A CPS system remote state estimation method, device and readable storage medium
Through K-means and GMM clustering method, Kalman filter and core principal component analysis method, the weight and distribution probability of the CPS system sensor data are calculated, and the problem of being unable to accurately estimate the system status in the existing technology is solved, and the system status is realized Reliable estimation of system status and operational parameter adjustment in attack situations.
Patent Information
- Application Number
- CN202410809444.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-21
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-06-21
AI Technical Summary
The existing remote state estimation method of CPS system cannot obtain relatively accurate and reliable system status based on the revised data after detecting an attack, resulting in the control system being unable to adjust the CPS operating parameters, affecting production safety and efficiency.
The sensor data is clustered using K-means clustering method and GMM clustering method, and the distribution probability density function and mixing coefficient of sensor data under the Gaussian distributed cluster are calculated, and the weight of each sensor is calculated based on the weight and measurement data calculation system. State, data fusion and dimensionality reduction are used using Kalman filter and core principal component analysis method or variational autoencoder.
Through weight calculation and data fusion, the credibility and accuracy of sensor data are improved, and the current operating status of the system can be reflected more accurately, providing a reliable basis for adjusting system operating parameters, and enhancing production safety and efficiency.
Smart Images

Figure CN118400182B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of attack detection technology, and in particular to a CPS system remote state estimation method, device and computer-readable storage medium. Background Art
[0002] Cyber-physical systems (CPS) consist of actuators, sensors, controllers, and controlled objects, and use the connectivity of physical components and networks to collaboratively transmit and process data. CPS systems seamlessly integrate sensing, control, communication, and computing, and are applied to multiple fields such as smart grids, smart transportation, and environmental monitoring. For example, in power systems, generators, transformers, transmission lines, switchgear, sensors, and actuators are physical components, and computers, servers, and data storage devices are computing components. The communication network connects the physical components and computing components together to achieve real-time data transmission and sharing, thereby realizing the detection, control, optimization, and protection of the power system. As the application of CPS systems becomes more and more widespread, their security issues are also receiving increasing attention.
[0003] The existing remote state estimation method of CPS system first converts the measurement data of each sensor in the CPS system into system state estimation data, and then clusters the system state estimation data corresponding to each sensor, thereby dividing the sensors into attacked clusters and non-attacked clusters. By detecting whether the sensors in the CPS system are attacked, it is determined whether the CPS system is under malicious attack.
[0004] However, for some CPS systems, it is not only necessary to detect whether the system is under malicious attack, but also to be able to fuse the CPS system status data based on the measurement data of each sensor when some sensors are attacked and the data are tampered with. For example, in industrial production CPS systems, real-time and accuracy are key factors to ensure production safety and improve production efficiency. When the industrial control system is under malicious attack, the attacker will tamper with some sensor data in the system, causing the control system to receive wrong information and make wrong decisions. For example, in the chemical production process, when key parameters such as temperature, pressure or flow are tampered with, if the relatively accurate and reliable system status cannot be obtained through the tampered data, the control system will not be able to accurately judge the current production status and equipment operation status, and thus cannot adjust production parameters such as temperature and pressure, and may even cause equipment damage, production accidents or casualties.
[0005] In summary, the existing CPS remote system state estimation method can only determine whether the CPS system is under malicious attack by detecting whether the data of the sensors in the CPS system has been tampered with. It is unable to obtain a relatively accurate and reliable system state based on the tampered data, which results in the control system being unable to adjust the CPS system operating parameters based on the system state. Summary of the invention
[0006] To this end, the technical problem to be solved by the present invention is to overcome the problem that the CPS system remote estimation method in the prior art cannot obtain a relatively accurate and reliable system status based on the tampered data after detecting that the CPS system is attacked, thereby causing the control system to be unable to adjust the CPS operating parameters based on the system status.
