Abstract Manager

Travel demand models, whether 4-step or Agent-Based (ABMs) were traditionally based on special travel surveys such as Household Travel Surveys (HTS’s) that served as the main source for estimation of the model coefficients. Other types of data such as transit on-board surveys, traffic counts, transit ridership by line, etc., were primarily used for model validation and possible calibration of the coefficients.

With the ABMs becoming more common, one of the frequently expressed concerns is that large-scale HTS’s have become increasingly expensive and difficult to recruit. Moreover, the process of conducting the household travel survey and making them ready for model development can sometimes take over 1-2 years. Due to such a long process, travel models cannot be updated on a frequent basis and many of them have not been updated post pandemic.

A frequently suggested alternative option is to utilize passive datasets as a replacement for model development. Big data, traffic counts, transit ridership data are already available post pandemic. The obvious drawback of these passive datasets is that they are not behavioral (no detailed trips purpose or individual attributes are available) and the available trip tables do not have a person ID for an identification of tours or person daily activity patterns.

The reality of transportation industry pushes model developers to minimize the need for large local HTS’s and at the same time take full advantage of passive datasets in model calibration. The central question in this regard is how big data or traffic counts could be effectively used for calibration on mode choice or tour/trip frequency or car ownership of disaggregate models.

The paper shows a new general approach where different types of data including HTS, big data, traffic counts, and transit ridership can be used in one systematic process of travel model calibration in an automated manner using Machine Learning techniques. The paper aims to illustrate the approach using the Person Travel Model of the City of Edmonton. Currently the City of Edmonton is beginning to conduct a multiple-year continuous household travel survey to better understand post-pandemic travel behavior. However, the survey cannot be used for model calibration and validation as the first-year limited samples are not sufficient to fully understand Edmontonian's travel behavior. On the other hand, post-pandemic traffic and transit counts were already available and could be used for model calibration in an automated manner. This paper shows how the traffic counts and transit ridership data were used in an automated calibration of the key components of Person Travel Model of the City of Edmonton to better reflect post-pandemic travel patterns including hybrid work programs.