Transformer fault diagnosis method based on near infrared spectrum and related equipment

Abstract

The invention discloses a transformer fault diagnosis method based on a near infrared spectrum and related equipment, and relates to the technical field of transformer fault diagnosis, and the method comprises the steps: S1, obtaining the near infrared spectrum information of a transformer, and carrying out the standard normal transformation of the near infrared spectrum information, and obtaining a first spectrum signal; s2, performing dimension reduction calculation on the first spectral signal to obtain spectral features corresponding to different operation states of the transformer; s3, constructing a transformer fault diagnosis model based on the spectral features and the corresponding insulating oil category labels; and S4, detecting a to-be-detected near infrared spectrum data set of a to-be-detected transformer based on the transformer fault diagnosis model, outputting the probability of each insulating oil type, determining the insulating oil type of the to-be-detected transformer according to the maximum probability, and judging whether the to-be-detected transformer has a fault or not according to the insulating oil type to complete transformer fault detection. And the problems of narrow detection range and poor adaptability of the transformer detection model due to the fact that spectral characteristics of different operation states of the transformer are not considered are solved.

Description

Technical Field

[0001] The invention relates to the technical field of transformer fault diagnosis, and in particular to a transformer fault diagnosis method based on near infrared spectroscopy and related equipment. Background Art

[0002] There are many reasons for transformer failures and accidents, including external damage, natural disasters, equipment defects, insulation aging, material degradation, etc. Existing transformer spectral detection mostly uses gas chromatography, etc., which can only be tested within a certain period of time each time, and has poor applicability; or some detection methods mainly obtain the infrared spectrum and Raman spectrum of aged insulating oil paper, fuse the spectral features to obtain the degree of polymerization value, and evaluate the real-time insulation aging of the transformer; because the transformer insulating oil category is not identified, the infrared spectral characteristics of the transformer in different operating states are not comprehensively considered, and the focus is on data sampling of the aged insulating oil, resulting in the transformer diagnosis model being unable to fully learn the different insulating oil states of the transformer, which in turn leads to low fault detection accuracy and inability to accurately identify the fault type of the transformer.

[0003] Chinese patent, publication number: CN118209537A, publication date: June 18, 2024, discloses a transformer oil-paper insulation aging diagnosis method based on fusion spectrum and neural network. The oil-paper aging sample is obtained through accelerated thermal aging experiment. Based on the diffusion balance of substances between oil and paper, the spectral data of insulating oil is analyzed to indirectly reflect the aging state of insulating paper. Combined with the self-help method, the sample training set and test set are divided, and the oil-paper insulation aging diagnosis model is established based on the one-dimensional convolutional neural network and the neural architecture search algorithm. The infrared spectrum and Raman spectrum characteristic variables of the transformer aging insulating oil are used as input to output the polymerization degree value of the insulating paper. However, the invention does not take into account the complexity and diversity of transformer data, lacks the identification of transformer insulating oil categories, resulting in the established oil-paper insulation aging diagnosis model can only learn the infrared light characteristics of the aged ones, which leads to low model learning and generalization performance. In addition, the neural architecture search algorithm will fall into a local solution, thereby affecting the model calculation performance and the final transformer fault identification results. Summary of the invention

[0004] The purpose of the present invention is to solve the problem that the existing transformer spectrum detection method does not comprehensively consider the near-infrared spectrum characteristics of the transformer in different operating states, resulting in a narrow detection range and poor adaptability; a transformer fault diagnosis method based on near-infrared spectroscopy and related equipment are proposed, a first spectrum signal is obtained by performing a standard normal transformation on the near-infrared spectrum information of the transformer, and the spectrum characteristics of the transformer in different operating states are obtained after the first spectrum signal is reduced in dimension; a transformer fault diagnosis model is constructed based on the spectrum characteristics and the corresponding insulating oil category labels, so that the model can comprehensively and accurately detect and warn various potential faults of the transformer; based on the optimization model, the near-infrared spectrum data set to be tested is detected, the probability of the insulating oil type is output, and the type with the largest probability is selected as the type of the transformer insulating oil to be tested, and compared with the standard insulating oil state database to determine whether the transformer to be tested is faulty; the problem that the detection range is narrow and the adaptability is poor due to the failure to comprehensively consider the near-infrared spectrum characteristics of the transformer in different operating states is overcome, the adaptability and detection efficiency of the transformer spectrum detection method are significantly improved, and the accuracy and timeliness of transformer fault diagnosis are further improved.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a transformer fault diagnosis method based on near infrared spectroscopy, comprising the following steps: S1, obtaining near-infrared spectrum information of the transformer and performing standard normal transformation on it to obtain a first spectrum signal; S2. Perform dimensionality reduction calculation on the first spectral signal to obtain spectral features corresponding to different operating states of the transformer; S3, constructing a transformer fault diagnosis model based on the spectral characteristics and the corresponding insulating oil category label; S4. Based on the transformer fault diagnosis model, the near-infrared spectrum data set of the transformer to be tested is detected, the probability of each insulating oil type is output, and the insulating oil type of the transformer to be tested is determined according to the maximum probability value, and whether the transformer to be tested is faulty is judged by the insulating oil type, thereby completing the transformer fault detection.

