The Evaluation Methods Of Robustness For Multi-Agent Networks Under Malicious Attack

Multi-agent system(MAS)is a new concept formed with the development of embedded sys-tem and network technology.It can accomplish complex tasks that can not be accomplished by a single agent or a person.Consensus is the most widely studied problem in MASs.It mainly in-cludes two parts:distributed control protocol and network communication topology.At present,network topological connectivity is a necessary condition to achieve consensus control algorithm.However,existing attacks can easily destroy this topological connectivity.Network robustness or(r,s)-robustness is a network topology attribute commonly used to measure the ability to resist attacks,but its evaluation and determination is a NP-hard problem.In this work,the evaluation of network robustness of MASs is studied in depth based on intelligent optimization algorithm.This work proposed a method of evaluating network robustness based on multi-layer percep-tron(MLP)to avoid exponential time complexity of algorithm based on exhaustive idea.Firstly,features were extracted from degree distribution and spectrum space of graph data,and then MLP model was trained and optimized by using this feature.Finally,the model was able to evaluate network robustness.The simulation results showed that the average accuracy of this method was at least 4%higher than that of the original method.This work proposed a method of evaluating network robustness based on representation learning to reduce manual process in traditional neural network method.In the pre-processing stage,the adjacency matrix was reordered,then the convolutional neural network(CNN)was used to automatically learn the features from the adjacency matrix.Finally,the method was able to evaluate network robustness with a small amount of manual processing.The simulation results showed that the average accuracy of this nethod was at least 4%higher than that of the original method.This work proposed a network robustness evaluation method based on simulated anneal-ing(SA)algorithm to avoid the training process of neural network method.This method mainly designed three state transition modes including adding,deleting and replacing,and realized the evaluation of network robustness by SA.The simulation results showed that the performance of the proposed method in evaluating network robustness without training was comparable to that of the neural network-based method in the test data.However,with the increase of the number of nodes,the computational efficiency of this method will gradually decrease.In summary,this work opened up a new way to evaluate the robustness of multi-agent net-works under malicious attacks.The simulation results showed that the proposed method can effectively evaluate the robustness of networks.Therefore,it is of certain theoretical value.