ICPS comprehensive safety control system and method based on GCRA-LightGBM under D-ADETCS

Abstract

The invention discloses an ICPS comprehensive safety control system and method based on GCRA-Light GBM under D-ADETCS, and the system comprises a sampler which carries out the equal-period sampling of the measurement output data of a sensor; the self-adaptive discrete event trigger A is used for transmitting and screening the sampling values and transmitting the data meeting the data transmission condition A through the sensing side communication network; the data reconstruction restorer A is used for restoring system output data subjected to FDI attack on a sensing side, and the restored data is acted by a robust observer and a controller to obtain a control quantity; the self-adaptive discrete event trigger B is used for carrying out transmission screening on the control quantity and transmitting the control quantity meeting a data transmission condition B through an execution side communication network; and the data reconstruction restorer B is used for restoring the control data subjected to the FDI attack on the execution side, maintaining the restored control quantity by using a zero-order retainer, and acting on the controlled object through the executor to complete the integrated security control of the ICPS.

Description

Technical Field

[0001] The present invention relates to the field of industrial cyber-physical system security technology, and more particularly to an ICPS integrated security control system and method based on GCRA-LightGBM. Background Art

[0002] Industrial Cyber-Physical Systems (ICPS), which deeply integrate communication networks with industrial physical systems, are the mainstream control architecture for modern large-scale, complex industrial systems. They enable core functions such as resource sharing, precise control, and remote collaboration in industrial scenarios, and have become a key technology driving the global transformation to intelligent industry. However, as system deployments expand and control structures become increasingly complex, ICPS faces various risks and challenges in actual operation and maintenance:

[0003] In terms of network security, ICPS faces threats from network attacks such as Denial of Service (DoS), False Data Injection (FDI), and data replay. FDI refers to attackers injecting false data into the system by exploiting vulnerabilities in the data transmission network. Such data can circumvent conventional industrial residual detection mechanisms and maliciously tamper with key data in the system, thereby undermining the integrity of system state estimation and the effectiveness of control decisions. Therefore, building a highly robust intrusion-tolerant control architecture and an active intrusion-tolerant strategy against FDI attacks has become the focus and difficulty of current ICPS network security control research.

[0004] In view of the characteristics of FDI attacks, such as strong temporal and spatial concealment, high detection difficulty and complex defense mechanism, scholars have conducted research from the two dimensions of attack modeling and defense strategy; in terms of attack mechanism, some scholars have constructed a type of attack model that can circumvent the traditional industrial residual detection mechanism, and complete the malicious attack task by destroying the accuracy of system state estimation; in the field of defense technology, some scholars have proposed an active and passive hybrid intrusion tolerance strategy based on the fusion of data-driven and mechanism analysis methods for ICPS attacked by covert FDI; these two studies have deepened the FDI attack and defense theoretical framework from the perspectives of attack mechanism model and defense system construction, respectively. However, in addition to concealment, distributed network attacks and FDI target selectivity also bring a series of huge challenges to intrusion tolerance control research.

[0005] In terms of physical security, functional degradation or failure of some key components in ICPS may lead to risks such as system performance degradation or even global collapse.

[0006] In addition, with the rapid development of ICPS, the complexity of system control decisions continues to increase, and the limited network data transmission bandwidth resources are unable to meet its development needs. To address this problem, some scholars have proposed a communication trigger mechanism solution: by monitoring the system operation status in real time and dynamically deciding the timing of data transmission, an adaptive discrete event triggered communication mechanism (ADETCS) is used, in which the trigger parameters are dynamically adjusted according to the system status. This can effectively alleviate the data transmission pressure of ICPS. However, the traditional single-ended ADETCS has insufficient communication resource configuration optimization capabilities for ICPS with a multi-network node architecture.

[0007] Therefore, how to study a comprehensive security control method that is both fault-tolerant and attack-tolerant for ICPS where multi-source covert FDI attacks and actuator failures coexist, and optimize the configuration and utilization of communication resources in ICPS with a multi-network node architecture, is an urgent problem that technical personnel in this field need to solve. Summary of the Invention

[0008] In view of this, the present invention provides an ICPS integrated safety control system and method based on GCRA-LightGBM under D-ADETCS to solve some of the technical problems mentioned in the background technology.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] An ICPS integrated safety control system based on GCRA-LightGBM under D-ADETCS includes: a controlled object, an intelligent sensor unit, a bilateral communication network, an intelligent control unit, and an intelligent execution unit;

[0011] The intelligent sensing unit includes a sensor, a sampler and an adaptive discrete event trigger A;

[0012] A sampler, used for performing equal-periodic sampling of sensor measurement output data;

[0013] Adaptive discrete event trigger A, used to filter the system output sample values ​​for transmission, and transmit the system output data that meets the data transmission condition A to the intelligent control unit through the communication network on the sensor side;

[0014] The intelligent control unit data includes a data reconstruction repairer A, a robust observer, a controller and an adaptive discrete event trigger B;

[0015] The data reconstruction and repair device A is used to repair the system output data attacked by FDI on the sensor side. The repaired output data is then processed by the robust observer and controller to obtain the control variable.

