Fault identification method based on ultrahigh frequency characteristic change

Abstract

The invention discloses a fault identification method based on ultrahigh frequency characteristic change, and relates to the technical field of transformer fault identification, and the method comprises the steps: carrying out the fault probability prediction of a target transformer, and determining a fault identification aging coefficient; performing data value evaluation on sensors deployed on the target transformer according to the working operation information and the regional environment information, and performing value evaluation on the spectrum characteristic parameters to obtain a key sensor group and a key spectrum characteristic parameter group; screening the ultrahigh-frequency electromagnetic wave signal sequence set to obtain a key ultrahigh-frequency electromagnetic wave signal sequence set; and the UHF sensor returns the comprehensive ultrahigh frequency electromagnetic wave signal sequence set to carry out fault fine diagnosis, and a fault identification result is output. The problem that in the prior art, due to the fact that an adaptive signal monitoring and recognition scheme cannot be set according to a transformer operation scene, timeliness and accuracy of transformer fault recognition are insufficient in a method for recognizing transformer faults based on ultrahigh frequency characteristics is solved.

Description

Technical Field

[0001] This application relates to the field of transformer fault identification technology, specifically to a fault identification method based on ultra-high frequency characteristic changes. Background Technology

[0002] With the continuous development of power systems, transformers, as core equipment for power transmission and distribution, are of paramount importance in terms of operational stability and reliability. However, traditional transformer fault identification methods mainly rely on periodic manual inspections and offline tests, which suffer from problems such as long detection cycles and poor real-time performance, making it difficult to meet the needs of modern power systems for real-time monitoring of transformer operating status and rapid fault diagnosis. Summary of the Invention

[0003] This application provides a fault identification method based on changes in UHF frequency characteristics, which solves the technical problem in existing methods for identifying transformer faults based on UHF frequency characteristics that suffer from insufficient timeliness and accuracy in fault identification due to the inability to set up an appropriate signal monitoring and identification scheme according to the transformer operating scenario.

[0004] The technical solution to the above-mentioned technical problems in this application is as follows: In a first aspect, this application provides a fault identification method based on ultra-high frequency characteristic variations, the method comprising: Based on the target transformer’s operation information and regional environmental information within the historical time window, the failure probability of the target transformer is predicted. Based on the predicted failure probability within the preset time zone, the timeliness of fault identification is analyzed, and the fault identification timeliness coefficient is determined. Based on the operational information and regional environmental information, the data value of several UHF sensors deployed on the target transformer is evaluated, and the parameter value of preset spectral characteristic parameters is evaluated. Based on the fault identification timeliness coefficient, key sensor groups and key spectral characteristic parameter groups are selected. Within the preset time zone, UHF electromagnetic wave signals are acquired according to the key sensor group and key spectral characteristic parameter group. Based on the fault identification time efficiency coefficient, a key data screening scheme is set to screen the UHF electromagnetic wave signal sequence set to obtain the key UHF electromagnetic wave signal sequence set. Initial fault diagnosis is performed based on the key UHF electromagnetic wave signal sequence set. If the target transformer is abnormal, detailed fault diagnosis is performed by transmitting a comprehensive UHF electromagnetic wave signal sequence set back from the several UHF sensors, and the fault identification result is output.

[0005] This application provides one or more technical solutions, which have at least the following technical effects or advantages: This application provides a fault identification method based on changes in UHF frequency characteristics. First, fault probability prediction and timeliness analysis are performed based on the transformer's operation and regional environmental information to determine the fault identification timeliness coefficient, making subsequent fault identification more aligned with the actual operating scenario of the transformer and improving the timeliness of identification. Second, the value of UHF sensors and spectral characteristic parameters is evaluated, and key groups are selected. Targeted collection of key data avoids interference from invalid data and improves data accuracy. Then, a key sequence set is obtained by filtering UHF electromagnetic wave signal sequence sets, reducing data processing volume and accelerating fault identification speed. Finally, an initial fault diagnosis is performed first; if anomalies are found, a detailed diagnosis is conducted, ensuring both the comprehensiveness and accuracy of fault identification.

[0006] The above technical solution determines the timeliness coefficient by predicting the fault probability, selects key sensor groups and key spectral feature parameter groups, and rationally filters and analyzes the collected signals to achieve efficient and accurate identification of transformer faults. The signal monitoring and identification scheme is flexibly adjusted according to the actual operating scenario of the transformer, improving the timeliness of fault identification. Furthermore, the accuracy of fault identification is enhanced through the filtering of key data and comprehensive signal feedback analysis. Simultaneously, the fault identification model constructed using a transformer fault mechanism knowledge graph and graph neural network can deeply analyze fault characteristics, improving the accuracy and reliability of fault diagnosis. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart illustrating a fault identification method based on changes in ultra-high frequency characteristics provided in an embodiment of this application. Figure 2 This is a schematic diagram of the process for obtaining a set of key spectral feature parameters in a fault identification method based on changes in ultra-high frequency characteristics provided in an embodiment of this application. Detailed Implementation

[0009] This application provides a fault identification method based on changes in ultra-high frequency characteristics, which addresses the technical problem in existing methods for identifying transformer faults based on ultra-high frequency characteristics, where the inability to set up an appropriate signal monitoring and identification scheme according to the transformer's operating scenario leads to insufficient timeliness and accuracy in transformer fault identification.

