Transformer fault early warning method and system under multi-source data

Abstract

The invention relates to the technical field of transformer fault early warning, and provides a transformer fault early warning method and system under multi-source data. The method comprises the following steps: carrying out fault-free temperature distribution prediction on environment temperature and transformer control parameters to obtain reference temperature distribution information; performing fault-free audio feature prediction on the transformer control parameters to obtain reference audio feature information; performing fault-free electrical parameter prediction on the transformer control parameters to obtain reference electrical characteristic information; performing fault prediction based on the reference temperature distribution information, the reference audio feature information and the reference electrical feature information to obtain a fault prediction type; and when the fault prediction type is non-null, generating a transformer fault early warning signal to give an alarm. According to the method and the device, the technical problem of insufficient analysis result accuracy caused by single feature information source in traditional transformer fault analysis is solved, and the technical effect of improving transformer fault prediction accuracy and early warning timeliness through a multi-source data fusion and comprehensive prediction method is realized.

Description

Technical Field

[0001] The present application relates to the technical field of transformer fault early warning, and in particular to a transformer fault early warning method and system under multi-source data. Background Art

[0002] Transformers are vital equipment in power systems, and their operating status is directly related to the stability and safety of the entire power grid. However, in the long-term operation process, transformers are easily affected by multiple factors such as load fluctuations, ambient temperature changes, and equipment aging, resulting in failures. These failures may not only cause damage to the equipment itself, but also cause large-scale power outages in the power system. Therefore, how to effectively carry out transformer fault warnings and reduce the occurrence of failures has always been an important research direction in the power field. Traditional transformer fault analysis methods mainly rely on a single data source, such as temperature monitoring data, electrical parameters, or partial discharge signals. These methods can reflect the operating status of the transformer to a certain extent, but due to the single data source, their analysis capabilities have obvious limitations. The fault information of a single data source often cannot fully cover the various characteristics of the transformer operation. For example, temperature data can reflect the characteristics of thermal faults, but it is difficult to effectively identify mechanical faults, electrical short circuits, and other types of faults. In addition, environmental interference such as external temperature changes may mask the real fault signals inside the transformer, further reducing the accuracy of fault diagnosis. At the same time, traditional methods rely on a fixed threshold judgment mechanism and cannot adapt to complex and changing operating environments, resulting in insufficient real-time and accuracy of fault warnings. Summary of the invention

[0003] The present application provides a transformer fault early warning method and system under multi-source data, aiming to solve the technical problem that the source of characteristic information in traditional transformer fault analysis is single, resulting in insufficient accuracy of analysis results.

[0004] In view of the above problems, the present application provides a transformer fault early warning method and system under multi-source data.

[0005] The first aspect disclosed in the present application provides a transformer fault warning method under multi-source data, the method comprising: predicting the fault-free temperature distribution of ambient temperature and transformer control parameters to obtain transformer benchmark temperature distribution information; predicting the fault-free audio characteristics of the transformer control parameters to obtain transformer benchmark audio characteristic information; predicting the fault-free electrical parameters of the transformer control parameters to obtain transformer benchmark electrical characteristic information; predicting faults based on the transformer benchmark temperature distribution information, the transformer benchmark audio characteristic information and the transformer benchmark electrical characteristic information to obtain a transformer fault prediction type; and generating a transformer fault warning signal for alarm when the transformer fault prediction type is non-empty.

[0006] Another aspect disclosed in the present application provides a transformer fault warning system under multi-source data, the system comprising: a fault-free temperature distribution prediction module: predicting the fault-free temperature distribution of the ambient temperature and the transformer control parameters to obtain the transformer benchmark temperature distribution information; a fault-free audio feature prediction module: predicting the fault-free audio features of the transformer control parameters to obtain the transformer benchmark audio feature information; a fault-free electrical parameter prediction module: predicting the fault-free electrical parameters of the transformer control parameters to obtain the transformer benchmark electrical feature information; a fault prediction module: predicting the fault based on the transformer benchmark temperature distribution information, the transformer benchmark audio feature information and the transformer benchmark electrical feature information to obtain the transformer fault prediction type; a fault alarm module: generating a transformer fault warning signal for alarm when the transformer fault prediction type is non-empty.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: The transformer fault early warning method under the above multi-source data predicts the temperature distribution of the transformer in a fault-free state by analyzing the ambient temperature and transformer control parameters, thereby obtaining its benchmark temperature distribution information and providing a basis for subsequent fault judgment; further predicts the fault-free audio characteristics of the transformer control parameters and extracts its benchmark audio feature information to ensure that the acoustic changes that may be caused by abnormal operating conditions can be captured; at the same time, predicts the fault-free electrical parameters of the transformer control parameters and generates benchmark electrical feature information to cover the electrical behavior characteristics of the equipment operation; based on the fusion of the above multi-source benchmark information, the difference between the current state and the benchmark state of the transformer is comprehensively analyzed to predict the fault type, effectively improving the accuracy and comprehensiveness of fault diagnosis; when the prediction result shows the existence of a potential fault, a fault warning signal is generated and an alarm is issued, thereby realizing early detection and response to potential fault hazards.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1The figure is a flow chart of a transformer fault early warning method under multi-source data in one embodiment.

