A transformer fault diagnosis method, device and equipment based on data processing

Abstract

The application discloses a transformer fault diagnosis method, device and equipment based on data processing, comprising: acquiring characteristic quantity real-time data of transformer oil chromatogram; based on a preset auxiliary feature vector and a preset FCM-iForest algorithm, performing abnormal data detection on the characteristic quantity real-time data, and eliminating invalid abnormal data to obtain target characteristic quantity data; adopting a preset three-ratio method to perform fault discrimination on the target characteristic quantity data to obtain a three-ratio method fault discrimination result, and adopting a preset neural network model to perform fault discrimination on the target characteristic quantity data to obtain a neural network fault discrimination result; and determining a transformer fault cause according to the similarity between the three-ratio method fault discrimination result and the neural network fault discrimination result. The application realizes efficient utilization of transformer state detection data and efficient fault discrimination.

Description

Technical Field

[0001] This invention relates to the field of transformer abnormal data detection technology, and specifically to a transformer fault diagnosis method, apparatus, and equipment based on data processing. Background Technology

[0002] Power transformers are the most important and critical equipment in the power grid. Their function is to convert electrical energy from high voltage to low voltage during transmission and distribution to meet user needs. However, due to the wide variety of sources and types of power equipment status data, the quality of the collected data varies greatly. Furthermore, data transmission may be affected by signal interference or interruptions, inevitably resulting in abnormal data in the raw data. Low-quality detection data severely impacts the accuracy of power transformer fault identification, leading to low transformer early warning efficiency. Summary of the Invention

[0003] The purpose of this invention is to overcome the above-mentioned technical deficiencies and provide a transformer fault diagnosis method, device, electronic device and storage medium based on data processing, thereby solving the technical problem of low transformer early warning efficiency caused by low-quality detection data in the prior art.

[0004] To solve the above problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides a transformer fault diagnosis method based on data processing, comprising: Acquire real-time data of characteristic quantities in transformer oil chromatography; Based on the preset auxiliary feature vector and the preset FCM-iForest algorithm, abnormal data detection is performed on the real-time data of the feature quantity, and invalid abnormal data is removed to obtain the target feature quantity data. The target feature data is subjected to fault identification using a preset three-ratio method to obtain a fault identification result using the three-ratio method, and the target feature data is subjected to fault identification using a preset neural network model to obtain a neural network fault identification result. The cause of transformer failure is determined based on the similarity between the fault discrimination results of the three-ratio method and the fault discrimination results of the neural network.

[0005] In some embodiments, the step of performing anomaly detection on the real-time feature data based on a preset auxiliary feature vector and a preset FCM-iForest algorithm, and removing invalid anomaly data to obtain target feature data, includes: The pre-defined FCM-iForest algorithm is used to detect abnormal data in the real-time data of the feature quantity to obtain the first abnormal dataset. A transformer detection auxiliary dataset is constructed based on a preset auxiliary feature vector. Anomalies are detected in the transformer detection auxiliary dataset using a preset FCM-iForest algorithm to obtain a second abnormal dataset. Based on the comparison results between the first abnormal dataset and the second abnormal dataset, valid abnormal data and invalid abnormal data are determined, and the invalid abnormal data is removed to obtain the target feature data.

[0006] In some embodiments, the construction of a transformer detection auxiliary dataset based on a preset auxiliary feature vector, and the use of a preset FCM-iForest algorithm to perform anomaly detection on the transformer detection auxiliary dataset to obtain a second anomaly dataset, includes: The transformer detection auxiliary dataset is clustered using the preset FCM-iForest algorithm to obtain the clustering results of the transformer detection auxiliary dataset; Based on the clustering results, a pre-defined isolated forest method is used to detect anomalies in different clusters of data, resulting in a second abnormal dataset.

[0007] In some embodiments, determining valid and invalid abnormal data based on the comparison result between the first abnormal dataset and the second abnormal dataset includes: By comparing the first abnormal dataset and the second abnormal dataset, common abnormal data and non-common abnormal data in the first abnormal dataset and the second abnormal dataset are determined. The common anomalous data is defined as the valid anomalous data, and the non-common anomalous data is defined as the invalid anomalous data.

