Transformer fault diagnosis method and system

Abstract

The invention relates to the technical field of transformers, and particularly discloses a transformer fault diagnosis method and system.The transformer fault diagnosis method comprises the steps that firstly, concentration data of gas dissolved in transformer oil is collected, the potential fault type of a transformer is preliminarily determined based on a three-ratio method, and a diagnosis result confidence evaluation mechanism is introduced; a confidence score for the potential fault type is calculated. And when the confidence coefficient is lower than a preset threshold value, further combining the potential fault type and the confidence coefficient thereof with basic attribute data of the transformer to form fault diagnosis priori data. And then, deep conjoint analysis is performed on the concentration data of the gas dissolved in the oil and the fault diagnosis prior data by using the trained fault diagnosis model, so that more comprehensive fault features can be captured, and an accurate transformer fault diagnosis result can be obtained. According to the method, a deep learning algorithm is introduced during low-confidence diagnosis of a traditional three-ratio method, so that the accuracy and reliability of transformer fault diagnosis can be remarkably improved.

Description

Technical Field

[0001] The present application relates to the technical field of transformers, and more specifically, to a transformer fault diagnosis method and system. Background Art

[0002] As a core component of the power system, the stable and reliable operation of power transformers is directly related to the safe and economical operation of the entire power grid. However, during long-term operation, transformers are inevitably susceptible to various latent faults due to the combined effects of various stresses, including electrical, thermal, and mechanical stresses. Failure to promptly detect and accurately diagnose these faults can lead to escalation or even serious power outages, resulting in significant economic losses and adverse social impacts. Therefore, researching and developing efficient transformer fault diagnosis solutions is crucial for ensuring the safe and stable operation of power systems, improving power supply reliability, and optimizing equipment operation and maintenance strategies.

[0003] Currently, dissolved gas analysis (DGA) is one of the most widely used and effective methods for diagnosing latent faults within transformers. Traditional DGA data interpretation methods, such as the IEC three-ratio method, analyze the concentrations or ratios of specific gas components such as H2, CH4, C2H2, C2H4, and C2H6, and compare them with preset fault type coding rules to determine the health of the transformer and the type of potential fault. However, the traditional IEC three-ratio method can only output discrete fault codes and cannot quantify the reliability of the diagnostic results. It has significant limitations when the gas ratios approach the critical values ​​of the interval. For example, slight measurement errors near the ratio boundaries can cause the fault type to jump, resulting in frequent misjudgments or missed detections in practical applications. Existing technologies lack a dynamic assessment mechanism for the confidence level of diagnostic results, which limits the accuracy and reliability of diagnostic results.

[0004] Therefore, an optimized transformer fault diagnosis method and system are expected. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. An embodiment of the present application provides a transformer fault diagnosis method and system, which first collects the dissolved gas concentration data in the transformer oil, preliminarily determines the potential fault type of the transformer based on the three-ratio method, and introduces a diagnosis result confidence evaluation mechanism to calculate the confidence score of the potential fault type. When this confidence is lower than the preset threshold, the above potential fault type and its confidence are further combined with the basic attribute data of the transformer to form fault diagnosis prior data. Subsequently, the dissolved gas concentration data in the oil and the fault diagnosis prior data are deeply jointly analyzed by using the trained fault diagnosis model to capture more comprehensive fault characteristics, thereby obtaining accurate transformer fault diagnosis results. This method can significantly improve the accuracy and reliability of transformer fault diagnosis by introducing a deep learning algorithm in the low-confidence diagnosis of the traditional three-ratio method.

[0006] According to one aspect of the present application, a transformer fault diagnosis method is provided, which includes: Collecting dissolved gas concentration data in transformer oil by online gas chromatograph, wherein the dissolved gases in the oil include H2, CH4, C2H2, C2H4, and C2H6; Processing the dissolved gas concentration data in the transformer oil based on a three-ratio method to determine a potential fault type of the transformer and a confidence score thereof; In response to the confidence score being lower than a preset threshold, obtaining basic attribute data of the transformer, and adding the potential fault type of the transformer and its confidence score to the basic attribute data to obtain transformer fault diagnosis prior data; The dissolved gas concentration data in the transformer oil and the transformer fault diagnosis priori data are input into the trained fault diagnosis model to obtain the fault diagnosis result of the transformer.

[0007] According to another aspect of the present application, a transformer fault diagnosis system is provided, comprising: A gas concentration data acquisition module is used to collect dissolved gas concentration data in transformer oil through an online gas chromatograph, wherein the dissolved gases in the oil include H2, CH4, C2H2, C2H4, and C2H6; a fault pre-diagnosis module, configured to process the dissolved gas concentration data in the transformer oil based on a three-ratio method to determine a potential fault type of the transformer and a confidence score thereof; a priori data generating module, configured to, in response to the confidence score being lower than a preset threshold, obtain basic attribute data of the transformer, and add the potential fault type of the transformer and its confidence score to the basic attribute data to obtain priori data for transformer fault diagnosis; The fault diagnosis result generating module is used to input the dissolved gas concentration data in the transformer oil and the transformer fault diagnosis priori data into the trained fault diagnosis model to obtain the fault diagnosis result of the transformer.

