Early warning method and device for partial discharge of switch cabinet, electronic equipment and storage medium

Abstract

The invention provides an early warning method and device for partial discharge of a switch cabinet, electronic equipment and a storage medium, and relates to the technical field of power monitoring, and the method comprises the steps: collecting a high-frequency pulse current signal through an HFCT; performing noise reduction processing on the high-frequency pulse current signal to obtain a partial discharge signal; preprocessing the partial discharge signal by using the target neural network model to obtain a plurality of frames of preprocessed signals, performing feature extraction on each frame of preprocessed signal in the plurality of frames of preprocessed signals to obtain a plurality of local features, and predicting a discharge type corresponding to each local feature and a probability value of the discharge type, the target neural network model is obtained by training a historical sample data set; and triggering a partial discharge early warning signal under the condition that a group of prediction results meet a preset condition. By implementing the technical scheme provided by the invention, the effect of improving the accuracy and timeliness of fault early warning is achieved.

Description

Technical Field

[0001] The present application relates to the technical field of power monitoring, and particularly relates to a warning method, device, electronic device and storage medium for partial discharge of a switch cabinet. Background Art

[0002] During the operation of high-voltage power equipment (such as switch cabinets, GIS), partial discharge (PD) is an important sign of insulation deterioration, and its detection is crucial for preventing equipment failures. In related technologies, the partial discharge monitoring of switch cabinets mainly relies on traditional earth current wave, ultrasonic wave, pulse current method and ultra-high frequency method. These methods have problems such as weak anti-interference ability, low sensitivity and low reliability, and it is difficult to accurately detect partial discharge events, resulting in low maintenance efficiency of power equipment and increasing the risk of equipment failure; in addition, existing systems mostly rely on manual analysis of spectrograms, adopt threshold judgment or fixed algorithms, and judge whether partial discharge occurs according to the real-time amplitude detected currently, lacking the analysis and utilization of historical data, restricting the accuracy and timeliness of fault warning, and it is also difficult to meet the automation requirements of the smart grid for condition monitoring. Summary of the Invention

[0003] In order to solve the above technical problems, the present application provides a warning method, device, electronic device and storage medium for partial discharge of a switch cabinet.

[0004] In a first aspect, the present application provides a warning method for partial discharge of a switch cabinet, including: collecting high-frequency pulse current signals through a high-frequency current transformer HFCT, where the high-frequency pulse current signals include partial discharge signals generated when a target switch cabinet has a partial discharge; performing noise reduction processing on the high-frequency pulse current signals to obtain partial discharge signals; using a target neural network model to preprocess the partial discharge signals to obtain multiple frames of preprocessed signals, extracting features from each frame of the preprocessed signals among the multiple frames of preprocessed signals to obtain multiple local features, and respectively predicting the discharge type and the probability value of the discharge type corresponding to each local feature to obtain a set of prediction results, where the target neural network model is trained using a historical sample data set, and the historical sample data set includes partial discharge signal sample data of different discharge types, and each prediction result includes the discharge type and probability value of a corresponding local feature; triggering a partial discharge warning signal when a set of prediction results meets a preset condition.

[0005] By adopting the above technical solution, the high-frequency pulse current transformer HFCT is used to collect high-frequency pulse current signals containing partial discharge signals, and noise reduction processing is performed on them to obtain more accurate partial discharge signals, which can improve the anti-interference ability and detection sensitivity; the target neural network model trained based on the historical sample data set is used to preprocess, extract features, and predict the discharge type and probability value of the partial discharge signal, which can make full use of historical data and improve the accuracy and timeliness of fault warning; when the prediction result meets the preset conditions, the partial discharge warning signal is triggered, which can detect the partial discharge event of the switch cabinet in advance, reduce the equipment failure risk, and improve the maintenance efficiency of power equipment.

[0006] Optionally, when a set of prediction results meets the preset conditions, triggering the partial discharge warning signal includes: when a set of prediction results indicates that the discharge types corresponding to the preprocessing signals of consecutive preset frames in multiple frames of preprocessing signals are preset discharge types, and the probability values of the preset discharge types are all greater than or equal to the preset probability threshold, triggering the partial discharge warning signal.

