Transformer partial discharge early warning system and method combined with time-frequency characteristics

Abstract

The invention provides a transformer partial discharge early warning system and method in combination with time-frequency characteristics, and relates to the technical field of partial discharge monitoring and early warning, and the method comprises the steps: inputting a partial discharge signal into a multipath signal separation model for superposed signal separation, and outputting a plurality of separated partial discharge signals; performing time-frequency feature extraction, and outputting multiple groups of time-frequency feature vectors; performing pattern recognition on the multiple groups of time-frequency feature vectors by using a partial discharge feature library, and outputting multiple discharge early warning signals; and carrying out partial discharge early warning reminding on an upper computer of the transformer according to the plurality of discharge early warning signals. According to the invention, the technical problem that the accuracy and precision of partial discharge early warning are insufficient due to the fact that complex aliasing signals of different propagation paths cannot be effectively separated in a traditional early warning method can be solved; signal interference of different paths can be effectively eliminated, the characteristic of each signal source is ensured to be clearer, and thus the precision and accuracy of transformer partial discharge early warning are remarkably improved.

Description

Technical Field

[0001] The present application relates to the technical field of partial discharge monitoring and early warning, and in particular to a transformer partial discharge early warning system and method combining time-frequency characteristics. Background Art

[0002] Traditional transformer partial discharge early warning methods usually rely on the collection of discharge signals by partial discharge sensors (such as ultrasonic sensors, current sensors, voltage sensors, etc.). These methods usually determine whether partial discharge occurs by monitoring the time domain or frequency domain signals collected by the sensors. However, when partial discharge signals propagate inside the transformer, they will experience a complex propagation path and may be affected by multiple factors such as equipment structure and environmental noise, resulting in signal aliasing and distortion.

[0003] To identify partial discharge signals, traditional methods usually rely on time domain analysis, frequency domain analysis or simple waveform analysis. However, these methods are difficult to effectively deal with complex mixed signals, especially in the presence of multiple signal sources (i.e., multipath propagation). Even if some filtering or denoising techniques are used, the interference components in the signal will still affect the accuracy and precision of the detection.

[0004] In summary, the traditional transformer partial discharge warning method has the technical problem of insufficient accuracy and precision of partial discharge warning due to the inability to effectively separate the complex aliasing signals of different propagation paths. Summary of the invention

[0005] The purpose of this application is to provide a transformer partial discharge warning system and method combining time-frequency characteristics, so as to solve the technical problem that the accuracy and precision of the partial discharge warning are insufficient due to the inability to effectively separate the complex aliasing signals of different propagation paths in the traditional transformer partial discharge warning method.

[0006] In view of the above problems, the present application provides a transformer partial discharge early warning system and method combining time-frequency characteristics.

[0007] In a first aspect, the present application provides a transformer partial discharge warning system combined with time-frequency features, including: a partial discharge signal acquisition module, used to obtain a partial discharge signal of a transformer according to a partial discharge sensing device; a superposition signal separation module, used to input the partial discharge signal into a multi-path signal separation model, perform superposition signal separation on the partial discharge signal according to the multi-path signal separation model, and output multiple separated partial discharge signals; a time-frequency feature extraction module, used to extract time-frequency features of the multiple separated partial discharge signals, and output multiple groups of time-frequency feature vectors, wherein each group of time-frequency feature vectors includes a time domain feature vector, a frequency domain feature vector and a time-frequency difference feature vector; a partial discharge feature library training module, used to train a partial discharge feature library, the partial discharge feature library includes time-frequency feature vector samples corresponding to partial discharge type samples; a time-frequency feature vector recognition module, used to use the partial discharge feature library to perform pattern recognition on the multiple groups of time-frequency feature vectors respectively, and output multiple discharge warning signals; a partial discharge warning module, used to issue a partial discharge warning reminder to the host computer of the transformer according to the multiple discharge warning signals.

