| With the rapid development of wireless communication and localization technology,the computing nodes distributed in the network can quickly communicate and cooperate with each other.The major problem in the multi-agent system composed of computing nodes is how to reach a consensus through distributed protocols.In order to cooperate to complete the task on the premise of consensus,each agent with limited computing resources and limited sensing range relies on the communication network topology composed of agents.Therefore,the optimization of network topology plays an important role in the convergence performance.This thesis studies the multi-agent network topology optimization and its application in the connected and automated vehicles.Specifically,this thesis firstly studies the network topology optimization under two classical consensus protocols of multi-agent systems,and designs two different topology optimization methods,and then proposes a new architecture of connected and automated vehicles,which provides a reliable and efficient scheme for the implementation of the above proposed topology optimization framework and algorithm.The specific works of this thesis are as follows:Firstly,this thesis studies a framework of the multi-agent system based on hierarchical backbone network topology.Specifically,a new hierarchical organization form of the multi-agent system is designed,which transforms the traditional peer-to-peer structure into a hierarchical backbone network structure.The convergence evolution process of agents is guided by adaptively switching the hierarchical backbone network at any time.The convergence and connectivity of the designed framework are proved by exhaustive method and geometric analysis method.Then,extensive simulations are conducted to verify the performance of connectivity maintenance and convergence through.Besides,compared with the traditional method,this framework can effectively reduce the connectivity constraints of the neighborhood,and then accelerate the convergence process.For the large-scale and high-density multi-agent system,this architecture has stable convergence performance and strong generalization ability.Then,this thesis studies a multi-agent consensus algorithm based on reinforcement learning with neighbor selection.Taking the multi-agent opinion dynamics scene under social network as an example,this thesis designs the decision-making model of neighbor selection in the Markov decision-making process under this scene.Besides,this thesis designs the difference reward function of three different objectives in this scenario,and makes the multi-agent adaptively learn the decision-making ability of multi-objective trade-off in the process of interacting with the environment through the reinforcement learning algorithm of policy gradient.Through simulation experiments,the effectiveness of the designed decision-making model based on neighborhood selection under the difference reward function of three kinds of objectives is verified.Compared with the traditional dynamics model,the proposed model can effectively reconcile the conflicting opinions among agents and enhance the consensus among all agents,that is,reduce the number of clusters formed when the multi-agent system finally stabilizes.Finally,taking a prototypical lane reduction scenario as an example,this thesis designs a new connected and automated vehicles architecture under multi-agent mobileedge computing networks based on software defined network and its corresponding core operation process.In this thesis,the network topology optimization methods proposed in the above two works are integrated into the proposed vehicular architecture.This architecture can provide a low delay,efficient and reliable network topology for connected and automated vehicles in the platoon control process of the lane reduction scenario,and provide a sustainable and scalable application example for the application of connected and automated vehicles. |
PDF Full Text Request