[0007] By adopting the above technical solution, when a set of prediction results indicates that the discharge types corresponding to the preprocessing signals of consecutive preset frames in multiple frames of preprocessing signals are preset discharge types and the probability values are all greater than or equal to the preset probability threshold, triggering the partial discharge warning signal can improve the anti-interference ability, sensitivity and reliability of the partial discharge detection of the switch cabinet, and based on the training of the historical sample data set and the comprehensive judgment of multiple frames of signals for early warning, the accuracy and timeliness of the fault warning are improved. Through the dual judgment mechanism of multi-frame continuity and discharge type recognition probability, the problem of false alarms caused by single-frame misjudgment or noise interference in traditional methods is solved, and the accuracy, stability and intelligent level of the early warning system are significantly improved.

[0008] Optionally, feature extraction is respectively performed on each frame of the preprocessing signals in multiple frames of preprocessing signals to obtain multiple local features, and the discharge type and the probability value of the discharge type corresponding to each local feature are respectively predicted to obtain a set of prediction results, including: the target frame preprocessing signal is any one of the preprocessing signals in multiple frames of preprocessing signals, and the target prediction result is obtained by processing the target frame preprocessing signal in the following manner: using the target neural network model to extract features from the target frame preprocessing signal to obtain the target local feature, and performing classification and recognition based on the target local feature to obtain the target prediction result, where the target prediction result includes the target discharge type and the probability value of the target discharge type, the target discharge type is used to represent the discharge type corresponding to the target local feature predicted by the target neural network model, and the probability value of the target discharge type is used to represent the probability that the discharge type of the target local feature predicted by the target neural network model is the target discharge type.

[0009] By adopting the above technical solution, local features can be accurately extracted from each frame of the preprocessed signals among multiple frames of preprocessed signals. For each local feature, the discharge type and the corresponding probability value are accurately predicted. For the preprocessed signal of the target frame, a target prediction result including the target discharge type and the probability value can be obtained, improving the accuracy and reliability of the recognition of the local discharge type in the switchgear cabinet. Furthermore, it helps to more timely and accurately warn of the local discharge situation, improve the maintenance efficiency of power equipment, and reduce the equipment failure risk.

[0010] Optionally, the target neural network model is used to extract features from the preprocessed signal of the target frame to obtain the target local features, and, based on the target local features, classification and recognition are performed to obtain the target prediction result, including: The target neural network model includes an input layer, a convolutional layer, an activation function layer, a pooling layer, and a fully connected layer. The input layer receives the partial discharge signal and preprocesses the partial discharge signal to obtain multiple frames of preprocessed signals. The convolutional layer extracts the target local features through a sliding convolutional kernel; the activation function layer introduces nonlinearity, the pooling layer reduces the data dimension, and the fully connected layer maps the extracted features to the classification result to obtain the target prediction result.

[0011] By adopting the above technical solution, the input layer of the target neural network model receives the partial discharge signal and preprocesses it to obtain multiple frames of preprocessed signals. The convolutional layer extracts the target local features through a sliding convolutional kernel. The activation function layer introduces nonlinearity, the pooling layer reduces the data dimension, and the fully connected layer maps the extracted features to the classification result to obtain the target prediction result, which can improve the anti-interference ability and sensitivity of the partial discharge detection in the switchgear cabinet, accurately identify the partial discharge signal and the corresponding discharge type and probability value, and improve the accuracy and timeliness of the fault warning. The entire feature extraction and classification and recognition process is automatically completed by the neural network model without manual intervention, realizing the intelligence of the partial discharge detection and meeting the requirements of the smart grid for automatic monitoring.

[0012] Optionally, the target neural network model is used to preprocess the partial discharge signal to obtain multiple frames of preprocessed signals, including: The target neural network model is used to perform normalization processing on the partial discharge signal to obtain a normalized discharge signal, where the high-frequency pulse current signal is a one-dimensional time series signal; the normalized discharge signal is framed to obtain multiple frames of preprocessed signals, where each frame of the preprocessed signals among the multiple frames of preprocessed signals is a signal of a preset length.