[0008] In the second aspect, the present application also provides a transformer partial discharge warning method combined with time-frequency features, including: obtaining a partial discharge signal of the transformer according to a partial discharge sensing device; inputting the partial discharge signal into a multi-path signal separation model, performing superposition signal separation on the partial discharge signal according to the multi-path signal separation model, and outputting multiple separated partial discharge signals; extracting time-frequency features of the multiple separated partial discharge signals, and outputting multiple groups of time-frequency feature vectors, wherein each group of time-frequency feature vectors includes a time domain feature vector, a frequency domain feature vector and a time-frequency difference feature vector; training a partial discharge feature library, the partial discharge feature library includes time-frequency feature vector samples corresponding to partial discharge type samples; using the partial discharge feature library to perform pattern recognition on the multiple groups of time-frequency feature vectors respectively, and outputting multiple discharge warning signals; and providing a partial discharge warning reminder to the host computer of the transformer according to the multiple discharge warning signals.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages: By inputting the local discharge signal into the multi-path signal separation model, the local discharge signal is superimposed and separated according to the multi-path signal separation model, and multiple separated local discharge signals are output; then the time-frequency feature extraction is performed on the multiple separated local discharge signals, and multiple groups of time-frequency feature vectors are output, wherein each group of time-frequency feature vectors includes a time domain feature vector, a frequency domain feature vector and a time-frequency difference feature vector; on the other hand, a local discharge feature library is trained, and the local discharge feature library includes time-frequency feature vector samples corresponding to local discharge type samples; then the local discharge feature library is used to perform pattern recognition on the multiple groups of time-frequency feature vectors respectively, and multiple discharge warning signals are output; finally, a local discharge warning reminder is given to the host computer of the transformer according to the multiple discharge warning signals. In other words, by combining technologies such as multi-path signal separation, time-frequency feature extraction and pattern recognition, signal interference of different paths can be effectively eliminated, ensuring that the characteristics of each signal source are clearer, thereby significantly improving the precision and accuracy of transformer local discharge warning, and ensuring the safe and stable operation of the transformer.

[0010] 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 according to the contents of the specification, and 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 specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0012] Figure 1 This is a schematic diagram of the structure of a transformer partial discharge warning system combining time-frequency characteristics in this application; Figure 2 A flowchart of a transformer partial discharge early warning method combining time-frequency characteristics is provided for this application; Description of reference numerals: A partial discharge signal acquisition module 11, a superposition signal separation module 12, a time-frequency feature extraction module 13, a partial discharge feature library training module 14, a time-frequency feature vector recognition module 15, and a partial discharge early warning module 16. DETAILED DESCRIPTION

[0013] This application provides a transformer partial discharge warning system and method combined with time-frequency features, which solves the technical problem that the accuracy and precision of the partial discharge warning are insufficient in the traditional transformer partial discharge warning method due to the inability to effectively separate the complex aliasing signals of different propagation paths. By combining technologies such as multi-path signal separation, time-frequency feature extraction and pattern recognition, signal interference from different paths can be effectively eliminated, ensuring that the characteristics of each signal source are clearer, thereby significantly improving the precision and accuracy of the transformer partial discharge warning and ensuring the safe and stable operation of the transformer.

[0014] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0015] For example, please refer to the attached Figure 1 The present application provides a transformer partial discharge early warning system combining time-frequency characteristics, including: The partial discharge signal acquisition module 11 is used to acquire the partial discharge signal of the transformer according to the partial discharge sensor device.

[0016] Furthermore, the partial discharge signal acquisition module 11 is also used for: A partial discharge signal of the transformer is obtained according to a partial discharge sensor device, wherein the partial discharge signal is a superimposed partial discharge signal corresponding to a plurality of discharge propagation paths at the same defect position; wherein a starting point of each discharge propagation path in the plurality of discharge propagation paths is the defect position, and an end point of the path is the partial discharge sensor device.

[0017] Specifically, the source of the partial discharge signal is located at a defective position inside the transformer, which may be a partial discharge phenomenon caused by damage, aging or other structural defects of the insulating material. The discharge at the defective position is the source of the partial discharge.

[0018] Then, the partial discharge signal of the transformer is obtained through a partial discharge sensing device (such as a current sensor, a voltage sensor, etc.), wherein the transformer partial discharge signal will be transmitted along multiple paths during the propagation process, and the signals of the multiple paths will be superimposed when reaching the sensor detection point to form a complex mixed signal; the partial discharge signal is a superimposed partial discharge signal corresponding to multiple discharge propagation paths at the same defect location, and the multiple discharge propagation paths include cable paths, insulating material paths (propagated through media such as insulating oil and solid insulating materials), metal structure paths (propagated through the metal frame or other structures of the transformer), etc. The propagation characteristics of each path may be different, such as the signal propagates faster in the metal path, but may propagate slower in the insulating material, and the signal may be attenuated or distorted.

[0019] The starting point of each of the multiple discharge propagation paths is the defect location, and the end point of the path is the partial discharge sensor device. The partial discharge signals propagated through different paths will eventually converge at the partial discharge sensor outside the transformer. Since the signals of each path have different propagation time, propagation attenuation and distortion effects, when these signals reach the sensor, they will form a mixed signal in a superimposed manner. This mixed signal contains the partial discharge information of all paths.

[0020] The superposition signal separation module 12 is used to input the local discharge signal into a multi-path signal separation model, perform superposition signal separation on the local discharge signal according to the multi-path signal separation model, and output a plurality of separated local discharge signals.

