Optimizing Arterial Coordination Using Trajectory Data

Roger Lloret,Minjun Liu,Jianfeng (Jeff) Zheng,Henry X. Liu,Shengbo Eben Li,Keqiang Li

DOI: https://doi.org/10.2139/ssrn.3983111

2021-01-01

SSRN Electronic Journal

Abstract:Trajectory data from connected vehicles is becoming increasing available, such as GPS data from Transportation Network Companies (TNC). With TNC trajectory data, a sample of vehicle delays, queues and stops is measured directly. Therefore, this data is particularly useful for enhancing traffic control on signalized arterials. Existing efforts on utilizing trajectory data mainly focus on traffic state estimation and performance evaluation, while applications in control optimization are under-explored. In this work, we propose two optimization models for arterial coordination using trajectory data. The first model optimizes a weighted bandwidth consisting of link bandwidth, contiguous two-link bandwidths and the full-arterial bandwidths, so that the optimized bandwidth not only can accommodates through traffic, but also traffic with different Origin-Destination (OD) pairs along the arterial. Extending the first model, the second model further optimizes time-variant bandwidth within green band considering traffic arrival patterns. Therefore, better coordination efficiency can be achieved. In contrast with existing bandwidth optimization models which typically require a common cycle length, these two models can accommodate coordination with different cycle lengths along an arterial. In this way, coordination can be designed with higher flexibility. The optimizations are formulated as a Mixed Integer Linear Problem (MILP). Thus, to solve the problem efficiently, we develop heuristic solution algorithms based on genetic algorithm and dynamic programming, which reduce the time complexity from exponential to polynomial. Both models have been extensively tested in real world using DiDi trajectory data. Field experiments demonstrate promising results, with considerable reduction of delays and stops.