[0006] In this scheme, by performing a standard normal transformation on the collected near-infrared spectral information of the transformer, the noise signal in the fluorescence spectral data can be reduced, and the spectral differences caused by factors such as sample particle size, scattering, and optical path changes can be eliminated, so that the spectral characteristics between different insulating oil states are more comparable, which is convenient for the extraction of spectral features; the first spectral signal is reduced in dimension, so that feature optimization and feature extraction reduce the data dimension, improve the efficiency of feature extraction, and remove those redundant and useless information for classification tasks, so as to provide high-quality data support for strengthening the learning ability of the transformer fault diagnosis model; by constructing an optimization model with the extracted spectral features and the corresponding insulating oil category labels, the model can comprehensively learn the spectral characteristics of different insulating oil states, ensuring that when diagnosing the fault of the transformer to be tested, the insulating oil type of the transformer to be tested can be quickly and accurately determined, and the fault type of the transformer can be accurately identified, so as to efficiently complete the detection of the transformer fault state; at the same time, it is ensured that the optimization model can be applied to the complex and changeable transformer detection environment, and the accuracy, reliability and timeliness of the fault detection of the transformer to be tested can be improved.

[0007] Preferably, the S1 comprises the following sub-steps: S11, collecting near-infrared spectrum data sets of insulating oil in different operating states of transformers, and calculating the spectral average value of near-infrared spectrum samples of insulating oil of different categories at all spectral wavelengths; S12, calculating the spectral standard deviation of each insulating oil near-infrared spectrum sample based on the spectral average value; S13. Using a standard normal transformation function to calculate the spectrum average and the spectrum standard deviation, so as to perform a standard normal transformation on the insulating oil near-infrared spectrum data set to obtain the first spectrum signal.

[0008] In this scheme, by calculating the mean and standard deviation of each type of insulating oil spectrum sample at all wavelengths and then performing a standard normal transformation, it helps to eliminate the mean and variance differences between different spectral data, so that the spectral data of different insulating oil states have a unified scale and comparability, thereby improving the quality and stability of the data.

[0009] Preferably, S2 comprises the following sub-steps: S21, establishing a first spectral signal matrix based on the first spectral signal; S22, based on the first spectral signal matrix, using a linear discriminant analysis algorithm to calculate the covariance matrix of each type of insulating oil sample according to different types of insulating oil near-infrared spectral data sets; S23, establishing a divergence matrix of different types of insulating oils based on the first spectral signal matrix and the covariance matrix, and performing a low-dimensional space projection on the first spectral signal matrix according to the divergence matrix to obtain a second spectral signal matrix with reduced dimension; S24. Extracting near-infrared spectral signal features of different types of insulating oils based on the second spectral signal matrix, and using them as the spectral features.

[0010] Preferably, the S23 includes the following sub-steps: S231, performing weighted averaging on the covariance matrix of each type of insulating oil sample to obtain intra-class scatter matrices of different types of insulating oils; S232, calculating the mean vector and the global mean vector of each type of insulating oil based on the first spectral signal matrix, and calculating the vector difference between the mean vector and the global mean vector, and constructing the inter-class scatter matrix of different types of insulating oil according to the vector difference; S233, multiplying the intra-class scatter matrix by the generalized inverse of the inter-class scatter matrix to obtain a generalized feature matrix, performing eigenvalue decomposition on the generalized feature matrix, selecting the eigenvector of the maximum eigenvalue as the low-dimensional space projection direction, and constructing a projection matrix; S234. Project the first spectral signal matrix based on the projection matrix to obtain a second spectral signal matrix after dimensionality reduction.