[0008] Compared with the prior art, the transformer fault diagnosis method and system provided by the present application first collects the dissolved gas concentration data in the transformer oil, preliminarily determines the potential fault type of the transformer based on the three-ratio method, and introduces a diagnostic result confidence assessment mechanism to calculate the confidence score of the potential fault type. When this confidence is lower than the preset threshold, the above-mentioned potential fault type and its confidence are further combined with the basic attribute data of the transformer to form fault diagnosis prior data. Subsequently, the dissolved gas concentration data in the oil and the fault diagnosis prior data are deeply jointly analyzed by using the trained fault diagnosis model to capture more comprehensive fault characteristics, thereby obtaining accurate transformer fault diagnosis results. This method can significantly improve the accuracy and reliability of transformer fault diagnosis by introducing a deep learning algorithm in the low-confidence diagnosis of the traditional three-ratio method. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 Flowchart of a transformer fault diagnosis method according to an embodiment of the present application.

[0011] Figure 2 4 is a data flow diagram of a transformer fault diagnosis method according to an embodiment of the present application.

[0012] Figure 3 Flowchart of sub-step S2 of the transformer fault diagnosis method according to an embodiment of the present application.

[0013] Figure 4 This is a flowchart of sub-step S4 of the transformer fault diagnosis method according to an embodiment of the present application.

[0014] Figure 5 4 is a flowchart of sub-step S43 of the transformer fault diagnosis method according to an embodiment of the present application.

[0015] Figure 6 4 is a block diagram of a transformer fault diagnosis system according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0017] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0018] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0019] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0020] It is worth noting that in this application, all actions to obtain data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0021] In response to the technical problems described in the above background technology, the present application proposes a transformer fault diagnosis method, which first collects the dissolved gas concentration data in the transformer oil, preliminarily determines the potential fault type of the transformer based on the three-ratio method, and introduces a diagnosis result confidence assessment mechanism to calculate the confidence score of the potential fault type. When this confidence is lower than the preset threshold, the above potential fault type and its confidence are further combined with the basic attribute data of the transformer to form fault diagnosis prior data. Subsequently, the dissolved gas concentration data in the oil and the fault diagnosis prior data are deeply jointly analyzed by using the trained fault diagnosis model to capture more comprehensive fault characteristics, thereby obtaining accurate transformer fault diagnosis results. This method can significantly improve the accuracy and reliability of transformer fault diagnosis by introducing a deep learning algorithm in the low-confidence diagnosis of the traditional three-ratio method.

[0022] Figure 1 Flowchart of a transformer fault diagnosis method according to an embodiment of the present application. Figure 2 Figure 1 is a data flow diagram of a transformer fault diagnosis method according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the transformer fault diagnosis method includes the following steps: S1, collecting dissolved gas concentration data in transformer oil through an online gas chromatograph, wherein the dissolved gases in the oil include H2, CH4, C2H2, C2H4, and C2H6; S2, processing the dissolved gas concentration data in the transformer oil based on a three-ratio method to determine the potential fault type of the transformer and its confidence score; S3, in response to the confidence score being lower than a preset threshold, obtaining basic attribute data of the transformer, and adding the potential fault type of the transformer and its confidence score to the basic attribute data to obtain transformer fault diagnosis prior data; S4, inputting the dissolved gas concentration data in the transformer oil and the transformer fault diagnosis prior data into a trained fault diagnosis model to obtain a fault diagnosis result of the transformer.

[0023] In the above-mentioned transformer fault diagnosis method, step S1 collects dissolved gas concentration data in transformer oil using an online gas chromatograph, where the dissolved gases in the oil include H2, CH4, C2H2, C2H4, and C2H6. It should be understood that during transformer operation, the insulating oil and solid insulation materials within the transformer undergo varying degrees of decomposition under electrical and thermal stresses, resulting in the generation of specific dissolved gases in the oil. The type and content of these gases are key indicators reflecting the presence and nature of latent faults within the transformer. H2, CH4, C2H2, C2H4, and C2H6 are generally recognized as core characteristic gases closely associated with major fault types (such as discharge and overheating). Therefore, to obtain the most direct and fundamental raw characterization data for diagnosing internal transformer fault conditions, this application, based on the fundamental principle of dissolved gas analysis (DGA), which states that specific faults generate specific gas combinations, uses an online gas chromatograph to automatically monitor and quantify the concentrations of five key dissolved gases—H2, CH4, C2H2, C2H4, and C2H6—in transformer oil, enabling accurate and timely collection of DGA data. Specifically, an online gas chromatograph typically integrates an oil-gas separation device, an injection system, a chromatographic separation column, and a high-sensitivity detector (such as a thermal conductivity detector (TCD) and a flame ionization detector (FID). During operation, the instrument automatically draws a certain amount of insulating oil sample from the transformer oil line, extracts the gases dissolved in the oil using a built-in oil-gas separation unit (such as a headspace, bubbling, or permeation membrane method), and then injects the gas sample into the chromatographic column. Due to their varying physical and chemical properties, the different gas components have different retention times in the chromatographic column, enabling separation. The separated gas components sequentially enter the detector, which converts their concentrations into corresponding electrical signals. The data processing system then calculates the precise concentration of each gas component (typically in μL / L or ppm). This method continuously and automatically acquires concentration data for the five key gases that reflect the transformer's internal operating status. This provides a dynamic and reliable data source for subsequent preliminary fault diagnosis and in-depth diagnosis, and lays a solid foundation for early warning and analysis of fault development trends.

