Method and device for identifying potential fault modes of a transformer based on back derivation

Abstract

The present application is suitable for the field of transformer fault analysis, and provides a transformer potential fault mode recognition method and device based on reverse deduction, which generates a multi-field coupling feature jigsaw model through real-time collection of four-dimensional physical quantity space-time data such as temperature, vibration, oil chromatogram and partial discharge, matches the model with a historical health record topology, reversely deduces a fault coupling path through a constraint satisfaction algorithm for an abnormal missing area, reconstructs a missing feature block and decouples and extracts interface abnormal features, combines a device aging factor dynamic optimization model topology, finally quantifies a full life cycle risk accumulation intensity and locates a composite fault source. The method breaks through the limitation of a traditional single parameter threshold, realizes early identification and accurate tracing of a multi-field coupling fault by using a reverse deduction mechanism of a feature jigsaw block, is suitable for full life cycle health management of a transformer, and can significantly improve timeliness and accuracy of potential fault early warning.

Description

Technical Field

[0001] The present invention belongs to the field of transformer fault analysis, and in particular relates to a method and device for identifying potential transformer fault patterns based on reverse deduction. Background Art

[0002] Transformers are core components of power generation equipment, and early identification of potential faults is crucial for the safe operation of power grids. The industry has evolved from single-parameter monitoring to integrated monitoring of multiple physical fields (such as temperature, vibration, oil chromatography, and partial discharge). This involves collecting multidimensional data in real time through sensor arrays and leveraging machine learning algorithms to assess condition. However, existing technologies still face challenges in addressing the early signs of multi-field coupled faults, including insufficient depth in feature correlation analysis, poor model adaptability to equipment aging, and difficulty quantifying gradual risk.

[0003] Existing technologies primarily employ single-physical-value threshold alarm mechanisms or static pattern matching methods based on historical data. For example, faults are identified by setting characteristic gas concentration thresholds in oil chromatograms, or by using neural networks to classify single vibration spectra. Some approaches attempt to construct multi-parameter correlation models, but these often rely on fixed topological rules, fail to implement dynamic deduction based on physical field coupling mechanisms, and lack the ability to reverse engineer missing features.

[0004] Existing technologies are unable to effectively capture the early signs of multi-physical field coupling anomalies, and single-parameter threshold alarms are prone to missed or false alarms; static topology models are difficult to adapt to changes in the interaction between physical fields during equipment aging; there is a lack of quantitative means for the risk accumulation process of gradual failures, and it is impossible to achieve trend predictions throughout the entire life cycle; when faced with complex failures, it is difficult to locate the true source of the failure, and diagnostic deviations are often caused by superficial anomalies. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and device for identifying potential fault modes of a transformer based on reverse deduction, aiming to solve the technical problems existing in the prior art identified in the background technology.

[0006] The present invention is implemented by a method for identifying potential fault modes of a transformer based on reverse deduction, the method comprising:

[0007] Real-time collection of transformer physical quantity monitoring data is performed and parsed into standardized feature modules with temporal and spatial correlations. Each standardized feature module represents the correlation pattern of a specific physical quantity in the temporal and spatial dimensions. All standardized feature modules are topologically connected according to the multi-physics field coupling rules to generate a feature puzzle model.

[0008] Topologically match the real-time feature puzzle model with the transformer's historical health records. When missing feature puzzle pieces or deviations from the established topology are detected, the abnormal vacant areas are marked and the reverse filling engine is triggered.

[0009] The fault coupling path that leads to the abnormal vacancy area is deduced through the constraint satisfaction algorithm, and the potential form of the missing feature puzzle piece is generated in reverse.

[0010] Perform multi-physics field signal decoupling on the missing feature puzzle pieces generated by reverse engineering, extract abnormal interface effect features and feed them back to the feature puzzle model;

[0011] Dynamically adjust the topological connection rules of the feature puzzle model according to the equipment aging index to control the self-organization optimization of the feature puzzle model;

[0012] Establish a correlation mapping channel between the currently missing feature puzzle pieces and historical operating fluctuations to quantify the intensity of gradual risk accumulation throughout the transformer life cycle;

[0013] Based on the reverse puzzle verification mechanism, the composite fault source location results are output.

[0014] As a further solution of the present invention, the real-time collection of transformer physical quantity monitoring data and the generation of the feature puzzle model specifically include:

[0015] Synchronously collect the spatiotemporal sequence data of four-dimensional physical quantities including temperature, vibration, oil chromatography and partial discharge to generate a multi-dimensional monitoring time series matrix;

[0016] The frequency domain energy entropy of each physical quantity is extracted as the basic feature vector by wavelet packet decomposition, and the spatial and temporal correlation degree across physical quantities is calculated by combining the mutual information entropy.

[0017] A correlation threshold is set, and the basic feature vectors whose spatiotemporal correlation is higher than the correlation threshold are combined and encoded into standardized puzzle pieces;

[0018] Based on the physical field coupling mechanism, puzzle piece connection rules are established, and all standardized puzzle pieces are assembled according to the puzzle piece connection rules to form a characteristic puzzle model.

[0019] As a further solution of the present invention, the marking of abnormal vacant areas and triggering of the reverse filling engine specifically includes:

[0020] A topological map of historical health records is constructed based on a graph convolutional network, where nodes represent feature puzzle pieces and edge weights represent the strength of association between modules.

[0021] Obtain a real-time feature puzzle model and calculate the structural similarity index between the feature puzzle model and the topological map :

[0022] ;

[0023] in, Represents a collection of real-time feature puzzle model nodes, Represents a collection of historical graph nodes, represents the edge set of the historical graph, represents the historical edge weight, represents the corresponding edge weight in the real-time feature puzzle model, is the weight coefficient;

[0024] when When the value is less than 0.85 and the area of ​​the missing region is greater than 5% of the total area of ​​the atlas, it is determined to be an abnormal missing region;

[0025] Generate a reverse filling task instruction set containing the position coordinates of abnormal vacant areas and the association constraints of adjacent modules.