[0011] Figure 2 The figure is an architecture diagram of a transformer fault warning system under multi-source data in one embodiment.

[0012] Explanation of the accompanying drawings: fault-free temperature distribution prediction module 1, fault-free audio feature prediction module 2, fault-free electrical parameter prediction module 3, fault prediction module 4, fault alarm module 5. DETAILED DESCRIPTION

[0013] The embodiments of the present application provide a transformer fault early warning method and system under multi-source data to solve the technical problem that the source of characteristic information in traditional transformer fault analysis is single, resulting in insufficient accuracy of analysis results.

[0014] The following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0015] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0016] Embodiment 1, as Figure 1 As shown, the present application provides a transformer fault early warning method under multi-source data, the method comprising: The fault-free temperature distribution of the ambient temperature and transformer control parameters is predicted to obtain the transformer benchmark temperature distribution information.

[0017] In an embodiment of the present application, the system terminal collects the ambient temperature data and related control parameters of the transformer during operation, such as load current, cooling system status, etc., and uses the temperature distribution information of fault-free samples of the same model under a fault-free state to perform hierarchical clustering of single-sample temperature distribution and aggregation of multiple samples at the same position, calculates the theoretical temperature distribution of different parts of the transformer under various operating conditions, and generates the reference temperature distribution information of the transformer under a fault-free state. This reference data can comprehensively reflect the impact of ambient temperature, heat dissipation efficiency and load status on the transformer temperature, and provide a comparative basis for discovering temperature deviations caused by poor heat dissipation, overload or abnormal external environment in the subsequent real-time monitoring process, thereby improving the early identification capability of thermal faults.

[0018] Furthermore, the present application provides a method for predicting the fault-free temperature distribution of the ambient temperature and the transformer control parameters to obtain the transformer reference temperature distribution information, including: Taking the ambient temperature and the transformer control parameters as constraints, the fault-free sample temperature distribution information of the transformer model is collected; the fault-free sample temperature distribution information is traversed to perform single sample temperature distribution hierarchical clustering to obtain the fault-free sample update temperature distribution information; the fault-free sample update temperature distribution information is aggregated with multiple samples at the same position to obtain the transformer reference temperature distribution information.

[0019] Preferably, the system terminal first screens out qualified transformer operating condition data according to the variation range of ambient temperature and transformer control parameters (such as load current, cooling fan status, etc.). These data are all from the operation records of the same model in a fault-free state to ensure their authenticity and stability. By arranging multi-point temperature sensors, the temperature information of different parts of the transformer (such as windings, cores, oil tanks, etc.) is collected to form a temperature distribution data set of fault-free samples; then, the fault-free sample temperature distribution information is traversed, and in each traversal, the current fault-free sample temperature distribution information is compared with the temperature deviation of the neighbors until any neighboring temperature difference is greater than the preset temperature deviation threshold, thereby forming the updated temperature distribution information of the fault-free sample; then, in the updated fault-free temperature distribution data of all samples, grouping and aggregation are performed according to the position of the temperature measurement point, for example, the temperature characteristics of fixed positions such as the top, bottom and core of the transformer winding are integrated respectively, and the reference temperature distribution value of each position is generated by using a statistical method. At the same time, the temperature variation range of each measurement point in a fault-free state is recorded as the reference range for subsequent fault analysis. Through the above process, the transformer benchmark temperature distribution information finally generated can accurately reflect the stable temperature state of the transformer under fault-free conditions, providing a reliable reference standard for real-time monitoring and fault diagnosis, while improving the ability to identify temperature anomalies and the judgment accuracy.

[0020] Furthermore, the present application provides a method of traversing the temperature distribution information of the fault-free samples to perform single sample temperature distribution hierarchical clustering to obtain updated temperature distribution information of the fault-free samples, including: Obtain the first fault-free sample temperature distribution information of the fault-free sample temperature distribution information; extract the first position temperature and the second position temperature that are neighbors of the first fault-free sample temperature distribution information; when the temperature deviation of the first position temperature and the second position temperature is less than or equal to the temperature deviation threshold, merge the first position and the second position into a third position; at the same time, set the average of the first position temperature and the second position temperature as the third position temperature; perform a cyclic analysis until the temperature difference of any neighborhood of the first fault-free sample temperature distribution information is greater than the temperature deviation threshold, output the first fault-free sample updated temperature distribution information; add the first fault-free sample updated temperature distribution information into the fault-free sample updated temperature distribution information.

