A group-based traffic signal control with adaptive learning ability

Abstract

A group-based traffic signal control with adaptive learning ability. Engineering applications of artificial intelligenceAccess to the published version may require subscription. N.B. When citing this work, cite the original published paper. AbstractGroup-based control is an advanced traffic signal strategy capable of dynamically generating phase sequences at intersection. Combined with the phasing scheme, vehicle actuated timing is often adopted to respond to the detected traffic. However, the parameters of a signal controller are often predetermined in practice, and the control performance may suffer from deterioration when dealing with highly fluctuating traffic demand. This study proposes a groupbased signal control approach capable of making decisions based on its understanding of traffic conditions at the intersection level. In particular, the control problem is formulated using a framework of stochastic optimal control for multi-agent system in which each signal group is modeled as an intelligent agent. The agents learn how to react to traffic environment and make optimal timing decisions according to the perceived system states. Reinforcement learning, enhanced by multiple-step backups, is applied as the kernel of the intelligent control algorithm, where each agent updates its knowledge on-line based on a sequence of states during the process. In addition, the proposed system is designated to be compatible with the prevailing signal system. A case study was carried out in a simulation environment to compare the proposed control approach with a benchmark controller used in practice, group-based vehicle actuated (GBVA) controller, whose parameters were off-line optimized using a genetic algorithm. Simulation results show that the proposed adaptive group-based control system outperforms the optimized GBVA control system mainly because of its real-time adaptive learning capacity in response to the changes of traffic demand.

A non-parametric Bayesian framework for traffic-state estimation at signalized intersections

“…In particular, an on-line learning process will be facilitated to adaptively enhance the models when streaming data is received. In addition, the approach will be integrated into a group-based intelligent signal control system proposed by Jin and Ma [8] with a proper understanding of traffic system. The conditional distribution, P(y N+1 |X, x N+1 , y, θ, η), can be written as P(y N+1 |X, x N+1 , y, θ, η) = P(y, y N+1 |X, x N+1 , θ, η) P(y|X, x N+1 , y, θ, η)…”

“…While many studies in literature were devoted to link-based TSE with regards to intersections controlled by fixedtime traffic signal, group-based and lane-based control approaches provided benefits due to more flexible phase structures and sequences for both conventional traffic signal systems [e.g., 9,15] and adaptive signal systems [8]. Lanebased traffic states are one of the most important inputs for enabling the functionalities of group-based and lane-based controllers.…”

Hierarchical multi-agent control of traffic lights based on collective learning

“…Group-based phasing approaches have proved their utilities by dynamically generating phase structures and sequences with respect to traffic detected at intersection (e.g., Wong and Wong, 2003;Jin et al, 2017b). A previous study by the same authors proposed an RL-based adaptive TLC system suitable for group-based phasing strategies (Jin and Ma, 2017). The system showed the benefits in improving traffic mobility when compared to a conventional logic-based timing approach.…”

“…This paper extends the RL-based intersection adaptive control approach proposed in Jin and Ma (2017) for operating a network of signalized intersections. The rest of this paper is organized as follows.…”

An Adaptive Signal Control Method with Optimal Detector Locations

“…However, most adaptive signal control systems work with stage-based control in which phase sequence is predetermined. To generate phase sequences dynamically, some group-based control methods [18][19][20][21] have been proposed. Recently, some innovative methods such as machine learning, artificial intelligence and reinforcement learning have been used in the adaptive signal control method [22][23][24][25][26][27], and Jing et al [28] proposed an adaptive signal control method in a connected vehicle environment.…”