TRAVEL TIME ESTIMATION AND PREDICTION FOR URBAN ARTERIAL ROADS

Abstract

Travel time information has been accepted as the core of advanced traveler information systems (ATIS) and advanced traffic management systems (ATMS). Providing the accurate travel time information to traffic operators and travelers allows them to make informed decisions, leading to more advantage for individual road users and the entire transportation system. Most of the traffic information providers normally deliver the current traffic conditions or current travel times to public assuming the state of traffic remains constant in the near future. Aimed at the more effective applications, short-term future traffic conditions have been proposed as a valuable piece of information in ATIS and ATMS, apart from instantaneous or estimated travel time for representing current traffic conditions. This dissertation aims at formulating the approaches for travel time estimation and short-term travel time prediction using probe data. The urban roadways in CBD area of Bangkok metropolis with highly complex and nonlinear behaviors were selected as the study corridors for confirming the applicability of the proposed techniques. First, a modified algorithm for calculating travel time and travel speed on urban roadways from high-resolution GPS probe data called “Running Speed and Stopped Delay (RSSD) method” has been proposed. This technique was modified from the average speed method using the advantage of movable sensor in which the location and speed of the tracked vehicle could be automatically detected. Secondly, for the real world application, the new analytical algorithm for allocating travel time from low-resolution GPS probe data into individual road sections by integrating instantaneous speed together with tracked locations and time stamp has been proposed. The performance of the proposed model in travel time allocation was tested and compared with the widely used technique using real field data. Results indicated that the proposed technique provided a significant improvement in travel time allocation at both complete section and intersection levels compared to the baseline technique. Thirdly, a traffic data collection system from Bluetooth MAC Scanner (BMS) was developed and the framework for constructing link travel time information from Bluetooth probe data and the preliminary analysis was also provided. Next, the short-term travel time prediction model using multilayer feedforward neural networks with the information from both target section and neighboring sections as the candidates for model inputs has been proposed. The real Bluetooth dataset obtained from BMS systems installed on urban roadway networks in Bangkok CBD was used in verifying the applicability of the proposed technique. Results indicated the proposed forecast technique was superior in traffic condition with moderate and highly fluctuated travel time profiles (CV>0.4) which could be experienced on most urban road sections.