Information-theoretic sensor network design and reliable route guidance: planning and enriching advanced traveler information systems

This dissertation aims to develop an innovative and improved paradigm for real-time large-scale traffic system estimation and mobility optimization. To fully utilize heterogeneous data sources in a complex spatial environment, this dissertation proposes an integrated and unified estimation-optimization framework capable of interpreting different types of traffic measurements into various decision-making processes. With a particular emphasis on the end-to-end travel time prediction problem, this dissertation proposes an information-theoretic sensor location model that aims to maximize information gains from a set of point, point-to-point and probe sensors in a traffic network. After thoroughly examining a number of possible measures of information gain, this dissertation selects a path travel time prediction uncertainty criterion to construct a joint sensor location and travel time estimation/prediction framework. To better measure the quality of service for ransportation systems, this dissertation investigates the path travel time reliability from two perspectives: variability and robustness. Based on calibrated travel disutility functions, the path travel time variability in this research is represented by its standard deviation in addition to the mean travel time. To handle the nonlinear and nonadditive cost functions introduced by the quadratic forms of the standard deviation term, a novel Lagrangian substitution approach is introduced to estimate the lower bound of the most reliable path solution through solving a sequence of standard shortest path problems. To recognize the asymmetrical and heavy-tailed travel time distributions, this dissertation proposes Lagrangian relaxation based iterative search algorithms for finding the absolute and percentile robust shortest paths. Moreover, this research develops a sampling-based method to dynamically construct a proxy objective function in terms of travel time observations from multiple days. Comprehensive numerical experiment results with real-world travel time measurements show that 10-20 iterations of standard shortest path algorithms for the reformulated models can offer a very small relative duality gap of about 2-6%, for both reliability measure models. This broadly-defined research has successfully addressed a number of theoretically challenging and practically important issues for building the next-generation Advanced Traveler Information Systems, and is expected to offer a rich foundation beneficial to the model and algorithmic development of sensor network design, traffic forecasting and personalized navigation.