Cluster Analysis for a Series of Microscopic Traffic Simulation Results

Abstract

The use of traffic simulators is getting increasingly popular for the assessment of policies to reduce traffic jams. However, simulators based on multi-agent models show some variability in results even if the input data and parameters are identical, because they use probabilistic phenomena, such as lane change of vehicles, which is determined by random numbers. Results of such simulations have been evaluated and analyzed by taking the mean of several trials, but such an approach fails to account for phenomena that have a low probability of occurring, but are still possible nonetheless, and therefore appropriate decisions may not be made. This paper verifies that possible phenomena can be taken into account by the cluster analysis combing a self-organizing map (SOM) and hierarchical clustering. This study clustered traffic volume data obtained from 600 traffic simulations near Okayama Station, grouped the traffic patterns, and analyzed the results.