[0011] In this scheme, by performing dimensionality reduction processing on the first spectral signal, redundant information in high-dimensional data can be removed, and only important classification information in the original spectral signal can be retained, which significantly reduces the amount of calculation for subsequent feature extraction and improves calculation efficiency. Among them, the use of linear discriminant analysis can find the best projection direction, so that spectral signals of different categories are separated as much as possible in the projected space, making the distinction between different categories of spectral signals greater, which helps to improve the classification accuracy and generalization ability of the transformer fault diagnosis model for insulating oil types.

[0012] Preferably, S3 includes the following sub-steps: S31, setting corresponding category labels based on different types of insulating oil, configuring learning parameters of a recurrent neural network algorithm in combination with the dimension of the second spectral signal matrix to perform model training, and obtaining a preliminary transformer fault diagnosis model; S32, mapping the spectral features to the probability distribution of the category labels through the output layer of the preliminary transformer fault diagnosis model, and obtaining weight information of the output layer of the preliminary transformer fault diagnosis model; S33. Establishing an optimization target of the preliminary transformer fault diagnosis model based on the weight information, performing weight optimization in response to the optimization target, and constructing a transformer fault diagnosis model.

[0013] Preferably, before executing S32, the following sub-steps are included: Based on the activation function, the linear output of the preliminary transformer fault diagnosis model for the spectral features is converted into a probability distribution, and a mapping relationship between the spectral features and the category labels is established.

[0014] Preferably, the S33 includes the following sub-steps: Using the cross entropy loss of the preliminary transformer fault diagnosis model as a loss function through a coati optimization algorithm; The weight information is adjusted with minimizing the loss function as the optimization goal, the output layer weight of the preliminary transformer fault diagnosis model is optimized, and the transformer fault diagnosis model is established.

[0015] Preferably, S4 includes the following sub-steps: S41, using the near infrared spectrum data set to be tested as the input of the transformer fault diagnosis model, and outputting the probability value of the category label corresponding to each spectral feature in the near infrared spectrum data set to be tested; S42, sorting the probability values, retrieving the category label corresponding to the largest probability value and taking it as the target type label, and taking the insulating oil type matched by the target type label as the insulating oil type of the transformer to be tested; S43, comparing the insulating oil type of the transformer to be tested with a standard insulating oil status database to determine the fault status of the transformer to be tested, and completing the fault diagnosis of the transformer to be tested.

[0016] In this scheme, the near-infrared spectral data set of the transformer to be tested is detected through the transformer fault diagnosis model. The model will output the probability of each insulating oil type, select the insulating oil type with the largest probability as the category of the transformer to be tested, and then compare it with the standard insulating oil status database, that is, combining the correspondence between the category and the health status of the transformer, evaluate the current status of the transformer, and make decisions on maintenance or replacement of the insulating oil, so as to achieve effective monitoring and fault diagnosis of the transformer insulating oil status, and provide strong support for the safe and stable operation of the power system.

[0017] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, steps of a transformer fault diagnosis method based on near infrared spectroscopy are implemented.

[0018] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a transformer fault diagnosis method based on near infrared spectroscopy.

[0019] Beneficial effects of the present invention: 1. Reduce the dimension of the first spectral signal, so that feature optimization and feature extraction reduce the data dimension, improve the efficiency of feature extraction, and remove those redundant and useless information for classification tasks, providing high-quality data support for strengthening the learning ability of the transformer fault diagnosis model; 2. By constructing an optimization model based on the extracted spectral features and the corresponding insulating oil category labels, the model can comprehensively learn the spectral features of different insulating oil states, ensuring that the insulating oil type of the transformer under test can be quickly and accurately determined during fault diagnosis of the transformer under test, so as to efficiently complete the detection of the transformer fault state; at the same time, ensure that the optimization model can be applied to complex and changeable transformer detection environments, and improve the accuracy, reliability and timeliness of fault detection of the transformer under test. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Other features, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings. The drawings are only for the purpose of illustrating preferred embodiments and are not to be considered as limiting the present invention. Also, the same reference symbols are used throughout the drawings to represent the same parts.

