Dry-type bushing fault diagnosis method and system

Abstract

The invention discloses a dry-type bushing fault diagnosis method and system, and the method comprises the steps: obtaining a partial discharge signal and a surface vibration signal of a to-be-diagnosed dry-type bushing in an operation process during data collection, carrying out the feature analysis of data through combining the capability of the wavelet analysis for finely depicting signal features in a time-frequency domain, obtaining the feature quantities in different operation states, and carrying out the fault diagnosis of the dry-type bushing. According to the method, the characteristics of non-stationary and transient signals are better processed, a normal state and various fault states can be more accurately distinguished, all possible fault types can be comprehensively covered, weak fault characteristic information is separated and extracted from the signals, and the detection sensitivity of latent and early faults is improved in combination with a classification model.

Description

Technical Field

[0001] The invention belongs to the technical field of power systems and relates to a dry-type bushing fault diagnosis method and system. Background Art

[0002] As a critical insulation device in power systems, the operational status of dry-type bushings is directly linked to the safety and stability of the power grid. Due to long-term exposure to outdoor environments and high-voltage electric fields, dry-type bushings are prone to insulation defects such as partial discharge, aging, and cracks. These defects can cause equipment failures and even power accidents. Therefore, accurately diagnosing the insulation status of dry-type bushings and promptly identifying potential faults are crucial to ensuring the safe operation of power systems.

[0003] Traditional dry bushing fault diagnosis methods primarily fall into two categories: offline testing and online monitoring. Offline testing typically requires a power outage, which not only impacts power supply reliability but also prevents real-time monitoring of equipment status changes. Online monitoring, on the other hand, allows for continuous monitoring of bushing status without power outages. However, existing online monitoring technologies have limitations in practical applications. For example, relative capacitance and dielectric loss tests can reflect the overall insulation performance of the bushing but are not very sensitive to local defects. While end-screen voltage monitoring can capture certain types of partial discharge signals, its detection range is limited and cannot fully cover all possible fault types.

[0004] Specifically, the following deficiencies exist:

[0005] First, offline detection has limitations: it requires equipment to be powered off, which affects power supply reliability and cannot reflect the evolution of operating status in real time.

[0006] Second, the shortcomings of online monitoring methods:

[0007] (1) Infrared thermal imaging: It mainly targets thermal faults and is not sensitive to non-thermal early insulation defects (such as weak internal discharge).

[0008] (2) Partial discharge detection: Partial discharge signals usually have weak energy and complex waveforms, and are easily affected by strong electromagnetic interference on site. Traditional signal processing and analysis methods are difficult to effectively extract reliable fault feature information.

[0009] (3) One-sidedness of a single monitoring method: Relying on only one monitoring method can often only reflect one aspect of the fault, making it difficult to fully and accurately judge the equipment status, and there is a risk of missed judgment and misjudgment.

[0010] (4) Low intelligence and diagnostic accuracy: Many existing diagnostic methods rely on the experience of operation and maintenance personnel or simple threshold judgment logic. They have limited ability to identify complex fault modes, multiple concurrent faults, or early weak fault characteristics. The accuracy and sensitivity of diagnosis need to be improved.

[0011] Third, existing technologies fail to effectively combine advanced signal processing with intelligent algorithms: they fail to fully utilize the advantages of time-frequency analysis tools such as wavelet analysis in processing non-stationary and transient signals, and they fail to effectively apply the powerful capabilities of machine learning algorithms such as support vector machines in high-dimensional and nonlinear classification problems to improve the intelligence and accuracy of diagnosis. Summary of the Invention

[0012] The purpose of the present invention is to solve the problems in the prior art of low sensitivity to local defects, limited detection range, inability to fully cover all possible fault types, failure of existing classification methods to handle the advantages of non-stationary and transient signals, and low accuracy of classification results, and to provide a dry casing fault diagnosis method and system.