[0016] Adaptive discrete event trigger B, used to filter the control variables for transmission, and transmit the control variables that meet the data transmission condition B to the intelligent execution unit through the communication network on the execution side;

[0017] The intelligent execution unit includes a data reconstruction repairer B, a zero-order holder and an executor;

[0018] The data reconstruction repairer B is used to repair the control data attacked by FDI on the execution side. It uses the zero-order holder to hold the repaired control quantity and transmits the holding result to the actuator. Finally, it acts on the controlled object to complete the comprehensive security control of ICPS.

[0019] Preferably, the triggering conditions of the adaptive discrete event trigger A and the adaptive discrete event trigger B are specifically:

[0020]

[0021] Among them, Φ A ,Φ B are the event trigger weight matrices of the sensing side and the execution side respectively; σ A (t k h),σ B (t k h) is the distributed adaptive discrete event communication trigger parameter; e) y (i k h)=y(i k h)-y(t k h) is the event triggering error on the sensor side, e u (t k h)=u(t k h)-u(j k h) is the execution side event triggering error; y(i k h),y(t k h) are the current sampling value of the sensor and the system output that meets the trigger condition at the previous moment; u(t k h),u(j k h) are the current control quantity of the controller and the control quantity after the trigger condition is satisfied at the last moment, and i k h=t k h+mh(m∈N),t k h=j k h+nh(n∈N); h is the sampling period;

[0022] Aiming at the non-uniform characteristics of data transmission and resource constraints in the system, the trigger parameters are made dynamic. Taking into account the negative correlation logical relationship between the change rate of the adaptive discrete event communication trigger threshold and the system data fluctuation, a trigger threshold function that is adaptively adjusted according to the system behavior is designed based on the cosine function.

[0023] Preferably, the trigger threshold function of the adaptive discrete event trigger A is specifically:

[0024]

[0025] Among them, e y (t k h)=y(i k h)-y(t k-1 h) is the transmission event error at the sensor end; The upper and lower bounds of the trigger parameters for adaptive discrete event communication at the sensor end; is the allowed transmission event error threshold; α>0 is the sensor error impact factor; z A (t k h)=1-||y(t k-1 h)|| / ||y(t k h)|| is the normalized value of the transmission event error at the sensor end;

[0026] The trigger threshold function of the adaptive discrete event trigger B is specifically:

[0027]

[0028] Among them, e u (j k h)=u(t k h)-u(j k-1 h) is the error in the transmission event of the execution end; The upper and lower bounds of trigger parameters for adaptive discrete event communication on the execution side; is the allowed transmission event error threshold; β>0 is the execution end error impact factor; z B (t k h)=1-||u(j k-1 h)|| / ||u(j k h)|| is the normalized value of the transmission event error at the sensor end;

[0029] Preferably, the back-end output of the system's two-side communication network is specifically:

[0030]

[0031] in, are the communication network backend output signals of the system sensing side and execution side under attack respectively; y(t k h),u(j k h) are the communication network backend output signals for the original system sensing side and execution side respectively; a) y (t kh) is the FDI attack signal on the sensing side; a) u (j k h) is the FDI attack signal on the execution side.

[0032] Preferably, for the FDI attack signal, according to the effectiveness and concealment conditions of the multi-source covert FDI attack, a multi-source covert FDI attack model based on system structure data is constructed, specifically:

[0033]

[0034] in, and are the multiplicative attack information of FDI on the sensing side and the execution side, and are the additive attack information of FDI on the sensing side and the execution side, and These are the progressive attack target selection matrices for FDI on the sensing side and the execution side, respectively.

[0035] Preferably, the specific content of the data reconstruction and repairing device A repairing the system output data of the sensor side that is attacked by FDI is:

[0036] By collecting the system network backend data of the sensor side network that is attacked by FDI And the normal system output data y(t k h) and perform differential operation to complete the sensor side FDI attack information a y (t k h) extraction; As feature data, a y (t k h) Used as target data, it drives the training of the LightGBM attack reconstruction model optimized by the GCRA algorithm, and combines it with the machine learning analysis method to build the data repairer A:

[0037]

[0038] in, Output data for the repaired system, is the reconstructed FDI information;

[0039] During the system operation, the actual network backend data The data repairer A is driven to actively repair the system output data of the sensor side network that is attacked by FDI.

[0040] Preferably, the specific content of the data reconstruction and repairing device B used to repair the control data attacked by FDI on the execution side is:

[0041] By collecting the system network backend data that is attacked by FDI on the execution side network And normal control data u(j k h) and perform differential operation to complete the execution side FDI attack information a u (j k h) extraction; As feature data, a u (j k h) Used as target data, it drives the training of the LightGBM attack reconstruction model optimized by the GCRA algorithm, and combines the machine learning analysis method to build the data reconstruction repairer B:

[0042]

[0043] in, Output data for the repaired system; is the reconstructed FDI information;

[0044] During the system operation, the actual network backend data The data reconstruction repairer B is driven to actively repair the control volume of the execution side network that is attacked by FDI.

[0045] Preferably, the data reconstruction repair error of the data reconstruction repairer Considered as special measurement noise, a robust observer is designed, specifically:

[0046]

[0047] in, are the estimated values ​​of the system state and output respectively, u(t) is the control input vector, A, B, C, E f are matrices of known appropriate dimensions, L2 is the state gain matrix, F is the fault gain matrix, is the fault estimation value, τ A (t) is the data transmission delay of the sensing side system, and τ(t) is the data transmission delay of the entire system.