[0008] In some embodiments, after removing the invalid abnormal data, the method further includes: A preset linear interpolation method is used to fill in the missing values ​​after removing the invalid and abnormal data.

[0009] In some embodiments, determining the cause of a transformer fault based on the similarity between the fault identification result of the three-ratio method and the fault identification result of the neural network includes: If the fault identification result of the three-ratio method is consistent with the fault identification result of the neural network, the cause of the transformer fault is determined and an alarm is triggered for the transformer fault.

[0010] In some embodiments, after acquiring the real-time characteristic data of transformer oil chromatography, the method further includes: A preset random forest algorithm is used to fill in the missing data in the real-time data of the feature quantities.

[0011] Secondly, the present invention also provides a transformer fault diagnosis device based on data processing, comprising: The acquisition module is used to acquire real-time data of characteristic quantities in transformer oil chromatography. The target feature quantity determination module is used to perform abnormal data detection on the real-time data of the feature quantity based on the preset auxiliary feature vector and the preset FCM-iForest algorithm, and remove invalid abnormal data to obtain the target feature quantity data. The fault identification module is used to identify faults in the target feature data using a preset three-ratio method to obtain a fault identification result using the three-ratio method, and to identify faults in the target feature data using a preset neural network model to obtain a neural network fault identification result. The fault cause determination module is used to determine the cause of transformer faults based on the similarity between the fault discrimination results of the three-ratio method and the fault discrimination results of the neural network.

[0012] Thirdly, the present invention also provides an electronic device, comprising: a processor and a memory; The memory stores a computer-readable program that can be executed by the processor; When the processor executes the computer-readable program, it implements the steps in the data processing-based transformer fault diagnosis method described above.

[0013] Fourthly, the present invention also provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the data processing-based transformer fault diagnosis method described above.

[0014] Compared with existing technologies, the transformer fault diagnosis method, apparatus, and equipment based on data processing provided by this invention first acquire real-time characteristic data of transformer oil chromatograms. Then, based on a preset auxiliary feature vector and a preset FCM-iForest algorithm, abnormal data detection is performed on the real-time characteristic data, and invalid abnormal data is removed to obtain target characteristic data. By removing erroneous or interfering data and retaining valid data, errors in transformer fault diagnosis caused by invalid data can be avoided, thereby improving the accuracy of transformer fault diagnosis. Furthermore, a preset three-ratio method is used to perform fault discrimination on the target characteristic data, obtaining a three-ratio method fault discrimination result, and a preset neural network model is used to perform fault discrimination on the target characteristic data, obtaining a neural network fault discrimination result. By using two methods—the three-ratio method and the neural network model—to discriminate faults, the accuracy and reliability of fault diagnosis are improved. Finally, the cause of the transformer fault is determined based on the similarity between the three-ratio method fault discrimination result and the neural network fault discrimination result. This invention improves the effective utilization rate of transformer characteristic data by processing it, thereby improving the accuracy of determining transformer faults through characteristic data. Furthermore, by using two methods to identify faults, the accuracy of fault identification is improved. Attached Figure Description

[0015] Figure 1 This is a flowchart of an embodiment of the transformer fault diagnosis method based on data processing provided by the present invention; Figure 2 This is a schematic diagram of an embodiment of step S102 in the transformer fault diagnosis method based on data processing provided by the present invention; Figure 3 This is a schematic diagram of an embodiment of step S202 in the transformer fault diagnosis method based on data processing provided by the present invention; Figure 4 This is a schematic diagram of an embodiment of the transformer fault diagnosis device based on data processing provided by the present invention; Figure 5 This is a schematic diagram of the operating environment of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] This invention provides a transformer fault diagnosis method based on data processing. Please refer to [link to relevant documentation]. Figure 1 ,include: S101. Obtain real-time data of characteristic quantities from transformer oil chromatography; S102. Based on the preset auxiliary feature vector and the preset FCM-iForest algorithm, perform abnormal data detection on the real-time data of the feature quantity, and remove invalid abnormal data to obtain the target feature quantity data. S103. The target feature data is subjected to fault identification using a preset three-ratio method to obtain the fault identification result of the three-ratio method, and the target feature data is subjected to fault identification using a preset neural network model to obtain the neural network fault identification result. S104. Based on the similarity between the fault discrimination results of the three-ratio method and the fault discrimination results of the neural network, determine the cause of the transformer fault.