[0024] In practice, an online gas chromatograph is typically installed in the transformer's oil system, extracting dissolved gas samples from the insulating oil in real time or at a set interval. The instrument integrates key components such as the oil-gas separation device, sample injection system, chromatographic column, and detector. The oil-gas separation unit is the core module that ensures efficient and accurate gas extraction. Common oil-gas separation methods include headspace, bubbling, and permeable membrane methods. Each method has its own advantages in terms of gas recovery rate, response speed, and stability, and is suitable for different operating environments and working conditions. For example, the headspace method separates oil and gas by injecting transformer oil into a sealed container and maintaining a constant temperature, allowing dissolved gases to gradually diffuse into the gas phase. The bubbling method uses an inert gas as a carrier gas, which is introduced into the oil sample to form bubbles, prompting the rapid release of dissolved gases into the gas phase. The permeable membrane method utilizes a selectively permeable membrane material, allowing gas molecules to penetrate the membrane and enter the detection system, while effectively isolating the liquid. All of these methods achieve high gas recovery rates while reducing interference introduced by the operating process, ensuring the authenticity and representativeness of the obtained gas samples.

[0025] After oil and gas separation, the gas sample is introduced into the injection system and propelled by a carrier gas (such as helium or nitrogen) into the chromatographic column for separation. The chromatographic column is filled with a specific stationary phase material, whose physicochemical properties determine the different retention times of different gas components. Because the five gases—H2, CH4, C2H2, C2H4, and C2H6—distinctly differ in molecular weight, polarity, and adsorption characteristics, their migration velocities within the chromatographic column also vary. After separation along a certain length of the chromatographic column, the gases sequentially exit the column end and enter the detector. The detector typically uses a thermal conductivity detector (TCD) or a flame ionization detector (FID). The former has excellent response to inorganic gases such as hydrogen, while the latter offers higher sensitivity and selectivity for hydrocarbon gases. The coordinated use of these two detectors enables comprehensive detection of the five key gases, ensuring accurate identification and quantification of all target gases.

[0026] When gas components pass through the detector, their concentration information is converted into an electrical signal output, which is then amplified, filtered, and digitized by the built-in data processing module. The system compares the standard gas curves, combines correction factors with a baseline subtraction algorithm, and calculates the specific concentration of each gas, typically expressed in μL / L or ppm. This value not only reflects the gas distribution within the transformer at the current moment but also provides a direct basis for subsequent three-ratio analysis. To further improve measurement accuracy, some advanced online chromatography systems are equipped with an automatic calibration function, which regularly calibrates the detector using standard gases of known concentrations to eliminate errors caused by sensor drift or environmental changes.

[0027] Furthermore, online gas chromatographs possess strong environmental adaptability and long-term operational stability. Considering that power transformers often operate in complex environments characterized by high temperature, high humidity, and electromagnetic interference, the equipment was designed with full consideration given to its protection level, anti-interference capabilities, and heat dissipation performance. The outer casing is typically constructed of explosion-proof and corrosion-resistant materials, and the internal circuit boards undergo special packaging to protect against the effects of harsh external conditions. The system also supports remote monitoring and fault self-diagnosis. Operations and maintenance personnel can view the equipment's operating status, historical data records, and alarm information in real time through the backend management system. If abnormal fluctuations or hardware failures are detected, maintenance measures can be taken promptly to avoid impacting the reliability of the overall diagnostic system.

[0028] The application of online gas chromatographs is not limited to the detection of single gas components. More importantly, by continuously and stably collecting concentration data for five gases—H2, CH4, C2H2, C2H4, and C2H6—they construct a dynamic data chain reflecting the internal operating status of the transformer. The generation and accumulation of these gases are influenced by various factors, including various fault modes such as partial discharge, overheating, and arcing. For example, hydrogen (H2) is often produced in large quantities during corona discharge or partial discharge, while ethylene (C2H4) and acetylene (C2H2) are more often associated with high-temperature overheating or arcing. Methane (CH4) and ethane (C2H6) may occur in slow pyrolysis reactions at lower temperatures. Therefore, by continuously tracking the concentration trends of these five gases, it is possible to preliminarily determine whether the transformer is experiencing abnormal conditions and the likely fault type, providing a solid data foundation for subsequent in-depth analysis.

[0029] In the above-mentioned transformer fault diagnosis method, the step S2 processes the dissolved gas concentration data in the transformer oil based on the three-ratio method to determine the potential fault type of the transformer and its confidence score. It should be understood that the IEC three-ratio method, as a widely used DGA data interpretation method, can classify faults into different coding areas according to the ratios between characteristic gases, thereby preliminarily judging the fault type. Therefore, in order to preliminarily determine the potential fault type of the transformer, the present application uses the three-ratio method to calculate the ratio of the concentrations of specific gas components in the transformer oil, and compares the calculation results with the preset fault type coding rules to obtain the potential fault type of the transformer. Among them, Figure 3 FIG. 1 is a flow chart of sub-step S2 of the transformer fault diagnosis method according to an embodiment of the present application. Figure 3As shown, step S2 includes the following steps: S21, calculating the IEC three ratios based on the dissolved gas concentration data in the transformer oil, wherein the IEC three ratios include R1=C2H2 / C2H4, R2=CH4 / H2, and R3=C2H4 / C2H6; S22, determining the fault code of the transformer based on the range interval of the IEC60599 standard and the IEC three ratios; S23, based on the fault code, looking up a table to obtain the potential fault type of the transformer; S24, calculating the confidence score of the potential fault type based on the normalized distance of each ratio in the IEC three ratios relative to the interval boundary of the range interval to which it belongs.

[0030] Specifically, step S21 calculates the IEC three ratios based on the dissolved gas concentration data in the transformer oil, wherein the IEC three ratios include R1=C2H2 / C2H4, R2=CH4 / H2, and R3=C2H4 / C2H6. Specifically, in order to establish a preliminary correlation between gas concentration and fault type, this application, based on the basic logic of fault diagnosis based on the IEC three-ratio method, divides the measured concentration values ​​of H2, CH4, C2H2, C2H4, and C2H6 according to the corresponding formula to obtain the three ratios R1, R2, and R3, respectively, to preliminarily establish a relationship between gas composition and fault type.

