The central goal of this research is to develop a systematic framework and the support tools to ease, streamline and speed up the calibration of micro simulation projects. Part III of the final report documents the accomplishments achieved in the second phase of the research project. They include the following.
First, to overcome the lengthy time it takes for GA to obtain local and global driving behavior modeling parameters, we implemented a faster heuristic optimization technique, the simultaneous perturbation stochastic approximation (SPSA) and compared its performance with other heuristic optimization methods. Results indicate that SPSA can achieve comparable calibration accuracy with much less computational time than the often used Genetic Algorithm (GA) method.
Second, we developed a much faster O-D estimation tool to obtain an initial time-dependent O-D trip table. This O-D trip table can be used as a seed table in Paramics’ own O-D estimator for further refinement, or directly used in a micro simulation. In either case, the estimation time of O-D trip tables can be considerably shortened. Since our O-D estimation tool makes use of a macroscopic traffic model (logit path flow estimator, or LPFE), a network conversion tool is therefore developed to convert Paramics’s detailed network settings to those of LPFE and vice versa.
Third, we enhanced the vehicle actuated signal control APIs in Paramics, making it more versatile to implement and simulate various types of actuated traffic control strategies found in practice. We also developed a set of guidelines to help micro simulation users to set up and check signal settings in a micro simulation project.
Finally, we developed a summary statistics tools to track, diagnose and report on the calibration as it progresses or after it terminates, and carried out a case study using the SR-41 network in Fresno to demonstrate the use of the developed tools, identify potential problems and summarizing our calibration experiences with large scale networks.
Our case study indicates that the developed calibration tools can indeed ease, streamline and speed up the calibration of micro simulation, particularly when the network concerned is large. It also reveals that the calibration of a micro simulation is a complex task that involves numerous engineering judgments and cannot be fully automated. In a micro simulation, every modeling detail matters and each must be treated properly to ensure a good simulation outcome.
