A sensor additive switching watermark design method for detecting and identifying physical faults and network attacks

Through the sensor additive switching watermark design method, combining the two-layer decision strategy and the time seed of the Lorentz system, accurate detection and distinction of physical failures and network attacks is achieved, and the watermark detection affects control performance and confidentiality in the existing technology is solved. The problem of sexuality improves the security and reliability of the system.

CN116155523BActive Publication Date: 2024-11-15NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211245875.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-10-12
Publication Date
2024-11-15
Estimated Expiration
2042-10-12

AI Technical Summary

Technical Problem

In the prior art, when detecting and identifying physical failures and network attacks, typical additive watermark detection methods will lead to degradation in control performance, and the method of adding watermarks at the sensor requires additional information sharing channels, which reduces the confidentiality of watermarks and lacks A method that can simultaneously detect and distinguish between general physical failures and man-in-the-middle attacks.

Method used

The watermark design method of sensor additive switching is adopted, and the watermark switch trigger mechanism is realized by establishing a two-layer decision strategy and adaptive threshold, combined with the time seeds generated by the Lorentz system, to ensure that the watermark generation is synchronously or asynchronously switched in the sensor data transmission channel. and removers to enhance detection of man-in-the-middle attacks and physical failures.

Benefits of technology

Improves the reliability and security of the system, and can accurately detect and identify physical failures and network attacks without affecting control performance, enhances the security and reliability of the system, especially in scenarios containing unknown information. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116155523B_ABST
    Figure CN116155523B_ABST
Patent Text Reader

Abstract

The present invention discloses a sensor additive switching watermark design method for detecting and identifying physical faults and network attacks, including: facing man-in-the-middle attacks represented by denial of service attacks and integrity attacks. The present invention loads a watermark generator and a watermark remover on the physical end and the network end of the sensor respectively, so that they intervene in the data transmission channel in the sensor, and complete the switching between data transmission quality sets according to the trigger protocol and time seed; a trigger mechanism integrating inner and outer loop switching is designed, which can effectively diagnose man-in-the-middle attacks and identify threats, even if the trigger protocol and time seed contain unknown information of the attacker. The diagnostic identification method based on switching watermarks of the present invention performs well in detecting and identifying attacks and faults on the system, even in scenarios containing unknown information, greatly improving the reliability and security of the system. At the same time, the method of the present invention has the characteristics of simple structure, few parameters, and easy implementation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of automatic control technology, and in particular relates to a sensor additive switching watermark design method for detecting and identifying physical faults and network attacks. Background Art

[0002] Cyber-physical systems (CPS) integrate technologies from the fields of control, computing, and communication. In addition to traditional physical failures, CPS also creates new security vulnerabilities due to this integration, such as cyber attacks. Cyber ​​attack incidents have been witnessed in the past decade. Physical failures and cyber attacks require different remediation measures. Man-in-the-middle (MITM) attacks appear in the form of service interruption (DoS) and integrity attacks, which have attracted widespread attention in the field of CPS security. Threat identification aims to identify the type of threat, that is, to determine what type of threat (physical failure or cyber attack) has occurred, to help operators make the right decision, and to take appropriate remedial measures. The watermark detection method that has emerged, which involves adding a watermark to the control input of the physical device, and then comparing it with the measured watermark on the network side, reveals the occurrence of the attack by the difference between the added watermark and the measured watermark. The typical additive watermark detection method will lead to a degradation of control performance even under nominal conditions. The reason is that the watermark is injected into the system control input channel and acts as an additional input interference to affect the normal operation of the system. Based on the detection method of adding watermarks at the sensor end, a watermark is added to the sensor measurement value at the physical device end, and the watermark is removed at the network end. This method can detect replay attacks and does not affect the control performance under nominal conditions. However, it requires some additional information sharing channels, which reduces the confidentiality of the watermark. Therefore, when using the watermark method to effectively detect faults and attacks in physical information systems, it is necessary to ensure that the control performance of the system is not affected, and to ensure the confidentiality of the watermark, and to detect and identify the faults and attacks on the system as accurately as possible, so as to enhance the security and reliability of the system, making it a research hotspot in this field.

[0003] At present, algorithms for detecting and identifying physical faults and network attacks mainly focus on the following aspects: (1) In previous detection and identification algorithms, typical additive watermark detection methods will lead to a decrease in control performance even under nominal conditions, because the watermark is injected into the system control input and acts as an additional input to disturb the normal operation of the system. (2) The watermark detection method at the sensor end requires some additional information sharing channels, which reduces the confidentiality of the watermark scheme. (3) Previous algorithms mainly focus on attack detection problems, while ignoring threat identification problems. There is a lack of research on watermark methods that can simultaneously detect and distinguish general physical faults and MITM attacks. Summary of the invention

[0004] Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention discloses a sensor additive switching watermark design method for detecting and identifying physical faults and network attacks, which can greatly improve the reliability and security of the system even in scenarios containing unknown information.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a sensor additive switching watermark design method for detecting and identifying physical faults and network attacks, comprising the following steps:

[0006] (1) Establish a physical information system with man-in-the-middle attacks and physical failures, including constructing two adaptive thresholds with different amplitudes for the anomaly detector in the system to correspond to the nominal scenario and the failure scenario respectively. Based on the anomaly detector and adaptive threshold in the system, combined with the sensor additive switching watermark, a two-layer decision strategy is constructed;

[0007] (2) According to step (1), a two-layer decision strategy and detection threshold are established, and a watermark generator and a watermark remover are loaded on the physical end and the network end of the sensor respectively, so that they are inserted into the data transmission channel of the sensor, and the generated additive switching watermark is switched in a set of data transmission quality sets;

[0008] (3) Based on the data transmission quality set established in step (2), the switching trigger mechanism of the additive watermark includes inner loop switching and outer loop switching;

[0009] In the outer loop switching, the watermark generator and the watermark remover are switched synchronously in the nominal scenario and the fault scenario, and are switched asynchronously in the man-in-the-middle attack detection;

[0010] In the inner loop switching, a set of Lorentz time seeds is generated by the Lorentz system, in which any element of a time seed is different from an element of another time seed;

[0011] The switching trigger mechanism of the additive watermark and the generated time seed both contain unpredictable attacker information.