[0021] Figure 1 This is a flow chart of a transformer fault diagnosis method based on near infrared spectroscopy in this embodiment.

[0022] Figure 2 Schematic diagram of a standard normal transformation process of near-infrared spectral data in this embodiment.

[0023] Figure 3 The figure is a schematic diagram of a process of constructing a transformer fault diagnosis model according to the present embodiment.

[0024] Figure 4 This is a diagram of an electronic device architecture of this embodiment. DETAILED DESCRIPTION

[0025] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] Example 1: Figure 1 As shown, a transformer fault diagnosis method based on near infrared spectroscopy includes steps S1-S4, wherein: S1. Acquire near-infrared spectrum information of the transformer and perform standard normal transformation on it to obtain a first spectrum signal.

[0027] Specifically, S1 includes the following sub-steps: S11, collecting near-infrared spectrum data sets of insulating oil in different operating states of transformers, and calculating the spectral average value of near-infrared spectrum samples of insulating oil of different categories at all spectral wavelengths; S12, calculating the spectral standard deviation of each insulating oil near-infrared spectrum sample based on the spectral average value; S13. Using a standard normal transformation function to calculate the spectral average and the spectral standard deviation, so as to perform a standard normal transformation on the insulating oil near-infrared spectrum data set to obtain a first spectral signal.

[0028] Specifically, the insulating oil collected in different operating states of the transformer includes at least: the insulating oil when the transformer is operating normally, the insulating oil when the transformer has a short circuit fault, the insulating oil when the transformer insulation is damp, and the insulating oil when the transformer has a thermal fault, to ensure that the transformer fault diagnosis model comprehensively learns the spectral characteristics of the insulating oil under various operating states of the transformer.

[0029] Specifically, the calculation formula for the spectral average is as follows: In the formula, n represents the number of wavelength points, represents the average spectrum of the i-th insulating oil near-infrared spectrum sample at all wavelengths, x ij represents the original spectrum value of the i-th insulating oil near-infrared spectrum sample at the j-th wavelength; The calculation formula for the spectral standard deviation is as follows: In the formula, s i represents the spectral standard deviation of the i-th insulating oil near-infrared spectrum sample; The standard normal transformation calculation formula is as follows: In the formula, X snv is the standard normal transformation form of the insulating oil near-infrared spectrum data set.

[0030] In this embodiment, by calculating the mean value and standard deviation of each type of insulating oil spectrum sample at all wavelengths, and then performing a standard normal transformation, it is helpful to eliminate the mean and variance differences between different spectral data, so that the spectral data of different insulating oil states have a unified scale and comparability, and improve the quality and stability of the data. Among them, the standard deviation of the insulating oil spectrum sample reflects the degree of discreteness of the spectral data. After further converting the spectrum number of each insulating oil spectrum sample into a standard normal form, the spectral difference caused by factors such as sample particle size, scattering, and optical path change is eliminated, making the spectral data between different insulating oil states more comparable.

[0031] S2. Perform dimensionality reduction calculation on the first spectral signal to obtain spectral features corresponding to different operating states of the transformer.

[0032] Specifically, S2 includes the following sub-steps: S21, establishing a first spectral signal matrix based on the first spectral signal; S22, based on the first spectral signal matrix, using a linear discriminant analysis algorithm to calculate the covariance matrix of each type of insulating oil sample according to different types of insulating oil near-infrared spectral data sets; S23, establishing a divergence matrix of different types of insulating oils based on the first spectral signal matrix and the covariance matrix, and performing a low-dimensional space projection on the first spectral signal matrix according to the divergence matrix to obtain a second spectral signal matrix with reduced dimension; S24. Extract near-infrared spectral signal features of different types of insulating oils based on the second spectral signal matrix and use them as spectral features.