[0021] Optionally, the system terminal traverses the collected fault-free sample temperature distribution information, and uses the first traversed fault-free sample temperature distribution information as the first fault-free sample temperature distribution information. The data contains temperature values ​​at different positions of the transformer, such as the temperature of the top, bottom and core of the winding; then, based on the physical position relationship of the measurement points (such as proximity or structural connectivity), identify the temperature measurement points that are neighbors of each other, extract the temperature values ​​of these points, and record them as the first position temperature and the second position temperature, respectively. For example, the measurement points at the top and bottom of a certain layer of the winding can be regarded as neighborhood positions; then, the absolute deviation value of the temperature at the first position and the temperature at the second position is calculated, and the deviation value is compared with the preset temperature deviation threshold. If the absolute deviation value is less than or equal to the temperature deviation threshold, it is considered that the temperature changes at the two positions are similar. This temperature deviation threshold is determined based on the transformer operation experience data, the temperature sensor accuracy, etc.; if the temperature deviation meets the conditions, the first position and the second position are merged into a new position, recorded as the third position, and its temperature value is set to the average of the temperature at the first position and the temperature at the second position. The physical coordinates of the third position are the midpoint of the two positions or other suitable fusion methods, depending on the actual arrangement of the measurement points; according to the above method, all neighborhood positions in the temperature distribution information of the first fault-free sample are iteratively processed. After each merge and update, the new neighborhood temperature deviation is recalculated, and it is determined whether further fusion is required. When the temperature difference of all neighboring positions is greater than the temperature deviation threshold, the loop is stopped; then, the temperature distribution information obtained after the loop processing is used as the first fault-free sample to update the temperature distribution information, and the measurement point position and its temperature value are recorded as new distribution data, and then the processed first fault-free sample update temperature distribution information is integrated into the fault-free sample update temperature distribution information set, so that these fused data can be used for multi-sample aggregation later. Through the above processing, redundant data points are effectively reduced, the representation of temperature distribution is optimized, and the temperature difference between different measurement points is more consistent with the actual operating status. This fusion process enhances the accuracy and stability of the fault-free sample temperature distribution, provides high-quality basic data for the subsequent construction of the benchmark temperature distribution, and improves the sensitivity to abnormal temperature changes.

[0022] Furthermore, the present application provides a method of performing co-location multi-sample aggregation on the updated temperature distribution information of the fault-free samples to obtain the transformer reference temperature distribution information, including: The temperature distribution information is updated from the fault-free samples, and the first position temperatures of all samples are extracted; a central trend analysis is performed on the first position temperatures of all samples to obtain a first position temperature concentration interval; the first position temperature concentration interval is set as a first position reference temperature, and added to the transformer reference temperature distribution information.

[0023] Optionally, the system terminal updates the temperature distribution information of the fault-free samples and filters the first position temperature data of all samples, where the first position usually refers to a fixed physical position of the transformer (such as the top of the winding or the side of the oil tank), to form a temperature data set; then, for the extracted temperature data set, a central trend analysis method is used to calculate its representative interval, that is, the first position temperature in the temperature data set is firstly averaged to obtain the first position temperature mean, which is used to reflect the overall trend of the temperature; then, based on the first position temperature mean, the first position temperature standard deviation is calculated to measure the degree of dispersion of the temperature data, and the standard deviation is calculated by combining the mean. The first position temperature concentration interval is formed by taking the mean minus multiple standard deviations as the lower limit and the mean plus multiple standard deviations as the upper limit, where the first position temperature concentration interval is usually 1 or 2 standard deviations, which can be selected as needed. This concentration interval can reflect the normal range of the first position temperature; then, the first position temperature concentration interval is set as the reference temperature value of the first position to provide a basis for abnormal temperature judgment, and then the calculated first position reference temperature value is added to the transformer reference temperature distribution information, and its corresponding position identification and reference value are recorded. This information will be used as a reference for comparison and analysis in subsequent fault detection to ensure that temperature anomalies can be quickly identified. This process not only improves the representativeness of the transformer reference temperature distribution information, but also provides an accurate judgment basis for temperature anomaly detection, so that the system terminal can identify potential faults more quickly and reliably.

[0024] Fault-free audio feature prediction is performed on the transformer control parameters to obtain transformer reference audio feature information.

[0025] In one embodiment, when the transformer is in a fault-free state, the system terminal inputs the transformer control parameters into a pre-built transformer audio feature analysis model, analyzes these parameters through the first-order audio feature analysis channel and the second-order audio feature analysis channel inside the model, and generates transformer reference audio feature information, including the frequency distribution range of each position, typical time domain feature values, etc. This reference information can not only reflect the normal audio characteristics of the transformer operation, but also provide a comparative reference for subsequent fault audio analysis.