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0012] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0013] Example 1, as Figure 1 As shown, this application provides a fault identification method based on ultra-high frequency characteristic changes, including: S10: Based on the target transformer’s operating information and regional environmental information within the historical time window, predict the fault probability of the target transformer, perform fault identification timeliness analysis based on the predicted fault probability within the preset time zone, and determine the fault identification timeliness coefficient. In this embodiment, firstly, operational information and regional environmental information of the target transformer within a historical time window are collected. By collecting historical data and utilizing machine learning algorithms, the probability of failure of the target transformer in a future preset time zone is predicted.

[0014] For example, when a transformer operates under high load for an extended period and the ambient temperature is high, the probability of failure increases significantly. Based on the predicted failure probability, a fault identification timeliness analysis is performed. If the predicted failure probability is high, the timeliness of fault identification needs to be improved, i.e., the fault identification time interval needs to be shortened, and the fault identification timeliness coefficient should be increased accordingly. Conversely, if the predicted failure probability is low, the timeliness of fault identification can be appropriately reduced, the fault identification time interval can be extended, and the fault identification timeliness coefficient can be decreased.

[0015] Specifically, step S10 in the method includes: The target transformer's operational data sequence set and regional environmental data sequence set within the historical time window are monitored and uploaded to the cloud server as operational information and regional environmental information. The operational data includes at least load current, voltage, oil temperature, and winding temperature, and the regional environmental data includes at least electromagnetic interference intensity, relative humidity, and ambient humidity. On the cloud server, a fault probability predictor is used to predict the fault probability of the target transformer within a preset time zone based on the operating information and regional environmental information, and the predicted fault probability is output. The ratio of the preset fault probability scalar of the target transformer to the predicted fault probability is set as the fault identification timeliness coefficient.

[0016] In this embodiment of the application, firstly, the target transformer's operating data sequence set and regional environmental data sequence set within a historical time window are monitored and acquired. The operating data includes load current, voltage, oil temperature, and winding temperature; the regional environmental data includes electromagnetic interference intensity, relative humidity, and ambient humidity. The acquired historical operating data sequence set and regional environmental data sequence set are uploaded to a cloud server as operating information and regional environmental information.

[0017] Among them, load current is an indicator of the transformer's load level. Excessive load current will cause the transformer windings to heat up more, accelerate insulation aging, and increase the possibility of faults. Excessive voltage fluctuations may lead to uneven electric field distribution inside the transformer, causing faults such as partial discharge. Oil temperature reflects the thermal state inside the transformer. Excessive oil temperature means poor heat dissipation or internal short circuit faults. Winding temperature is related to the insulation performance of the windings. Excessive winding temperature will reduce the insulation life and may even lead to insulation breakdown.

[0018] The intensity of electromagnetic interference can affect the acquisition and transmission of UHF electromagnetic wave signals. Strong electromagnetic interference may distort the acquired signals and affect the accuracy of fault identification. Relative temperature and ambient humidity affect the insulation performance of transformers. High humidity environments may cause insulation to become damp, reduce insulation resistance, and increase the risk of leakage and flashover.

[0019] Secondly, in the cloud server, the fault probability predictor is trained based on historical data. For example, a deep learning neural network model is used to establish a mapping relationship between fault probability and operational information and regional environmental information through learning and training on a large amount of historical data. The monitored operational data sequence set and regional environmental data sequence set are uploaded to the cloud server, and the fault probability predictor is used to predict the fault probability of the target transformer within a preset time zone, and then output the predicted fault probability.

[0020] Finally, a preset fault probability scalar is set based on factors such as the transformer's design parameters and historical fault conditions. For example, a standard value of 1% is set, and the ratio of the preset fault probability to the predicted fault probability is calculated to obtain the fault identification timeliness coefficient. The fault identification timeliness coefficient can reflect the requirements for fault identification timeliness under the current conditions.

[0021] For example, when the predicted failure probability is high, the ratio is small, indicating that failure identification needs to be performed in a shorter time, so the failure identification timeliness coefficient is smaller; when the predicted failure probability is low, the ratio is large, and the failure identification timeliness coefficient is correspondingly larger.

[0022] By determining the fault identification timeliness coefficient, the subsequent fault identification process can be flexibly adjusted according to the actual operating conditions of the transformer, thereby improving the timeliness and accuracy of fault identification and better ensuring the stable operation of the transformer.

[0023] The method for constructing the fault probability predictor includes: Based on the historical operation records of similar transformers to the target transformer, a sample operation information set and a sample area environmental information set are collected. The proportion of fault events in different sample operation information and sample area environmental information within the historical time zone is collected as the sample fault probability, thus obtaining the sample fault probability set. Using the sample operational information set and sample area environmental information set as input, and the sample fault probability set as supervision, a long short-term memory network is trained until convergence to obtain a fault probability predictor, which is then deployed on a cloud server.

[0024] In this embodiment of the application, firstly, historical operation records of similar transformers to the target transformer are collected to obtain sample operation information set and sample area environmental information set. At the same time, the proportion of fault events in different sample operation information and sample area environmental information in the historical time zone is statistically analyzed to obtain the sample fault probability.

[0025] Among them, Long Short-Term Memory (LSTM) networks are a special type of recurrent neural network that can effectively process sequential data and capture long-term dependencies in the data. They are suitable for tasks that need to consider historical data and time series, such as fault probability prediction.

[0026] Secondly, the sample operational information set and sample area environmental information set are used as inputs, and the sample failure probability set is used as supervision information to train the Long Short-Term Memory (LSTM) network. During training, the network continuously adjusts its weights and parameters to make the predicted failure probability as close as possible to the actual sample failure probability. When the network's loss function converges to a small value, it indicates that the network has learned the mapping relationship between the input data and the failure probability. At this point, training is complete, and the failure probability predictor is obtained.

