Risk assessment method of power plant industrial control information security based on Bayesian attack graph

Abstract

In view of the current fault isolation and single-fault assessment methods for power plant industrial control information security, there are problems of low attack point capture accuracy, long time,and poor evaluation effect. A Bayesian attack graph-based intelligent risk assessment of power plantindustrial control information security is proposed. method. The attack graph technology is used tomodel the risk elements identified in the risk analysis, and the probabilistic model and Bayesianprobabilistic attack graph are used to describe the relationship between system threats and attackbehaviors. Deeply understand the basic elements of attack graph modeling for informationcollection, automatically generate tools to construct and optimize attack graphs, use sampleinformation to modify the original estimates of parameters, and the hyperparameters of the priordistribution are determined by the node probability value. Analyze the attack sample data to obtainthe attribute data of the node, so as to complete the posterior estimation parameter learning. Thebasic credibility of each proposition in the recognition framework is determined, and the weightvector of evidence is determined. Analyze the specific information security threats of the industrialcontrol system of the power plant, obtain the basic credibility function, and establish the riskcalculation formula. Quantify the language assessment of security threats by experts, construct anintelligent risk assessment model for power plant industrial control information security, and designthe implementation process of risk assessment based on the risk analysis of attack graphs. Theweight value of security threats and the result of credibility distribution are determined to completethe intelligent assessment. It can be seen from the experimental results that the short-term energyof this method is gradually invalid, the information is in the storage stage, and no new evaluationinformation will be added. The highest accuracy of capturing attack points under passive and activeattacks is 91% and 93%, respectively, and the longest capturing of attack points. The time does notexceed 20 min