A control system performance maintenance method, device, equipment and storage medium

Abstract

The present invention discloses a performance maintenance method, device, equipment and computer-readable storage medium for a control system, and relates to the field of automatic control technology. The method comprises: obtaining re-identification input data of a process model of the control system; wherein the re-identification input data comprises data before model mismatch and data at the time of model mismatch; constructing algorithm data of a target type according to the re-identification input data; substituting a matrix in the algorithm data into an objective function based on similarity extraction for solving, and extracting a common matrix between the data before model mismatch and the data at the time of model mismatch; calculating model parameters of the process model using a subspace identification algorithm according to the acquired data after model mismatch and the common matrix; the present invention extracts similarity information from the original past data, and transfers it to the calculation of the re-identification process, thereby reducing the data demand for re-identification and shortening the acquisition time of the model parameters.

Description

Technical Field

[0001] The present invention relates to the field of automation control technology, and in particular to a performance maintenance method, device, equipment and computer-readable storage medium for a control system. Background Art

[0002] Process model mismatch is an important issue when it comes to process control and system modeling. Process models are mathematical representations used to describe and predict the behavior of real systems. Mismatch refers to significant differences or inconsistencies between the process model and the actual system. Process model mismatch can lead to the following problems: 1. Degraded control performance; if the process model does not match the actual system, the controller may not be able to accurately predict and respond to the behavior of the system; this may lead to degraded control performance and fail to meet the required control requirements, such as stability, robustness, and fast response. 2. Unpredictable behavior; the mismatched model may not capture all the dynamic characteristics and nonlinear behavior of the actual system; this may cause unexpected behavior of the system and fail to predict and explain the response of the system. 3. Algorithm failure; some control algorithms require accurate process models for design and analysis; if the model is mismatched, these algorithms may not work properly or produce unreliable results. Therefore, in order to maintain efficient process operation, it is necessary to re-identify the mismatched process model. However, system identification experiments are complex and time-consuming, and conducting these experiments can result in considerable economic losses. How to effectively use data and use less data to achieve acceptable modeling performance is a problem that needs to be explored.

[0003] In the prior art, model training with a small amount of data is also called few-sample learning. An intuitive solution strategy is to combine some prior knowledge in modeling training. However, the required prior knowledge is difficult to obtain in practical applications, making it difficult to use existing system identification methods. In addition, existing system identification methods do not directly target scenarios where re-identification is performed after model mismatch. Even if the various prior knowledge mentioned in the method can be obtained, this prior knowledge is not directly related to the re-identification of the mismatched model, and re-identification after model mismatch cannot be achieved. Therefore, how to achieve re-identification after process model mismatch without using prior knowledge is a problem that urgently needs to be solved today. Summary of the invention

[0004] The object of the present invention is to provide a control system performance maintenance method, device, equipment and computer-readable storage medium to achieve re-identification after process model mismatch without using prior knowledge.

[0005] In order to solve the above technical problems, the present invention provides a performance maintenance method of a control system, comprising:

[0006] Acquire re-identification input data of a process model of the control system; wherein the re-identification input data includes data before model mismatch and data at the time of model mismatch;

[0007] According to the re-identification input data, construct algorithm data of target type; wherein the target form is a matrix and vector form in a subspace identification algorithm, and the algorithm data includes matrices and vectors corresponding to the data before the model mismatch and the data when the model mismatch occurs;

[0008] Substituting the algorithm data into the objective function based on similarity extraction for solving, and extracting a common matrix between the data before the model mismatch and the data when the model mismatch occurs;

[0009] The model parameters of the process model are calculated using a subspace identification algorithm according to the acquired model mismatch data and the common matrix, so as to predict the system predicted output data of the control system using the process model corresponding to the model parameters and the input data of the control system.

[0010] In some embodiments, constructing algorithm data of a target type according to the re-identification input data includes:

[0011] Using the input data and output data of the data before the model mismatch and the data at the time of the model mismatch, respectively, constructing the future Hankel matrix and the past Hankel matrix of the data before the model mismatch and the data at the time of the model mismatch, respectively; wherein the future Hankel matrix includes a future output Hankel matrix and a future input Hankel matrix, and the past Hankel matrix includes a past output Hankel matrix and a past input Hankel matrix;

[0012] According to the future Hankel matrix and the past Hankel matrix, the algorithm data is obtained; wherein the algorithm data includes the matrix Z corresponding to the data before the model mismatch match and vector y match And the matrix Z corresponding to the data when the model is mismatched mismatch and vector y mismatch ;y match =vec(Y f,match ), y mismatch =vec(Y f,mismatch ), Y f,match and Y f,mismatch are the future output Hankel matrices corresponding to the data before the model mismatch and the data when the model mismatch respectively; W p,match and W p,mismatch are the past Hankel matrices corresponding to the data before the model mismatch and the data when the model mismatch occurs, respectively, U f,match and Uf,mismatch are the future input Hankel matrices corresponding to the data before the model mismatch and the data when the model mismatch respectively; I is the unit matrix, the size of I is Nm×Nm, N is the length of the preset dynamic window for constructing the Hankel matrix, and m is the number of output variables; N 0,1 A matrix constructed with preset 0s and 1s.

[0013] In some embodiments, the step of calculating the model parameters of the process model using a subspace identification algorithm based on the acquired model mismatch data and the common matrix includes:

[0014] Get the matrix Z corresponding to the data after the model mismatch exp and vector y exp ; Among them, y exp =vec(Y f,exp ), Y f,exp , W p,exp and U f,exp are the future output Hankel matrix, past Hankel matrix and future input Hankel matrix corresponding to the model mismatch data, respectively. 0,1 A matrix constructed for the preset 0s and 1s;

[0015] pass Calculate the estimated parameter α exp ; Wherein, Φ is the common matrix; α exp The matrix size is r·NmJ×1, where N is the length of the preset dynamic window for constructing the Hankel matrix, m is the number of output variables, J is the number of dynamic windows for constructing the Hankel matrix, and r≤(N 2 m(m+l)+Nml), r is the number of preset individual vectors, l is the number of input variables;

[0016] Through θ exp =Φα exp , calculate the identification parameter θ exp θ exp The matrix size is (N 2 m(m+l)+Nml)·NmJ;

[0017] According to the identification parameter θ exp The parameter l in u , construct the target Hankel matrix H; where, to They are the parameter descriptions of the state space model respectively;