[0031] Specifically, step S22 determines the fault code of the transformer based on the range interval of the IEC60599 standard and the IEC three ratios. It should be understood that since the IEC60599 standard clearly stipulates the three ratio range intervals corresponding to different fault types, and it is an empirical standard summarized from long-term practice and research, it can provide a standardized basis for fault type judgment. Therefore, in order to convert the calculated IEC three ratios into a code that can intuitively judge the fault type, this application is based on the correspondence between the fault type and the three ratio range intervals specified in the IEC60599 standard. The calculated three ratios R1, R2, and R3 are compared with each range interval specified in the IEC60599 standard. According to the interval where each ratio is located, according to the coding rules set in the standard, the corresponding fault code is generated in combination. For example, if R1 = 0.0417, then the code corresponding to its range interval in the IEC60599 standard is 0. By classifying R1, R2, and R3 as a number among 0, 1, or 2 according to the rules, and combining them into a complete fault code, a preliminary classification of the fault type can be achieved.

[0032] Specifically, step S23 involves looking up a table based on the fault code to obtain the potential fault type of the transformer. That is, based on the generated fault code, by consulting the table of fault type and fault code correspondences preset in the IEC60599 standard, the potential fault type of the transformer can be quickly and preliminarily determined. For example, if the fault code is "121," the table retrieves the potential fault type corresponding to this code as "high temperature (above 700°C)." This approach not only simplifies the complex fault diagnosis process but also improves diagnostic efficiency and accuracy.

[0033] Specifically, step S24 calculates the confidence score for the potential fault type based on the normalized distance of each of the three IEC ratios relative to the interval boundary of the range to which it belongs. It should be understood that the traditional IEC three-ratio method can contain slight measurement errors when the gas ratio approaches the interval critical value, which can cause the fault type to jump, resulting in inaccurate diagnosis results. The existing technology lacks effective means for assessing the reliability of diagnostic results. Therefore, to quantitatively assess the confidence of the potential fault type of the transformer determined based on the IEC three-ratio method, the present application further calculates the confidence score for the potential fault type based on the normalized distance of each of the three IEC ratios relative to the interval boundary of the range to which it belongs. Specifically, for each ratio, the relative distance from the closer boundary of the range to which it belongs is first calculated, and then the relative distance is divided by the half-width of the range to which it belongs to obtain the normalized distance. For open intervals (e.g., "greater than 3.0"), a reasonable threshold is set as the boundary based on statistical analysis of historical transformer fault data, and corresponding normalization processing is performed. It should be understood that the closer the normalized distance is to 0, the closer the ratio is to the interval boundary, and the less robust the diagnostic result it represents. Conversely, the closer the normalized distance is to 1, the closer the ratio is to the center of the interval, and the more robust the diagnostic result it represents. By calculating the normalized distances of the three ratios R1, R2, and R3 relative to the interval boundary of the range to which they belong, and taking the average of the three normalized distances as the final confidence score, a quantitative assessment of the credibility of potential fault types determined based on the IEC three-ratio method is achieved.

[0034] In the above-mentioned transformer fault diagnosis method, in step S3, in response to the confidence score being lower than a preset threshold, the basic attribute data of the transformer is obtained, and the potential fault type of the transformer and its confidence score are added to the basic attribute data to obtain transformer fault diagnosis prior data. It should be understood that the higher the confidence score, the more consistent the potential fault type is with the actual fault condition, and the higher the reliability of the diagnosis result. On the contrary, if the confidence score is low, it means that the diagnosis result may be affected by measurement error or data fluctuation, and further analysis or verification using other diagnostic methods is required to ensure the accuracy and reliability of the fault diagnosis. Therefore, when the confidence score is lower than the preset threshold, the present application further compensates for the shortcomings of the DGA data by integrating the transformer's own characteristics and operating information, and uses the above-diagnosed potential fault type of the transformer and its confidence score as prior knowledge, and associates and integrates them with the basic attribute data of the transformer to form transformer fault diagnosis prior data containing multi-dimensional information, thereby providing a more comprehensive and accurate contextual reference basis for subsequent fault analysis.

[0035] In the above-mentioned transformer fault diagnosis method, in step S4, the dissolved gas concentration data in the transformer oil and the transformer fault diagnosis prior data are input into the trained fault diagnosis model to obtain the fault diagnosis result of the transformer. That is, in order to make full use of the acquired dissolved gas concentration data and fault diagnosis prior data, the present application adopts machine learning technology to construct a fault diagnosis model. Based on the powerful data fitting and feature extraction capabilities of the deep learning model, the inherent correlation and complementarity between the dissolved gas data in the oil and the transformer fault diagnosis prior data are deeply explored, thereby overcoming the limitations of traditional methods and achieving a more accurate and reliable final judgment on the transformer fault type. Among them, Figure 4 FIG. 4 is a flow chart of sub-step S4 of the transformer fault diagnosis method according to an embodiment of the present application. Figure 4 As shown, the step S4 includes the following steps: S41, performing multi-source information association coding on the dissolved gas concentration data in the transformer oil to obtain a multi-source dissolved gas concentration feature association coding vector; S42, performing semantic embedding coding on the transformer fault diagnosis prior data to obtain a transformer fault diagnosis prior data semantic coding vector; S43, performing semantic interactive response reasoning on the multi-source dissolved gas concentration feature association coding vector and the transformer fault diagnosis prior data semantic coding vector to obtain a transformer fault state comprehensive feature vector; S44, inputting the transformer fault state comprehensive feature vector into the trained fault diagnosis model to obtain the fault diagnosis result, which is used to represent the fault type of the transformer.

