Distributed Traffic Signal Adaptive Control Based On Multi-agent Coordination

Based on data-driven modeling,queue balancing idea and multi-agent system theory,this paper focuses on the multi-agent data-driven distributed adaptive coordination control from the perspective of single signal control and road network signal control.The main research content and innovation of this paper are summarized as follows:1.In view of the current spillback and starvation problems generated by the traffic signal control system,the store-and-forward model is used to abstract the dynamics of traffic flow.Based on the queue balancing idea,the traffic flow data obtained only through the front-end sensor network is applied to the linear mathematical transformation of the unknown function with the system mechanism by parameter separation technology.The data-driven modeling of signal control system with complex nonlinear characteristics is realized.Secondly,a multi-agent coordination strategy under the single signal control intersection is proposed,and the distributed consistency error is constructed.An improved multi-agent data-driven adaptive coordination control algorithm(i-MA-DDACC)with saturated green light constraints is designed,so as to achieve the goal of quickly releasing the queuing vehicles and improving the efficiency of green time utilization.Finally,using Lyapunov stability theory,it is strictly proved that the controller monotonically decreases and has upper and lower boundaries,that is,it converges to the true value.2.In order to avoid the adverse effects of queue overflow and green time waste caused by unbalanced traffic flow in urban traffic network,a multi-agent data-driven distributed adaptive coordinated control strategy(D3AC2)is proposed.In the case of unknown network traffic flow mechanism,only use traffic data generated in the traffic network to model,which solves the problems of limited observation and low expansion ability of network control unit.Firstly,the internal and external coordination adjacency matrix in traffic network communication topology is studied.On this basis,a new fully scalable distributed consistent coordinated vehicle queuing error model is designed,which involves the information flow interaction between agents inside and outside the intersection.Under the constraint of saturated green time,the control law and parameter learning law of D3AC2 are constructed,independent of the precise traffic mechanism model.Through Lyapunov stability analysis,a complete and strict convergence proof is given.At the same time,the principle of selecting the gain of the adjustable controller is discussed.3.On the basis of i-MA-DDACC and D3AC2 algorithms proposed in this paper,a Python-Sumo joint simulation platform is built.In the numerical experiment of i-MADDACC algorithm,the comparison simulation of the delay level and queuing time with fixed and inductive timing proves the effectiveness of the algorithm.Furthermore,the comparative experiments under different traffic conditions prove the adaptive ability and robustness of the algorithm.In the numerical experiment of D3AC2 algorithm,based on Open Street Map to build a real traffic network model,four different datadriven adaptive control methods are compared in multiple dimensions such as delay,change of green time input and queue output,and pollution emission level.The simulation results verify the superiority of the algorithm.