[0018] Perform singular value decomposition on the target Hankel matrix H to obtain a singular value decomposition result; wherein H = USV T, the singular value decomposition result includes a first matrix U, a second matrix S and a third matrix V;

[0019] The model parameters are generated using the observation matrix Γ and the extended controllable matrix Δ; where Γ=U(:,1:n)S 1 / 2 , Δ=S 1 / 2 V(:,1:n) T , n is a preset model order; the model parameters include a first model parameter A, a second model parameter B, a third model parameter C and a fourth model parameter D; B=Δ(:,1:m),C=Γ(1:l,:),

[0020] In some embodiments, before constructing the algorithm data of the target type according to the re-identification input data, the method further includes:

[0021] Preprocessing the re-identification input data to obtain the preprocessed re-identification input data;

[0022] Correspondingly, constructing algorithm data of a target type according to the re-identified input data includes:

[0023] The algorithm data is constructed according to the pre-processed re-identification input data.

[0024] In some embodiments, obtaining re-identified input data of a process model of the control system includes:

[0025] The pre-model mismatch data is generated using an excitation signal according to the process model.

[0026] In some embodiments, the objective function is and are the future output Hankel matrix, past Hankel matrix and future input Hankel matrix corresponding to the data match before model mismatch and the data mismatch when model mismatch respectively, N 0,1 A matrix constructed with preset 0 and 1, α g are the individual parameter vectors corresponding to the data match before model mismatch and the data mismatch when the model mismatch occurs, Φ is the common parameter matrix corresponding to the data match before model mismatch and the data mismatch when the model mismatch occurs, and I is the unit matrix.

[0027] In some embodiments, the step of substituting the algorithm data into an objective function based on similarity extraction for solving, and extracting a common matrix between the data before the model mismatch and the data at the time of the model mismatch, includes:

[0028] According to the matrix in the algorithm data, a data matrix is ​​constructed; wherein the data matrix includes a first data matrix Second data matrix and the third data matrix

[0029] pass Using self-consistent field iteration, the i-th individual vector is calculated The corresponding conversion parameter γ i ; Where i = 1, ..., r, r is the number of preset individual vectors,

[0030] pass Get the i-th individual vector

[0031] If we get r individual vectors Then use r individual vectors Generate a common matrix Φ; where,

[0032] If the individual vectors obtained If the number is less than r, the currently acquired i-th individual vector is used pass Update Z match and Z mismatch ;

[0033] Using the updated Z match and Z mismatch , update the data matrix, and execute the Using self-consistent field iteration, the i-th individual vector is calculated The corresponding conversion parameter γ i steps.

[0034] The present invention also provides a performance maintenance device for a control system, comprising:

[0035] An acquisition module, used for acquiring re-identification input data of a process model of a control system; wherein the re-identification input data includes data before model mismatch and data at the time of model mismatch;

[0036] A construction module, configured to construct algorithm data of a target type according to the re-identification input data; wherein the target form is a matrix and vector form in a subspace identification algorithm, and the algorithm data includes matrices and vectors corresponding to the data before the model mismatch and the data when the model mismatch occurs;

[0037] An extraction module, used for substituting the algorithm data into the target function based on similarity extraction for solving, and extracting a common matrix between the data before the model mismatch and the data when the model mismatch occurs;

[0038] The identification module is used to calculate the model parameters of the process model using a subspace identification algorithm based on the acquired model mismatch data and the common matrix, so as to use the process model corresponding to the model parameters to predict the system predicted output data of the control system using the input data of the control system.

[0039] The present invention also provides a performance maintenance device for a control system, comprising:

[0040] Memory for storing computer programs;

[0041] A processor is used to implement the steps of the performance maintenance method of the control system as described above when executing the computer program.

[0042] 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 performance maintenance method of the control system as described above are implemented.

[0043] A performance maintenance method for a control system provided by the present invention comprises: obtaining re-identification input data of a process model of a control system; wherein the re-identification input data comprises data before model mismatch and data at the time of model mismatch; constructing algorithm data of a target type according to the re-identification input data; wherein the target form is a matrix and vector type in a subspace identification algorithm, and the algorithm data comprises matrices and vectors corresponding to the data before model mismatch and the data at the time of model mismatch respectively; substituting the algorithm data into an objective function based on similarity extraction for solving, and extracting a common matrix between the data before model mismatch and the data at the time of model mismatch; according to the acquired data after model mismatch and the common matrix, using the subspace identification algorithm to calculate the model parameters of the process model, so as to use the process model corresponding to the model parameters to predict the system predicted output data of the control system using the input data of the control system;

[0044] It can be seen that the present invention substitutes the algorithm data into the objective function based on similarity extraction to solve it, and extracts the common matrix between the data before the model mismatch and the data at the time of model mismatch. It can extract similarity information from the original past data and transfer it to the calculation of the re-identification process without the need to know the prior knowledge of the process model in advance, which is easy to implement; by using the subspace identification algorithm to calculate the model parameters of the process model based on the acquired data after the model mismatch and the common matrix, it can use the similarity information extracted from the past data to reduce the data demand for re-identification, thereby shortening the time required to obtain the model parameters, and has a high application value. In addition, the present invention also provides a performance maintenance device, equipment and computer-readable storage medium for a control system, which also have the above-mentioned beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0046] Figure 1 A flow chart of a performance maintenance method of a control system provided by an embodiment of the present invention;

[0047] Figure 2 A flowchart of a common matrix extraction process of another control system performance maintenance method provided by an embodiment of the present invention;

[0048] Figure 3 A structural block diagram of a performance maintenance device for a control system provided by an embodiment of the present invention;

[0049] Figure 4 A simplified structural diagram of a performance maintenance device for a control system provided by an embodiment of the present invention;

[0050] Figure 5 A schematic diagram of the specific structure of a performance maintenance device for a control system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0052] Please refer to Figure 1 , Figure 1 This is a flow chart of a performance maintenance method of a control system provided by an embodiment of the present invention. The method may include:

[0053] Step 101: Obtain re-identification input data of a process model of a control system; wherein the re-identification input data includes data before model mismatch and data during model mismatch.