[0048] Preferably, the controller adopts a dynamic feedback comprehensive safety control strategy, specifically:

[0049]

[0050] in, and is the observation result, K is the controller gain matrix, is the state feedback control quantity, B * is the fault adjustment matrix, satisfying (I-BB * )E f =0; is the active compensation for actuator failure.

[0051] A GCRA-LightGBM-based ICPS integrated safety control method under D-ADETCS, based on the GCRA-LightGBM-based ICPS integrated safety control system under D-ADETCS, comprising:

[0052] The sampler performs isocyclic sampling on the sensor measurement output data;

[0053] The system output sampling values ​​are filtered for transmission, and the system output data that meets the data transmission condition A is transmitted through the communication network on the sensor side;

[0054] Repair the system output data attacked by FDI on the sensor side, and obtain the control quantity after the repaired output data is acted upon by the robust observer and controller;

[0055] Perform transmission screening on the control quantities and transmit the control quantities that meet the data transmission condition B through the communication network on the execution side;

[0056] Repair the control data attacked by FDI on the execution side, maintain the repaired control quantity, transmit it to the actuator to act on the controlled object, and complete the comprehensive security control of ICPS.

[0057] Through the above technical solutions, it can be seen that compared with the existing technology, the present invention discloses an ICPS integrated security control system and method based on GCRA-LightGBM under D-ADETCS, which comprehensively considers the concealment of FDI and the ability to select attack targets in actual industrial scenarios, and constructs a new type of multi-source hidden FDI attack model to study the ICPS integrated security control problem of coexistence of distributed network attacks and actuator failures under the distributed adaptive discrete event triggered communication mechanism D-ADETCS architecture; based on the GCRA algorithm, some key hyperparameters of the LightGBM model are optimized to effectively improve the repair accuracy of the data reconstruction repairer; by integrating data-driven and mechanism analysis methods, a comprehensive security control architecture integrating active fault tolerance, active and passive hybrid intrusion tolerance and communication collaboration is constructed; the negative correlation logical relationship between trigger parameters and system data fluctuations is deeply analyzed, and a new type of adaptive discrete event trigger threshold function based on the cosine law is designed to achieve dynamic balanced regulation between optimized control performance and communication resources; the present invention provides a theoretical basis and technical support for the realization of comprehensive security control with both fault tolerance and intrusion tolerance in complex industrial systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0059] Figure 1 Schematic diagram of an ICPS integrated safety control system based on GCRA-LightGBM under D-ADETCS provided by the present invention;

[0060] Figure 2 A timing diagram of non-uniform data transmission under the dual-terminal ADETCS provided by the present invention;

[0061] Figure 3 Schematic diagram of the data repair process based on GCRA-LightGBM provided by the present invention;

[0062] Figure 4 The hyperparameter optimization flow chart based on GCRA provided by the present invention;

[0063] Figure 5 A schematic diagram of the simulation verification of the multi-source covert FDI attack provided by the present invention;

[0064] Figure 6 A schematic diagram of the system output response based on the existing elastic control method provided by the present invention;

[0065] Figure 7 This is a schematic diagram of the system output response under the ICPS comprehensive security control strategy provided by the present invention. DETAILED DESCRIPTION

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0067] The embodiment of the present invention discloses an ICPS integrated safety control system based on GCRA-LightGBM under D-ADETCS, such as Figure 1 and Figure 2 , including: controlled object, intelligent sensing unit, two-side communication network, intelligent control unit and intelligent execution unit;

[0068] The intelligent sensing unit includes a sensor, a sampler and an adaptive discrete event trigger A;

[0069] A sampler, used for performing equal-periodic sampling of sensor measurement output data;

[0070] Adaptive discrete event trigger A, used to filter the system output sample values ​​for transmission, and transmit the system output data that meets the data transmission condition A to the intelligent control unit through the communication network on the sensor side;

[0071] The intelligent control unit data includes a data reconstruction repairer A, a robust observer, a controller and an adaptive discrete event trigger B;

[0072] The data reconstruction and repair device A is used to repair the system output data attacked by FDI on the sensor side. The repaired output data is then processed by the robust observer and controller to obtain the control variable.

[0073] Adaptive discrete event trigger B, used to filter the control variables for transmission, and transmit the control variables that meet the data transmission condition B to the intelligent execution unit through the communication network on the execution side;

[0074] The intelligent execution unit includes a data reconstruction repairer B, a zero-order holder and an executor;

[0075] The data reconstruction repairer B is used to repair the control data attacked by FDI on the execution side. It uses the zero-order holder to hold the repaired control quantity and transmits the holding result to the actuator. Finally, it acts on the controlled object to complete the comprehensive security control of ICPS.

[0076] Adaptive discrete event triggers A and B constitute D-ADETCS to realize multiple controls on system data transmission; data repairers A and B constitute a distributed active intrusion tolerance control framework to realize active intrusion tolerance control against multi-source covert FDI attacks.