[0033] Specifically, S23 includes the following sub-steps: S231, weighted averaging the covariance matrix of each type of insulating oil sample to obtain the intra-class scatter matrix of different types of insulating oil; S232, calculating the mean vector and the global mean vector of each type of insulating oil based on the first spectral signal matrix, and calculating the vector difference between the mean vector and the global mean vector, and constructing the inter-class scatter matrix of different types of insulating oil according to the vector difference; S233, multiplying the generalized inverse of the intra-class scatter matrix and the inter-class scatter matrix to obtain a generalized feature matrix, performing eigenvalue decomposition on the generalized feature matrix, selecting the eigenvector of the maximum eigenvalue as the low-dimensional space projection direction, and constructing a projection matrix; S234. Project the first spectral signal matrix based on the projection matrix to obtain a second spectral signal matrix after dimensionality reduction.

[0034] As an implementation method, Figure 2 As shown in the figure, the first spectral signal obtained by the standard normal transformation is organized into a matrix form, the first spectral signal is separated by category, and the covariance matrix of each category is calculated; then, the covariance matrices of all categories are weighted averaged (usually the weight is the proportion of the number of samples of each category to the total number of samples) to obtain the intra-class divergence matrix B; for example, there are N insulating oil spectral samples, each sample contains M features (i.e., the wavelength points of the spectrum), then the dimension of the first spectral signal matrix A is N×M; if there are C insulating oil categories in the first spectral signal, the cth category sample set is A c , then the intra-class scatter matrix B can be expressed as: Where N crepresents the number of samples in the cth class, Cov(A c ) represents the covariance matrix of the c-th class sample; After calculating the intra-class scatter matrix B, calculate the mean vector of each category and the global mean vector, then calculate the difference between the mean vector of each category and the global mean vector, and use these differences to construct the inter-class scatter matrix F, which is expressed as follows: In the formula, μ c represents the mean vector of the cth class, μ represents the global mean vector; Solve the eigenvalues ​​and eigenvectors of the product of the generalized inverse (such as Moore-Penrose pseudoinverse) of the inter-class scatter matrix F and the intra-class scatter matrix B, select the eigenvector corresponding to the maximum eigenvalue as the projection direction, and construct the projection matrix D; D is expressed as follows: D=F + B(6) Among them, the characteristic polynomial of the projection matrix D is solved to obtain the eigenvalue, and then the corresponding eigenvector is solved according to the obtained eigenvalue; the characteristic polynomial calculation formula is as follows: det(D-λI)=0(7) Where λ represents the eigenvalue, I represents the unit matrix, and its size is the same as D. Then, we solve (D-λI)v=0 to find the corresponding eigenvector v. Finally, the projection matrix D is used to project the first spectral signal matrix A to obtain the second spectral signal matrix E after dimension reduction. E is expressed as follows: E=A·D(8).

[0035] In this embodiment, by performing dimensionality reduction processing on the first spectral signal, redundant information in high-dimensional data can be removed, and only important classification information in the original spectral signal can be retained, which significantly reduces the amount of calculation for subsequent feature extraction and improves calculation efficiency. Among them, the use of linear discriminant analysis can find the best projection direction, so that spectral signals of different categories are separated as much as possible in the projected space, so that the distinction between different categories of spectral signals is greater, which helps to improve the classification accuracy and generalization ability of the transformer fault diagnosis model for insulating oil types.

[0036] S3. Construct a transformer fault diagnosis model based on spectral features and corresponding insulating oil category labels.

[0037] Specifically, S3 includes the following sub-steps: S31, setting corresponding category labels based on different types of insulating oil, configuring learning parameters of the recurrent neural network algorithm in combination with the dimension of the second spectral signal matrix to perform model training, and obtaining a preliminary transformer fault diagnosis model; S32, mapping the spectral features to the probability distribution of the category label through the output layer of the preliminary transformer fault diagnosis model, and obtaining the weight information of the output layer of the preliminary transformer fault diagnosis model; S33. Establish an optimization target for the preliminary transformer fault diagnosis model based on the weight information, perform weight optimization in response to the optimization target, and construct a transformer fault diagnosis model.

[0038] Specifically, before executing S32, the following sub-steps are included: Based on the activation function, the linear output of the preliminary transformer fault diagnosis model for spectral features is converted into probability distribution, and the mapping relationship between spectral features and category labels is established.

[0039] Specifically, S33 includes the following sub-steps: The cross entropy loss of the preliminary transformer fault diagnosis model is used as the loss function through the coati optimization algorithm; The weight information is adjusted with the minimization of the loss function as the optimization objective, the output layer weights of the preliminary transformer fault diagnosis model are optimized, and the transformer fault diagnosis model is established.