[0026] Furthermore, the present application provides a method for predicting the fault-free audio characteristics of the transformer control parameters to obtain transformer reference audio characteristic information, including: A transformer audio feature analysis model is obtained, wherein the transformer audio feature analysis model comprises a first-order audio feature analysis channel and a second-order audio feature analysis channel, the first-order audio feature analysis channel comprises a plurality of parallel nodes with different topological structures, and an input layer of the second-order audio feature analysis channel and an output layer of the first-order audio feature analysis channel are fully connected; the transformer control parameter is input into the first-order audio feature analysis channel to obtain a plurality of first-order audio feature information; the plurality of first-order audio feature information is input into the second-order audio feature analysis channel to obtain the transformer reference audio feature information.

[0027] Preferably, the system terminal constructs a transformer audio feature analysis model based on the control parameter record data and audio feature record data of the fault-free work log. The model consists of two-level audio feature analysis channels, including a first-order audio feature analysis channel and a second-order audio feature analysis channel; wherein the first-order audio feature analysis channel consists of multiple parallel nodes, each node is responsible for processing the transformer control parameters from different angles and generating primary audio feature information, and the topological structure of the node can adopt a convolutional neural network (CNN), a long short-term memory network (LSTM) or a support vector machine (SVM), etc., to capture different association patterns between the transformer control parameters and the audio features; the second-order audio feature analysis channel can be constructed by a deep neural network (such as a multi-layer perceptron or LSTM), which is responsible for further processing the first-order audio feature information, and its input layer is fully connected to the output layer of the first-order audio feature analysis channel, and the complex relationship between the control parameters and the high-order audio features is captured through deeper modeling; the system terminal takes the control parameters of the transformer as input and passes them to the first-order audio feature Each node of the analysis channel uses its specific topological structure to map the transformer control parameters into intermediate variables related to the audio features. For example, node A calculates the contribution of the control parameters to the main audio frequency components through a linear regression model, node B uses a neural network structure to learn the influence of the control parameters on the signal amplitude characteristics, and node C uses a timing model to analyze the dynamic association of the control parameters with the audio time domain characteristics. Each node generates a set of first-order audio feature information, such as time domain amplitude characteristics, frequency characteristics, or preliminary estimates of time-frequency distribution. Subsequently, the obtained multiple first-order audio feature information is used as input and passed to the second-order audio feature analysis channel for further processing. The input layer of the second-order channel is fully connected to the output layer of the first-order channel, and the features output by each node are integrated for nonlinear mapping to extract hidden high-order audio features. After being processed by the second-order audio feature analysis channel, the transformer benchmark audio feature information is generated, and the output results include the audio feature values ​​at each position, the comprehensive representation between different features, the distribution pattern of multi-dimensional audio features (such as principal component distribution), etc. These reference audio feature information comprehensively reflect the normal audio characteristics of the transformer in a fault-free state, providing high-precision basic data for the analysis and comparison of fault audio features.

[0028] Furthermore, the present application provides the steps of constructing the transformer audio feature analysis model, including: Construct the transformer audio feature analysis model loss function: ;in, Characterize the overall loss, Characterize second-order audio characteristics to analyze channel loss, Characterize the i-th first-order audio feature analysis channel loss, Characterize the total number of first-order audio feature analysis channels; according to the transformer audio feature analysis model loss function, retrieve the control parameter record data and audio feature record data of the fault-free working log of the preset transformer model for training to generate the transformer audio feature analysis model.

[0029] Optionally, in order to train the transformer audio feature analysis model, the system terminal first constructs a loss function of the transformer audio feature analysis model to optimize the model. The transformer audio feature analysis model loss function is as follows: ;in, Represents the overall loss and is used to evaluate the performance of the entire model. Characterizes the second-order audio feature analysis channel loss, reflects the prediction error associated with high-order features, Characterizes the i-th first-order audio feature analysis channel loss, which is used to measure the extraction effect of each node on the primary audio features. The total number of first-order audio feature analysis channels is characterized; the loss function combines the errors of the first-order and second-order channels, and highlights the impact of primary feature extraction accuracy on the overall model performance through an exponential amplification mechanism, thereby achieving global optimization during the training process; subsequently, training data are extracted from the preset transformer fault-free work log, which are divided into control parameter recording data and audio feature recording data. The control parameter recording data includes fault-free operating parameters such as load current, cooling fan status, coil voltage, etc., which serve as model input. The audio feature recording data records the audio features under the fault-free state, such as spectrum distribution, signal time domain characteristics and frequency characteristics. As the target output, the extracted data needs to be normalized to eliminate the dimensional differences between different parameters, reduce the computational complexity and improve the training effect of the model. The normalization is performed in the form of maximum and minimum values. After that, the transformer audio feature analysis model is trained. The model training process includes the training of the first-order audio feature analysis channel and the training of the second-order audio feature analysis channel. In the training of the first-order audio feature analysis channel, the control parameter recording data is input into multiple parallel nodes of the first-order channel. Each node has a different network structure (such as CNN, LSTM or regression network) to extract audio features of a specific dimension, and then the loss of the output of each node is calculated based on the mean square error. , and optimize the parameters of each node through the back propagation algorithm to minimize its loss ; In the training of the second-order audio feature analysis channel, the outputs of all first-order channel nodes are integrated into a comprehensive feature matrix and input into the second-order audio feature analysis channel. The second-order channel performs high-order correlation analysis on the comprehensive features through a fully connected layer and a deep network structure (such as a multi-layer perceptron). Then, the loss of the second-order channel is calculated. , combined with the loss of the first-order and second-order channels, the loss function of the model is analyzed through the transformer audio feature Calculate the overall loss of the model; then, use an optimization algorithm (such as Adam or RMSProp) to adjust the parameters of all channels to minimize the overall loss, and gradually optimize the model parameters through multiple rounds of iterative training until the overall loss converges or the maximum number of training times is reached; during the training process, it is also necessary to evaluate the generalization ability of the model through cross-validation to ensure that the model can generate accurate audio feature information under different control parameter inputs; after the training is completed, the generated transformer audio feature analysis model includes a first-order audio feature analysis channel and a second-order audio feature analysis channel. The trained model can generate audio feature information in a fault-free state according to the control parameter input, providing strong support for transformer fault detection.