[0027] Finally, the trained fault probability predictor is deployed on a cloud server. Leveraging the cloud server's computing and storage capabilities, efficient and accurate prediction of the target transformer's fault probability is achieved. The cloud server can receive real-time operating information of the target transformer and regional environmental information, and quickly calculate the predicted fault probability using the fault probability predictor, providing a reliable basis for subsequent fault identification timeliness coefficient determination and fault identification work.

[0028] For example, a fault probability predictor is constructed and trained based on a long short-term memory network, and the specific steps are as follows: First, data acquisition involves taking the sample operation information set and the sample area environmental information set as inputs and the predicted sample failure probability as output.

[0029] Secondly, in model construction, the number of nodes in the input layer is equal to the dimension of the input features. For example, if there are 7 features in the sample load current, voltage, oil temperature, winding temperature, electromagnetic interference intensity, relative temperature, and ambient humidity, then the input layer contains 7 nodes. Set 1-3 hidden layers, and adjust the number of nodes in each layer through experiments, such as 64, 32, etc. The activation function is ReLU. The number of nodes in the output layer is equal to the number of predicted fault probabilities. The output layer generally does not use an activation function and directly outputs continuous values.

[0030] Finally, for model training, the sample operational information set and the sample regional environmental information set are used as the model input features, and the sample failure probability set is used as the supervision label. The Adam optimizer and the mean squared error (MSE) loss function are used to calculate the loss. The batch size is set to 32 and the total number of training rounds is 100. An early stopping mechanism (patience=5) is introduced. When the validation set loss does not decrease for 5 consecutive rounds, the training process is automatically terminated, and the trained failure probability predictor is obtained. This effectively avoids model overfitting and ensures that the model reaches a convergent state. The trained failure probability predictor can accurately output the data credibility.

[0031] S20: Based on the operational information and regional environmental information, evaluate the data value of several UHF sensors deployed on the target transformer, and evaluate the parameter value of preset spectral characteristic parameters. Based on the fault identification timeliness coefficient, select key sensor groups and key spectral characteristic parameter groups. In this embodiment, firstly, because the contribution of data collected by different UHF sensors to fault identification varies, and the importance of different spectral characteristic parameters in fault diagnosis differs, data value evaluation and parameter value evaluation are performed. For the data value evaluation of UHF sensors, factors such as the sensor's installation location, the stability of the collected data, and its correlation with fault characteristics are comprehensively considered. Sensors installed in critical parts of the transformer, whose collected data exhibits small fluctuations and a high correlation with fault characteristics have relatively high data value. For the parameter value evaluation of preset spectral characteristic parameters, the variation patterns of the parameters under different fault types and their sensitivity to faults are analyzed. Parameters that change significantly when a fault occurs and can effectively distinguish different fault types have greater parameter value.

[0032] A UHF sensor array is deployed, and the data value of each sensor in the array is evaluated based on the transformer's working scenario and environmental parameters, such as fault correlation and signal interference. Multiple key sensor groups are selected, and the parameter value of the spectral characteristic parameters is evaluated, such as fault correlation and signal interference, resulting in multiple spectral characteristic parameter groups.

[0033] Then, after evaluation, the data is filtered based on the previously determined fault identification timeliness coefficient. If the fault identification timeliness coefficient is low, it indicates a need for faster and more accurate fault identification. In this case, sensors with high data value are selected to form the key sensor group, and spectral feature parameters with high parameter value are selected to form the key spectral feature parameter group, ensuring that the most critical and effective data is collected. Conversely, if the fault identification timeliness coefficient is high, the screening criteria can be appropriately relaxed to include more sensors and spectral feature parameters, while ensuring a certain level of fault identification capability.

[0034] The key sensor group and key spectral feature parameter group selected through the above methods improve the targeting and accuracy of fault identification while reducing the amount of data acquisition and processing. The key sensor group can accurately acquire ultra-high frequency electromagnetic wave signals related to the fault, while the key spectral feature parameter group can more effectively extract fault features from the signal, laying a solid foundation for subsequent fault identification work.

[0035] The preset spectral characteristic parameters include the dominant frequency position, bandwidth range, power spectral density, energy distribution characteristics, and spectral centroid.

[0036] In this embodiment, the preset spectral characteristic parameters include the dominant frequency position, bandwidth range, power spectral density, energy distribution characteristics, and spectral centroid. The dominant frequency position is the frequency position where energy is most concentrated in the ultra-high frequency signal, and different fault types often correspond to different dominant frequency positions. For example, a partial discharge fault may cause the dominant frequency position to shift. By monitoring the change in the dominant frequency position, it is possible to preliminarily determine whether a fault has occurred.

[0037] The bandwidth range reflects the frequency range covered by the UHF signal. When a fault occurs, the bandwidth range of the signal may become wider or narrower. The power spectral density represents the power distribution of the signal at various frequencies. When a transformer fails, the power spectral density at certain frequencies will change. The energy distribution characteristics describe the proportion of signal energy distributed in different frequency bands. Different faults will lead to a redistribution of energy at frequencies. The spectral centroid is the location of the energy center of the spectrum. Changes in the spectral centroid reflect the overall frequency shift trend.