[0021] Specifically, in the partial discharge monitoring system of the transformer, since the partial discharge signal comes from the same defect position but propagates through multiple different paths, the sensor receives the superposition result of these signals. In order to accurately extract the independent discharge characteristics of each propagation path from the mixed signal, it is necessary to construct a multi-path signal separation model. The goal of the multi-path signal separation model is to decompose these aliased signals into signal components on each independent path through mathematical or signal processing methods, thereby recovering the original partial discharge signal of each path. Then, the partial discharge signal is input into the multi-path signal separation model, and the multi-path signal separation model performs superposition signal separation on the partial discharge signal, and outputs multiple separated partial discharge signals, wherein each signal corresponds to a specific propagation path, thereby recovering the independent partial discharge characteristics on each path. By accurately separating the signal of each propagation path, the characteristics of partial discharge can be more accurately identified, and the accuracy and precision of discharge detection can be improved.

[0022] Furthermore, the superposition signal separation module 12 is also used for: Acquire the internal structure data and insulation distribution data of the transformer; model the transformer according to the internal structure data and insulation distribution data, and output a transformer simulation model; simulate the discharge propagation paths under different defect positions and intensities, and the time-frequency characteristic changes under each discharge propagation path according to the transformer simulation model, and output a simulation database.

[0023] Specifically, first, the internal structure data and insulation distribution data of the transformer are obtained, wherein the internal structure data includes the position, size, material properties, etc. of each component. The layout and material properties of the metal structure, coil, insulation layer, cable and other components inside the transformer will affect the propagation path and signal characteristics of partial discharge. Through detailed structural data, the electrical and mechanical properties inside the transformer can be modeled; the insulation distribution data includes the distribution, type, thickness and aging degree of the insulating material, etc. The insulation distribution data is crucial for predicting the occurrence and propagation path of partial discharge. The insulating material may be oil, paper, gas, solid or other composite materials, and their dielectric properties directly affect the nature and propagation behavior of partial discharge.

[0024] Next, the transformer is modeled according to the internal structure data and the insulation distribution data. For example, the internal structure and insulation distribution of the transformer are digitally modeled using computer-aided design (CAD) tools or finite element analysis (FEA) technology. The model should include various electrical components, insulation layers, grounding systems, etc. of the transformer, and set its electrical characteristics and physical properties (such as conductivity, dielectric constant, etc.). A transformer simulation model is obtained based on the modeling results. The transformer simulation model can be used to simulate partial discharge behavior under different operating conditions.

[0025] Further, the transformer simulation model is used to simulate the discharge propagation paths under different defect positions and intensities, as well as the time-frequency characteristic changes under each discharge propagation path, such as simulating the influence of defects at different positions in the transformer (such as cracks in the insulation layer, air gaps, aging, etc.) on the discharge path. The simulation will generate different local discharge propagation paths according to the different defect positions (such as high-voltage windings, low-voltage windings, grounding parts, etc.); simulate the influence of local discharges of different intensities on signal propagation, including the difference between weak discharges and larger discharge intensities. Depending on the defect intensity, the discharge signal may have different degrees of attenuation, distortion or time delay during the propagation process. Based on the simulation model and the simulation results of the discharge propagation path under different defect conditions, a simulation database containing time-frequency characteristic changes is generated. The simulation database includes the discharge propagation path under each simulation condition, and provides detailed data for the discharge path under different defect positions and discharge intensities; at the same time, the time-frequency characteristics of the local discharge signal (such as time domain waveform, spectrum characteristics, time-frequency domain changes, etc.) will be different on different paths, and the simulation database should contain the time-frequency analysis results of each case. Through these steps, a detailed transformer simulation model can be used to simulate various partial discharge scenarios and generate diverse time-frequency feature data, providing reliable data support and analysis basis for subsequent partial discharge feature extraction.

[0026] The simulation database is connected to train a multipath signal separation model according to an independent component analysis algorithm.

[0027] Furthermore, the present application also includes the following steps: An initialization separation model is constructed, wherein the initialization separation model includes an initial random separation matrix, wherein the initial random separation matrix is ​​a random orthogonal matrix.

[0028] Furthermore, the present application also includes the following steps: Construct an initial separation model, the expression includes: ;in, is the multipath superposition signal matrix extracted from the simulation database, represents the signal that has not been propagated and superimposed through the path, represents the attenuation coefficient of signal propagation on the i-th path, represents the delay of the signal on the i-th path, N is the total number of paths, and is a random interference signal.