[0036] Specifically, in a specific example of the present application, step S41 includes inputting the transformer oil dissolved gas concentration data into a multi-source information association encoder based on a fully connected layer to obtain the multi-source dissolved gas concentration feature association encoding vector. It should be understood that the transformer oil dissolved gas concentration data contains multiple gas concentration information, which are interrelated, and different fault types will cause gas concentrations to exhibit specific combination patterns. Therefore, to fully explore the potential correlations between gas concentration data, the present application uses a fully connected layer to construct a multi-source information association encoder. By stacking a series of fully connected layers (including activation functions such as ReLU and normalization layers such as BatchNorm), the encoder learns the nonlinear relationships between the components within the gas concentration data and the joint indicative effect of multidimensional gas concentration data on fault types. This maps the original discrete multi-dimensional gas concentration data into a low-dimensional feature space, extracting feature information that is critical for fault diagnosis, and obtaining a more discriminative multi-source dissolved gas concentration feature association encoding vector.

[0037] Specifically, in a specific example of the present application, step S42 includes: inputting the transformer fault diagnosis prior data into a semantic embedding encoder based on the Bert model to obtain a semantic encoding vector of the transformer fault diagnosis prior data. It should be understood that since the transformer fault diagnosis prior data contains a variety of basic attribute data such as the rated capacity, load rate, and operating years of the transformer, as well as the potential fault type and confidence score of the preliminary diagnosis, it has different data types and feature dimensions, and there is a certain semantic association between them. Therefore, in order to convert the transformer fault diagnosis prior data into a form that can be understood and processed by the machine learning model, and fully explore the implicit semantic information and association relationships within it, the present application further utilizes the powerful semantic understanding and feature extraction capabilities of the Bert model in the field of natural language processing to perform semantic embedding encoding on the transformer fault diagnosis prior data to obtain a semantic encoding vector of the transformer fault diagnosis prior data. In the specific implementation process, first, the transformer fault diagnosis prior data is preprocessed, that is, various attribute data and diagnostic information are subjected to word segmentation and word embedding processing, and converted into the form of a word embedding vector sequence that can be processed by the Bert model. Then, the processed word embedding vector sequence is input into the pre-trained Bert model for deep semantic encoding. The bidirectional Transformer architecture of the Bert model captures the semantic associations and feature dependencies within the data, enhances the correlation and interpretability between multidimensional data, and generates a semantic encoding vector of transformer fault diagnosis prior data rich in semantic information, providing strong feature support for subsequent fault diagnosis reasoning.

[0038] Specifically, step S43 performs semantic interactive response reasoning on the multi-source dissolved gas concentration feature association encoding vector and the transformer fault diagnosis prior data semantic encoding vector to obtain a comprehensive transformer fault state feature vector. It should be understood that the multi-source dissolved gas concentration feature association encoding vector and the transformer fault diagnosis prior data semantic encoding vector each reflect transformer status information from different perspectives, and there is a certain inherent connection between the two. For example, as a transformer ages, its internal insulation material gradually ages, degrading insulation performance, which in turn increases the probability of faults such as partial discharge. At this time, the concentrations of dissolved gases H2 and C2H2 in the oil may also increase accordingly. For another example, if the transformer load factor is too high, excessive heat will be generated within the transformer, causing a high-temperature overheating fault, which manifests as a significant increase in the concentrations of gases such as CH4 and C2H4. Furthermore, different types of transformer insulating oils also have different gas generation characteristics under fault conditions, which directly affect the proportions of the various dissolved gas components. Therefore, in order to fully explore the potential semantic associations and synergistic effects between dissolved gas in oil and transformer fault diagnosis prior data, this application further performs deep semantic interactive response reasoning on the multi-source dissolved gas concentration feature association coding vector and the transformer fault diagnosis prior data semantic coding vector to explore the intrinsic connection and complementary information between the two, and generate a comprehensive feature representation that fully reflects the actual fault state of the transformer, that is, the transformer fault state comprehensive feature vector, thereby providing a more accurate feature description for subsequent fault diagnosis. Among them, Figure 5 FIG. 4 is a flow chart of sub-step S43 of the transformer fault diagnosis method according to an embodiment of the present application. Figure 5 As shown, the step S43 includes the steps of: S431, calculating the semantic response matrix of the multi-source dissolved gas concentration feature association coding vector relative to the transformer fault diagnosis prior data semantic coding vector to obtain the transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix; S432, performing matrix decomposition on the transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix to obtain the sequence distribution of the transformer fault attribute-dissolved gas local semantic response feature coding vector; S433, performing global semantic interactive reasoning on the sequence distribution of the transformer fault attribute-dissolved gas local semantic response feature coding vector to obtain the transformer fault state comprehensive feature vector.

[0039] More specifically, step S431 is expressed as follows:

[0040] in, represents the hyperbolic tangent function, Represents the characteristic correlation coding vector of multi-source dissolved gas concentration, represents the semantic encoding vector of transformer fault diagnosis prior data, represents the transpose of a vector, represents the sigmoid activation function, is the dynamic weight coefficient matrix, is the projection transformation matrix, Indicates point multiplication by position, It means adding by position. represents matrix multiplication, is the hyperbolic tangent asymmetric interaction, For element-level gated interactions, Represents the global fine-grained semantic response matrix of transformer fault attributes-dissolved gas characteristics.

