations. A comparative analysis is carried out to highlight the feasibility and applicability of the proposed approach in generating consistent multilevel agents while adhering to geography-based heterogeneity. The current work is similar in spirit to other recent simulation-based generators, however, there are two important contributions. First, a hierarchical transition structure is proposed in the HMM-based model, to capture the dependencies across household and person-level attributes. Thus, the procedure ensures that both household and person level attributes are controlled simultaneously. Second, the transition matrices are estimated at the geography level incorporating the sample as well as marginal information available. This helps synthesize populations that are more accurate and consistent with available information.143