[0054] Among them, the process model in this embodiment can be a model for predicting the output of the control system, that is, the process model can simulate the control process of the control system, and use the input data of the control system to predict the output of the control system (that is, the system predicted output data). For example, the process model can be a process model of a temperature control system or a motor control system; for example, the process model of a temperature control system can use the input data of the temperature control system (such as the current temperature and the fan speed, etc.) to predict the fan speed control value of the temperature control system (that is, the system predicted output data). The process model in this step can be a model that needs to be re-identified after mismatch. The re-identification input data in this embodiment can be the input data required for re-identification of the mismatched process model; the re-identification input data can include data before the model mismatch and data at the time of model mismatch, that is, past data of the process model collected before re-identification; data before the model mismatch can be data of the process model before the mismatch occurs, and data at the time of model mismatch can be data of the process model after the mismatch occurs.

[0055] Correspondingly, the specific content of the re-identification input data in this embodiment can be set by the designer according to the usage scenario and user needs, such as the pre-model mismatch data can include the input data and output data of the process model collected before the process model mismatch occurs (i.e., the system predicted output data). The pre-model mismatch data can also include the input data and output data with a continuous excitation signal generated by the previous identification experiment on the process model. That is to say, in this embodiment, the input and output data generated when the model identification of the process model was originally performed can be used as a reference; however, sometimes the data used when the model identification was originally performed may not be obtained; in this case, according to the existing accurate model (i.e., the process model), the simulation data (i.e., the pre-model mismatch data) can be generated by giving an excitation signal; the excitation signal can be a generalized binary noise signal (GBNS) or a pseudo-binary random signal (PRBS), etc.; that is, in this step, the processor can generate the pre-model mismatch data according to the process model using the excitation signal.

[0056] Accordingly, the mismatch of the process model is generally detected through an algorithm when the system is running; however, in actual industry, it is unlikely to immediately deal with it in a targeted manner as soon as a mismatch occurs, such as replacing equipment or re-conducting system identification experiments. In addition, most control designs themselves contain a certain degree of robustness; therefore, the process model will continue to operate within the tolerance range after a mismatch occurs. In this embodiment, the data when the model mismatch occurs may include the input data and output data of the process model operating within the tolerance range collected after the process model mismatch occurs. The data when the model mismatch occurs may also include the input data and output data generated by a re-identification experiment on the process model; although the amount of data when the model mismatch occurs obtained from the re-identification experiment may be limited, such similarity information (such as a common matrix) can greatly assist the identification calculation.

[0057] Among them, in order to successfully carry out model identification, a large amount of data is required to ensure that these data can fully reflect the characteristics of the system. Usually, in order to obtain these data, it is necessary to stop the production and conduct an open-loop identification experiment (such as the previous identification experiment and re-identification experiment on the process model mentioned above); however, this means that during the experiment, the production process will be interrupted, resulting in the inability to continue production or output. The longer the experiment takes, the greater the loss of economic benefits. Therefore, if the data generated during the above re-identification experiment is too scarce or it is inconvenient to conduct a re-identification experiment, the data running within the tolerance range can be directly used as the data when the model mismatch occurs.

[0058] Furthermore, before step 102, after obtaining the re-identification input data of the process model, the obtained re-identification input data can also be preprocessed to obtain the preprocessed re-identification input data to facilitate subsequent use; for example, since data preprocessing operations such as noise reduction and outlier removal have been performed in the previous identification experiment; and since the simulation data generated by the old model (i.e., the process model) is basically clean, most of the data preprocessing process can be omitted for the non-mismatched data (i.e., the data before the model mismatch), and only normalization processing is required before the next calculation, that is, the processor can normalize the data before the model mismatch to obtain the preprocessed data before the model mismatch. Accordingly, if the data when the model mismatch occurs includes input data and output data generated by a re-identification experiment on the process model, the processor can perform normalization processing on the data when the model mismatch occurs to obtain the preprocessed data when the model mismatch occurs; if the data when the model mismatch occurs includes input data and output data of the process model operating within a tolerable range that are collected after a mismatch occurs in the process model, the processor can perform preprocessing on the data when the model mismatch occurs that may include noise reduction, outlier removal and normalization processing to obtain the preprocessed data when the model mismatch occurs.

[0059] Step 102: constructing algorithm data of a target type according to the re-identification input data; wherein the target form is the matrix and vector form in the subspace identification algorithm, and the algorithm data includes the matrices and vectors corresponding to the data before the model mismatch and the data during the model mismatch.

[0060] It can be understood that the target form in this step can be the matrix and vector form in the subspace identification algorithm, so as to construct the corresponding algorithm data in the matrix and vector form in the subspace identification algorithm based on the input and output data of the data before the model mismatch and the data at the time of model mismatch, that is, the matrix and vector corresponding to the data before the model mismatch and the data at the time of model mismatch.

[0061] Correspondingly, the specific method for the processor in this step to construct the target type of algorithm data according to the re-identified input data can be set by the designer. For example, in order to construct the relationship between variables, the following linear time-invariant state space model can be used:

[0062]

[0063] In the above formula, is the state vector, is the input vector, is the output vector; Represents a matrix of n×1 dimensions; and Represent process noise and measurement noise respectively, both are zero-mean white noise sequences, is the model parameter; n is the preset model order; m is the number of output variables, that is, the number of preset output variables; l is the number of input variables, that is, the number of preset input variables.

[0064] Correspondingly, according to the above state space model, the following relationship can be obtained:

[0065]

[0066] In the above formula, is the future output Hankel matrix without noise terms, W p with U f are the past Hankel matrix and the future input Hankel matrix respectively; I is the identity matrix; N is a matrix constructed by 0 and 1, that is, N 0,1 The input and output of the data before model mismatch and the data at model mismatch are respectively established according to the definition of the above relationship, and the future and past Hankel matrices are then expressed as y g =Z g θ g; Among them, g represents the index of the data class, g∈{match,mismatch}, which can represent the data match before the model mismatch and the data mismatch when the model mismatches.

[0067] For example, the outputs of time periods N+1 to 2N+J-1 are arranged into a future output Hankel matrix (i.e., a future output Hankel matrix) as follows:

[0068]

[0069] in, Output the Hankel matrix for the future, y N+1 is the output variable at the N+1 sampling time; N and J can be the dynamic window describing the dynamic relationship of the sequence and the number of custom dynamic windows, that is, N is the length of the preset dynamic window for constructing the Hankel matrix, and J is the number of dynamic windows for constructing the Hankel matrix; usually N can be debugged and configured, and J can be set accordingly according to the amount of data. Similarly, in the past, the Hankel matrix With the future input Hankel matrix U f The same arrangement can be used to obtain; Y p and U p are the past output Hankel matrix and the past input Hankel matrix in the past Hankel matrix respectively. According to the above relationship, the non-mismatched (match) and mismatched (mismatch) data are arranged respectively to obtain the following algorithm data:

[0070] y match =vec(Y f,match );

[0071] y mismatch =vec(Y f,mismatch );

[0072] In the above formula, I is a unit matrix of size Nm×Nm.

