Industrial signal fault recognition method based on graph regularized semi-supervised dictionary learning

Abstract

In the disclosed embodiment, a method for industrial signal fault identification based on graph regularized semi-supervised dictionary learning is provided, which belongs to the field of data identification technology, and specifically includes: collecting industrial signal data of target equipment; using labeled data and unlabeled data as training data sets and inputting them into a graph regularized semi-supervised dictionary learning model to obtain training set sparse coding, confidence matrix and structural dictionary; bringing the test set signal and the structural dictionary into the classification model to obtain the test set sparse coding of the test set signal; calculating the reconstruction error of the test set signal, and identifying the fault category of the test set signal according to the reconstruction error. Through the scheme disclosed in the present invention, dictionary learning is combined with semi-supervised learning and popular learning, which can not only preserve the manifold structure of the data, but also learn the characteristics of unlabeled data, thereby enhancing the discriminability and reliability of model learning, and improving the efficiency, accuracy and adaptability of recognition.

Description

Technical Field

[0001] The disclosed embodiments relate to the technical field of data recognition, and in particular to an industrial signal fault recognition method based on graph regularization semi-supervised dictionary learning. Background Art

[0002] At present, the operating status of the production process and field equipment in the industrial environment will directly affect the reliability of the production process and the service life of the equipment. Accurately judging the operating status of the equipment and the production process will largely avoid accidents. Therefore, it is of great significance to perform fault diagnosis on industrial signal analysis. The existing industrial process fault diagnosis methods mainly include fault diagnosis methods based on mechanism models and fault diagnosis methods based on data-driven. The fault diagnosis method based on mechanism models is too dependent on industrial background knowledge, the model reliability is low, and thus the generalization ability is not strong. With the rapid development of computer technology and Internet of Things technology, data-driven methods have made great progress in the field of equipment fault diagnosis. Data-driven methods generally use field signals collected by sensors, use deep learning and machine learning methods to analyze signals, learn the internal features of signals, and finally identify the fault category. However, deep learning methods have problems such as long training time, high requirements on computers, large amount of learning process calculations, long time consumption, and difficulty in application in industrial sites. Dictionary learning methods are not constrained by the above problems, have high model accuracy and good reliability, and are widely used in the field of fault diagnosis. However, many signals collected in on-site industrial environments are unlabeled, and most traditional dictionary learning methods focus on studying fully supervised fault diagnosis, which is too dependent on labeled data. To address this problem, an effective method is needed to achieve better diagnostic results even when there is less labeled data. The learning process of dictionary learning is to minimize the reconstruction error under constraints, and then use the corresponding solution algorithm to solve the dictionary and sparse coding. However, the existing dictionary learning algorithm makes good use of the labeled characteristics of data, so the model can achieve ideal results. However, the common problems of such models are: 1. Over-reliance on a large amount of labeled data, 2. Only the data characteristics of the data itself are learned without learning the internal structure characteristics of the data.

[0003] It can be seen that there is an urgent need for an efficient, accurate and adaptable industrial signal fault recognition method based on graph regularized semi-supervised dictionary learning. Summary of the invention

[0004] In view of this, an embodiment of the present disclosure provides an industrial signal fault identification method based on graph regularization semi-supervised dictionary learning, which at least partially solves the problems of poor recognition efficiency, recognition accuracy and adaptability in the prior art.

[0005] The present disclosure provides an industrial signal fault identification method based on graph regularization semi-supervised dictionary learning, comprising:

[0006] Collecting industrial signal data of a target device, wherein the industrial signal data includes labeled data and unlabeled data;

[0007] The labeled data and the unlabeled data are used as training data sets and input into a graph-regularized semi-supervised dictionary learning model to obtain a training set sparse code, a confidence matrix, and a structure dictionary, wherein the training set sparse code includes a labeled sparse code corresponding to the labeled data and an unlabeled sparse code corresponding to the unlabeled data;

[0008] Collecting a test set signal of the target device, and bringing the test set signal and the structure dictionary into a classification model to obtain a test set sparse coding of the test set signal;

[0009] A reconstruction error of the test set signal is calculated according to the test set sparse coding, and a fault category of the test set signal is identified according to the reconstruction error.