[0026] As a further solution of the present invention, the reverse generation of the potential form of the missing feature puzzle piece specifically includes:

[0027] Establish a fault propagation directed graph centered on the abnormal vacancy area, where the nodes are faulty transformer components and the edges are the fault transmission probabilities.

[0028] A Monte Carlo constraint satisfaction algorithm is used to iteratively generate candidate filling solutions while satisfying the physical boundary conditions of adjacent standardized puzzle pieces.

[0029] The energy conservation of candidate paths is verified through multi-physics field coupling simulation, and the filling solution set with the highest failure probability is screened. Based on the filling solution set, the reconstruction parameters of the missing modules are generated to generate the reconstructed standardized puzzle pieces.

[0030] As a further solution of the present invention, the multi-physics field signal decoupling of the missing feature puzzle pieces generated inversely, extracting the abnormal interface effect features and feeding them back to the feature puzzle model specifically includes:

[0031] Decompose the reconstructed normalized puzzle pieces to generate hidden fault feature vectors;

[0032] The hidden fault feature vector is injected as a new dimension into the standardized puzzle piece corresponding to the feature puzzle model, and the model feature descriptor is updated.

[0033] As a further solution of the present invention, the dynamic adjustment of the topological connection rules of the feature puzzle model and the control of the self-organization optimization of the feature puzzle model specifically include:

[0034] Quantifying equipment aging factors α =∫(operating years, load factor, fault history weight);

[0035] Dynamically adjust the topology rules of the feature puzzle model according to the value of the quantitative device aging factor;

[0036] Based on the reinforcement learning mechanism, the topology update strategy is optimized with the fault recognition accuracy as the reward function, and the reward function is: R=1 / (false alarm rate + missed alarm rate).

[0037] As a further solution of the present invention, the quantification of the cumulative intensity of gradual risk in the entire life cycle of the transformer specifically includes:

[0038] Constructing a time scale mapping function , align the evolution rate of the current missing feature puzzle pieces with the feature puzzle pieces in the historical health records;

[0039] Calculating the distribution similarity between the current abnormal pattern and the historical fluctuation pattern by using Wasserstein distance, wherein the current abnormal pattern refers to the physical field parameter distribution of the inversely reconstructed characteristic puzzle piece;

[0040] Generate a risk accumulation path diagram and output the risk accumulation intensity of the entire life cycle :

[0041] ;

[0042] ;

[0043] in, represents the risk acceleration, The set of physical parameters representing the characteristic puzzle pieces for inverse reconstruction, Indicates the health status benchmark parameters of the corresponding spatial location in the historical health archive, Indicates the time when the relay equipment is put into operation. is the current time.

[0044] As a further solution of the present invention, the outputting of the composite fault source location result specifically includes:

[0045] Reversely embed the inversely reconstructed feature puzzle pieces into the original feature puzzle model and verify the physical field continuity;

[0046] Comparison of the deviation between the fault probability determined by a single physical quantity threshold and the fault probability determined by multi-field coupling deduction :

[0047] ;

[0048] in, represents the failure probability of a single physical quantity threshold judgment, represents the failure probability of multi-field coupling simulation;

[0049] When the deviation is greater than 40%, it is determined that a superficial fault exists, and the composite fault source location result and the multi-field coupling contribution weight are output.

[0050] Another object of the present invention is to provide a device for identifying potential fault modes of a transformer based on reverse deduction, the device comprising:

[0051] A topological connection module is used to collect transformer physical quantity monitoring data in real time and parse it into standardized feature modules with spatiotemporal correlation. Each standardized feature module represents the correlation pattern of a specific physical quantity in the spatiotemporal dimension. All standardized feature modules are topologically connected according to the multi-physical field coupling rules to generate a feature puzzle model.

[0052] Abnormal vacancy area marking module, which is used to topologically match the real-time feature puzzle model with the historical health records of the transformer. When it is detected that the feature puzzle pieces are missing or deviate from the established topology structure, the abnormal vacancy area is marked and the reverse filling engine is triggered;

[0053] The missing feature generation module is used to deduce the fault coupling path that leads to the abnormal vacancy area through the constraint satisfaction algorithm, and reversely generate the potential form of the missing feature puzzle piece;

[0054] Abnormal feature extraction module, used to decouple multi-physics field signals from the missing feature puzzle pieces generated by reverse engineering, extract abnormal interface effect features and feed them back to the feature puzzle model;

[0055] The topology rule adjustment module is used to dynamically adjust the topology connection rules of the feature puzzle model according to the device aging index and control the self-organization optimization of the feature puzzle model;

[0056] The correlation mapping channel establishment module is used to establish the correlation mapping channel between the current missing feature puzzle pieces and the historical operation fluctuations, and quantify the intensity of the gradual risk accumulation in the entire life cycle of the transformer;

[0057] The result output module is used to output the composite fault source location results based on the reverse puzzle verification mechanism.

[0058] The beneficial effects of the present invention are:

[0059] This solution converts transformer multi-physical field monitoring data into standardized feature puzzle blocks with spatiotemporal correlation through a closed-loop mechanism of "feature puzzle model construction - abnormal area marking - reverse deduction of missing blocks - multi-field decoupling feedback - dynamic topology optimization". When missing blocks are found by topological matching between the real-time model and the historical health archive, the fault coupling path is reversely deduced through the constraint satisfaction algorithm, the potential fault morphology is reconstructed, and the abnormal characteristics of the interface effect are decoupled and extracted.