[0073] That is to say, in this step, the processor can use the input data and output data of the data before the model mismatch and the data at the time of the model mismatch to construct the future Hankel matrix and the past Hankel matrix of the data before the model mismatch and the data at the time of the model mismatch; wherein the future Hankel matrix includes the future output Hankel matrix and the future input Hankel matrix, and the past Hankel matrix includes the past output Hankel matrix and the past input Hankel matrix; according to the future Hankel matrix and the past Hankel matrix, the algorithm data is obtained; wherein the algorithm data includes the matrix Z corresponding to the data before the model mismatch match and vector y match And the matrix Z corresponding to the data when the model mismatchesmismatch and vector y mismatch ;y match =vec(Y f,match ), y mismatch =vec(Y f,mismatch ), Y f,match and Y f,mismatch are the future output Hankel matrices corresponding to the data before model mismatch and the data during model mismatch respectively; W p,match and W p,mismatch are the past Hankel matrices corresponding to the data before model mismatch and the data during model mismatch, respectively, U f,match and U f,mismatch are the future input Hankel matrices corresponding to the data before model mismatch and the data at model mismatch, respectively; I is the unit matrix, the size of I is Nm×Nm, N is the length of the preset dynamic window for constructing the Hankel matrix, and m is the number of output variables; N 0,1 A matrix constructed with preset 0s and 1s.

[0074] Step 103: Substitute the algorithm data into the objective function based on similarity extraction to solve it, and extract the common matrix between the data before the model mismatch and the data when the model mismatch occurs.

[0075] The specific content of the objective function based on similarity extraction in this embodiment can be set by the designer. For example, the above parameter θ g Decompose into the common parameter matrix Φ and the matrix α of individual parameters g The multiplication of g =Φα g , so that the original relationship becomes y g =Z g Φα g ; g∈{match, mismatch}; In order to extract the common parameter matrix, the objective function can be defined as follows:

[0076]

[0077] In the above formula, ||·|| F is the Frobenius norm (abbreviated as F-norm, a matrix norm). The above objective function can be designed based on the concept of minimum sum of squared errors; the interpretation of this objective function is: find the common parameter matrix Φ that exists in both the mismatched and non-mismatched data sets, and the individual parameter vectors α of the mismatched and non-mismatched data sets. g , that is, α match and α mismatch .

[0078] That is to say, the objective function in this embodiment can be and are the future output Hankel matrix, past Hankel matrix and future input Hankel matrix corresponding to the data match before model mismatch and the data mismatch when model mismatch respectively, N 0,1 A matrix constructed with preset 0 and 1, α g are the individual parameter vectors corresponding to the data match before model mismatch and the data mismatch when model mismatch occurs, and Φ is the common parameter matrix corresponding to the data match before model mismatch and the data mismatch when model mismatch occurs. g =Φα g , l w,g Belongs to parameter θ g The part is the parameter corresponding to the past information, which can be expressed in MATLAB (a mathematical software) programming terms as l w,g =θ g (N 2 m(m+l)·NmJ,:);l u Belongs to parameter θ g The rest of the part is the parameter corresponding to the future output information, which can be expressed in MATLAB (a mathematical software) programming terms as l u =θ g (Nml·NmJ,:).

[0079] It should be noted that the specific method in which the processor in this embodiment substitutes the algorithm data into the objective function based on similarity extraction to solve and extract the common matrix between the data before the model mismatch and the data when the model mismatch occurs can be set by the designer, such as Figure 2 As shown, this step may include:

[0080] Step 1031: Construct a data matrix according to the matrix in the algorithm data.

[0081] Among them, due to the mismatched and unmismatched individual parameter vectors α g (α match and α mismatch ) are coupled together. In order to avoid the problem of non-convergence in parameter alternating optimization, the objective function can be further processed to g Substitute the least squares analytical solution of into the original objective function for sorting, introduce the structural constraint of orthogonality, and finally obtain another form of solving only the common parameter Φ, written as:

[0082]

[0083] stΦT Φ=I

[0084] In the above formula, tr(·) is the trace of the matrix. In order to reduce the complexity of its nonlinear optimization, the parameter Φ can be decomposed into individual vectors And r vectors are obtained by calculating them one by one. r is the number of preset individual vectors, which is used to determine the structure of the common parameter matrix. The larger its value is, the lower the similarity is, that is, the lower the degree of information transfer is. At the same time, in order to ensure the convergence of the solution, the objective function will be organized into a form that conforms to the use of self-consistent-field iteration (SCF iteration). Perform a projective transformation via the unmismatched matrix; Substituting into the optimization solution of the vector and introducing the Lagrange multiplier, the optimal expression for the i-th vector can be as follows:

[0085]

[0086] In the above formula,

[0087] Correspondingly, in this step, the processor can calculate the matrix (such as Z) in the algorithm data. match and Z mismatch ), construct a data matrix; wherein the data matrix includes a first data matrix Second data matrix and the third data matrix

[0088] Step 1032: Apply self-consistent field iterative calculation to the data matrix to obtain current conversion parameters.

[0089] The current conversion parameter in this step can be the i-th conversion parameter γ obtained by the current calculation. i , that is, the i-th individual vector The corresponding conversion parameter γ i That is to say, in this step, the processor can Using self-consistent field iteration, the i-th individual vector is calculated The corresponding conversion parameter γ i .

[0090] Correspondingly, in this step, the processor uses self-consistent field iteration calculation on the data matrix to obtain the specific method of the current conversion parameters, which can be set by the designer. For example, the self-consistent field iteration can be: Perform singular value decomposition (SVD decomposition) and use the eigenvector corresponding to the maximum eigenvalue as the parameter of the tth iteration Iterate until the parameters converge (such as reaching a preset number of iterations). That is, in this step, the processor can obtain the current (i.e., tth) iteration right Perform singular value decomposition and obtain the eigenvector corresponding to the maximum eigenvalue as the conversion parameter of the current iteration If the current iteration reaches the preset number of iterations, the conversion parameters of the current iteration are As the current conversion parameter γ i ; If the current iteration does not reach the preset number of iterations, the next iteration will be used as the current iteration, that is, Using conversion parameters Get the current iteration And execute Perform singular value decomposition and obtain the eigenvector corresponding to the maximum eigenvalue as the conversion parameter of the current iteration steps.