[0013] By adopting the above technical solution, the partial discharge signal is normalized and framed by using the target neural network model to obtain multi-frame preprocessed signals of a preset length, which is beneficial to subsequent feature extraction, classification and recognition, and prediction of the discharge type and probability value. The signal frames after normalization and framing have a unified numerical range and a fixed length, which can better adapt to the input requirements of the neural network model, improve the training efficiency and stability of the model, and provide a better data basis for subsequent feature extraction and classification and recognition, thereby improving the performance of the entire partial discharge warning system, including detection accuracy, reliability and real-time performance.

[0014] Optionally, the high-frequency pulse current signal is denoised to obtain a partial discharge signal, including: performing 5-layer wavelet decomposition on the high-frequency pulse current signal by using the db4 wavelet basis to obtain a target decomposition signal; performing hierarchical threshold processing on a group of high-frequency coefficients in the target decomposition signal to remove high-frequency noise and obtain processed coefficients, wherein the third to fifth layer high-frequency coefficients are retained as effective partial discharge features and the first to second layer high-frequency noise components are suppressed in the processed coefficients; performing wavelet reconstruction based on the processed coefficients to obtain a denoised partial discharge signal.

[0015] By adopting the above technical solution, the high-frequency pulse current signal is subjected to 5-layer wavelet decomposition by using the db4 wavelet basis, the high-frequency coefficients are subjected to hierarchical threshold processing to remove high-frequency noise and retain effective partial discharge features, and then wavelet reconstruction is performed. Through wavelet decomposition and threshold processing, high-frequency noise can be effectively removed, partial discharge features can be effectively retained, and noise can be suppressed, significantly improving the anti-interference ability of the signal.

[0016] Optionally, before the partial discharge signal is preprocessed by using the target neural network model to obtain multi-frame preprocessed signals, the above method further includes: obtaining a historical sample data set, which includes partial discharge signal sample data of different discharge types and corresponding actual sample results, wherein the actual sample results are used to represent the results of the actual discharge types corresponding to the partial discharge signal sample data; training the original neural network model by using the historical sample data set until the loss value between the predicted sample results output by the original neural network model and the pre-determined actual sample results meets a preset convergence condition, ending the training, and determining the original neural network model at the end of the training as the target neural network model.

[0017] By adopting the above technical solutions, a historical sample data set containing partial discharge signal sample data of different discharge types and corresponding actual sample results is obtained, and the original neural network model is trained, enabling the model to learn the characteristic laws of different discharge types. When the loss value between the predicted sample result and the actual sample result satisfies the convergence condition during training, the target neural network model is determined, which can ensure that the target neural network model has high accuracy and reliability. Furthermore, when processing the partial discharge signal and predicting the discharge type subsequently, it is more accurate, improving the accuracy and timeliness of the partial discharge warning of the switchgear and solving the problems that the existing system lacks the analysis and utilization of historical data and the fault warning is inaccurate and untimely.

[0018] Optionally, the high-voltage electrode and the outer shell of the target switchgear are respectively connected to the high-frequency pulse current sampling circuit. The high-frequency pulse current sampling circuit includes a coupling capacitor, a detection impedance, a calibration capacitor, a calibration pulse generator, and an AC power supply. Among them, the first end of the coupling capacitor is connected to the high-voltage electrode of the target switchgear, the second end of the coupling capacitor is connected to the ground terminal through the detection impedance, the outer shell of the target switchgear is connected to the ground terminal, the first end and the second end of the coupling capacitor are respectively connected to both ends of the AC power supply, the calibration capacitor and the calibration pulse generator are connected in series and then connected in parallel to both ends of the AC power supply, the HFCT is sleeved in the grounding loop, and the coupling capacitor, the detection impedance, and the target switchgear form a grounding loop. The coupling capacitor is used to capture the voltage signal generated by the partial discharge inside the target switchgear.

[0019] By adopting the above technical solutions, the high-voltage electrode and the outer shell of the target switchgear are respectively connected to the high-frequency pulse current sampling circuit. The coupling capacitor in this circuit can capture the voltage signal generated by the partial discharge inside the target switchgear. Cooperating with components such as the detection impedance, the calibration capacitor, the calibration pulse generator, and the AC power supply, the HFCT can be sleeved in the grounding loop formed by the coupling capacitor, the detection impedance, and the target switchgear to collect high-frequency pulse current signals, which is beneficial to accurately monitoring and warning the partial discharge situation of the switchgear subsequently.