[0018] In this embodiment, real-time characteristic data of transformer oil chromatography are first acquired. Then, based on a preset auxiliary feature vector and a preset FCM-iForest algorithm, abnormal data detection is performed on the real-time characteristic data, and invalid abnormal data is removed to obtain target characteristic data. By removing erroneous or interfering data and retaining valid data, errors in transformer fault diagnosis caused by invalid data can be avoided, thereby improving the accuracy of transformer fault diagnosis. Furthermore, a preset three-ratio method is used to perform fault discrimination on the target characteristic data, obtaining a three-ratio method fault discrimination result. A preset neural network model is also used to perform fault discrimination on the target characteristic data, obtaining a neural network fault discrimination result. By using two methods—the three-ratio method and the neural network model—to discriminate faults, the accuracy and reliability of fault diagnosis are improved. Finally, based on the similarity between the three-ratio method fault discrimination result and the neural network fault discrimination result, the cause of the transformer fault is determined. This invention improves the effective utilization rate of transformer characteristic data by processing it, thereby increasing the accuracy of determining transformer faults through characteristic data. Furthermore, by using two methods to identify faults, the accuracy of fault identification is improved. It should be noted that outlier data, also known as outliers or anomalies, refers to outliers that are significantly different from the majority of data samples in some characteristic or behavior, or that deviate significantly from the normal pattern.

[0019] It should be noted that real-time characteristic data refers to various data related to the transformer's status and performance collected during transformer operation, including but not limited to transformer operation time information, hydrogen concentration, oxygen concentration, nitrogen concentration, carbon monoxide concentration, carbon dioxide concentration, acetylene concentration, ethylene concentration, ethane concentration, and methane concentration. These characteristic quantities are obtained through monitoring methods such as transformer oil chromatography analysis. Transformer oil is an important insulating and cooling medium in transformers, reflecting the internal operating status and potential faults. By acquiring this real-time characteristic data and combining it with the FCM-iForest algorithm, the three-ratio method, and neural network models, the transformer's operating status can be monitored and faults diagnosed, enabling preventative maintenance and fault prediction, thereby improving the reliability and stability of the power system.

[0020] It should be noted that the causes of transformer failure include: low energy density partial discharge, high energy density partial discharge, low energy discharge, high energy discharge, overheating below 150℃, overheating between 150℃ and 300℃, overheating between 300℃ and 700℃, and overheating above 700℃.

[0021] In some embodiments, the real-time feature data is analyzed based on a preset auxiliary feature vector and a preset FCM-iForest algorithm to detect abnormal data and remove invalid abnormal data, thereby obtaining the target feature data. (See also...) Figure 2 ,include: S201. The preset FCM-iForest algorithm is used to detect abnormal data in the real-time data of the feature quantity to obtain the first abnormal dataset. S202. Construct a transformer detection auxiliary dataset based on a preset auxiliary feature vector, and use a preset FCM-iForest algorithm to detect abnormal data in the transformer detection auxiliary dataset to obtain a second abnormal dataset. S203. Based on the comparison results between the first abnormal dataset and the second abnormal dataset, determine the valid abnormal data and the invalid abnormal data, and remove the invalid abnormal data to obtain the target feature data.

[0022] In this embodiment, the preset FCM-iForest algorithm is used to detect abnormal data in the real-time feature data to obtain the first abnormal dataset. The FCM-iForest algorithm is an algorithm that combines fuzzy C-means clustering (FCM) and isolated forest (iForest), which can effectively identify abnormal data points and is used to remove noise and outliers in the real-time feature data.

[0023] A transformer detection auxiliary dataset is constructed based on preset auxiliary feature vectors, and the preset FCM-iForest algorithm is used to detect anomalies in this auxiliary dataset, resulting in a second anomaly dataset. The preset auxiliary feature vectors refer to mapping transformer feature data to a high-dimensional space to expand the feature space, improve feature representation capabilities, and enhance anomaly detection performance.