[0040] As an implementation method, Figure 3 As shown in Figure 2, the specific steps for establishing a transformer fault diagnosis model are as follows: A1, the second spectrum signal matrix E∈R after dimension reduction N×E′ and the corresponding insulating oil category label t∈{t1,t2,...,C} N , where E′ represents the feature dimension after dimensionality reduction; A2. Adopt a variant of the recurrent neural network: the gated recurrent unit (GRU) network, set the input layer of the GRU network to E′ dimension, and the output layer to C dimension (corresponding to the probabilities of C categories); and define the reset gate and update gate of the GRU network to construct a preliminary transformer fault diagnosis model; A3. Through the output layer, the hidden state h T Mapping to the probability distribution of C categories, converting the linear output of the preliminary transformer fault diagnosis model to the spectral features into a probability distribution, establishing the mapping relationship between the spectral features and the category labels, and determining the weight and bias of the output layer; A4. Based on the Coyote Optimization Algorithm (COA) algorithm, the cross entropy of the GRU network is used as the loss function. The learning parameters, weights and biases in the GRU network are iteratively adjusted with the goal of minimizing the loss function to construct a transformer fault diagnosis model.

[0041] Furthermore, in A2: the reset gate and the update gate are expressed as follows: Reset the gate: 1xfz r =σ(W r x t +U r h t-1 +b r )(9) Among them, σ is the activation function, x t is the input of the tth time step (the E′-dimensional feature of each sample can be regarded as the input of one time step. In addition, if the data has a time series nature, it is processed according to the actual time step), h t-1 represents the hidden state of the previous time step, W r , U r With b r They represent learning parameters respectively; Update Gate: 1xfz t =σ(W z x t +U z h t-1 +b z )(10) Where W z , U z With b z They represent the learning parameters respectively.

[0042] Furthermore, in A3, the hidden state h T The mapping to the category probability distribution is expressed as follows: mathbfp=softmax(W ο h T +b ο )(11) Where W ο and b ο They represent the weight and bias of the output layer respectively. Softmax is a probability distribution function, which is used to convert the linear output of the preliminary transformer fault diagnosis model for spectral features into a probability distribution.

[0043] In this embodiment, the GRU network, as a variant of the recurrent neural network, can effectively capture and process the dependency relationship in the near-infrared spectral data sequence of the transformer insulating oil by introducing an update gate and a reset gate (i.e., the probability distribution relationship between the spectral features and the category labels that can be used to process). Compared with another variant of the recurrent neural network, such as the long short-term memory network, the GRU network is more concise in structure and has fewer parameters, so that it has higher computational efficiency while maintaining high performance; the COA algorithm has efficient traversal and randomness, can perform efficient global search in the solution space, and has relatively low requirements for initial parameters and search space, so as to avoid falling into the local optimal solution, that is, the COA algorithm is used to find the minimum loss function to adjust the learning parameters, weights and biases of the GRU network, so that the preparatory transformer fault diagnosis model can accurately and fully learn the spectral features of different types of insulating oils, so that the model has the probability distribution of identifying spectral features and corresponding category labels, which significantly improves the accuracy and generalization ability of the transformer fault diagnosis model. In addition, by optimizing the parameters of the GRU model through COA, the model structure can be customized to adapt to different diagnostic needs according to the specific transformer fault type and spectral data characteristics, and the adaptability of the optimization model is significantly improved.

[0044] S4. Based on the transformer fault diagnosis model, the near-infrared spectral data set of the transformer to be tested is detected, the probability of each insulating oil type is output, and the insulating oil type of the transformer to be tested is determined according to the maximum probability. Whether the transformer to be tested is faulty is determined by the insulating oil type, and the transformer fault detection is completed.

[0045] Specifically, S4 includes the following sub-steps: S41, using the near infrared spectrum data set to be tested as the input of the transformer fault diagnosis model, and outputting the probability value of the category label corresponding to each spectral feature in the near infrared spectrum data set to be tested; S42, sorting the probability values, retrieving the category label corresponding to the largest probability value and taking it as the target type label, and taking the insulating oil type matching the target type label as the insulating oil type of the transformer to be tested; S43, comparing the insulating oil type of the transformer to be tested with the standard insulating oil status database to determine the fault status of the transformer to be tested, and completing the fault diagnosis of the transformer to be tested.