[0077] In this embodiment, the controlled object is first described. Specifically, for a continuous-time ICPS with actuator faults, external disturbances, and noise, it can be expressed as:

[0078]

[0079] Where, is the system state variable, is the control input vector, is the output vector measured and sampled by the sensor; and are measurement noise and external disturbance, respectively; The actuator is a continuous time-varying fault, and the derivative range of f(t) is bounded, that is, there is a constant f1 such that A,B,C,E f ,E w ,Ev are matrices of known appropriate dimensions.

[0080] Under the combined action of the sampler and D-ADETCS, the data transmission in the closed-loop control system is non-uniform, while the controlled object is continuous. To address the problem of non-uniform data transmission, a time delay function is defined:

[0081] τ A (t) = tt k h,t∈[t k h,t k+1 h]

[0082] τ B (t) = t k hj k h,t k h∈[j k h,j k+1 h]

[0083] τ(t)=tj k h,t∈[j k h,j k+1 h]

[0084] Where, τ A (t) is the data transmission delay of the sensor side system, τ B (t) is the data transmission delay of the execution side system, τ(t) is the data transmission delay of the entire system, and τ(t)∈[h,h1), h1 is the data transmission delay of the D-ADETCS. The maximum allowed delay is determined jointly.

[0085] In order to further implement the above technical solution, the triggering conditions of the adaptive discrete event trigger A and the adaptive discrete event trigger B are specifically as follows:

[0086]

[0087] Among them, Φ A ,Φ B are the event trigger weight matrices of the sensing side and the execution side respectively; σ A (t k h),σ B (t k h) is the distributed adaptive discrete event communication trigger parameter; e) y (i k h)=y(i k h)-y(t k h) is the event triggering error on the sensor side, e u (t k h)=u(t k h)-u(j kh) is the execution side event triggering error; y(i k h),y(t k h) are the current sampling value of the sensor and the system output that meets the trigger condition at the previous moment; u(t k h),u(j k h) are the current control quantity of the controller and the control quantity after the trigger condition is satisfied at the last moment, and i k h=t k h+mh(m∈N),t k h=j k h+nh(n∈N); h is the sampling period;

[0088] Aiming at the non-uniform characteristics of data transmission and resource constraints in the system, the trigger parameters are made dynamic. Taking into account the negative correlation logical relationship between the change rate of the adaptive discrete event communication trigger threshold and the system data fluctuation, a trigger threshold function that is adaptively adjusted according to the system behavior is designed based on the cosine function.

[0089] In order to further implement the above technical solution, the trigger threshold function of the adaptive discrete event trigger A is specifically:

[0090]

[0091] Among them, e y (t k h)=y(i k h)-y(t k-1 h) is the transmission event error at the sensor end; The upper and lower bounds of the trigger parameters for adaptive discrete event communication at the sensor end; is the allowed transmission event error threshold; α>0 is the sensor error impact factor; z A (t k h)=1-||y(t k-1 h)|| / ||y(t k h)|| is the normalized value of the transmission event error at the sensor end;

[0092] The trigger threshold function of the adaptive discrete event trigger B is specifically:

[0093]

[0094]

[0095] Among them, e u (j k h)=u(t k h)-u(j k-1 h) is the error in the transmission event of the execution end; The upper and lower bounds of trigger parameters for adaptive discrete event communication on the execution side; is the allowed transmission event error threshold; β>0 is the execution end error impact factor; z B (t k h)=1-||u(j k-1 h)|| / ||u(j k h)|| is the normalized value of the transmission event error at the sensor end;

[0096] In this embodiment, the trigger threshold functions of the adaptive discrete event triggers A and B established based on the mathematical characteristics of the cosine function both take into account the negative correlation logical relationship between the communication trigger threshold change rate and the system data fluctuation, that is, when the system data fluctuation increases, the communication trigger threshold decreases faster; when the system data fluctuation is smaller, the communication trigger parameter decreases slower.

[0097] Specifically, when the system data transmission error is too large, that is, The trigger parameters are The restriction on data transmission bandwidth is relaxed to a great extent, so that the system enters a high-frequency data transmission state, thereby improving the system control accuracy; when the system data transmission error is too small, in order to solve the monitoring failure problem that may be caused when the system data transmission error is lower than the set threshold, the data transmission delay is strictly limited to a controllable range by presetting the upper limit of the trigger parameter, thereby effectively avoiding the risk of misjudgment of the control unit due to failure of the transmission channel; when the data transmission error is within the preset range, the correlation between the normalized value of the transmission event error and the trigger threshold is adjusted by the preset error influence factor, and a nonlinear trigger parameter adjustment mechanism is constructed based on the cosine law to realize the reasonable regulation of the trigger threshold adjustment rate with data fluctuations. When the normalized value of the transmission event error exceeds the critical threshold, Λ A (t k h) Set to zero to make the system enter a high-frequency data transmission state to increase the overall adjustment efficiency of the system.

[0098] In order to further implement the above technical solution, the back-end output of the communication network on both sides of the system is specifically as follows:

[0099]

[0100] in, are the communication network backend output signals of the system sensing side and execution side under attack respectively; y(t k h),u(j k h) are the communication network backend output signals for the original system sensing side and execution side respectively; a) y (t k h) is the FDI attack signal on the sensing side; a) u (j kh) is the FDI attack signal on the execution side.

