Method and system for carrying out fault diagnosis on power transformer based on gas dissolved in oil

Abstract

The invention discloses a power transformer fault diagnosis method and system based on gas dissolved in oil, and belongs to the field of power transformer fault diagnosis. The method comprises the following steps: acquiring a plurality of one-dimensional structured data samples of the dissolved gas in the power transformer oil; converting the one-dimensional structured data sample into a two-dimensional grayscale image by adopting an improved Grubrum angle field transformation method, and forming an original power transformer fault diagnosis sample set; enhancing a two-dimensional grayscale image in the original power transformer fault diagnosis sample set to obtain an expanded power transformer fault diagnosis sample set, and dividing the expanded power transformer fault diagnosis sample set into a training set and a test set according to a preset proportion; training the target model according to the training set, testing the trained power transformer fault diagnosis model according to the test set, and outputting the state category of the power transformer; according to the method, the problems of data scarcity, complex data potential rule, low fault diagnosis precision and the like of the data of the gas dissolved in the one-dimensional structured oil are solved.

Description

Technical Field

[0001] The invention relates to a method and a system for diagnosing faults of a power transformer based on gas dissolved in oil, and belongs to the field of power transformer fault diagnosis. Background Art

[0002] Power transformers are the core equipment for realizing energy conversion and transmission in power grids. They are characterized by large number of in-service, wide distribution range and long working time. Once a fault occurs, its fault impact and loss will be much higher than other equipment. When faults such as discharge and overheating, medium-low temperature and high temperature overheating, low-energy and high-energy discharge occur inside the power transformer, the insulating oil near the fault point of the transformer will decompose into different amounts of low-molecular gases (including hydrogen (H2), methane (CH4), acetylene (C2H2), ethane (C2H6) and ethylene (C2H4), etc.), and diagnosed into different categories (including normal, discharge and overheating, low-energy and high-energy discharge, medium-low temperature and high temperature overheating). Therefore, fault diagnosis of power transformers is particularly important. At present, in the field of transformer fault diagnosis, one-dimensional structured dissolved gas data in oil has problems such as scarcity of fault sample data and complex potential laws of data. In addition, transformer fault diagnosis mostly adopts traditional methods such as three-ratio method and encoding method, as well as simple machine learning modules such as Back Propagation neural network module and random forest module, which can easily lead to large errors in transformer fault diagnosis and low accuracy. Summary of the invention

[0003] The present invention provides a method and system for fault diagnosis of power transformers based on dissolved gas in oil, which are used for image conversion and image enhancement expansion of power transformer fault sample data, thereby improving data samples and data quality and better applying fault diagnosis modules to perform fault diagnosis work.

[0004] The technical solution of the present invention is:

[0005] According to a first aspect of the present invention, there is provided a method for fault diagnosis of a power transformer based on dissolved gas in oil, comprising the following steps:

[0006] Step 1, obtaining one-dimensional structured data samples of dissolved gas in multiple power transformer oils; the one-dimensional structured data samples are constructed using the content of dissolved gas in the oil;

[0007] Step 2, converting the one-dimensional structured data samples into two-dimensional grayscale images by using the improved Gram angular field transform method to form the original power transformer fault diagnosis sample set;

[0008] Step 3, enhancing the two-dimensional grayscale image in the original power transformer fault diagnosis sample set to obtain an expanded power transformer fault diagnosis sample set;

[0009] Step 4, dividing the expanded power transformer fault diagnosis sample set into a training set and a test set according to a preset ratio;

[0010] Step 5. Train the target model based on the training set to obtain a trained power transformer fault diagnosis model; test the trained power transformer fault diagnosis model based on the test set to output the power transformer status category; the power transformer status category includes normal, discharge and overheating, low-energy discharge, high-energy discharge, medium and low temperature overheating, and high temperature overheating.

