Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
1Overview The goal of the SHRP 2 L33 Validation of Urban Freeways project is to assess and enhance the predictive travel time reliability models developed in the SHRP 2 L03 project, Analytical Procedures for Determining the Impacts of Reliability Mitigation Strategies. SHRP 2 L03, which concluded in 2010, developed two categories of reliability models to be used for the estimation or prediction of travel time reliability within planning, programming, and systems management contexts: data-rich and data-poor models. The L33 project was tasked with (1) validating the data-rich and data-poor equations with new data sets; (2) assessing the validation outcomes to recommend potential enhancements; (3) exploring enhancements and developing a final set of predictive equations; (4) validating the enhanced models; and (5) developing a clear set of application guidelines for practitioners to use with the project outputs. The work outputs of this project include a set of recalibrated data-poor models, a set of enhanced data-rich models, and revised application guidelines for using the models. Models Data-poor, the first category of L03 models, predicts a select set of travel time reliability measures for urban freeway sections based only on the mean travel time for the section. This category of models was intended for use in locations with low availability of traffic and related types of data. The second categoryâdata-rich modelsâpredicts a similar set of travel time reliability measures based on the following input measures related to the causes of congestion and unreliability: (1) the lane-hours lost caused by incidents; (2) the hours of precipitation exceeding 1/20th of an inch; (3) the average demand-to-capacity ratio; and/or (4) the 99th-percentile demand-to-capacity ratio. In cases where one or more of these pieces of data are not available, the L03 project provides heuristic approaches to estimate them from commonly available data. Both models predict reli- ability over the course of the year. The data-poor models can be fit to any defined time period. The data-rich models estimate or predict travel time reliability within four defined time periods: peak hour, peak period, midday, and weekday. The L03 models were calibrated using data collected in metropolitan areas from around the country. They were validated using data collected in the Seattle, Washington, region. Method Both the validation and the enhancement tasks rely on the collection of traffic, incident, weather, and capacity data sets collected from a diverse set of metropolitan areas, with care taken to avoid using the same data collected in the L03 project. The L33 project collected the data sets needed Executive Summary
2for the data-rich and data-poor models in the Los Angeles, California; MinneapolisâSaint Paul, Minnesota; Sacramento, California; Salt Lake City, Utah; San Diego and San Francisco, California; and Spokane, Washington, metropolitan regions. The validation was performed by processing the collected data in accordance with L03-established methodologies and using it to compare the predicted reliability metrics with the measured ones. The validation assessed both the error of each predictive equation and whether the results met the generalized assumptions of regression modeling. The enhancement was performed in two ways: (1) recalibration of all of the L03 models using data collected in the California regions and the Minneapolis region; and (2) testing new model forms with the L33 data sets. Performance of the recalibrated and new models was measured in the same way as the validation: assessing the error of each predictive equation and whether the results met the generalized assumptions of regression modeling. The data-poor enhancement process tested the performance of three new model forms to predict the 95th-, 90th-, and 80th-percentile Travel Time Indices (TTIs): (1) a single parameter power form model; (2) a two-parameter power form model; and (3) a two-parameter polynomial model. Validation Findings The validation for the data-poor models yielded the following findings: 1. The errors for most of the models across the study regions were acceptable. 2. The models violated the assumptions of regression, particularly in that the residuals did not average to zero. From these results, it was concluded that the best way to proceed with the data-poor work was to compare the results of recalibrating the existing equations using the new data sets with the results of the performance of new model forms. The validation for the data-rich models revealed the following key insights: 1. The performance of the models varied regionally, with the highest errors measured in the California regions. Additionally, the models tended to systematically overpredict reliability in some regions and underpredict it in others. 2. The performance of the models varied by time period, with the highest errors measured during the peak period. 3. The error was highest for the equations that predict higher moments of the travel time distri- bution (such as the 95th- and 99th-percentile TTIs). 4. The models systematically violated a number of the assumptions of regression, leading the team to conclude that they may be missing one or more important variables for predicting reliability. In addition to these validation findings, the team noted that a number of the study segments, particularly those in Salt Lake City and Spokane, did not meet the L03 requirements for having a peak hour and a peak period. This suggested the need for a revision of these definitions, to ensure that all segments can get estimation results for these important time periods. From the data-rich validation results, the team concluded that the models would best be improved by enhancement, specifically through seeking additional variables to include in the equations and exploring modifications to the model functional form. Enhancement Findings Enhancement was explored for the data-rich models, but no suitable enhancements resulting in performance improvements were found.
3 The data-poor enhancement process yielded the following findings: 1. In general, the recalibrated L03 models yielded reasonable error values (measured by the mean square error), but still violated the assumptions of generalized regression. 2. The new models yielded similar or improved error values compared with the recalibrated L03 models and better satisfied the assumptions of regression. Overall, the L33 research team recommends that the SHRP 2 program adopt the new data- poor models. They allow for a consistent model form between all of the predictive equations and better capture the intuitive increasing rate of change between the mean TTI and the percentile TTI as the mean TTI increases.