[0007] In order to solve the above technical problems, the present invention provides a remote state estimation method for a CPS system, comprising:
[0008] Based on the CPS system to be detected The measurement data of each sensor at the moment is obtained CPS system state estimation data of each sensor at each moment;
[0009] Using K-means clustering method The CPS system state estimation data of all sensors at all times are clustered to obtain the attacked cluster and the unattacked cluster;
[0010] The cluster centers of the attacked cluster and the unattacked cluster are used as the initial values of the cluster centers, and the GMM clustering method is used to The CPS system state estimation data of all sensors at the moment are re-clustered to obtain the attacked Gaussian distribution clusters and their mixing coefficients, as well as the unattacked Gaussian distribution clusters and their mixing coefficients;
[0011] calculate The distribution probability density function of the CPS system state estimation data of each sensor at the moment under the attacked Gaussian distribution cluster and the first product of the mixing coefficient of the attacked Gaussian distribution cluster, and The distribution probability density function of the CPS system state estimation data of the sensor at the moment under the unattacked Gaussian distribution cluster and the second product of the mixing coefficient of the unattacked Gaussian distribution cluster; the proportion of the second product to the sum of the first product and the second product is taken as The CPS system state estimation data of the sensor at the moment belongs to the weight of the Gaussian distribution cluster that has not been attacked;
[0012] based on The CPS system state estimation data of each sensor at the moment belongs to the weight of the Gaussian distribution cluster that has not been attacked, The measurement data of each sensor at all times and State estimation data calculation of the CPS system to be detected at any time State estimation data of the CPS system to be detected at all times.
[0013] Preferably, The calculation formula for the weight of the CPS system state estimation data of the sensor at the moment belonging to the Gaussian distribution cluster that has not been attacked is:
[0014] ,
[0015] in, express Moment The CPS system state estimation data of each sensor belongs to the weight of the Gaussian distribution cluster that has not been attacked; Express The mixing coefficient of the unattacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; express Moment The distribution probability density function of the CPS system state estimation data of the sensors under the Gaussian distribution cluster without attack; express Moment CPS system state estimation data of sensors; Express The mean of the unattacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; express Moment The estimated error covariance matrix of the sensors; Express The mixing coefficient of the attacked Gaussian distribution cluster is obtained by re-clustering the CPS system state estimation data of all sensors at the moment; express Moment The distribution probability density function of the CPS system state estimation data of each sensor under the attacked Gaussian distribution cluster; Express The mean of the attacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; Express The covariance of the attacked Gaussian distribution cluster is obtained by re-clustering the CPS system state estimation data of all sensors at the moment.
[0016] Preferably, The state estimation data of the CPS system to be detected at any time is expressed as:
[0017] ,
[0018] in, express State estimation data of the CPS system to be tested at all times; express State estimation data of the CPS system to be tested at all times; represents the number of sensors in the CPS system to be detected; express Moment The CPS system state estimation data of each sensor belongs to the weight of the Gaussian distribution cluster that has not been attacked; express The estimated error covariance matrix of the CPS system to be detected at any time; Indicates The observation matrix of sensors, Represents the transpose of a matrix; Indicates The inverse matrix of the measurement noise covariance matrix of the sensors; express Moment The measurement data of each sensor; Represents the process matrix of the CPS system to be tested.
[0019] Preferably, The calculation formula of the distribution probability density function of the CPS system state estimation data of the sensor at the moment under the Gaussian distribution cluster under attack is:
[0020] ,
[0021] in, express Moment The distribution probability density function of the CPS system state estimation data of each sensor under the attacked Gaussian distribution cluster; express Moment CPS system state estimation data of sensors; Express The mean of the attacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; Express The covariance of the attacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; express Moment The dimension of the CPS system state estimation data of each sensor; Represents the transpose of a matrix.
[0022] Preferably, The calculation formula of the distribution probability density function of the CPS system state estimation data of the sensor at the moment in the unattacked Gaussian distribution cluster is:
[0023] ,
[0024] in, express Moment The distribution probability density function of the CPS system state estimation data of the sensors under the Gaussian distribution cluster without attack; express Moment CPS system state estimation data of sensors; Express The mean of the unattacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; express Moment The estimated error covariance matrix of the sensors; express Moment The dimension of the CPS system state estimation data of each sensor; Represents the transpose of a matrix.
[0025] Preferably, based on the CPS system to be detected The measurement data of each sensor at the moment is obtained The CPS system state estimation data of each sensor at the moment includes:
[0026] Construct a linear time-invariant state space model of the CPS system to be tested;
[0027] The Kalman filter is based on the linear time-invariant state space model and The measurement data of each sensor at the moment is obtained CPS system state estimation data of each sensor at each moment.
[0028] Preferably, the linear time-invariant state space model of the CPS system to be detected is expressed as:
[0029] ,
[0030] ,
[0031] in, express The state vector of the CPS system to be detected at any time; The process matrix representing the CPS system to be tested; express The state vector of the CPS system to be detected at any time; express The process noise at each moment; express Moment The measurement data of each sensor; Indicates The observation matrix of sensors; express Moment The noise of a sensor.