[0041] That is, in order to deeply explore the semantic interaction relationship between the multi-source dissolved gas concentration feature association coding vector and the transformer fault diagnosis prior data semantic coding vector, by calculating the transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix of the two, all potential association strengths between the multi-source dissolved gas concentration feature association coding vector and the transformer fault diagnosis prior data semantic coding vector are fully captured, ensuring the integrity of the interactive analysis in a full-domain coverage manner, and at the same time accurately capturing subtle semantic differences and association resonances with fine-grained processing logic, thereby constructing an information map that can comprehensively reflect the mutual influence of gas characteristics and fault attributes, providing a high-density interactive information foundation for subsequent feature extraction and semantic reasoning based on matrix decomposition.

[0042] In particular, when calculating the global fine-grained semantic response matrix of transformer fault attributes-dissolved gas characteristics, due to the fine-grained correlation matrix Dynamic weighting coefficient matrix and the projection transformation matrix Perform multi-directional modulation operations, and the function and Having different power levels of exponential functions based on natural constants, while ensuring the integrity of global topological interactions, it will also stimulate differentiated dynamic coupling channels for fine-grained implicit semantic representations, thereby affecting the fusion effect of the two parts. Therefore, in order to achieve the equivalence of coupling channels, it is expected to ensure that the topological feature flow type is based on the topological irreducibility of heterogeneous coupling channels, thereby improving the multimodal fusion efficiency of implicit semantic encoding at fine granularity. In a preferred example of the present application, the step S432 includes: first, performing a dual-interaction path equivalence correction on the transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix to obtain an optimized transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix.

[0043] Specifically, first let:

[0044]

[0045] in, Represents the transformer fault attribute-dissolved gas characteristic hyperbolic tangent asymmetric interaction matrix, Represents the transformer fault attribute-dissolved gas feature cell-level gating interaction matrix.

[0046] Secondly, the relay state mapping matrix is ​​obtained through projection verification, which satisfies:

[0047] in, Represents the relay state mapping matrix.

[0048] That is, it is considered that different projection parts have an interactive relationship with each other, so that the relay state mapping matrix Based on different interaction orders, there are corresponding algebraic generating kernels and inverse element mappers.

[0049] Then, the relay state mapping matrix can be The ground state coherence degree of the dual interaction path is regulated, namely:

[0050] in, represents the inverse matrix, express norm, represents the correction matrix.

[0051] That is, the relay state mapping matrix and its inverse matrix As the group representation of the algebraic generating kernel and the inverse element mapper, respectively, through the matrix The norm is used to perform distribution state degeneration, that is, the group theory distribution state characteristic description is irreducibly degenerated to obtain the ground state coupling topological mapping relationship, so that the directed evolution of the algebraic generating kernel and the inverse element mapper can be realized.

[0052] Finally, by correcting the matrix To calibrate and , and then fusion can improve the fusion expression effect of the transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix, that is:

[0053]

[0054]

[0055] in, represents the corrected transformer fault attribute-dissolved gas characteristic hyperbolic tangent asymmetric interaction matrix, represents the corrected transformer fault attribute-dissolved gas feature unit-level gated interaction matrix, Represents the optimized transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix.

[0056] Then, the optimized transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix is ​​decomposed into column vectors to obtain the sequence distribution of the transformer fault attribute-dissolved gas local semantic response feature encoding vector, which is expressed as follows:

[0057] in, represents the matrix decomposition function, 、 、 and They represent the first, second, and third in the sequence distribution of transformer fault attribute-dissolved gas local semantic response feature encoding vectors. and Transformer fault attribute-dissolved gas local semantic response feature encoding vector.

[0058] That is, the optimized transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix is ​​converted into a one-dimensional sequence distribution of transformer fault attribute-dissolved gas local semantic response feature encoding vectors. By decomposing it into column vectors, the static global interaction information is reconstructed into a local information flow with temporal dynamic characteristics, so that each transformer fault attribute-dissolved gas local semantic response feature encoding vector corresponds to the semantic association between gas features and fault attributes under a specific interaction dimension. This not only retains the complete information of the global interaction, but also enables the model to capture the subtle semantic response between gas concentration changes and fault attributes in stages through temporal decomposition, providing structured feature input for subsequent global semantic reasoning, and improving the comprehensiveness of fault feature extraction and the reliability of diagnostic results in low-confidence scenarios.

[0059] More specifically, in a specific example of the present application, step S433 includes: inputting the sequence distribution of the transformer fault attribute-dissolved gas local semantic response feature encoding vector into a response inference generation engine based on a forward LSTM model to obtain the transformer fault state comprehensive feature vector, which is expressed as follows:

[0060] in, represents the forward LSTM model, Represents the comprehensive feature vector of transformer fault state.

[0061] Specifically, the forward LSTM model leverages the temporal modeling capabilities of the sequential distribution of transformer fault attribute-dissolved gas local semantic response feature encoding vectors to dynamically integrate information. Its gating mechanism captures the distance-dependent dependencies between local features in the sequence, thereby transforming the dispersed, phased semantic responses into a global feature representation that incorporates temporal correlations and dynamic evolutionary logic, providing a deeper feature foundation for comprehensive fault state judgment. Specifically, the forward LSTM model processes the sequential distribution of the transformer fault attribute-dissolved gas local semantic response feature encoding vectors time-by-time to generate a highly condensed comprehensive transformer fault state feature vector. This vector not only integrates the specific dimensional semantic information carried by each transformer fault attribute-dissolved gas local semantic response feature encoding vector, but also encodes the temporal correlation patterns and dynamic evolutionary logic between gas features and fault attributes from different interactive perspectives. This enables the model to capture complex fault feature patterns from the sequential semantic interactions of multi-source data, which are difficult to identify using traditional methods. This provides a powerful representational support that integrates global temporal dependencies and local semantic details for the precise output of the final fault diagnosis results, effectively improving the accuracy and reliability of fault diagnosis in low-confidence scenarios.