[0012] Furthermore, the physical information system described in step (1) is composed of physical devices Actuators, sensors, actuator communication networks and sensor networks, controllers and anomaly detectors constitute;

[0013] The operation of the physical information system can be expressed as a set of differential equations as follows:

[0014]

[0015] y p =Cxp +f s (t)+D d d(t),

[0016] y q =y p +a y (t),

[0017] in, is the system status, Controller meets is the measured value of the sensor, Sensor measurements received by the controller, also anomaly detector Sensor measurements obtained from the sensor network, A, B, C, B d ,D d is a known system matrix with suitable dimension, the matrix pair (A, C) satisfies the observability of the system, represents the lumped disturbance and measurement noise, and has an upper bound satisfy t represents time;

[0018] matrix Represents the nonlinear part of the physical information system and satisfies the following Lipschitz condition:

[0019]

[0020] Among them, γ>0 is the Lipschitz constant, and They refer to the man-in-the-middle attack signals received by the actuator and sensor, respectively. u and a y Cooperate to implement man-in-the-middle attack;

[0021] and They refer to the physical process faults, actuator faults, and sensor faults to which the system is subjected, respectively;

[0022] In the physical information system constructed based on the above step (1), the method defines the following expression to characterize the differential equations of the system operation:

[0023]

[0024] We can get: f p (t)+Bf a (t) = B f f(t),B f =[I,B,0],f s =D f f,D f=[0,0,I], and at the same time:

[0025] Furthermore, step (1) is for anomaly detector The model design expression is as follows:

[0026]

[0027] r=y q -Cx r ,

[0028] in, is the system state quantity of the detector, is the system residual used for system anomaly detection, is the system gain matrix, satisfying Become the Hurwitz Matrix;

[0029] Based on this, the residual system expression established is as follows:

[0030]

[0031] r=Ce+D d d+D f f+a y .

[0032] in, is the state quantity of the residual system;

[0033] Based on anomaly detector The model expression, in the nominal scenario, the system satisfies f(t)=0,a u (t) = 0 and a y (t) = 0, r(t) represents the residual of the system, which satisfies the following adaptive threshold constraint:

[0034]

[0035] Among them, the coefficients κ>0 and μ>0 satisfy in addition, satisfy

[0036] In a fault scenario, a u (t)=0,a y (t) = 0, the residual r(t) satisfies The adaptive threshold Specifically

[0037]

[0038]

[0039] Furthermore, the method is based on the established anomaly detector And adaptive threshold and Combined with the sensor additive switching watermark, the two-layer decision strategy is as follows:

[0040] Layer-1: At a certain time t d ,if This indicates that at t d At this moment, the system is attacked, otherwise, it indicates that the system is not attacked;

[0041] Layer-2: When at t d Always present If at t a Always Satisfied And t a >t d , indicating that at t a The system is attacked at the moment, or vice versa, the system fails;

[0042] When layer-1 detects that the system is under threat, layer-2 is activated to identify that the system is under attack.

[0043] Furthermore, based on the two-layer decision strategy and detection threshold, in order to achieve the threat detection and identification goals, the method switches the watermark to the data y transmitted in the sensor communication channel. p Add a watermark to drive the residual |r(t)| beyond the threshold and Rather than a failure scenario;

[0044] The method introduces an additive switching watermark mechanism into the sensor data transmission channel, which specifically includes the following process:

[0045] Represents the input of the sensor data transmission network on the physical side, The output of the network-side sensor data transmission network, the switching watermark fusion mechanism is composed of the switching watermark generator and watermark remover Generate separately and To achieve this, we can get:

[0046]

[0047] Watermark Generator With Watermark Remover Share a set of time seeds;

[0048] Watermark Generator Generate watermark by switching between {-θ1,θ1,θ1+θ2,-θ1,-θ2} w , and watermark w The time switching sequence is determined by the watermark generator The trigger strategy determines the watermark remover The watermark q is also generated by switching between {-θ1,θ1,θ1+θ2,-θ1,-θ2}, but its time switching sequence is determined by the watermark remover and a common time seed to determine.

[0049] For step (2), the switching trigger mechanism of the additive watermark includes the following calculation process:

[0050] Define the symbol h∈{w,q} to describe the watermark generator With Watermark Remover Behavior, definition As the triggering time sequence of the watermark generator w (h = w) and the watermark remover (h = q), at the same time, define As the time series of the k-th time seed, And for any and h∈{w,q} are both fixed values. represents the switching time sequence triggered by the time seed, forming an inner-loop switching mechanism;

[0051] Correspondingly, the watermark generator and watermark remover The triggered switching mechanism is called the outer loop switching mechanism;

[0052] further, and represents the switching time constant, starting point and switching time of the k-th outer loop switching mechanism. Based on this, the switching expression for the additive watermark in the method is:

[0053]

[0054] in, and all its elements are positive, For any i∈{1,…,n y}, when h i When (t)≥0, sign(h i (t))=1, otherwise sign(h i (t)) = 1, Describes the relationship between the switching timing and quantity values ​​of w and q;

