Concurrent Estimation of Origin-Destination Flows and Calibration of Microscopic Traffic Simulation Parameters in a High-Performance Computing Cluster

摘要

This paper is aimed at developing an optimization framework for the concurrent calibration of demand and supply parameters in a dynamic traffic assignment (DTA) model. The proposed approach calibrates route choice, along with drivers' behavioral parameters, and estimates origin-destination (OD) flows in a large-scale network in a Paramics microscopic traffic simulation model. A mathematical formulation is defined to quantify the reliability of the observations. A genetic algorithm (GA) is selected as a suitable solution algorithm for the resulting nonlinear stochastic optimization problem. The application of the proposed methodology is implemented in the large-scale network in the business district core of downtown Toronto, Ontario, Canada. For this network, the emerging traffic surveillance data from in-vehicle navigation system technology provide an enriched source of disaggregated speed data. The empirical results from various experiments support the hypothesis that incorporating in-vehicle navigation system speed data can improve the calibration accuracy and minimize the reliance of the calibration process on a priori OD flows. The quality of the solution and convergence speed of a GA is further enhanced by dividing the GA population into multiple demes and running the GA on a high-performance computing cluster (HPCC) with multiple processors (i.e., parallel distributed GA, PDGA). In addition, this research takes a further step toward analyzing the temporal variations of the driving behavior of travelers. The case study establishes an example for modelers and practitioners who are interested in calibrating a large-scale traffic simulation model. The developed simulation model for traffic has the potential to serve as a test bed on a HPCC for more efficient computation and integration with other optimization tools such as GAs.