[0010] According to a specific implementation of an embodiment of the present disclosure, the step of using the labeled data and the unlabeled data as training data sets and inputting them into a graph-regularized semi-supervised dictionary learning model to obtain a training set sparse coding, a confidence matrix, and a structure dictionary includes:

[0011] Preserving the flow structure of the labeled data and simplifying it into a graph-regularized labeled sparse coding;

[0012] Initialize the supervised dictionary using the K-SVD algorithm, and then use the matching pursuit algorithm to obtain the initialized sparse code corresponding to the unlabeled data;

[0013] Obtaining the confidence matrix using the reconstruction error of the unlabeled data;

[0014] Updating an extended dictionary according to the confidence matrix;

[0015] Solving the initialized sparse code using the matching pursuit algorithm to obtain the unlabeled sparse code;

[0016] Updating the labeled sparse code using the FISTA back-propagation method;

[0017] Update the supervision dictionary using the Metaface method;

[0018] The structure dictionary is formed according to the updated extended dictionary and the updated supervised dictionary.

[0019] According to a specific implementation of an embodiment of the present disclosure, the step of saving the flow structure of the labeled data and simplifying it into a graph-regularized labeled sparse coding includes:

[0020] Calculate k nearest neighbor points of each labeled data using k nearest neighbor algorithm;

[0021] Calculate a weight coefficient matrix according to the k nearest neighbor points;

[0022] All the labeled data are mapped to a low-dimensional space according to the weight coefficient matrix, and simplified into a graph-regularized labeled sparse code through a Laplacian matrix.

[0023] According to a specific implementation of the embodiment of the present disclosure, the step of obtaining the confidence matrix using the reconstruction error of the unlabeled data includes:

[0024] The confidence matrix is ​​obtained by calculating the confidence value that each of the unlabeled data belongs to a different class.

[0025] According to a specific implementation of the embodiment of the present disclosure, the step of updating the extended dictionary according to the confidence matrix includes:

[0026] The extended dictionary is updated according to the confidence matrix using singular value decomposition.

[0027] According to a specific implementation of an embodiment of the present disclosure, the step of collecting the test set signal of the target device, bringing the test set signal and the structure dictionary into a classification model, and obtaining a test set sparse coding of the test set signal includes:

[0028] Bringing the test set signal and the structure dictionary into the classification model, and obtaining a sparse representation corresponding to the test set signal according to the structure dictionary;

[0029] The sparse representation is solved using the matching pursuit algorithm to obtain the test set sparse coding.

[0030] The industrial signal fault identification scheme based on graph regularized semi-supervised dictionary learning in the embodiment of the present disclosure includes: collecting industrial signal data of the target device, wherein the industrial signal data includes labeled data and unlabeled data; using the labeled data and the unlabeled data as training data sets and inputting them into a graph regularized semi-supervised dictionary learning model to obtain training set sparse coding, a confidence matrix and a structural dictionary, wherein the training set sparse coding includes labeled sparse coding corresponding to the labeled data and unlabeled sparse coding corresponding to the unlabeled data; collecting a test set signal of the target device, and bringing the test set signal and the structural dictionary into a classification model to obtain a test set sparse coding of the test set signal; calculating a reconstruction error of the test set signal according to the test set sparse coding, and identifying the fault category of the test set signal according to the reconstruction error.

