Transformer operation state monitoring method and device

Abstract

The invention provides a transformer operation state monitoring method and device, and the method comprises the steps: obtaining N groups of sample data of a transformer in a historical time period, carrying out the processing of DGA original data according to a preset method, obtaining a real ratio feature, and carrying out the prediction according to the sample data and an initial convolutional neural network model, and obtaining a sample prediction feature; the initial convolutional neural network model predicts the operation state of the transformer according to the sample prediction features; and training the initial convolutional neural network model according to the sample prediction features and the real ratio features, the operation state corresponding to the DGA original data and the operation state of the transformer obtained through prediction, and obtaining a target convolutional neural network model. A traditional method is integrated into the target convolutional neural network model, so that when the target convolutional neural network model monitors the operation state of the transformer by using the DGA original data, the monitoring accuracy and efficiency are improved.

Description

Technical Field

[0001] The present application relates to the technical field of transformers, and in particular to a method and device for monitoring the operating status of a transformer. Background Art

[0002] Transformers are key components of power grids, and their health is crucial to the safe and stable operation of power systems. However, transformers have complex structures and operating environments. Over time, they can be susceptible to surface contamination, moisture, and insulation aging, which can easily lead to faults such as partial discharge and overheating. If internal faults such as localized overheating or discharge occur, heat from the insulating oil and solid insulation materials generates gases, which dissolve in the oil. Chromatographic analysis can detect the composition and content of these gases, thereby determining the nature and severity of the fault within the transformer.

[0003] In the prior art, the composition and content of gas in oil are analyzed by chromatography, usually using traditional methods such as the three-ratio method, the four-ratio method, and the David triangle method. Although these traditional methods have the advantage of simple operation, they have limitations in monitoring the fault type when analyzing transformer faults. For example, the three-ratio method uses ratios for linear classification, which is prone to the problem of overly absolute classification boundaries. Compared with the three-ratio method, the four-ratio method has higher accuracy for overheating faults in conductive circuits and magnetic circuits. Combined with the three-ratio method, it can distinguish whether the overheating fault occurs in the magnetic circuit or the conductive circuit, but it is easy to misjudge other faults.

[0004] Therefore, it is difficult to analyze and determine transformer faults using a traditional method, resulting in low efficiency in intelligent monitoring and early warning of transformer operating status. Summary of the Invention

[0005] The present application provides a transformer operating status monitoring method and device to solve the technical problems mentioned in the background technology.

[0006] In a first aspect, the present application provides a transformer operating status monitoring method, comprising: obtaining N sets of sample data within a historical time period of the transformer, the sample data comprising N sets of DGA raw data and corresponding operating status, the DGA raw data being data obtained by analyzing the dissolved gas content of the transformer's insulating oil, the dissolved gases comprising hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, and carbon dioxide, where N is a positive integer greater than or equal to 1; Processing the DGA raw data according to a preset method to obtain a true ratio feature, wherein the preset method includes at least one of the following: a two-ratio method, a three-ratio method, a four-ratio method, an uncoded ratio method, a furan compound analysis method, a David triangle method, and an IEC new guideline; Predicting a sample prediction feature based on the sample data and the initial convolutional neural network model, where the sample prediction feature is a feature related to the corresponding operating state in the DGA raw data predicted by the initial convolutional neural network model based on the operating state corresponding to the DGA raw data, and the composition of the sample prediction feature is consistent with the composition of the true ratio feature; The initial convolutional neural network model predicts the operating state of the transformer according to the sample prediction features; According to the sample prediction features and the true ratio features, the operating status corresponding to the DGA original data and the predicted operating status of the transformer, the initial convolutional neural network model is trained to obtain a target convolutional neural network model, and the target convolutional neural network model is used to predict the operating status of the transformer based on the DGA original data.

[0007] Optionally, the training of the initial convolutional neural network model to obtain a target convolutional neural network model based on the sample prediction feature and the true ratio feature, the operating state corresponding to the DGA original data, and the predicted operating state of the transformer includes: Obtaining a first loss value according to the sample prediction feature, the true ratio feature, and a first loss function; Obtaining a second loss value according to the operating state corresponding to the DGA original data, the predicted operating state of the transformer, and a second loss function; Obtaining a target loss function according to the first loss function and the second loss function; The initial convolutional neural network model is trained according to the target loss function, the first loss value, and the second loss value to obtain a target convolutional neural network model.

[0008] Optionally, the first loss function is: in, Indicates the i The first j Features, Indicates the i The first in the group true ratio feature j Features, J Indicates the number of features contained in each set of sample prediction features, L k represents the first loss value; The second loss function is: in, xi Indicates the i Set the DGA raw data, y i express x i The corresponding operating status, l express x i The corresponding operating status, L Indicates the number of types corresponding to the running status, L c represents the second loss value, log ( h ( x i ) l ) means that according to i The operating status of the transformer obtained by group DGA raw data prediction; The objective loss function is: in, represents the weight of the second loss value, represents the weight of the first loss value, p ki Represents the true ratio feature corresponding to the i-th group of DGA raw data.

[0009] Optionally, obtaining N groups of sample data within a historical time period of the transformer includes: Obtain K groups of real sample data within a historical time period of the transformer, where the real sample data is the real DGA original data and corresponding operating status of the transformer within the historical time period; Based on the K groups of real sample data, sample data corresponding to the operating states with less sample data in the K groups of real sample data are expanded to obtain the M groups of expanded sample data; The N groups of sample data are obtained according to the K groups of real sample data and the M groups of expanded sample data.