[0091] In terms of distance, in this step, for the first individual vector (i.e., i=1) is calculated, and a random number of size (N 2 m(m+l)+Nml)×1 initial individual vector Combine it with and Merge and Using the above matrix calculation, we can get And Perform SVD decomposition Collect the eigenvector corresponding to the maximum eigenvalue as the result of this iteration This process is repeated until the parameter γ1 converges.

[0092] Step 1033: Project the current transformation parameters to obtain the current individual vector.

[0093] Among them, in this step, the processor can Get the i-th individual vector For example, the first parameter vector γ1 can be projected back to the original parameter space Right now

[0094] Step 1034: Determine whether the current individual vector is the rth individual vector; if so, proceed to step 1035; if not, proceed to step 1036.

[0095] It is understandable that in this step, the processor can determine the current i-th individual vector Is it the rth individual vector? Determine whether r individual vectors have been obtained Thus, we can obtain r individual vectors Then, the process proceeds to step 1035, where a common matrix is ​​generated using the obtained r individual vectors; When r individual vectors are obtained, the process proceeds to step 1036, where the data matrix is ​​updated using the current individual vectors to continue obtaining new individual vectors until r individual vectors are calculated.

[0096] Step 1035: Generate a common matrix based on the obtained r individual vectors.

[0097] It is understandable that in this step, the processor can obtain r individual vectors When r individual vectors are used Generate the common matrix Φ. For example, by Generate the common matrix Φ.

[0098] Step 1036: Use the current individual vector to update the data matrix and enter step 1032.

[0099] It is understandable that in this step, the processor can obtain individual vectors If the number is less than r, the currently acquired i-th individual vector is used (i.e. the current individual vector), update the data matrix to continue to obtain the next individual vector.

[0100] Correspondingly, the specific way in which the processor uses the current individual vectors to update the data matrix in this step can be set by the designer. For example, the processor can use reduction techniques to respectively match With Z mismatch The i-th individual vector Related information is eliminated, such as the processor can use the currently obtained i-th individual vector pass Update Z match and Z mismatch ; Using the updated Z match and Z mismatch , update the data matrix (such as and ), and proceed to step 1032 to obtain the i+1th individual vector For example, when i=1, Z can be excluded by the following formula match With Z mismatch Already extracted from Information:

[0101]

[0102]

[0103] Step 104: Calculate the model parameters of the process model using the subspace identification algorithm based on the acquired model mismatch data and the common matrix, and use the process model corresponding to the model parameters to predict the input data of the control system to obtain the system predicted output data of the control system.

[0104] It can be understood that the post-model mismatch data in this embodiment can be data of the process model after the mismatch occurs. In this embodiment, the data after the process model mismatch occurs is collected, and it is combined with the extracted common matrix in the calculation of subspace identification to obtain the model parameters of the process model.

[0105] Correspondingly, in this embodiment, after calculating the model parameters of the process model using the subspace identification algorithm based on the acquired model mismatch data and the common matrix, the process model using the calculated model parameters can also be used to predict the system predicted output data of the control system based on the input data of the control system, so as to use the system predicted output data for the design and fault detection of the control system; for example, the method provided in this embodiment can also include a fault detection process of the control system, for example, the processor can compare the actual output data of the control system with the system predicted output data to detect the fault condition of the control system; wherein the fault condition is a fault state or a normal state.

[0106] Specifically, the specific content of the data after the model mismatch in this embodiment can be set by the designer. For example, the data after the model mismatch can be the same as the data when the model mismatch is in the re-identification input data, or it can be different from the data when the model mismatch is. For example, in order to identify a process model that can reflect more comprehensive characteristics of the process, the collection of data after the model mismatch can be completed through a system identification experiment (i.e., a re-identification experiment) that gives a sufficient excitation signal.

[0107] In this step, the processor can arrange the acquired model mismatch data to obtain the Hankel matrix Then, the common matrix Φ is combined with it, and the parameter estimation optimization problem of the mismatched process model can be written as:

[0108]

[0109] In the above formula, α expis the individual parameter describing the data after model mismatch. This problem can be directly solved by least squares, expressed as:

[0110]

[0111] The estimated parameter α exp It can be combined with parameter Φ to form parameter θ exp θ exp =Φα exp As the operation for subsequent subspace identification; θ exp It can include parameters corresponding to past information l w Parameter l corresponding to future output information u , to are the state space model parameter descriptions respectively.

[0112] Regarding the parameter dimensions that need to be calculated, parameter α exp The matrix size is r·NmJ×1, and the parameter θ exp The size of (N 2 m(m+l)+Nml)·NmJ. Since r≤(N 2 m(m+l)+Nml), if r is smaller, the data demand in this link will be lower.

[0113] For the calculation of state space model parameters, the existing state space model algorithm (such as the algorithm proposed by Kung) can be used to solve the model parameter matrices A, B, C and D. Take θ exp Parameters in Construct the following Hankel matrix H:

[0114]

[0115] By performing SVD decomposition on the H matrix, we can get H = USV T ; According to the decomposition results, the observation matrix and the extended controllable matrix in the input and output equations can be expressed as:

[0116] Γ=U(:,1:n)S 1 / 2 ;

[0117] Δ=S 1 / 2 V(:,1:n) T .

[0118] Where n is the preset model order. Furthermore, the matrices A, B, C and D of the model parameters can be calculated using the relationship between the above two matrices, such as B=Δ(:,1:m),C=Γ(1:l,:), in, To find the pseudo-inverse of a matrix.