[0030] Fault-free electrical parameter prediction is performed on the transformer control parameters to obtain transformer benchmark electrical characteristic information.

[0031] In one embodiment, when the transformer is in a fault-free state, the system terminal inputs the transformer control parameters into a pre-built transformer electrical parameter analysis model (built in the same manner as above), analyzes these parameters through the first-order electrical parameter analysis channel and the second-order electrical parameter analysis channel inside the model, and generates transformer benchmark electrical characteristic information, including winding current and winding voltage distribution at each position, etc. This benchmark information can not only reflect the normal electrical characteristics of the transformer operation, but also provide a comparative reference for subsequent fault electrical analysis.

[0032] Fault prediction is performed based on the transformer reference temperature distribution information, the transformer reference audio feature information, and the transformer reference electrical feature information to obtain a transformer fault prediction type.

[0033] In one embodiment, the system terminal collects the current operation data of the transformer in real time, including temperature distribution data, audio feature data and electrical parameter data. After being collected by the sensor, these data will be transmitted to the system terminal for normalization processing to ensure the accuracy and consistency of the data; then, the collected real-time data is compared with the benchmark feature information in the fault-free state to determine the deviation information of each dimension; then, the analysis results of temperature, audio and electrical features are comprehensively correlated and analyzed, and the type of fault is evaluated by retrieving the fault sample set and added to the transformer fault prediction type. Common fault types include overheating fault, partial discharge fault, mechanical fault, load abnormality, etc.; among them, overheating fault means that the temperature is significantly higher and the gradient distribution is abnormal, partial discharge fault means that the audio high-frequency noise is significantly abnormal, accompanied by electrical parameter fluctuations, mechanical fault means that the low-frequency audio feature is abnormal, which may be accompanied by temperature and electrical parameter changes, and load abnormality means that the electrical parameters deviate from the benchmark range but the temperature and audio features are normal; through this multi-source data fusion fault prediction method, the system terminal can significantly improve the accuracy and timeliness of fault detection, not only can it quickly identify the potential faults of the transformer, but also can accurately classify the fault type, providing comprehensive support for subsequent fault handling and operation optimization.

[0034] Further, the present application provides a method for performing fault prediction based on the transformer reference temperature distribution information, the transformer reference audio feature information and the transformer reference electrical feature information to obtain a transformer fault prediction type, including: Obtain transformer monitoring temperature distribution information, transformer monitoring audio feature information and transformer monitoring electrical feature information; compare the transformer monitoring temperature distribution information with the transformer benchmark temperature distribution information to obtain temperature deviation vector distribution information; compare the transformer monitoring audio feature information with the transformer benchmark audio feature information to obtain an audio deviation vector; compare the transformer monitoring electrical feature information with the transformer benchmark electrical feature information to obtain an electrical deviation vector; retrieve a fault sample set that simultaneously satisfies the temperature deviation vector distribution information, the audio deviation vector and the electrical deviation vector; extract the fault type whose occurrence frequency meets the trigger frequency threshold in the fault sample set, and add it to the transformer fault prediction type.

