Research On Attack&Defense Model Of Industrial Cyber-Physical Systems Based On Physical Fingerprinting

Industrial cyber-physical Systems(ICPS)is a multidimensional complex system integrating physics,network and computing,which has a wide range of application value and prospect in the field of modern Industrial production and manufacturing.However,in recent years,ICPS has exposed a number of security vulnerabilities,vulnerable to different types of attacks.Therefore,how to build a secure ICPS network has important theoretical and practical significance.At present,there are the following problems in the research of ICPS security field:1)the influence of physical layer attack on security is not considered too much;2)high-value traffic samples in ICPS nodes are difficult to obtain;3)there are deviations in the labeling of high-value samples for attackers.In view of these shortcomings,this master's thesis studies the fingerprinting based on multi-physical features mining,active learning tracking attack based on multi-layer sampling strategy,active learning tracking attack based on ?-greedy expert decision screening.Specifically,this paper's main work includes the following three aspects:1.Fingerprinting based on multi-physical features mining(FMPFM)is proposed for the low update frequency of existing ICPS software and hardware and the fragile security capability,which makes it difficult to prevent hidden and destructive internal attacks.For sorting in industrial production process,the actual sorting system for abstract a simple physical model,a variety of physical characteristics extracted from physical devices(speed,quality,height,etc.),and the only fingerprinting equipment,and has set up a small sorting test bed to simulate the actual situations of industrial production process of sorting.Finally,the existence of fingerprinting based on multiple physical features is further verified,and the characteristic relationship of fingerprinting is analyzed.At the same time,the influence of random attack model on the application of FMPFM in ICPS is discussed,and the defense capability of FMPFM against internal random attack is proved.2.Aiming at the attack problem of source nodes in industrial information physics system,active learning tracking attack based on multi-layer sampling strategy model(ALTA-MLSS)is proposed.This algorithm is an iterative learning model of "random walk+ active learning",which is a model in which the attacker adds active learning to improve the performance of tracking walk on the basis of random walk of network nodes.In order to solve the problem that outliers and similar samples may be selected in traditional active learning,a multi-sampling strategy algorithm based on sample information degree is constructed.At the same time,a sampling strategy algorithm based on the spatial property of samples is constructed to effectively express the distribution of the whole sample space.Considering the problem of information redundancy in the sampling process,a sampling strategy algorithm based on sample diversity is proposed.Finally,through simulation experiments,it is verified that the traceable attack of active learning has better traceable attack capability than the traceable attack of random walk,and ALTA-MLSS algorithm has better sampling performance than other active learning algorithms.3.In order to solve the problem of expert labeling error in traditional active learning attack model,active learning tracking attack based on ?-greedy expert decision screening algorithm(ALTA-GEDS)is proposed.In order to avoid the error sample annotation caused by the matching process between the attack model and the expert,an attack model expert filtering algorithm(AMEF)is designed to reduce the waste of the attack resources and the probability of being discovered by the system.In addition,an expert decision optimization algorithm based on ?-greedy(EDOPG)is proposed to solve the instability of experts in sample labeling and attack model screening.Finally,the simulation results show that ALTA-GEDS algorithm has better tracking performance than other active learning algorithms.This paper establishes a small physical platform to verify the proposed defense mechanism,and discusses the attack performance of random attack model against FMPFM in ICPS,proving the defense capability of FMPFM against internal random attack.This paper carries out simulation experiments on the proposed attack model from the perspective of the attacker,and the experimental results prove that the proposed attack model performs better in terms of wandering performance and classifier performance than similar attack models.