Project Summary
In order to design a deployable real‐world response plan for major incidents on the roadway network, prediction of alternative routes likely to be used to avoid the accident is necessary in order to adjust signal timings and ramp metering plans for the resulting changes in demand. A number of recent Integrated Corridor Management (ICM) projects rely on microsimulation to generate and test catalogs of potential response plans in order to understand and compare the impact of each response plan. The quality of results from the microsimulations depends greatly on the ability of the calibrated model to reproduce actual vehicle routes. Model calibration is extremely labor intensive, expensive, and reliant on limited data. However, the wealth of data collected via the I-210 Connected Corridors Program can be used to take a data‐driven approach to understand driver behavior, in terms of travel paths, by mining actual travel paths that emerge as a consequence of past events. There exist few empirically‐based, large‐scale studies on route choice that leverage detailed trajectories from GPS data. Studies of this type are generally severely limited in geographical scope, number of participants, and often involve only a few days of data. Researchers will employ data science techniques on years of GPS‐based point‐speed data and associated transportation infrastructure data to generate an inventory of observed driver paths in the I‐210 Corridor. This inventory will be semantically categorized to facilitate investigation of key arterial and freeway paths and to track the observed popularity of these paths. This will be the first large‐scale driver response study ‐ both temporally and geospatially ‐ to establish a consistent methodology for processing GPS‐based trajectory data that can be used to augment efforts of transportation system performance measurement, evaluation, and optimization. The work plan consists of three main tasks. First is to identify key features of the trajectories to enable classification and to distinguish data fitness for a purpose, mode, and congested or free‐flow traffic conditions. Second is to characterize large‐scale patterns in the data such as baseline activity, anomalies, locations of blockages, and potential network inefficiencies. Third is to create semantic classifications to understand routing preferences and to compare unusual patterns on known major‐incident days. Together, these three tasks will result in algorithms and code that can be implemented and run at scale over four plust years of data.