Evolutionary Optimization Algorithm And Application On The Robustness And Influence Of Complex Networks

The complex network is an important tool to describe and analyze complex systems,which is composed of interconnected units with certain interactive patterns.Typical complex systems,such as power grid,transportation network,social network and Internet,can be constructed and modeled as complex networks.These networks exist widely in society and play an important role,and it is necessary to ensure their efficient and stable operation.As one of the main research topics in network science,network robustness refers to the ability of a system to maintain its basic functions when suffering internal interference and external changes.At the same time,the rapid development of the Internet and various social media provide a good platform for the information exchange and dissemination.Therefore,the analysis of social influence dissemination has become the focus of research in sociology,economics,computer science and other fields.One of the goals of studying social influence is influence maximization,which can be applied to big data analysis,viral marketing,recommendation system and many other directions.Taking network robustness and influence as the core,using evolutionary algorithms as the optimization tool,this thesis focuses on studying the following four questions:"what characteristics make network robustness","what types of attack strategies make network vulnerable","the relationship between network robustness and influence",and "how to maximize the network influence".These studies provide ideas for the research and application of related fields,and the main work and innovations of this thesis are summarized as follows:1.The relationship between network structure parameters and network robustness is analyzed from the perspective of protectors.The network structure not only reflects the connection between nodes in the network,but also reflects the basic characteristics of the network through the measurement of network structure parameters.In the context of catastrophic cascade attacks,the classical scale-free networks and small world networks are taken as examples,the power exponent and assortativity that can represent the characteristics of the network are selected as the structural parameters.Then the relation between network robustness under different parameter values is comprehensively analyzed.The experimental results show that the network type and network structure have obvious impact on the network robustness.2.A cost-minimizing attack strategy based on memetic algorithm is proposed from the perspective of attackers.In order to improve the disadvantage traditional malicious attacks without considering the cost of attacking nodes,the proposed strategy considers the attack cost which is proportional to the node importance,and has reasonable attack termination conditions.In order to minimize the attack cost,the corresponding memetic algorithm is designed to optimize the combination of attack node sequences.Based on a variety of network data,experimental results show that the algorithm can take the attack cost and the structural importance of the attacked node into account at the same time.The algorithm not only shows good performance in minimizing the attack cost,but also effectively improves the attack accuracy and efficiency.3.A cascading failures model based on multi-attribute influence propagation is proposed from the perspective of attackers.The load distribution of failed components is the key factor affecting the efficiency of cascading failure.The global attribute influence of nodes is extracted from the K-shell decomposition process and the local attribute influence is extracted from neighbors.The above two attributes can accurately quantify the influence of each node.The comprehensive influence of nodes is adopted as the basis for load redistribution of failed nodes.The experimental results confirm the effectiveness of the proposed multi-attribute influence propagation strategy to improve the efficiency of cascading failure,and also verify the important role of the node influence on the effect of information dissemination.4.A niching memetic algorithm for multilayer network influence maximization problem is proposed.The node diversity and the propagation probability are the keys to the success of information dissemination.In terms of the quantification of node diversity,an expandable node influence representation vector is designed,which can change the influence evaluation type according to actual needs.Meanwhile,a two-stage information propagation model is constructed using various influencing factors in the process of information propagation,which reasonably quantifies the propagation probability between node pairs.By introducing niching technology,the optimization algorithm can provide multiple high-quality solutions,which satisfies the multi-solution characteristic of the problem itself.5.Modeling,analysis and optimization of military communication networks.Taking the military communication network as an application example,firstly,combined with the characteristics of military communication network,a structural model including vertical hierarchical relationship and horizontal collaborative relationship is constructed.Then,based on this structural model,the performance of generated military communication network is analyzed from three aspects:equipment entity importance,link importance and network robustness by using the basic statistical indexes of complex network.Finally,an optimization algorithm is designed to improve the network robustness.Through the analysis and optimization of the network generated by the proposed structure model,and the feasibility and effectiveness of the structural model and optimization algorithm are verified.