Optimization methodology for the calibration of transportation network micro-simulation models

Traffic micro-simulation models have been used extensively because of their ability to model the dynamic and stochastic nature of transportation systems. The proper selection of the input parameters is critical if the model is to replicate accurately the supply characteristics, demand characteristics, and their interaction. While the use of traffic micro-simulation models has become widespread, there has been little research on the methodology for calibrating the parameter set. The main reasons for this are a paucity of available data, the cost of data collection, and a lack of systematic calibration methodology.; An automatic calibration methodology for micro-simulation models was developed in order to select the “best” parameter set based on observed intelligent transportation system (ITS) data. TRANSIMS and CORSIM models were employed because they both capture the detail of individual vehicles but are based on different traffic flow theories, namely cellular automata and car-following theories, respectively. The methodology was applied to two components of the transportation system, namely a limited-access highway and a signalized diamond interchange. Three highway corridors were used as test beds: Interstate 10 and US 290 in Houston, Texas, and Interstate I-37, in San Antonio, Texas. In addition, a diamond interchange in College Station, Texas, was used.; A genetic algorithm and a sequential simplex algorithm were examined for use in the optimization component. They were successful in that both identified similar parameter sets. Three origin-destination (OD) matrices were evaluated as input parameters in the calibration process and the OD information was found to have a profound effect on the simulation results. A bi-level programming technique was applied in order to calibrate the parameter set and OD matrix simultaneously. A sensitivity analysis was performed on the parameters of traffic micro-simulations, the test beds, traffic congestion levels, OD estimates, and random seed numbers in order to ascertain the effects on the results. It was also found that, while TRANSIMS is a low-fidelity model and CORSIM is a high-fidelity model, both models could replicate the baseline traffic data to the same level of accuracy for all components of the transportation system tested.