[0011] Furthermore, the one-dimensional structured data sample is converted into a two-dimensional grayscale image using an improved Gram angular field transform method, specifically:

[0012] Step 2.1, preprocess the one-dimensional structured data sample, the expression is as follows:

[0013]

[0014] In the formula, g i is the i-th variable in sample g, For pre-processed samples The i-th variable in ; max(g) and min(g) represent the maximum and minimum values ​​of the variable in sample g, respectively;

[0015] Step 2.2, Perform polar coordinate transformation:

[0016]

[0017] Where: N is a constant factor used to adjust the span of polar coordinates, Φ i is the polar angle of point i, r i is the polar diameter of point i;

[0018] Step 2.3. After converting the data to polar coordinates, use GASF and GADF to convert the data into two-dimensional image data, where GASF and GADF are as follows:

[0019] GASF (i,j) =[cos(Φ i +Φ j )];

[0020] GADF (i,j) =[sin(Φ i -Φ j )]

[0021] Where: Φ j represents the jth polar angle;

[0022] Step 2.4: Improve the matrix converted in Step 2.3, denoted as GAF, and the expression is as follows:

[0023]

[0024] Step 2.5, map the GAF matrix to the interval [0,255].

[0025] Furthermore, the two-dimensional grayscale image in the original power transformer fault diagnosis sample set is enhanced, and the enhancement method includes at least one of the following: an image enhancement method based on image brightness, an image enhancement method based on salt and pepper noise; an image enhancement method based on Gaussian noise, and an image enhancement method based on a vector quantization variational autoencoder.

[0026] Furthermore, the target model adopts at least one of the following: VGG19 network model, ResNet50 model.

[0027] According to a second aspect of the present invention, there is provided a system for diagnosing faults of a power transformer based on dissolved gas in oil, comprising:

[0028] An acquisition module is used to acquire one-dimensional structured data samples of dissolved gas in multiple power transformer oils; the one-dimensional structured data samples are constructed using the content of dissolved gas in the oil;

[0029] An image conversion module is used to convert a one-dimensional structured data sample into a two-dimensional grayscale image by using an improved Gram angular field transform method to form an original power transformer fault diagnosis sample set;

[0030] An image enhancement module is used to enhance the two-dimensional grayscale image in the original power transformer fault diagnosis sample set to obtain an expanded power transformer fault diagnosis sample set; and is used to divide the expanded power transformer fault diagnosis sample set into a training set and a test set according to a preset ratio;

[0031] The fault diagnosis module is used to train the target model according to the training set to obtain a trained power transformer fault diagnosis model; test the trained power transformer fault diagnosis model according to the test set, and output the power transformer status category; the power transformer status category includes normal, discharge and overheating, low-energy discharge, high-energy discharge, medium and low temperature overheating, and high temperature overheating.

[0032] The beneficial effects of the present invention are as follows: on the one hand, the present invention provides a method for diagnosing faults of power transformers based on dissolved gas in oil, which can solve the problems of scarcity of one-dimensional structured dissolved gas in oil data, complex underlying data rules and low fault diagnosis accuracy, and greatly improve the work efficiency of staff in diagnosing faults of power transformers; further, a system for diagnosing faults of power transformers based on dissolved gas in oil is provided, which provides a visual software platform for fault diagnosis of power transformers, enables monitoring personnel to understand the operating conditions of power transformers more intuitively, has high practicality, and is worth further promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a flow chart of the method of the present invention;

[0034] Figure 2 is an example of a Gram matrix; where Figure 2 (a) Gram summation angular field, Figure 2 (b) Gram difference angular field;

[0035] Figure 3 It is the fused Gram Corner Field;

[0036] Figure 4 is the grayscale image of the Gram angle field;

[0037] Figure 5 These are examples of images of each category after one-dimensional data conversion;

[0038] Figure 6 is an example of an enhanced normal category image;

[0039] Figure 7 This is the VGG19 model structure diagram;

[0040] Figure 8 This is the ResNet50 model structure diagram;

[0041] Fig. 9 is the residual structure diagram;

[0042] Fig.10 are image samples after improved Gram angular field transform, continuous wavelet transform and short-time Fourier transform respectively;

[0043] Fig.11 is the time (in seconds) required to convert 1000 samples using the three data conversion methods;

[0044] Fig.12 These are the training results of the VGG19 and ResNet50 models on the GAF dataset;

[0045] Fig.13These are the training results of the VGG19 and ResNet50 models on the CWT dataset;

[0046] Fig.14 It is the training result of VGG19 and ResNet50 models on the STFT dataset;

[0047] Fig.15 This is the training result of VGG19 and ResNet50 models on the enhanced GAF ​​dataset.

