Identification Of Vital Nodes In Complex Networks Based On Deep Reinforcement Learning

Complex network theory is widely applied in the area of business intelligence.The identification of vital nodes is the key technology of complex network theory,which attracts the attention of academic community.Many scholars have carried out in-depth research on the vital node identification or node importance ordering of complex networks and achieved a lot of research results.But the scale of complex networks is growing exponentially as artificial intelligence and big data technologies are applied to businesses.The accuracy and real-time performance of traditional vital node identification methods can no longer meet the real demand.The main research works of this thesis are as follows:(1)Firstly,on the basis of graph theory,the existing classical complex network evolution models such as regular network,random network,small-world network and scale-free network are analyzed to understand the statistical characteristics of various models,which lays a foundation for the definition and identification of vital nodes.(2)This thesis researches the vital node indentification algorithm of traditional classical and new complex networks.The traditional classical algorithms include degree centrality,medium centrality,k-kernel decomposition and PageRank algorithm,etc.The new ones mainly include the improved algorithm and comprehensive method for the classical algorithm.Based on the static indexes such as the reliability and topology of the complex network,the importance level method of the vital node identification is divided into the method based on propagation dynamics and the method based on network robustness.(3)The evaluation model of the vital node indentification algorithm of complex network is constructed based on the static indexes such as the reliability and topology of complex network.Based on the analysis of the current situation of deep reinforcement learning and complex network research,the scheme of combining deep Q network with vital nodes identification is explored.(4)Combining the advantages of Network robustness and Network propagation dynamics,a method for identifying vital nodes of complex networks based on Deep q-network(DQN)is proposed.Through DQN reward matrix design to integrate the existing indicators,training the neural network parameters can be given nodes between the model of the optimal path strategy,define DQNRank value is through a node number of the optimal path and the ratio of the total number of article path,said the index node toother nodes in the network control degree,according to the size of the value of node importance ranking.In order to verify the applicability and effectiveness of the proposed method,two simulation experiments are designed.Experiment No.1 applies the method respectively without to have no right to ARPA and ARPA undirected weighted networks,and the indentification result with other four methods do contrast analysis,use based on the robustness of the evaluation standard for inspection method,observation results sorted by importance to remove node in turn after impact on the network structure and connectivity.the results show that the method of the number of subgraph is bigger and graph on a smaller scale,this method is suitable for the identification of vital nodes in complex networks,and the precision is higher.In experiment No.2,BBV network model is used to simulate the construction of large scale undirected weighted real networks.The results show that the method presented in this thesis has good applicability to large-scale complex networks and can be applied to the identification of vital nodes in such networks.