[0038] Specifically, step S20 in the method includes: Several UHF sensors are deployed at several key locations of the target transformer. The UHF sensors are either built-in or external UHF sensors, and the number of sensors is not less than 5. Based on the historical operating records of similar transformers, the correlation between the aforementioned key locations and transformer faults is analyzed, and several fault correlation coefficients are output. Based on several key locations and sensor attribute information of the aforementioned UHF sensors, data interference intensity simulation is performed on the aforementioned UHF sensors according to the operational information and regional environmental information, and several data reliability coefficients are output. The data value of the several UHF sensors is evaluated based on the several fault correlation coefficients and several data reliability coefficients, resulting in several data value coefficients. The data value coefficients are positively correlated with the fault correlation coefficients and the data reliability coefficients. Based on the historical operating records of similar transformers, the correlation between multiple spectral feature parameters and transformer fault identification in the preset spectral feature parameters is analyzed to obtain the correlation coefficients of multiple parameters. Based on the operational information and regional environmental information, the reliability of the multiple spectral characteristic parameters is evaluated, and multiple parameter reliability coefficients are output. The parameter value is evaluated based on the correlation coefficients and confidence coefficients of the multiple parameters, and multiple parameter value coefficients are output. The parameter value coefficients are positively correlated with the correlation coefficients and confidence coefficients of the parameters.

[0039] In this embodiment, firstly, several UHF sensors are deployed at key locations on the target transformer. These sensors are either built-in or external, and the number is no less than five. The selection of these key locations affects whether the sensors can accurately acquire UHF electromagnetic wave signals related to the fault.

[0040] Built-in UHF sensors can get closer to the fault source inside the transformer and can more sensitively capture weak fault signals, but installation and maintenance are relatively complex; external UHF sensors are easy to install and can be flexibly adjusted in position as needed, but may be affected by external environmental interference.

[0041] Secondly, based on the historical operating records of similar transformers, the correlation between each key location and transformer faults is analyzed to obtain several fault correlation coefficients. This process requires in-depth mining and analysis of a large amount of historical data to identify patterns in the behavior of different locations during fault occurrences. For example, if significant signal changes are detected at certain locations in multiple faults, then the fault correlation coefficient is relatively high.

[0042] Simultaneously, based on the key locations and sensor attribute information of the UHF sensors, combined with operational information and regional environmental information, a three-dimensional simulation of data interference intensity was performed on each sensor, yielding several data reliability coefficients. Sensors can be built-in or external, and vary in type. Sensor attribute information includes sensitivity, anti-interference capability, etc., while operational information and regional environmental information, such as load current and electromagnetic interference intensity, all affect data reliability. In environments with strong electromagnetic interference, the data collected by the sensors may be distorted, thus reducing the data reliability coefficients.

[0043] Then, the data value of the UHF sensors is evaluated based on the fault correlation coefficient and the data reliability coefficient, resulting in several data value coefficients. The data value coefficient is positively correlated with the fault correlation coefficient and the data reliability coefficient; that is, the higher the fault correlation coefficient and the data reliability coefficient, the greater the data value coefficient. This allows for the selection of sensors with high data value, providing more reliable data for subsequent fault identification.

[0044] For the preset spectral characteristic parameters, based on the historical operating records of similar transformers, the correlation between multiple spectral characteristic parameters and transformer fault identification is analyzed to obtain the correlation coefficients of multiple parameters. Different spectral characteristic parameters behave differently under different fault types. By analyzing historical data, the degree of correlation between each parameter and the fault can be found.

[0045] Secondly, based on operational and regional environmental information, the reliability of multiple spectral characteristic parameters is evaluated, and multiple parameter reliability coefficients are output. For example, under high temperature and high humidity environments, some spectral characteristic parameters may be affected, and their reliability will decrease.

[0046] Finally, parameter value is evaluated based on parameter correlation coefficients and parameter confidence coefficients, resulting in multiple parameter value coefficients. The parameter value coefficients are positively correlated with both the parameter correlation coefficients and parameter confidence coefficients. High-value spectral feature parameters are selected for subsequent fault identification.

[0047] Furthermore, based on the fault identification timeliness coefficient, key sensor groups and key spectral feature parameter groups are obtained through screening, including: The fault identification timeliness coefficient is multiplied by the initial number of sensors and the initial number of spectral feature parameters and then rounded to obtain the number of key sensors and the number of key feature parameters. The initial number of sensors is half the number of deployed sensors, the initial number of spectral feature parameters is 3, and the number of key sensors is greater than or equal to 2 and the number of key feature parameters is greater than or equal to 2. Based on the aforementioned data value coefficients, the key sensors are selected from largest to smallest to obtain multiple key sensors and construct a key sensor group. Based on the value coefficients of the multiple parameters, the key spectral feature parameters are filtered from largest to smallest to obtain a group of key spectral feature parameters.

[0048] In this embodiment, the number of key sensors and key feature parameters are first determined based on the fault identification timeliness coefficient. The initial number of sensors is half the number of deployed sensors, and the initial number of spectral feature parameters is set to 3. The fault identification timeliness coefficient is multiplied by the initial number of sensors and the initial number of spectral feature parameters, and then rounded down. At the same time, it is ensured that the number of key sensors is greater than or equal to 2, and the number of key feature parameters is greater than or equal to 2, so as to reasonably determine the number of sensors and spectral feature parameters used for subsequent fault identification under different fault identification timeliness requirements.

[0049] For example, if the fault identification timeliness coefficient is 0.6 and the number of deployed sensors is 10, the initial number of sensors is 5. After multiplying, we get 3, which meets the requirement that the number of critical sensors is greater than or equal to 2. The initial number of spectral feature parameters is 3, which are multiplied and rounded down to 2, satisfying the requirement that the number of key feature parameters is greater than or equal to 2.

[0050] Then, based on several data value coefficients, key sensors are selected in descending order of quantity. The data value coefficients combine fault correlation coefficients and data reliability coefficients, reflecting the importance of the data collected by each sensor for fault identification. Sensors with higher data value coefficients are then grouped together to construct key sensor groups.