[0013] In order to achieve the above object, the present invention adopts the following technical solutions:

[0014] A dry-type bushing fault diagnosis method comprises the following steps:

[0015] Obtain the partial discharge signal and surface vibration signal of the dry-type bushing to be diagnosed during operation;

[0016] Perform multi-scale time-frequency analysis on the partial discharge signal and surface vibration signal respectively to obtain feature quantities under different operating conditions. All feature quantities are fused to obtain a fused feature vector.

[0017] Obtain a classification model, use the fused feature vector as input to the classification model, and obtain the fault diagnosis result.

[0018] A further improvement of the present invention is:

[0019] The multi-scale time-frequency analysis of the partial discharge signal and the surface vibration signal further includes:

[0020] Perform wavelet denoising and standardization on the partial discharge signal to obtain the denoised partial discharge PD signal x ′ {pd}[n] ;

[0021] Filter and denoise the surface vibration signal to obtain the denoised surface vibration signal x ′ {vib,f}[n] ;

[0022] For partial discharge PD signal x v {pd}[n] and surface vibration signal x′ {vib,f}[n] Perform normalization processing to obtain the processed partial discharge signal x {pd,norn}[n] and surface vibration signal x {vib,norm}[n] .

[0023] Extract the characteristics of partial discharge signals, including:

[0024] For partial discharge signal x {pd,norm}[n] Perform wavelet decomposition;

[0025] The detail coefficients of each layer after decomposition are calculated, wherein the detail coefficients include the energy, kurtosis and information entropy of the partial discharge signal of each layer.

[0026] The feature extraction of the surface vibration signal includes:

[0027] Calculate the surface vibration signal x {vib,norm}[n] The detail coefficients of each layer include energy proportion and standard deviation.

[0028] The fused feature vector is expressed by the following formula:

[0029] X=

[0030] [E {pd,1} ,…,E {pd,6} ,H {pd,1} ,…,H {pd,6} ,

[0031] H {pd,1} ,…,H {pd,6} ,p {vib,1} ,…,p {vib,8} ,σ {vib,1} ,…,σ {vib,8} ]

[0032] Among them, E {pd,1} Indicates the energy of the partial discharge signal; K {pd,1} Indicates the kurtosis of the partial discharge signal; H {pd,1} represents the information entropy of the partial discharge signal; p {vib,1} Indicates the energy ratio of surface vibration signal; σ {vib,8} ] represents the standard deviation of the surface vibration signal.

[0033] The classification model includes M*(M-1) / 2 binary classification SVMs {ij} Sub-model, where i and j represent the predicted fault type and M represents the number of categories.

[0034] The obtaining of the fault diagnosis result includes:

[0035] Input the fused feature vector into M*(M-1) / 2 binary classification SVMs{ij} In the sub-model, all prediction results are obtained and voted on. The category with the most predictions is the diagnosed fault category.

[0036] A dry-type casing fault diagnosis system, comprising:

[0037] A signal acquisition module is used to acquire partial discharge signals and surface vibration signals of the dry-type bushing to be diagnosed during operation;

[0038] The feature vector extraction module is used to perform multi-scale time-frequency analysis on the partial discharge signal and the surface vibration signal, obtain feature quantities under different operating conditions, and fuse all feature quantities to obtain a fused feature vector;

[0039] The fault diagnosis module is used to obtain a classification model, use the fused feature vector as the input of the classification model, and obtain the fault diagnosis result.

[0040] A terminal device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0041] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of any method described in the present invention.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The present invention discloses a dry-type bushing fault diagnosis method. During data acquisition, partial discharge signals and surface vibration signals of the dry-type bushing to be diagnosed during operation are obtained. The ability of wavelet analysis to finely characterize signal characteristics in the time and frequency domains is combined to perform feature analysis on the data, obtain feature quantities under different operating states, better process the characteristics of non-stationary and transient signals, more accurately distinguish between normal states and various fault states, comprehensively cover all possible fault types, separate and extract weak fault feature information from the signal, and combine with a classification model to improve the detection sensitivity of latent and early faults.

