A monitoring system and method for distribution transformers

Abstract

This invention discloses a monitoring system and method for distribution transformers, relating to the field of intelligent power monitoring technology. The method includes: acquiring the three-dimensional temperature gradient field distribution and mechanical stress distribution of the distribution transformer based on a standardized multi-source dataset, and outputting a comprehensive field matrix using field coupling analysis; performing spatial analysis on the comprehensive field matrix using spatial hotspot analysis to identify and mark abnormal temperature and mechanical stress regions, performing spatial aggregation and noise filtering using the OPTICS density clustering algorithm, and outputting anomaly feature vectors; inputting the anomaly feature vectors into a pre-trained intelligent diagnostic model to output a comprehensive fault risk score and perform a preliminary diagnosis; retrieving fault data from a historical case database based on the preliminary diagnosis results for comparison and verification, optimizing the preliminary diagnosis results using machine learning algorithms, and generating a final diagnosis result; this invention uses field coupling analysis to establish a thermo-mechanical bidirectional interaction model, accurately reflecting the spatial distribution characteristics of temperature gradient and stress tensor.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power intelligent monitoring, in particular to a power distribution transformer monitoring system and a monitoring method. BACKGROUND

[0002] With the rapid development of smart grid, the power distribution transformer state monitoring technology continues to innovate, relying on sensor networks and machine learning to realize the intelligentization of data collection and fault diagnosis. The current technology mainly obtains real-time temperature, vibration, electrical parameters and other multi-dimensional physical quantity data of the transformer during operation through infrared temperature measurement, vibration detection, oil chromatographic analysis and other means. Based on threshold judgment, statistical analysis and signal processing algorithm, the system can preliminarily identify abnormal states.

[0003] In the technical field of power intelligent monitoring, the traditional method usually adopts a step-by-step processing flow, that is, the temperature gradient field and the mechanical stress distribution are calculated independently first, and then a simple correlation is made through an empirical formula or linear superposition. This processing method is difficult to accurately reflect the nonlinear interaction effect of multiple physical fields under complex working conditions inside the transformer, especially in the scene of transient overload or local insulation degradation, which is easy to cause misjudgment and missed detection of abnormal areas. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a power distribution transformer monitoring method to solve the problem of low fault diagnosis precision caused by the lack of coordinated analysis of multiple physical fields of the power distribution transformer.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a power distribution transformer monitoring method, which comprises,

[0008] Real-time collection of power distribution transformer operating state data and preprocessing to obtain a standardized multi-source data set; based on the standardized multi-source data set, obtaining the three-dimensional temperature gradient field distribution and mechanical stress distribution of the power distribution transformer, and outputting a comprehensive field matrix through field coupling analysis; using spatial hotspot analysis to implement spatial analysis on the comprehensive field matrix, identifying and marking temperature and mechanical stress abnormal areas, and outputting an abnormal feature vector through OPTICS density clustering algorithm for spatial aggregation and noise filtering; inputting the abnormal feature vector into a pre-trained intelligent diagnosis model to output a comprehensive fault risk score and perform preliminary diagnosis; based on the preliminary diagnosis result, searching and comparing the fault data in the historical case database, and optimizing the preliminary diagnosis result through a machine learning algorithm to generate a final diagnosis result; performing dynamic early warning through the final diagnosis result, and generating a fault detection report.

[0009] As a preferred scheme of the power distribution transformer monitoring method, wherein: the power distribution transformer operating state data includes temperature data, core vibration data, shell vibration data, load current data, operating voltage data, and insulation oil gas data.

[0010] The preprocessing includes noise suppression by sliding window filtering, data normalization by Z-score standardization method, and missing value compensation by linear interpolation.

[0011] As a preferred scheme of the power distribution transformer monitoring method, wherein: based on the standardized multi-source data set, three-dimensional temperature gradient field distribution and mechanical stress distribution of the power distribution transformer are obtained, and a comprehensive field matrix is output through field coupling analysis method, the specific steps are as follows,

[0012] Based on the standardized multi-source data set, the three-dimensional temperature gradient field distribution is calculated by finite element analysis method, and the mechanical stress distribution is obtained through the generalized Hooke's law;

[0013] The three-dimensional temperature gradient field distribution and the mechanical stress distribution are input into the field coupling analysis method, the thermal-mechanical coupling relationship is established, and the comprehensive field matrix containing multi-dimensional data of spatial coordinates, three-dimensional temperature gradient field distribution and mechanical stress distribution is output.

[0014] As a preferred scheme of the power distribution transformer monitoring method, wherein: the spatial analysis is performed on the comprehensive field matrix by using the spatial hotspot analysis, the temperature and mechanical stress abnormal areas are identified and marked, the spatial aggregation and noise filtering are performed through the OPTICS density clustering algorithm, and the abnormal feature vector is output, the specific steps are as follows,

[0015] The comprehensive field matrix is input into the spatial hotspot analysis, the temperature gradient and mechanical stress abnormal areas are identified, and the abnormal points are output;

[0016] The abnormal points are spatially clustered by the OPTICS density clustering algorithm, the adjacent abnormal points are merged, and the isolated noise points are filtered, and the clustering core area is output;

[0017] Based on the clustering core area, the spatial coordinates, the average temperature gradient and the maximum mechanical stress of the abnormal points are extracted by the spatial statistical feature extraction method, and the abnormal feature vector is generated.