[0055] There are N time seeds in the method, and the time seed refers to a set of time lengths between watermark switches. The k-th time seed is defined as

[0056] Furthermore, the step (3) includes watermark detection, watermark generator and watermark remover The watermark detection mechanism and its corresponding network control system include the following processing and calculation processes:

[0057] (31) The sensor measurement value y received by the controller q for:

[0058] y q =y p +w+a y -q

[0059] Define y q ,y p The error between eq , specifically:

[0060]

[0061] Among them, T a The moment the attack occurred. Satisfy:β(tT a )=1,t≥T a , in other cases, β(tT a )=0;

[0062] (32) According to the attack type of the system, the replay attack signal is defined as:

[0063] a y (t) = β(tT a )(y w (tT)-y w (t))

[0064] Similarly, the DoS attack signal is defined as

[0065]

[0066] in, represents the time sequence of closing / opening transitions caused by DoS attack; τ i ,0<τ i <t i+1 -t i Indicates the length of time, meaning that in the time interval (t i ,t i +τ i ] can communicate normally;

[0067] For hidden / covert attacks, the method defines the attack signal as:

[0068] a y (t) = β(tT a )(y pn (t)-y p (t))

[0069] (33) What is synchronized is the switching time between the watermark generator and the watermark remover, that is, First, the following watermark remover switching time is introduced:

[0070]

[0071] Among them, η>0 is the adjustable parameter of the design, because y p Both w(t) and w(t) can be adjusted by q (t) triggering the switching mechanism in step (32) to achieve synchronization: the watermark generator and the watermark remover Self-trigger synchronization, while A new switching mechanism is driven in the middle, specifically:

[0072]

[0073]

[0074] Among them, α k is an unpredictable random parameter that satisfies Furthermore, we design a watermark generator The switching mechanism is:

[0075]

[0076] (33) For the physical information system and watermark switching trigger mechanism, if θ1, θ2 and η satisfy:

[0077] |2θ1+θ2|≥η+2(|C|σ+|D d |δ d ),

[0078] |2θ1|<η-2(|C|σ+|D d |δ d ),

[0079] |θ2|<η-2(|C|σ+|D d |δ d ),

[0080] In the scenario with or without a fault, the following conclusions can be drawn:

[0081] 1) Watermark generator With Watermark Remover The outer loop trigger mechanism is synchronous and satisfies

[0082] 2) Watermark Generator The inner loop trigger mechanism will not cause the corresponding watermark remover The trigger of the outer loop switching can be obtained:

[0083]

[0084] at the same time,

[0085] (34) The design of the time seed needs to satisfy the following requirements: for any k1≠k2, and Any element in is different. To satisfy the condition that w(t)≠q(t), the watermark generator and watermark remover The asynchrony of the Lorenz system designed by the method has the following mathematical relationship:

[0086]

[0087]

[0088]

[0089] The characteristics of the Lorenz system are determined by the Pointcare mapping Φ:p→Φ(p), where p is the sequence Any point in

[0090] The method designs the key

[0091] On this basis, the following is introduced from A set of initial values and meet Then, combined with the g function of the Lorenz system, the following time seed generator is generated, and its mathematical expression is as follows:

[0092]

[0093] (35) According to the Lorentz time seed In step (34) Initial value set Combined with the time-varying function s k (t), for any k1≠k2, if The choice makes If true, then there exists a non-empty time set This makes the following result hold:

[0094]

[0095] in,

[0096] At this time, if the attack causes asynchrony between w and q, there must be Make Based on a eq As well as the designed parameters θ1, the anomaly detector It can detect a eq , thus, obtaining a time seed that helps threat detection and identification

[0097] It is further pointed out that the method includes the following mathematical relationship:

[0098] (a) In the fault scenario, for all t ≥ T f have When it holds, there exists t>T f , so that satisfy:

[0099]

[0100] Among them, G f (jω)=C(jωI-A r ) -1 (B f +LD f )+D f ;

[0101] (b) In the man-in-the-middle attack scenario (a u =0), there exists t>T a , so that If and only if a eq satisfy

[0102]

[0103] Among them, G a (jω)=C(jωI-A r ) -1 L+I;

[0104] (c) When T a ∈(t w,k-1 ,t w,k ] A replay attack occurred at Then the corresponding time span T satisfies Then the method described will cause when t>Ta When, meet Make Complete the detection of the attack;

[0105] In addition, when T a A DoS attack occurs at the moment. Under the detection mechanism of the method, θ1 satisfies:

[0106]

[0107] Make Complete detection of DoS attacks;

[0108] (d) When T a If an invisible attack occurs at time Make Established, thus, causing the watermark remover to a The outer loop switches at time t, and the watermark generator switches at time t a There is no outer loop switching at all times;

[0109] Consider a covert attack, when t a >T a When the watermark generator and the watermark remover are asynchronous, the method has the following parameters Make Complete the detection of hidden attacks.

