Estimation of Traffic State Transition Probabilities and Its Application to Travel Time Prediction

Real-time travel time information is important for proactive traffic management and provision of advanced traveller information systems. Although many approaches have been proposed to predict expected travel times, very few studies focus on the prediction of travel time distribution (TTD). The prediction is more challenging when the travel times have multimodal distributions that are associated with the underlying traffic states. In this case, it is important to predict the traffic state under which the stochastic travel times can be characterized accordingly. This paper develops an approach to estimate the state transition probabilities and use them to predict the downstream link TTD. The approach consists of three components, namely state definition, state transition probability estimation, and TTD prediction. The state is defined using a heuristic clustering algorithm based on Gaussian Mixture Models. The state transition probabilities are predicted as a function of link characteristics and trip conditions using a logit model. The downstream link TTD is predicted as the sum of historical link TTDs conditional on states weighted by the predicted state probabilities. The model is validated in a transit case study using an integrated database. The important factors for state transitions are traffic conditions on the previous link, traffic conditions on the current link at preceding time interval, and recurrent traffic conditions on the current link. The results show that the proposed approach provides better point and interval predictions of travel times than its alternatives.