NCHRP Report 765 lays out guidelines for project-level travel forecasting. I was privileged to have served as the technical lead on this report; many people contributed mightily to its success, including Rob Bostrom and his team from CDMSmith, Tom Creasy, Paul Herskowitz, Mei Chen, and Ram Pendyala. All of us brought to this report many years of experience with project-level travel forecasting. We learned from each other, found common ground, and, in some cases, broke new ground. For my part, NCHRP Report 765 reflects what QRS II users have been accomplishing already and my aspirations for QRS II in the future. Many of those aspirations will be revealed in QRS II Version 9.
At the same time as I was working on the NCHRP report, I was also building several travel models for the Wisconsin Department of Transportation. Collectively known as RADIUS, these models are an ambitious attempt to create a framework for evaluating work zone traffic management strategies along I-39 and along vast stretches of I-94 and I-894 in southern Wisconsin. As part of the RADIUS effort, I agreed to make certain changes, gratis, to QRS II. Some of these have already appeared in QRS II Version 8, such as actuated signals from the 2010 HCM, a work zone diversion procedure at the assignment step, and a queue length report. RADIUS’s work zone diversion procedure found its way into NCHRP Report 765. The RADIUS 94 models were especially challenging, given their size and the limited resources available to build them. Each RADIUS 94 model contained approximately 22,000 links, 2300 zones, and 6 time intervals. The dynamic OD tables were estimated using gravity models, whose outputs were heavily refined using counts at approximately 6000 counting stations. The refinement was the largest statistical estimation problem I had ever heard of – potentially more than 30,000,000 variables to be estimated with constrained, nonlinear least squares. Some very serious changes to QRS II were required to pull this off.
NCHRP Report #765 contains many sections related to refinement. These sections convey a well-founded sense that in most cases outputs of travel forecasting models are insufficiently accurate for project-level work. As a counterargument I penned the section, “The Travel Forecasting Model Ideal”, which expresses the notion that a well-constructed travel model, perhaps with the assistance of OD table refinement, is indeed suitable for project-level work. Still another section compares the QRS II-based model from Parkersburg WV/OH to the Ohio standard, demonstrating that excellent results may be obtained with enough TLC and a good-enough modelling platform. QRS II provides sufficient firepower to implement the “Ideal”.
NCHRP Report 765 makes several references to OD table refinement. All of those sections, taken together, could be used as a primer on the technique. While the report does not endorse any of the many methods out there, QRS II’s implementation of whole table least squares is highlighted as an example. I am advising QRS II users to read those sections before attempting their own OD table refinements.
So as you might guess from this discussion, QRS II Version 9 implements the philosophy, but not necessarily the exact substance of NCHRP Report 765.
Alan Horowitz, December 27, 2014