[0029] Specifically, first, an initialization separation model is constructed. The main purpose of the initialization separation model is to provide a preliminary separation matrix for the subsequent signal separation process. The initial setting of the model must be as close as possible to the independent components of the final target signal. For this reason, a random orthogonal matrix is ​​used as the initial separation matrix. Its main function is to generate an initial state containing multiple independent path signals. In signal processing and blind source separation, an orthogonal matrix is ​​usually used to represent different components of a signal. An orthogonal matrix is ​​a square matrix with orthogonal column vectors or row vectors, that is, the transpose of the matrix is ​​equal to its inverse, which means that the columns (or rows) of the matrix are independent of each other and will not interfere with each other. Initializing the separation matrix is ​​usually achieved by randomly generating orthogonal matrices. The elements of these matrices are randomly initialized and normalized by orthogonalization methods (such as Schmidt orthogonalization). In this way, the initial state of the model can maintain appropriate randomness but still have orthogonality, which provides a suitable starting point for subsequent signal separation. Among them, the random initialization of the matrix helps to avoid the problem of local optimal solutions. Randomness can prevent the signal separation process from falling into the wrong separation mode too early; orthogonality ensures that during the signal separation process, the signals of each path are mathematically independent, thereby avoiding mutual interference between signals.

[0030] In the initialization separation model, The multi-path superposition signal matrix extracted from the simulation database represents the signals gathered from various discharge propagation paths. These signals are mixed signals after multi-path propagation and include the superposition of local discharge signals of multiple paths; It refers to the signal that has not been propagated and superimposed on the path, that is, the original form of the partial discharge signal propagated on different paths, without being affected by attenuation, delay or other propagation characteristics; It represents the attenuation coefficient of signal propagation on the i-th path, indicating the degree of signal attenuation; represents the delay of the signal on the i-th path, which means the delay of the signal from the source to the receiving end; N is the total number of paths, is a random interference signal, which is the noise or external interference that may be introduced during the signal propagation process. Its existence may affect the quality of the signal and cause errors in the separation process. Through this expression, a multipath signal model can be established, and subsequent signal separation and processing can be performed on this basis.

[0031] Connecting to the simulation database to iteratively update the initial random separation matrix to obtain a converged separation matrix that satisfies a preset independence; Furthermore, the present application also includes the following steps: The simulation database is connected to iteratively update the initial random separation matrix, and the expression includes: ;in, is the initial random separation matrix, is the converged separation matrix after iterative update, is the multipath superposition signal matrix extracted from the simulation database, is a nonlinear comparison function used to enhance the independence of signals, T is the transpose of the matrix, is the derivative of the contrast function, used to calculate the gradient value of the signal to update the initial random separation matrix, and E is the expected operation for calculating the statistical characteristics of the multipath superposition signal matrix.

[0032] Specifically, in the signal separation process, the simulation database is used to provide training data for the model, and the initial random separation matrix is ​​optimized through an iterative update process. Through this iterative update, the goal is to enable the separation matrix to extract the independent components of the signal, thereby accurately separating the local discharge signals of each propagation path. Then, the initial random separation matrix is ​​iteratively updated through the simulation database, wherein the core of the iterative process is to optimize the signal through the derivative (gradient) of the contrast function, so that the signal separation matrix gradually converges to a separation matrix that meets the preset independence; by continuously adjusting the elements of the separation matrix, the dependence between the signal sources is gradually reduced until the signal reaches an ideal independent state.

[0033] In the iterative update process, first, the initial random separation matrix and the convergence matrix, where the initial random separation matrix is ​​usually a randomly generated orthogonal matrix; after iterative update, the convergence separation matrix, the ultimate goal is to make this matrix able to effectively separate the signal. Then, the multipath superposition signal matrix extracted from the simulation database represents the mixed signal of multiple signal sources. This matrix contains the superposition of signals on multiple paths, and there is cross interference between the signals. A nonlinear comparison function is used to measure the independence of the signal. This function is continuously optimized during the iteration process to enhance the independence of the signal source. Then, in order to realize the matrix update, it is necessary to calculate the derivative of the comparison function. This derivative represents the rate of change of the separation matrix to the comparison function in the current iteration state. It is used to indicate how to adjust the matrix to enhance the independence of the signal. The expectation of the statistical characteristics of the signal matrix is ​​further calculated. This operation is usually used to estimate the mean, variance or high-order statistics of the signal. The expectation operation helps to obtain the statistical characteristics of the data, which is used to guide the update direction of the matrix; repeat the above steps for iterative update until a converged separation matrix that meets the preset independence (such as convergence through the comparison function value, or the independence of the signal source reaches a predetermined threshold) is obtained. This matrix can effectively separate the local discharge signals on each path from the mixed signal, so that the independence of the signal is fully enhanced. By continuously iteratively updating the initial random separation matrix and using the gradient information and expectation operation in the above expression, the goal is to make the separation matrix converge to a state that maximizes the independence of the signal, thereby accurately separating the independent signal of each path.