[0060] This solution breaks through the limitations of traditional single-parameter analysis and realizes the early identification of multi-field coupled faults: through the topological association of feature puzzle blocks, it can capture early signs of a sudden change in the strength of inter-field correlations but without exceeding the threshold value of a single parameter; the reverse deduction mechanism can dynamically generate the potential form of missing features and quantify the probability of fault propagation; the model topology is dynamically adjusted in combination with the equipment aging factor, so that the diagnostic capability can adaptively evolve with the equipment life cycle; finally, through the reverse puzzle verification mechanism, the complex fault source is accurately located, which solves the shortcomings of existing technologies in multi-field coupled anomaly identification, aging adaptability and fault tracing, extends the lead time of early fault warning, and significantly improves the operating reliability and condition-based maintenance efficiency of transformers. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A flowchart of a method for identifying potential transformer fault patterns based on reverse deduction provided by an embodiment of the present invention;

[0062] Figure 2 A flowchart of real-time collection of transformer physical quantity monitoring data and generation of a feature puzzle model provided by an embodiment of the present invention;

[0063] Figure 3 A flowchart of marking abnormal vacant areas and triggering a reverse filling engine provided by an embodiment of the present invention;

[0064] Figure 4 A flowchart of reversely generating potential forms of missing feature puzzle pieces provided by an embodiment of the present invention;

[0065] Figure 5 A flowchart of extracting abnormal interface effect features and feeding them back to a feature puzzle model provided by an embodiment of the present invention;

[0066] Figure 6 A flowchart of dynamically adjusting the topological connection rules of a feature puzzle model and controlling the self-organizing optimization of the feature puzzle model provided by an embodiment of the present invention;

[0067] Figure 7 A flowchart of quantifying the cumulative intensity of gradual risk in the entire life cycle of a transformer provided by an embodiment of the present invention;

[0068] Figure 8 A flowchart of outputting composite fault source location results provided by an embodiment of the present invention;

[0069] Figure 9 This is a structural block diagram of a transformer potential fault mode identification device based on reverse deduction provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0071] Figure 1 The flow chart of the transformer potential fault mode recognition method based on reverse deduction provided by the embodiment of the present application is shown in Figure 1 The method comprises the following steps:

[0072] S100, real-time acquisition of transformer physical quantity monitoring data, and parsing into a space-time correlated standardized feature module, each standardized feature module representing the correlation mode of a specific physical quantity in the space-time dimension, and generating a feature jigsaw model by topological connection according to the multi-physical field coupling rule;

[0073] This step synchronously acquires the space-time sequence data of four-dimensional physical quantities such as temperature field distribution (such as three-dimensional temperature gradient of winding, iron core and oil), mechanical vibration spectrum (covering high-frequency abnormal sound and low-frequency resonance signal under different working conditions), oil chromatographic component concentration (including real-time volume fraction of characteristic gases such as hydrogen, acetylene and carbon monoxide) and partial discharge quantity time domain waveform by deploying a sensor array at each key position of the transformer with a millisecond-level sampling frequency, and generates a multi-dimensional monitoring time sequence matrix containing time stamp, spatial coordinates and physical quantity amplitude after preliminary preprocessing by an edge computing node.

[0074] In the feature analysis stage, wavelet packet decomposition is used to perform multi-scale decomposition in the time-frequency domain of each physical quantity, and the energy entropy of each frequency band is extracted as the basic feature vector representing the equipment operating state, and the correlation degree of different physical quantities in the space-time dimension is quantified by mutual information entropy, for example, when the acetylene concentration mutation in the oil chromatogram and the partial discharge pulse show strong correlation on the time axis, it can be determined that there is a potential discharge decomposition coupling relationship between them.

[0075] With the preset correlation degree threshold as the screening condition, the feature vectors with strong coupling relationship are combined and coded as standardized jigsaw blocks, and each jigsaw block not only contains the feature parameters of a single physical quantity, but also embeds the correlation weight matrix across physical fields.

[0076] Based on the coupling mechanism of electromagnetic field-thermal field-mechanical field-chemical field, the jigsaw block connection rule is established, such as the thermal-electric coupling rule which defines the nonlinear mapping relationship between winding temperature rise and partial discharge quantity, so as to topologically assemble all standardized jigsaw blocks into a dynamic feature jigsaw model, which intuitively presents the space-time coupling relationship of the transformer multi-physical field in the form of a graph structure, with the feature jigsaw blocks as the nodes and the physical field interaction strength as the edges.

[0077] This step, through multi-physics field spatiotemporal fusion modeling, breaks through the limitations of traditional single-parameter threshold alarms and achieves holographic characterization and dynamic correlation analysis of equipment status. Specifically, the simultaneous acquisition of four-dimensional physical quantities and the construction of a spatiotemporal correlation feature module can capture early signs of faults before a single parameter anomaly becomes apparent. For example, when a weak partial discharge occurs inside a transformer, it may first manifest as a small, coordinated change in the high-frequency components of the vibration spectrum and the hydrogen content in the oil chromatogram. This weakly correlated signal across physical fields can be effectively identified through mutual information entropy calculations, while traditional single-parameter monitoring often misses detection because it does not reach the alarm threshold.

[0078] In addition, the feature puzzle model constructs topological connections based on the physical field coupling mechanism, converting abstract device states into a visual graph structure. This not only makes it easier for engineers to intuitively understand the interactive relationship between various physical quantities, but also provides a structured fault propagation path carrier for subsequent reverse deduction. For example, when an abnormality occurs in the temperature-vibration coupling block, the model can quickly locate the possible fault source area based on preset thermal-mechanical coupling rules.

