Incorporating Big Data in an Activity-based Model for Chattanooga
Corresponding Author: Vince Bernardin, RSG
Presented By: Vince Bernardin, RSG
Abstract
After their last long range plan update, the Chattanooga TPO undertook a thoughtful review first of their data resources and needs and then of their travel model. The TPO chose to invest in several exciting new types of “Big Data” including HERE, AirSage, and ATRI in addition to traditional traffic counts and floating car travel time runs and chose to develop a new Daysim activity-based model since it cost only marginally more than the a new trip-based or hybrid model and would support additional planning features, particularly for bicycle and pedestrian planning.
Chattanooga made efforts to validate their new data against traditional data sources, comparing new HERE data to floating car GPS travel time runs, and AirSage data to traffic counts, Census journey-to-work patterns, and their previous travel survey. These investigations identified key issues and limitations of the new data, but none that could not be accounted for and corrected.
The development of the new activity-based travel demand model then incorporated passenger origin-destination (OD) data based on the AirSage data after removing truck trips based on the ATRI data. This is believed to be the first time such passively collected, anonymous “big” OD data has been incorporated in an activity-based modeling system. The process used the cell-phone based data in conjunction with data on commuting flows from the Census Bureau to develop district level origin-destination constants for inclusion in the utility functions of the spatial choice models. In this initial application, the constants were developed iteratively using shadow pricing techniques by minimizing error versus the big data sources, holding fixed the other utility function parameters originally estimated from household survey data.
The process successfully significantly improved the ability of the spatial choice models to reproduce the travel patterns observed in the big data and contributed to good overall model validation against traffic counts and transit ridership. The resulting model combines the accuracy of big data and the sensitivity of activity-based models to produce a travel model that is both grounded in a rich behavioral framework and data driven, leveraging the large sample size and relative completeness of spatial big data.