[0034] Through the above iterative update process, the multi-path superposition signal matrix in the simulation database is used to continuously optimize the initial random separation matrix through nonlinear comparison function and gradient descent method, so that the final converged separation matrix can accurately separate the partial discharge signals of each path. This process ensures the independence of the signal source, thereby improving the accuracy and reliability of transformer partial discharge warning.

[0035] The converged separation matrix is ​​used as the core parameter of the initialization separation model, and a training multipath signal separation model is output.

[0036] Specifically, after multiple iterations and updates, a converged separation matrix that meets the preset independence requirements is obtained. This converged matrix has been optimized to the best state and can effectively separate the signals of each independent path from the multi-path superposition signal. Then the converged separation matrix is ​​used as the core parameter of the initialization separation model, and the training multi-path signal separation model is output. The model separates the new multi-path superposition signal through the converged separation matrix, and outputs the estimation result of each independent signal. The trained model can separate the new mixed signal (for example, the signal obtained during the transformer monitoring process) into the signals of each independent path. By constructing a multi-path signal separation model, the model effectively extracts the independent components of each signal path from power equipment such as transformers through the core parameters optimized by the converged separation matrix, thereby providing high-precision data support for partial discharge detection and analysis, and improving the accuracy and reliability of partial discharge detection.

[0037] The time-frequency feature extraction module 13 is used to extract time-frequency features from the multiple separated partial discharge signals and output multiple groups of time-frequency feature vectors, wherein each group of time-frequency feature vectors includes a time domain feature vector, a frequency domain feature vector and a time-frequency difference feature vector.

[0038] Specifically, after signal separation, each signal obtained represents a local discharge signal on a different path. Next, time-frequency feature extraction is performed on these separated signals to extract time-frequency features that are helpful for analysis and identification. Through these features, different types of local discharges and abnormal behaviors can be effectively distinguished. First, time-frequency feature extraction is performed on the multiple separated local discharge signals. The goal of time-frequency feature extraction is to extract information representing the time and frequency characteristics of the separated local discharge signals. These features will be used for subsequent pattern recognition and fault diagnosis. The time-frequency features of each signal can reflect different aspects of the discharge activity, such as the time of occurrence of the discharge, frequency changes, and time-frequency differences. Multiple groups of time-frequency feature vectors are obtained, where each group of time-frequency feature vectors includes time domain feature vectors, frequency domain feature vectors, and time-frequency difference feature vectors. The time domain feature vectors mainly focus on the changes in the signal over time, including the signal's Amplitude, waveform, fluctuation frequency, etc. Typical time domain features include the maximum value, mean, standard deviation, root mean square, zero crossing rate, kurtosis and skewness of the signal; frequency domain feature vectors focus on the spectral characteristics of the signal, reflecting the distribution of the signal in frequency. Fourier transform can be used to convert the signal from the time domain to the frequency domain for analysis. Typical frequency domain features include power spectrum density, main frequency, bandwidth, frequency center, etc.; time-frequency difference feature vectors combine time domain and frequency domain information and are often used to analyze non-stationary signals, especially the instantaneous changes of partial discharge signals, including instantaneous frequency, instantaneous energy, time-frequency diagram, wavelet transform, etc. Multiple groups of time-frequency feature vectors are generated by extracting time-frequency features of multiple separated partial discharge signals. Each group of vectors includes time domain, frequency domain and time-frequency difference features. These feature vectors provide data support for subsequent partial discharge type identification and early warning analysis.

[0039] The partial discharge feature library training module 14 is used to train the partial discharge feature library, wherein the partial discharge feature library includes time-frequency feature vector samples corresponding to partial discharge type samples.

[0040] Specifically, a partial discharge feature library is trained, which contains time-frequency feature vector samples of different types of partial discharges. Through these samples, the relationship between the partial discharge signal and its type can be established, thereby providing a basis for subsequent fault diagnosis and early warning, wherein the partial discharge feature library includes time-frequency feature vector samples corresponding to the partial discharge type samples, and the partial discharge type represents a certain type of partial discharge (for example, gas discharge, insulation discharge, etc.), which can be pre-defined by manual calibration or simulation model; the time-frequency feature vector includes time domain features, frequency domain features and time-frequency difference features, which are extracted by the time-frequency analysis part in the previous step, wherein the partial discharge type and the time-frequency feature vector have a corresponding relationship. The steps of training the partial discharge feature library include, first, by collecting time-frequency features from different partial discharge scenarios, a sample library is established, and these samples can come from experimental data, simulation data or historical fault records; then, the corresponding partial discharge type is marked for each time-frequency feature sample. For example, different discharge types can be labeled as gas discharge, local contact discharge, tree branch discharge, etc. through simulation data or empirical rules; then, in order to improve the stability and accuracy of the model, all extracted time-frequency feature vectors can be standardized or normalized to eliminate the influence of different feature dimensions; finally, each local discharge type and its corresponding time-frequency feature vector are added to the feature library to form the final training data set.