[0024] Therefore, the auxiliary dataset can more accurately reflect the transformer's state and performance, and is used to further detect outliers and noise in the real-time feature quantity data. Finally, based on the comparison results between the first and second outlier datasets, valid and invalid outlier data are identified, and invalid outlier data is removed to obtain the target feature quantity data. By comparing the outliers in the two outlier datasets, it can be determined which outliers are valid outlier data related to the transformer's true state and which outlier data are invalid outlier data related to measurement errors or other external factors. Based on the comparison results, invalid outlier data can be removed, thus obtaining more accurate and reliable target feature quantity data.

[0025] Furthermore, based on the preset auxiliary feature vector and the preset FCM-iForest algorithm, outliers and noise in the real-time feature data can be effectively removed, improving the quality and reliability of the feature data.

[0026] In step S201, the original data is first standardized using the z-score method to eliminate computational errors caused by data of different types and orders of magnitude. An auxiliary feature vector is then used to replace the original data for anomaly detection. The expression for the auxiliary feature vector is as follows: ,in After processing using the z-score method, the first Class state quantity The measurements of each sample. There are 9 categories of state quantities, namely hydrogen concentration, oxygen concentration, nitrogen concentration, carbon monoxide concentration, carbon dioxide concentration, acetylene concentration, ethylene concentration, ethane concentration, and methane concentration.

[0027] Furthermore, Taking the hydrogen concentration in transformer operation and maintenance data as an example, the differential change rate is calculated as follows:

[0028] Furthermore, This is a warning value. When the measured value of a status variable exceeds its warning value, a fault alarm will be triggered. The warning value is calculated as follows:

[0029] in It is the first The warning value for a gas category is set as follows: when the measured value is lower than the warning value for that gas, the output is the measured value itself; when the measured value is higher than the warning value, the output is zero. The warning value can be specified based on the abnormal range of each type of characteristic gas data. Generally, it is 200 ppm for hydrogen; 100 ppm for ethane; 50 ppm for ethylene; and 100 ppm for methane.

[0030] It should be noted that the z-score method, taking the hydrogen concentration in transformer operation and maintenance data as an example, involves the following specific steps: In the formula This represents the average hydrogen concentration from the transformer operation and maintenance data. The standard deviation of hydrogen concentration, The first in the hydrogen concentration data There are 1 sample, and a total of m samples.

[0031] In some embodiments, the transformer detection auxiliary dataset is constructed based on a preset auxiliary feature vector, and anomaly detection is performed on the transformer detection auxiliary dataset using a preset FCM-iForest algorithm to obtain a second abnormal dataset. (See [link to relevant documentation]). Figure 3 ,include: S301. The transformer detection auxiliary dataset is clustered using the preset FCM-iForest algorithm to obtain the clustering results of the transformer detection auxiliary dataset; S302. Based on the clustering results, anomaly detection is performed on different cluster data using a preset isolated forest method to obtain a second abnormal dataset.

[0032] It should be noted that, compared to directly using the FCM-iForest algorithm, constructing an auxiliary dataset for transformer detection based on auxiliary feature vectors can better reflect the internal state and performance of the transformer, improving the robustness and reliability of the data. Furthermore, by performing anomaly detection on different clusters of data, outliers and noise in the transformer feature data can be identified more accurately, thereby improving the accuracy and reliability of anomaly detection.

[0033] Furthermore, the FCM algorithm is used to cluster the feature dataset. The calculation steps are as follows: (1) Let the dataset be Based on similarity Divided into There are several clusters, each cluster center is represented by a matrix. Representation. Partitionable matrix. It means that among them satisfy:

[0034] The objective function of the FCM algorithm is:

[0035] in, is the Euclidean distance, and m is the fuzziness index, which is generally taken as 2.

[0036] (2) Fixed iteration termination threshold Number of clusters Randomly initialize cluster centers .

[0037] (3) Update membership ; membership degree The membership function iterative formula is solved using the Lagrange multiplier method:

[0038] (4) Use Update cluster center ; The cluster center The iterative formula for the cluster central function is solved using the Lagrange multiplier method:

[0039] (5) If If yes, then output the result; otherwise, let Return to step S203 and continue the iteration.

[0040] Finally, based on the clustering results, the isolated forest method was used to detect abnormal data in different clusters, resulting in a second abnormal dataset based on auxiliary feature vectors.