[0110] Beneficial effects: Compared with the prior art, the substantial progress and significant effects of the present invention are that the sensor additive switching watermark design method for detecting and identifying physical faults and network attacks described in the present invention establishes a framework that can simultaneously detect and distinguish general physical faults and MITM attacks. On the one hand, the watermark generator and the watermark remover are loaded on the physical end and the network end of the sensor so that they intervene in the data transmission channel in the sensor, which will not cause a decrease in the performance of the control system; on the other hand, the proposed detection and identification mechanism integrates the trigger mechanism of the inner and outer loop switching, which can effectively diagnose man-in-the-middle attacks and identify threats, even the trigger protocol and time seed containing unknown information of the attacker, thereby improving the accuracy of detection and identification of physical faults and network attacks, and enhancing the security and reliability of the system. The detection and identification framework based on additive switching watermarks has a simple structure, and the system parameters can be directly obtained according to inequalities, so it is easy to implement. BRIEF DESCRIPTION OF THE DRAWINGS

[0111] Figure 1 It is a schematic diagram of CPS under MITM attack and physical failure;

[0112] Figure 2 is a schematic diagram of a sensor watermarking method for threat detection and identification;

[0113] Figure 3 It is a graph of the relationship between switching timing and quantity value;

[0114] Figure 4 It is the flow chart of Lorentz time seed generation;

[0115] Figure 5 It is the detection and recognition result: the watermark detection result in the fault scenario;

[0116] Figure 6 These are the detection and recognition results: and the watermark detection results in the replay attack scenario. DETAILED DESCRIPTION

[0117] The specific implementation of the present invention will be further described below in conjunction with the accompanying drawings.

[0118] The present invention provides a sensor additive switching watermark design method for detecting and identifying physical faults and network attacks, including: facing man-in-the-middle attacks represented by denial of service attacks and integrity attacks, the present invention loads a watermark generator and a watermark remover on the physical end and the network end of the sensor respectively, so that they intervene in the data transmission channel in the sensor, and complete the switching between data transmission quality sets according to the trigger protocol and time seed; a trigger mechanism integrating inner and outer loop switching is designed, which can effectively diagnose man-in-the-middle attacks and identify threats, even if the trigger protocol and time seed contain unknown information of the attacker. The diagnostic identification method based on the switching watermark of the present invention has excellent performance in detecting and identifying attacks and faults on the system, even in scenarios containing unknown information, greatly improving the reliability and security of the system.

[0119] A sensor additive switching watermark design method for detecting and identifying physical faults and network attacks. The specific design and implementation process can be described as follows:

[0120] Step 1: According to the watermark method for detecting and identifying system failures and MITM attacks, a two-layer decision strategy is adopted; based on the existing anomaly detectors (failures, attacks, etc.) in the system, the present invention adopts two adaptive thresholds with different amplitudes to correspond to the nominal scenario and the failure scenario respectively;

[0121] Step 2: According to the two-layer decision strategy and detection threshold established in step 1, the watermark generator and watermark remover are loaded on the physical end and network end of the sensor respectively, so that they are inserted into the data transmission channel in the sensor, and the generated additive switching watermark is switched in a set of specially designed mass sets;

[0122] Step 3: Based on the data transmission quality set established in step 2, the switching trigger mechanism of the additive watermark is composed of inner and outer loop switching methods; in the outer loop switching method, the watermark generator and the watermark remover are switched synchronously in the nominal and fault scenarios, while in the man-in-the-middle attack detection, the two are asynchronous; in the inner loop switching method, a set of Lorenz time seeds is generated by the Lorenz system, the purpose of which is to make any element in a time seed different from the element in another time seed; in addition, the switching trigger mechanism designed by the present invention and the generated time seeds both contain unpredictable attacker information.

[0123] Step 4: For the detection and identification of man-in-the-middle attacks and faults, the additive switching watermark method proposed in the present invention can effectively detect and identify, especially in the face of replay attacks, DoS attacks and invisible integrity attacks, even in scenarios containing unknown information, greatly improving the reliability and security of the system.

[0124] Combining the overall steps of the above method, the specific description is as follows.

[0125] Step 1.1: Establish a physical cyber system with man-in-the-middle attacks and physical failures. The system consists of physical devices Actuators, sensors, actuator communication networks and sensor networks, controllers and anomaly detectors The operation of the system is expressed as a set of differential equations as follows:

[0126]

[0127] y p =Cx p +f s (t)+D d d(t),

[0128] y q =y p +a y (t),

[0129] Among them, the following Figure 1 As shown, is the system status, Controller meets is the measured value of the sensor, Sensor measurements received by the controller, also anomaly detector Sensor measurements obtained from the sensor network. A,B,C,B d ,D d is a known system matrix with suitable dimension, and the matrix pair (A, C) satisfies the observability of the system. represents the lumped disturbance and measurement noise, and has an upper bound satisfy t represents time. represents the nonlinear part of the physical information system and satisfies the following Lipschitz condition:

[0130]

[0131] Among them, γ>0 is the Lipschitz constant. and They refer to the man-in-the-middle attack signals received by the actuator and sensor, respectively. u and a y Cooperate to implement a man-in-the-middle attack. and They refer to the physical process fault, actuator fault and sensor fault of the system respectively. In order to simplify the symbols, the present invention defines Further, we can get f p (t)+Bf a (t) = B f f(t),B f =[I,B,0],f s =D f f,D f =[0,0,I]. At the same time,

[0132] Step 1.2: Figure 1 As shown, the model-based anomaly detector The design is as follows:

[0133]

[0134] r=y q -Cx r ,

[0135] in, is the system state quantity of the detector, is the system residual used for system anomaly detection, is the system gain matrix, satisfying becomes the Hurwitz matrix. Further, the following residual system is established:

[0136]

[0137] r=Ce+D d d+D f f+a y .

