Abstract
The traditional Travel Demand Forecasting Model (TDFM), used within the Long Range Transportation Planning (LRTP) process, has mainly focused on the evaluation of transportation system effectiveness and environmental impact with various performance measures to assess transportation investment alternatives. However, TDFM does not explicitly account for delays due to incidents that contribute non-recurrent urban congestion. Previous studies have developed ad-hoc techniques to consider incident impacts (e.g., safety studies that identify crash hotspots or pre-defined incident scenarios at the subarea level).
This dissertation research developed an approach to integrate the large amount of increasingly available incident data with a region’s TDFM. This dissertation research has explored incident data and their impacts (the number of blocked lanes, duration, etc.) on the network and shown how incident data should be prepared to be integrated into traditional TDFM networks. Known as a Travel Demand Forecasting Model with Incidents (TDFMI), the approach incorporates historical incident information (the duration and reduced capacity due to the incidents) into the corresponding links and nodes of the traditional TDFM network.
Incident impacts were accommodated in the traffic assignment step by modifying the functional form of volume delay functions (VDFs) to consider incident duration and capacity reduction. Field traffic data and crash data in Virginia DOT’s database were explored to find crash-involved traffic data by using common temporal and spatial information. The prepared crash-involved traffic data were split into subgroups by facility types to calibrate VDFs separately. The Bureau of Public Roads (BPR) and Akcelik VDFs were modified with additional variables for considering incident impacts (duration and reduced capacity) at link segments and intersections. The parameters of modified VDFs were calibrated using crash involved traffic data and application results showed better performance measures compared to the TDFM results.
The approach is demonstrated in the Hampton Roads, Virginia region. Prepared incident data were successfully matched with corresponding segments and intersections on the networks of traditional TDFM. For the base year comparisons, TDFMI offers better percent root mean square error (%RMSE) than TDFM for all facility types even without the calibration and validation of TDFMI; with larger improvements in %RMSE for higher volume groups (over 40,000 vehicles per day). Especially, TDFMI results for interstate freeways and principal arterials, and rural area showed improvements in both %RMSE and volume/count ratio.
For the future year evaluation of scenario investment, TDFMI results were evaluated by three major criteria: project utility, economic vitality, and project viability. From the three criteria, six quantitative sub-criteria, contributing 85 points out of a total 300 points, were evaluated and scored. Relative to TDFM, the applications of TDFMI to nine candidate major investments show that the TDFMI notably affected the prioritization of investments by explicitly considering each investment’s impact on incidents. While the top ranked project is unaffected, three projects changed their ranking by one position and another three projects changed their ranking by three positions. The paired t-Tests for the nine projects showed that the evaluation scores for three projects in the TDFMI were significantly different than those generated by the TDFM. These changes in prioritization demonstrate that the explicit consideration of a project’s ability to reduce incidents is feasible with TDFMI and can materially influence which investments are selected during LRTP process.