[0048] Fig.16 It is a flow chart of the control method of the system. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Based on the embodiments 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. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other arbitrarily without conflict.

[0050] Example 1: Figure 1-16 As shown, according to a first aspect of an embodiment of the present invention, a method for diagnosing a fault of a power transformer based on dissolved gas in oil is provided, comprising the following steps:

[0051] Step 1, obtaining one-dimensional structured data samples of dissolved gases in multiple power transformer oils; the one-dimensional structured data samples are constructed using the content of dissolved gases in the oil; the dissolved gases in the oil include hydrogen (H2), methane (CH4), acetylene (C2H2), ethane (C2H6) and ethylene (C2H4);

[0052] Step 2, converting the one-dimensional structured data samples into two-dimensional grayscale images by using the improved Gram angular field transform method to form an original power transformer fault diagnosis sample set;

[0053] Step 3, enhancing the two-dimensional grayscale image in the original power transformer fault diagnosis sample set to obtain an expanded power transformer fault diagnosis sample set;

[0054] Step 4, dividing the expanded power transformer fault diagnosis sample set into a training set and a test set according to a preset ratio;

[0055] Step 5. Train the target model based on the training set to obtain a trained power transformer fault diagnosis model; test the trained power transformer fault diagnosis model based on the test set to output the power transformer status category; the power transformer status category includes normal, discharge and overheating, low-energy discharge, high-energy discharge, medium and low temperature overheating, and high temperature overheating.

[0056] Furthermore, the one-dimensional structured data sample is converted into a two-dimensional grayscale image using an improved Gram angular field transform method, specifically:

[0057] Step 2.1, preprocess the one-dimensional structured data sample, the expression is as follows:

[0058]

[0059] In the formula, g i is the i-th variable in sample g, For pre-processed samples The i-th variable in ; max(g) and min(g) represent the maximum and minimum values ​​of the variable in sample g, respectively;

[0060] Step 2.2, Perform polar coordinate transformation, as shown in formula (2):

[0061]

[0062] Where: N is a constant factor used to adjust the span of polar coordinates, Φ i is the polar angle of point i, r i is the polar diameter of point i.

[0063] Step 2.3, after converting the data to polar coordinates, use GASF and GADF to convert the data into two-dimensional image data, where GASF and GADF are shown in equations (3) and (4). The exemplary conversion results are as follows: Figure 2 (a) and Figure 2 (b) as shown:

[0064] GASF (i,j) =[cos(Φ i +Φ j )] (3)

[0065] GADF (i,j) =[sin(Φ i -Φ j )] (4)

[0066] Where: Φ j represents the jth polar angle in Φ.

[0067] From equations (3) and (4), we can see that the Gram sum angular field converts the one-dimensional structured data into a symmetric matrix, and the matrix elements store the original data information; the matrix obtained by the Gram difference angular field is a matrix in which all diagonal elements are 0, and the symmetric elements on both sides of the main diagonal are opposite to each other. Therefore, the numerical matrices after the two types of angular field transformations have symmetric elements on both sides of the diagonal that are redundant with each other.

[0068] Step 2.4, in order to solve the problem of matrix element redundancy, the present invention improves the matrices obtained by the two types of conversion models, as shown in formula (5), denoted as GAF. It can be seen from formula (5) that the lower triangular and diagonal elements of the GAF matrix are composed of the lower triangular and diagonal elements of the Gram sum angular field, and the upper triangular elements are composed of the upper triangular elements of the Gram difference angular field, as shown in Figure 3 As shown, the redundancy problem is effectively solved and the data features are enriched.

[0069]

[0070] Step 2.5, map the GAF matrix to the interval [0,255], such as Figure 4 The grayscale image is formed as shown in formula (6).

[0071]

[0072] Among them, GAF max , GAF min Represent the maximum and minimum values ​​in GAF respectively.

