.png){kind=icon}
.png){kind=icon}
Return
Adaptive distributed multi-objective collaborative traffic signal control framework based on multi-agent reinforcement learning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.future.2026.108526.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
With the development of urban road traffic, the adaptive coordinated control of traffic lights at multiple intersections has become one of the key issues in the field of intelligent transportation systems. With the introduction of various algorithms, deep reinforcement learning (DRL) has shown outstanding results. However, with the increase of traffic network scale, the traditional deep Q network (dqn) is difficult to adapt to large-scale action space due to the limitation of model structure. To solve this problem, this paper proposes a multi-objective distributed adaptive cooperative communication framework. The main contributions are as follows: (i) A general local traffic network partitioning method is proposed to divide the traffic network topology into smaller sub-networks. (ii) Traffic safety is incorporated into consideration to reduce the signal conflict rate and effectively lower the incidence of traffic accidents, while achieving superior performance with reduced algorithmic and communication complexity. (iii) A fully adaptive distributed framework is proposed, enabling each intersection to coordinate the entire traffic network solely through local information, thereby improving the training efficiency and scalability of Traffic Signal Control (TSC). The experimental results show that the proposed multi-objective distributed adaptive neighborhood information fusion model can effectively reduce the dimension of the action space and greatly improve the efficiency and safety of multi-agent traffic signal control. Compared with the existing most advanced benchmark models such as IVPL, EHC-HDQN, CODQN, our adaptive distributed intelligent learning (ADIL) model has a significant lead. At the same time, through ablation experiments, we reduced the conflict rate by 7% compared with ADIL-NS, verifying the traffic safety of multi-objective control. Compared with existing methods, the proposed ADIL method not only shows better performance, but also has faster training speed, stronger robustness and security.
Keywords:
Multi-agent reinforcement learning
Traffic signal control
Distributed framework
Adaptive coordination
Multi-objective optimization
Journal
F
IF:
0
Papers:
642
Citations:
0