[0051] Finally, based on multiple parameter value coefficients, key feature parameters are selected from largest to smallest quantity. The parameter value coefficients are positively correlated with the parameter correlation coefficient and parameter confidence coefficient, reflecting the importance of each spectral feature parameter in fault identification. Spectral feature parameters with higher parameter value coefficients are selected to form a key spectral feature parameter group. This key spectral feature parameter group can more effectively extract fault features from the acquired signals, helping to improve the accuracy and efficiency of fault identification.

[0052] S30: Within the preset time zone, UHF electromagnetic wave signals are acquired according to the key sensor group and key spectral characteristic parameter group. Based on the fault identification time efficiency coefficient, a key data screening scheme is set to screen the UHF electromagnetic wave signal sequence set to obtain the key UHF electromagnetic wave signal sequence set. In this embodiment, firstly, within a preset time zone, UHF electromagnetic wave signals are acquired according to the selected key sensor group and key spectral feature parameter group. Based on the fault identification time efficiency coefficient, a key data screening scheme is set to screen the UHF electromagnetic wave signal sequence set. Combined with the transformer's working scenario and environmental parameters, fault probability prediction is performed. Based on the probability prediction results, the proportion of key indicator extraction is optimized for data dimensionality reduction.

[0053] Then, initial fault identification is performed based on multiple key parameter sequences from multiple key sensors. If an anomaly is found, comprehensive data collection is then conducted for accurate identification.

[0054] Specifically, step S30 in the method includes: Within the preset time zone, the corresponding UHF sensor is activated according to the key sensor group to collect ultra-high frequency electromagnetic wave signals. In the edge processing unit of the target transformer, the collection results are filtered according to the key spectral feature parameter group to obtain a set of ultra-high frequency electromagnetic wave signal sequences. The product of the fault identification timeliness coefficient and the preset data extraction ratio is used as the adaptation data extraction ratio, wherein the preset data extraction ratio is 50%. According to the specified adaptation data extraction ratio, the UHF electromagnetic wave signal sequence set is subjected to data similarity dimensionality reduction to obtain the key UHF electromagnetic wave signal sequence set, which is then uploaded to the cloud server.

[0055] In this embodiment, firstly, within a preset time zone, the corresponding UHF sensors are activated according to the key sensor group to collect ultra-high frequency electromagnetic wave signals. The key sensors are selected through the above steps and can accurately collect signals related to the fault. After the acquisition is completed, in the edge processing unit of the target transformer, the acquisition results are filtered according to the key spectral feature parameter group, and only the acquisition results corresponding to the key parameters are retained, thereby obtaining a set of ultra-high frequency electromagnetic wave signal sequences.

[0056] Then, the fault identification timeliness coefficient is multiplied by the preset data extraction ratio to obtain the appropriate data extraction ratio. The preset data extraction ratio is set at 50%, and by combining it with the fault identification timeliness coefficient, the data extraction ratio can be flexibly adjusted according to different timeliness requirements. If the fault identification timeliness coefficient is large, it indicates that the timeliness requirement for fault identification is not so high, and the appropriate data extraction ratio may be relatively low; conversely, if the fault identification timeliness coefficient is small, the appropriate data extraction ratio will be relatively high to ensure that key data can be obtained in a timely manner.

[0057] Subsequently, the UHF electromagnetic wave signal sequence set is subjected to data similarity dimensionality reduction according to the appropriate data extraction ratio. Data similarity dimensionality reduction can remove redundant data, reduce the amount of data, and retain key information. During the dimensionality reduction process, the data in the signal sequence set is analyzed, only key data is saved, similar data is identified and merged or removed, and finally the key UHF electromagnetic wave signal sequence set is obtained, improving the efficiency of initial diagnosis.

[0058] For example, Radapt performs data similarity dimensionality reduction on the UHF electromagnetic wave signal sequence set according to the following steps: First, a predefined spectral feature vector is extracted for each UHF signal sequence, such as dominant frequency, bandwidth, energy distribution, and spectral centroid. Second, the similarity between sequences is calculated based on the feature vector, such as cosine similarity, dynamic time warping, or similarity based on spectral distance. Then, clustering is performed according to similarity, such as K-means, spectral clustering, or hierarchical clustering, or representative samples are extracted based on similarity. Under the constraint of maintaining overall similarity and information entropy, representative sequences are selected from each type of sensor according to the Radapt ratio to form a key UHF electromagnetic wave signal sequence set.

[0059] Finally, the key ultra-high frequency electromagnetic wave signal sequence set is uploaded to a cloud server. The cloud server, with its computing and storage capabilities, can further analyze and process the critical data. On the cloud server, advanced algorithms and models can be used for in-depth data mining, and combined with previous fault probability prediction results, the extraction ratio of key indicators can be optimized, thereby more accurately identifying transformer fault conditions and providing strong support for the safe and stable operation of the transformer.

[0060] S40: Perform initial fault diagnosis based on the key UHF electromagnetic wave signal sequence set. If the target transformer is abnormal, perform detailed fault diagnosis by transmitting the comprehensive UHF electromagnetic wave signal sequence set back by the several UHF sensors, and output the fault identification result.

[0061] In this embodiment, a preliminary fault diagnosis is first performed based on a set of key ultra-high frequency electromagnetic wave signal sequences. This set of key ultra-high frequency electromagnetic wave signal sequences is filtered and dimensionality-reduced data containing information closely related to transformer faults. The data is analyzed using a cloud server, and by comparing the characteristic data during normal operation with the currently collected key data, it is determined whether the target transformer exhibits any abnormalities.