[0020] Optionally, the preset discharge types include at least one of the following: surface discharge type, internal discharge type, corona discharge type; the output layer of the target neural network model includes three independent output nodes, corresponding to the probability of internal discharge type, the probability of surface discharge type, and the probability of corona discharge type respectively.

[0021] By adopting the above technical solution, it is clear that the preset discharge types cover surface discharge type, internal discharge type, and corona discharge type. When a set of prediction results meet specific conditions to trigger a partial discharge warning signal, it can more accurately conduct partial discharge warning for these common discharge types, improving the accuracy and pertinence of the partial discharge warning of the switch cabinet, thereby improving the maintenance efficiency of power equipment and reducing the equipment failure risk. The output layer of the target neural network model is provided with three independent output nodes corresponding to the probability of internal discharge type, the probability of surface discharge type, and the probability of corona discharge type respectively, which can output the probabilities of different discharge types more clearly and specifically, helping to accurately judge the partial discharge situation of the switch cabinet and improving the accuracy of the partial discharge warning.

[0022] In the second aspect of the present application, there is also provided a warning device for partial discharge of a switch cabinet, including: a collection module for collecting high-frequency pulse current signals through a high-frequency pulse current transformer HFCT, where the high-frequency pulse current signals include partial discharge signals generated when partial discharge occurs in the target switch cabinet; a noise reduction module for performing noise reduction processing on the high-frequency pulse current signals to obtain partial discharge signals; a processing module for preprocessing the partial discharge signals by using the target neural network model to obtain multiple frames of preprocessed signals, extracting features from each frame of the preprocessed signals among the multiple frames of preprocessed signals to obtain multiple local features, and respectively predicting the discharge type and the probability value of the discharge type corresponding to each local feature to obtain a set of prediction results, where the target neural network model is trained by using a historical sample data set, and the historical sample data set includes partial discharge signal sample data of different discharge types, and each prediction result includes the discharge type and probability value of a corresponding local feature; a trigger module for triggering a partial discharge warning signal when a set of prediction results meet the preset conditions.

[0023] In the third aspect of the present application, there is also provided an electronic device, including a memory and a processor, where a computer program is stored on the memory, and when the processor executes the program, the method steps of any one of the above are implemented.

[0024] In the fourth aspect of the present application, there is also provided a computer-readable storage medium, where instructions are stored on the computer-readable storage medium, and when the instructions are executed, the method steps of any one of the above are executed.

[0025] In summary, one or more technical solutions provided in the present application have at least the following technical effects or advantages: 1. It can improve the anti-interference ability and detection sensitivity, make full use of historical data, improve the accuracy and timeliness of fault warning, trigger a partial discharge warning signal when the prediction results meet the preset conditions, can detect partial discharge events of the switch cabinet in advance, and reduce the equipment failure risk; 2. The accuracy, stability and intelligence level of the early warning system are significantly improved through a dual judgment mechanism of multi-frame continuity and discharge type recognition probability; 3. The high-frequency pulse current signal is decomposed by 5 layers using the db4 wavelet basis, the high-frequency coefficients are processed by hierarchical thresholding to remove high-frequency noise and retain effective partial discharge characteristics, and then wavelet reconstruction is performed. Through wavelet decomposition and threshold processing, high-frequency noise can be effectively removed, partial discharge characteristics can be effectively retained, and noise can be suppressed, significantly improving the anti-interference ability of the signal. Description of the Drawings

[0026] Figure 1 is a flowchart of a method for early warning of partial discharge in a switch cabinet provided by an embodiment of the present application; Figure 2 is a schematic diagram of a high-frequency pulse current sampling circuit provided by an embodiment of the present application; Figure 3 is a schematic diagram of a new type of partial discharge early warning principle for a switch cabinet provided by an embodiment of the present application; Figure 4 is a structural block diagram of an early warning device for partial discharge in a switch cabinet provided by an embodiment of the present application; Figure 5 is a schematic diagram of the structure of an electronic device disclosed by an embodiment of the present application.