[0032] Preferably, The CPS system state estimation data of each sensor at the moment also includes: using kernel principal component analysis or variational autoencoder to The CPS system state estimation data of each sensor at each moment is reduced in dimension.
[0033] The present invention also provides a CPS system remote state estimation device, comprising:
[0034] Data estimation module is used to estimate the The measurement data of each sensor at the moment is obtained CPS system state estimation data of each sensor at each moment;
[0035] K-means clustering module, used to use K-means clustering method The CPS system state estimation data of all sensors at all times are clustered to obtain the attacked cluster and the unattacked cluster;
[0036] The GMM clustering module is used to use the cluster centers of the attacked clusters and the cluster centers of the unattacked clusters as the initial values of the cluster centers, and use the GMM clustering method to The CPS system state estimation data of all sensors at the moment are re-clustered to obtain the attacked Gaussian distribution clusters and their mixing coefficients, as well as the unattacked Gaussian distribution clusters and their mixing coefficients;
[0037] Weight calculation module, used to calculate The distribution probability density function of the CPS system state estimation data of each sensor at the moment under the attacked Gaussian distribution cluster and the first product of the mixing coefficient of the attacked Gaussian distribution cluster, and The distribution probability density function of the CPS system state estimation data of the sensor at the moment under the unattacked Gaussian distribution cluster and the second product of the mixing coefficient of the unattacked Gaussian distribution cluster; the proportion of the second product to the sum of the first product and the second product is taken as The CPS system state estimation data of the sensor at the moment belongs to the weight of the Gaussian distribution cluster that has not been attacked;
[0038] CPS system state estimation module is used to The CPS system state estimation data of each sensor at the moment belongs to the weight of the Gaussian distribution cluster that has not been attacked, The measurement data of each sensor at all times and State estimation data calculation of the CPS system to be detected at any time State estimation data of the CPS system to be detected at all times.
[0039] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned CPS system remote state estimation method are implemented.
[0040] The CPS system remote state estimation method provided by the present invention clusters the CPS system state estimation data of each sensor to obtain an attacked cluster and an unattacked cluster, and then uses the GMM clustering method to re-cluster the CPS system state estimation data of all sensors to obtain an attacked Gaussian distribution cluster and its mixing coefficient, an unattacked Gaussian distribution cluster and its mixing coefficient, and the distribution probability density of the CPS system state estimation data of each sensor under the attacked Gaussian distribution cluster and the distribution probability density under the unattacked Gaussian distribution cluster, so as to calculate whether the CPS system state estimation data of each sensor belongs to an unattacked cluster. The weight of the attacked Gaussian distribution cluster. Since the smaller the possibility of a sensor being maliciously attacked, the more accurate the sensor's measurement data is, and its corresponding CPS system state estimation data is also more accurate and more reliable. Therefore, after calculating the weight value to reflect the reliability of the CPS system state estimation data of each sensor, when fusing the data of all sensors to calculate the CPS system state estimation data, focus on the measurement data of sensors that have not been attacked, so that the fused CPS system state estimation data can more accurately reflect the current operating status of the system, thereby providing a reliable basis for adjusting the system CPS system operating parameters. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:
[0042] Figure 1It is a flow chart of the remote state estimation method for CPS system;
[0043] Figure 2 A schematic diagram of the covariance trace of the remote state estimation error of a CPS system;
[0044] Figure 3 Schematic diagram of the covariance trace of the remote state estimation error of the second CPS system;
[0045] Figure 4 Schematic diagram of the covariance trace of the remote state estimation error of the third CPS system;
[0046] Figure 5 Schematic diagram of the structure of the remote state estimation device of the CPS system. DETAILED DESCRIPTION
[0047] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention. Example 1
[0048] See also Figure 1 , Figure 1 The figure is a flow chart of a remote state estimation method for a CPS system provided in an embodiment of the present application, and the method specifically includes:
[0049] S10: Based on the CPS system to be detected The measurement data of each sensor at the moment is obtained CPS system state estimation data of each sensor at each moment;
[0050] Since each sensor in the CPS system measures the state of a linear time-invariant process, the specific implementation of step S10 is as follows:
[0051] S100: Construct a linear time-invariant state space model of the CPS system to be detected; specifically, the model is expressed as:
[0052] ,
[0053] ,
[0054] in, express The state vector of the CPS system to be detected at any time; The process matrix representing the CPS system to be tested; express The state vector of the CPS system to be detected at any time; express The process noise at each moment; express Moment The measurement data of each sensor; Indicates The observation matrix of sensors; express Moment The noise of each sensor;
[0055] S200: Using Kalman filter based on linear time-invariant state space model and The measurement data of each sensor at the moment is obtained CPS system state estimation data of each sensor at each moment;
[0056] Specifically, the Kalman filter algorithm is used to filter and update the data to obtain The CPS system state estimation data of each sensor at the moment, the filtering process is expressed as:
[0057] ,
[0058] ,
[0059] in, express Moment CPS system state estimation data of sensors; express Moment The a posteriori minimum mean square error estimation of the CPS system state estimation data of the sensors; express Moment The a posteriori estimation error covariance matrix of the sensors; express Moment The estimated error covariance matrix of the sensors; Indicates The measurement noise covariance matrix of the sensors.