[0101] In order to further implement the above technical solution, based on the conventional industrial residual detection mechanism, the following observer with fault tolerance and anti-interference capabilities is preset:

[0102]

[0103] in, are the estimated values ​​of the system state and output respectively; L1 represents the conventional state gain matrix to be solved.

[0104] The normal system state estimation error is And the system state estimation error under attack is To ensure the effectiveness of FDI, the state estimation error components introduced by multi-source covert FDI attacks Not less than the error threshold e allowed when the system is running stably th :

[0105]

[0106] The system output residual is To ensure that FDI meets the attack concealment, the system output residual of the multi-source covert FDI attack is Does not exceed the residual threshold J set when the system is running normally and stably th :

[0107]

[0108] According to the effectiveness and concealment conditions of multi-source covert FDI attacks, a multi-source covert FDI attack model based on system structure data is constructed as follows:

[0109]

[0110]

[0111] in, and are the multiplicative attack information of FDI on the sensing side and the execution side, and are the additive attack information of FDI on the sensing side and the execution side, and These are the progressive attack target selection matrices for FDI on the sensing side and the execution side, respectively.

[0112] In order to further implement the above technical solutions, Figure 3 ,The specific content of the system output data that the data reconstruction repairer A repairs on the sensor side that is attacked by FDI is:

[0113] By collecting the system network backend data of the sensor side network that is attacked by FDI And the normal system output data y(t k h) and perform differential operation to complete the sensor side FDI attack information a y (t k h) extraction; As feature data, a y (t k h) Used as target data, it drives the training of the LightGBM attack reconstruction model optimized by the GCRA algorithm, and combines it with the machine learning analysis method to build the data repairer A:

[0114]

[0115] in, Output data for the repaired system, is the reconstructed FDI information;

[0116] During the system operation, the actual network backend data The data repairer A is driven to actively repair the system output data of the sensor side network that is attacked by FDI.

[0117] To further implement the above technical solution, the data reconstruction and repair device B is used to repair the control data attacked by FDI on the execution side. The specific content is:

[0118] By collecting the system network backend data that is attacked by FDI on the execution side network And normal control data u(j k h) and perform differential operation to complete the execution side FDI attack information a u (j k h) extraction; As feature data, a u (j k h) Used as target data, it drives the training of the LightGBM attack reconstruction model optimized by the GCRA algorithm, and combines it with the machine learning analysis method to build the data reconstruction repairer B:

[0119]

[0120] in, Output data for the repaired system; is the reconstructed FDI information;

[0121] During the system operation, the actual network backend data The data reconstruction repairer B is driven to actively repair the control volume of the execution side network that is attacked by FDI.

[0122] In an embodiment, Figure 4 , the specific content of training the LightGBM attack reconstruction model optimized by the GCRA algorithm is:

[0123] Based on the GCRA algorithm, some key hyperparameters of LightGBM (learning rate, number of decision trees, maximum depth of decision trees and number of leaf nodes) are optimized, such as Figure 4 As shown in Figure 2, the optimization process of GCRA is divided into two stages: search and development:

[0124] During the search phase, the cane rat population (X) sets out from different nests to forage for food. Each cane rat plans its own search trajectory based on the path information left by the dominant male rat. The mathematical expression is:

[0125]

[0126] Where x i,j represents the individual mouse at the i-th position in the j-th dimension, Indicates the new individual mouse position, x k,j represents the position of the dominant male rat, C is a random number defined within the boundary of the problem space, and r is the coefficient that simulates the reduction of food sources.

[0127] After each search, the system evaluates the fitness of each male mouse to update the dominant male mouse, achieving global traversal within the search space and ensuring that the algorithm can find the potential optimal solution. Its mathematical expression is:

[0128]

[0129] Where, X i represents the i-th upcoming or new cane rat population state, F i represents the target fitness value of the dominant mouse, and α and β are preset parameters.

[0130] During the development phase, male mice will leave the group, and the remaining mice will gather in areas with abundant food resources to forage. During this process, the mice will randomly select the location of a female mouse to determine a new foraging location and forage around it. This mechanism enables the algorithm to conduct detailed exploration within a specific area, thereby achieving local optimization. Its mathematical expression is:

[0131]

[0132] Where x m,j Indicates the position of the selected female mouse, and u is the preset parameter.

[0133] In order to further implement the above technical solution, the data reconstruction repair error of the data reconstruction repairer Considered as special measurement noise, a robust observer is designed, specifically:

[0134]

[0135] in, are the estimated values ​​of the system state and output respectively, u(t) is the control input vector, A, B, C, E f are matrices of known appropriate dimensions, L2 is the state gain matrix, F is the fault gain matrix, is the fault estimation value, τ A (t) is the data transmission delay of the sensing side system, and τ(t) is the data transmission delay of the entire system.

