Improving Railroad Grade Crossing Safety: Accurate Prediction of Train Arrival Times for Emergency Response Management and Driver Decision Support

2019Final ReportPDFTech ReportCTS 19-03DTRT13-G-UTC35AlgorithmsDecision support systemsEmergency managementEmergency response timeMachine learningRailroad grade crossingsRailroad safetyRailroad vehicle operationsSchedules and schedulingTravel timeUnited StatesUnited States. Dept. of Transportation. Office of the Assistant Secretary for Research and TechnologyWork, Daniel B.Wang, RenBarbour, WilliamVanderbilt UniversityUniversity of Minnesota. Center for Transportation StudiesUS Transportation CollectionIncidents at highway-rail grade crossings, locations where railroad tracks intersect surface streets at grade, are a primary driver of safety in rail transportation in the United States. Addressing safety at these locations through technology is a focus of the Federal Railroad Administration and United States Department of Transportation. Effective management of emergency response resources on the road network requires knowledge of when trains will arrive at grade crossings and temporarily disconnect emergency vehicles from parts of the community they serve. Generating estimated times of arrival (ETAs) for trains at grade crossings on a long time horizon can be used to proactively address surface transportation safety and emergency response management. This project investigates train delays to accurately estimate train arrival times at grade crossings to support in-vehicle driver alerts. The prediction of arrival times uses train-positioning information, properties of the train, properties of the network, and properties of potentially conflicting traffic on the network as input. A historical algorithm is developed to accurately model delays using train positioning information and an online algorithm integrates real-time train position information into the forecasts. Amtrak data and over two years of CSX freight rail data are used to test and validate the proposed algorithms. Results on ETA prediction are presented for various sets of input features, machine learning algorithms, and prediction locations. ETAs at control points located close to grade crossings are dramatically improved over baseline algorithms, particularly for predictions made multiple hours from a crossing that are useful for proactive safety measures.78