Research On Cyber-Physical Threat Models And Security Mechanisms For Critical Infrastructure Networks

Critical infrastructure networks(CIN)are closely related to people's daily life and support the whole society to normally run.CIN are large-scale distributed cyber physical systems(CPSs)and typical complex cyber-physical integrated networks.In CIN,cyber domain and physical domain depend on and integrate with each other,which introduce cyber-physical dependence and make cyber-physical cross-domain attack possible.Moreover,by comprehensively applying physical attacks,network attacks and social engineering methods,CIN are faced with great challenges of unknown attacks and highly hidden attacks.Focusing on CIN' cyber-physical security and taking the large-scale distributed cyber-physical integrated CPSs as research object,this paper investigates into the following several areas:1.Quantifying the CIN cyber-physical dependences: nodes in CIN are classified into cyber nodes and physical nodes.Cyber-physical dependence model is constructed based on weighted directed graph.Moreover,two cyber-physical dependence metrics named as Cross-Entropy model and Control-Measure model are proposed to quantify the dependences among cyber and physical nodes.Cross-Entropy model and Control-Measure model are used to quantify the dependence between two physical nodes and the dependence between two cyber and physical nodes respectively.2.Quantitative CIN cyber-physical security assessment: security indicators including critical level,attack loss,defense strength and infected probability are introduced to describe the security attributes of each cyber and physical nodes,and cyber-physical security metric is defined to quantify each node's security status.Furthermore,cyber-physical security assessment algorithm is proposed based on the cyber-physical dependence model,which can compute each node's security metric values and discover potential vulnerable paths.Case study and evaluations verify the effectiveness of the proposed algorithm and demonstrate its high time efficiency.3.Quantitative CIN cyber-physical threat model: the colored Petri net is extended to define the probabilistic colored Petri net,based on which the cyber-physical threat model fragments are proposed.We further clarify how to construct a complete cyber-physical threat model for a given CIN based on cyber-physical dependence model and threat model fragments.To calculate connection weights in cyber-physical threat model,mixed-strategy game theory is applied to build the attack-defense game model and mixed-strategy Nash equilibrium is solved.Case study verifies the effectiveness and usability of the proposed threat modeling method.4.Quantitative CIN cyber-physical attack path analysis: to calculate connection weights in cyber-physical threat model,incomplete information Bayesian game theory is applied to build the attack-defense game model and perfect Nash equilibrium is solved.Incomplete information Bayesian attack-defense dynamic game algorithm is proposed to depict the attack and defense dynamics between the attacker and defender.Furthermore,cyber-physical attack path analysis algorithm is proposed to discover all possible attack paths with probabilities from attacked nodes.Case study verifies the usability of the proposed algorithm.Evaluations show that the algorithm has high time efficiency within limited node number.5.Anomaly and unknown attack detections based on cyber-physical dependence metrics: non-parameter probability density estimation is used to estimate the distributions of cyber and physical nodes' measurements.Cross-Entropy model and Control-Measure model are used to quantify dependences among cyber and physical nodes.By large dataset and machine learning,the normal ranges of cyber-physical dependence metrics are trained and based on that anomalies and unknown attacks are detected.Evaluations verify the high accuracy of cyber-physical dependence-metric based anomaly detections and verify that dependence metrics can be combined with KNN classifier.6.Context based detections against highly hidden legal false price data and malicious commands: attackers can steal legal credentials to make their false price data and malicious commands legal,which attacks are highly hidden.The communication contexts of true price data and normal commands are defined.Corresponding attack contexts are also defined.Based on that,the detection unit on smart meter side is designed and legal false price data and malicious commands detection algorithms are proposed.Simulations and evaluations verify the effectiveness of the proposed detection algorithm against highly hidden attacks.The detection accuracy can be improved by selecting proper parameter values.