[0073] For example, Figure 5 The figure shows the six status categories after GAF conversion.

[0074] Furthermore, the two-dimensional grayscale image in the original power transformer fault diagnosis sample set is enhanced, and the enhancement method includes at least one of the following: an image enhancement method based on image brightness, an image enhancement method based on salt and pepper noise, an image enhancement method based on Gaussian noise, and an image enhancement method based on vector quantization variational autoencoder. For example, Figure 6 As shown, the image enhancement method based on image brightness is used for the original image to obtain Figure 6 (b) Figure 6 (c) Two cases are shown: the image enhancement method based on salt and pepper noise is used to obtain Figure 6 (d) Schematic diagram; the image enhancement method based on Gaussian noise is used to obtain Figure 6 (e) Schematic diagram; the image enhancement method based on vector quantization variational autoencoder is used to obtain the following Figure 6 (f) Indication.

[0075] Furthermore, the target model adopts at least one of the following: VGG19 network model, ResNet50 model.

[0076] Furthermore, the VGG19 network model consists of 16 convolution layers (CL), 3 fully connected layers (FCL) and 5 pooling layers (PL), and its structure is as follows: Figure 7 As shown in the figure. This model achieves a deeper network structure through simple stacking and has strong nonlinear fitting capabilities. The VGG19 network contains 38942534 parameters and a size of 152.204Mb, of which the front-end convolution and pooling layer contains 20035392 parameters and a size of 78.465Mb, and the terminal fully connected layer contains 18907142 parameters and a size of 73.748Mb.

[0077] Furthermore, the ResNet50 model network structure is as follows Figure 8 As shown in the figure. The ResNet50 network contains 38,129,670 parameters and the model size is 149.436MB. The schematic diagram of the residual structure is as follows Fig. 9 shown.

[0078] In order to further illustrate the effectiveness of the present invention, the following experiments are conducted:

[0079] The experiments of the present invention are all built using pytorch, and the experimental operating environment is a computer equipped with an Intel Core i512400F central processing unit (CPU) and an NVIDIA Geforce RTX3060 graphics processing unit (GPU). The model training adopts the stochastic gradient descent method, the number of training rounds is set to 175, and the initial learning rate is 0.01. The cosine annealing method is used to adjust the learning rate, which can prevent the model from stopping training at the local optimal solution and effectively improve the model training accuracy.

[0080] The performance evaluation method uses the time required for model recognition (unit: seconds (s)) and the accuracy (A), as shown in formula (7).

[0081]

[0082] Where: T p is the total number of correctly classified samples, F p is the total number of misclassified samples.

[0083] 1. Construction of original and expanded power transformer fault diagnosis sample sets

[0084] The transformer oil chromatographic fault data used in this experiment comes from some data collected from a substation and some literature. After sorting, the original one-dimensional structured data samples of dissolved gas in power transformer oil totaled 1,678, including normal and fault modes, namely "normal", "high energy discharge", "low energy discharge", "high temperature overheating", "medium and low temperature overheating" and "discharge and overheating". In order to better carry out the performance evaluation of the fault identification model, the original one-dimensional structured data samples are divided into training set and test set, with a ratio of 8:2.

[0085] The original one-dimensional structured data samples are transformed using the improved Gram angular field transform (GAF) proposed in the present invention and the traditional continuous wavelet transform (CWT) and short time Fourier transform (STFT) methods to obtain different types of two-dimensional image samples as the original power transformer fault diagnosis sample set. Figure 4 The fault sample conversion results in the case are as follows Fig.10 Furthermore, the two-dimensional images obtained by each comparative data transformation method are enhanced using a data enhancement method to obtain an expanded power transformer fault diagnosis sample set.

[0086] like Fig.11 The time required for converting 1000 one-dimensional samples by three data transformation methods is shown. The experimental results show that the GAF method proposed in the present invention is significantly faster than the CWT method, but slightly slower than the STFT method.