[0138] in, is the state quantity of the residual system. In the nominal scenario, the system satisfies f(t)=0,a u (t) = 0 and a y (t) = 0, r(t) represents the residual of the system, which satisfies the following adaptive threshold constraint:

[0139]

[0140] Among them, the coefficients κ>0 and μ>0 satisfy in addition, satisfy When in a fault scenario, a u (t)=0,a y (t) = 0, the residual r(t) satisfies The adaptive threshold Specifically

[0141]

[0142]

[0143] Step 1.3: Based on the established anomaly detector And adaptive threshold and Combined with the sensor additive switching watermark, a two-layer decision strategy of the present invention is formed:

[0144] Layer-1: At a certain time t d ,if This indicates that at t d At this moment, the system is attacked, otherwise, it indicates that the system is not attacked;

[0145] Layer-2: When at t d Always present If at t a Always Satisfied And t a >t d , indicating that at t a The system is under attack at the moment, otherwise, the system fails.

[0146] When layer-1 detects that the system is compromised, layer-2 is activated to identify that the system is under attack.

[0147] For step 2, the specific implementation is as follows:

[0148] Step 2.1: According to the two-layer decision strategy and detection threshold established in step 1, the watermark generator and watermark remover are loaded on the physical end and network end of the sensor respectively, so that they are inserted into the data transmission channel in the sensor, and the generated watermark is switched in a set of specially designed quality sets;

[0149] Step 2.2: To achieve the goal of threat detection and identification, according to the two-level decision strategy established in step 1, it is necessary to drive the residual |r(t)| to exceed the threshold and To this end, the present invention switches the watermark to the data y transmitted in the sensor communication channel. p Add watermark;

[0150] Step 2.3: If Figure 2 As shown, the present invention introduces an additive switching watermark mechanism into the sensor data transmission channel. Represents the input of the sensor data transmission network on the physical side. is the output of the network-side sensor data transmission network. The switching watermark fusion mechanism is composed of the switching watermark generator and watermark remover Generate separately and To achieve (ref. Figure 2 ), further, we can get:

[0151]

[0152] Step 2.4: The watermark introduction mechanism designed in step 2.3 does not degrade the performance of the system because the specially designed q can eliminate w The impact brought about by this is that when w = q, y q =y p . Watermark Generator With Watermark Remover Share a set of time seeds. Watermark generator The watermark w is generated by switching between {-θ1,θ1,θ1+θ2,-θ1,-θ2}. The temporal switching sequence of the watermark w is determined by the watermark generator The trigger strategy (including a state-dependent trigger, a time-dependent trigger, and a common time seed) determines the watermark remover. The watermark q is also generated by switching between {-θ1,θ1,θ1+θ2,-θ1,-θ2}, but its time switching sequence is determined by the watermark remover and a common time seed to determine.

[0153] Step 2.5: After determining the generation mechanism of the watermark generator and remover, it is necessary to further clarify the relationship between the generation time and magnitude of w and q. There are N time seeds in the present invention. The time seed refers to a set of time lengths between watermark switching. The k-th time seed is defined as This paper introduces the symbol h∈{w,q} to uniformly describe the watermark generator With Watermark Remover Behavior, definition As the triggering time sequence of the watermark generator w (h = w) and the watermark remover (h = q), we introduce the definition As the k-th time seed time series, satisfying And for any and h∈{w,q} are both fixed values. To represent the switching time sequence triggered by the time seed, forming an inner loop switching mechanism. Correspondingly, the watermark generator and watermark remover The triggered switching mechanism is called the outer loop switching mechanism. and Represents the switching time constant, starting point and switching time of the k-th outer loop switching mechanism. The present invention designs the following switching watermark:

[0154]

[0155] in, and all its elements are positive. For any i∈{1,…,n y}, when h i When (t)≥0, sign(h i (t))=1, otherwise sign(h i (t)) = 1. h(t + ) describes the relationship between the switching timing and quantity value of w, q, as shown below Figure 3 .

[0156] The design of the present invention enables the watermark generator Outer loop trigger mechanism and watermark remover The outer loop trigger mechanism can work synchronously, and the watermark generator The inner loop trigger mechanism is triggered by the time seed, and the time seed cannot be synchronized to trigger the watermark remover This makes the inner loop triggering mechanism of the watermark generator and the watermark remover asynchronous.

[0157] For step 3, the specific implementation includes the following sub-steps:

[0158] Step 3.1: For the watermark detection mechanism (watermark generator and watermark remover ) network control system (such as Figure 2 As shown), we can get y q for:

[0159] yq =y p +w+a y -q

[0160] Furthermore, define y q ,y p The error between eq , specifically:

[0161]

[0162] Among them, T a The moment the attack occurred. Satisfy:β(tT a )=1,t≥T a , in other cases, β(tT a )=0.

[0163] Step 3.2: According to the attack category of the system, the replay attack signal is defined as:

[0164] a y (t) = β(tT a )(y w (tT)-y w (t))

[0165] Similarly, the DoS attack signal is defined as:

[0166]

[0167] in, Represents the time sequence of close / open transitions caused by DoS attack. τ i ,0<τ i <t i+1 -t i Indicates the length of time, meaning that in the time interval (t i ,t i +τ i ] can communicate normally.

