This study examines the calibration and validation of an AIMSUN microscopic traffic simulation model of the Western Brisbane Commuting Corridor. In particular, the study focuses on the two areas of data collection/collation and comparison of model outputs with field data.
Collection of travel times for Milton Road and Coronation Drive during morning, evening and off peak conditions was performed using GPS and analyzed with the aid of the Geographic Information System Mapinfo. Achieving the ideal 90% statistical reliability in travel times for peak traffic conditions is a rigorous process, as it requires ongoing data analysis and computation of sample mean and standard deviation. Statistics on vehicle composition and vehicle attributes suitable for model input were gathered from Austroads and Australian Bureau of Statistics.
Results of travel times produced from an AIMSUN simulation model of Coronation Drive under morning peak conditions show poor correspondence with recorded travel times. Model error is most pronounced for travel times heading inbound towards the Brisbane CBD, where the simulator produces a mean trip time of less than 4 minutes compared with a observed mean trip time of 13 minutes. This disparity exists because the model considers a 50/50 directional split in traffic volumes, but collected field data indicates a strong imbalance of traffic volumes with a higher portion of commuters traveling inbound under morning peak conditions. Further research into attaining updated statistics regarding traffic volumes and travel demands is therefore recommended.
Results also show high error rates in modeling segments of road that are closer to driver destinations, and where lane changes and merges among vehicles are more common. Under congested conditions, parameters that describe driver behavior, or how each vehicle will move and interact in a network, have greater impact. In light of this, key areas of model calibration that will need to be reassessed in order for the traffic simulator to achieve closer correspondence with observed data are driver parameters and dynamic route selection.