[0041] The time-frequency feature vector recognition module 15 is used to use the partial discharge feature library to perform pattern recognition on the multiple groups of time-frequency feature vectors respectively, and output multiple discharge warning signals.

[0042] Furthermore, the time-frequency feature vector identification module 15 is also used for: Construct a pattern recognizer, the pattern recognizer is obtained by identifying the correspondence between the local discharge feature library and the local discharge type sample; the multiple groups of time-frequency feature vectors are input into the pattern recognizer for feature similarity matching to obtain multiple matching discharge patterns; severity levels are identified according to the multiple matching discharge patterns, and multiple identification results are output; and multiple discharge warning signals are output according to the multiple identification results.

[0043] Specifically, first, a pattern recognizer is constructed by identifying the correspondence between the local discharge feature library and the local discharge type samples. The pattern recognizer is a classification model obtained through training, which can identify the type and severity of the discharge according to the time-frequency feature vector of the local discharge, wherein the local discharge feature library contains time-frequency feature samples of different discharge types, and these samples are obtained through previous time-frequency feature extraction and annotation; then these time-frequency feature samples are used to train the pattern recognizer to learn how to judge different types of local discharges and severity according to the time-frequency feature vectors, such as by calculating the similarity between the input features and the samples in the feature library (such as using Euclidean distance, cosine similarity, etc.), the discharge type that best matches the input signal is determined; based on the matching results, the pattern recognizer will output the corresponding discharge type and the corresponding severity level.

[0044] Furthermore, the multiple groups of time-frequency feature vectors are input into the pattern recognizer for feature similarity matching, and multiple matching discharge patterns are identified by calculating the similarity between the input time-frequency feature vectors and the samples in the local discharge feature library, and multiple matching discharge patterns are output. Then, severity levels are identified based on the multiple matching discharge patterns, that is, for each matching discharge pattern, the severity of the discharge is judged according to its type and characteristics (such as frequency, intensity, etc.), such as normal, warning, dangerous, etc., and the severity level is identified according to the characteristics of the matching pattern. Finally, multiple discharge warning signals are generated based on the matched discharge patterns and the identified severity levels.

[0045] The partial discharge warning module 16 is used to issue a partial discharge warning reminder to the host computer of the transformer according to the multiple discharge warning signals.

[0046] Specifically, after time-frequency feature extraction and pattern recognition, multiple discharge warning signals are finally generated. These signals will be used to send local discharge warning reminders to the transformer's host computer system, helping operators to detect potential electrical faults in a timely manner and take corresponding measures to prevent accidents.

[0047] In summary, the transformer partial discharge early warning system combined with time-frequency characteristics provided by the present application has the following technical effects: By inputting the local discharge signal into the multi-path signal separation model, the local discharge signal is superimposed and separated according to the multi-path signal separation model, and multiple separated local discharge signals are output; then the time-frequency feature extraction is performed on the multiple separated local discharge signals, and multiple groups of time-frequency feature vectors are output, wherein each group of time-frequency feature vectors includes a time domain feature vector, a frequency domain feature vector and a time-frequency difference feature vector; on the other hand, a local discharge feature library is trained, and the local discharge feature library includes time-frequency feature vector samples corresponding to local discharge type samples; then the local discharge feature library is used to perform pattern recognition on the multiple groups of time-frequency feature vectors respectively, and multiple discharge warning signals are output; finally, a local discharge warning reminder is given to the host computer of the transformer according to the multiple discharge warning signals. In other words, by combining technologies such as multi-path signal separation, time-frequency feature extraction and pattern recognition, signal interference of different paths can be effectively eliminated, ensuring that the characteristics of each signal source are clearer, thereby significantly improving the precision and accuracy of transformer local discharge warning, and ensuring the safe and stable operation of the transformer.

[0048] Embodiment 2: Based on the transformer partial discharge warning system combined with time-frequency characteristics in the above embodiment, the present application also provides a transformer partial discharge warning method combined with time-frequency characteristics, please refer to the attached Figure 2 ,include: A partial discharge signal of the transformer is obtained according to a partial discharge sensing device; the partial discharge signal is input into a multi-path signal separation model, superposition signal separation is performed on the partial discharge signal according to the multi-path signal separation model, and a plurality of separated partial discharge signals are output; time-frequency feature extraction is performed on the plurality of separated partial discharge signals, and a plurality of groups of time-frequency feature vectors are output, wherein each group of time-frequency feature vectors includes a time domain feature vector, a frequency domain feature vector and a time-frequency difference feature vector; a partial discharge feature library is trained, and the partial discharge feature library includes time-frequency feature vector samples corresponding to partial discharge type samples; pattern recognition is performed on the plurality of groups of time-frequency feature vectors respectively using the partial discharge feature library, and a plurality of discharge warning signals are output; and a partial discharge warning reminder is given to a host computer of the transformer according to the plurality of discharge warning signals.