[0087] 2. Diagnosis based on the original power transformer fault diagnosis sample set

[0088] In order to verify the effectiveness of the GAF proposed in this paper, the two-dimensional image data converted by the three methods of GAF, CWT and STFT were input into the VGG19 and ResNet50 networks and tested. The test accuracy and loss values ​​of GAF, CWT and STFT on the original data set are shown in Figure 2. Figures 12 to 14 The test results of VGG19 and ResNet50 models on the original dataset are shown in Table 1.

[0089] Table 1 Test accuracy of VGG19 and ResNet50 models in the original dataset

[0090] Image Type VGG19 / % ResNet50 / % GAF 80.06 80.68 CWT 88.63 88.81 STFT 51.28 50.41

[0091] Depend on Figures 12 to 14As shown in Table 1, the image obtained by GAF conversion constructed by the present invention is much higher than the image obtained by STFT conversion in terms of model recognition accuracy. Although slightly lower than the image obtained by CWT conversion, the time required for GAF conversion is significantly less than that required for CWT conversion. Therefore, GAF is used as a method for converting one-dimensional structured data into two-dimensional image data in subsequent experiments.

[0092] 3. Diagnosis based on the expanded power transformer fault diagnosis sample set

[0093] In order to verify the effectiveness of the data enhancement method proposed in this invention, the expanded power transformer fault diagnosis sample set obtained by GAF conversion and then enhancement is input into the VGG19 and ResNet50 networks and tested. The test accuracy and loss value of GAF on the expanded power transformer fault diagnosis sample set are shown in Figure 2. Fig.15 shown.

[0094] By comparison Fig.15 and Fig.12 It can be seen that the VGG19 and ResNet50 network models trained using the expanded power transformer fault diagnosis sample set after data enhancement are more accurate than the networks trained using the original data sets, indicating that data enhancement work is effective and necessary.

[0095] According to the above experimental results, the subsequent experiments of the present invention select the expanded power transformer fault diagnosis sample set obtained by GAF and data enhancement to carry out fault diagnosis.

[0096] According to a second aspect of an embodiment of the present invention, there is provided a system for diagnosing faults of a power transformer based on dissolved gas in oil, comprising:

[0097] An acquisition module is used to acquire one-dimensional structured data samples of dissolved gas in multiple power transformer oils; the one-dimensional structured data samples are constructed using the content of dissolved gas in the oil;

[0098] An image conversion module is used to convert a one-dimensional structured data sample into a two-dimensional grayscale image by using an improved Gram angular field transform method to form an original power transformer fault diagnosis sample set;

[0099] An image enhancement module is used to enhance the two-dimensional grayscale image in the original power transformer fault diagnosis sample set to obtain an expanded power transformer fault diagnosis sample set; and is used to divide the expanded power transformer fault diagnosis sample set into a training set and a test set according to a preset ratio;

[0100] The fault diagnosis module is used to train the target model according to the training set to obtain a trained power transformer fault diagnosis model; test the trained power transformer fault diagnosis model according to the test set, and output the power transformer status category; the power transformer status category includes normal, discharge and overheating, low-energy discharge, high-energy discharge, medium and low temperature overheating, and high temperature overheating.

[0101] For the parts of the modules not described in detail above, please refer to the relevant description in this embodiment.

[0102] For example, Fig.16 As shown, a control method for a system for diagnosing faults of a power transformer based on dissolved gas in oil is provided, comprising:

[0103] The first step is to obtain one-dimensional structured data samples of dissolved gas in multiple power transformer oils according to an acquisition module;

[0104] The second step is to convert the one-dimensional structured data samples into two-dimensional grayscale images based on the improved Gram angular field transform method according to the image conversion module to form an original power transformer fault diagnosis sample set, and divide it into a first training set and a first test set;

[0105] The third step is to determine whether enhancement and expansion are needed: under the premise of satisfying the need for enhancement and expansion, the image enhancement module is called to enhance the two-dimensional grayscale image in the original power transformer fault diagnosis sample set to obtain the expanded power transformer fault diagnosis sample set, and divide it into a second training set and a second test set;

[0106] In the fourth step, on the premise that it is determined in the third step that enhancement and expansion are not needed, the fault diagnosis module is called according to the first training set to perform model training to obtain the best trained model, and the output diagnosis result is tested through the first test set; on the premise that it is determined in the third step that enhancement and expansion are needed, the fault diagnosis module is called according to the second training set to perform model training to obtain the best trained model, and the output diagnosis result is tested through the second test set.