[0136] In this embodiment, the design method of the robust observer is:

[0137] Reconstructing multi-source covert FDI attack information via GCRA-LightGBM and Based on this information, the system data attacked by FDI was proactively repaired. The output of the network backend after the repair is described as follows:

[0138]

[0139] make are the data reconstruction and repair errors at the sensor end and the execution end, respectively. These errors exist objectively during the system operation and are uncertain. Therefore, they can be regarded as special measurement noise. Combining the system description and the description of the network backend output after repair, the following system description can be obtained:

[0140]

[0141] definition The following augmented error system description can be obtained:

[0142]

[0143] In order to facilitate calculation, let Integrate the above error systems into an augmented error system:

[0144]

[0145] Where,

[0146]

[0147] Furthermore, the following Lyapunov-Krasovskii functional is constructed:

[0148]

[0149] in,

[0150]

[0151] According to Lyapunov's second method, the state and fault gain matrices L2 and F are solved so that the augmented error system is asymptotically stable in the absence of disturbances and meets the performance index in the presence of disturbances:

[0152]

[0153] In this embodiment, for the data reconstruction and repair errors under multi-source covert FDI attacks, when designing the observer, based on the machine-learning analysis method, the data reconstruction and repair errors are regarded as a special noise and are processed through a robust response strategy, thereby achieving a clever fusion of data-driven and machine-learning analysis methods; at the same time, a double γ constraint mechanism is introduced to reduce the multi-source data reconstruction and repair errors. and Targeted constraints are imposed separately to further enhance ICPS's defense capabilities against multi-source covert FDI attacks.

[0154] Since the observation data after attack compensation contains sensor-end data reconstruction and repair errors, although the data reconstruction and repair device and the robust observer have implemented reconstruction, repair and robustness effects on multi-source covert FDI attacks, there is still residual attack energy that has not been processed cleanly.

[0155] To further implement the above technical solutions, the controller design not only has fault tolerance capabilities, but also implicitly has the ability to tolerate multi-source FDI attacks. Taking both fault tolerance and intrusion tolerance into account, a dynamic feedback integrated security control strategy is adopted. Specifically:

[0156]

[0157] in, and is the observation result, K is the controller gain matrix, is the state feedback control quantity, B * is the fault adjustment matrix, satisfying (I-BB * )E f =0; is the active compensation for actuator failure.

[0158] The ICPS closed-loop model is further obtained as:

[0159]

[0160] in, Reconstruct and repair the error for the execution end data; state estimation error e x (t-τ A (t)) and the fault estimation error ef (t-τ A (t)) all converge asymptotically; In order to describe the characteristics of continuous fault sampling, the cross-term inequality technique is used to deal with it in order to obtain less conservative results.

[0161] Furthermore, the following Lyapunov-Krasovskii functional is constructed:

[0162]

[0163] in,

[0164]

[0165] According to Lyapunov's second method, the control gain matrix K and the event trigger weight matrix Φ are obtained. A ,Φ B The closed-loop control system is asymptotically stable in the absence of disturbances and meets the performance indicators when there are disturbances:

[0166]

[0167] In another embodiment, the effectiveness and performance advantages of the present invention are verified based on a typical industrial four-tank simulation example, specifically:

[0168] (1) Simulate and verify the effectiveness, concealment and target selectivity of multi-source covert FDI, such as Figure 5 As shown, Figure 5 Figure (a) shows the changing curve of the state estimation error component introduced by the multi-source covert FDI attack, which is used to verify the effectiveness of the multi-source covert FDI. Figure 5 Figure (b) shows the change curve of the system output residual under the multi-source covert FDI attack, which is used to verify the concealment of multi-source covert FDI;

[0169] When 200s≤t≤650s, the system is attacked by multi-source covert FDI. The internal state estimation error component of the system under attack is significantly higher than the error threshold (0.2), while the system output residuals are all lower than the residual threshold (0.47). This shows that the FDI model preset by the present invention not only meets the requirements of concealment, but also achieves the expected attack effect.

[0170] Figure 5 Figure (c) and Figure 5 Figure (d) shows the change curves of the front-end / back-end output errors of the communication network on the sensor side and the execution side of the system under the multi-source covert FDI attack, which is used to verify the attack target selectivity of the multi-source covert FDI;

[0171] When 200s≤t≤400s, the execution-side network of the system is attacked by FDI and exhibits the ability to expand progressively with different output channels. When 250s≤t≤650s, the sensing-side network of the system is also invaded by FDI and also exhibits the ability to expand progressively with different output channels.

[0172] (2) In order to reflect the performance advantages of the D-ADETCS designed in the present invention, Table 1 shows the system data transmission under different triggering schemes DETCS, ADETCS and D-ADETCS. It can be seen that the data transmission volume under the D-ADETCS of the present invention is lower, and the energy saving rate (η = (Nm) / N×100%, N is the total number of data packets, m is the data transmission volume) is significantly improved compared with the control scheme, indicating that D-ADETCS has a better effect in optimizing the dynamic balance between the control performance and communication resource allocation of ICPS with multiple network nodes;

[0173] Table 1 System data transmission under different triggering schemes:

[0174]

[0175] (3) To verify the effectiveness and advantages of the comprehensive security control strategy constructed by the present invention, Figure 6 The system output response curve based on the elastic control method is given. Figure 7 The system output response curve of the present invention is given, and it can be seen that Figure 6 The system output shows obvious fluctuations and cannot be maintained within the controllable fluctuation range; analysis Figure 7 It can be obtained that: when t<200s, the system output response quickly tends to a stable state; when 200≤t≤250s, the system execution side network is attacked by FDI, and the system output response is in a stable state with no obvious fluctuation; when 250≤t≤300s, the system bilateral network is attacked by FDI, and the system output response fluctuates slightly and is in a stable state; when 300≤t≤400s, the system bilateral network is attacked by FDI and an actuator failure occurs, and the system output response fluctuates, and the system output response at t=300s suddenly changes to -0.12, with a small fluctuation amplitude, and the impact on the system can be ignored; when 400≤t≤650s, the system sensor side network is attacked by FDI and affected by the actuator failure, and the system output response fluctuation is within ±0.06, and the impact on the system can also be ignored; when 650≤t≤800s, the system is only affected by the actuator failure, and the system output response has no obvious fluctuation and is in a stable state.

