The stochastic dynamic journey time reliability analysis by considering the spatial and temporal correlation

Abstract

Travel time, which may be the most intuitive network performance index that can be perceived by the travelers and planners, is a fundamental measure in transportation systems. Because of their importance in traffic surveillance, management and control, path planning, and routing, travel time estimation and prediction have attracted significant research interests. On the other hand, traffic networks are fragile to the demand and supply uncertainties to which they exposed, especially under incident scenarios and adverse weather conditions. Due to the importance of transportation networks in economics and daily lives of citizens, the reliability of transportation networks cannot be overemphasized. Meanwhile, travel time reliability (TTR) has been widely recognized as one of the key performance measures that describe the reliability of a transportation network. To this end, this dissertation investigates two important topics: 1) estimating and predicting the distribution of stochastic dynamic travel time for short-term planning and intelligent transportation systems (ITSs) applications; 2) evaluating the travel time reliability index based on the stochastic dynamic travel time distribution obtained previously and analyzing the relationship between the TTR and the skewness of travel time distribution. The thesis extends the definition of (deterministic) link travel time to a stochastic version by defining a kind of likelihood between the stochastic link inflow and outflow profiles. The physical meaning of the proposed likelihood is the probability that the difference between the cumulative link inflow and the cumulative link outflow be less than or equal to a prescribed bound, e.g one unit vehicle. Based on this likelihood definition, the probability mass function (PMF) of the link travel time is evaluated by defining some appropriate sampling interval. The dynamic link travel time distribution is evaluated by fitting this PMF with skew normal distribution. The dissertation then extends the deterministic nested delay operator to evaluate stochastic journey time distribution. The PMF of journey time is obtained by a series of "nested" conditional probabilities along the links on the route. By the same distribution fitting mechanism, the stochastic journey time distribution is deduced. Two empirical studies are conducted to verify the proposed method. The results prove a satisfactory performance of the proposed method for estimating and predicting stochastic dynamic travel time and its distribution. It is also been validated that the shewness analysis is consistent with the empirical observations reported in the transportation literature. The results indicate the proposed methods are adaptive to abnormal traffic conditions. To increase the accuracy of travel time prediction and to handle the abnormal traffic conditions such as traffic incidents, adverse weather conditions, the dissertation extends the SCTM to consider the temporal and spatial correlations of traffic flow for short-term traffic state prediction. Meanwhile, this dissertation also extends the methodology proposed previously to predict journey travel time distribution for TTR prediction analysis and real-time applications. To incorporate the correlations, the SCTM is expanded with a best linear predictor. This predictor is utilized to predict the inflow demand and supply functions. Historical traffic flow profiles are taken as inputs to the predictor to forecast the demand and supply functions. Meanwhile, the real-time measurement, as another input to the predictor, is utilized to correct the prediction. For real-time application, the prediction is conducted in a rolling horizon manner. The rolling horizon method is useful especially under abnormal traffic conditions, e.g. traffic incidents, adverse weather conditions. Finally, the traffic state and journey time predictions are verified by empirical studies, which prove that significant improvement can be achieved by incorporating the correlations into the SCTM framework. In conclusion, this thesis contributes to the literature on estimation and prediction of stochastic dynamic traffic state and travel time distribution, as well as the travel time reliability analysis.