This paper is focused on investigating calibration issues of the Volume-Delay Functions (VDFs) for the traffic assignment step within Travel Demand Models (TDMs). VDFs, such as the Bureau of Public Roads (BPR) function, have parameters that allow the analyst to calibrate the model to local conditions. Finding the right parameters for a given TDM is challenging and there is no established guidance on how to determine them. Overall, two potential approaches can be taken: calibration based on link travel time/speed data or calibration based on link counts or observed flows over the network. It is shown that calibration based on link travel time/speed yields varying results depending on the congestion level; and consequently does not result in a consistent set of optimal VDF parameters for all traffic conditions. It is also shown that calibrating VDFs for link travel times (in macroscopic static TDMs) does not necessarily result in accurate route choice or distribution of trips over the network. A unique aspect of this research is the development of a Genetic Algorithm (GA) that can provide optimal parameters for VDFs. Specifically, for the calibration based on link counts, a GA is developed to search for the optimal set of VDF parameters while minimizing the difference between model link volumes and link counts from the field – measured in terms of root mean square error (RMSE). The GA-based calibration method is implemented on three different TDMs from Virginia. It is shown that the GA is an effective tool for calibrating VDFs as the algorithm produces results that have better RMSE values than those produced by the VDFs in current practice. The analyses are performed for three different common VDFs (i.e., BPR, Conical, and Akcelik) to investigate how these functions perform in terms of RMSE. For the three TDMs in Virginia, it is found that BPR function produced the best RMSEs. The GA algorithm developed in this study can be used as a tool to obtain optimal parameters for VDFs.
Stored only when both papers are in the corpus — 8 citations exist upstream.
Imported from OpenAlex (CC0) — nothing here is generated.