[0010] Optionally, performing sample expansion on sample data corresponding to operating states with a small amount of sample data in the K groups of real sample data to obtain the M groups of expanded sample data includes: According to the K groups of real sample data, the WANG model is used to perform sample expansion on the sample data corresponding to the operating state with a small amount of sample data in the K groups of real sample data to obtain the M groups of expanded sample data.

[0011] Optionally, also include: Obtain the DGA raw data of the target transformer; According to the DGA original data of the target transformer and the target convolutional neural network model, the predicted ratio characteristics and the operating status of the target transformer are predicted.

[0012] Optionally, after obtaining the predicted ratio feature and the operating status of the target transformer, the prediction further includes: According to the prediction ratio characteristics and the preset method, the operating status of the target transformer determined according to each preset method is determined, and the operating status of the target transformer determined according to each preset method is used to correct the predicted operating status of the target transformer.

[0013] In a second aspect, the present application provides a transformer operating status monitoring device, comprising: A sample acquisition module is configured to acquire N sets of sample data within a historical period of the transformer, wherein the sample data includes N sets of DGA raw data and corresponding operating status, wherein the DGA raw data is data obtained by analyzing the dissolved gas content of the transformer's insulating oil, wherein the dissolved gases include hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, and carbon dioxide, and N is a positive integer greater than or equal to 1; a ratio feature acquisition module, configured to process the DGA raw data according to a preset method to obtain a true ratio feature, wherein the preset method includes at least one of the following: a two-ratio method, a three-ratio method, a four-ratio method, an uncoded ratio method, a furan compound analysis method, a David triangle method, and an IEC new guideline; a first prediction module, configured to predict and obtain a sample prediction feature based on the sample data and the initial convolutional neural network model, wherein the sample prediction feature is a feature related to the corresponding operating state in the DGA raw data predicted by the initial convolutional neural network model based on the operating state corresponding to the DGA raw data, and the composition of the sample prediction feature is consistent with the composition of the true ratio feature; A second prediction module is used for the initial convolutional neural network model to predict the operating state of the transformer according to the sample prediction features; A training module is used to train the initial convolutional neural network model based on the sample prediction features and the true ratio features, the operating status corresponding to the DGA original data, and the predicted operating status of the transformer to obtain a target convolutional neural network model. The target convolutional neural network model is used to predict the operating status of the transformer based on the DGA original data.

[0014] Optionally, the training module trains the initial convolutional neural network model according to the sample prediction feature and the true ratio feature, as well as the operating state corresponding to the DGA original data and the predicted operating state of the transformer to obtain a target convolutional neural network model, specifically for: Obtaining a first loss value according to the sample prediction feature, the true ratio feature, and a first loss function; Obtaining a second loss value according to the operating state corresponding to the DGA original data, the predicted operating state of the transformer, and a second loss function; Obtaining a target loss function according to the first loss function and the second loss function; The initial convolutional neural network model is trained according to the target loss function, the first loss value, and the second loss value to obtain a target convolutional neural network model.

[0015] Optionally, the first loss function is: in, Indicates the i The first j Features, Indicates the i The first in the group true ratio feature j Features, J Indicates the number of features contained in each set of sample prediction features, L k represents the first loss value; The second loss function is: in, x i Indicates the i Set the DGA raw data, y i express x i The corresponding operating status, l express x i The corresponding operating status, L Indicates the number of types corresponding to the running status, L c represents the second loss value, log ( h ( x i ) l ) means that according to i The operating status of the transformer obtained by group DGA raw data prediction; The objective loss function is: in, represents the weight of the second loss value, represents the weight of the first loss value, p ki Represents the true ratio feature corresponding to the i-th group of DGA raw data.

[0016] Optionally, the sample acquisition module acquires N groups of sample data within a historical time period of the transformer, specifically for: Obtain K groups of real sample data within a historical time period of the transformer, where the real sample data is the real DGA original data and corresponding operating status of the transformer within the historical time period; Based on the K groups of real sample data, sample data corresponding to the operating states with less sample data in the K groups of real sample data are expanded to obtain the M groups of expanded sample data; The N groups of sample data are obtained according to the K groups of real sample data and the M groups of expanded sample data.

[0017] Optionally, the sample acquisition module performs sample expansion on the sample data corresponding to the operating state with less sample data in the K groups of real sample data to obtain the M groups of expanded sample data, specifically for: According to the K groups of real sample data, the WANG model is used to perform sample expansion on the sample data corresponding to the operating state with a small amount of sample data in the K groups of real sample data to obtain the M groups of expanded sample data.

[0018] Optionally, it also includes: a monitoring data acquisition module and a monitoring module; Monitoring data acquisition module, used to obtain the DGA raw data of the target transformer; A monitoring module is used to predict and obtain the predicted ratio characteristics and the operating status of the target transformer based on the DGA raw data of the target transformer and the target convolutional neural network model.

[0019] Optionally, after the monitoring module predicts and obtains the predicted ratio feature and the operating status of the target transformer, it is further used to: According to the prediction ratio characteristics and the preset method, the operating status of the target transformer determined according to each preset method is determined, and the operating status of the target transformer determined according to each preset method is used to correct the predicted operating status of the target transformer.

[0020] In a third aspect, the present application provides an electronic device, comprising: a processor and a memory; Memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method as described in any one of the first aspects.