[0035] Optionally, when monitoring the operating status of the transformer, firstly, the real-time operating data of the transformer is collected respectively through the temperature sensor, audio sensor and electrical sensor to obtain the monitoring temperature distribution information, monitoring audio feature information and monitoring electrical feature information of the transformer. These real-time monitoring information reflect the dynamic characteristics of the transformer under the current operating conditions and provide basic data for fault prediction; then, the real-time collected transformer monitoring temperature distribution information is compared with the pre-established reference temperature distribution information, and the deviation between the two is calculated point by point to form temperature deviation vector distribution information. For example, for the monitoring temperature value at the top of the transformer winding, the deviation between it and the reference temperature value is calculated. For the temperature distribution of the entire winding, the deviation values ​​of each point are combined to construct the overall temperature deviation distribution. This distribution information can not only quantify the degree of temperature anomaly, but also mark the abnormal location; similarly, the monitoring audio feature The information is compared with the reference audio feature information, and the deviations between the audio features (such as frequency, energy, waveform envelope, etc.) are calculated one by one to form audio deviation vectors. These deviation vectors can reflect the abnormal patterns that may exist in the audio signal, such as abnormal changes in high-frequency noise or vibration frequency, indicating the possibility of mechanical vibration or partial discharge failure; in the electrical feature comparison, the electrical deviation vectors are extracted by analyzing the differences between the monitored electrical parameters (such as winding current, reactive power, coil voltage, etc.) and the reference electrical features. For example, when the winding current deviates from the normal range, it may point to overload or insulation loss problems. When the reactive power fluctuation exceeds the reference value, it may reflect the instability of the power grid or internal load abnormalities. These deviation vectors provide multi-dimensional basic data for subsequent comprehensive analysis; then, the temperature deviation vector distribution information, audio deviation vector and electrical deviation vector are input into the fault sample library for retrieval. In the fault sample library, each record stores the temperature, audio and electrical deviation information and the corresponding fault type under known fault conditions. By matching the current deviation vector with the records in the sample library, the fault sample set that meets these deviations is screened out; then, the retrieved fault sample set is statistically analyzed to calculate the frequency of occurrence of each fault type in the sample set. For example, if the matching samples of a certain fault type account for more than 80% of the total samples, it is considered that the fault type has a high correlation. The system terminal screens out the fault type that meets the conditions according to the set trigger frequency threshold (such as 80%) as the fault prediction type under the current operating conditions; finally, the extracted fault prediction type is added to the transformer fault prediction type and output to the monitoring system to prompt the operation and maintenance personnel to take corresponding measures. Through this process, the system terminal realizes the accurate comparison of real-time monitoring data and benchmark feature information, multi-dimensional deviation information extraction, and high-precision fault type prediction based on sample matching, providing reliable support for transformer operation safety.

[0036] When the transformer fault prediction type is not empty, a transformer fault warning signal is generated for alarm.

[0037] In one embodiment, when the transformer fault prediction type is not empty, the system terminal will immediately generate a transformer fault warning signal and trigger an alarm mechanism, which means that a certain type of potential fault has been identified through the previous real-time monitoring data analysis and benchmark feature comparison, and the fault type has a certain degree of credibility. The system terminal will associate the predicted fault type with the corresponding risk level, for example, a minor, serious or critical fault, and determine the intensity and priority of the warning signal according to the risk level; the warning signal will be output and notified in a variety of ways, and the fault warning information will be displayed in an intuitive form on the transformer monitoring platform, including the fault type, location, abnormal parameters and their deviation values, etc. The system will also record detailed information about the fault prediction, including the fault type, corresponding deviation information, benchmark feature comparison results, etc., and store these data in the database. These records not only provide a basis for subsequent fault handling, but can also be used to further optimize fault prediction and improve system reliability.

[0038] In summary, the embodiments of the present application have at least the following technical effects: The embodiment of the present application combines ambient temperature, control parameters, etc. to perform baseline predictions on the temperature, audio, and electrical characteristics of the transformer in a fault-free state, and construct baseline feature information; generates baseline temperature distribution information through hierarchical clustering and multi-sample aggregation methods; performs high-order modeling of audio features of control parameters through an audio feature analysis model containing a multi-layer structure; extracts temperature, audio, and electrical deviation vectors by comparing monitoring data with baseline feature information, retrieves a fault sample library based on the deviation information, and identifies the fault type in combination with the trigger frequency threshold; finally, when a non-empty fault type is predicted, a warning signal is generated and an alarm is triggered, achieving efficient prediction and warning of potential transformer faults, and improving the safety and reliability of transformer operation. These technical effects jointly solve the technical problem that the single source of feature information in traditional transformer fault analysis leads to insufficient accuracy of analysis results, and achieve the technical effect of improving the accuracy of transformer fault prediction and the timeliness of warning through multi-source data fusion and comprehensive prediction methods.

[0039] Embodiment 2, based on the same inventive concept as the transformer fault early warning method under multi-source data in the above embodiment, Figure 2As shown, the present application provides a transformer fault early warning system under multi-source data, the system comprising: Fault-free temperature distribution prediction module 1: performs fault-free temperature distribution prediction on ambient temperature and transformer control parameters to obtain transformer benchmark temperature distribution information; fault-free audio feature prediction module 2: performs fault-free audio feature prediction on the transformer control parameters to obtain transformer benchmark audio feature information; fault-free electrical parameter prediction module 3: performs fault-free electrical parameter prediction on the transformer control parameters to obtain transformer benchmark electrical feature information; fault prediction module 4: performs fault prediction based on the transformer benchmark temperature distribution information, the transformer benchmark audio feature information and the transformer benchmark electrical feature information to obtain the transformer fault prediction type; fault alarm module 5: generates a transformer fault early warning signal for alarm when the transformer fault prediction type is not empty.