[0018] As a preferred scheme of the power distribution transformer monitoring method, wherein: the abnormal feature vector is input into the pre-trained intelligent diagnosis model, and the comprehensive fault risk score is output and preliminary diagnosis is performed, the specific steps are as follows,

[0019] Based on the historical abnormal feature vector, the DNN deep neural network is trained by the back propagation algorithm method, and the intelligent diagnosis model capable of outputting the comprehensive fault risk score is obtained.

[0020] Input the abnormal feature vector into the intelligent diagnosis model to analyze the comprehensive fault risk score of each abnormal area;

[0021] Based on historical fault data, define low risk threshold, high risk threshold and fault classification rule, and judge the comprehensive fault risk score to obtain the preliminary diagnosis result.

[0022] As a preferred scheme of the power distribution transformer monitoring method, wherein: based on the preliminary diagnosis result, the fault data in the historical case library is searched for comparison and verification, the preliminary diagnosis result is optimized through a machine learning algorithm, and a final diagnosis result is generated, and the specific steps are as follows,

[0023] Based on the preliminary diagnosis result, the fault data records under similar working conditions are matched in the historical case library, and Top-K similar cases are output.

[0024] The preliminary diagnosis result and the treatment effect data of the Top-K similar cases are input into the machine learning algorithm, the diagnosis conclusion is optimized through probability weighting, and the final diagnosis result is generated.

[0025] As a preferred scheme of the power distribution transformer monitoring method, wherein: based on the preliminary diagnosis result, the fault data in the historical case library is searched for comparison and verification, the preliminary diagnosis result is optimized through a machine learning algorithm, and a final diagnosis result is generated, and the specific steps are as follows,

[0026] Based on the final diagnosis result, a rule engine is used to automatically match a warning rule library to trigger multi-level warning;

[0027] The final diagnosis result, multi-level warning and power distribution transformer operation state data are time and space aligned through a feature fusion algorithm to construct structured diagnosis data.

[0028] Based on the structured diagnosis data, intelligent analysis is performed through a natural language generation method to obtain a fault detection report.

[0029] In a second aspect, the present application provides a power distribution transformer monitoring system, comprising,

[0030] The data acquisition module is used for collecting power distribution transformer operation state data in real time and performing preprocessing to obtain a standardized multi-source data set; the coupling analysis module is used for obtaining a three-dimensional temperature gradient field distribution and a mechanical stress distribution of the power distribution transformer based on the standardized multi-source data set, and outputting a comprehensive field matrix through a field coupling analysis method; the abnormality extraction module is used for implementing spatial analysis on the comprehensive field matrix by using spatial hotspot analysis, identifying and marking temperature and mechanical stress abnormal regions, performing spatial aggregation and noise filtering through an OPTICS density clustering algorithm, and outputting an abnormal feature vector; the diagnosis module is used for inputting the abnormal feature vector into a pre-trained intelligent diagnosis model, outputting a comprehensive fault risk score and performing preliminary diagnosis; the optimization module is used for retrieving fault data in a historical case library based on the preliminary diagnosis result for comparison and verification, optimizing the preliminary diagnosis result through a machine learning algorithm, and generating a final diagnosis result; and the report generation module is used for performing dynamic early warning through the final diagnosis result and generating a fault detection report.

[0031] In a third aspect, the present application provides a computer device comprising a memory and a processor, and the memory stores a computer program, wherein the computer program is executed by the processor to implement any step of the power distribution transformer monitoring method according to the first aspect of the present application.

[0032] In a fourth aspect, the present application provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement any step of the power distribution transformer monitoring method according to the first aspect of the present application.

[0033] The present application has the following beneficial effects: the three-dimensional temperature gradient field distribution and the mechanical stress distribution are calculated through finite element analysis and generalized Hooke's law, and a thermal-mechanical bidirectional action model is established by using a field coupling analysis method, so that the collaborative calculation and closed-loop feedback of the temperature field and the stress field are realized, and the spatial distribution characteristics of the temperature gradient and the stress tensor are accurately reflected; meanwhile, the spatial hotspot analysis result is intelligently aggregated through an OPTICS density clustering algorithm, the spatial coordinates, the average temperature gradient and the maximum mechanical stress features of the clustering core region are extracted, the automatic identification and quantitative characterization of the distributed fault region are realized, and the fault diagnosis accuracy and reliability are improved. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0035] Fig. 1 Flow chart for power distribution transformer monitoring method.

[0036] Fig. 2 Flow chart for power distribution transformer operating state data preprocessing.

[0037] Fig. 3 Flow chart for field coupling analysis.