[0021] In a fourth aspect, an embodiment of the present application provides a readable storage medium, including a program or instruction. When the program or instruction runs on a computer, the method described in any one of the above-mentioned first aspects is executed.

[0022] In a fifth aspect, an embodiment of the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method as described in any one of the first aspects.

[0023] The transformer operating status monitoring method and device provided in the present application obtain N groups of sample data within a historical time period of the transformer, where the sample data include N groups of DGA raw data and corresponding operating status, process the DGA raw data according to a preset method to obtain true ratio features, and predict sample prediction features based on the sample data and an initial convolutional neural network model. The sample prediction features are features related to the corresponding operating status that are predicted in the DGA raw data by the initial convolutional neural network model based on the operating status corresponding to the DGA raw data, and the composition of the sample prediction features is consistent with the composition of the true ratio features. The initial convolutional neural network model predicts the operating status of the transformer based on the sample prediction features. The initial convolutional neural network model is trained based on the sample prediction features and the true ratio features, the operating status corresponding to the DGA raw data, and the predicted operating status of the transformer to obtain a target convolutional neural network model. The target convolutional neural network model is used to predict the operating status of the transformer based on the DGA raw data. The traditional methods such as the three-ratio method, the four-ratio method, and the David triangle method were integrated into the target convolutional neural network model. Through a large amount of sample data, the target convolutional neural network model not only learned the relationship between the DGA raw data and the transformer operating status, but also learned the advantages of various traditional chromatographic analysis methods in analyzing transformer faults. This improved the accuracy and efficiency of monitoring when the target convolutional neural network model used the DGA raw data to monitor the transformer operating status. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0025] Figure 1 A flowchart of a transformer operating status monitoring method provided in one embodiment of the present application; Figure 2 A schematic diagram of the structure of a convolutional neural network model provided in one embodiment of the present application; Figure 3 A WGAN model convergence curve diagram provided in one embodiment of the present application; Figure 4 A comparison chart of the augmented sample data and the real sample data corresponding to the WGAN model provided in one embodiment of the present application; Figure 5 A transformer state characteristic map provided in an embodiment of the present application; Figure 6 A schematic diagram of the structure of a transformer operating status monitoring device provided in one embodiment of the present application; Figure 7 A schematic diagram of the structure of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0026] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application are clearly and completely described below. Obviously, the described embodiments are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts also fall within the scope of protection of this application.

[0027] The health of transformers is crucial to the safe and stable operation of power systems. However, transformers have complex structures and operating environments. Over long-term operation, they can be susceptible to surface contamination, moisture, and insulation aging, which can easily lead to faults such as partial discharge and overheating. If a transformer experiences localized overheating or discharge, the insulating oil and solid insulation materials generate heat and generate gases. These gases dissolve in the oil. Chromatographic analysis can detect the composition and content of these gases, thereby determining the nature and severity of the fault within the transformer.

[0028] Among them, chromatographic analysis of the composition and content of gas in oil typically uses traditional methods such as the three-ratio method, the four-ratio method, and the David triangle method. While these traditional methods offer the advantage of simple operation, they have limitations in monitoring fault types when analyzing transformer faults. For example, the three-ratio method uses ratios for linear classification, which can easily lead to overly absolute classification boundaries. Compared to the three-ratio method, the four-ratio method has higher accuracy for overheating faults in both the conductive and magnetic circuits. Combined with the three-ratio method, it can distinguish whether an overheating fault occurs in the magnetic or conductive circuit, but it is prone to misdiagnosing other faults. In other words, each traditional method has a specific focus when analyzing fault types. Using one method may only detect one fault type, while other existing faults may go undetected. Alternatively, a fault may not be detected when the transformer is already faulty, affecting the determination of the transformer fault and resulting in low efficiency and accuracy in transformer fault diagnosis.

[0029] Therefore, in order to solve the above-mentioned technical problems existing in the prior art, the present application proposes a transformer operating status monitoring method and device, which collects a large amount of historical data of the transformer, that is, obtains the DGA original data obtained during historical detection, that is, the content of hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide and carbon dioxide dissolved in the insulating oil of the transformer, and the corresponding operating status of the transformer, and uses a large amount of historical data to train the convolutional neural network model. During the training, traditional methods such as the three-ratio method, the four-ratio method, and the David triangle method are integrated, so that the obtained convolutional neural network model integrates the advantages of various traditional methods in analyzing transformer faults. On this basis, the relationship between the DGA original data and the fault type is established, so that the convolutional neural network model is more accurate when analyzing the transformer operating status through the DGA original data, thereby improving the efficiency of transformer operating status monitoring.

[0030] Figure 1 This is a flow chart of a transformer operating status monitoring method provided in one embodiment of the present application. Figure 1 As shown, the execution subject of this embodiment may be, for example, a server, and the method includes: S101 : Acquire N groups of sample data within a historical time period of the transformer, where the sample data include N groups of DGA original data and corresponding operating status.

[0031] The DGA raw data is the data obtained by analyzing the dissolved gas content of the transformer insulating oil. The dissolved gases include hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide and carbon dioxide. N is a positive integer greater than or equal to 1.

[0032] In this embodiment, taking a 500kV transformer as an example, according to the standard requirements, the oil chromatogram of the operating transformer should be tested once every three months. For some transformers with problems, the oil chromatogram may need to be tested once a month.

[0033] The sample data used for analysis this time does not take special circumstances into consideration. Instead, the chromatographic data of the transformer in operation is tested for three months. In this way, the obtained sample data has strong regularity and time series.