[0040] Furthermore, the fault-free temperature distribution prediction module 1 is also used to perform the following method: Taking the ambient temperature and the transformer control parameters as constraints, the fault-free sample temperature distribution information of the transformer model is collected; the fault-free sample temperature distribution information is traversed to perform single sample temperature distribution hierarchical clustering to obtain the fault-free sample update temperature distribution information; the fault-free sample update temperature distribution information is aggregated with multiple samples at the same position to obtain the transformer reference temperature distribution information.

[0041] Furthermore, the fault-free temperature distribution prediction module 1 is also used to perform the following method: Obtain the first fault-free sample temperature distribution information of the fault-free sample temperature distribution information; extract the first position temperature and the second position temperature that are neighbors of the first fault-free sample temperature distribution information; when the temperature deviation of the first position temperature and the second position temperature is less than or equal to the temperature deviation threshold, merge the first position and the second position into a third position; at the same time, set the average of the first position temperature and the second position temperature as the third position temperature; perform a cyclic analysis until the temperature difference of any neighborhood of the first fault-free sample temperature distribution information is greater than the temperature deviation threshold, output the first fault-free sample updated temperature distribution information; add the first fault-free sample updated temperature distribution information into the fault-free sample updated temperature distribution information.

[0042] Furthermore, the fault-free temperature distribution prediction module 1 is also used to perform the following method: The temperature distribution information is updated from the fault-free samples, and the first position temperatures of all samples are extracted; a central trend analysis is performed on the first position temperatures of all samples to obtain a first position temperature concentration interval; the first position temperature concentration interval is set as a first position reference temperature, and added to the transformer reference temperature distribution information.

[0043] Furthermore, the fault-free audio feature prediction module 2 is also used to perform the following method: A transformer audio feature analysis model is obtained, wherein the transformer audio feature analysis model comprises a first-order audio feature analysis channel and a second-order audio feature analysis channel, the first-order audio feature analysis channel comprises a plurality of parallel nodes with different topological structures, and an input layer of the second-order audio feature analysis channel and an output layer of the first-order audio feature analysis channel are fully connected; the transformer control parameter is input into the first-order audio feature analysis channel to obtain a plurality of first-order audio feature information; the plurality of first-order audio feature information is input into the second-order audio feature analysis channel to obtain the transformer reference audio feature information.

[0044] Furthermore, the fault-free audio feature prediction module 2 is also used to perform the following method: Construct the transformer audio feature analysis model loss function: in, Characterize the overall loss, Characterize second-order audio characteristics to analyze channel loss, Characterize the i-th first-order audio feature analysis channel loss, Characterize the total number of first-order audio feature analysis channels; according to the transformer audio feature analysis model loss function, retrieve the control parameter record data and audio feature record data of the fault-free working log of the preset transformer model for training to generate the transformer audio feature analysis model.

[0045] Furthermore, the fault prediction module 4 is also used to execute the following method: Obtain transformer monitoring temperature distribution information, transformer monitoring audio feature information and transformer monitoring electrical feature information; compare the transformer monitoring temperature distribution information with the transformer benchmark temperature distribution information to obtain temperature deviation vector distribution information; compare the transformer monitoring audio feature information with the transformer benchmark audio feature information to obtain an audio deviation vector; compare the transformer monitoring electrical feature information with the transformer benchmark electrical feature information to obtain an electrical deviation vector; retrieve a fault sample set that simultaneously satisfies the temperature deviation vector distribution information, the audio deviation vector and the electrical deviation vector; extract the fault type whose occurrence frequency meets the trigger frequency threshold in the fault sample set, and add it to the transformer fault prediction type.

[0046] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0047] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0048] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A transformer fault early warning method based on multi-source data, characterized in that: include: Predict the fault-free temperature distribution of the ambient temperature and transformer control parameters to obtain the transformer benchmark temperature distribution information; Predicting fault-free audio characteristics of the transformer control parameters to obtain transformer reference audio characteristic information; Performing fault-free electrical parameter prediction on the transformer control parameters to obtain transformer benchmark electrical characteristic information; Perform fault prediction based on the transformer reference temperature distribution information, the transformer reference audio feature information, and the transformer reference electrical feature information to obtain a transformer fault prediction type; When the transformer fault prediction type is not empty, a transformer fault warning signal is generated for alarm.

2. The transformer fault early warning method under multi-source data according to claim 1, characterized in that: The fault-free temperature distribution prediction of the ambient temperature and transformer control parameters is carried out to obtain the transformer reference temperature distribution information, including: Taking the ambient temperature and the transformer control parameters as constraints, collecting temperature distribution information of fault-free samples of the transformer model; Traversing the temperature distribution information of the fault-free samples to perform single sample temperature distribution hierarchical clustering to obtain updated temperature distribution information of the fault-free samples; The temperature distribution information of the fault-free samples is updated by performing multi-sample aggregation at the same position to obtain the transformer reference temperature distribution information.