[0041] In some embodiments, determining valid and invalid abnormal data based on the comparison result between the first abnormal dataset and the second abnormal dataset includes: By comparing the first abnormal dataset and the second abnormal dataset, common abnormal data and non-common abnormal data in the first abnormal dataset and the second abnormal dataset are determined. The common anomalous data is defined as the valid anomalous data, and the non-common anomalous data is defined as the invalid anomalous data.

[0042] In this embodiment, by comparing the first abnormal dataset and the second abnormal dataset, and distinguishing between common abnormal data and non-common abnormal data, it can be ensured that the obtained target feature data has high quality and reliability.

[0043] Such data cleaning and filtering processes help to accurately analyze the condition and performance of transformers, thereby supporting transformer fault diagnosis and prediction.

[0044] In some embodiments, after removing the invalid abnormal data, the method further includes: A preset linear interpolation method is used to fill in the missing values ​​after removing the invalid and abnormal data.

[0045] In this embodiment, linear interpolation is a commonly used interpolation method to predict missing values ​​based on the known linear relationship between data points. During data imputation, linear interpolation fills in the missing values ​​according to the known linear trend between data points. By imputing the dataset after removing invalid and outlier data using a preset linear interpolation method, the integrity of the data can be restored, making the dataset usable in subsequent analysis and processing.

[0046] It is understood that in other embodiments, other interpolation methods, such as polynomial interpolation and spline interpolation, may be considered, with the appropriate method selected based on the specific circumstances. Data completion aims to avoid losing valuable data and ensure the integrity of the dataset, providing a more accurate basis for subsequent analysis and decision-making.

[0047] In some embodiments, determining the cause of a transformer fault based on the similarity between the fault identification result of the three-ratio method and the fault identification result of the neural network includes: If the fault identification result of the three-ratio method is consistent with the fault identification result of the neural network, the cause of the transformer fault is determined and an alarm is triggered for the transformer fault.

[0048] In this embodiment, by comprehensively comparing different fault identification methods, the accuracy and reliability of fault identification can be improved. If the results of the three-ratio method and the neural network identification are consistent, the cause of the transformer fault can be determined more reliably, and corresponding maintenance measures can be taken in a timely manner. This approach of combining multiple fault identification algorithms helps to improve the accuracy and efficiency of transformer fault diagnosis.

[0049] The three-ratio method was applied to determine transformer faults, analyzing the concentrations of various gases in the abnormal data. The principle of the three-ratio method is to determine the transformer fault type by analyzing the relative concentrations and temperatures of the characteristic gases H2, C2H2, C2H4, C2H6, and CH4 produced after a transformer fault. This is done by combining two gases with similar solubility and diffusion coefficients in three groups, and then encoding these ratios. The C2H2 / C2H4, CH4 / H2, and C2H4 / C2H6 ratios were calculated, and the fault results were compared with the fault type comparison chart (see Table 1) and the existing coding rule chart (see Table 2) to determine the cause of the transformer anomaly.

[0050] Table 1: Comparison Table of Fault Type Judgment Methods

[0051] Table 2: Gas Ratio Code Table for the Three Ratio Method

[0052] Furthermore, the reliability of the three-ratio method for determining the cause of transformer faults is tested using a BP neural network. A pre-trained neural network is employed, and the Duval method is used to detect CH4, C2H2, and C2H4 in the abnormal dataset. The Duval method's diagnostic model input vector consists of the relative percentages of the three characteristic gas concentrations. Input feature data vector: Based on the output of the BP neural network, the transformer fault is determined.

[0053] Finally, the transformer fault results based on the BP neural network are compared with those based on the three-ratio method to determine the final cause of the fault. In this comparison method, if the three-ratio method and the BP neural network method determine the same cause of the transformer fault, an alarm is triggered for the corresponding fault operation. Optionally, if the three-ratio method and the BP neural network method determine different causes of the transformer fault, a suspected transformer fault warning is issued.

[0054] In some embodiments, after acquiring the real-time characteristic data of transformer oil chromatography, the method further includes: A preset random forest algorithm is used to fill in the missing data in the real-time data of the feature quantities.

[0055] In this embodiment, the interpolation method uses the random forest algorithm, with the first... Taking class variables as an example, the first The observed values ​​in the class variable are denoted as Missing values ​​are denoted as In the Other variables with observed values ​​on class variables (except the first) Other variables besides class variables). In the... Other variables with missing values ​​in the class variable (except the first) (variables other than class variables).