[0168] In addition, for hidden / covert attacks, the present invention defines the attack signal as:

[0169] a y (t) = β(tT a )(y pn (t)-y p (t))

[0170] Step 3.3: Based on the switching method designed in step 2, in order to ensure the synchronization of the outer loop switching mechanism, the present invention needs to synchronize the switching time of the watermark generator and the watermark remover. First, the following watermark remover switching time is introduced:

[0171]

[0172] Among them, η>0 is the adjustable parameter of the design. p Both w(t) and w(t) can be adjusted by q (t) Trigger the switching mechanism in step 3.2 to achieve synchronization: the watermark generator and the watermark remover Self-trigger synchronization, while A new switching mechanism is driven in

[0173]

[0174]

[0175] Among them, α k is an unpredictable random parameter that satisfies Furthermore, we design a watermark generator The switching mechanism is:

[0176]

[0177] Step 3.3: For the network system in step 1, Figure 2 The described switching watermark method, combined with the first two steps of step 3, if θ1, θ2 and η satisfy:

[0178] |2θ1+θ2|≥η+2(|C|σ+|D d |δ d ),

[0179] |2θ1|<η-2(|C|σ+|D d |δ d ),

[0180] |θ2|<η-2(|C|σ+|D d |δ d ),

[0181] In the case of a fault or no fault, we can get:

[0182] 1) Watermark generator With Watermark Remover The outer loop trigger mechanism is synchronous and satisfies

[0183] 2) Watermark Generator The inner loop trigger mechanism will not cause the corresponding watermark remover The trigger of the outer loop switching can be obtained:

[0184]

[0185] at the same time,

[0186] Step 3.4: If Figure 4 As shown, to generate the Lorenz time seed, it is necessary to satisfy: for any k1≠k2, and In order to satisfy the condition that w(t)≠q(t), the watermark generator and watermark remover The present invention designs the following Lorenz system:

[0187]

[0188]

[0189]

[0190] The properties of the above Lorenz system are determined by the Pointcare mapping Φ:p→Φ(p), where p is the sequence Any point in .

[0191] Furthermore, the present invention designs a secret key

[0192] On this basis, the following is introduced from A set of initial values and meet Then, combined with the g function of the Lorenz system, the following time seed generator is generated:

[0193]

[0194] Step 3.5: Based on the Lorentz time seed In step 3.4 Initial value set Combined with the time-varying function s k (t), for any k1≠k2, if The choice makes If true, then there exists a non-empty time set This makes the following result hold

[0195]

[0196] in,

[0197] At this time, if the attack causes asynchrony between w and q, there must be Make Based on a eq As well as the designed parameters θ1, the anomaly detector It can detect a eq Therefore, a time seed that helps in threat detection and identification can be obtained.

[0198] Step 4 includes the following sub-steps:

[0199] Step 4.1: Consider the network control system in claim 1, the anomaly detector Threshold and Using the switching watermark h(t + ), and And the parameters θ1, θ2 and η required in step 3.3 are met. In the face of failure and attack scenarios, the present invention can achieve

[0200] 1) In the fault scenario, for all t ≥ T f have When it holds, there exists t>T f , so that satisfy:

[0201]

[0202] Among them, G f (jω)=C(jωI-A r ) -1 (B f +LD f )+D f .

[0203] 2) In the man-in-the-middle attack scenario (a u =0), there exists t>T a , so that If and only if a eq satisfy:

[0204]

[0205] Among them, G a (jω)=C(jωI-A r ) -1 L+I.

[0206] Step 4.2: When T a ∈(t w,k-1 ,t w,k ] a replay attack occurred at Then the corresponding time span T satisfies In the present invention, when t>T aWhen, meet Make Complete the attack detection.

[0207] In addition, when T a A DoS attack occurs at the moment. Under the detection mechanism of the present invention, θ1 satisfies:

[0208]

[0209] Make Complete the detection of DoS attacks.

[0210] Step 4.3: When T a If an invisible attack occurs at time Make Established, thus, causing the watermark remover to a The outer loop switches at time t, and the watermark generator switches at time t a There is no outer loop switching at the moment. Further, consider the covert attack when t a >T a When the watermark generator and the watermark remover are asynchronous, the parameter Make Complete the detection of hidden attacks.

[0211] Example

[0212] This embodiment conducts a simulation experiment on the sensor additive switching watermark design method for detecting and identifying physical faults and network attacks described in the present invention:

[0213] Step 1: Consider the following cyber-physical system:

[0214]

[0215] y p =Cx p +f s (t)+D d d(t),

[0216] y q =y p +a y (t),

[0217] in, (C, A) is the observable matrix. In addition, the nonlinear function g(x) is g(x) = [sin x1, cos(x1+x2), 0.2sin(x1)], the Lipschitz constant γ = 0.2, the system state boundary is set to σ = 20, the interference boundary is set to |d(t)| ≤ 5, and the fault |f(t)| ≤ 10, and the parameter is δd =5 and δ f = 10. Threshold satisfy Anomaly Detector The gain matrix The switching parameter η in the watermark generator and the watermark remover is 80.

[0218] θ1=20,θ2=50. In the Lorenz time seed, the parameters are N=200,n0=20,Δ=1,key0=1,λ=80. In addition, for any Generated watermark Set to 0.2s. At this time, the interference d(t) = [sin(50t), d1(t)] T , where d1(t) is selected from an evenly distributed random distribution in the interval [-2,2]. In addition, the fault occurs T f =10s, the fault signal is set to f p =[1+2sin(15t); 2+sin(25t); 0.5+0.3sin(35t)], f s =[0.5+0.3sin(35t); 1+0.6sin(35t)]f a =[0.2sin(10t)+0.2sin(15t); 0.2+0.1sin(25t); 0.7+0.3sin(20t)].

[0219] Step 2: Use Matlab / Simulink simulation to build a physical information system model and corresponding fault model and attack model in Matlab / Simulink, and perform simulation based on this to verify the algorithms proposed in the present invention.