[0079] like Figure 2 As shown, the real-time collection of transformer physical quantity monitoring data and the generation of the feature puzzle model specifically include:

[0080] S110, synchronously collecting spatiotemporal sequence data of four-dimensional physical quantities of temperature, vibration, oil chromatography, and partial discharge, and generating a multi-dimensional monitoring time series matrix;

[0081] S120, extracting frequency domain energy entropy of each physical quantity as a basic feature vector by wavelet packet decomposition, and calculating the spatiotemporal correlation degree across physical quantities in combination with mutual information entropy;

[0082] S130, setting a correlation threshold, and encoding the basic feature vectors whose spatiotemporal correlation is higher than the correlation threshold into a standardized puzzle piece;

[0083] S140, establishing puzzle piece connection rules based on the physical field coupling mechanism, and assembling all standardized puzzle pieces according to the puzzle piece connection rules to form a feature puzzle model.

[0084] S200 performs topological matching between the real-time feature puzzle model and the historical health records of the transformer. When it detects that a feature puzzle piece is missing or deviates from the established topological structure, it marks the abnormal vacant area and triggers the reverse filling engine;

[0085] This step first constructs a topological map of the health record based on historical operating data, and uses a graph convolutional network (GCN) to perform deep learning on the spatiotemporal correlation patterns of historical feature puzzle blocks, where nodes correspond to standardized feature modules (such as temperature-oil chromatography coupling blocks), and edge weights are dynamically assigned by the mean mutual information entropy between modules in the historical data to characterize the long-term coupling strength of physical fields such as mechanical vibration and partial discharge.

[0086] In the real-time matching stage, the current feature puzzle model is aligned with the historical topology map through the graph isomorphism algorithm to perform node alignment and edge weight comparison. When calculating the structural similarity index SSI, the weight coefficient ω can be dynamically adjusted according to the device type (such as oil-immersed transformer ω =0.6 to focus on node matching, dry-type transformer ω =0.4 focuses on edge correlation), when SSI is less than 0.85 and the proportion of vacant areas exceeds 5%, the device not only marks the coordinates of the abnormal area, but also extracts the physical field constraints of adjacent modules (for example, the temperature module requires the gas generation rate gradient of the adjacent oil chromatography module to be ≤0.5ppm / ℃), and generates a reverse filling task package containing spatial position, correlation constraints and historical coupling probability.

[0087] By incorporating historical operating fluctuation data into the construction of the topological map, early fault signs can be captured where a single parameter has not exceeded the threshold but the associated structure has shifted. For example, when the transformer core is slightly loose, the change in the high-frequency component of the vibration spectrum may not have reached the threshold, but the weight of the associated edge with the temperature module will drop significantly due to abnormal mechanical-thermal coupling. This structural change can be identified in advance through SSI calculation.

[0088] In addition, the SSI indicator considers both node existence and edge association strength, and can effectively distinguish between normal operating fluctuations and structural anomalies in the early stages of a fault. For example, a temperature increase caused by a sudden load change will maintain the integrity of the topological structure, while anomalies caused by inter-turn short circuits will be accompanied by node loss and a sudden change in edge weights, and the SSI value can drop sharply from the normal 0.92 to 0.78.

[0089] like Figure 3 As shown, the marking of abnormal vacant areas and triggering of the reverse filling engine specifically includes:

[0090] S210, constructs a topological map of historical health records based on a graph convolutional network, where nodes represent feature puzzle pieces and edge weights represent the strength of associations between modules;

[0091] S220, obtaining a real-time feature puzzle model and calculating a structural similarity index between the feature puzzle model and the topological map :

[0092] ;

[0093] in, Represents a collection of real-time feature puzzle model nodes, Represents a collection of historical graph nodes, represents the edge set of the historical graph, represents the historical edge weight, represents the corresponding edge weight in the real-time feature puzzle model, is the weight coefficient;

[0094] S230, when the structural similarity index is less than 0.85 and the area of ​​the vacant region is greater than 5% of the total area of ​​the map, it is determined to be an abnormal vacant region;

[0095] S240, generating a reverse filling task instruction set including the position coordinates of the abnormal vacant area and the association constraints of adjacent modules.

[0096] S300, uses a constraint satisfaction algorithm to deduce the fault coupling path that leads to the abnormal vacancy area, and reversely generates the potential form of the missing feature puzzle piece;

[0097] This step uses the abnormally vacant area as the topological center to construct a fault propagation directed graph that includes key components such as the transformer winding, iron core, and oil pillow. The edge weights in the graph are assigned by the conditional probability matrix obtained through training with historical fault data (for example, the transmission probability of winding overheating to insulation aging is 0.72).

[0098] When using the Monte Carlo constraint satisfaction algorithm, the physical field boundary conditions of adjacent standardized puzzle blocks (for example, the temperature module requires the field strength gradient of adjacent discharge modules to be ≤1.2kV / mm) are first randomly sampled to generate more than 1,000 sets of candidate filling schemes. Each set of schemes includes a combination of faulty components, parameter distribution, and propagation path.

[0099] The energy conservation of candidate solutions was then verified through a multi-physics field coupling simulation platform. When simulating inter-turn short-circuit faults, the energy conversion balance between the winding temperature field distribution and the oil chromatography gas production rate needed to be verified (the error needed to be less than 5%). Finally, a set of solutions with the top 10% failure probability was selected, and standardized puzzle blocks containing temperature-discharge-chromatography coupling characteristics were generated based on their reconstruction parameters. For example, a nonlinear mapping function (y=0.3x²+1.5x) between discharge amount and acetylene concentration was embedded in the reconstruction block.

[0100] The fault propagation directed graph, combined with historical fault probability data, can quantify the possibility of chain reactions from different component failures. For example, when an abnormal oil temperature is detected, the algorithm can simultaneously deduce potential fault paths such as winding overheating (probability 0.65) and abnormal core hysteresis loss (probability 0.28), rather than attributing a single cause.

