Microscopic traffic simulation has been seeing more and more applications in the formulation and assessment of transportation policy alternatives and Intelligent Transportation System (ITS) measures. Calibration of Micro-simulation is a vital step towards a successful application because it serves as an additional check of the model's validity and ensures model parameters reflect local conditions. In this study, a five-step procedure for microsimulation calibration is established that differentiates the driving behavior and departure time and route choice (D-R choice) behavior parameters explicitly. The procedure divides the calibration problem into successive, and sometimes iterative, subproblems and solves the subproblems one by one in a much smaller scale. This paper focuses on the calibration of D-R choice behavior parameters, developing a genetic algorithm tool to optimize the parameters in D-R choice. To incorporate users' knowledge about the local network into the calibration process, the tool applies a flexibly expandable structure to increase the algorithm computation efficiency. With other algorithmic improvements, the tool has been tested in both a trial network and a medium-size southern California network. Calibration results show good convergence and better performance in terms of matching the macro measurements (link counts and trip travel times).
Abstract