Localization of Spatial and Temporal Parameters in Microscopic Traffic Simulation Models Based on Genetic Algorithms

Hierarchical and K-MEANS clustering methods is used to cluster historical traffic data into different clusters from space and time. After the completion of the clustering, a Genetic Algorithms (GA)-based method is used for all the time segments (temporal clusters) in each (spatial) cluster to search the optimal combination of parameters for models such as car following, lane changing, event responding and intersection delay models. The solution from GA is set as the parameters for the specified position and time segment When finished the GA-based parameter calibration for all the time segments in all clusters, the calibrated parameters are stored as a file with a space and time tag so that the system can reload such parameters while running simulations after calibration. As a result, the localization of simulation model parameters is achieved. The paper also presents a case study to further illustrate the process of model parameter localization. The result from the case study reveals that local simulation models outperform global ones.