[0101] The combination of Monte Carlo algorithm and physical field simulation enables the generated candidate solutions to satisfy both randomness and physical laws, avoiding the rigidity of traditional rule-based reasoning.

[0102] like Figure 4 As shown, the potential form of the reverse generation of missing feature puzzle pieces specifically includes:

[0103] S310, establishing a fault propagation directed graph centered on the abnormal vacancy area, where nodes are faulty transformer components and edges are fault transmission probabilities;

[0104] S320, using a Monte Carlo constraint satisfaction algorithm to iteratively generate candidate filling solutions while satisfying the physical field boundary conditions of adjacent standardized puzzle pieces;

[0105] S330 verifies the energy conservation of candidate paths through multi-physics field coupling simulation, selects the filling solution set with the highest failure probability, and generates reconstruction parameters of missing modules based on the filling solution set to generate reconstruction standardized puzzle pieces.

[0106] S400, decoupling the multi-physics field signals of the missing feature puzzle pieces generated by reverse engineering, extracting the abnormal characteristics of the interface effect and feeding them back to the feature puzzle model;

[0107] This step uses a tensor decomposition algorithm to couple and separate the temperature-vibration-oil chromatography-discharge four-dimensional physical field signals contained in the reconstructed standardized puzzle blocks. For example, the mixed signal is decomposed into independent heat conduction components, mechanical vibration components, gas diffusion components, and electromagnetic discharge components, and then the abnormal characteristic parameters of the interface effects between each field are extracted, such as the gas generation rate gradient of the temperature-oil chromatography interface and the energy transfer efficiency attenuation coefficient of the vibration-discharge interface.

[0108] When generating latent fault feature vectors, blind source separation technology is used to extract hidden features representing interface failure from the coupled signal. When the transformer winding insulation deteriorates, after decoupling, an abnormal coupled signal with a characteristic frequency of 10kHz is found at the interface between the temperature field and the discharge field. This signal is easily submerged by noise in traditional single-field analysis.

[0109] Subsequently, these feature vectors containing interface effect anomalies are injected into the corresponding modules of the feature puzzle model as new dimensions, and the feature descriptors of the model are updated synchronously. For example, an "insulation interface degradation index" attribute is added to the temperature-discharge coupling block. This attribute is dynamically calculated through the nonlinear mapping relationship between the discharge energy and temperature gradient obtained by decoupling.

[0110] The signal decoupling process in this step can decompose the composite signal generated by multi-field interactions into independent components, thereby capturing subtle abnormal changes at the interface. When the mechanical interface between the transformer core and winding is slightly loose, the decoupled vibration signal will show energy concentration in a specific frequency band (such as 200-500Hz), which is difficult to identify with traditional time domain analysis.

[0111] In addition, feeding back the implicit features obtained by decoupling to the model can enable the feature puzzle model to have dynamic learning capabilities. When decoupling finds that the C2H2 generation rate at the oil chromatography and temperature interface is abnormal, the model automatically updates the association rules of the coupling block and adjusts the mapping relationship between C2H2 concentration and temperature from linear y=0.2x to nonlinear y=0.2x+0.05x², thereby more accurately identifying similar faults in subsequent monitoring.

[0112] In the actual application of power equipment, when the main transformer oil chromatographic data showed a slight anomaly (acetylene concentration of 22 ppm), this method discovered the nonlinear coupling anomaly between the winding partial discharge and the oil temperature increase by decoupling the gas generation rate characteristics of the temperature-chromatographic interface. Compared with the traditional single-parameter analysis, the method located the micro-damage fault of the winding inter-turn insulation 56 hours earlier, avoiding the sudden tripping accident caused by the propagation of the fault, and demonstrating the key role of multi-physics field decoupling in early fault mechanism identification and model adaptive optimization.

[0113] like Figure 5 As shown, the multi-physics field signal decoupling of the missing feature puzzle pieces generated inversely, extracting the abnormal interface effect features and feeding them back to the feature puzzle model specifically includes:

[0114] S410, decomposing the reconstructed standardized puzzle pieces to generate a hidden fault feature vector;

[0115] S420: inject the hidden fault feature vector as a new dimension into the standardized puzzle piece corresponding to the feature puzzle model, and update the model feature descriptor.

[0116] S500, dynamically adjusts the topological connection rules of the feature puzzle model according to the device aging index, and controls the self-organization optimization of the feature puzzle model;

[0117] This step builds a multi-dimensional aging assessment system, quantifying parameters such as operating years (the aging weight of transformers over 15 years is increased by 20%), long-term load rate (the aging acceleration factor is 1.5 when it exceeds 80% of the rated value), and the number of historical faults (the aging factor is added by 0.1 for each internal fault) into equipment aging factors through integral calculation. α ,when α When the value is >0.7, the dynamic adjustment mechanism of topology rules is triggered:

[0118] Expanding the correlation radius of the temperature-oil chromatogram puzzle piece from the original d1 to 1.3d1 allows the model to capture the delayed correlation between temperature and gas generation caused by the decrease in heat transfer efficiency in aging devices;

[0119] At the same time, the angle constraint of the mechanical-chemical coupling block is changed from θ∈[30°,60°] is expanded to [20°,70°] to accommodate the vibration mode shift caused by loose silicon steel sheets in the aging core.

[0120] During the reinforcement learning optimization phase, guided by fault identification accuracy, the topology update strategy is iteratively adjusted through real-time feedback of false alarm and missed alarm rate data. For example, if the missed alarm rate of a certain type of aging fault is consistently higher than 15%, the association weight threshold of the corresponding coupling block is automatically increased, making the model more sensitive to aging-related anomalies.