[0044] Furthermore, in the present invention, the wavelet denoising in the signal preprocessing can effectively suppress the on-site environmental noise and interference, and improve the accuracy of the subsequent classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0048] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0049] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0050] In the description of the embodiments of the present invention, it should be noted that if the terms "upper," "lower," "horizontal," "inner," etc. appear, the orientation or positional relationship indicated is based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the inventive product is typically placed when in use. These terms are merely for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or component referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. In addition, the terms "first," "second," etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0051] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0052] In the description of the embodiments of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0053] The present invention is described in further detail below with reference to the accompanying drawings:

[0054] See also Figure 1 The embodiment of the present invention discloses a dry casing fault diagnosis method, which includes the specific type and parameters of signal acquisition, the specific method of preprocessing and wavelet basis selection, the specific wavelet basis of feature extraction, the number of decomposition layers and the type of feature quantity, the kernel function selection of the SVM model, the parameter optimization method, the multi-classification strategy and the final diagnosis and alarm process.

[0055] The present invention combines the signal feature extraction capability of wavelet analysis with the intelligent classification capability of support vector machine to achieve accurate and automatic diagnosis of dry-type bushing faults. The specific technical solution includes the following steps:

[0056] Step 1: Condition monitoring signal acquisition

[0057] Obtain the partial discharge (PD) signal and equipment surface vibration signal of the dry-type bushing to be diagnosed during operation.

[0058] Among them, the signal acquisition system uses an ultra-high frequency (UHF) sensor to collect PD signals; a piezoelectric accelerometer is used to collect vibration signals to ensure sufficient sampling rate (PD signal is not less than 1GS / s, vibration signal is not less than 50kHz) and accuracy.

[0059] Assume that the collected discrete time series signal is x[n], where n=1,2,...,N, N is the number of sampling points, and the PD signal is x {pd}[n] , the vibration signal is x {vib}[n] .

[0060] Step 2: Signal preprocessing

[0061] The collected raw PD and vibration signals are preprocessed to eliminate noise interference and standardize them, including:

[0062] Step 2.1: Denoising:

[0063] PD signal wavelet threshold denoising:

[0064] For PD signal x {pd}[n] Perform L=5 layers of db4 wavelet decomposition to obtain the detail coefficients cD of each layer {pd,j} (j=1...5) and approximation coefficient cA {pd,5} .

[0065] Estimated noise standard deviation σ noise (For example, use the median absolute deviation MAD estimate of the first layer detail coefficients: σ noise ≈MAD(cD {pd,1} ) / 0.6745;

[0066] Calculate the threshold T j (For example, a common threshold where N j is the length of the j-th layer coefficients).

[0067] For detail coefficient d = cD {pd,j,k} (representing the kth specific coefficient value in the jth layer detail coefficient set) is processed using a soft threshold function:

[0068] d′=η{soft}{d,T j}=sgn(d)*max(0,|d|-T j )

[0069] Among them, sgn() is the sign function and max() takes the larger value.

[0070] Using the processed coefficient d′ and the unprocessed approximate coefficient cA {pd,5} Perform wavelet reconstruction to obtain the denoised PD signal x ′ {pd}[n] .

[0071] Vibration signal filtering and denoising:

[0072] For vibration signal x {vib}[n] Apply a 50Hz notch filter (or band-stop filter) to remove the power frequency interference and obtain the denoised x ′ {vib,f}[n] .

[0073] x ′ {vib,f}[n] The sym6 wavelet base is used to perform similar wavelet threshold denoising (the appropriate number of layers and threshold rules can be selected according to the characteristics of the vibration signal) to obtain the denoised vibration signal x ′ {vib}[n] .

[0074] Step 2.2: Perform normalization:

[0075] For the denoised PD signal x ′{pd}[n] and vibration signal x ′ {vib}[n] The maximum-minimum normalization method is used to scale the signal amplitude to the [-1,1] interval:

[0076]

[0077] where x orig is the original signal value, min(x orig ) and max(x orig ) are the minimum and maximum values ​​of the original signal, respectively, x norm is the normalized signal value. The processed signal is recorded as x {pd,norm}[n] and x {vin,norm}[n] .