[0038] Fig. 4 Schematic diagram of power distribution transformer monitoring system. DETAILED DESCRIPTION

[0039] In order to make the above objectives, features and advantages of the present application more apparent, specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0040] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details given herein, that the present application can be practiced with other than the described implementations, and that the present application can be practiced with different or additional components. Therefore, the specific details set forth in the following description are by way of examples and not intended to limit the present application.

[0041] Secondly, the "one embodiment" or "embodiment" referred to herein can include specific features, structures or characteristics in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments.

[0042] Reference Figs. 1-4 The embodiment provides a power distribution transformer monitoring method, comprising the following steps:

[0043] S1, real-time acquisition of power distribution transformer operating state data, and preprocessing to obtain standardized multi-source data set.

[0044] S1.1, the power distribution transformer operating state data includes temperature data, core vibration data, shell vibration data, load current data, operating voltage data, and insulation oil gas data.

[0045] It should be noted that the power distribution transformer operating state data is achieved by deploying various sensors at the power distribution transformer. The temperature data is collected by the optical fiber temperature sensor embedded in the winding and the iron core and the infrared temperature measurement module attached to the shell, which can reflect the overall and local heating state of the device; The core and shell vibration data are obtained by high-precision acceleration sensors, which monitor the internal mechanical vibration caused by magnetostriction and external structure resonance, respectively; The load current and operating voltage are measured by the Hall sensor and voltage transformer installed on the high and low voltage side in real time, which is used to analyze the electrical load working condition; The insulation oil gas data is collected by the oil chromatography online monitoring device, which detects the dissolved gas composition and its concentration change to evaluate the insulation aging degree.

[0046] S1.2, the pretreatment includes noise suppression by sliding window filtering, data normalization by Z-score standardization method, and missing value compensation by linear interpolation.

[0047] It should be noted that the sliding window filtering technique is used to suppress high frequency noise. For vibration signals, a window width adjustment mechanism based on signal energy analysis is used, which automatically reduces the window width to retain detailed features when random pulse interference is detected. For temperature data, the window width is dynamically expanded according to the temperature change rate, effectively smoothing the abnormal jumps caused by sensor transient response. The Z-score standardization method is used to process heterogeneous physical quantities, eliminating the numerical scale differences between temperature data, core vibration data, shell vibration data, load current data, operating voltage data, and insulation oil gas data. The temperature data, core vibration data, shell vibration data, load current data, operating voltage data, and insulation oil gas data are comparable. For missing values caused by network delay or device failure during transmission, linear interpolation algorithm is used based on time series correlation to compensate, ensuring the continuity of temperature data, core vibration data, shell vibration data, load current data, operating voltage data, and insulation oil gas data. The multi-source data after hierarchical processing is synchronized and aligned in time domain, forming a standardized multi-source data set that can be used for advanced analysis.

[0048] S2, based on the standardized multi-source data set, the three-dimensional temperature gradient field distribution and mechanical stress distribution of the power distribution transformer are obtained, and the comprehensive field matrix is output by the field coupling analysis method.

[0049] S2.1, based on the standardized multi-source data set, the three-dimensional temperature gradient field distribution is calculated by the finite element analysis method, and the mechanical stress distribution is obtained by the generalized Hook's law.

[0050] It should be noted that after importing the standardized multi-source dataset into the finite element analysis environment, a three-dimensional geometric model of the distribution transformer is established based on its actual physical structure, including key components such as the core, windings, and insulation layers. Tetrahedral elements are used for spatial discretization. Temperature data is applied as boundary conditions to the corresponding node positions, and the heat conduction partial differential equation is solved to obtain a continuously distributed three-dimensional temperature gradient field.

[0051] Solving the partial differential equation for heat conduction yields the expression for a continuously distributed temperature gradient field:

[0052] ;

[0053] in, It is a three-dimensional temperature gradient field distribution, representing the temperature in space. Rate of change in position It is the differential change in the temperature field. It is along Differential displacements of the coordinate axes, It is along Differential displacements of the coordinate axes, It is along Differential displacements of the coordinate axes, It is the total number of nodes. It is an index variable representing the total number of nodes. It is the first Temperature values ​​at each node It is the first The shape function of each node in relation to spatial coordinates The partial derivatives;

[0054] Vibration data of the iron core and shell are used as displacement boundary conditions and applied to the corresponding node positions of the finite element model. A stiffness matrix is ​​constructed using parameters such as the material's elastic modulus and Poisson's ratio. A stress-strain relationship σ = C:ε is established based on the generalized Hooke's law, where σ is the stress tensor, C is the elastic stiffness tensor, and ε is the strain tensor. The strain components are calculated using the displacement field gradient: ε = (∇u + ∇uᵀ) / 2, where ∇u is the gradient of the displacement field u, ∇uᵀ is the transpose of the displacement gradient, and u is the vibration displacement vector. The Newton-Raphson iterative method is used for nonlinear solution: initial assumptions are made regarding the displacement field distribution, the residual force vector is calculated, the linearized equilibrium equations are solved to update the displacement field, and the residual force is recalculated until the convergence criterion ‖R‖ < ε is met. The Jacobian matrix is ​​updated in each iteration, and the linear equations are solved using the preconditional conjugate gradient method. After convergence, the stress components at the Gaussian integral points of each element are calculated using constitutive relations, and the nodal stresses are interpolated. Finally, the mechanical stress distribution is output.