[0121] The aging factor quantification mechanism converts the gradual change of the equipment health status into a computable model parameter, so that the topology rules can automatically evolve with the degree of equipment aging. When the thermal-chromatographic correlation of the transformer is weakened due to the deterioration of the insulating oil, the model can be used to calculate the aging factor. α The value (0.75) triggers the rule adjustment, which reduces the correlation threshold of the temperature-oil chromatogram block from 0.6 to 0.5, thereby continuing to effectively capture the weak coupling anomaly between oil temperature and gas concentration.

[0122] The reinforcement learning mechanism empowers the model with autonomous learning capabilities, significantly reducing monitoring blind spots caused by equipment aging compared to traditional fixed-topology models. Furthermore, the adaptive expansion of coupling block parameters (such as correlation radius and angle constraints) during dynamic adjustment effectively compensates for changes in the physical field propagation characteristics of aging equipment.

[0123] like Figure 6 As shown, the dynamic adjustment of the topological connection rules of the feature puzzle model and the control of the self-organization optimization of the feature puzzle model specifically include:

[0124] S510, quantified equipment aging factor α =∫(operating years, load factor, fault history weight);

[0125] S520: Dynamically adjust the topology rules of the feature puzzle model according to the value of the quantified device aging factor:

[0126] when α When >0.7, the puzzle piece association radius is allowed to expand by 30%, the association threshold τ Reduce by 15%;

[0127] when α When >0.7, the maximum spacing of the thermal-electric coupling blocks is relaxed from d1 to 1.3d1;

[0128] Mechano-chemical coupling block angle constraints extended to θ ∈[20°,70°];

[0129] S530, based on the reinforcement learning mechanism, optimizes the topology update strategy with the fault recognition accuracy as the reward function, and the reward function is: R = 1 / (false alarm rate + missed alarm rate).

[0130] S600: Establish a correlation mapping channel between the currently missing feature puzzle pieces and historical operating fluctuations to quantify the intensity of the gradual risk accumulation throughout the transformer life cycle;

[0131] This step constructs a time scale mapping function to nonlinearly align the parameters of the current reverse-generated feature puzzle block with the corresponding modules in the historical health record in the time dimension, thereby solving the problem of inconsistent data rates in different operating stages of the equipment (such as the difference in temperature change rate in the initial operation period and the aging period).

[0132] When using the Wasserstein distance to calculate the distribution similarity between the current anomaly pattern and the historical fluctuation pattern, it not only considers the mean deviation of the physical quantity parameters but also quantifies the overall shift of the probability density function. When the distribution of hydrogen concentration in the oil chromatogram changes from a single peak to a double peak, this distance can sensitively capture the potential failure mode transition. In the risk accumulation calculation, the second-order derivative a(t) is used to characterize the risk acceleration, and the inflection point of the risk mutation can be identified (for example, when a(t) increases from 0.05 to 0.12, it indicates that the fault development rate is accelerating). Then, the integral operation F is used to generate a risk accumulation path diagram from the commissioning time t0 to the current time t. This diagram superimposes the evolutionary trajectories of multiple physical field parameters such as temperature and discharge, intuitively presenting the development trend of gradual faults.

[0133] The time scale mapping function solves the comparability problem of data with different operating years. For example, the vibration data of transformers with 5 and 15 years of operation are unified to the same time scale through logarithmic transformation, so that the historical fluctuation pattern can accurately map the current anomaly.

[0134] Compared with the Euclidean distance, the Wasserstein distance is more suitable for analyzing fault characteristics with asymmetric distributions. In addition, the introduction of risk acceleration enables the device to capture the nonlinear characteristics of fault development, issuing replacement warnings earlier than traditional life assessments, thus avoiding sudden failures caused by insulation aging.

[0135] like Figure 7 As shown in FIG, the quantified gradual risk accumulation intensity in the entire life cycle of the transformer specifically includes:

[0136] S610, constructing a time scale mapping function , align the evolution rate of the current missing feature puzzle pieces with the feature puzzle pieces in the historical health records;

[0137] S620, calculating the distribution similarity between the current abnormal pattern and the historical fluctuation pattern by using Wasserstein distance, wherein the current abnormal pattern refers to the distribution of physical field parameters of the inversely reconstructed characteristic puzzle piece;

[0138] S630, generate a risk accumulation path diagram and output the risk accumulation intensity of the entire life cycle :

[0139] ;

[0140] ;

[0141] in, represents the risk acceleration, The set of physical parameters representing the characteristic puzzle pieces for inverse reconstruction, Indicates the health status benchmark parameters of the corresponding spatial location in the historical health archive, Indicates the time when the relay equipment is put into operation. is the current time.

[0142] S700, based on the reverse puzzle verification mechanism, outputs composite fault source location results.

[0143] This step embeds the reversely reconstructed feature puzzle blocks (such as modules containing temperature-discharge-chromatography coupling features) into the abnormally vacant areas of the original feature puzzle model, and verifies the continuity of the physical field after embedding through the finite element simulation platform (such as the temperature field gradient change rate must be less than 0.3℃ / mm, and the discharge field intensity distribution must satisfy the Poisson equation), ensuring the self-consistency of the field interaction logic between the reconstructed module and the existing model.

[0144] During the deviation calculation stage, the single physical quantity threshold judgment engine (for example, determining the fault probability based solely on the 30ppm threshold of the oil chromatogram acetylene concentration) and the multi-field coupling deduction engine (combining parameters such as the temperature rise rate and the degree of vibration spectrum distortion) are synchronously called. When the fault probability deviation Δ between the two is greater than 40%, it is determined that a superficial fault exists (for example, the oil temperature increase may be caused by both winding overheating and cooling device abnormalities). At this time, the output composite fault source location result will include a weight matrix of the contribution of each physical field (for example, 45% contribution from the temperature field, 35% contribution from the discharge field, and 20% contribution from the chromatographic field), and the critical path of fault propagation (such as the thermal-electrical-mechanical coupling chain of winding → oil gap → iron core) will be highlighted through a topological diagram.