[0078] Step 3: Feature extraction based on wavelet analysis

[0079] The preprocessed signal is subjected to multi-scale time-frequency analysis using wavelet transform to extract characteristic quantities that can effectively distinguish different operating states.

[0080] Select the wavelet basis and the number of decomposition levels:

[0081] x {pd,norm}[n] , using db5 wavelet basis for J {pd} =6-layer decomposition.

[0082] x {vib,norm}[n] , using db5 wavelet basis for J {vib} =8 layers of decomposition.

[0083] Specifically, wavelet decomposition:

[0084] PD signal: x {pd,norm}[n] →{cA {pd,6} ,cD {pd,6} ,cD {pd,5} ,…,cD {pd,1} ,}

[0085] Vibration signal: x {vib,norm}[n] →{cA {vib,8} ,cD {vib,8} ,cD {vib,7} ,…,cD {vib,1} ,}

[0086] Feature calculation: Calculate the feature from the detail coefficients of each layer. Let the detail coefficient sequence of the jth layer be cD j ={d {j,K} |k=1,…,M j}, where M j is the number of coefficients in this layer.

[0087] For each layer detail coefficient of PD signal (cD {pd,j} ,j=1,2,3,4,5,6):

[0088] Energy:

[0089] E {pd,j} =∑ k |d {pd,j,K} | 2

[0090] Kurtosis: Let μ {pd,j} cD {pd,j} The mean of {pd,j} is its standard deviation.

[0091]

[0092] M j is the number of coefficients in this layer. Kurtosis reflects the sharpness of the signal waveform

[0093] Information Entropy: Calculate cD {pd,j} Shannon entropy of the coefficient value distribution. First, we need to estimate the probability distribution p of the coefficient {j,b} (The coefficient amplitude is discretized into intervals, and the frequency falling in each interval is counted to obtain the probability).

[0094] H{pd,j}=-∑ b p{j,b}*log2(p{j,b})

[0095] For each layer detail coefficient of vibration signal (cD {vib,j} ,j=1,2,3,4,5,6,7,8):

[0096] Energy ratio: first calculate the energy of each layer

[0097] E {vib,j} =∑ k |d {vib,j,K} | 2

[0098] Then calculate the total detail energy:

[0099] E {vib,totaldetail}=∑ j E {vib,j}

[0100] Then the energy proportion of the jth layer is:

[0101]

[0102] Standard Deviation:

[0103]

[0104] Where μ{vib,j} is the mean value of the vibration detail coefficient of the jth layer.

[0105] Constructing a feature vector: All calculated feature quantities are combined in a predetermined order to form a feature vector X. Its dimensions are: (6-dimensional PD energy + 6-dimensional PD kurtosis + 6-dimensional PD information entropy) + (8-dimensional vibration energy proportion + 8-dimensional vibration standard deviation) = 18 + 16 = 34 dimensions.

[0106] X=

[0107] [E {pd,1} ,…,E {pd,6} ,K {pd,1} ,…,K {pd,6} ,

[0108] H {pd,1} ,…,H {pd,6} ,p {vib,1} ,…,p {vib,8} ,σ {vib,v1} ,…,σ {vib,8} ]

[0109] Step 4: Build and train the support vector machine (SVM) classifier model

[0110] Using sample data with known states, a SVM model capable of classifying dry casing states is trained, including:

[0111] Prepare training data set: collect L samples to form training set D = {(X i ,Y i |i=1,…,L}, where X i is the 34-dimensional feature vector of the i-th sample, Y i It is the corresponding status label (such as: 0-normal, 1-internal discharge, 2-surface contamination discharge, 3-bad connection, etc.).

[0112] Select SVM kernel function: Use radial basis kernel function (RBF):

[0113]

[0114] where γ>0 is the kernel parameter, ||X i -X j || is the squared Euclidean distance between two eigenvectors.

[0115] Model parameter optimization: Grid search combined with K-fold cross validation (here K = 5) is used to find the optimal hyperparameters (C, γ). For each set of (C, γ) combinations:

[0116] The training set D is randomly divided into K mutually exclusive subsets.