[0055] S2.2, import the three-dimensional temperature gradient field distribution and mechanical stress distribution into the field coupling analysis platform, perform field matching alignment operation under the unified grid system. The temperature gradient field distribution data is embedded into the mechanical equilibrium equation as a thermal load item, and the mechanical stress distribution data is used to feedback and correct the thermal physical parameters. A two-way coupled control equation set is constructed, the temperature gradient field distribution is converted into equivalent thermal stress item through thermal expansion coefficient, and the mechanical stress distribution is used to affect the temperature field distribution through deformation work;

[0056] It should be noted that the three-dimensional temperature gradient field distribution and the mechanical stress distribution are imported into the coupling analysis platform, and the field matching alignment operation is performed under the unified grid system. The temperature gradient field distribution data is embedded into the mechanical equilibrium equation as a thermal load item, and the mechanical stress distribution data is used to feedback and correct the thermal physical parameters. A two-way coupled control equation set is constructed, the temperature gradient field distribution is converted into equivalent thermal stress item through thermal expansion coefficient, and the mechanical stress distribution is used to affect the temperature field distribution through deformation work;

[0057] Perform step-by-step iterative solution process, temperature field solver updates three-dimensional temperature gradient field distribution data, mechanical solver updates mechanical stress distribution data, alternating boundary conditions until the full field residual reaches the convergence threshold. After coupling convergence, extract all grid node complete physical quantity, including node space coordinates, three direction temperature gradient components, six independent stress tensor components, according to the spatial topological relationship organization storage, output contains space coordinates, three-dimensional temperature gradient field distribution and mechanical stress distribution of multi-dimensional data of comprehensive field matrix.

[0058] S3, adopt space hot spot analysis to implement spatial analysis on comprehensive field matrix, identify and mark temperature and mechanical stress abnormal area, output abnormal feature vector through OPTICS density clustering algorithm for spatial aggregation and noise filtering.

[0059] S3.1, input the comprehensive field matrix into the space hot spot analysis, identify the temperature gradient and mechanical stress abnormal area, output the abnormal point.

[0060] It should be noted that the comprehensive field matrix is analyzed into the spatial coordinates, temperature gradient components and mechanical stress components data of the grid nodes. The temperature gradient field distribution data is identified by calculating the standard deviation of the temperature gradient values of each node and the adjacent nodes, and the abnormal high temperature gradient area exceeding the set threshold is identified. The mechanical stress distribution data uses von Mises stress criterion to evaluate the equivalent stress of each node, and marks the area exceeding the material yield strength limit. The spatial correlation rules are established, and the nodes exceeding the limit of temperature gradient and mechanical stress are cross-verified to exclude isolated noise points. The verified abnormal nodes are recorded as abnormal points, including complete position coordinates, temperature gradient abnormal degree and mechanical stress exceeding value, and finally output the abnormal points.

[0061] S3.2, perform spatial clustering on the abnormal points through the OPTICS density clustering algorithm, merge adjacent abnormal points and filter isolated noise points, output the clustering core area.

[0062] It should be noted that the abnormal points are loaded into the computing environment of the OPTICS density clustering algorithm to construct the point set topology based on the spatial coordinate information. The Euclidean distance to the kth nearest neighbor point is calculated for each abnormal point as the core distance, and the minimum sample size parameter is set to determine the neighborhood range. All abnormal points are traversed to generate an ordered reachable distance sequence, and potential clusters are identified by analyzing the significant concave regions in the sequence curve. Density expansion is performed on adjacent abnormal points in space that meet the core distance condition to form connected abnormal point clusters. A statistical-based noise filtering strategy is used to eliminate isolated points that are consistently above the dynamic threshold range. Convex hull calculation is performed on each connected cluster to determine the spatial boundary range, and the dense region containing the majority of abnormal points is retained as the effective cluster. The cluster core region is output.

[0063] S3.3, based on the cluster core region, the spatial coordinate, the average temperature gradient and the maximum mechanical stress of the abnormal point are extracted by the spatial statistical feature extraction method to generate the abnormal feature vector.