[0049] Furthermore, the transformer partial discharge early warning method combined with time-frequency characteristics also includes: obtaining internal structure data and insulation distribution data of the transformer; modeling the transformer according to the internal structure data and insulation distribution data, and outputting a transformer simulation model; simulating the discharge propagation paths under different defect positions and intensities, and the time-frequency characteristic changes under each discharge propagation path according to the transformer simulation model, and outputting a simulation database; connecting the simulation database, and training a multi-path signal separation model according to an independent component analysis algorithm.

[0050] Furthermore, the transformer partial discharge warning method combined with time-frequency characteristics also includes: constructing an initialization separation model, the initialization separation model includes an initial random separation matrix, wherein the initial random separation matrix is ​​a random orthogonal matrix; connecting the simulation database to iteratively update the initial random separation matrix to obtain a converged separation matrix that meets the preset independence; using the converged separation matrix as the core parameter of the initialization separation model, and outputting a trained multipath signal separation model.

[0051] Furthermore, the transformer partial discharge early warning method combining time-frequency characteristics further includes: constructing an initialization separation model, the expression includes: ;in, is the multipath superposition signal matrix extracted from the simulation database, represents the signal that has not been propagated and superimposed through the path, represents the attenuation coefficient of signal propagation on the i-th path, represents the delay of the signal on the i-th path, N is the total number of paths, is a random interference signal.

[0052] Furthermore, the transformer partial discharge early warning method combined with time-frequency characteristics further includes: connecting the simulation database to iteratively update the initial random separation matrix, and the expression includes: ;in, is the initial random separation matrix, is the converged separation matrix after iterative update, is the multipath superposition signal matrix extracted from the simulation database, is a nonlinear comparison function used to enhance the independence of signals, T is the transpose of the matrix, is the derivative of the contrast function, used to calculate the gradient value of the signal to update the initial random separation matrix, and E is the expected operation for calculating the statistical characteristics of the multipath superposition signal matrix.

[0053] Furthermore, the transformer partial discharge early warning method combined with time-frequency characteristics also includes: obtaining a partial discharge signal of the transformer according to a partial discharge sensor device, wherein the partial discharge signal is a superimposed partial discharge signal corresponding to multiple discharge propagation paths at the same defect location; wherein the starting point of each discharge propagation path in the multiple discharge propagation paths is the defect location, and the end point of the path is the partial discharge sensor device.

[0054] Furthermore, the transformer partial discharge warning method combined with time-frequency features also includes: constructing a pattern recognizer, which is obtained by identifying the correspondence between the partial discharge feature library and the partial discharge type sample; inputting the multiple groups of time-frequency feature vectors into the pattern recognizer for feature similarity matching to obtain multiple matching discharge patterns; identifying the severity level according to the multiple matching discharge patterns, and outputting multiple identification results; and outputting multiple discharge warning signals according to the multiple identification results.

[0055] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The transformer partial discharge warning system and specific examples in the aforementioned embodiment 1 are also applicable to a transformer partial discharge warning method in combination with time-frequency characteristics in this embodiment. Through the aforementioned detailed description of a transformer partial discharge warning system in combination with time-frequency characteristics, those skilled in the art can clearly know a transformer partial discharge warning method in combination with time-frequency characteristics in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0056] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0057] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A transformer partial discharge early warning system combining time-frequency characteristics, characterized in that: The transformer partial discharge early warning system combining time-frequency characteristics includes: A partial discharge signal acquisition module, used for acquiring a partial discharge signal of the transformer according to a partial discharge sensor device; A superposition signal separation module, used for inputting the local discharge signal into a multi-path signal separation model, performing superposition signal separation on the local discharge signal according to the multi-path signal separation model, and outputting a plurality of separated local discharge signals; A time-frequency feature extraction module, used for extracting time-frequency features from the plurality of separated partial discharge signals, and outputting a plurality of groups of time-frequency feature vectors, wherein each group of time-frequency feature vectors includes a time domain feature vector, a frequency domain feature vector and a time-frequency difference feature vector; A partial discharge feature library training module, used for training a partial discharge feature library, wherein the partial discharge feature library includes time-frequency feature vector samples corresponding to partial discharge type samples; A time-frequency feature vector recognition module, used to use the local discharge feature library to perform pattern recognition on the multiple groups of time-frequency feature vectors respectively, and output multiple discharge warning signals; The partial discharge warning module is used to provide a partial discharge warning reminder to the host computer of the transformer according to the multiple discharge warning signals.

