Research On (r,s)-robustness Evaluation And Optimization Of Multi-agent Systems Based On Deep Learning

A Multi-agent system is a group system composed of a group of independent and cooperative individuals through the form of a network.They can be coupled together in different forms,sensing and sharing information through sensors and communication modules,and performing tasks through actuators.Multi-agent system is widely used in People’s Daily life,but as the multiagent system is usually deployed in an open environment,the communication link and intelligent node in the system will inevitably be affected by the environment or even malicious attack,which makes the network security of multi-agent system become the focus of researchers in recent years.Multi-agent network robustness refers to the ability of a system to perform and complete its tasks despite a cyber-attack.The communication graph properties of r-robustness and(r,s)-robustness have received close attention for their ability to assess the resilience of a system in the event of a failure of a node or link in the network due to an attack.However,the traditional evaluation methods of r-robustness and(r,s)-robustness,such as the exhaustive method,are difficult to evaluate the large-scale multi-agent network with a large number of nodes due to their high time complexity.In recent years,deep learning has opened up new ideas for solving problems with high algorithmic complexity.Given this,this paper uses deep learning,graph theory,and other methods to study the multi-agent system(r,s)-robustness evaluation and optimization methods under network attacks.The main research work of this paper is as follows:(1)Based on the commonness of the multi-agent network cooperation and information transmission mechanism with the cooperative communication network between intelligent vehicles,the vulnerability detection problem of an intelligent transportation system communication network is modeled as the r-robustness value solving the problem of multi-agent network.Aiming at the problems of high complexity and poor real-time performance of the existing robustness detection methods for multi-agent networks,the solution of r-robustness value is transformed into a deep learning graph classification problem with the help of the deep learning algorithm,and a deep learning algorithm integrated into the residual network is designed.Finally,a set of typical traffic scenes are simulated to verify the effectiveness of the proposed method.(2)Aiming at the problem that the(r,s)-robustness evaluation of multi-agent systems cannot be solved in polynomial time,a deep learning model with cross-shaped convolution kernel structure is designed to quickly and accurately evaluate the(r,s)-robustness of the systems.In addition to this,the concept of key nodes is proposed and studied in networks based on(r,s)-robustness,transforming the problem of key node detection into a classification problem with multiple labels for deep learning.Finally,the validity of the proposed model was verified experimentally,and the details of the model are discussed in depth through ablation experiments.(3)Aiming at the(r,s)-robustness optimization problem of multi-agent systems,an optimization strategy based on key nodes is designed.Firstly,the existing multi-agent systems and common optimization methods in complex network scenarios are classified.Then,based on the concept of critical nodes,an optimization strategy is proposed by adding extra links to the key nodes in the network under cost constraints.The proposed method can effectively increase the redundant information flow in the network.Finally,four sets of comparison experiments verify the effectiveness of the optimization strategy.