Abstract
Synthetic population generator is the core component of the microsimulation in activity-based travel demand model. Typically, synthetic population is used in the way that their decisions on activity-travel pattern are simulated. Traditionally, household sample survey data is used to synthesize the population. The estimated results can be biased due to such as low-sampling size and inaccurate household sample data. To deal with this issue, a statistical maximum-likelihood method to calibrate synthetic population using the roadside observations (link counts) is proposed. Statistical performances of the proposed method are evaluated on the illustrative network and real network with census and household sample survey data. Multiday link counts are simulated from (true) activity-based model parameters and synthetic population. Tests are carried out assuming different number of observations and observation variations. The results illustrate the efficiency of the model calibration based on link counts and its potential for large and complex applications.
