【作者】 董林；

【导师】 卢玉峰；

【作者基本信息】 大连理工大学 ， 人工智能， 2025， 硕士

【摘要】 随着信息技术的迅速发展,复杂网络在现代社会中扮演着至关重要的角色,被广泛应用于电力系统、交通运输以及社交平台等诸多实际场景中。关键节点识别是复杂网络分析中的重要研究方向之一,在提升网络整体效能、降低系统故障风险以及推动社会生产效率等方面发挥着关键作用。本文针对现有关键节点识别方法在捕捉邻域信息全面性和适应时变网络方面的局限性,提出了一种基于邻域K-shell分布与图卷积网络(GCN)相结合的关键节点检测方法K-GCN。该方法通过计算节点的邻域K-shell分布熵值,量化节点在网络中的重要性,并将这些熵值作为关键状态信息输入到GCN网络中,实现对关键节点的精确识别。本文还针对复杂网络中核心节点的识别问题,提出了一种多标准决策方法(MCT-NDI)。在模型构建过程中,融合了多种指标,包括接近中心性、介数中心性、H指数以及网络约束系数,同时还整合了局部邻域影响力、网络中节点的拓扑结构位置、路径中心性与节点互信息等多维度特征,从信息层次和结构层级两个角度提升了评估的全面性,有效克服了单一指标在衡量节点重要性时所存在的局限。实验环节选用了多个实际网络数据集,结合SIR传播模型及牵制控制实验,对所提出的方法进行了有效性验证。实验结果表明,K-GCN和MCTNDI在识别关键节点方面表现出色,尤其在传播速度和传播范围上优于现有方法。本研究所取得的成果为复杂网络中关键节点的识别提供了全新的方法路径,不仅在理论探讨层面具有积极推动作用,并在实际场景中体现出较强的应用前景与实践意义。更多还原

【Abstract】 With the rapid development of information technology,complex networks play an in-creasingly critical role in modern society,with widespread applications in areas such as elec-tric power systems,transportation,and social media platforms.Identifying key nodes is a central issue in complex network research,as it is essential for enhancing overall network performance,reducing system failure risks,and improving productivity.To address the limitations of existing key node identification methods—particularly their inability to comprehensively capture neighborhood information and adapt to time-varying network structures—we propose a novel method called K-GCN,which integrates neighbor-hood K-shell distribution with a graph convolutional network(GCN).This approach eval-uates the importance of each node by calculating the entropy of its neighborhood K-shell distribution,and then feeds this entropy as crucial state information into the GCN to accu-rately identify key nodes.Moreover,this study proposes a Multi-Attribute Decision-Making Method(MCTNDI)aimed at identifying key nodes within complex networks.The developed model incorpo-rates a combination of metrics such as closeness centrality,betweenness centrality,H-index,and the network constraint coefficient.Furthermore,it incorporates a range of features such as local neighborhood influence,node topological position,path centrality,and mutual in-formation between nodes.By combining both informational and structural dimensions,this approach enhances the comprehensiveness of node importance evaluation and effectively overcomes the limitations of relying on a single metric.In the experimental section,multiple real-world network datasets are utilized to vali-date the effectiveness of the proposed methods through both SIR propagation simulations and pinning control experiments.The results demonstrate that K-GCN and MCTNDI ex-hibit superior performance in identifying key nodes,particularly outperforming existing ap-proaches in terms of propagation speed and coverage.The outcomes of this study offer a novel methodological framework for critical node detection in complex networks,contribut-ing significantly to theoretical research while also showcasing strong potential for practical applications.更多还原