[0060] S20: Using K-means clustering method The CPS system state estimation data of all sensors at all times are clustered to obtain the attacked cluster and the unattacked cluster;
[0061] S30: The cluster centers of the attacked clusters and the cluster centers of the unattacked clusters are used as the initial values of the cluster centers, and the GMM clustering method is used to The CPS system state estimation data of all sensors at the moment are re-clustered to obtain the attacked Gaussian distribution clusters and their mixing coefficients, as well as the unattacked Gaussian distribution clusters and their mixing coefficients;
[0062] S40: Calculation The distribution probability density function of the CPS system state estimation data of each sensor at the moment under the attacked Gaussian distribution cluster and the first product of the mixing coefficient of the attacked Gaussian distribution cluster, and The distribution probability density function of the CPS system state estimation data of the sensor at the moment under the unattacked Gaussian distribution cluster and the second product of the mixing coefficient of the unattacked Gaussian distribution cluster; the proportion of the second product to the sum of the first product and the second product is taken as The CPS system state estimation data of the sensor at the moment belongs to the weight of the Gaussian distribution cluster that has not been attacked;
[0063] S50: Based on The CPS system state estimation data of each sensor at the moment belongs to the weight of the Gaussian distribution cluster that has not been attacked, The measurement data of each sensor at all times and State estimation data calculation of the CPS system to be detected at any time State estimation data of the CPS system to be detected at all times.
[0064] The CPS system remote state estimation method provided in the present application clusters the CPS system state estimation data of each sensor to obtain an attacked cluster and an unattacked cluster, and then uses the GMM clustering method to re-cluster the CPS system state estimation data of all sensors to obtain an attacked Gaussian distribution cluster and its mixing coefficient, an unattacked Gaussian distribution cluster and its mixing coefficient, and the distribution probability density of the CPS system state estimation data of each sensor under the attacked Gaussian distribution cluster and the distribution probability density under the unattacked Gaussian distribution cluster, so as to calculate the weight of the CPS system state estimation data of each sensor belonging to the unattacked Gaussian distribution cluster. Since the smaller the possibility of a sensor being maliciously attacked, the more accurate and credible the CPS system state estimation data of the sensor is, the credibility of the CPS system state estimation data of each sensor is reflected by calculating the weight value, so that when the data of all sensors are fused to calculate the CPS system state estimation data, the measurement data of the unattacked sensors is focused on, so that the fused CPS system state estimation data can more accurately reflect the current operating status of the system, thereby providing a reliable basis for adjusting the system CPS system operating parameters.
[0065] Specifically, in some embodiments of the present application, in step S40 The calculation formula of the distribution probability density function of the CPS system state estimation data of the sensor at the moment under the Gaussian distribution cluster under attack is:
[0066] ,
[0067] in, express Moment The distribution probability density function of the CPS system state estimation data of each sensor under the attacked Gaussian distribution cluster; express Moment CPS system state estimation data of sensors; Express The mean of the attacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; Express The covariance of the attacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; express Moment The dimension of the CPS system state estimation data of each sensor; Indicates transpose.
[0068] The calculation formula of the distribution probability density function of the CPS system state estimation data of the sensor at the moment in the unattacked Gaussian distribution cluster is:
[0069] ,
[0070] in, express Moment The distribution probability density function of the CPS system state estimation data of the sensors under the Gaussian distribution cluster without attack; Express The mean of the unattacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; express Moment The estimated error covariance matrix of the sensors.