[0119] That is to say, in this step, the processor can obtain the matrix Z corresponding to the data after the model mismatch exp and vector y exp ; Among them, y exp =vec(Y f,exp ), Y f,exp , W p,exp and U f,exp are the future output Hankel matrix, past Hankel matrix and future input Hankel matrix corresponding to the model mismatch data, respectively. 0,1 A matrix constructed for the preset 0s and 1s; by Calculate the estimated parameter α exp ; where Φ is the common matrix; α exp The matrix size is r·NmJ×1, where N is the length of the preset dynamic window for constructing the Hankel matrix, m is the number of output variables, J is the number of dynamic windows for constructing the Hankel matrix, and r≤(N 2 m(m+l)+Nml), r is the number of preset individual vectors, l is the number of input variables; through θ exp =Φα exp , calculate the identification parameter θ exp θ exp The matrix size is (N 2 m(m+l)+Nml)·NmJ; according to the identification parameter θ exp The parameter l in u , construct the target Hankel matrix H; where, to are the state space model parameter descriptions respectively; the target Hankel matrix H is subjected to singular value decomposition to obtain the singular value decomposition result; where H = USV T , the singular value decomposition results include the first matrix U, the second matrix S and the third matrix V; the model parameters are generated using the observation matrix Γ and the extended controllable matrix Δ; where Γ=U(:,1:n)S 1 / 2 , Δ=S 1 / 2 V(:,1:n) T , n is the preset model order; the model parameters include a first model parameter A, a second model parameter B, a third model parameter C and a fourth model parameter D; B=Δ(:,1:m),C=Γ(1:l,:),

[0120] In this embodiment, the embodiment of the present invention substitutes the algorithm data into the objective function based on similarity extraction to solve it, and extracts the common matrix between the data before the model mismatch and the data at the time of model mismatch. It can extract similarity information from the original past data and transfer it to the calculation of the re-identification process without the need to know the prior knowledge of the process model in advance, which is easy to implement; by calculating the model parameters of the process model based on the obtained data after the model mismatch and the common matrix using the subspace identification algorithm, it is possible to use the similarity information extracted from the past data to reduce the data demand for re-identification, thereby shortening the time required to obtain the model parameters, and has a high application value.

[0121] Corresponding to the above method embodiment, an embodiment of the present invention further provides a performance maintenance device for a control system. The performance maintenance device for a control system described below and the performance maintenance method for a control system described above can refer to each other.

[0122] Please refer to Figure 3 , Figure 3 This is a structural block diagram of a performance maintenance device for a control system provided by an embodiment of the present invention. The device may include:

[0123] An acquisition module 10 is used to acquire re-identification input data of a process model of a control system; wherein the re-identification input data includes data before model mismatch and data at the time of model mismatch;

[0124] A construction module 20 is used to construct algorithm data of a target type according to the re-identification input data; wherein the target form is a matrix and vector form in the subspace identification algorithm, and the algorithm data includes matrices and vectors corresponding to the data before the model mismatch and the data when the model mismatch occurs;

[0125] An extraction module 30 is used to substitute the algorithm data into the target function based on similarity extraction for solving, and extract the common matrix between the data before the model mismatch and the data when the model mismatch occurs;

[0126] The identification module 40 is used to calculate the model parameters of the process model using the subspace identification algorithm according to the acquired model mismatch data and the common matrix, so as to use the process model corresponding to the model parameters to predict the input data of the control system to obtain the system predicted output data of the control system.

[0127] In some embodiments, building blocks 20 may include:

[0128] The arrangement submodule is used to construct the future Hankel matrix and the past Hankel matrix of the data before the model mismatch and the data at the time of the model mismatch by using the input data and the output data of the data before the model mismatch and the data at the time of the model mismatch, respectively; wherein the future Hankel matrix includes the future output Hankel matrix and the future input Hankel matrix, and the past Hankel matrix includes the past output Hankel matrix and the past input Hankel matrix;

[0129] Construct a submodule for obtaining algorithm data based on the future Hankel matrix and the past Hankel matrix; wherein the algorithm data includes the matrix Z corresponding to the data before the model mismatch match and vector y match And the matrix Z corresponding to the data when the model mismatches mismatch and vector y mismatch ;y match =vec(Y f,match ), y mismatch =vec(Y f,mismatch ), Y f,match and Y f,mismatch are the future output Hankel matrices corresponding to the data before model mismatch and the data during model mismatch respectively; W p,match and W p,mismatch are the past Hankel matrices corresponding to the data before model mismatch and the data during model mismatch, respectively, U f,match and U f,mismatch are the future input Hankel matrices corresponding to the data before model mismatch and the data at model mismatch, respectively; I is the unit matrix, the size of I is Nm×Nm, N is the length of the preset dynamic window for constructing the Hankel matrix, and m is the number of output variables; N 0,1 A matrix constructed with preset 0s and 1s.

[0130] In some embodiments, the identification module 40 may include:

[0131] Arrangement construction submodule, used to obtain the matrix Z corresponding to the data after model mismatch exp and vector y exp ;in, Y f,exp , W p,exp and U f,exp are the future output Hankel matrix, past Hankel matrix and future input Hankel matrix corresponding to the data after model mismatch, respectively. 0,1 A matrix constructed for the preset 0s and 1s;

[0132] The estimation submodule is used to Calculate the estimated parameter α exp ; where Φ is the common matrix; α expThe matrix size is r·NmJ×1, where N is the length of the preset dynamic window for constructing the Hankel matrix, m is the number of output variables, J is the number of dynamic windows for constructing the Hankel matrix, and r≤(N 2 m(m+l)+Nml), r is the number of preset individual vectors, l is the number of input variables;

[0133] The calculation submodule is used to calculate the exp =Φα exp , calculate the identification parameter θ exp θ exp The matrix size is (N 2 m(m+l)+Nml)·NmJ;

[0134] The matrix construction submodule is used to identify the parameters θ exp The parameter l in u , construct the target Hankel matrix H; where, to They are the parameter descriptions of the state space model respectively;

[0135] The decomposition submodule is used to perform singular value decomposition on the target Hankel matrix H to obtain the singular value decomposition result; where H = USV T , the singular value decomposition results include the first matrix U, the second matrix S and the third matrix V;

[0136] The generation submodule is used to generate model parameters using the observation matrix Γ and the extended controllable matrix Δ; where Γ=U(:,1:n)S 1 / 2 , Δ=S 1 / 2 V(:,1:n) T , n is the preset model order; the model parameters include a first model parameter A, a second model parameter B, a third model parameter C and a fourth model parameter D; B=Δ(:,1:m),C=Γ(1:l,:),

[0137] In some embodiments, the apparatus may further include:

[0138] A preprocessing module, used for preprocessing the re-identification input data to obtain the preprocessed re-identification input data;

[0139] Correspondingly, the construction module 20 can be specifically used to construct algorithm data according to the pre-processed re-identification input data.

