Real-time Traffic Management Strategy Generation: Framework, Application and Performanc
COM2 Level 4
Executive Classroom, COM2-04-02
Abstract:
The importance of real-time traffic management strategies in modern transportation networks is well established especially when the network has evolved and become increasingly complicated. The advance of information and sensing technology has resulted in the emergence of Advanced Traveler Information System (ATIS) which provides a large amount of traffic data containing speed, volume, density, etc. The usage of this rich source of data then gives rise to a challenging question of how to extract the important information from the data and turn it into a series of strategic actions in order to improve the traffic condition, save cost, enrich travelers' experience and increase their level of satisfaction.
In this thesis, we focus on three perspectives of the aforementioned problem: framework, application and performance. First, we work on a simulation-based framework for traffic management strategy generation. The framework utilizes the power of both microscopic traffic simulation and mesoscopic traffic simulation in order to create a closed feedback loop between the model and the system and at the same time support scenario creation, strategy evaluation, strategy simulation, strategy optimization and the usage of predictive information. The proposed framework is also versatile in supporting different types of Intelligent Transport Systems (ITS), information used, objective functions and optimization methodologies.
Second, we tackle the problem of road pricing strategy generation covering both the existing area-based road pricing system and the new distance-based road pricing system. Our approach utilizes the proposed framework for strategy generation with more advanced strategy generation component in order to generate dynamic, predictive, optimal strategies in real-time which are consistent with drivers' behavior. A synthesis study between area-based and distance-based toll strategies is also conducted in order to have a better insight into the economics of pricing and the effectiveness of different pricing policies in traffic management. This study is extremely meaningful in the context of Singapore since area-based road pricing will be replaced by distance-based road pricing in near future.
Third, we investigate an approach to improve the performance of the traffic management strategy generation in order to meet real-time requirements. Our approach aims to speed up the two most time-consuming processes of the strategy generation: the strategy optimization and the traffic simulation. The strategy optimization is accelerated by utilizing parallel computing in order to execute multiple What-If simulation runs in parallel while the traffic simulation is accelerated by taking advantage of GPU technology in order to massively parallelize and speed up both the demand and the supply models of mesoscopic traffic simulation without compromising its realistic modeling and correctness. The power of GPU is even further harnessed by employing novel data structures and GPU optimization techniques to improve memory access patterns and execution efficiency.
With the above approach, we aim to fill in the gaps and solve the issues of existing traffic management strategy generation research and lay a foundation for future work in this area.