[0064] It should be noted that the cluster core region is loaded into the feature extraction environment for spatial statistical analysis and processing. For each identified cluster core region, all abnormal point data records contained in the region are traversed. The geometric center position coordinates of the cluster core region are calculated, which are obtained by averaging all abnormal point spatial coordinates in the region. For temperature gradient field distribution data, the temperature gradient component values of each abnormal point in the cluster core region in three orthogonal directions are extracted, and the average gradient values in X, Y and Z directions are calculated, which together constitute the representative temperature gradient feature of the region. For mechanical stress distribution data, the six independent stress components of each abnormal point in the cluster core region are checked, the absolute values of all stress components are compared, and the stress component with the largest value is recorded as the maximum mechanical stress feature of the region. The obtained geometric center coordinates, average temperature gradient values in three directions and maximum mechanical stress value are arranged and combined in a predetermined order to form a fixed-dimensional feature description vector. Each cluster core region corresponds to a complete abnormal feature vector, which contains key parameters in three aspects of spatial position information, temperature gradient feature and mechanical stress feature. The final output result is the abnormal feature vector corresponding to all cluster core regions.

[0065] S4, input the abnormal feature vector into the pre-trained intelligent diagnosis model, output the comprehensive fault risk score and perform preliminary diagnosis.

[0066] S4.1, based on the historical abnormal feature vector, the DNN deep neural network is trained by the back propagation algorithm method to obtain an intelligent diagnosis model that can output a comprehensive fault risk score.

[0067] It should be noted that the historical abnormal feature vector set is loaded into the DNN deep neural network training environment, and a DNN deep neural network architecture including an input layer, multiple hidden layers and an output layer is constructed. The input layer nodes strictly correspond to the spatial coordinates, average temperature gradient and maximum mechanical stress feature dimensions of the abnormal feature vector. Each hidden layer is configured with an activation function to implement a nonlinear transformation, and the output layer uses a specific activation function to map the result to a probability value.

[0068] The training process initializes the weight parameters, sets the learning rate control parameter update amplitude, and defines the batch size to determine the sample size for each parameter update. The backpropagation algorithm is used for iterative optimization. In the forward propagation stage, the abnormal feature vector is passed through each layer and the predicted output is calculated. The loss function value is obtained by comparing the predicted output with the true fault type labeled in the historical fault record. In the backpropagation stage, the network weight parameters are adjusted layer by layer according to the gradient of the loss function. The validation set is used independently to monitor the training process. When the validation loss does not decrease for multiple consecutive iterations, the early stopping mechanism is triggered.

[0069] The trained DNN deep neural network solidifies the parameters of each layer, establishing a stable mapping relationship from abnormal features to fault risk. When a new abnormal feature vector is input, the network can calculate the standardized comprehensive fault risk score. The score value reflects the probability of failure, and the final intelligent diagnostic model can output the comprehensive fault risk score.

[0070] S4.2, input the abnormal feature vector into the intelligent diagnostic model, and analyze the comprehensive fault risk score of each abnormal area.

[0071] It should be noted that the abnormal feature vector is loaded into the input interface of the trained intelligent diagnostic model. After the intelligent diagnostic model receives the abnormal feature vector, the input layer nodes are activated to pass the spatial coordinates, average temperature gradient and maximum mechanical stress feature values to the first hidden layer. The hidden layer performs weighted summation operation and applies activation function for nonlinear transformation. The processing results are sequentially passed to the subsequent hidden layers. Each hidden layer repeatedly performs feature extraction and conversion operations, gradually abstracting high-level feature representations. The output layer receives the output of the last hidden layer, and calculates the scalar value in the interval of zero to one as the comprehensive fault risk score through a specific activation function. The score result establishes a mapping relationship with the abnormal feature vector. For each input abnormal feature vector, the intelligent diagnostic model outputs the corresponding comprehensive fault risk score.

[0072] S4.3, based on the historical fault data, define low risk threshold, high risk threshold and fault classification rules, and judge the comprehensive fault risk score to obtain the preliminary diagnosis result.

[0073] It should be noted that the correspondence between the comprehensive failure risk score and the actual failure in the past cases is analyzed. According to the statistical distribution of historical failure data, the threshold boundary for distinguishing low-risk and high-risk states is determined, the low-risk threshold is set as the upper limit of the score interval with low failure probability, and the high-risk threshold is set as the lower limit of the score interval with significantly increased failure probability. A multi-level failure classification rule is established. When the comprehensive failure risk score is located in the interval from zero to the low-risk threshold, it is determined as "normal state", indicating that the current abnormal area is in a safe operation range and only needs regular monitoring; when the comprehensive failure risk score exceeds the low-risk threshold but does not reach the high-risk threshold, it is determined as "warning state", indicating that the abnormal area has potential failure signs and needs to be monitored and prepared for maintenance plan; when the comprehensive failure risk score exceeds the high-risk threshold, it is determined as "failure state";

[0074] The comprehensive failure risk score output by the intelligent diagnosis model is compared with the defined threshold interval, and the risk level of each abnormal area is labeled according to the preset classification rule. The spatial coordinates of the abnormal area, the comprehensive failure risk score and the risk level label form a structured diagnosis record, and the determination results of all abnormal areas are collected to generate a preliminary diagnosis report.

[0075] S5, based on the preliminary diagnosis result, retrieve the failure data in the historical case library for comparison and verification, optimize the preliminary diagnosis result through machine learning algorithm, and generate the final diagnosis result.

