Adaptive estimation of O-D demands for an incident-induced congested freeway under ATIS environment

An integrated freeway traffic management system requires a dynamic traffic control model that operates in real-time and efficiently allocates freeway traffic diversion onto less congested arterials. The distribution of congested freeway traffic onto neighboring arterials is often a cost-effective and practical way to mitigate the effect of non-recurrent congestion on urban freeways. However, the key tasks associated with the redistribution of traffic in the network involve reliable information provision, realistic modeling of driver behavior, and projecting future traffic conditions in the network. All three of these tasks are interdependent, and the accuracy of the driver behavior model contributes to the efficiency and precision of the other two tasks. Hence, it is essential to construct a reliable, realistic, and real-time driver behavior modeling framework that can efficiently represent drivers\u27 dynamic route switching behavior. While it is difficult to capture individual travel behavior characteristics in the field using currently available technology, it is relatively easy to obtain aggregate level responses in real-time. This research proposes an aggregate level dynamic route diversion model framework that can adapt to rapidly varying traffic characteristics as well as to short term and long term changes in drivers\u27 route choice behaviors. A Bayesian updating framework was developed to update the time varying parameters of the aggregate diversion model. The proposed route diversion model updating strategy was integrated with an adaptive freeway origin-destination (O-D) flow estimation framework. The framework was applied to estimate time-dependent O-D flow matrices in a test freeway network in northern Indiana under incident induced congested situations. Data required for the models were generated using INTEGRATION, a traffic simulation software developed at Queen\u27s University, Kingston, Canada. Experimental results showed significant improvement in O-D estimates when the route diversion propensity of drivers was properly accounted for. The adaptive characteristics of the route diversion model were captured using a Dynamic Linear Model which proved to be a computationally efficient approach yielding satisfactory predictions