[0062] Specifically, step S44 inputs the transformer fault state comprehensive feature vector into the trained fault diagnosis model to obtain the fault diagnosis result, which is used to represent the transformer fault type. It should be understood that transformer fault diagnosis is essentially a multi-classification problem, requiring further mapping of the transformer fault state comprehensive feature vector to specific fault types (e.g., partial discharge, overheating, arc discharge, etc.). Therefore, to ensure the accuracy and interpretability of the diagnostic results, this application utilizes a supervised learning framework to perform classification decisions on the transformer fault state comprehensive feature vector using the trained fault diagnosis model. Specifically, the fault diagnosis model utilizes a multi-layer perceptron (MLP) architecture, comprising multiple fully connected layers. Nonlinear activation functions (e.g., ReLU) and appropriate regularization strategies (e.g., Dropout) are used to enhance the model's generalization capabilities. During the training phase, the fault diagnosis model is pre-trained using a large historical dataset with real fault labels (including gas data and basic attribute data under various normal and fault conditions). Model parameters are optimized using a backpropagation algorithm, enabling the model to learn the mapping relationship between fault features and fault types. In the inference stage, the comprehensive feature vector of the transformer fault state to be diagnosed is input into the trained fault diagnosis model. Through layer-by-layer transmission and calculation of the multi-layer perceptron, the terminal classification layer of the model outputs the probability distribution of the corresponding fault category (such as no fault, partial discharge, low-temperature overheating, high-temperature overheating, low-energy discharge, high-energy discharge, etc.) through the Softmax function. By selecting the fault type with the highest probability as the final fault diagnosis result, accurate identification of the transformer fault type can be achieved.

[0063] In summary, a transformer fault diagnosis method based on an embodiment of the present application is illustrated, which first collects data on the concentration of dissolved gas in transformer oil, preliminarily determines the potential fault type of the transformer based on the three-ratio method, and introduces a diagnostic result confidence assessment mechanism to calculate the confidence score of the potential fault type. When this confidence is lower than a preset threshold, the above-mentioned potential fault type and its confidence are further combined with the basic attribute data of the transformer to form fault diagnosis prior data. Subsequently, the dissolved gas concentration data in the oil and the fault diagnosis prior data are deeply jointly analyzed using the trained fault diagnosis model to capture more comprehensive fault characteristics, thereby obtaining accurate transformer fault diagnosis results. This method can significantly improve the accuracy and reliability of transformer fault diagnosis by introducing a deep learning algorithm in the low-confidence diagnosis of the traditional three-ratio method.

[0064] Furthermore, a transformer fault diagnosis system is provided.

[0065] Figure 6 FIG. 1 is a block diagram of a transformer fault diagnosis system according to an embodiment of the present application. Figure 6 As shown, the transformer fault diagnosis system 100 according to an embodiment of the present application includes: a gas concentration data acquisition module 110, which is used to collect dissolved gas concentration data in transformer oil through an online gas chromatograph, wherein the dissolved gases in the oil include H2, CH4, C2H2, C2H4, and C2H6; a fault pre-diagnosis module 120, which is used to process the dissolved gas concentration data in the transformer oil based on a three-ratio method to determine the potential fault type of the transformer and its confidence score; a priori data generation module 130, which is used to obtain basic attribute data of the transformer in response to the confidence score being lower than a preset threshold, and add the potential fault type of the transformer and its confidence score to the basic attribute data to obtain transformer fault diagnosis priori data; a fault diagnosis result generation module 140, which is used to input the dissolved gas concentration data in the transformer oil and the transformer fault diagnosis priori data into a trained fault diagnosis model to obtain a fault diagnosis result of the transformer.

[0066] Here, those skilled in the art will appreciate that the specific operations of each module in the above transformer fault diagnosis system have been described in detail above. Figures 1 to 5 The transformer fault diagnosis method has been described in detail in the description of the present invention, and therefore, its repeated description will be omitted.

[0067] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in the present invention are merely illustrative and non-limiting, and should not be construed as necessarily possessed by each embodiment of the present invention. Furthermore, the specific details of the above embodiments are provided for illustrative purposes and to facilitate understanding, and are not intended to be limiting. These details do not necessarily limit the present invention to being implemented using these specific details.

[0068] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0069] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be encompassed therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0070] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0071] Finally, it should be noted that the above description has been provided for the purpose of illustration and description. In addition, the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to be limiting. Although the technical solutions may be modified or replaced with equivalents with reference to the preferred embodiments, they do not depart from the spirit and scope of the technical solutions of the present invention.

Claims

1. A transformer fault diagnosis method, characterized in that: include: Collecting dissolved gas concentration data in transformer oil by online gas chromatograph, wherein the dissolved gases in the oil include H2, CH4, C2H2, C2H4, and C2H6; Processing the dissolved gas concentration data in the transformer oil based on a three-ratio method to determine a potential fault type of the transformer and a confidence score thereof; In response to the confidence score being lower than a preset threshold, obtaining basic attribute data of the transformer, and adding the potential fault type of the transformer and its confidence score to the basic attribute data to obtain transformer fault diagnosis prior data; The dissolved gas concentration data in the transformer oil and the transformer fault diagnosis priori data are input into the trained fault diagnosis model to obtain the fault diagnosis result of the transformer.