[0176] A GCRA-LightGBM-based ICPS integrated safety control method under D-ADETCS is based on a GCRA-LightGBM-based ICPS integrated safety control system under D-ADETCS, comprising:

[0177] The sampler performs isocyclic sampling on the sensor measurement output data;

[0178] The system output sampling values ​​are filtered for transmission, and the system output data that meets the data transmission condition A is transmitted through the communication network on the sensor side;

[0179] Repair the system output data attacked by FDI on the sensor side, and obtain the control quantity after the repaired output data is acted upon by the robust observer and controller;

[0180] Perform transmission screening on the control quantities and transmit the control quantities that meet the data transmission condition B through the communication network on the execution side;

[0181] Repair the control data attacked by FDI on the execution side, maintain the repaired control quantity, transmit it to the actuator to act on the controlled object, and complete the comprehensive security control of ICPS.

[0182] In this embodiment, while deploying an adaptive discrete event trigger mechanism on the sensing end, an adaptive discrete event trigger mechanism is also introduced on the execution end. D-ADETCS screens the system's output data and control quantity respectively, thereby further alleviating the contradiction between the massive control data transmission and communication resources of each network node of the system while ensuring the control performance; according to the characteristics of actual FDI, the mechanism of FDI intercepting system data and injecting false data is simulated in the communication network module, aiming to achieve more practical FDI intrusion tolerance control through in-depth simulation and analysis of typical FDI behaviors; for multi-source covert FDI attacks, the first data reconstruction and repairer and the second data reconstruction and repairer based on GCRA-LightGBM are introduced in the system control unit and the execution unit respectively, so as to reconstruct the covert FDI attacks appearing in each network node of the system, and repair the data under attack with the attack reconstruction information, in order to achieve active intrusion tolerance control of multi-source covert FDI attacks.

[0183] A processing terminal includes a memory and a processor. The memory stores a computer program that can be run on the processor. When the processor executes the computer program, an ICPS integrated security control method based on GCRA-LightGBM is implemented.

[0184] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0185] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An ICPS integrated safety control system based on GCRA-LightGBM under D-ADETCS, characterized by: include: Controlled object, intelligent sensing unit, bilateral communication network, intelligent control unit and intelligent execution unit; The intelligent sensing unit includes a sensor, a sampler and an adaptive discrete event trigger A; A sampler, used for performing equal-periodic sampling of sensor measurement output data; Adaptive discrete event trigger A, used to filter the system output sample values ​​for transmission, and transmit the system output data that meets the data transmission condition A to the intelligent control unit through the communication network on the sensor side; The intelligent control unit data includes a data reconstruction repairer A, a robust observer, a controller and an adaptive discrete event trigger B; The data reconstruction and repair device A is used to repair the system output data attacked by FDI on the sensor side. The repaired output data is then processed by the robust observer and controller to obtain the control variable. Adaptive discrete event trigger B, used to filter the control variables for transmission, and transmit the control variables that meet the data transmission condition B to the intelligent execution unit through the communication network on the execution side; The intelligent execution unit includes a data reconstruction repairer B, a zero-order holder and an executor; The data reconstruction repairer B is used to repair the control data attacked by FDI on the execution side. It uses the zero-order holder to hold the repaired control quantity and transmits the holding result to the actuator. Finally, it acts on the controlled object to complete the comprehensive security control of ICPS.

2. The ICPS integrated safety control method based on GCRA-LightGBM under D-ADETCS according to claim 1 is characterized in that: The triggering conditions of the adaptive discrete event trigger A and the adaptive discrete event trigger B are specifically as follows: Among them, Φ A ,Φ B are the event trigger weight matrices of the sensing side and the execution side respectively; σ A (t k h),σ B (t k h) is the distributed adaptive discrete event communication trigger parameter; e) y (i k h)=y(i k h)-y(t k h) is the event triggering error on the sensor side, e u (t k h)=u(t k h)-u(j k h) is the execution side event triggering error; y(i k h),y(t k h) are the current sampling value of the sensor and the system output that meets the trigger condition at the previous moment; u(t k h),u(j k h) are the current control quantity of the controller and the control quantity after the trigger condition is satisfied at the last moment, and i k h=t k h+mh(m∈N),t k h=j k h+nh(n∈N); h is the sampling period; Aiming at the non-uniform characteristics of data transmission and resource constraints in the system, the trigger parameters are made dynamic. Taking into account the negative correlation logical relationship between the change rate of the adaptive discrete event communication trigger threshold and the system data fluctuation, a trigger threshold function that is adaptively adjusted according to the system behavior is designed based on the cosine function.