[0046] Beneficial effects of the embodiment: by performing a standard normal transformation on the collected near-infrared spectral information of the transformer, the noise signal in the fluorescence spectral data can be reduced, and the spectral differences caused by factors such as sample particle size, scattering and optical path changes can be eliminated, so that the spectral characteristics between different insulating oil states are more comparable, so as to facilitate the extraction of spectral characteristics; the first spectral signal is reduced in dimension, so that feature optimization and feature extraction reduce the data dimension, improve the feature extraction efficiency, and remove those redundant and useless information for the classification task, so as to provide high-quality data support for strengthening the learning ability of the transformer fault diagnosis model; by constructing an optimization model with the extracted spectral features and the corresponding insulating oil category labels, the model can comprehensively learn the spectral characteristics of different insulating oil states, ensuring that the insulating oil type of the transformer to be tested can be quickly and accurately determined when diagnosing the fault of the transformer to be tested, so as to efficiently complete the detection of the transformer fault state; at the same time, it is ensured that the optimization model can be applied to the complex and changeable transformer detection environment, and the accuracy, reliability and timeliness of the fault detection of the transformer to be tested are improved. The near-infrared spectral data set of the transformer to be tested is detected through the transformer fault diagnosis model. The model will output the probability of each insulating oil type, select the insulating oil type with the highest probability as the category of the transformer insulating oil to be tested, and then compare it with the standard insulating oil status database, that is, combining the correspondence between the category and the health status of the transformer, evaluate the current status of the transformer, and make decisions on maintenance or replacement of the insulating oil, so as to achieve effective monitoring and fault diagnosis of the transformer insulating oil status, and provide strong support for the safe and stable operation of the power system.

[0047] An electronic device, such as Figure 4 As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a transformer fault diagnosis method based on near infrared spectroscopy is implemented.

[0048] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, a transformer fault diagnosis method based on near infrared spectroscopy is implemented.

[0049] In the embodiments provided in the present application, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the embodiments of the structure described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another structure, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, structures or units, which can be electrical, mechanical or other forms.

[0050] The units described as separate components may or may not be physically separated, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0051] In addition, each functional unit in the embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0052] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor (processor) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0053] Through the description of the above implementation methods, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the specific device can be divided into different functional modules to complete all or part of the functions described above.

[0054] In the embodiments provided in the present application, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the embodiments of the structure described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another structure, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, structures or units, which can be electrical, mechanical or other forms.

[0055] The units described as separate components may or may not be physically separated, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0056] In addition, each functional unit in the embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0057] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor (processor) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0058] The above specific embodiments are preferred embodiments of the present invention, and are not intended to limit the specific implementation scope of the present invention. The scope of the present invention includes but is not limited to the present specific embodiments. All equivalent changes made in accordance with the shape, structure, and method of the present invention are within the protection scope of the present invention.

Claims

1. A transformer fault diagnosis method based on near infrared spectroscopy, characterized in that: S1, obtaining near-infrared spectrum information of the transformer and performing standard normal transformation on it to obtain a first spectrum signal; S2. Perform dimensionality reduction calculation on the first spectral signal to obtain spectral features corresponding to different operating states of the transformer; S3, constructing a transformer fault diagnosis optimization model based on the spectral characteristics and the corresponding insulating oil category labels; S4. Based on the transformer fault diagnosis optimization model, the near-infrared spectrum data set of the transformer to be tested is detected, the probability of each insulating oil type is output, and the insulating oil type of the transformer to be tested is determined according to the maximum probability value, and whether the transformer to be tested is faulty is judged by the insulating oil type, thereby completing the transformer fault detection.

2. The transformer fault diagnosis method based on near infrared spectroscopy according to claim 1 is characterized in that: The S1 comprises the following sub-steps: S11, collecting near-infrared spectrum data sets of insulating oil in different operating states of transformers, and calculating the spectral average value of near-infrared spectrum samples of insulating oil of different categories at all spectral wavelengths; S12, calculating the spectral standard deviation of each insulating oil near-infrared spectrum sample based on the spectral average value; S13. Using a standard normal transformation function to calculate the spectrum average and the spectrum standard deviation, so as to perform a standard normal transformation on the insulating oil near-infrared spectrum data set to obtain the first spectrum signal.

