Calibration and validation of micro-simulation models using measurable variables

Abstract

Traffic micro-simulation models have been extensively used to evaluate the impacts of traffic design alternatives. These models help stakeholders make informed decisions about any new change in the transport network. Calibration and validation of micro-simulation models are essential steps to ensure the reliability of the models. Most of the existing calibration efforts focus on experimental designs of driver behavior and lane changing parameters that have no physical meaning, or at least, cannot be easily measured. This paper suggests an approach for calibrating micro-simulation models using measurable variables. We advocate using an easy-to-understand, yet appropriate measures of the traffic stream such as spot speed as a calibration parameter. The approach was applied to a suburban corridor of about 12 km located in Greater Cairo Region. Video data were collected for about 9 hours on the corridor for both directions and were used to obtain traffic volumes and composition. Spot speeds were collected using a speed gun for a sample that comprised more than 2,000 vehicles belonging to different vehicle classes. VISSIM was used to model the corridor for only one direction that encompassed uninterrupted flow conditions. Spot speeds distributions were edited extensively in VISSIM to replicate the empirical distributions of the collected data. Furthermore, the distributions were developed for each vehicle type to achieve a higher level of accuracy. Statistical tests were carried out to ensure that the modeled speed distributions match the observed distributions. Corridor travel time was used as a validation parameter. Error measurements were computed for different analysis intervals (e.g., average travel time during a particular range), and it was shown that the validation errors decrease as the analysis period increases.