3. The ICPS integrated safety control method based on GCRA-LightGBM under D-ADETCS according to claim 2 is characterized in that: The trigger threshold function of the adaptive discrete event trigger A is specifically: Among them, e y (t k h)=y(i k h)-y(t k-1 h) is the transmission event error at the sensor end; The upper and lower bounds of the trigger parameters for adaptive discrete event communication at the sensor end; is the allowed transmission event error threshold; α>0 is the sensor error impact factor; z A (t k h)=1-||y(t k-1 h)|| / ||y(t k h)|| is the normalized value of the transmission event error at the sensor end; The trigger threshold function of the adaptive discrete event trigger B is specifically: Among them, e u (j k h)=u(t k h)-u(j k-1 h) is the error in the transmission event of the execution end; The upper and lower bounds of trigger parameters for adaptive discrete event communication on the execution side; is the allowed transmission event error threshold; β>0 is the execution end error impact factor; z B (t k h)=1-||u(j k-1 h)|| / ||u(j k h)|| is the normalized value of the transmission event error at the sensor end; 4. The ICPS integrated safety control method based on GCRA-LightGBM under D-ADETCS according to claim 1 is characterized in that: The back-end output of the system's two-side communication network is specifically: in, are the communication network backend output signals of the system sensing side and execution side under attack respectively; y(t k h),u(j k h) are the communication network backend output signals for the original system sensing side and execution side respectively; a) y (t k h) is the FDI attack signal on the sensing side; a) u (j k h) is the FDI attack signal on the execution side.

5. The ICPS integrated safety control method based on GCRA-LightGBM under D-ADETCS according to claim 1 is characterized in that: For FDI attack signals, according to the effectiveness and concealment conditions of multi-source covert FDI attacks, a multi-source covert FDI attack model based on system structure data is constructed, specifically: in, and are the multiplicative attack information of FDI on the sensing side and the execution side, and are the additive attack information of FDI on the sensing side and the execution side, and These are the progressive attack target selection matrices for FDI on the sensing side and the execution side, respectively.

6. The ICPS integrated safety control method based on GCRA-LightGBM under D-ADETCS according to claim 1 is characterized in that: The specific content of the data reconstruction and repair device A repairing the system output data of the sensor side that has been attacked by FDI is as follows: By collecting the system network backend data of the sensor side network that is attacked by FDI And the normal system output data y(t k h) and perform differential operation to complete the sensor side FDI attack information a y (t k h) extraction; As feature data, a y (t k h) Used as target data, it drives the training of the LightGBM attack reconstruction model optimized by the GCRA algorithm, and combines it with the machine learning analysis method to build the data repairer A: in, Output data for the repaired system, is the reconstructed FDI information; During the system operation, the actual network backend data The data repairer A is driven to actively repair the system output data of the sensor side network that is attacked by FDI.

7. The ICPS integrated safety control method based on GCRA-LightGBM under D-ADETCS according to claim 1 is characterized in that: The specific content of the data reconstruction repairer B used to repair the control data attacked by FDI on the execution side is: By collecting the system network backend data that is attacked by FDI on the execution side network And normal control data u(j k h) and perform differential operation to complete the execution side FDI attack information a u (j k h) extraction; As feature data, a u (j k h) Used as target data, it drives the training of the LightGBM attack reconstruction model optimized by the GCRA algorithm, and combines the machine learning analysis method to build the data reconstruction repairer B: in, Output data for the repaired system; is the reconstructed FDI information; During the system operation, the actual network backend data The data reconstruction repairer B is driven to actively repair the control volume of the execution side network that is attacked by FDI.

8. The ICPS integrated safety control method based on GCRA-LightGBM under D-ADETCS according to claim 5 is characterized in that: Reconstruct the data of the data repairer to repair the error Considered as special measurement noise, a robust observer is designed, specifically: in, are the estimated values ​​of the system state and output respectively, u(t) is the control input vector, A, B, C, E f are matrices of known appropriate dimensions, L2 is the state gain matrix, F is the fault gain matrix, is the fault estimation value, τ A (t) is the data transmission delay of the sensing side system, and τ(t) is the data transmission delay of the entire system.

9. The ICPS integrated safety control method based on GCRA-LightGBM under D-ADETCS according to claim 1 is characterized in that: The controller adopts a dynamic feedback comprehensive safety control strategy, specifically: in, and is the observation result, K is the controller gain matrix, is the state feedback control quantity, B * is the fault adjustment matrix, satisfying (I-BB * )E f =0; is the active compensation for actuator failure.

10. A GCRA-LightGBM-based ICPS integrated security control method under D-ADETCS, characterized by: An ICPS integrated safety control system based on GCRA-LightGBM under D-ADETCS according to any one of claims 1 to 9, comprising: The sampler performs isocyclic sampling on the sensor measurement output data; The system output sampling values ​​are filtered for transmission, and the system output data that meets the data transmission condition A is transmitted through the communication network on the sensor side; Repair the system output data attacked by FDI on the sensor side, and obtain the control quantity after the repaired output data is acted upon by the robust observer and controller; Perform transmission screening on the control quantities and transmit the control quantities that meet the data transmission condition B through the communication network on the execution side; Repair the control data attacked by FDI on the execution side, maintain the repaired control quantity, transmit it to the actuator to act on the controlled object, and complete the comprehensive security control of ICPS.

Images