[0056] The first All individuals with no missing values ​​in the class variable are used as training samples, and those with missing values ​​are used as test samples. The random forest imputation process is as follows: (1) First apply variables and independent variable Fit a random forest.

[0057] (2) As feature variables input, a trained random forest is used to predict missing values. The data matrix is ​​obtained after prediction and interpolation. ; (3) Repeat the interpolation process until it meets the standard. That is, the newly interpolated data matrix and the previous data matrix When the difference first begins to increase.

[0058] The missing values ​​in the acquired data are filled in using the random forest method described above, resulting in a complete dataset after imputation.

[0059] Based on the above-described data-processing-based transformer fault diagnosis method, this embodiment of the invention also provides a data-processing-based transformer fault diagnosis device 400. Please refer to [link to relevant documentation]. Figure 4 The data processing-based transformer fault diagnosis device 400 includes an acquisition module 410, a target feature quantity determination module 420, a fault discrimination module 430, and a fault cause determination module 440.

[0060] The acquisition module 410 is used to acquire real-time data of characteristic quantities in transformer oil chromatography. The target feature quantity determination module 420 is used to perform abnormal data detection on the real-time data of the feature quantity based on the preset auxiliary feature vector and the preset FCM-iForest algorithm, and remove invalid abnormal data to obtain the target feature quantity data. The fault identification module 430 is used to perform fault identification on the target feature data using a preset three-ratio method to obtain a fault identification result using the three-ratio method, and to perform fault identification on the target feature data using a preset neural network model to obtain a neural network fault identification result. The fault cause determination module 440 is used to determine the cause of transformer faults based on the similarity between the fault discrimination results of the three-ratio method and the fault discrimination results of the neural network.

[0061] like Figure 5As shown, based on the above-described data processing-based transformer fault diagnosis method, the present invention also provides an electronic device, which can be a mobile terminal, desktop computer, laptop, handheld computer, server, or other computing electronic device. The electronic device includes a processor 510, a memory 520, and a display 530. Figure 5 Only some components of the electronic device are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0062] In some embodiments, memory 520 may be an internal storage unit of the electronic device, such as a hard disk or memory. In other embodiments, memory 520 may be an external storage device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Furthermore, memory 520 may include both internal and external storage devices. Memory 520 is used to store application software and various types of data installed on the electronic device, such as program code installed on the electronic device. Memory 520 may also be used to temporarily store data that has been output or will be output. In one embodiment, memory 520 stores a data processing-based transformer fault diagnosis program 540, which can be executed by processor 510 to implement the data processing-based transformer fault diagnosis method of the embodiments of this application.

[0063] In some embodiments, processor 510 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 520 or process data, such as executing a transformer fault diagnosis method based on data processing.

[0064] Display 530 may, in some embodiments, be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. Display 730 is used to display information from the data-processing-based transformer fault diagnosis electronic equipment and to display a visual user interface. Components 510-530 of the electronic equipment communicate with each other via a system bus.

[0065] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The storage medium can be a memory, magnetic disk, optical disk, etc.

[0066] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A transformer fault diagnosis method based on data processing, characterized in that, include: Acquire real-time data of characteristic quantities in transformer oil chromatography; Based on the preset auxiliary feature vector and the preset FCM-iForest algorithm, abnormal data detection is performed on the real-time data of the feature quantity, and invalid abnormal data is removed to obtain the target feature quantity data. The target feature data is subjected to fault identification using a preset three-ratio method to obtain a fault identification result using the three-ratio method, and the target feature data is subjected to fault identification using a preset neural network model to obtain a neural network fault identification result. The cause of the transformer fault is determined based on the similarity between the fault identification results of the three-ratio method and the fault identification results of the neural network. The method of detecting abnormal data and removing invalid abnormal data based on a preset auxiliary feature vector and a preset FCM-iForest algorithm to obtain target feature data includes: The pre-defined FCM-iForest algorithm is used to detect abnormal data in the real-time data of the feature quantity to obtain the first abnormal dataset. A transformer detection auxiliary dataset is constructed based on a preset auxiliary feature vector. Anomalies are detected in the transformer detection auxiliary dataset using a preset FCM-iForest algorithm to obtain a second abnormal dataset. Based on the comparison results between the first abnormal dataset and the second abnormal dataset, valid abnormal data and invalid abnormal data are determined, and the invalid abnormal data is removed to obtain the target feature data. The expression for the preset auxiliary feature vector is as follows: ,in After processing using the z-score method, the first Class state quantity Measurement values ​​of a sample; The difference rate of change This is a warning value.