[0031] The beneficial effects of the embodiments of the present disclosure are as follows: through the scheme of the present disclosure, dictionary learning is combined with semi-supervised learning to solve the problem that the model learning effect is not ideal due to the small amount of labeled data in the industrial environment, and dictionary learning is combined with graph regularization to solve the problem that the traditional method only considers the data domain characteristics of the data itself but ignores the internal geometric structure characteristics of the data, thereby enhancing the discriminability and reliability of model learning and improving the efficiency, accuracy and adaptability of recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 A flowchart of an industrial signal fault identification method based on graph regularization semi-supervised dictionary learning provided by an embodiment of the present disclosure;

[0034] Figure 2 A schematic diagram of an industrial signal data processing flow provided by an embodiment of the present disclosure;

[0035] Figure 3 A schematic diagram of a specific process of industrial signal data processing provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0036] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0037] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0038] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0039] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0040] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0041] The embodiment of the present disclosure provides an industrial signal fault identification method based on graph regularization semi-supervised dictionary learning, which can be applied to the equipment operation status identification process in industrial production scenarios.

[0042] See also Figure 1 , is a flow chart of an industrial signal fault identification method based on graph regularization semi-supervised dictionary learning provided by an embodiment of the present disclosure. Figure 1 and Figure 2 As shown, the method mainly comprises the following steps:

[0043] S101, collecting industrial signal data of a target device, wherein the industrial signal data includes labeled data and unlabeled data;

[0044] In specific implementation, when it is necessary to identify the industrial signal corresponding to a certain device to determine the fault type, it can be first used as the target device, and then the industrial signal data generated by the target device during its historical movement process can be collected, wherein the industrial signal data includes labeled data and unlabeled data.

[0045] S102, using the labeled data and the unlabeled data as training data sets and inputting them into a graph-regularized semi-supervised dictionary learning model to obtain a training set sparse code, a confidence matrix, and a structure dictionary, wherein the training set sparse code includes a labeled sparse code corresponding to the labeled data and an unlabeled sparse code corresponding to the unlabeled data;

[0046] Furthermore, in step S102, the labeled data and the unlabeled data are used as training data sets and input into a graph-regularized semi-supervised dictionary learning model to obtain a training set sparse coding, a confidence matrix, and a structure dictionary, including:

[0047] Preserving the flow structure of the labeled data and simplifying it into a graph-regularized labeled sparse coding;

[0048] Initialize the supervised dictionary using the K-SVD algorithm, and then use the matching pursuit algorithm to obtain the initialized sparse code corresponding to the unlabeled data;

[0049] Obtaining the confidence matrix using the reconstruction error of the unlabeled data;

[0050] Updating an extended dictionary according to the confidence matrix;

[0051] Solving the initialized sparse code using the matching pursuit algorithm to obtain the unlabeled sparse code;

[0052] Updating the labeled sparse code using the FISTA back-propagation method;

[0053] Update the supervision dictionary using the Metaface method;

[0054] The structure dictionary is formed according to the updated extended dictionary and the updated supervised dictionary.

[0055] Optionally, the step of saving the flow structure of the labeled data and simplifying it into graph-regularized labeled sparse coding includes:

[0056] Calculate k nearest neighbor points of each labeled data using k nearest neighbor algorithm;

[0057] Calculate a weight coefficient matrix according to the k nearest neighbor points;

[0058] All the labeled data are mapped to a low-dimensional space according to the weight coefficient matrix, and simplified into a graph-regularized labeled sparse code through a Laplacian matrix.

[0059] Optionally, the step of obtaining the confidence matrix by using the reconstruction error of the unlabeled data includes:

[0060] The confidence matrix is ​​obtained by calculating the confidence value that each of the unlabeled data belongs to a different class.

[0061] Optionally, the step of updating the extended dictionary according to the confidence matrix includes:

[0062] The extended dictionary is updated according to the confidence matrix using singular value decomposition.