[0034] Therefore, based on the data recorded in the transformer log, we obtained the DGA raw data from the regular monitoring of chromatographic data since 2006. The DGA raw data since 2006 corresponds to the actual operating status of the transformer. Therefore, the transformer operating status in the sample data is not determined by the DGA raw data, but is the actual operating status of the transformer at the corresponding time.

[0035] Therefore, these sample data are used to train the convolutional neural network model, and the trained model is more accurate.

[0036] Among them, the division of the operating status of the transformer is shown in Table 1: Running status Running status code Low temperature overheating 1 Partial discharge 2 Low-energy discharge 3 Arc discharge 4 Normal state 5 S102: Process the DGA raw data according to a preset method to obtain a true ratio feature.

[0037] The preset methods include at least one of the following: two-ratio method, three-ratio method, four-ratio method, uncoded ratio method, furan compound analysis method, David triangle method, and IEC new guidelines.

[0038] In this embodiment, as shown in Table 2, the ratio characteristics of the preset method are adopted, wherein the preset method is as shown in the table, and a set of ratio characteristics corresponding to multiple preset methods is obtained to form a true ratio characteristic, that is, the true ratio characteristics include: C2H2 / C2H4, CH4 / H2, C2H4 / C2H6, C2H6 / CH4, C2H2 / C2H6, H2 / CH4, CO2 / CO, C2H2%, C2H4%, and CH4%.

[0039] Table 2: S103. Predict and obtain sample prediction features based on the sample data and the initial convolutional neural network model.

[0040] The sample prediction features are the features related to the corresponding operating status in the DGA original data predicted by the initial convolutional neural network model based on the operating status corresponding to the DGA original data. The composition of the sample prediction features is consistent with the composition of the true ratio features.

[0041] In this embodiment, Figure 2 As shown, the convolutional neural network model includes two output terminals, one of which is used to output sample prediction features, wherein the composition of the sample prediction features is consistent with the composition of the true ratio features, that is, the number of feature data contained in the sample prediction features is consistent with the number of feature data contained in the true ratio features, and the order of the feature data in the sample prediction features is consistent with the order of the feature data in the true ratio features. For example, the true ratio features include C2H2 / C2H4, CH4 / H2, C2H4 / C2H6, C2H6 / CH4, C2H2 / C There are 10 data in total, including 2H6, H2 / CH4, CO2 / CO, C2H2%, C2H4%, and CH4%. The sample prediction feature also contains 10 feature data. The first feature data in the true ratio feature is C2H2 / C2H4. Then, the first feature data in the sample prediction feature also represents C2H2 / C2H4. However, this C2H2 / C2H4 is not obtained directly by comparing C2H2 and C2H4 in the input DGA raw data, but is predicted by the convolutional neural network model based on the input DGA raw data.

[0042] Therefore, according to the composition of the true ratio features, the convolutional neural network model predicts the sample prediction features based on the input DGA raw data.

[0043] S104. The initial convolutional neural network model predicts the operating status of the transformer based on the sample prediction features.

[0044] In this embodiment, according to S103, the composition of the sample prediction feature is consistent with the composition of the true ratio feature, and according to S102, the true ratio feature is obtained by processing the DGA raw data according to a preset method, and the preset method is a traditional method for analyzing the chromatogram in the prior art, such as the two-ratio method, the three-ratio method, the four-ratio method, the uncoded ratio method, the furan compound analysis method, etc. Therefore, the operating status of the transformer can be monitored according to the true ratio feature. Therefore, the operating status of the transformer can also be monitored through the sample prediction feature.

[0045] Therefore, during model training, the initial convolutional neural network model obtains the operating state of the transformer according to the sample prediction features, and outputs the operating state of the transformer through another output terminal.

[0046] S105. Based on the sample prediction features and the true ratio features, the operating status corresponding to the DGA original data and the predicted operating status of the transformer, the initial convolutional neural network model is trained to obtain a target convolutional neural network model.

[0047] The target convolutional neural network model is used to predict the operating status of the transformer based on the DGA raw data.

[0048] In this embodiment, since the sample prediction features are obtained by the initial convolutional neural network model through the DGA original data, there are differences between them and the true ratio features; the predicted operating state of the transformer is obtained by the initial convolutional neural network model through the sample prediction features, and there are also differences between them and the operating state corresponding to the DGA original data. These differences are caused by the initial convolutional neural network model. Therefore, the initial convolutional neural network model is trained based on these differences to reduce the difference between the sample prediction features and the true ratio features, as well as the difference between the operating state corresponding to the DGA original data and the predicted operating state of the transformer, thereby obtaining the target convolutional neural network model.

[0049] In this way, when the operating status of the transformer is monitored according to the DGA raw data through the target convolutional neural network model, the operating status of the transformer can be determined more accurately.