[0117] K-1 subsets are used in turn as training sets to train an SVM model, and the remaining 1 subset is used as a validation set to test the model performance (such as classification accuracy Acc).

[0118] Calculate the average value Acc_{avg}(C,γ) of K validation results.

[0119] Select the set (Cmax,γmax) that maximizes Acc_{avg}(C,γ) as the optimal parameter.

[0120] Model training: Using the optimal parameters (Cmax, γmax) and the entire training sample set D, the final multi-classification SVM model is trained using a one-vs-one strategy. For M categories, this strategy will train M*(M-1) / 2 binary classification SVM sub-models. Each sub-SVM {ij} Used to distinguish category i from category j.

[0121] Step 5: Online fault diagnosis and status judgment

[0122] The signal collected in real time from the dry casing to be diagnosed is processed through step 2 (preprocessing) and step 3 (feature extraction) to obtain its current feature vector X {test} .

[0123] X {test} Input all M*(M-1) / 2 binary classification SVMs trained in step 4 {ij} sub-model.

[0124] Each SVM {ij} Will X {test} Predict whether it belongs to category i or category j.

[0125] Use the "majority voting" method: count the prediction results of all sub-models, X {test} The category predicted the most times is the final diagnosis conclusion y {pred} .

[0126] For a single binary classifier SVM {ij} , its decision function is usually:

[0127]

[0128] Among them SV {ij} is the support vector set of the sub-model, α k is the Lagrange multiplier, y k is the support vector X kThe category label (+1 or -1), K(X k ,X) is the kernel function,b {ij} is the bias.

[0129] According to f {ij} The symbol of (X) determines whether it tends to be category i or j, and then votes

[0130] Step 6: Alarm and output

[0131] According to the diagnostic result y output in step 5 {pred} , if y {pred} Corresponding to a certain fault state (such as internal discharge, surface contamination, etc.), the alarm mechanism (sound, light, text message, etc.) is triggered. At the same time, the diagnosis time and diagnosis result y {pred} , the corresponding eigenvector X {test} The relevant original signal segments are recorded in the database for subsequent analysis and tracing.

[0132] The method disclosed in the present invention has the following advantages:

[0133] High diagnostic accuracy: Combining the ability of wavelet analysis to accurately characterize signal characteristics in the time and frequency domain (especially good at processing non-stationary and transient signals), and the superior performance and good generalization ability of SVM in processing small sample, high-dimensional, and nonlinear classification problems, it can more accurately distinguish between normal and various fault states.

[0134] Strong ability to identify early faults: The multi-scale characteristics of wavelet analysis help to separate and extract weak fault feature information from the signal. Combined with the sensitivity of SVM, it improves the detection sensitivity of latent and early faults.

[0135] Good anti-interference ability and robustness: Wavelet denoising in signal preprocessing can effectively suppress the interference of on-site environmental noise; the SVM algorithm itself is insensitive to the dimension of the feature space (kernel technique) and has good learning effect on small samples, making the diagnostic method more robust to data quality and feature selection.

[0136] High degree of intelligence and automation: The diagnostic process is based on a data-driven machine learning model. Once the model training is completed, automated online monitoring and diagnosis can be achieved, reducing dependence on manual experience and improving diagnostic efficiency and objectivity.

[0137] Easy to integrate multi-source information: This method has a flexible framework and can easily fuse feature information from multiple sensors (such as PD, vibration, temperature, etc.), build a multi-information fusion diagnosis model, and further improve the comprehensiveness and reliability of the diagnosis.

[0138] This embodiment discloses a dry-type bushing fault diagnosis system, including:

[0139] A signal acquisition module is used to acquire partial discharge signals and surface vibration signals of the dry-type bushing to be diagnosed during operation;

[0140] The feature vector extraction module is used to perform multi-scale time-frequency analysis on the partial discharge signal and the surface vibration signal, obtain feature quantities under different operating conditions, and fuse all feature quantities to obtain a fused feature vector;

[0141] The fault diagnosis module is used to obtain a classification model, use the fused feature vector as the input of the classification model, and obtain the fault diagnosis result.

