ARTERIAL NETWORK TRAVEL TIME ESTIMATION USING CONDITIONAL INDEPENDENCE GRAPHS AND STATE SPACE NEURAL NETWORKS

摘要

The travel time estimation and prediction on urban arterials is an important component of Advanced Traveler Information Systems (ATIS) and Advanced Traffic Management Systems (ATMS). This study developed robust travel time estimation models that work in both congested and non-congested conditions typically experienced on urban arterials throughout the day. The proposed travel time models explicitly account for turning movements, geometrics, and signal control settings besides traffic demand fluctuations which are lacking in current models. The result is generalized travel time estimation models for all three types of traffic movements i.e. through, left and right-turning movements. The state-space notion of traffic processes was found useful and State-Space Neural Network models are proposed. Conditional Independence graphs were utilized to identify independence and interaction between observable traffic parameters/variables that can be used to estimate travel time. Key variables were identified from among a larger group of potentially usable independent variables. The performance and computational efficiency of the Conditional Independence graphs coupled with the State-^sSpace Neural Network outperformed traditional Neural Network models. Mean absolute percentage error of modeled travel time ranged between 6.5% and 15% on testing sets.