[0050] In this embodiment, N groups of sample data within a historical time period of the transformer are obtained, and the sample data include N groups of DGA raw data and corresponding operating states. The DGA raw data are processed according to a preset method to obtain true ratio features. Sample prediction features are predicted based on the sample data and the initial convolutional neural network model. The sample prediction features are features related to the corresponding operating states in the DGA raw data predicted by the initial convolutional neural network model based on the operating states corresponding to the DGA raw data. The composition of the sample prediction features is consistent with the composition of the true ratio features. The initial convolutional neural network model predicts the operating state of the transformer based on the sample prediction features. The initial convolutional neural network model is trained based on the sample prediction features and the true ratio features, the operating state corresponding to the DGA raw data, and the predicted operating state of the transformer to obtain a target convolutional neural network model. The target convolutional neural network model is used to predict the operating state of the transformer based on the DGA raw data. The traditional methods such as the three-ratio method, the four-ratio method, and the David triangle method were integrated into the target convolutional neural network model. Through a large amount of sample data, the target convolutional neural network model not only learned the relationship between the DGA raw data and the transformer operating status, but also learned the advantages of various traditional chromatographic analysis methods in analyzing transformer faults. This improved the accuracy and efficiency of monitoring when the target convolutional neural network model used the DGA raw data to monitor the transformer operating status.

[0051] Optionally, a specific implementation of S105 is: S1051. Obtain a first loss value based on the sample prediction feature and the true ratio feature, as well as a first loss function.

[0052] In this step, the first loss function is used to calculate the difference between the sample prediction feature and the true ratio feature. Optionally, the first loss function is: in, Indicates the i The first j Features, Indicates the i The first in the group true ratio feature j Features, J Indicates the number of features contained in each set of sample prediction features, L k Represents the first loss value.

[0053] S1052: Obtain a first loss value according to the operating state corresponding to the DGA original data and the operating state corresponding to the DGA original data, and a second loss function.

[0054] In this step, the second loss function is used to calculate the difference between the operating state corresponding to the DGA original data and the operating state corresponding to the DGA original data. The second loss function is: in, x i Indicates the i Set DGA raw data, y i express x i The corresponding operating status, l express x i The corresponding running state, L represents the number of types corresponding to the running state, L c represents the second loss value, log ( h ( x i ) l ) means that according to i The operating status of the transformer is obtained by predicting the original DGA data.

[0055] S1053. Obtain a target loss function according to the first loss function and the second loss function.

[0056] In this step, the target loss function includes the first loss function and the second loss function. Specifically, the target loss function is: in, represents the weight of the second loss value, represents the weight of the first loss value, p ki Indicates the i The true ratio characteristics corresponding to the original data of the group DGA, L c ( f ( x i ), y i )express f ( x i ) and i The running status corresponding to the original data of the group DGA y i The first loss value between , at this time, f ( x i ) represents the initial convolutional neural network model according to i The operating status of the transformer output by the DGA raw data, L k ( f ( x i ), p ki )express f ( x i ) and i The true ratio characteristics corresponding to the original data of the group DGA p ki The second loss value between , at this time, f ( x i ) represents the initial convolutional neural network model according to i The sample prediction features of the original data output of the group DGA.

[0057] S1054. Train the initial convolutional neural network model according to the target loss function and the first loss value and the second loss value to obtain a target convolutional neural network model.

[0058] In this step, according to the target loss function, the goal of the target loss function is to minimize the first loss value and the second loss value as a whole. At this time, the operating status of the transformer predicted by the target convolutional neural network model is more accurate.

[0059] Therefore, during training, the value of the target loss function is made as small as possible to obtain the target convolutional neural network model.

[0060] In this embodiment, a first loss value between a sample prediction feature and a true ratio feature is obtained through a first loss function, and a second loss value between an operating state corresponding to the DGA original data and an operating state corresponding to the DGA original data is obtained through a second loss function. The target loss value obtained is made as small as possible according to the target loss function and the first and second loss values. In this way, the accuracy of the obtained target convolutional neural network model is higher.

[0061] Optionally, the sample data obtained in this embodiment is data from regular chromatographic testing of transformers since 2006. Although the sample size is large, the actual incidence rate of each type of transformer fault varies significantly, resulting in significant differences in the number of sample data for each fault type. This leads to imbalanced sample data and affects the performance of the target convolutional neural network model obtained through training. Therefore, it is necessary to address the problem of unbalanced sample data.

[0062] Specifically, a specific implementation of S101 is: S1011. Obtain K groups of real sample data within a historical time period of the transformer. The real sample data are the real DGA original data and corresponding operating status of the transformer within the historical time period.

[0063] Specifically, using the relevant description in S101, the DGA raw data recorded by regular monitoring of chromatographic data and the corresponding transformer operating status since 2006 are obtained, and a total of K groups of real sample data are obtained. Among them, the sample data obtained in this way are real sample data.

[0064] S1011. Based on K groups of real sample data, sample data corresponding to the operating states with small sample data amounts in the K groups of real sample data are expanded to obtain M groups of expanded sample data.

[0065] Specifically, when performing sample expansion on the sample data corresponding to the operating state with a small amount of sample data in the K groups of real sample data, the sample expansion can be performed using methods in the prior art, for example, using the SMOTE algorithm to expand a small number of samples.

[0066] After expanding the sample data with less sample data, M groups of expanded sample data are obtained.

[0067] Alternatively, the WANG model has the advantages of high training stability, high-quality sample generation, strong learning generalization ability, mode collapse mitigation, strong interpretability of loss metrics, and no need to assume probability distribution. For example, taking low-energy discharge as an example, after 2000 iterations of training, the WGAN model convergence curve is as follows: Figure 3 shown.

[0068] Therefore, the specific implementation of S1011 may also be: According to K groups of real sample data, the WANG model is used to expand the sample data corresponding to the operating state with small sample data volume in the K groups of real sample data to obtain M groups of expanded sample data.