[0142] A schematic diagram of a terminal device provided in one embodiment of the present invention. The terminal device in this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of each of the aforementioned method embodiments are implemented. Alternatively, when the processor executes the computer program, the functions of each module / unit in each of the aforementioned device embodiments are implemented.

[0143] The computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to accomplish the present invention.

[0144] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0145] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0146] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0147] If the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0148] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A dry bushing fault diagnosis method, characterized in that: The following steps are involved: Obtain the partial discharge signal and surface vibration signal of the dry-type bushing to be diagnosed during operation; Perform multi-scale time-frequency analysis on the partial discharge signal and surface vibration signal respectively to obtain feature quantities under different operating conditions. All feature quantities are fused to obtain a fused feature vector. Obtain a classification model, use the fused feature vector as input to the classification model, and obtain the fault diagnosis result.

2. A dry bushing fault diagnosis method according to claim 1, characterized in that: The multi-scale time-frequency analysis of the partial discharge signal and the surface vibration signal further includes: The partial discharge signal is subjected to wavelet denoising and normalization processing to obtain the denoised partial discharge PD signal x′ {pd}[n] ; Filter and denoise the surface vibration signal to obtain the denoised surface vibration signal x′ {vib,f}[n] ; For partial discharge PD signal x′ {pd}[n] and surface vibration signal x′ {vib,f}[n] Perform normalization processing to obtain the processed partial discharge signal x {pd,norm}[n] and surface vibration signal x {vib,norm}[n] .

3. A dry bushing fault diagnosis method according to claim 2, characterized in that: Extract the characteristics of partial discharge signals, including: For partial discharge signal x {pd,norm}[n] Perform wavelet decomposition; The detail coefficients of each layer after decomposition are calculated, wherein the detail coefficients include the energy, kurtosis and information entropy of the partial discharge signal of each layer.

4. A dry bushing fault diagnosis method according to claim 2, characterized in that: The feature extraction of the surface vibration signal includes: Calculate the surface vibration signal x {vib,norm}[n] The detail coefficients of each layer include energy proportion and standard deviation.

5. A dry bushing fault diagnosis method according to claim 4, characterized in that: The fused feature vector is expressed by the following formula: X＝ [E {pd,1} ,…,E {pd,6} ,K {pd,1} ,…,K {pd,6} , H {pd,1} ,…,H {pd,6} ,p {vib,1} ,…,p {vib,8} ,s {vib,v1} ,…,s {vib,8} ] Among them, E {pd,1} Indicates the energy of the partial discharge signal; K {pd,1} Indicates the kurtosis of the partial discharge signal; H {pd,1} represents the information entropy of the partial discharge signal; p {vib,1} Indicates the energy ratio of surface vibration signal; σ {vib,8} ] represents the standard deviation of the surface vibration signal.

6. A dry bushing fault diagnosis method according to claim 1, characterized in that: The classification model includes M*(M-1) / 2 binary classification SVMs {ij} Sub-model, where i and j represent the predicted fault type and M represents the number of categories.

7. A dry bushing fault diagnosis method according to claim 6, characterized in that: The obtaining of the fault diagnosis result includes: Input the fused feature vector into M*(M-1) / 2 binary classification SVMs {ij} In the sub-model, all prediction results are obtained and voted on. The category with the most predictions is the diagnosed fault category.

8. A dry casing fault diagnosis system, characterized in that: include: A signal acquisition module is used to acquire partial discharge signals and surface vibration signals of the dry-type bushing to be diagnosed during operation; The feature vector extraction module is used to perform multi-scale time-frequency analysis on the partial discharge signal and the surface vibration signal, obtain feature quantities under different operating conditions, and fuse all feature quantities to obtain a fused feature vector; The fault diagnosis module is used to obtain a classification model, use the fused feature vector as the input of the classification model, and obtain the fault diagnosis result.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 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 method according to any one of claims 1 to 7 are implemented.

Images