[0076] S5.1, based on the preliminary diagnosis result, retrieve the historical case library, match the failure data records under similar working conditions, and output Top-K similar cases.

[0077] It should be noted that the preliminary diagnosis result is input into the case retrieval engine, and the spatial coordinates of the abnormal area, the comprehensive failure risk score and the risk level in the diagnosis result are extracted as retrieval features. The historical case library establishes a multi-dimensional index structure, and the index fields include the abnormal feature vector, failure type and disposal record of the historical cases. Similarity calculation algorithm is adopted to compare the features of the current abnormal area with the records of the historical case library one by one, and the spatial position similarity, risk score deviation and failure type matching degree are calculated. Set the composite weight coefficient, the spatial position similarity weight is the highest, the risk score deviation is the second, and the failure type matching degree weight is the lowest. According to the descending order of the comprehensive similarity score, the top K cases are selected. The effectiveness of the Top-K similar cases is verified, and the output Top-K similar case set contains case number, similarity score, historical failure details and disposal scheme and other complete information.

[0078] S5.2, input the preliminary diagnosis result and the disposal effect data of the Top-K similar cases into the machine learning algorithm, optimize the diagnosis conclusion through probability weighting, and generate the final diagnosis result.

[0079] It should be noted that the preliminary diagnosis result and the Top-K similar case set are loaded into the machine learning optimization environment, and the risk level classification in the preliminary diagnosis result and the treatment effect record in the Top-K similar cases are extracted as optimization basis;

[0080] The treatment effect record contains complete maintenance data of the actual maintenance measures taken in the historical cases, the change of the equipment operating state after maintenance, the fault recurrence period, etc., which are quantitatively coded to form computable treatment effect data. The treatment effect data specifically includes evaluation values of three dimensions of the matching degree of maintenance measures and fault types, the degree of recovery of equipment performance after maintenance, and the interval time of fault recurrence. Each dimension is converted into a standardized numerical value according to a predefined scoring standard. The matching degree of maintenance measures and fault types reflects the pertinence of the treatment scheme, the degree of recovery of equipment performance after maintenance measures the repair effect, and the interval time of fault recurrence evaluates the durability of the repair scheme. The evaluation values of the three dimensions are fused by weighting to form a comprehensive treatment effect score, and the higher the score value, the more successful the treatment scheme of the historical case. The comprehensive treatment effect score and the case similarity score together constitute the dual basis for the optimization diagnosis conclusion;

[0081] A probability weighting matrix is constructed, the rows of the matrix correspond to the possible fault types of the preliminary diagnosis result, the columns correspond to the treatment effects of the Top-K similar cases, and the matrix element values are the normalized values of the case similarity scores. Using the Bayesian inference framework, the treatment effects of the Top-K similar cases are taken as observed evidence, and the preliminary diagnosis result is taken as prior probability, and the posterior probability distribution is calculated. For each fault type hypothesis, the weighted treatment effect data of the relevant cases is accumulated, and the modified probability is obtained after normalization. Based on the accuracy rate of the historical diagnosis cases, a probability confidence threshold is set. When the modified probability exceeds the probability confidence threshold (0.7-0.9), the diagnosis conclusion is confirmed, otherwise the preliminary diagnosis result is retained. The fault type judgment after probability weighting optimization is integrated with the original risk level classification to form the final diagnosis result containing fault type, risk score, confidence level and recommended treatment scheme.

[0082] S6, dynamic early warning is carried out through the final diagnosis result, and a fault detection report is generated.

[0083] S6.1, based on the final diagnosis result, the rule engine automatically matches the early warning rule library to trigger multi-level early warning.

[0084] It should be noted that the final diagnosis result is input into the rule engine execution environment, and the fault type, risk score and confidence level fields in the diagnosis result are parsed. The early warning rule library establishes a multi-level index structure according to the fault type and risk level, and each early warning rule clearly defines the trigger condition, early warning level and response measure. The rule engine matches the final diagnosis result with the condition expression in the early warning rule library piece by piece, and activates the corresponding early warning rule when the fault type is completely matched and the risk score reaches the rule threshold. For the matched early warning rule, the preset early warning level and response measure are extracted, and a standardized early warning notification is generated. The early warning notification contains structured fields such as fault location coordinates, early warning level code, recommended treatment measures and time limit requirements. The multi-level early warning is sorted according to the emergency degree, and the high-risk early warning is immediately pushed, and the medium and low-risk early warning is put into the processing queue. All triggered early warning records are stored in the early warning log database.

[0085] S6.2, the final diagnosis result, multi-level early warning and power distribution transformer operation state data are spatio-temporally aligned through a feature fusion algorithm to construct structured diagnostic data.