[0071] Further, The calculation formula for the weight of the CPS system state estimation data of the sensor at the moment belonging to the Gaussian distribution cluster that has not been attacked is:
[0072] ,
[0073] in, express Moment The CPS system state estimation data of each sensor belongs to the weight of the Gaussian distribution cluster that has not been attacked; Express The mixing coefficient of the unattacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; express Moment The distribution probability density function of the CPS system state estimation data of the sensors under the Gaussian distribution cluster without attack; express Moment CPS system state estimation data of sensors; Express The mean of the unattacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; express Moment The estimated error covariance matrix of the sensors; Express The mixing coefficient of the attacked Gaussian distribution cluster is obtained by re-clustering the CPS system state estimation data of all sensors at the moment; express Moment The distribution probability density function of the CPS system state estimation data of each sensor under the attacked Gaussian distribution cluster; Express The mean of the attacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at the moment; Express The covariance of the attacked Gaussian distribution cluster is obtained by re-clustering the CPS system state estimation data of all sensors at the moment.
[0074] Based on the above The CPS system state estimation data of the sensor at the moment belongs to the weight of the Gaussian distribution cluster that has not been attacked. In step S50 The state estimation data of the CPS system to be detected at any time is expressed as:
[0075] ,
[0076] in, express State estimation data of the CPS system to be tested at all times; express State estimation data of the CPS system to be tested at all times; represents the number of sensors in the CPS system to be detected; express Moment The CPS system state estimation data of each sensor belongs to the weight of the Gaussian distribution cluster that has not been attacked; express The estimated error covariance matrix of the CPS system to be detected at any time; Indicates The observation matrix of sensors, Represents the transpose of a matrix; Indicates The inverse matrix of the measurement noise covariance matrix of the sensors; express Moment The measurement data of each sensor; Represents the process matrix of the CPS system to be tested;
[0077] in, The estimated error covariance matrix of the CPS system to be detected at any time The calculation formula is:
[0078] .
[0079] Optionally, in some embodiments of the present application, the The CPS system state estimation data of each sensor at the moment may be high-dimensional data. Therefore, after step S10, the method further includes: using a kernel principal component analysis method or a variational autoencoder to The CPS system state estimation data of each sensor at each moment is reduced in dimension;
[0080] Specifically, kernel principal component analysis (KPCA) is used to analyze The steps of reducing the dimension of the CPS system state estimation data of the sensor at the moment include:
[0081] 1. Calculate the values of each element of the kernel matrix required for KPCA:
[0082] ,
[0083] in, Represents and matrix Line The element value of the column, and are all hyperparameters, express Moment The CPS system state estimation data of the sensor elements, express Moment The CPS system state estimation data of the sensor elements;
[0084] 2. Calculate the average kernel matrix based on the values of each element of the kernel matrix:
[0085] ,
[0086] in, represents the average kernel matrix, Represents an n-row and n-column matrix with all elements being 1;
[0087] 3. Perform eigenvalue decomposition on the average kernel matrix to obtain multiple standard orthogonal eigenvectors and their corresponding eigenvalues:
[0088] ,
[0089] in, represents the orthonormal eigenvectors, , express The corresponding eigenvalues are ;
[0090] 4. Normalize each standard orthogonal eigenvector and its corresponding eigenvalue to obtain multiple eigenvectors, select the d eigenvectors with the largest cumulative variance contribution rate, obtain the reduced dimension feature matrix, and use the reduced dimension feature matrix kernel average kernel matrix to obtain the reduced dimension feature matrix kernel average kernel matrix. The CPS system state estimation data of the sensor at each moment is reduced in dimension; The normalization formula is:
[0091] ,
[0092] The reduced dimension feature matrix is expressed as:
[0093] ,
[0094] After dimensionality reduction The CPS system state estimation data of the sensor at the moment is expressed as:
[0095] .