[0220] According to the above designed parameters, the sensor additive switching watermark design method for detecting and identifying physical faults and network attacks of the present invention is simulated, and it can be obtained that Figure 5 Detection and recognition results: Watermark detection results in fault scenarios and Figure 6 Watermark detection results in the replay attack scenario. It can be observed that the residual r exceeds 15.2s. and remain at that level throughout the simulation time. Therefore, based on the two-layer decision strategy, the threat is detected at 15.2s and identified as a physical fault within the simulation time duration. Figure 5 Medium watermark w The switch of q is kept in sync. In this invention, the replay attack is taken as an example to demonstrate the effectiveness of the designed algorithm. Figure 5 and Figure 6 In the simulation, the replay attack is set up as follows: the simulation records are from Ta - Starting at t = 19s, at T a = 25s, and then the replay attack ends at T a Therefore, T = 6s, we can verify that in the replay attack, the recording duration satisfies After setting the observation matrix gain matrix L, we can get According to the given θ1, we can judge Is satisfied.

[0221] According to the detection and identification results in the replay attack scenario, Figure 6 The results show that the residual r exceeds the threshold at 25.1s. , exceeding the threshold at 25.2s Therefore, based on the two-layer decision strategy, the threat is detected at 25.1s and the threat scenario is identified as an attack at 25.2s. Figure 6 The watermarks w(t) and q(t) in show that before 25 seconds, w and q are synchronized, but at 25 seconds, w and q become asynchronous and no longer synchronized.