[0086] It should be noted that the final diagnosis result, multi-level early warning record and power distribution transformer operation state data are loaded into the feature fusion processing environment. According to the unified time stamp, the three types of data sources are time-aligned to ensure the time consistency of the data records. Based on the power distribution transformer physical structure topology map, the abnormal area coordinates in the final diagnosis result and the spatial positions of the operation state monitoring points are mapped. The early warning level field of the multi-level early warning record is correlated with the change trend of the operation state parameters. A feature-level fusion strategy is adopted to combine the fault type code of the final diagnosis result, the early warning level identifier of the multi-level early warning and the key parameters of the power distribution transformer operation state. A five-tuple data structure is constructed, including time dimension, space dimension, diagnosis dimension, early warning dimension and operation dimension. Each data unit records complete time mark, space position, diagnosis conclusion, early warning level and operation parameter, forming structured diagnostic data with spatio-temporal correlation.

[0087] S6.3, based on the structured diagnostic data, intelligent analysis is carried out through a natural language generation method to obtain a fault detection report.

[0088] It should be noted that the structured diagnostic data is input into the natural language generation engine, and the time mark, space position, diagnosis conclusion, early warning level and operation parameter in the five-tuple data structure are parsed. A report template library is established, and chapter structures such as fault overview, abnormal positioning, risk analysis, treatment suggestion are defined. The key features in the structured diagnostic data are extracted, the time dimension data is converted into readable time description, the space position data is mapped to the physical position description of the equipment, the diagnosis conclusion and the early warning level are converted into risk level expression, and the operation parameter change trend is generated quantitative analysis sentence;

[0089] The rule-based text generation method is adopted to fill the structured fields into the corresponding positions of the report template according to the semantic relationship. The diagnostic conclusion part generates the severity description in combination with the fault type and the risk score, the abnormal positioning part generates the location description in combination with the spatial coordinates and the distribution transformer structure diagram, the risk analysis part generates the trend prediction in combination with the historical cases and the current parameters, and the disposal suggestion part generates the operation guidance in combination with the early warning rule library. The generated text segments are checked for grammar and logical coherence, and then combined into a complete fault detection report conforming to the technical document specification. The fault detection report includes standardized chapters such as title, equipment information, detection time, fault description, analysis conclusion and suggested measures.

[0090] The embodiment also provides a power distribution transformer monitoring system, comprising:

[0091] a data acquisition module, a coupling analysis module, an abnormality extraction module, a diagnosis module, an optimization module and a report generation module;

[0092] The data acquisition module is configured to acquire power distribution transformer operation state data in real time and perform preprocessing to obtain a standardized multi-source data set;

[0093] The coupling analysis module is configured to acquire a three-dimensional temperature gradient field distribution and a mechanical stress distribution of the power distribution transformer based on the standardized multi-source data set, and output a comprehensive field matrix through a field coupling analysis method;

[0094] The abnormality extraction module is configured to perform spatial analysis on the comprehensive field matrix by using spatial hotspot analysis, identify and mark temperature and mechanical stress abnormal regions, perform spatial aggregation and noise filtering through an OPTICS density clustering algorithm, and output an abnormal feature vector;

[0095] The diagnosis module is configured to input the abnormal feature vector into a pre-trained intelligent diagnosis model, output a comprehensive fault risk score and perform preliminary diagnosis;

[0096] The optimization module is configured to retrieve fault data in a historical case library for comparison and verification based on the preliminary diagnosis result, optimize the preliminary diagnosis result through a machine learning algorithm, and generate a final diagnosis result;

[0097] The report generation module is configured to perform dynamic early warning through the final diagnosis result and generate a fault detection report.

[0098] The embodiment also provides a computer device suitable for the power distribution transformer monitoring method, comprising a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement the power distribution transformer monitoring method proposed in the above embodiment.

[0099] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0100] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the power distribution transformer monitoring method proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk.

[0101] To sum up, the application achieves the collaborative calculation and closed-loop feedback of the temperature field and the stress field by calculating the three-dimensional temperature gradient field distribution and the mechanical stress distribution through finite element analysis and generalized Hooke's law and establishing a thermal-mechanical bidirectional action model by using field coupling analysis, and accurately reflects the spatial distribution characteristics of the temperature gradient and the stress tensor. Meanwhile, the spatial hot spot analysis result is intelligently aggregated by using the OPTICS density clustering algorithm, the spatial coordinates, the average temperature gradient and the maximum mechanical stress characteristics of the clustering core region are extracted, the automatic identification and quantitative characterization of the distributed fault region are achieved, and the fault diagnosis accuracy and reliability are improved.

