A Novel Decision Support Tool to Develop Link Driving Schedules for Moves.

2015PDFTech ReportNEXTRANS Project No. 153OSUY2.2DTRT12-G-UTC05AlgorithmsDecision support systemsLinks (Networks)Nitrogen oxidesParticulatesPollutantsVehicle trajectoriesDynamic time warpingEmissions inventoriesHierarchical clusteringLInk driving schedulesMachine learningNorth AmericaUnited States. Dept. of Transportation. Research and Innovative Technology AdministrationUkkusuri, SatishPurdue University.NextransNTL-ENERGY AND ENVIRONMENT-ENERGY AND ENVIRONMENTNTL-HIGHWAY/ROAD TRANSPORTATION-HIGHWAY/ROAD TRANSPORTATIONUS Transportation CollectionA system or user level strategy that aims to reduce emissions from transportation networks requires a rigorous assessment of emissions inventory for the system to justify its effectiveness. It is important to estimate the total emissions for a transportation network before and after the implementation of a particular policy. For instance, a traffic signal control scheme that is optimized for environmental goal is expected to cause less emissions compared with the scheme without environmental goal. This research proposes a novel technique to find the representative vehicle trajectories and the corresponding LInk Driving Schedule (LDS) for links on transportation networks. The technique uses the dynamic time warping distance as the similarity measure in clustering which is more appropriate for curve alignment compared with Euclidean distances and its variants which is more common in the literature. Findings of this research are as follows: (1) The hierarchical clustering with dynamic time warping (HC-DTW) technique provides higher accuracy compared with average speed technique as seen from seven test cases. (2) The error percentage for PM10 is found to be very high in most cases. The results for PM2.5 have similar trends and are not reported here for the sake of brevity. (3) The number of links in a cluster affects the accuracy of the estimation. As the number of vehicles increases in a cluster, the degree of similarity (closeness) decreases. Although the average similarity remains the same, some details are lost. Further, estimation with a very high volume traffic adds the accumulated error. This implies that when the number of trajectories are reasonably high the analyst needs to carefully decide the number of clusters to include as input to the Motor Vehicle Emission Simulator (MOVES). (4) Except PM10, the error percentage ranges from 1% to 10% for all other pollutants. Case 2 and case 6 exhibit higher errors for NOx. Recommendations are as follows: (1) Machine learning algorithms are effective techniques to estimate emissions at the city wide level. (2) The developed tool is an effective add-on to MOVES to estimate emissions effectively. (3) HC-DTW is most effective when the variation in vehicular activities on a link is high which is typically found during peak hour conditions and in urban areas. If the congestion level is low, and the vehicular activities are similar, average speed technique provides estimation with reasonable accuracy