3. The transformer fault diagnosis method based on near infrared spectroscopy according to claim 1 is characterized in that: The S2 comprises the following sub-steps: S21, establishing a first spectral signal matrix based on the first spectral signal; S22, based on the first spectral signal matrix, using a linear discriminant analysis algorithm to calculate the covariance matrix of each type of insulating oil sample according to different types of insulating oil near-infrared spectral data sets; S23, establishing a divergence matrix of different types of insulating oils based on the first spectral signal matrix and the covariance matrix, and performing a low-dimensional space projection on the first spectral signal matrix according to the divergence matrix to obtain a second spectral signal matrix with reduced dimension; S24. Extracting near-infrared spectral signal features of different types of insulating oils based on the second spectral signal matrix, and using them as the spectral features.

4. The transformer fault diagnosis method based on near infrared spectroscopy according to claim 3 is characterized in that: The S23 comprises the following sub-steps: S231, performing weighted averaging on the covariance matrix of each type of insulating oil sample to obtain intra-class scatter matrices of different types of insulating oils; S232, calculating the mean vector and the global mean vector of each type of insulating oil based on the first spectral signal matrix, and calculating the vector difference between the mean vector and the global mean vector, and constructing the inter-class scatter matrix of different types of insulating oil according to the vector difference; S233, multiplying the intra-class scatter matrix by the generalized inverse of the inter-class scatter matrix to obtain a generalized feature matrix, performing eigenvalue decomposition on the generalized feature matrix, selecting the eigenvector of the maximum eigenvalue as the low-dimensional space projection direction, and constructing a projection matrix; S234. Project the first spectral signal matrix based on the projection matrix to obtain a second spectral signal matrix after dimensionality reduction.

5. The transformer fault diagnosis method based on near infrared spectroscopy according to claim 3 is characterized in that: The S3 comprises the following sub-steps: S31, setting corresponding category labels based on different types of insulating oil, configuring learning parameters of a recurrent neural network algorithm in combination with the dimension of the second spectral signal matrix to perform model training, and obtaining a preliminary transformer fault diagnosis model; S32, mapping the spectral features to the probability distribution of the category labels through the output layer of the preliminary transformer fault diagnosis model, and obtaining weight information of the output layer of the preliminary transformer fault diagnosis model; S33. Establishing an optimization target of the preliminary transformer fault diagnosis model based on the weight information, performing weight optimization in response to the optimization target, and constructing a transformer fault diagnosis optimization model.

6. The transformer fault diagnosis method based on near infrared spectroscopy according to claim 5 is characterized in that: Before executing S32, the following sub-steps are included: Based on the activation function, the linear output of the preliminary transformer fault diagnosis model for the spectral features is converted into a probability distribution, and a mapping relationship between the spectral features and the category labels is established.

7. The transformer fault diagnosis method based on near infrared spectroscopy according to claim 5 is characterized in that: The S33 comprises the following sub-steps: Using the cross entropy loss of the preliminary transformer fault diagnosis model as a loss function through a coati optimization algorithm; The weight information is adjusted with minimizing the loss function as the optimization goal, the output layer weight of the preliminary transformer fault diagnosis model is optimized, and a transformer fault diagnosis optimization model is established.

8. The transformer fault diagnosis method based on near infrared spectroscopy according to claim 1 is characterized in that: The S4 comprises the following sub-steps: S41, using the near-infrared spectral data set to be tested as the input of the transformer fault diagnosis optimization model, and outputting the probability value of the category label corresponding to each spectral feature in the near-infrared spectral data set to be tested; S42, sorting the probability values, retrieving the category label corresponding to the largest probability value and taking it as the target type label, and taking the insulating oil type matched by the target type label as the insulating oil type of the transformer to be tested; S43, comparing the insulating oil type of the transformer to be tested with a standard insulating oil status database to determine the fault status of the transformer to be tested, and completing the fault diagnosis of the transformer to be tested.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the transformer fault diagnosis method based on near infrared spectroscopy as claimed in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the transformer fault diagnosis method based on near infrared spectroscopy as claimed in any one of claims 1 to 8 are implemented.

Images