[0062] Furthermore, if an anomaly is detected in the target transformer during the initial diagnosis, a more precise fault diagnosis is required. This is achieved by transmitting a comprehensive set of ultra-high frequency electromagnetic wave signal sequences from several previously deployed UHF sensors. This comprehensive set of ultra-high frequency electromagnetic wave signal sequences contains richer and more detailed information about ultra-high frequency electromagnetic waves.

[0063] Then, after receiving a comprehensive set of ultra-high frequency electromagnetic wave signal sequences, more complex and precise algorithms and models are used for detailed fault diagnosis. A thorough analysis of various characteristic parameters of the signal is conducted, combined with historical fault data and operating conditions of the transformer, to accurately determine the type, location, and severity of the fault. For example, by analyzing changes in parameters such as the signal's dominant frequency, bandwidth range, power spectral density, energy distribution characteristics, and spectral centroid, it can be determined whether the fault is a partial discharge fault, an insulation fault, or another type of fault.

[0064] Finally, after detailed fault diagnosis, the final fault identification result is output. The fault identification result should indicate the specific nature of the fault, providing accurate guidance for subsequent repair and maintenance work. Simultaneously, the fault identification result and related data are stored and recorded to facilitate subsequent tracking and analysis of the transformer's operating status, continuously optimizing the fault identification method and improving the accuracy of fault diagnosis.

[0065] Specifically, step S40 in the method includes: Based on the preset signal warning threshold, the key ultra-high frequency electromagnetic wave signal sequence set is traversed and judged. If the target transformer does not have any abnormalities, monitoring continues according to the key sensor group and key spectral characteristic parameter group. If the target transformer is abnormal, the aforementioned UHF sensors are activated to collect ultra-high frequency electromagnetic wave signals, and a comprehensive set of ultra-high frequency electromagnetic wave signal sequences is obtained and uploaded to the cloud server. Within the cloud server, a transformer fault identification model is used to perform detailed fault diagnosis on the comprehensive ultra-high frequency electromagnetic wave signal sequence set and output the fault identification results. The transformer fault identification model is constructed based on a transformer fault mechanism knowledge graph and a graph neural network.

[0066] In this embodiment, firstly, the key UHF electromagnetic wave signal sequence set is traversed and judged based on a preset signal warning threshold. The preset signal warning threshold is determined based on a large amount of experimental data and actual operating experience, and serves as the basis for judging whether the transformer has any abnormalities. When traversing the key UHF electromagnetic wave signal sequence set, the characteristic parameters of each signal in the sequence set are compared with the preset signal warning threshold. If the characteristic parameters of all signals are within the warning threshold range, it indicates that the target transformer has no abnormalities. At this time, monitoring continues according to the key sensor group and the key spectral characteristic parameter group to continuously monitor the transformer's operating status.

[0067] Furthermore, if during the traversal, the characteristic parameters of certain signals exceed the preset signal warning threshold, it indicates an anomaly in the target transformer. At this point, several previously deployed UHF sensors are immediately activated to acquire ultra-high frequency electromagnetic wave signals. The UHF sensors are distributed at different key locations on the transformer, acquiring a comprehensive set of ultra-high frequency electromagnetic wave signal sequences by collecting ultra-high frequency electromagnetic wave signals around the transformer. After acquisition, the comprehensive set of ultra-high frequency electromagnetic wave signal sequences is uploaded to the cloud server.

[0068] Then, within the cloud server, a transformer fault identification model built on a transformer fault mechanism knowledge graph and graph neural network is used to perform detailed fault diagnosis on a comprehensive set of UHF electromagnetic wave signal sequences. The transformer fault mechanism knowledge graph integrates various knowledge and information about transformer faults, including typical partial discharge types, discharge source location characteristics, discharge development trends, and environmental interference patterns, which are used to assist in identifying different types of transformer faults. The graph neural network has powerful learning and reasoning capabilities, enabling it to process complex graph-structured data. By inputting the comprehensive set of UHF electromagnetic wave signal sequences into the model, the model analyzes various characteristic parameters of the signals and, combined with information from the knowledge graph, accurately determines the type, location, and severity of the fault.

[0069] For example, a transformer fault identification model is constructed based on a transformer fault mechanism knowledge graph and a graph neural network. The steps are as follows: First, knowledge acquisition involves collecting operating logs, partial discharge monitoring data, and key operating parameters such as temperature, load, voltage, and current from the target transformer and similar equipment. Second, fault case collection involves extracting typical fault modes from maintenance records, repair reports, and accident analyses, such as inter-turn short circuits, insulation breakdown, bushing discharge, and gas evolution in oil.

[0070] Secondly, knowledge representation involves representing the collected knowledge in a graph format, constructing a "node-relationship-attribute" triple. Node types include fault type nodes, such as winding turn-to-turn short circuits, insulation partial discharge, surface creepage, and oil breakdown; signal characteristic nodes, such as UHF amplitude, energy spectrum distribution, pulse repetition rate, and discharge spectrum bandwidth; environmental interference nodes, such as electromagnetic noise sources, wireless communication interference, lightning electromagnetic pulses, and switch operation interference; and operating condition nodes, such as high load, temperature rise, and excessive gas in oil. Relationship types include trigger relationships, characterization relationships, interference relationships, and evolutionary relationships. Attribute information consists of attributes attached to each node and relationship, such as occurrence probability, eigenvalue, and confidence level.