[0063] In the specific implementation, considering the situation where there is only a small amount of labeled data and a large amount of unlabeled data, we can use semi-supervised learning to fully explore the features in the unlabeled data to make up for the shortcomings of insufficient labeled data. In this invention, we propose a new semi-supervised dictionary learning method. When learning the dictionary, we learn the corresponding category probability distribution P for each unlabeled data i,j , the category probability distribution of each data is combined into a confidence matrix P. In addition, we learn an extended dictionary through unlabeled data, and use the extended dictionary learned from unlabeled data as a supplement to the overall dictionary, which can enhance the reliability of the model and further improve the dictionary with high confidence samples, which shows that this method is robust to label assignment errors. Traditional dictionary learning methods only learn the data domain features that exist in the data itself. The study of manifold learning found that in addition to the features of the data domain, the data itself also has internal geometric structure features. Nonlinear high-dimensional data has an inherent low-dimensional structure, namely the manifold structure. The manifold learning method is an effective method to discover the inherent manifold structure of nonlinear data. By learning the internal geometric structure features of the data, the discriminability of the model can be enhanced. Local linear embedding is an effective popular learning method. In this invention, we consider using local linear embedding to save the manifold structure from the original data to the sparse coding space, which can be considered as the regularization of sparse coding, and finally transform it into a graph regularized form of sparse coding.

[0064] By combining dictionary learning with semi-supervised learning and manifold learning, the objective function of the graph-regularized semi-supervised dictionary learning model we proposed is as follows:

[0065]

[0066]

[0067] In the above formula, X=[X1,…,X i ,…,X C ] is a labeled training set, which contains a total of C different types of samples. is the i-th type of training samples, each type of labeled training samples contains N l samples, X i Each column of is a training sample data; It contains N u unlabeled training samples. i E i ] represents the learned dictionary, which is different from traditional dictionary learning. We divide the required dictionary into two parts, D = [D1,…D i ,…,D C ] and E=[E1,…E i ,…E C ], where D i represents the supervised dictionary of the i-th category, E i The extended dictionary represents the i-th extended dictionary. The extended dictionary is mainly used to explore the discriminative features of unlabeled data as a supplement to the overall dictionary. i E i ] is related to the i-th sample, we expect the dictionary [D i E i ] can better represent the i-th class samples, but poorly represent other class samples.

[0068] A=[A1,…A i ,…,A C ] indicates sparse coding of labeled data, Represents the i-th category sparse coding, each category contains N l A sparse code. is the sparse coding of unlabeled data, which represents the sparse coding of the i-th category of the j-th data. i,j Indicates the confidence that the unlabeled data belongs to a certain category, P i,j Specifically, it represents the confidence that the jth unlabeled data belongs to the i-th category. A It is the Laplace matrix calculated using labeled data samples in the local linear embedding method, L A Used for sparse coding graph regularization. Constraints It means that the sum of the confidences of the j-th unlabeled data belonging to each class of samples is 1. α and β are both positive parameters used to control the weights of the corresponding items of the objective function.

[0069] The specific meaning of each item in the objective function is explained as follows:

[0070] It represents the reconstruction error term of labeled data. We should try to make the reconstruction error as small as possible. Represents graph regularized sparse coding. In order to preserve the popular structure of the data, we use the local linear embedding method to preserve the popular structure of the data and finally simplify it into the graph regularized sparse coding form;

[0071] Represents the reconstruction error term of unlabeled data, and the dictionary [D i E i ] represents the data better and satisfies the requirement of minimizing the reconstruction error; the confidence level P i,j It can ensure that the unlabeled samples with high confidence have a greater impact, while the unlabeled samples with low confidence have a smaller impact. This makes the method more flexible in using unlabeled data and more robust to label assignment errors.

[0072] β||A i ||1. They are sparse coding of labeled data and unlabeled data respectively. These two items are to ensure that the sparse coding meets the corresponding sparse characteristics.

[0073] Based on the above analysis, we know that we introduce a confidence matrix P for unlabeled data, through which we can know the confidence of the sample belonging to each category. In order to learn the features of unlabeled data more flexibly, we make full use of samples with high confidence to learn an extended dictionary, which can be used to improve the dictionary and enhance the representation ability of the dictionary; in addition, we use the local linear embedding method to preserve the manifold structure of the data and simplify it into a graph regularization form to enhance the discriminability and reliability of the dictionary.

