Estimation of dynamic origin destination matrix: a genetic algorithm approach

Dynamic origin-destination (O-D) matrix estimation is one of the key components in the deployment of microscopic traffic simulation based real-time traffic predictions and estimations. Various theoretical methods have been proposed and tested via relatively small-scale networks. Very few practical studies have attempted to evaluate the performance of dynamic O-D matrix estimation methods for large-scale networks. This is because practical applications have not yet adopted dynamic O-D matrix estimation method, in part, due to the complexity and time requirements of advanced methods. This paper investigates the application of dynamic O-D matrix estimation methods for a large-scale network using a genetic algorithm (GA). The performance of GA-based method was compared with that of the QUEENSOD method using a microscopic traffic simulation program, PARAMICS. The evaluation results indicate that the GA-based method outperforms the QUEENSOD method.