Traditionally, traffic impacts of work zones have been assessed using planning software such as Quick Zone, custom spreadsheets, and others. These software programs generate delay, queuing, and other mobility measures but are difficult to validate due to the lack of field data necessary for validation. One alternative approach for assessing the travel time impacts of a work zone is through data mining. Historical data of travel times observed during work zones and normal conditions can be used for work zone planning and scheduling.
This project developed a prototype tool using historical data for work zones in the St. Louis region in Missouri. Data from 782 work zones on I-70, I-270, and MO 141 that occurred between January 2014 and October 2015 were used. Several data sources were utilized in this project. These included electronic alerts of work zone information such as start and end times, location, lane closure information, and travel times. Spatially, the data included the work zone segment, two upstream segments, and all segments within a 2-mile radius of the work zone. Two delay measures were used for quantifying impact of work zones on freeway segments: travel time delay based on historical average travel times for the segment and travel time delay based on historical 15th percentile travel time values.
A model was developed to estimate travel times for planned work zones at sites that may not have sufficient historical work zone data. The Random Forests technique was used to develop the model. Separate models were developed for interstate and arterial work zones using historical travel times and speed profiles, work zone and upstream segment lengths, lane closure information, and work zone schedule. The predicted travel times were then utilized to compute delays. A prototype of the data-driven traffic assessment tool was developed. The predicted travel times for both interstate and arterial work zones were within 5% error.
For demonstration purposes, the scope of the prototype was limited to two interstate corridors and one arterial corridor. The tool uses four types of input information: work zone location, roadway direction, work zone duration, and work zone type and lane closure information. The tool then uses this information to mine the historical data to identify any work zones that occurred at the same location in the past. If a match is found, the historical data is utilized to generate the expected delay measures. If a match is not found, the Random Forests prediction model is used to generate the expected delay measures.