[0074] Then, the objective function of the graph-regularized semi-supervised dictionary learning model is optimized. The specific steps may be: The main steps of the optimization process are as follows:

[0075] Step 1.1:

[0076] Initialization: We first use the K-SVD algorithm to initialize the dictionary D, and then use the matching pursuit algorithm (OMP) to initialize the sparse code y.

[0077] Step 1.2:

[0078] Update the confidence matrix P: We calculate the confidence P of each sample belonging to different classes i,j To update the confidence matrix P, for example, the confidence that the j-th unlabeled data belongs to the i-th category is updated as follows:

[0079]

[0080] in represents the i-th type reconstruction error of the j-th unlabeled data, and β is a positive parameter.

[0081] Step 1.3:

[0082] Update extended dictionary E i : We will Rewrite it as:

[0083]

[0084] Then using singular value decomposition:

[0085] So the dictionary E i Update by: E i =U(:,n)

[0086] Where n is the number of atoms in the extended dictionary.

[0087] Step 1.4:

[0088] Update the sparse code y j :We only update the sparse code y by fixing other terms j , the optimization problem is transformed into:

[0089]

[0090] Rewrite the above formula as

[0091] After rewriting, B i Represents the product of the confidence that all samples belong to the i-th category and the corresponding unlabeled samples Y i Represents the product of the confidence that all samples belong to the i-th category and their corresponding sparse code This way we can easily update the sparse code Y i , we use the OMP method to solve Y i =[D i E i ]([D i E i ] T [D i E i ]) -1 [D i E i ] T B i .

[0092] Step 1.5:

[0093] Update sparse coding A i :We first fix other items and only update the sparse code Ai , the optimization problem is transformed into:

[0094]

[0095] FISTA back propagation can be used to effectively solve sparse coding. When iterates to n+1 times:

[0096]

[0097] in λ, τ>0, Representation function right The derivative of

[0098] Step 1.6:

[0099] Update dictionary D i :We first fix other items and only update dictionary D i , the optimization problem is transformed into:

[0100]

[0101] We can use the Metaface method to update the dictionary atoms by class and effectively solve the dictionary D i , solve to get the lth dictionary atom of the i-th dictionary

[0102]

[0103] in represents the i-th dictionary, represents the jth dictionary atom of the i-th dictionary, each dictionary contains k dictionary atoms; all sparse codes A = [A1,…A i ,…,A C ] contains C types of sparse codes, the i-th type of sparse code represents the sparse code of the i-th type of the l-th labeled data; the sparse code of the i-th type of the unlabeled data represents the sparse coding of the i-th category of the l-th unlabeled data; X i is the labeled sample of the i-th category, B i Represents the product of the confidence that all samples belong to the i-th category and the corresponding unlabeled samples

[0104] S103, collecting a test set signal of the target device, and bringing the test set signal and the structure dictionary into a classification model to obtain a test set sparse coding of the test set signal;

[0105] On the basis of the above embodiment, the step of collecting the test set signal of the target device and bringing the test set signal and the structure dictionary into the classification model to obtain the test set sparse coding of the test set signal in step S103 includes:

[0106] Bringing the test set signal and the structure dictionary into the classification model, and obtaining a sparse representation corresponding to the test set signal according to the structure dictionary;

[0107] The sparse representation is solved using the matching pursuit algorithm to obtain the test set sparse coding.

[0108] In specific implementation, the classification model may be as follows:

[0109]

[0110] We will solve the dictionary [D i E i ] is introduced into the above classification model, and the dictionary obtained by graph regularized semi-supervised dictionary learning is used to sparsely represent the test set signal, and then the sparse coding A corresponding to the test set signal is solved by the OMP algorithm. i .

[0111] S104, calculating a reconstruction error of the test set signal according to the test set sparse coding, and identifying a fault category of the test set signal according to the reconstruction error.

[0112] In the specific implementation, after solving the sparse coding of the test sample, the reconstruction error of the sample is calculated ε=[ε1,…,ε C ],ε i The specific process of training and identifying industrial signal data using the industrial signal fault identification method based on graph regularization semi-supervised dictionary learning provided in this embodiment is as follows: Figure 3 shown.