[0071] Secondly, knowledge fusion and cleaning involve denoising, aligning, and unifying the encoding of knowledge from different data sources. For example, spectral energy is unified to dBm, and temperature sequences are unified to degrees Celsius. Rule-based reasoning and machine learning are used to fuse redundant knowledge from different sources and eliminate conflicts, such as through similarity calculation and graph embedding.

[0072] Finally, knowledge reasoning and updating utilize graph neural networks or rule-based reasoning mechanisms to achieve automatic matching from signal features to fault types. The knowledge map is dynamically updated based on newly acquired UHF data and operational feedback to ensure the model's real-time performance and adaptability.

[0073] The final output fault identification result, through "dominant frequency position + bandwidth + energy distribution characteristics + spectral centroid", can distinguish different discharge types, indicate the specific circumstances of the fault, the location of the fault, and the severity of the fault. For example, fault types include corona discharge faults, surface discharge faults, internal discharge faults, or arc discharge.

[0074] For example, corona discharge faults have a low dominant frequency position, a narrow bandwidth, low-frequency energy dominance, concentrated energy, and relatively small amplitude; surface discharge faults have a medium dominant frequency position, a medium bandwidth, both high-frequency and low-frequency energy, a large number of spectral peaks, and a slightly wider distribution; internal partial discharge faults have a high dominant frequency position, a narrow bandwidth, high-frequency energy dominance, complex spectral peaks, and strong transient signals; arc discharge faults have a high dominant frequency position, the widest bandwidth, significant high-frequency energy, sharp spectral peaks, high energy, and irregular waveforms.

[0075] In summary, the embodiments of this application have at least the following technical effects: This application provides a fault identification method based on changes in UHF frequency characteristics. First, fault probability prediction and timeliness analysis are performed based on the transformer's operation and regional environmental information to determine a fault identification timeliness coefficient, making subsequent fault identification more aligned with the actual operating scenario of the transformer and improving the timeliness of identification. Second, the value of UHF sensors and spectral characteristic parameters is evaluated, and key groups are selected. This allows for targeted collection of key data, avoiding interference from invalid data and improving data accuracy. Then, a key sequence set is obtained by filtering UHF electromagnetic wave signal sequence sets, reducing data processing volume and accelerating fault identification speed. Finally, an initial fault diagnosis is performed first, followed by a detailed diagnosis if anomalies are found, ensuring both comprehensiveness and accuracy of fault identification. Through the above technical solution, by predicting fault probability to determine the timeliness coefficient, selecting key sensor groups and key spectral characteristic parameter groups, and rationally filtering and analyzing the collected signals, efficient and accurate identification of transformer faults is achieved. The signal monitoring and identification scheme is flexibly adjusted according to the actual operating scenario of the transformer, improving the timeliness of fault identification. Furthermore, the accuracy of fault identification is enhanced through the filtering of key data and comprehensive signal feedback analysis. Meanwhile, the fault identification model built using transformer fault mechanism knowledge graph and graph neural network can deeply analyze fault characteristics and improve the accuracy and reliability of fault diagnosis.

[0076] Among them, the fault identification timeliness coefficient is processed to obtain a set of key spectral feature parameters, such as... Figure 2 As shown, it includes: The fault identification timeliness coefficient is multiplied by the initial number of sensors and the initial number of spectral feature parameters and then rounded to obtain the number of key sensors and the number of key feature parameters. The initial number of sensors is half the number of deployed sensors, the initial number of spectral feature parameters is 3, and the number of key sensors is greater than or equal to 2 and the number of key feature parameters is greater than or equal to 2. Based on the aforementioned data value coefficients, the key sensors are selected from largest to smallest to obtain multiple key sensors and construct a key sensor group. Based on the value coefficients of the multiple parameters, the key spectral feature parameters are filtered from largest to smallest to obtain a group of key spectral feature parameters.

[0077] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0078] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0079] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A fault identification method based on ultra-high frequency characteristic changes, characterized in that the method... include: Based on the target transformer’s operation information and regional environmental information within the historical time window, the failure probability of the target transformer is predicted. Based on the predicted failure probability within the preset time zone, the timeliness of fault identification is analyzed, and the fault identification timeliness coefficient is determined. Based on the operational information and regional environmental information, the data value of several UHF sensors deployed on the target transformer is evaluated, and the parameter value of preset spectral characteristic parameters is evaluated. Based on the fault identification timeliness coefficient, key sensor groups and key spectral characteristic parameter groups are selected. Within the preset time zone, UHF electromagnetic wave signals are acquired according to the key sensor group and key spectral characteristic parameter group. Based on the fault identification time efficiency coefficient, a key data screening scheme is set to screen the UHF electromagnetic wave signal sequence set to obtain the key UHF electromagnetic wave signal sequence set. Initial fault diagnosis is performed based on the key UHF electromagnetic wave signal sequence set. If the target transformer is abnormal, detailed fault diagnosis is performed by transmitting a comprehensive UHF electromagnetic wave signal sequence set back from the several UHF sensors, and the fault identification result is output.

2. The fault identification method based on ultra-high frequency characteristic changes according to claim 1, characterized in that, Based on the target transformer's operational information and regional environmental information within a historical time window, the failure probability of the target transformer is predicted. Based on the predicted failure probability within a preset time zone, a fault identification timeliness analysis is performed to determine the fault identification timeliness coefficient, including: The target transformer's operational data sequence set and regional environmental data sequence set within the historical time window are monitored and uploaded to the cloud server as operational information and regional environmental information. The operational data includes at least load current, voltage, oil temperature, and winding temperature, and the regional environmental data includes at least electromagnetic interference intensity, relative humidity, and ambient humidity. On the cloud server, a fault probability predictor is used to predict the fault probability of the target transformer within a preset time zone based on the operating information and regional environmental information, and the predicted fault probability is output. The ratio of the preset fault probability scalar of the target transformer to the predicted fault probability is set as the fault identification timeliness coefficient.