[0145] This step breaks through the one-sidedness of traditional single-parameter diagnosis and realizes the precise tracing of complex faults and identification of superficial faults through multi-field coupling deviation analysis.

[0146] The reverse embedding verification mechanism can verify the physical rationality of the reverse reconstruction results and avoid misjudgment due to algorithmic errors. For example, after decoupling and reconstruction, verification found that the energy conservation deviation between the temperature field and the discharge field was as much as 8%, thereby overturning the initial reconstruction plan and re-deducing it, and finally locating the real fault source of the abnormal hysteresis loss of the iron core.

[0147] The introduction of the deviation degree Δ effectively solves the problem of false alarms caused by single physical quantity alarms. For example, when a transformer cooler fan anomaly occurs, a single parameter may only show an increase in oil temperature (fault probability 60%). However, multi-field coupling deduction finds no abnormality in the fan speed-related frequency band in the vibration spectrum (fault probability 25%). Δ=58%>40%, indicating that the device determines that a superficial fault exists and requires further investigation.

[0148] like Figure 8 As shown, the output composite fault source location result specifically includes:

[0149] S710, reversely embedding the inversely reconstructed feature puzzle piece into the original feature puzzle model and verifying the physical field continuity;

[0150] S720: Compare the deviation between the fault probability determined by the single physical quantity threshold and the fault probability determined by the multi-field coupling deduction. :

[0151] ;

[0152] in, represents the failure probability of a single physical quantity threshold judgment, represents the failure probability of multi-field coupling deduction;

[0153] S730: When the deviation is greater than 40%, it is determined that a superficial fault exists, and the composite fault source location result and the multi-field coupling contribution weight are output.

[0154] Figure 9 The structural block diagram of the transformer potential fault mode identification device based on reverse deduction provided by the embodiment of the present invention is as follows: Figure 9 As shown, the device includes:

[0155] A topological connection module 100 is used to collect transformer physical quantity monitoring data in real time and parse it into standardized feature modules with spatiotemporal correlation. Each standardized feature module represents the correlation pattern of a specific physical quantity in the spatiotemporal dimension. All standardized feature modules are topologically connected according to the multi-physical field coupling rules to generate a feature puzzle model.

[0156] The abnormal vacancy area marking module 200 is used to perform topological matching between the real-time feature puzzle model and the historical health record of the transformer. When it is detected that the feature puzzle piece is missing or deviates from the established topological structure, the abnormal vacancy area is marked and the reverse filling engine is triggered;

[0157] The missing feature generation module 300 is used to deduce the fault coupling path that leads to the abnormal vacancy area through the constraint satisfaction algorithm, and reversely generate the potential form of the missing feature puzzle piece;

[0158] The anomaly feature extraction module 400 is configured to decouple the multi-physical field signals of the reversely generated missing feature jigsaw puzzle piece, extract interface effect anomaly features, and feed back the interface effect anomaly features to the feature jigsaw model;

[0159] The topology rule adjustment module 500 is configured to dynamically adjust a topology connection rule of the feature jigsaw model according to the equipment aging index, and control self-organization optimization of the feature jigsaw model.

[0160] The correlation mapping channel establishment module 600 is configured to establish a correlation mapping channel between the current missing feature jigsaw puzzle piece and historical operation fluctuations, and quantify the gradual risk accumulation intensity in the whole life cycle of the transformer.

[0161] The result output module 700 is configured to output a composite fault source positioning result based on a reverse jigsaw verification mechanism.

[0162] The technical features of the above-described embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above-described embodiments are not described, but as long as the combinations of the technical features do not contradict, they should be considered within the scope of the present disclosure.

[0163] The above-described embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as limiting the scope of the present patent. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application. Therefore, the protection scope of the present patent should be subject to the appended claims.

[0164] The above-described embodiments are only the preferred embodiments of the present application, and are not used to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for identifying potential fault patterns of transformers based on reverse deduction, characterized in that: The method comprises: Real-time collection of transformer physical quantity monitoring data is performed and parsed into standardized feature modules with temporal and spatial correlations. Each standardized feature module represents the correlation pattern of a specific physical quantity in the temporal and spatial dimensions. All standardized feature modules are topologically connected according to the multi-physics field coupling rules to generate a feature puzzle model. Topologically match the real-time feature puzzle model with the transformer's historical health records. When missing feature puzzle pieces or deviations from the established topology are detected, the abnormal vacant areas are marked and the reverse filling engine is triggered. The fault coupling path that leads to the abnormal vacancy area is deduced through the constraint satisfaction algorithm, and the potential form of the missing feature puzzle piece is generated in reverse; Perform multi-physics field signal decoupling on the missing feature puzzle pieces generated by reverse engineering, extract abnormal interface effect features and feed them back to the feature puzzle model; Dynamically adjust the topological connection rules of the feature puzzle model according to the equipment aging index to control the self-organization optimization of the feature puzzle model; Establish a correlation mapping channel between the currently missing feature puzzle pieces and historical operating fluctuations to quantify the intensity of gradual risk accumulation throughout the transformer life cycle; Based on the reverse puzzle verification mechanism, the composite fault source location results are output.

2. The method according to claim 1, characterized in that The real-time collection of transformer physical quantity monitoring data and the generation of a feature puzzle model specifically include: Synchronously collect the spatiotemporal sequence data of four-dimensional physical quantities including temperature, vibration, oil chromatography and partial discharge to generate a multi-dimensional monitoring time series matrix; The frequency domain energy entropy of each physical quantity is extracted as the basic feature vector by wavelet packet decomposition, and the spatial and temporal correlation degree across physical quantities is calculated by combining the mutual information entropy. A correlation threshold is set, and the basic feature vectors whose spatiotemporal correlation is higher than the correlation threshold are combined and encoded into standardized puzzle pieces; Based on the physical field coupling mechanism, puzzle piece connection rules are established, and all standardized puzzle pieces are assembled according to the puzzle piece connection rules to form a characteristic puzzle model.