[0113]

[0114] Sample x category = min{ε}

[0115] The industrial signal fault recognition method based on graph regularization semi-supervised dictionary learning provided in this embodiment can effectively solve the problem of unsatisfactory model learning effect due to less labeled data in the industrial environment by combining dictionary learning with semi-supervised learning and graph regularization.

[0116] Introducing a confidence level for unlabeled data to more flexibly learn unlabeled data features, which can be robust to label assignment errors;

[0117] Learning an extended dictionary from unlabeled data can enhance the reliability of the model and further improve the dictionary with high-confidence samples;

[0118] The present invention solves the problem that traditional methods only consider the data domain characteristics of the data itself but ignore the internal geometric structure characteristics of the data, thereby enhancing the discriminability and reliability of model learning and improving the efficiency, accuracy and adaptability of recognition.

[0119] The units involved in the embodiments described in the present disclosure may be implemented by software or by hardware.

[0120] It should be understood that various parts of the present disclosure may be implemented in hardware, software, firmware, or a combination thereof.

[0121] The above is only a specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present disclosure should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A method for industrial signal fault recognition based on graph regularization semi-supervised dictionary learning, characterized in that: include: Collecting industrial signal data of a target device, wherein the industrial signal data includes labeled data and unlabeled data; The labeled data and the unlabeled data are used as training data sets and input into a graph-regularized semi-supervised dictionary learning model to obtain a training set sparse code, a confidence matrix, and a structure dictionary, wherein the training set sparse code includes a labeled sparse code corresponding to the labeled data and an unlabeled sparse code corresponding to the unlabeled data; The step of using the labeled data and the unlabeled data as training data sets and inputting them into a graph-regularized semi-supervised dictionary learning model to obtain a training set sparse coding, a confidence matrix, and a structure dictionary includes: Preserving the flow structure of the labeled data and simplifying it into a graph-regularized labeled sparse coding; Initialize the supervised dictionary using the K-SVD algorithm, and then use the matching pursuit algorithm to obtain the initialized sparse code corresponding to the unlabeled data; Obtaining the confidence matrix using the reconstruction error of the unlabeled data; Updating an extended dictionary according to the confidence matrix; Solving the initialized sparse code using the matching pursuit algorithm to obtain the unlabeled sparse code; Updating the labeled sparse code using the FISTA back-propagation method; Update the supervision dictionary using the Metaface method; Forming the structure dictionary according to the updated extended dictionary and the updated supervised dictionary; The step of saving the flow structure of the labeled data and simplifying it into graph-regularized labeled sparse coding includes: Calculate k nearest neighbor points of each labeled data using k nearest neighbor algorithm; Calculate a weight coefficient matrix according to the k nearest neighbor points; Mapping all the labeled data to a low-dimensional space according to the weight coefficient matrix, and simplifying it into a graph-regularized labeled sparse code through a Laplacian matrix; Collecting a test set signal of the target device, and bringing the test set signal and the structure dictionary into a classification model to obtain a test set sparse coding of the test set signal; A reconstruction error of the test set signal is calculated according to the test set sparse coding, and a fault category of the test set signal is identified according to the reconstruction error.

2. The method according to claim 1, characterized in that The step of obtaining the confidence matrix by using the reconstruction error of the unlabeled data comprises: The confidence matrix is ​​obtained by calculating the confidence value that each of the unlabeled data belongs to a different class.

3. The method according to claim 1, characterized in that The step of updating the extended dictionary according to the confidence matrix comprises: The extended dictionary is updated according to the confidence matrix using singular value decomposition.

4. The method according to claim 1, characterized in that The step of collecting the test set signal of the target device, bringing the test set signal and the structure dictionary into the classification model to obtain the test set sparse coding of the test set signal includes: Bringing the test set signal and the structure dictionary into the classification model, and obtaining a sparse representation corresponding to the test set signal according to the structure dictionary; The sparse representation is solved using the matching pursuit algorithm to obtain the test set sparse coding.

Images