[0069] Specifically, K groups of real sample data are input into the WANG model. The WANG model expands the data volume of sample data corresponding to other operating states according to the data volume of sample data corresponding to the operating state with the largest data volume in the K groups of real sample data, so that the data volume of sample data corresponding to each operating state in the N groups of sample data is balanced.

[0070] For example, taking low-energy discharge as an example, the difference between the generated expanded sample data and the normalized real sample data is as follows: Figure 4 As shown, it can be seen that the characteristic trends of the real sample data and the expanded sample data are consistent.

[0071] S1011. Obtain N groups of sample data based on K groups of real sample data and M groups of expanded sample data.

[0072] Specifically, K groups of real sample data and M groups of expanded sample data are combined to obtain N groups of sample data.

[0073] In this embodiment, the above method is used to increase the number of sample data corresponding to the operating states with small sample data amounts in the sample data used for training the model, so that the number of sample data corresponding to the operating states of each transformer in the sample data used for training the model is balanced, thereby improving the performance of the target convolutional neural network model.

[0074] Optionally, after S105, the method further includes: S106: Obtain DGA raw data of the target transformer.

[0075] In this embodiment, after the training is completed to obtain the target convolutional neural network model, in the daily monitoring of the transformer, the target convolutional neural network model is applied to detect the operating status of the transformer according to the oil chromatogram of the transformer.

[0076] Specifically, when performing daily monitoring on the target transformer, the DGA raw data obtained through oil chromatography during the current monitoring is obtained.

[0077] S107. According to the DGA raw data of the target transformer and the target convolutional neural network model, the predicted ratio feature and the operating status of the target transformer are predicted.

[0078] In this embodiment, the DGA raw data of the target transformer is input into the target convolutional neural network model, and the target convolutional neural network model performs feature analysis on the DGA raw data of the target transformer to obtain and predict the ratio feature. The composition of the predicted ratio feature is consistent with the composition of the true ratio feature, that is, the predicted ratio feature can reflect the ratio feature obtained by using traditional methods, such as the three-ratio method, the two-ratio method, the four-ratio method, the uncoded ratio method, the furan compound analysis method, and the David triangle method.

[0079] The target convolutional neural network model obtains the operating status of the target transformer according to the prediction ratio feature prediction, thereby monitoring the operating status of the target transformer.

[0080] Optionally, after S107, the method further includes: S108. Determine the operating state of the target transformer determined according to each preset method based on the prediction ratio characteristics and the preset method, and use the operating state of the target transformer determined according to each preset method to correct the predicted operating state of the target transformer.

[0081] In this embodiment, the composition of the predicted ratio feature is consistent with the composition of the true ratio feature. When training the target convolutional neural network model, the true ratio feature is compared with the sample prediction feature, and the true ratio feature is obtained by traditional methods, such as the three-ratio method, the two-ratio method, the four-ratio method, the uncoded ratio method, the furan compound analysis method, and the David triangle method. Therefore, each feature in the predicted ratio feature is equivalent to being obtained by the corresponding traditional method. Since the predicted ratio feature is obtained by performing feature analysis on the DGA raw data of the target transformer by the target convolutional neural network model, the predicted ratio feature can better reflect the operating status of the target transformer than the true ratio feature corresponding to the DGA raw data of the target transformer.

[0082] Therefore, the operating state of the target transformer corresponding to each preset method is determined according to the prediction ratio characteristics, and the following Figure 5 The transformer state characteristic diagram is shown.

[0083] Figure 5 In the figure, the vertical axis represents the more specific operating status of the transformer, specifically: ND: not defined; OK: normal sample; S: stray gas; PD: partial discharge; T1: low temperature overheating; T2: medium overheating; T3: high temperature overheating; DT: discharge and overheating; D1: low energy discharge; D2: high energy discharge; C: Cellulose deterioration.

[0084] The horizontal axis represents the preset method, and each grid represents the operating state of the transformer that can be determined by the preset method. When the grid is shaded, it represents the operating state of the transformer determined by the preset method. For example, the first shaded grid on the left indicates that the operating state of the target transformer determined by the data corresponding to the three-ratio method in the predicted ratio feature is "high-energy discharge".

[0085] Figure 6 The schematic diagram of the structure of the transformer operation status monitoring device provided in one embodiment of the present application. The transformer operation status monitoring device can be a server, such as Figure 6 As shown, the transformer operating state monitoring device includes: a sample acquisition module 601 , a ratio feature acquisition module 602 , a first prediction module 603 , a second prediction module 604 and a training module 605 .

[0086] Optional, such as Figure 6 As shown, the transformer operation status monitoring device further includes: a monitoring data acquisition module 606 and a monitoring module 607 .

[0087] The sample acquisition module 601 is configured to acquire N sets of sample data within a historical period of the transformer, wherein the sample data includes N sets of DGA raw data and corresponding operating status. The DGA raw data is data obtained by analyzing the dissolved gas content of the transformer's insulating oil. The dissolved gases include hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, and carbon dioxide. N is a positive integer greater than or equal to 1. a ratio feature acquisition module 602 for processing the DGA raw data according to a preset method to obtain a true ratio feature, wherein the preset method includes at least one of the following: a two-ratio method, a three-ratio method, a four-ratio method, an uncoded ratio method, a furan compound analysis method, a David triangle method, and an IEC new guideline; A first prediction module 603 is configured to predict and obtain a sample prediction feature based on the sample data and the initial convolutional neural network model. The sample prediction feature is a feature related to the corresponding operating state in the DGA raw data predicted by the initial convolutional neural network model based on the operating state corresponding to the DGA raw data. The composition of the sample prediction feature is consistent with the composition of the true ratio feature. A second prediction module 604 is configured to use the initial convolutional neural network model to predict the operating state of the transformer based on the sample prediction features; The training module 605 is used to train the initial convolutional neural network model based on the sample prediction features and the true ratio features, the operating status corresponding to the DGA original data, and the predicted operating status of the transformer to obtain a target convolutional neural network model. The target convolutional neural network model is used to predict the operating status of the transformer based on the DGA original data.