3. The transformer fault early warning method under multi-source data according to claim 2, characterized in that: Traversing the temperature distribution information of the fault-free samples to perform single sample temperature distribution hierarchical clustering to obtain updated temperature distribution information of the fault-free samples, including: Obtain first fault-free sample temperature distribution information of the fault-free sample temperature distribution information; Extracting the temperature of a first position and a second position that are neighbors of each other from the temperature distribution information of the first fault-free sample; When a temperature deviation between the temperature of the first position and the temperature of the second position is less than or equal to a temperature deviation threshold, merging the first position and the second position into a third position; At the same time, the average of the first position temperature and the second position temperature is set as the third position temperature; Circular analysis is performed until the temperature difference of any neighborhood of the first fault-free sample temperature distribution information is greater than the temperature deviation threshold, and the first fault-free sample updated temperature distribution information is output; The first non-fault sample updated temperature distribution information is added to the non-fault sample updated temperature distribution information.

4. The transformer fault early warning method under multi-source data according to claim 2, characterized in that: Performing co-location multi-sample aggregation on the updated temperature distribution information of the fault-free samples to obtain the transformer reference temperature distribution information includes: Update the temperature distribution information from the fault-free samples and extract the first position temperature of all samples; Performing a central tendency analysis on the first position temperatures of all samples to obtain a first position temperature concentration interval; The first position temperature concentration interval is set as the first position reference temperature, and is added into the transformer reference temperature distribution information.

5. The transformer fault early warning method under multi-source data according to claim 1, characterized in that: Performing fault-free audio feature prediction on the transformer control parameter to obtain transformer reference audio feature information includes: Obtaining a transformer audio feature analysis model, wherein the transformer audio feature analysis model includes a first-order audio feature analysis channel and a second-order audio feature analysis channel, the first-order audio feature analysis channel includes a plurality of parallel nodes with different topological structures, and an input layer of the second-order audio feature analysis channel and an output layer of the first-order audio feature analysis channel are fully connected; Inputting the transformer control parameter into the first-order audio feature analysis channel to obtain a plurality of first-order audio feature information; The plurality of first-order audio feature information are input into the second-order audio feature analysis channel to obtain the transformer reference audio feature information.

6. The transformer fault early warning method under multi-source data according to claim 5, characterized in that: The transformer audio feature analysis model construction step includes: Construct the transformer audio feature analysis model loss function: in, Characterize the overall loss, Characterize second-order audio characteristics to analyze channel loss, Characterize the i-th first-order audio feature analysis channel loss, Characterize the total number of first-order audio feature analysis channels; According to the transformer audio feature analysis model loss function, the control parameter record data and audio feature record data of the fault-free working log of a preset transformer model are retrieved for training to generate the transformer audio feature analysis model.

7. The transformer fault early warning method under multi-source data according to claim 1, characterized in that: Performing fault prediction based on the transformer reference temperature distribution information, the transformer reference audio feature information, and the transformer reference electrical feature information to obtain a transformer fault prediction type includes: Obtain transformer monitoring temperature distribution information, transformer monitoring audio characteristic information and transformer monitoring electrical characteristic information; Comparing the transformer monitoring temperature distribution information with the transformer reference temperature distribution information to obtain temperature deviation vector distribution information; Comparing the transformer monitoring audio feature information with the transformer reference audio feature information to obtain an audio deviation vector; Comparing the transformer monitoring electrical characteristic information with the transformer reference electrical characteristic information to obtain an electrical deviation vector; Retrieving a fault sample set that simultaneously satisfies the temperature deviation vector distribution information, the audio deviation vector, and the electrical deviation vector; Fault types whose occurrence frequencies meet a trigger frequency threshold in the fault sample set are extracted and added into the transformer fault prediction type.

8. The transformer fault early warning system under multi-source data is characterized by: The steps for implementing the transformer fault early warning method under multi-source data according to any one of claims 1 to 7 include: Fault-free temperature distribution prediction module: predicts the fault-free temperature distribution of the ambient temperature and transformer control parameters to obtain the transformer benchmark temperature distribution information; A fault-free audio feature prediction module is used to predict the fault-free audio feature of the transformer control parameters to obtain transformer reference audio feature information; A fault-free electrical parameter prediction module is used to predict the fault-free electrical parameters of the transformer control parameters to obtain the transformer reference electrical characteristic information; Fault prediction module: performs fault prediction based on the transformer reference temperature distribution information, the transformer reference audio characteristic information and the transformer reference electrical characteristic information to obtain a transformer fault prediction type; Fault alarm module: when the transformer fault prediction type is not empty, a transformer fault warning signal is generated for alarm.

Images