[0140] In some embodiments, the acquisition module 10 may include:

[0141] The excitation generation submodule is used to generate pre-model mismatch data using an excitation signal according to a process model.

[0142] In some embodiments, the objective function is and are the future output Hankel matrix, past Hankel matrix and future input Hankel matrix corresponding to the data match before model mismatch and the data mismatch when model mismatch respectively, N 0,1 A matrix constructed with preset 0 and 1, α g are the individual parameter vectors corresponding to the data match before model mismatch and the data mismatch when the model mismatch occurs, Φ is the common parameter matrix corresponding to the data match before model mismatch and the data mismatch when the model mismatch occurs, and I is the unit matrix.

[0143] In some embodiments, the extraction module 30 may include:

[0144] The data construction submodule is used to construct a data matrix according to the matrix in the algorithm data; wherein the data matrix includes a first data matrix Second data matrix and the third data matrix

[0145] Iterator module, used to pass Using self-consistent field iteration, the i-th individual vector is calculated The corresponding conversion parameter γ i ; Where i = 1, ..., r, r is the number of preset individual vectors,

[0146] The projection submodule is used to Get the i-th individual vector

[0147] The common generation submodule is used to obtain r individual vectors Then use r individual vectors Generate a common matrix Φ; where,

[0148] Update submodule for obtaining individual vectors If the number is less than r, the currently acquired i-th individual vector is used pass Update Z match and Z mismatch ;

[0149] The matrix update submodule is used to utilize the updated Z matchand Z mismatch , update the data matrix, and execute through Using self-consistent field iteration, the i-th individual vector is calculated The corresponding conversion parameter γ i steps.

[0150] In this embodiment, the embodiment of the present invention substitutes the algorithm data into the objective function based on similarity extraction through the extraction module 30 to solve it, and extracts the common matrix between the data before the model mismatch and the data at the time of the model mismatch. It can extract the similarity information from the original past data and transfer it to the calculation of the re-identification process without the need to know the prior knowledge of the process model in advance, which is easy to implement; through the identification module 40, according to the acquired data after the model mismatch and the common matrix, the model parameters of the process model are calculated using the subspace identification algorithm, and the similarity information extracted from the past data can be used to reduce the data demand for re-identification, thereby shortening the time required to obtain the model parameters, which has a high application value.

[0151] Corresponding to the above method embodiment, an embodiment of the present invention further provides a performance maintenance device for a control system. The performance maintenance device for a control system described below and the performance maintenance method for a control system described above can refer to each other.

[0152] Please refer to Figure 4 , Figure 4 A simplified structural diagram of a performance maintenance device for a control system provided by an embodiment of the present invention. The performance maintenance device for the control system may include:

[0153] A memory D1, for storing computer programs;

[0154] The processor D2 is used to implement the steps of the control system performance maintenance method provided by the above method embodiment when executing a computer program.

[0155] Accordingly, please refer to Figure 5 , Figure 5A specific structural diagram of a performance maintenance device of a control system provided in an embodiment of the present invention, the performance maintenance device 310 of the control system may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 322 (for example, one or more processors) and memory 332, one or more storage media 330 (for example, one or more mass storage devices) storing application programs 342 or data 344. Among them, the memory 332 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each unit may include a series of instruction operations on the host. Furthermore, the central processor 322 can be configured to communicate with the storage medium 330 to execute a series of instruction operations in the storage medium 330 on the control device 310 of the air compressor.

[0156] The performance maintenance device 310 of the control system may further include one or more power supplies 326, one or more wired or wireless network interfaces 350, one or more input and output interfaces 358, and / or one or more operating systems 341. For example, Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0157] The performance maintenance device of the control system provided in this embodiment may specifically be a server or a computer.

[0158] The steps in the control system performance maintenance method described above can be implemented by the structure of the control system performance maintenance device.

[0159] Corresponding to the above method embodiment, an embodiment of the present invention further provides a computer-readable storage medium. The computer-readable storage medium described below and the performance maintenance method of a control system described above can be referred to each other.

[0160] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the control system performance maintenance method provided by the above method embodiment are implemented.

[0161] The computer-readable storage medium may be a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or any other readable storage medium that can store program codes.

[0162] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the devices, equipment, and computer-readable storage media 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 part description.

[0163] The above is a detailed introduction to a performance maintenance method, device, equipment and computer-readable storage medium for a control system provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A performance maintenance method for a control system, characterized in that: include: Acquire re-identification input data of a process model of the control system; wherein the re-identification input data includes data before model mismatch and data at the time of model mismatch; According to the re-identification input data, construct algorithm data of target type; wherein the target form is a matrix and vector form in a subspace identification algorithm, and the algorithm data includes matrices and vectors corresponding to the data before the model mismatch and the data when the model mismatch occurs; Substituting the algorithm data into the objective function based on similarity extraction for solving, and extracting a common matrix between the data before the model mismatch and the data when the model mismatch occurs; According to the acquired model mismatch data and the common matrix, the model parameters of the process model are calculated using a subspace identification algorithm, so as to obtain the system predicted output data of the control system by using the process model corresponding to the model parameters and the input data of the control system; The objective function is and are the future output Hankel matrix, past Hankel matrix and future input Hankel matrix corresponding to the data match before model mismatch and the data mismatch when model mismatch respectively, N 0,1 A matrix constructed with preset 0 and 1, α g are the individual parameter vectors corresponding to the data match before model mismatch and the data mismatch when model mismatch occurs, is the common parameter matrix corresponding to the data match before model mismatch and the data mismatch when model mismatch occurs, and I is the unit matrix; Substituting the algorithm data into the objective function based on similarity extraction for solving, and extracting a common matrix between the data before the model mismatch and the data when the model mismatch occurs, comprises: According to the matrix in the algorithm data, a data matrix is ​​constructed; wherein the data matrix includes a first data matrix The second data matrix and the third data matrix pass Using self-consistent field iteration, the i-th individual vector is calculated The corresponding conversion parameter γ i ; Where i = 1, ..., r, r is the number of preset individual vectors, pass Get the i-th individual vector If we get r individual vectors Then use r individual vectors Generate a common matrix Φ; where, If the individual vectors obtained If the number is less than r, the currently acquired i-th individual vector is used pass Update Z match and Z mismatch ; Using the updated Z match and Z mismatch , update the data matrix, and execute the Using self-consistent field iteration, the i-th individual vector is calculated The corresponding conversion parameter γ i steps.

