The performance of Transit Travel Time Reliability (TTR) influences service attractiveness, operating costs and system efficiency. Transit agencies have spent considerable effort on implementation of strategies related to advanced technologies capable of improving service reliability. Survey studies have shown that travelers tend to value a reduction in unreliability at least as important as a decrease in the average travel time. The increasing availability of data from automatic collection systems (e.g. automatic vehicle location, automatic fare collection, and etc.) provides opportunities in addressing transit TTR challenges. While most past studies estimate TTR for impact assessment of strategic and operational instruments, this research aims at developing generic models for TTR prediction that can fulfil different transit stakeholders’ requirements (e.g. operators, unreliability causes identification; passengers, trip and departure planning). Three main issues are addressed, namely TTR quantification, TTR modelling and Travel Time Distribution (TTD) estimation. A unique integrated data warehouse was established for case studies of this research using different sources of data across six months of a year in Southeast Queensland area, Australia.
For TTR quantification, a set of TTR measures from the perspective of passengers using the operational AVL data was proposed, considering different perceptions of TTR under different traf-fic states. The results show that the proposed measure can provide consistent TTR assessments with high-level of details, while the conventional TTR measures may give inconsistent assessments. For TTR modelling, the underlying determinants of travel time unreliability were identified and quanti-fied on links of different road types using Seemingly Unrelated Regression Equations (SURE) esti-mation to account for the cross-equation correlations across regression models caused by unobserved heterogeneity. Targeted strategies can be introduced to improve TTR under different scenarios. For TTD estimation, a novel evaluation approach was developed to assess the most appropriate probability distributions for travel time components (link running times and stop dwell times). The Gaussian Mixture Models (GMM) distribution was assessed to be superior to its alternatives, in terms of fitting accuracy, robustness and explanatory power. The correlation structures of travel time components were explored using both a global and a local correlation measures. On these basis, a generalized Markov chain model was proposed to estimate the trip TTDs for arbitrary origination-destination pairs at arbitrary times given the individual link TTDs, by considering their spatiotem-poral correlations. The proposed approach is generalizable and computationally more efficient, while it provides a comparable performance with reported models in literature.
A major contribution of the research is the establishment of a generic TTD estimation meth-odology that can be applied for a comprehensive analysis and prediction of TTR to fulfill different requirements of operators and passengers in transit. The methodology is applicable under general conditions as the link TTDs are derived conditional on the states of the current link and the transi-tion probabilities are estimated as a function of explanatory covariates using logit models. The re-sults of the research provide a better understanding on characterizing TTR from the perspective of passengers using the operational data, as well as the relationships between TTR and planning, oper-ational, and environmental factors on different types of roads. In addition, the research demonstrates the existence of multiple traffic states for a given time period and the GMM distribution can well approximate the underlying characteristics of travel times, including symmetric, asymmetric and multimodal distributions.
In practice, the proposed TTD estimation methodology provides a generic tool to analyse and predict TTR that enables transit agencies to implement strategies to improve quality of service, as well as help transit users to make smart travel decisions (e.g. fast and reliable path). Given the complexity of problems and the constraint of available data, the empirical findings on the causes of travel time unreliability and the probability distributions of travel time components are valid within the range of the used data and should be used with caution beyond this range.