2. The transformer fault diagnosis method based on data processing according to claim 1, characterized in that, The transformer detection auxiliary dataset is constructed based on a preset auxiliary feature vector. Anomaly detection is performed on the transformer detection auxiliary dataset using a preset FCM-iForest algorithm to obtain a second abnormal dataset, including: The transformer detection auxiliary dataset is clustered using the preset FCM-iForest algorithm to obtain the clustering results of the transformer detection auxiliary dataset; Based on the clustering results, a pre-defined isolated forest method is used to detect anomalies in different clusters of data, resulting in a second abnormal dataset.

3. The transformer fault diagnosis method based on data processing according to claim 1, characterized in that, The step of determining valid and invalid abnormal data based on the comparison results between the first abnormal dataset and the second abnormal dataset includes: By comparing the first abnormal dataset and the second abnormal dataset, common abnormal data and non-common abnormal data in the first abnormal dataset and the second abnormal dataset are determined. The common anomalous data is defined as the valid anomalous data, and the non-common anomalous data is defined as the invalid anomalous data.

4. The transformer fault diagnosis method based on data processing according to claim 1, characterized in that, After removing the invalid and abnormal data, the process also includes: A preset linear interpolation method is used to fill in the missing values ​​after removing the invalid and abnormal data.

5. The transformer fault diagnosis method based on data processing according to claim 1, characterized in that, The step of determining the cause of transformer faults based on the similarity between the fault discrimination results of the three-ratio method and the fault discrimination results of the neural network includes: If the fault identification result of the three-ratio method is consistent with the fault identification result of the neural network, the cause of the transformer fault is determined and an alarm is triggered for the transformer fault.

6. The transformer fault diagnosis method based on data processing according to claim 1, characterized in that, After obtaining the real-time characteristic data of transformer oil chromatography, the process also includes: A preset random forest algorithm is used to fill in the missing data in the real-time data of the features.

7. A transformer fault diagnosis device based on data processing, characterized in that, include: The acquisition module is used to acquire real-time data of characteristic quantities in transformer oil chromatography. The target feature quantity determination module is used to perform abnormal data detection on the real-time data of the feature quantity based on the preset auxiliary feature vector and the preset FCM-iForest algorithm, and remove invalid abnormal data to obtain the target feature quantity data. The fault identification module is used to identify faults in the target feature data using a preset three-ratio method to obtain a fault identification result using the three-ratio method, and to identify faults in the target feature data using a preset neural network model to obtain a neural network fault identification result. The fault cause determination module is used to determine the cause of transformer faults based on the similarity between the fault discrimination results of the three-ratio method and the fault discrimination results of the neural network; the step of performing abnormal data detection on the real-time feature data based on the preset auxiliary feature vector and the preset FCM-iForest algorithm, and removing invalid abnormal data to obtain target feature data, includes: The pre-defined FCM-iForest algorithm is used to detect abnormal data in the real-time data of the feature quantity to obtain the first abnormal dataset. A transformer detection auxiliary dataset is constructed based on a preset auxiliary feature vector. Anomalies are detected in the transformer detection auxiliary dataset using a preset FCM-iForest algorithm to obtain a second abnormal dataset. Based on the comparison results between the first abnormal dataset and the second abnormal dataset, valid abnormal data and invalid abnormal data are determined, and the invalid abnormal data is removed to obtain the target feature data. The expression for the preset auxiliary feature vector is as follows: ,in After processing using the z-score method, the first Class state quantity Measurement values ​​of a sample; The difference rate of change This is a warning value.

8. An electronic device, characterized in that, include: Processor and memory; The memory stores a computer-readable program that can be executed by the processor; When the processor executes the computer-readable program, it implements the steps in the data processing-based transformer fault diagnosis method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, which can be executed by one or more processors to implement the steps in the data processing-based transformer fault diagnosis method as described in any one of claims 1-6.

Images