[0096] Although the KPCA method is an effective dimensionality reduction method, it is necessary to select appropriate kernel functions and hyperparameters, which usually requires iterative experiments and tuning. In addition, when dealing with complex kernel functions, overfitting problems may occur. Therefore, the present invention also considers using variational auto-encoders (VAE) to reduce the dimensionality of the image. The CPS system state estimation data of the sensor at each moment is reduced in dimension. VAE is an end-to-end model that can jointly learn the encoder and decoder, so as to capture the distribution characteristics of the data more comprehensively. Unlike KPCA, VAE does not require additional steps to select the kernel function or calculate the kernel matrix; and, unlike the traditional autoencoder, VAE not only maps the input data to a single point in the latent space, but also maps it to the probability distribution within the latent space. Specifically, the encoder network of VAE is responsible for mapping the input data to the latent space, and the output of the encoder is composed of the latent variable distribution ( ) parameters, usually including the mean ( ) and standard deviation ( ), in order to enable optimization methods such as gradient descent to handle the probabilistic nature of the latent space, the present embodiment introduces a reparameterization technique to sample fixed noise from a standard normal distribution ( ), and then reparameterize the latent variables as , where the error function of VAE consists of the standard reconstruction error and KL divergence:
[0097] ,
[0098] ,
[0099] ,
[0100] in, Represents the input data, that is CPS system state estimation data from sensors at all times; Indicates division by The prior distribution of Indicates a given produce possibility; represents the approximate posterior distribution; Represents latent variables In the The mean of the dimension; Represents latent variables In the The standard deviation of the dimension.
[0101] The embodiment of the present application also applies the above-mentioned CPS system remote state estimation method to an optimal invisible information attack based on historical data on the system's sensors to verify the effectiveness of the method provided by the present application. Specifically, the attack model of the attack is:
[0102] ,
[0103] in, Represents local innovation and conforms to the steady-state Gaussian distribution , , attack matrix and all , ; is a zero-mean Gaussian distribution with a covariance of ;
[0104] Based on the above attack model, the optimal attack strategy is: attack from The moment begins, when When , the optimal attack strategy is , ;when When , the optimal attack strategy is the solution of the following convex optimization problem:
[0105] ,
[0106] St ,
[0107] ,
[0108] ,
[0109] ,
[0110] .
[0111] See also Figure 2 , Figure 2 A schematic diagram of the trace of the CPS system remote estimation error covariance obtained by using different CPS system remote estimation methods under the above-mentioned malicious attack provided in an embodiment of the present application, wherein GMM represents the CPS system remote state estimation method based on GMM clustering provided in the present application, BKA represents remote state estimation using only the Kalman filter, GEN represents remote state estimation with full knowledge of the sensors under malicious attack and ignoring the observation results of these sensors, and HEA represents remote state estimation without attack;
[0112] As can be seen from the figure, if only the Kalman filter is used for remote state estimation without considering whether the sensors in the system are attacked, the estimated result will deviate greatly from the actual state of the system and cannot reflect the actual state of the system; while the method provided in the present application can directly reflect the actual state of the system more accurately based on the observation results of all sensors in the current system even without knowing the attacked sensors.
[0113] like Figure 3As shown, in addition to the above four remote state estimation methods, the present application provides a trace schematic diagram of the remote state estimation error covariance after dimensionality reduction using KPCA based on the method provided in the present application. It can be seen from the figure that using the KPCA method to reduce the dimension of the system remote state estimation data of the sensor can further improve the accuracy of the final remote state estimation data and more accurately reflect the state of the CPS system;
[0114] like Figure 4 As shown, the present application also provides a trace diagram of the remote state estimation error covariance after dimensionality reduction using VAE on the basis of the method provided in the present application. It can be seen that using the VAE method to reduce the dimensionality of the sensor's system remote state estimation data can also improve the accuracy of the final remote state estimation data, and the final CPS system remote state data is also closer to the actual state of the system.
[0115] Based on the CPS system remote state estimation method provided in the above embodiment, the present application also provides a CPS system remote state estimation device, such as Figure 5 As shown, the device specifically includes:
[0116] The data estimation module 10 is used to estimate the The measurement data of each sensor at the moment is obtained CPS system state estimation data of each sensor at each moment;
[0117] K-means clustering module 20, used to use K-means clustering method to The CPS system state estimation data of all sensors at all times are clustered to obtain the attacked cluster and the unattacked cluster;
[0118] The GMM clustering module 30 is used to use the cluster center of the attacked cluster and the cluster center of the unattacked cluster as the initial value of the cluster center, and use the GMM clustering method to The CPS system state estimation data of all sensors at the moment are re-clustered to obtain the attacked Gaussian distribution clusters and their mixing coefficients, as well as the unattacked Gaussian distribution clusters and their mixing coefficients;
[0119] The weight calculation module 40 is used to calculate The distribution probability density function of the CPS system state estimation data of each sensor at the moment under the attacked Gaussian distribution cluster and the first product of the mixing coefficient of the attacked Gaussian distribution cluster, and The distribution probability density function of the CPS system state estimation data of the sensor at the moment under the unattacked Gaussian distribution cluster and the second product of the mixing coefficient of the unattacked Gaussian distribution cluster; the proportion of the second product to the sum of the first product and the second product is taken as The CPS system state estimation data of the sensor at the moment belongs to the weight of the Gaussian distribution cluster that has not been attacked;
[0120] The CPS system state estimation module 50 is used to estimate the state of the The CPS system state estimation data of each sensor at the moment belongs to the weight of the Gaussian distribution cluster that has not been attacked, The measurement data of each sensor at all times and State estimation data calculation of the CPS system to be detected at any time State estimation data of the CPS system to be detected at all times.