[0088] Optionally, the training module 605 trains the initial convolutional neural network model according to the sample prediction feature and the true ratio feature, as well as the operating state corresponding to the DGA original data and the predicted operating state of the transformer to obtain a target convolutional neural network model, specifically for: Obtaining a first loss value according to the sample prediction feature, the true ratio feature, and a first loss function; Obtaining a second loss value according to the operating state corresponding to the DGA original data, the predicted operating state of the transformer, and a second loss function; Obtaining a target loss function according to the first loss function and the second loss function; The initial convolutional neural network model is trained according to the target loss function, the first loss value, and the second loss value to obtain a target convolutional neural network model.

[0089] Optionally, the first loss function is: in, Indicates the i The first j Features, Indicates the i The first in the group true ratio feature j Features, J Indicates the number of features contained in each set of sample prediction features, L k represents the first loss value; The second loss function is: in, x i Indicates the i Set the DGA raw data, y i Indicates based on x i The operating state of the transformer obtained by prediction, l express x i The corresponding operating status,L Indicates the number of types corresponding to the running status, L c represents the second loss value, log ( h ( x i ) l ) means that according to i The operating status of the transformer obtained by group DGA raw data prediction; The objective loss function is: in, represents the weight of the second loss value, represents the weight of the first loss value, p ki Indicates i The true ratio characteristics corresponding to the original data of the group DGA, L c ( f ( x i ), y i )express f ( x i ) and i The running status corresponding to the original data of the group DGA y i The first loss value between , at this time, f ( x i ) represents the initial convolutional neural network model according to i The operating status of the transformer output by the DGA raw data, L k ( f ( x i ), p ki )express f ( x i ) and i The true ratio characteristics corresponding to the original data of the group DGA p ki The second loss value between , at this time, f ( x i ) represents the initial convolutional neural network model according to i The sample prediction features of the original data output of the group DGA.

[0090] Optionally, the sample acquisition module 601 acquires N groups of sample data within a historical time period of the transformer, specifically for: Obtain K groups of real sample data within a historical time period of the transformer, where the real sample data is the real DGA original data and corresponding operating status of the transformer within the historical time period; Based on the K groups of real sample data, sample data corresponding to the operating states with less sample data in the K groups of real sample data are expanded to obtain the M groups of expanded sample data; The N groups of sample data are obtained according to the K groups of real sample data and the M groups of expanded sample data.

[0091] Optionally, the sample acquisition module 601 performs sample expansion on the sample data corresponding to the operating state with less sample data in the K groups of real sample data to obtain the M groups of expanded sample data, specifically for: According to the K groups of real sample data, the WANG model is used to perform sample expansion on the sample data corresponding to the operating state with a small amount of sample data in the K groups of real sample data to obtain the M groups of expanded sample data.

[0092] Optionally, a monitoring data acquisition module 606 is used to obtain DGA raw data of the target transformer; The monitoring module 607 is used to predict and obtain the predicted ratio characteristics and the operating status of the target transformer based on the DGA raw data of the target transformer and the target convolutional neural network model.

[0093] Optionally, after the monitoring module 607 predicts and obtains the predicted ratio feature and the operating status of the target transformer, it is further configured to: According to the prediction ratio characteristics and the preset method, the operating status of the target transformer determined according to each preset method is determined, and the operating status of the target transformer determined according to each preset method is used to correct the predicted operating status of the target transformer.

[0094] The transformer operation status monitoring device provided in the embodiment of the present application has a specific implementation process, which can be found in the above-mentioned method embodiment. Its implementation principle and technical effects are similar, and will not be repeated here in this embodiment.

[0095] Figure 7 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present application. Figure 7 As shown, the electronic device includes: a processor 710 and a memory 720.

[0096] The memory 720 stores computer-executable instructions.

[0097] The processor 710 executes the computer-executable instructions stored in the memory 720 , so that the processor 710 performs the method described in any one of the above embodiments.

[0098] The specific implementation process of the electronic device provided in the embodiment of the present application can be found in the above-mentioned method embodiment. Its implementation principle and technical effects are similar, and will not be repeated here in this embodiment.

[0099] In the above Figure 7 In the illustrated embodiment, it should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly implemented by a hardware processor or implemented by a combination of hardware and software modules in the processor.

[0100] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk storage.

[0101] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be classified into address buses, data buses, and control buses. For ease of illustration, the buses in the drawings of this application are not limited to just one bus or just one type of bus.

[0102] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the method shown in the above method embodiment is implemented.

[0103] The computer-readable storage medium mentioned above can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The computer-readable storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0104] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist in a device as discrete components.