2. The transformer fault diagnosis method according to claim 1, characterized in that: The basic attribute data includes the rated capacity, load rate, top oil temperature, insulation paper type, operating years, cooling method and insulation oil type of the transformer.

3. The transformer fault diagnosis method according to claim 1, characterized in that: The dissolved gas concentration data in the transformer oil is processed based on a three-ratio method to determine a potential fault type of the transformer and its confidence score, including: Calculating the IEC three ratios based on the dissolved gas concentration data in the transformer oil, wherein the IEC three ratios include R1=C2H2 / C2H4, R2=CH4 / H2, and R3=C2H4 / C2H6; Determine a fault code of the transformer based on the range interval of the IEC60599 standard and the IEC three ratios; Based on the fault code, looking up a table to obtain a potential fault type of the transformer; The confidence score of the potential fault type is calculated based on the normalized distance of each ratio in the three IEC ratios relative to the interval boundary of the range interval to which it belongs.

4. The transformer fault diagnosis method according to claim 1, characterized in that: Inputting the dissolved gas concentration data in the transformer oil and the transformer fault diagnosis priori data into the trained fault diagnosis model to obtain the fault diagnosis result of the transformer, including: Performing multi-source information association coding on the dissolved gas concentration data in the transformer oil to obtain a multi-source dissolved gas concentration feature association coding vector; Performing semantic embedding coding on the transformer fault diagnosis prior data to obtain a semantic coding vector of the transformer fault diagnosis prior data; Performing semantic interactive response reasoning on the multi-source dissolved gas concentration feature association coding vector and the transformer fault diagnosis prior data semantic coding vector to obtain a comprehensive feature vector of the transformer fault state; The transformer fault state comprehensive feature vector is input into the trained fault diagnosis model to obtain the fault diagnosis result, and the fault diagnosis result is used to represent the fault type of the transformer.

5. The transformer fault diagnosis method according to claim 4, characterized in that: Performing multi-source information association coding on the dissolved gas concentration data in the transformer oil to obtain a multi-source dissolved gas concentration feature association coding vector includes: The dissolved gas concentration data in the transformer oil is input into a multi-source information association encoder based on a fully connected layer to obtain the multi-source dissolved gas concentration feature association coding vector.

6. The transformer fault diagnosis method according to claim 5, characterized in that: Performing semantic embedding coding on the transformer fault diagnosis prior data to obtain a semantic coding vector of the transformer fault diagnosis prior data, including: The transformer fault diagnosis prior data is input into a semantic embedding encoder based on a Bert model to obtain a semantic encoding vector of the transformer fault diagnosis prior data.

7. The transformer fault diagnosis method according to claim 6, characterized in that: Performing semantic interactive response reasoning on the multi-source dissolved gas concentration feature association coding vector and the transformer fault diagnosis prior data semantic coding vector to obtain a comprehensive feature vector of the transformer fault state, including: Calculating a semantic response matrix of the multi-source dissolved gas concentration feature association coding vector relative to the transformer fault diagnosis prior data semantic coding vector to obtain a transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix; Performing matrix decomposition on the transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix to obtain a sequence distribution of transformer fault attribute-dissolved gas local semantic response feature encoding vectors; Global semantic interactive reasoning is performed on the sequence distribution of the transformer fault attribute-dissolved gas local semantic response feature encoding vector to obtain the comprehensive feature vector of the transformer fault state.

8. The transformer fault diagnosis method according to claim 7, characterized in that: Performing matrix decomposition on the transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix to obtain a sequence distribution of transformer fault attribute-dissolved gas local semantic response feature encoding vectors, including: Performing a double-interaction-path equivalent correction on the transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix to obtain an optimized transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix; The optimized transformer fault attribute-dissolved gas feature global fine-grained semantic response matrix is ​​matrix decomposed according to column vectors to obtain the sequence distribution of the transformer fault attribute-dissolved gas local semantic response feature encoding vectors.

9. The transformer fault diagnosis method according to claim 8, characterized in that: Performing global semantic interactive reasoning on the sequence distribution of the transformer fault attribute-dissolved gas local semantic response feature encoding vector to obtain the transformer fault state comprehensive feature vector, including: The sequence distribution of the transformer fault attribute-dissolved gas local semantic response feature encoding vector is input into a response inference generation engine based on a forward LSTM model to obtain the comprehensive feature vector of the transformer fault state.

10. A transformer fault diagnosis system, characterized in that: include: A gas concentration data acquisition module is used to collect dissolved gas concentration data in transformer oil through an online gas chromatograph, wherein the dissolved gases in the oil include H2, CH4, C2H2, C2H4, and C2H6; a fault pre-diagnosis module, configured to process the dissolved gas concentration data in the transformer oil based on a three-ratio method to determine a potential fault type of the transformer and a confidence score thereof; a priori data generating module, configured to, in response to the confidence score being lower than a preset threshold, obtain basic attribute data of the transformer, and add the potential fault type of the transformer and its confidence score to the basic attribute data to obtain priori data for transformer fault diagnosis; The fault diagnosis result generating module is used to input the dissolved gas concentration data in the transformer oil and the transformer fault diagnosis priori data into the trained fault diagnosis model to obtain the fault diagnosis result of the transformer.

Images