[0121] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned CPS system remote state estimation method are implemented. Example 2
[0122] Based on the CPS system remote state estimation method provided in the above-mentioned embodiment 1, this embodiment also provides a specific example of applying it to the power Internet of Things CPS system;
[0123] Step 1: Get Always check the measurement data of each sensor in the power Internet of Things CPS system and obtain the power Internet of Things state estimation data of each sensor;
[0124] Specifically, the sensors in the power Internet of Things include current sensors, voltage sensors, wire temperature sensors, etc., and their measurement data include the current, voltage, transmission line wire temperature, etc. of each node in the power Internet of Things;
[0125] Step 2: Use K-means clustering method to The power IoT state estimation data of all sensors at all times are clustered to obtain attacked clusters and non-attacked clusters;
[0126] Step 3: Take the cluster centers of the attacked cluster and the unattacked cluster as the initial values of the cluster centers, and use the GMM clustering method to The power IoT state estimation data of all sensors at all times are re-clustered to obtain the attacked Gaussian distribution clusters and their mixing coefficients, as well as the unattacked Gaussian distribution clusters and their mixing coefficients;
[0127] Step 4: Calculation The first product of the distribution probability density function of the power IoT state estimation data of each sensor at the moment under the attacked Gaussian distribution cluster and the mixed coefficient of the attacked Gaussian distribution cluster, and The distribution probability density function of the power Internet of Things state estimation data of the sensor at the moment in the unattacked Gaussian distribution cluster and the second product of the mixing coefficient of the unattacked Gaussian distribution cluster; the proportion of the second product to the sum of the first product and the second product is taken as The power IoT state estimation data of the sensor at the moment belongs to the weight of the Gaussian distribution cluster that has not been attacked;
[0128] Step 5: Based on The power IoT state estimation data of each sensor at the moment belongs to the weight of the Gaussian distribution cluster that has not been attacked, The measurement data of each sensor at all times and State estimation data calculation of power IoT CPS system to be detected at any time State estimation data of the power Internet of Things CPS system to be tested at all times;
[0129] Finally, the status estimation data of the power Internet of Things CPS system to be tested reflects the operating status, energy efficiency, safety warning, fault diagnosis, environmental monitoring and other information of the power Internet of Things, thereby providing data support for the maintenance and management of various equipment in the power Internet of Things and the operating parameters and scheduling of power equipment.
[0130] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0131] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0132] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0133] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0134] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.
Claims
1. A remote state estimation method for a CPS system, characterized in that: include: Based on the measurement data of each sensor at time k in the CPS system to be detected, the CPS system state estimation data of each sensor at time k is obtained; The K-means clustering method is used to cluster the CPS system state estimation data of all sensors at time k to obtain the attacked cluster and the unattacked cluster. The cluster centers of the attacked cluster and the unattacked cluster are used as the initial values of the cluster centers. The GMM clustering method is used to re-cluster the CPS system state estimation data of all sensors at time k to obtain the attacked Gaussian distribution cluster and its mixing coefficient, as well as the unattacked Gaussian distribution cluster and its mixing coefficient. Calculate the first product of the distribution probability density function of the CPS system state estimation data of each sensor at time k under the attacked Gaussian distribution cluster and the mixed coefficient of the attacked Gaussian distribution cluster, and the second product of the distribution probability density function of the CPS system state estimation data of the sensor at time k under the unattacked Gaussian distribution cluster and the mixed coefficient of the unattacked Gaussian distribution cluster; take the proportion of the second product to the sum of the first product and the second product as the weight of the CPS system state estimation data of the sensor at time k belonging to the unattacked Gaussian distribution cluster; wherein, the calculation formula of the distribution probability density function of the CPS system state estimation data of the sensor at time k under the attacked Gaussian distribution cluster is: in, represents the distribution probability density function of the CPS system state estimation data of the i-th sensor at time k under the attacked Gaussian distribution cluster; represents the CPS system state estimation data of the i-th sensor at time k; represents the mean of the attacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at time k; represents the covariance of the attacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at time k; n represents the dimension