[0105] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented using hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for monitoring transformer operating status, characterized in that: include: Obtain N sets of sample data within a historical time period of the transformer, the sample data including N sets of DGA raw data and corresponding operating status, the DGA raw data being data obtained by analyzing the dissolved gas content of the transformer's insulating oil, the dissolved gases including hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, and carbon dioxide, where N is a positive integer greater than or equal to 1; Processing the DGA raw data according to a preset method to obtain a true ratio feature, wherein the preset method includes at least one of the following: a two-ratio method, a three-ratio method, a four-ratio method, an uncoded ratio method, a furan compound analysis method, a David triangle method, and an IEC new guideline; Predicting a sample prediction feature based on the sample data and the initial convolutional neural network model, where the sample prediction feature is a feature related to the corresponding operating state in the DGA raw data predicted by the initial convolutional neural network model based on the operating state corresponding to the DGA raw data, and the composition of the sample prediction feature is consistent with the composition of the true ratio feature; The initial convolutional neural network model predicts the operating state of the transformer according to the sample prediction features; According to the sample prediction features and the true ratio features, the operating status corresponding to the DGA original data and the predicted operating status of the transformer, the initial convolutional neural network model is trained to obtain a target convolutional neural network model, and the target convolutional neural network model is used to predict the operating status of the transformer based on the DGA original data.

2. The method according to claim 1, characterized in that The initial convolutional neural network model is trained according to the sample prediction feature and the true ratio feature, as well as the operating state corresponding to the DGA original data and the predicted operating state of the transformer to obtain a target convolutional neural network model, including: Obtaining a first loss value according to the sample prediction feature, the true ratio feature, and a first loss function; Obtaining a second loss value according to the operating state corresponding to the DGA original data, the predicted operating state of the transformer, and a second loss function; Obtaining a target loss function according to the first loss function and the second loss function; The initial convolutional neural network model is trained according to the target loss function, the first loss value, and the second loss value to obtain a target convolutional neural network model.

3. The method according to claim 2, characterized in that The first loss function is: in, Indicates the i The first j Features, Indicates the i The first in the group true ratio feature j Features, J Indicates the number of features contained in each set of sample prediction features, L k represents the first loss value; The second loss function is: in, x i Indicates the i Set the DGA raw data, y i express x i The corresponding operating status, l express x i The corresponding running state, L represents the number of types corresponding to the running state, L c represents the second loss value, log ( h ( x i ) l ) means that according to i The operating status of the transformer obtained by group DGA raw data prediction; The objective loss function is: in, represents the weight of the second loss value, represents the weight of the first loss value, p ki Represents the true ratio feature corresponding to the i-th group of DGA raw data.

4. The method according to claim 1, wherein The step of obtaining N groups of sample data within a historical time period of the transformer includes: Obtain K groups of real sample data within a historical time period of the transformer, where the real sample data is the real DGA original data and corresponding operating status of the transformer within the historical time period; Based on the K groups of real sample data, sample data corresponding to the operating states with less sample data in the K groups of real sample data are expanded to obtain the M groups of expanded sample data; The N groups of sample data are obtained according to the K groups of real sample data and the M groups of expanded sample data.

5. The method according to claim 4, characterized in that The method of performing sample expansion on the sample data corresponding to the operating state with a small amount of sample data in the K groups of real sample data to obtain the M groups of expanded sample data includes: According to the K groups of real sample data, the WANG model is used to perform sample expansion on the sample data corresponding to the operating state with a small amount of sample data in the K groups of real sample data to obtain the M groups of expanded sample data.

6. The method according to any one of claims 1 to 5, characterized in that Also includes: Obtain the DGA raw data of the target transformer; According to the DGA original data of the target transformer and the target convolutional neural network model, the predicted ratio characteristics and the operating status of the target transformer are predicted.

7. The method according to claim 6, characterized in that After the prediction ratio feature and the operating status of the target transformer are obtained, the method further includes: According to the prediction ratio characteristics and the preset method, the operating status of the target transformer determined according to each preset method is determined, and the operating status of the target transformer determined according to each preset method is used to correct the predicted operating status of the target transformer.

8. A transformer operating status monitoring device, characterized in that: include: A sample acquisition module is configured to acquire N sets of sample data within a historical period of the transformer, wherein the sample data includes N sets of DGA raw data and corresponding operating status, wherein the DGA raw data is data obtained by analyzing the dissolved gas content of the transformer's insulating oil, wherein the dissolved gases include hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, and carbon dioxide, and N is a positive integer greater than or equal to 1; a ratio feature acquisition module, configured to process the DGA raw data according to a preset method to obtain a true ratio feature, wherein the preset method includes at least one of the following: a two-ratio method, a three-ratio method, a four-ratio method, an uncoded ratio method, a furan compound analysis method, a David triangle method, and an IEC new guideline; a first prediction module, configured to predict and obtain a sample prediction feature based on the sample data and the initial convolutional neural network model, wherein the sample prediction feature is a feature related to the corresponding operating state in the DGA raw data predicted by the initial convolutional neural network model based on the operating state corresponding to the DGA raw data, and the composition of the sample prediction feature is consistent with the composition of the true ratio feature; A second prediction module is used for the initial convolutional neural network model to predict the operating state of the transformer according to the sample prediction features; A training module is used to train the initial convolutional neural network model based on the sample prediction features and the true ratio features, the operating status corresponding to the DGA original data, and the predicted operating status of the transformer to obtain a target convolutional neural network model. The target convolutional neural network model is used to predict the operating status of the transformer based on the DGA original data.

9. An electronic device, characterized in that: include: processor and memory; Memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that: include: The program or instruction, when the program or instruction is run on a computer, the method according to any one of claims 1 to 7 is executed.

Images