[0102] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A method of monitoring a distribution transformer, the method comprising: The application relates to a power distribution transformer fault diagnosis method and device. ​ Real-time acquisition of power distribution transformer operation state data and preprocessing to obtain a standardized multi-source data set; Based on the standardized multi-source data set, three-dimensional temperature gradient field distribution and mechanical stress distribution of the power distribution transformer are obtained, and a comprehensive field matrix is output through field coupling analysis; Spatial analysis is performed on the comprehensive field matrix by adopting spatial hotspot analysis, temperature and mechanical stress abnormal regions are identified and marked, and an abnormal feature vector is output through OPTICS density clustering algorithm for spatial aggregation and noise filtering; The abnormal feature vector is input into a pre-trained intelligent diagnosis model, a comprehensive fault risk score is output, and preliminary diagnosis is performed; Based on the preliminary diagnosis result, fault data in a historical case library are retrieved for comparison and verification, the preliminary diagnosis result is optimized through a machine learning algorithm, and a final diagnosis result is generated; A dynamic early warning is performed through the final diagnosis result, and a fault detection report is generated; Based on the preliminary diagnosis result, fault data records under similar working conditions are matched in the historical case library, and Top-K similar cases are output; The preliminary diagnosis result and disposal effect data of the Top-K similar cases are input into a machine learning algorithm, the diagnosis conclusion is optimized through probability weighting, and a final diagnosis result is generated. The power distribution transformer operation state data includes temperature data, core vibration data, shell vibration data, load current data, operating voltage data and insulating oil gas data. The preprocessing includes noise suppression through sliding window filtering, data normalization through Z-score standardization method and missing value compensation through linear interpolation.

2. The distribution transformer monitoring method of claim 1, wherein: Based on the standardized multi-source data set, three-dimensional temperature gradient field distribution and mechanical stress distribution of the power distribution transformer are obtained, and a comprehensive field matrix is output through field coupling analysis. Based on the standardized multi-source data set, three-dimensional temperature gradient field distribution is calculated through finite element analysis, and mechanical stress distribution is obtained through generalized Hooke's law; 3. The distribution transformer monitoring method of claim 2, wherein: The three-dimensional temperature gradient field distribution and the mechanical stress distribution are input into the field coupling analysis method, a thermal-mechanical coupling relationship is established, and a comprehensive field matrix of multi-dimensional data including spatial coordinates, three-dimensional temperature gradient field distribution and mechanical stress distribution is output. The comprehensive field matrix is input into spatial hotspot analysis, temperature gradient and mechanical stress abnormal regions are identified, and abnormal points are output; The abnormal points are spatially clustered through the OPTICS density clustering algorithm, adjacent abnormal points are merged and isolated noise points are filtered, and a clustering core region is output.

4. The distribution transformer monitoring method of claim 3, wherein: ​ ​ ​ Based on the clustering core area, the spatial coordinates, the average temperature gradient and the maximum mechanical stress of the abnormal points are extracted by the spatial statistical feature extraction method to generate an abnormal feature vector.

5. The distribution transformer monitoring method of claim 4, wherein: The abnormal feature vector is input into the pre-trained intelligent diagnosis model to output a comprehensive fault risk score and perform preliminary diagnosis, and the specific steps are as follows: Based on the historical abnormal feature vectors, the DNN deep neural network is trained by the back propagation algorithm method to obtain an intelligent diagnosis model that can output a comprehensive fault risk score; The abnormal feature vector is input into the intelligent diagnosis model to analyze the comprehensive fault risk score of each abnormal area; Based on the historical fault data, the low risk threshold, the high risk threshold and the fault classification rule are defined, and the comprehensive fault risk score is judged to obtain the preliminary diagnosis result.

6. The distribution transformer monitoring method of claim 1, wherein: The final diagnosis result is used for dynamic early warning, and a fault detection report is generated, and the specific steps are as follows: Based on the final diagnosis result, the rule engine is used to automatically match the warning rule library to trigger multi-level early warning; The final diagnosis result, multi-level early warning and power distribution transformer operation state data are spatio-temporally aligned by a feature fusion algorithm to construct structured diagnosis data; Based on the structured diagnosis data, intelligent analysis is performed by a natural language generation method to obtain a fault detection report.

7. A distribution transformer monitoring system based on the distribution transformer monitoring method according to any one of claims 1 to 6, characterized by: It comprises a data acquisition module, a coupling analysis module, an abnormality extraction module, a diagnosis module, an optimization module and a report generation module; The data acquisition module is used for real-time acquisition of power distribution transformer operation state data and pre-processing to obtain a standardized multi-source data set; The coupling analysis module is used for obtaining the three-dimensional temperature gradient field distribution and mechanical stress distribution of the power distribution transformer based on the standardized multi-source data set, and outputting a comprehensive field matrix by field coupling analysis; The abnormality extraction module is used for spatial analysis of the comprehensive field matrix by spatial hotspot analysis, identifying and marking temperature and mechanical stress abnormal areas, and outputting an abnormal feature vector by OPTICS density clustering algorithm for spatial aggregation and noise filtering; The diagnosis module is used for inputting the abnormal feature vector into the pre-trained intelligent diagnosis model to output a comprehensive fault risk score and perform preliminary diagnosis; The optimization module is used for retrieving fault data in the historical case library based on the preliminary diagnosis result for comparison and verification, optimizing the preliminary diagnosis result by a machine learning algorithm, and generating a final diagnosis result; The report generation module is used for dynamic early warning based on the final diagnosis result, and a fault detection report is generated.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that: The processor executes the computer program to realize the steps of the power distribution transformer monitoring method in any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the power distribution transformer monitoring method in any one of claims 1-6.

Images