of the CPS system state estimation data of the i-th sensor at time k; T represents the transpose of the matrix; The calculation formula of the distribution probability density function of the CPS system state estimation data of the sensor at time k in the Gaussian distribution cluster that is not attacked is: in, represents the distribution probability density function of the CPS system state estimation data of the i-th sensor at time k under the Gaussian distribution cluster that is not attacked; represents the CPS system state estimation data of the i-th sensor at time k; represents the mean of the unattacked Gaussian distribution clusters obtained by re-clustering the CPS system state estimation data of all sensors at time k; P k,i represents the estimated error covariance matrix of the i-th sensor at time k; n represents the dimension of the CPS system state estimation data of the i-th sensor at time k; T represents the transpose of the matrix; The state estimation data of the CPS system to be detected at time k is calculated based on the weight of the CPS system state estimation data of each sensor at time k belonging to the Gaussian distribution cluster that has not been attacked, the measurement data of each sensor at time k, and the state estimation data of the CPS system to be detected at time k-1; wherein the state estimation data of the CPS system to be detected at time k is expressed as: in, represents the state estimation data of the CPS system to be detected at time k; represents the state estimation data of the CPS system to be detected at time k-1; N represents the number of sensors in the CPS system to be detected; represents the weight of the CPS system state estimation data of the i-th sensor at time k belonging to the Gaussian distribution cluster that has not been attacked; P k represents the estimated error covariance matrix of the CPS system to be detected at time k; C i represents the observation matrix of the i-th sensor, and T represents the transpose of the matrix; represents the inverse matrix of the measurement noise covariance matrix of the i-th sensor; y k,i represents the measurement data of the i-th sensor at time k; A represents the process matrix of the CPS system to be detected.
2. The CPS system remote state estimation method according to claim 1, characterized in that: The calculation formula for the weight of the CPS system state estimation data of the sensor at time k belonging to the Gaussian distribution cluster that has not been attacked is: in, The weight indicating that the CPS system state estimation data of the i-th sensor at time k belongs to the Gaussian distribution cluster that has not been attacked; represents the mixing coefficient of the unattacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at time k; represents the distribution probability density function of the CPS system state estimation data of the i-th sensor at time k under the Gaussian distribution cluster that is not attacked; represents the CPS system state estimation data of the i-th sensor at time k; represents the mean of the unattacked Gaussian distribution clusters obtained by re-clustering the CPS system state estimation data of all sensors at time k; P k,i represents the estimated error covariance matrix of the i-th sensor at time k; represents the mixing coefficient of the attacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at time k; represents the distribution probability density function of the CPS system state estimation data of the i-th sensor at time k under the attacked Gaussian distribution cluster; represents the mean of the attacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at time k; It represents the covariance of the attacked Gaussian distribution cluster obtained by re-clustering the CPS system state estimation data of all sensors at time k.
3. The CPS system remote state estimation method according to claim 1, characterized in that: Based on the measurement data of each sensor at time k in the CPS system to be detected, the CPS system state estimation data of each sensor at time k is obtained, including: Construct a linear time-invariant state space model of the CPS system to be tested; The Kalman filter is used to obtain the CPS system state estimation data of each sensor at the k moment based on the linear time-invariant state space model and the measurement data of each sensor at the k moment.
4. The CPS system remote state estimation method according to claim 3, characterized in that: The linear time-invariant state space model of the CPS system to be detected is expressed as: X k+1 =AX k +ω k , y k,i =C i X k +v k,i , Among them, X k+1 represents the state vector of the CPS system to be detected at time k+1; A represents the process matrix of the CPS system to be detected; X k represents the state vector of the CPS system to be detected at time k; ω k represents the process noise at time k; y k,i represents the measurement data of the i-th sensor at time k; C i represents the observation matrix of the i-th sensor; v k,i represents the noise of the i-th sensor at time k.
5. The CPS system remote state estimation method according to claim 1, characterized in that: After obtaining the CPS system state estimation data of each sensor at time k, the method further includes: reducing the dimension of the CPS system state estimation data of each sensor at time k by using a kernel principal component analysis method or a variational autoencoder.
6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the CPS system remote state estimation method according to any one of claims 1 to 5 are implemented.
