Assessment of Data Quality Needs for Use in Transportation Applications

014913892013Final ReportPDFTech ReportMAUTC-2011-01DTRT07-G-0003Travel timeTraffic dataData qualityMathematical predictionAlgorithmsHighway corridorsReal time data processingPattern recognition systemsResearch HubPattern recognitionState spaceKalman filterData and Information TechnologyHighwaysOperations and Traffic ManagementPlanning and ForecastingI72: Traffic and Transport PlanningVirginiaMarylandUnited StatesUnited States. Dept. of Transportation. Research and Innovative Technology AdministrationRakha, HeshamChen, HaoHaghani, AliSadabadi, Kaveh FarokhiVirginia Polytechnic Institute and State UniversityUniversity of Maryland (College Park, Md.)NTL-HIGHWAY/ROAD TRANSPORTATION-HIGHWAY/ROAD TRANSPORTATIONNTL-PLANNING AND POLICY-PLANNING AND POLICYUS Transportation CollectionThe objective of this project is to investigate data quality measures and how they are applied to travel time prediction. This project showcases a short\u2010term travel time prediction method that takes into account the data needs of real\u2010time applications. The objective of this research is to prepare and disseminate accurate short\u2010term (up to 15 minutes ahead) travel time predictions on major highway corridors in the state of Maryland using real\u2010time and archived Bluetooth travel time samples, probe-based INRIX data, and stationary sensor data pooled together in Regional Integrated Transportation Information System (RITIS). In addition the research effort also develops a medium-term travel time prediction algorithm using pattern recognition techniques. The algorithm is used to predict travel times between Richmond and Virginia Beach in the state of Virginia. Unlike previous studies that use travel time as the variable, the traffic state spatiotemporal evolution is used to predict traffic patterns. The approach uses traffic state data for the current day matched with a historical data set to identify similar traffic patterns and predict travel times into the future. The tasks of this study start from data collection and analysis. The raw INRIX data, including data from I-64 and I-264 between Richmond to Virginia Beach for the past three years, are used in this study. Several problems with the raw data are analyzed, including geographically inconsistent sections, irregular time intervals of data collection, and missing data. Subsequently, a travel database is constructed to obtain daily spatiotemporal traffic states in which traffic state information and dynamic travel times are included. A travel time prediction algorithm is developed using speed measurements and which fully utilizes the relationship between traffic state and travel time. INRIX data for the selected 37-mile freeway stretch (Newport News to Virginia Beach) are used to test the proposed algorithm. The testing results indicate that the proposed algorithm outperforms the other three methods including using instantaneous measurements, using a Kalman filter, and using the k-nearest-neighbor method. Moreover, the case study on the entire 95-mile freeway stretch from Richmond to Virginia Beach demonstrates the superiority of the proposed algorithm over the instantaneous approach that is currently used by Virginia Department of Transportation (VDOT). The proposed prediction method reduces the prediction error by approximately 50% compared to the current instantaneous method, especially at the shoulders of the peak periods