[0107] The specific implementation modes of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A method for fault diagnosis of power transformer based on dissolved gas in oil, characterized in that: The following steps are involved: Step 1, obtaining one-dimensional structured data samples of dissolved gas in multiple power transformer oils; the one-dimensional structured data samples are constructed using the content of dissolved gas in the oil; Step 2, converting the one-dimensional structured data samples into two-dimensional grayscale images by using the improved Gram angular field transform method to form the original power transformer fault diagnosis sample set; Step 3, enhancing the two-dimensional grayscale image in the original power transformer fault diagnosis sample set to obtain an expanded power transformer fault diagnosis sample set; Step 4, dividing the expanded power transformer fault diagnosis sample set into a training set and a test set according to a preset ratio; Step 5. Train the target model based on the training set to obtain a trained power transformer fault diagnosis model; test the trained power transformer fault diagnosis model based on the test set to output the power transformer status category; the power transformer status category includes normal, discharge and overheating, low-energy discharge, high-energy discharge, medium and low temperature overheating, and high temperature overheating.

2. The method for fault diagnosis of power transformer based on dissolved gas in oil according to claim 1 is characterized in that: The one-dimensional structured data sample is converted into a two-dimensional grayscale image by using an improved Gram angular field transform method, specifically: Step 2.1, preprocess the one-dimensional structured data sample, the expression is as follows: In the formula, g i is the i-th variable in sample g, For pre-processed samples The i-th variable in ; max(g) and min(g) represent the maximum and minimum values ​​of the variable in sample g, respectively; Step 2.2, Perform polar coordinate transformation: Where: N is a constant factor used to adjust the span of polar coordinates, Φ i is the polar angle of point i, r i is the polar diameter of point i; Step 2.

3. After converting the data to polar coordinates, use GASF and GADF to convert the data into two-dimensional image data, where GASF and GADF are as follows: GASF (i,j) =[cos(Φ i +F j )]; GADF (i,j) =[sin(Φ i -F j )] Where: Φ j represents the jth polar angle; Step 2.4: Improve the matrix converted in Step 2.3, denoted as GAF, and the expression is as follows: Step 2.5, map the GAF matrix to the interval [0,255].

3. The method for fault diagnosis of power transformer based on dissolved gas in oil according to claim 1 is characterized in that: The two-dimensional grayscale image in the original power transformer fault diagnosis sample set is enhanced, and the enhancement method includes at least one of the following: an image enhancement method based on image brightness, and an image enhancement method based on salt and pepper noise; Image enhancement method based on Gaussian noise, image enhancement method based on vector quantization variational autoencoder.

4. The method for fault diagnosis of power transformer based on dissolved gas in oil according to claim 1 is characterized in that: The target model adopts at least one of the following: VGG19 network model, ResNet50 model.

5. A system for fault diagnosis of power transformer based on dissolved gas in oil, characterized in that: include: An acquisition module, used for acquiring one-dimensional structured data samples of dissolved gas in multiple power transformer oils; The one-dimensional structured data sample is constructed using the content of dissolved gas in oil; An image conversion module is used to convert a one-dimensional structured data sample into a two-dimensional grayscale image by using an improved Gram angular field transform method to form an original power transformer fault diagnosis sample set; An image enhancement module is used to enhance the two-dimensional grayscale image in the original power transformer fault diagnosis sample set to obtain an expanded power transformer fault diagnosis sample set; Used to divide the expanded power transformer fault diagnosis sample set into a training set and a test set according to a preset ratio; The fault diagnosis module is used to train the target model according to the training set to obtain a trained power transformer fault diagnosis model; test the trained power transformer fault diagnosis model according to the test set, and output the power transformer status category; the power transformer status category includes normal, discharge and overheating, low-energy discharge, high-energy discharge, medium and low temperature overheating, and high temperature overheating.

Images