[0222] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A sensor additive switching watermark design method for detecting and identifying physical faults and network attacks, characterized in that: The steps include: (1) Establish a physical information system with man-in-the-middle attacks and physical failures, including constructing two adaptive thresholds with different amplitudes for the anomaly detector in the system to correspond to the nominal scenario and the failure scenario respectively, and construct a two-layer decision strategy based on the anomaly detector and adaptive threshold in the system combined with the sensor additive switching watermark; The physical information system is composed of physical devices Actuators, sensors, actuator communication networks and sensor networks, controllers and anomaly detectors constitute; The operation of the physical information system can be expressed as a set of differential equations as follows: y p =Cx p +f s (t)+D d d(t), and q =and p +a y (t), in, is the system status, The controller satisfies u: is the measured value of the sensor, Sensor measurements received by the controller, also anomaly detector Sensor measurements obtained from the sensor network, A, B, C, B d ,D d is a known system matrix with suitable dimension, the matrix pair (A, C) satisfies the observability of the system, represents the lumped disturbance and measurement noise, and has an upper bound satisfy t represents time; Matrix g: Represents the nonlinear part of the physical information system and satisfies the following Lipschitz condition: Among them, γ>0 is the Lipschitz constant, and They refer to the man-in-the-middle attack signals received by the actuator and sensor, respectively. u and a y Cooperate to implement man-in-the-middle attack; and They refer to the physical process faults, actuator faults, and sensor faults to which the system is subjected, respectively; In the physical information system constructed based on the above step (1), the following expression is defined to characterize the differential equations of the system operation: We can get: f p (t)+Bf a (t) = B f f(t),B f =[I,B,0],f s =D f f,D f =[0,0,I], and at the same time: For anomaly detectors The model design expression is as follows: r=y q -Cx r , in, is the system state quantity of the detector, is the system residual used for system anomaly detection, is the system gain matrix, satisfying Become the Hurwitz Matrix; Based on this, the residual system expression established is as follows: r=Ce+D d d+D f f+a y . in, is the state quantity of the residual system; Based on anomaly detector The model expression, in the nominal scenario, the system satisfies f(t)=0,a u (t) = 0 and a y (t) = 0, r(t) represents the residual of the system, which satisfies the following adaptive threshold constraint: Among them, the coefficients κ>0 and μ>0 satisfy |Ce Art |≤κe -μt ,in addition, satisfy In a fault scenario, a u (t)=0,a y (t) = 0, the residual r(t) satisfies The adaptive threshold Specifically (2) According to step (1), a two-layer decision strategy and detection threshold are established, and a watermark generator and a watermark remover are loaded on the physical end and the network end of the sensor respectively, so that they are inserted into the sensor data transmission channel, and the generated additive switching watermark is switched in a set of data transmission quality sets; According to the established anomaly detector And adaptive threshold and Combined with the sensor additive switching watermark, the two-layer decision strategy is as follows: Layer-1: At a certain time t d ,if This indicates that at t d At this moment, the system is attacked, otherwise, it indicates that the system is not attacked; Layer-2: When at t d Always present If at t a Always Satisfied And t a >t d , indicating that at t a The system is attacked at the moment, or vice versa, the system fails; When layer-1 detects that the system is under threat, layer-2 is activated to identify that the system is under attack; (3) Based on the data transmission quality set and the two-layer decision strategy established in step (2), a switching trigger mechanism of the additive watermark is established, wherein the switching trigger mechanism of the additive watermark includes inner loop switching and outer loop switching; In the outer loop switching, the watermark generator and the watermark remover are synchronously switched in the nominal scenario and the fault scenario, and are asynchronously switched in the man-in-the-middle attack detection; In the inner loop switching, a set of Lorentz time seeds is generated by the Lorentz system, in which any element of a time seed is different from an element of another time seed; The switching trigger mechanism of the additive watermark and the generated time seed both contain unpredictable attacker information; Based on the two-layer decision strategy and detection threshold, in order to achieve the goal of threat detection and identification, the method switches the watermark to the data y transmitted in the sensor communication channel. p Add a watermark to drive the residual |r(t)| beyond the threshold and Rather than a failure scenario; The method introduces an additive switching watermark mechanism into the sensor data transmission channel, which specifically includes the following process: Represents the input of the sensor data transmission network on the physical side, The output of the network-side sensor data transmission network, the switching watermark fusion mechanism is composed of the switching watermark generator and watermark remover Generate separately and To achieve this, we can get: Watermark Generator With Watermark Remover Share a set of time seeds; Watermark Generator The watermark w is generated by switching between {-θ1,θ1,θ1+θ2,-θ1,-θ2}, and the time switching sequence of the watermark w is determined by the watermark generator The watermark q corresponding to the watermark remover Q is also generated by switching between {-θ1,θ1,θ1+θ2,-θ1,-θ2}, but its time switching sequence is determined by the watermark remover and common time seeds to determine; The switching trigger mechanism of the additive watermark includes the following calculation process: Define the symbol h∈{w,q} to describe the watermark generator With Watermark Remover Behavior, definition As a watermark generator h=w and watermark remover h = q trigger time sequence, and at the same time, define As the time series of the k-th time seed, And for any and h∈{w,q} are both fixed values. represents the switching time sequence triggered by the time seed, forming an inner-loop switching mechanism; Correspondingly, the watermark generator and watermark remover The triggered switching mechanism is called the outer loop switching mechanism; further, and represents the switching time constant, starting point and switching time of the k-th outer loop switching mechanism. Based on this, the switching expression for the additive watermark in the method is: in, and all its elements are positive, For any i∈{1,Λ,n y }, when h i When (t)≥0, sign(h i (t))=1, otherwise sign(h i (t))=1,h(t + ) describes the relationship between the switching timing and quantity value of w,q; There are N time seeds in the method, and the time seed refers to a set of time lengths between watermark switches. The k-th time seed is defined as The step (3) includes watermark detection, watermark generator and watermark remover The watermark detection mechanism and its corresponding network control system include the following processing and calculation processes: (31) The sensor measurement value y received by the controller q for: y q =y p +w+a y -q Define y q ,y p The error between eq , specifically: Among them, T a is the time when the attack occurs, β: Satisfy:β(tT a )=1,t≥T a , in other cases, β(tT a )=0; (32) According to the attack type of the system, the replay attack signal is defined as: a y (t)=β(t-T a )(y w (t-T)-y w (t)) Similarly, the DoS attack signal is defined as in, represents the time sequence of closing / opening transitions caused by DoS attack; τ i ,0<τ i <t i+1 -t i Indicates the length of time, meaning that in the time interval (t i ,t i +τ i ] can communicate normally; For hidden / covert attacks, the method defines the attack signal as: a y (t)=β(t-T a )(y pn (t)-y p (t)) (33) What is synchronized is the switching time between the watermark generator and the watermark remover, that is, First, the following watermark remover switching time is introduced: Among them, η>0 is the adjustable parameter of the design, because y p Both w(t) and w(t) can be adjusted by q (t) triggering the switching mechanism in step (32) to achieve synchronization: the watermark generator and the watermark remover Self-trigger synchronization, while A new switching mechanism is driven in the middle, specifically: Among them, α k is an unpredictable random parameter that satisfies Furthermore, design a watermark generator The switching mechanism is: (33) For the physical information system and watermark switching trigger mechanism, if θ1, θ2 and η satisfy: |2θ1+θ2|≥η+2(|C|σ+|D d |d d ), |2θ1|<η-2(|C|σ+|D d |d d ), |θ2|<η-2(|C|σ+|D d |d d ), In the scenario with or without a fault, the following conclusions can be drawn: 1) Watermark generator With Watermark Remover The outer loop trigger mechanism is synchronous and satisfies 2) Watermark Generator The inner loop trigger mechanism will not cause the corresponding watermark remover The trigger of the outer loop switching can be obtained: at the same time, (34) The design of the time seed needs to satisfy the following requirements: for any k1≠k2, and Any element in is different. To satisfy the condition that w(t)≠q(t), the watermark generator and watermark remover The asynchrony of the Lorenz system designed by the method has the following mathematical relationship: The characteristics of the Lorenz system are determined by the Pointcare mapping Φ:p→Φ(p), where p is the sequence Any point in The method designs the key On this basis, the following is introduced from A set of initial values and meet Then, combined with the g function of the Lorenz system, the following time seed generator is generated, and its mathematical expression is as follows: (35) According to the Lorentz time seed In step (34) Initial value set Combined with the time-varying function s k (t), for any k1≠k2, if The choice makes Established, Then there exists a non-empty time set This makes the following result hold: in, At this time, if the attack causes asynchrony between w and q, there must be Make Based on a eq As well as the designed parameters θ1, the anomaly detector It can detect a eq , thus, obtaining a time seed that helps threat detection and identification 2. The sensor additive switching watermark design method for detecting and identifying physical failures and network attacks according to claim 1 is characterized in that: The method includes the following mathematical relationship: (a) In the fault scenario, for all t ≥ T f have When it holds, there exists t>T f , so that satisfy: Among them, G f (jω)=C(jωI-A r ) -1 (B f +LD f )+D f ; (b) In the man-in-the-middle attack scenario, there exists t>T a , so that If and only if a eq satisfy Among them, G a (jω)=C(jωI-A r ) -1 L+I; (c) When T a ∈(t w,k-1 ,t w,k ] A replay attack occurred at Then the corresponding time span T satisfies: Then the method described will cause when t>T a When, meet Make Complete the detection of the attack; In addition, when T a A DoS attack occurs at the moment. Under the detection mechanism of the method, θ1 satisfies: Make Complete detection of DoS attacks; (d) When T a If an invisible attack occurs at time Make Established, thus, causing the watermark remover to a The outer loop switches at time t, and the watermark generator switches at time t a There is no outer loop switching at all times; Consider a covert attack, when t a >T a When the watermark generator and the watermark remover are asynchronous, the method has the following parameters Make Complete the detection of hidden attacks.