2. The transformer partial discharge early warning system combining time-frequency characteristics as claimed in claim 1 is characterized in that: Inputting the partial discharge signal into a multipath signal separation model comprises: Acquiring internal structure data and insulation distribution data of the transformer; Modeling the transformer according to the internal structure data and the insulation distribution data, and outputting a transformer simulation model; Simulating the discharge propagation paths under different defect positions and intensities and the time-frequency characteristic changes under each discharge propagation path according to the transformer simulation model, and outputting a simulation database; The simulation database is connected to train a multipath signal separation model according to an independent component analysis algorithm.

3. The transformer partial discharge early warning system combining time-frequency characteristics as claimed in claim 2 is characterized in that: Connecting the simulation database and training a multipath signal separation model according to an independent component analysis algorithm includes: Constructing an initialization separation model, wherein the initialization separation model includes an initial random separation matrix, wherein the initial random separation matrix is ​​a random orthogonal matrix; Connecting to the simulation database to iteratively update the initial random separation matrix to obtain a converged separation matrix that satisfies a preset independence; The converged separation matrix is ​​used as the core parameter of the initialization separation model, and a training multipath signal separation model is output.

4. The transformer partial discharge early warning system combining time-frequency characteristics as claimed in claim 3 is characterized in that: Construct an initial separation model, the expression includes: ； in, is the multipath superposition signal matrix extracted from the simulation database, represents the signal that has not been propagated and superimposed through the path, represents the attenuation coefficient of signal propagation on the i-th path, represents the delay of the signal on the i-th path, N is the total number of paths, is a random interference signal.

5. The transformer partial discharge early warning system combining time-frequency characteristics as claimed in claim 3 is characterized in that: The simulation database is connected to iteratively update the initial random separation matrix, and the expression includes: ； in, is the initial random separation matrix, is the converged separation matrix after iterative update, is the multipath superposition signal matrix extracted from the simulation database, is a nonlinear comparison function used to enhance the independence of signals, T is the transpose of the matrix, is the derivative of the contrast function, used to calculate the gradient value of the signal to update the initial random separation matrix, and E is the expected operation for calculating the statistical characteristics of the multipath superposition signal matrix.

6. The transformer partial discharge early warning system combining time-frequency characteristics as claimed in claim 1 is characterized in that: Acquire a partial discharge signal of the transformer according to the partial discharge sensor device, wherein the partial discharge signal is a superimposed partial discharge signal corresponding to multiple discharge propagation paths at the same defect location; Wherein, the starting point of each discharge propagation path among the multiple discharge propagation paths is the defect position, and the end point of the path is the local discharge sensor device.

7. The transformer partial discharge early warning system combining time-frequency characteristics according to claim 1 is characterized in that: The partial discharge feature library is used to perform pattern recognition on the multiple groups of time-frequency feature vectors respectively, and multiple discharge warning signals are output, including: Constructing a pattern recognizer, wherein the pattern recognizer is obtained by identifying the corresponding relationship between the partial discharge feature library and the partial discharge type sample; Inputting the multiple groups of time-frequency feature vectors into the pattern recognizer for feature similarity matching to obtain multiple matching discharge patterns; Perform severity level identification according to the multiple matching discharge modes, and output multiple identification results; According to the multiple identification results, multiple discharge warning signals are output.

8. A transformer partial discharge early warning method combining time-frequency characteristics, characterized in that: The method is implemented by the transformer partial discharge early warning system combined with time-frequency characteristics as described in any one of claims 1 to 7, comprising: Acquiring a partial discharge signal of the transformer according to a partial discharge sensor device; Inputting the local discharge signal into a multi-path signal separation model, performing superposition signal separation on the local discharge signal according to the multi-path signal separation model, and outputting a plurality of separated local discharge signals; Extracting time-frequency features from the plurality of separated partial discharge signals, and outputting a plurality of groups of time-frequency feature vectors, wherein each group of time-frequency feature vectors includes a time-domain feature vector, a frequency-domain feature vector, and a time-frequency difference feature vector; Training a partial discharge feature library, wherein the partial discharge feature library includes time-frequency feature vector samples corresponding to partial discharge type samples; Using the partial discharge feature library to perform pattern recognition on the multiple groups of time-frequency feature vectors respectively, and output multiple discharge warning signals; A partial discharge warning reminder is sent to a host computer of the transformer according to the multiple discharge warning signals.

Images