3. The fault identification method based on ultra-high frequency characteristic changes according to claim 2, characterized in that, The method for constructing the fault probability predictor includes: Based on the historical operation records of similar transformers to the target transformer, a sample operation information set and a sample area environmental information set are collected. The proportion of fault events in different sample operation information and sample area environmental information within the historical time zone is collected as the sample fault probability, thus obtaining the sample fault probability set. Using the sample operational information set and sample area environmental information set as input, and the sample fault probability set as supervision, a long short-term memory network is trained until convergence to obtain a fault probability predictor, which is then deployed on a cloud server.

4. The fault identification method based on ultra-high frequency characteristic changes according to claim 1, characterized in that, Based on the operational information and regional environmental information, the data value of several UHF sensors deployed on the target transformer is evaluated, and the parameter value of preset spectral characteristic parameters is evaluated, including: Several UHF sensors are deployed at several key locations of the target transformer. The UHF sensors are either built-in or external UHF sensors, and the number of sensors is not less than 5. Based on the historical operating records of similar transformers, the correlation between the aforementioned key locations and transformer faults is analyzed, and several fault correlation coefficients are output. Based on several key locations and sensor attribute information of the aforementioned UHF sensors, data interference intensity simulation is performed on the aforementioned UHF sensors according to the operational information and regional environmental information, and several data reliability coefficients are output. The data value of the several UHF sensors is evaluated based on the several fault correlation coefficients and several data reliability coefficients, resulting in several data value coefficients. The data value coefficients are positively correlated with the fault correlation coefficients and the data reliability coefficients. Based on the historical operating records of similar transformers, the correlation between multiple spectral feature parameters and transformer fault identification in the preset spectral feature parameters is analyzed to obtain the correlation coefficients of multiple parameters. Based on the operational information and regional environmental information, the reliability of the multiple spectral characteristic parameters is evaluated, and multiple parameter reliability coefficients are output. The parameter value is evaluated based on the correlation coefficients and confidence coefficients of the multiple parameters, and multiple parameter value coefficients are output. The parameter value coefficients are positively correlated with the correlation coefficients and confidence coefficients of the parameters.

5. The fault identification method based on ultra-high frequency characteristic changes according to claim 4, characterized in that, The preset spectral characteristic parameters include the dominant frequency position, bandwidth range, power spectral density, energy distribution characteristics, and spectral centroid.

6. The fault identification method based on ultra-high frequency characteristic changes according to claim 4, characterized in that, Based on the fault identification timeliness coefficient, key sensor groups and key spectral feature parameter groups are obtained, including: The fault identification timeliness coefficient is multiplied by the initial number of sensors and the initial number of spectral feature parameters and then rounded to obtain the number of key sensors and the number of key feature parameters. The initial number of sensors is half the number of deployed sensors, the initial number of spectral feature parameters is 3, and the number of key sensors is greater than or equal to 2 and the number of key feature parameters is greater than or equal to 2. Based on the aforementioned data value coefficients, the key sensors are selected from largest to smallest to form a key sensor group. Based on the value coefficients of the multiple parameters, the key spectral feature parameter groups are obtained by filtering according to the number of key feature parameters from largest to smallest.

7. The fault identification method based on ultra-high frequency characteristic changes according to claim 1, characterized in that, UHF electromagnetic wave signals are acquired according to the aforementioned key sensor group and key spectral characteristic parameter group. Based on the fault identification timeliness coefficient, a key data screening scheme is set to screen the UHF electromagnetic wave signal sequence set, resulting in a key UHF electromagnetic wave signal sequence set, including: Within the preset time zone, the corresponding UHF sensor is activated according to the key sensor group to collect ultra-high frequency electromagnetic wave signals. In the edge processing unit of the target transformer, the collection results are filtered according to the key spectral feature parameter group to obtain a set of ultra-high frequency electromagnetic wave signal sequences. The product of the fault identification timeliness coefficient and the preset data extraction ratio is used as the adaptation data extraction ratio, wherein the preset data extraction ratio is 50%. According to the specified adaptation data extraction ratio, the UHF electromagnetic wave signal sequence set is subjected to data similarity dimensionality reduction to obtain the key UHF electromagnetic wave signal sequence set, which is then uploaded to the cloud server.

8. The fault identification method based on ultra-high frequency characteristic changes according to claim 1, characterized in that, Initial fault diagnosis is performed based on the key UHF electromagnetic wave signal sequence set. If the target transformer is abnormal, a detailed fault diagnosis is performed by transmitting a comprehensive UHF electromagnetic wave signal sequence set back from the aforementioned UHF sensors, and the fault identification results are output, including: Based on the preset signal warning threshold, the key ultra-high frequency electromagnetic wave signal sequence set is traversed and judged. If the target transformer does not have any abnormalities, monitoring continues according to the key sensor group and key spectral characteristic parameter group. If the target transformer is abnormal, the aforementioned UHF sensors are activated to collect ultra-high frequency electromagnetic wave signals, and a comprehensive set of ultra-high frequency electromagnetic wave signal sequences is obtained and uploaded to the cloud server. Within the cloud server, a transformer fault identification model is used to perform detailed fault diagnosis on the comprehensive ultra-high frequency electromagnetic wave signal sequence set and output the fault identification results. The transformer fault identification model is constructed based on a transformer fault mechanism knowledge graph and a graph neural network.

Images