Using Automatic Calibration with Microscopic Traffic Simulation
The microscopic simulation is getting increasingly common in traffic planning and research because of the detailed analysis it can provide. The drawback of this development is that the calibration and validation of such a detailed simulation model can be very tedious. This paper summarizes the research on automatic calibration of a high-fidelity micro-simulation (HUTSIM) at the Helsinki University of Technology (TKK). In this research we used ramp operation as the case study. The automatic calibration of a detailed model requires a systematic approach. A key issue is the error-function, which provides a numeric value to the distance between simulated and measured results. Here we define the distance as combination of three distributions namely the speed distribution, gap distribution and lane distribution. We developed an automated environment that handles all the necessary operations. The system organizes the files, executes the simulations, evaluates the error and generates new parameter combinations. For searching of the parameter space we used a genetic algorithm (GA). The overall results of the research were good demonstrating the potential of using automatic processes in both calibration and validation of simulation models.