2. The performance maintenance method of the control system according to claim 1, characterized in that: The step of constructing algorithm data of a target type according to the re-identified input data comprises: Using the input data and output data of the data before the model mismatch and the data at the time of the model mismatch, respectively, constructing the future Hankel matrix and the past Hankel matrix of the data before the model mismatch and the data at the time of the model mismatch, respectively; wherein the future Hankel matrix includes a future output Hankel matrix and a future input Hankel matrix, and the past Hankel matrix includes a past output Hankel matrix and a past input Hankel matrix; According to the future Hankel matrix and the past Hankel matrix, the algorithm data is obtained; wherein the algorithm data includes the matrix Z corresponding to the data before the model mismatch match and vector y match And the matrix Z corresponding to the data when the model is mismatched mismatch and vector y mismatch ;y match =vec(Y f,match ), y mismatch =vec(Y f,mismatch ), Y f,match and Y f,mismatch are the future output Hankel matrices corresponding to the data before the model mismatch and the data when the model mismatch respectively; W p,match and W p,mismatch are the past Hankel matrices corresponding to the data before the model mismatch and the data when the model mismatch occurs, respectively, U f,match and U f,mismatch are the future input Hankel matrices corresponding to the data before the model mismatch and the data when the model mismatch respectively; I is the unit matrix, the size of I is Nm×Nm, N is the length of the preset dynamic window for constructing the Hankel matrix, and m is the number of output variables; N 0,1 A matrix constructed with preset 0s and 1s.

3. The performance maintenance method of the control system according to claim 1, characterized in that: The method of calculating the model parameters of the process model using a subspace identification algorithm based on the acquired model mismatch data and the common matrix includes: Get the matrix Z corresponding to the data after the model mismatch exp and vector y exp ; Among them, y exp =vec(Y f,exp ), Y f,exp , W p,exp and U f,exp are the future output Hankel matrix, past Hankel matrix and future input Hankel matrix corresponding to the model mismatch data, respectively. 0,1 A matrix constructed for the preset 0s and 1s; pass Calculate the estimated parameter α exp ; Wherein, Φ is the common matrix; α exp The matrix size is r·NmJ×1, where N is the length of the preset dynamic window for constructing the Hankel matrix, m is the number of output variables, J is the number of dynamic windows for constructing the Hankel matrix, and r≤(N 2 m(m+l)+Nml), r is the number of preset individual vectors, l is the number of input variables; Through θ exp =Φα exp , calculate the identification parameter θ exp θ exp The matrix size is (N 2 m(m+l)+Nml)·NmJ; According to the identification parameter θ exp The parameter l in u , construct the target Hankel matrix H; where, to They are the parameter descriptions of the state space model respectively; Perform singular value decomposition on the target Hankel matrix H to obtain a singular value decomposition result; wherein H = USV T , the singular value decomposition result includes a first matrix U, a second matrix S and a third matrix V; The model parameters are generated using the observation matrix Γ and the extended controllable matrix Δ; where Γ=U(:,1:n)S 1 / 2 , Δ=S 1 / 2 V(:,1:n) T , n is a preset model order; the model parameters include a first model parameter A, a second model parameter B, a third model parameter C and a fourth model parameter D; B=Δ(:,1:m),C=Γ(1:l,:), 4. The performance maintenance method of the control system according to claim 1, characterized in that: Before constructing the algorithm data of the target type according to the re-identified input data, the method further includes: Preprocessing the re-identification input data to obtain the preprocessed re-identification input data; Correspondingly, constructing algorithm data of a target type according to the re-identified input data includes: The algorithm data is constructed according to the pre-processed re-identification input data.

5. The performance maintenance method of the control system according to claim 1, characterized in that: The step of obtaining re-identified input data of a process model of the control system includes: The pre-model mismatch data is generated using an excitation signal according to the process model.

6. A performance maintenance device for a control system, characterized in that: include: An acquisition module, used for acquiring re-identification input data of a process model of a control system; wherein the re-identification input data includes data before model mismatch and data at the time of model mismatch; A construction module, configured to construct algorithm data of a target type according to the re-identification input data; wherein the target form is a matrix and vector form in a subspace identification algorithm, and the algorithm data includes matrices and vectors corresponding to the data before the model mismatch and the data when the model mismatch occurs; An extraction module, used for substituting the algorithm data into the target function based on similarity extraction for solving, and extracting a common matrix between the data before the model mismatch and the data when the model mismatch occurs; An identification module, used to calculate the model parameters of the process model using a subspace identification algorithm according to the acquired model mismatch data and the common matrix, so as to predict the system predicted output data of the control system using the process model corresponding to the model parameters and the input data of the control system; The objective function is and are the future output Hankel matrix, past Hankel matrix and future input Hankel matrix corresponding to the data match before model mismatch and the data mismatch when model mismatch respectively, N 0,1 A matrix constructed with preset 0 and 1, α g are the individual parameter vectors corresponding to the data match before model mismatch and the data mismatch when model mismatch occurs, is the common parameter matrix corresponding to the data match before model mismatch and the data mismatch when model mismatch occurs, and I is the unit matrix; The extraction module comprises: A data construction submodule is used to construct a data matrix according to the matrix in the algorithm data; wherein the data matrix includes a first data matrix The second data matrix and the third data matrix Iterator module, used to pass Using self-consistent field iteration, the i-th individual vector is calculated The corresponding conversion parameter γ i ; Where i = 1, ..., r, r is the number of preset individual vectors, The projection submodule is used to Get the i-th individual vector The common generation submodule is used to obtain r individual vectors Then use r individual vectors Generate a common matrix Φ; where, Update submodule for obtaining individual vectors If the number is less than r, the currently acquired i-th individual vector is used pass Update Z match and Z mismatch ; The matrix update submodule is used to utilize the updated Z match and Z mismatch , update the data matrix, and send a start signal to the iterative submodule.

7. A performance maintenance device for a control system, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the performance maintenance method of the control system as described in any one of claims 1 to 5 when executing the computer program.

8. 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 performance maintenance method of the control system according to any one of claims 1 to 5 are implemented.

Images