基于贝叶斯网络d-分隔定理的节点置信度计算方法

NODE CONFIDENCE CALCULATION METHOD BASED ON D-SEPARATION THEOREM OF BAYESIAN NETWORK

现有的贝叶斯网络节点置信度计算方法,存在着因条件概率的错误计算和节点的相关性导致的节点置信度错误计算问题。这些问题降低了节点置信度的准确性,影响了网络威胁传播路径预测的有效性。为此,提出基于贝叶斯网络d-分隔定理的节点置信度计算方法。首先,通过分析攻击成本和攻击行为发生的可能性之间的关系,提出攻击行为发生的条件概率计算方法,以解决条件概率的错误计算问题;其次,通过引入贝叶斯网络分隔定理,使存在关联性的节点在它们共有的d-分隔集合条件下相互独立,并提出节点置信度的计算方法,以有效地避免相关性导致的节点置信度错误计算;最后,实验结果表明,该方法有效地解决了节点置信度的错误计算问题,提高了节点置信度的准确性,实现了对网络威胁传播路径的有效预测。

Current node confidence calculation method for Bayesian network has the problem of node confidence miscalculation caused by the miscalculation of conditional probability and the correlation of nodes. This problem reduces the accuracy of node confidence and impacts the effectiveness of prediction on propagation paths of network threats. Therefore,we present a node confidence calculation method which is based on d-separation theorem of Bayesian network. First,by analysing the correlation between attack cost and the occurrence likelihood of attack activity,we propose an approach for calculating the conditional probability of attack activity occurrence so as to solve the problem of miscalculation in conditional probability. Secondly,by introducing separation theorem of Bayesian network,we make the nodes with correlation be independent to each other under the condition of their common d-separation set,and propose the node confidence calculation method so as to effectively avoid the miscalculation of node confidence caused by the correlation. Finally,experimental results show that our method effectively solves the miscalculation problem of node confidence and improves the accuracy of node confidence,consequently it achieves the effective prediction on propagation paths of network threats.