Abstract
This paper presents a high-tech solution to meet the challenges in calibrating transportation simulation models. Like any simulation software, model calibration prior to its application plays a crucial role in producing reliable results. However, transportation professionals face difficulties in performing the daunting tasks of calibrating a model for each transportation network design to satisfy the targeted traffic flow demand, especially during data collection and distillation. Our innovative approach utilizes sensor and geography networking technology to seamlessly collect data about real world network, traffic, and driver behavior. This data is then distilled as needed by data mining before feeding the data to a simulation model. The data is validated automatically to instantaneously reflect the real world and to avoid typographical errors often involved with human intervention, resulting in a more accurate model.
We conduct a feasibility study for our vision of model calibration automation. The research flexes multidisciplinary expertise in traffic flow simulation, geosciences, sensing/networking, and knowledge discovery. As a proof of concept, we implement a prototype that demonstrates how to convert sensor data about traffic flow collected by a state depaitment of transportation into a format taken.by CORSIM, a popular traffic simulation model. A running example shows encouraging results.