3. The method according to claim 2, characterized in that The marking of abnormal vacant areas and triggering the reverse filling engine specifically includes: A topological map of historical health records is constructed based on a graph convolutional network, where nodes represent feature puzzle pieces and edge weights represent the strength of association between modules. Obtain a real-time feature puzzle model and calculate the structural similarity index between the feature puzzle model and the topological map : ； in, Represents a collection of real-time feature puzzle model nodes, Represents a collection of historical graph nodes, represents the edge set of the historical graph, represents the historical edge weight, represents the corresponding edge weight in the real-time feature puzzle model, is the weight coefficient; when When the value is less than 0.85 and the area of ​​the missing region is greater than 5% of the total area of ​​the atlas, it is determined to be an abnormal missing region; Generate a reverse filling task instruction set containing the position coordinates of abnormal vacant areas and the association constraints of adjacent modules.

4. The method according to claim 3, characterized in that The reverse generation of the potential form of the missing feature puzzle piece specifically includes: Establish a fault propagation directed graph centered on the abnormal vacancy area, where the nodes are faulty transformer components and the edges are the fault transmission probabilities. A Monte Carlo constraint satisfaction algorithm is used to iteratively generate candidate filling solutions while satisfying the physical boundary conditions of adjacent standardized puzzle pieces. The energy conservation of candidate paths is verified through multi-physics field coupling simulation, and the filling solution set with the highest failure probability is screened. Based on the filling solution set, the reconstruction parameters of the missing modules are generated to generate the reconstructed standardized puzzle pieces.

5. The method according to claim 4, characterized in that The multi-physics field signal decoupling of the missing feature puzzle pieces generated inversely, extracting abnormal interface effect features and feeding them back to the feature puzzle model specifically includes: Decompose the reconstructed normalized puzzle pieces to generate hidden fault feature vectors; The hidden fault feature vector is injected as a new dimension into the standardized puzzle piece corresponding to the feature puzzle model, and the model feature descriptor is updated.

6. The method according to claim 5, characterized in that The dynamically adjusting the topological connection rules of the feature puzzle model and controlling the self-organization optimization of the feature puzzle model specifically include: Quantifying equipment aging factors α =∫(operating years, load factor, fault history weight); Dynamically adjust the topology rules of the feature puzzle model according to the value of the quantitative device aging factor; Based on the reinforcement learning mechanism, the topology update strategy is optimized with the fault recognition accuracy as the reward function, and the reward function is: R=1 / (false alarm rate + missed alarm rate).

7. The method according to claim 6, characterized in that The quantified cumulative intensity of gradual risk in the transformer life cycle specifically includes: Constructing a time scale mapping function , align the evolution rate of the current missing feature puzzle pieces with the feature puzzle pieces in the historical health records; Calculating the distribution similarity between the current abnormal pattern and the historical fluctuation pattern by using Wasserstein distance, wherein the current abnormal pattern refers to the distribution of physical field parameters of the inversely reconstructed characteristic puzzle piece; Generate a risk accumulation path diagram and output the risk accumulation intensity of the entire life cycle : ； ； in, represents the risk acceleration, The set of physical parameters representing the characteristic puzzle pieces for inverse reconstruction, Indicates the health status benchmark parameters of the corresponding spatial location in the historical health archive, Indicates the time when the relay equipment is put into operation. is the current time.

8. The method according to claim 7, characterized in that The output composite fault source location result specifically includes: Reversely embed the inversely reconstructed feature puzzle pieces into the original feature puzzle model and verify the physical field continuity; Comparison of the deviation between the fault probability determined by a single physical quantity threshold and the fault probability determined by multi-field coupling deduction : ； in, represents the failure probability of a single physical quantity threshold judgment, represents the failure probability of multi-field coupling simulation; When the deviation is greater than 40%, it is determined that a superficial fault exists, and the composite fault source location result and multi-field coupling contribution weight are output.

9. A device for identifying potential fault patterns of transformers based on reverse deduction, characterized in that: The device comprises: A topological connection module is used to collect transformer physical quantity monitoring data in real time and parse it into standardized feature modules with spatiotemporal correlation. Each standardized feature module represents the correlation pattern of a specific physical quantity in the spatiotemporal dimension. All standardized feature modules are topologically connected according to the multi-physical field coupling rules to generate a feature puzzle model. Abnormal vacancy area marking module, which is used to topologically match the real-time feature puzzle model with the historical health records of the transformer. When it is detected that the feature puzzle pieces are missing or deviate from the established topology structure, the abnormal vacancy area is marked and the reverse filling engine is triggered; The missing feature generation module is used to deduce the fault coupling path that leads to the abnormal vacant area through the constraint satisfaction algorithm, and reversely generate the potential form of the missing feature puzzle piece; The abnormal feature extraction module is used to decouple the multi-physics field signals of the missing feature puzzle pieces generated by reverse engineering, extract the abnormal features of the interface effect and feed them back to the feature puzzle model; A topology rule adjustment module is used to dynamically adjust the topology connection rules of the feature puzzle model according to the device aging index and control the self-organization optimization of the feature puzzle model; The correlation mapping channel establishment module is used to establish the correlation mapping channel between the current missing feature puzzle pieces and the historical operation fluctuations, and quantify the intensity of the gradual risk accumulation in the entire life cycle of the transformer; The result output module is used to output the composite fault source location results based on the reverse puzzle verification mechanism.

Images