[0027] Description of the reference numerals: 500 - electronic device; 501 - processor; 502 - communication bus; 503 - user interface; 504 - network interface; 505 - memory. Detailed Embodiments

[0028] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0029] In the description of the embodiments of the present application, words such as "for example" or "for instance" are used to indicate examples, illustrations or explanations. Any embodiment or design solution described as "for example" or "for instance" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "for example" or "for instance" is intended to present relevant concepts in a specific manner.

[0030] In the description of the embodiments of the present application, the term "a plurality of" means two or more. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. The terms "comprising", "including", "having" and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0031] The present application provides a method for warning of partial discharge in a switchgear cabinet. Referring to Figure 1 , Figure 1 FIG. is a flowchart of a method for warning of partial discharge in a switchgear cabinet provided by an embodiment of the present application. The method includes: Step S101: Collect high-frequency pulse current signals through a high-frequency pulse current transformer HFCT. Among them, the high-frequency pulse current signals include partial discharge signals generated when partial discharge occurs in the target switchgear cabinet; Step S102: Perform noise reduction processing on the high-frequency pulse current signals to obtain partial discharge signals; Step S103: Use the target neural network model to preprocess the partial discharge signals to obtain multiple frames of preprocessed signals. Extract features from each frame of the preprocessed signals among the multiple frames of preprocessed signals to obtain multiple local features, and respectively predict the discharge type and the probability value of the discharge type corresponding to each local feature to obtain a set of prediction results. Among them, the target neural network model is trained using a historical sample data set, and the historical sample data set includes partial discharge signal sample data of different discharge types. Each prediction result includes the discharge type and probability value of a corresponding local feature; Step S104: Trigger a partial discharge warning signal when a set of prediction results meets a preset condition.

[0032] Through the above steps, use the high-frequency pulse current transformer HFCT to collect high-frequency pulse current signals containing partial discharge signals, and perform noise reduction processing on them to obtain more accurate partial discharge signals, which can improve the anti-interference ability and detection sensitivity; use the target neural network model trained based on the historical sample data set to preprocess, extract features, and predict the discharge type and probability value of the partial discharge signals, which can make full use of historical data and improve the accuracy and timeliness of fault warning; trigger a partial discharge warning signal when the prediction results meet the preset conditions, which can detect partial discharge events in the switchgear cabinet in advance, reduce the risk of equipment failure, and improve the maintenance efficiency of power equipment.

[0033] This embodiment proposes a method for early warning of partial discharge in switchgear by combining the high-frequency pulse current method HFCT with a deep learning model. The high-frequency pulse current signals generated during partial discharge in the switchgear are collected by HFCT sensors, and the high-frequency pulse current signals are denoised to retain the true partial discharge signal components. The partial discharge signals contained therein are extracted. The pre-trained target neural network model is used to preprocess the denoised signals to obtain multiple frames of preprocessed signals. The local features of each frame are extracted respectively, and each local feature is classified and identified to predict the discharge type and its occurrence probability value corresponding to each frame of preprocessed signal. That is, the pre-trained target neural network model is used to process the signal in frames, extract local features frame by frame and predict the discharge type and probability value. An early warning is triggered through comprehensive judgment (preset conditions) of multiple frames of results, replacing the traditional threshold method or fixed algorithm, reducing human error and adapting to complex working conditions; when a set of prediction results meet the preset conditions, the system determines that there is a risk of partial discharge, thus triggering an early warning signal. The methods in the related art (such as earth electrical wave, ultrasonic wave, etc.) are easily affected by external interference, resulting in inaccurate detection results. However, the method of this embodiment collects signals through a high-frequency current transformer and combines a neural network model for processing, which can effectively reduce the influence of interference signals and improve the anti-interference ability of detection; the methods in the related art have low detection sensitivity to partial discharge signals and are difficult to accurately capture weak partial discharge signals, resulting in insufficient reliability. The method of this embodiment preprocesses and extracts features from signals through a neural network model, which can more sensitively identify partial discharge signals and improve the reliability of detection; in addition, this embodiment uses a historical sample data set to train the target neural network model, which can make full use of historical data to improve the accuracy and timeliness of fault early warning; and an early warning signal is triggered according to preset conditions, realizing the automation of partial discharge monitoring and meeting the automation requirements of the smart grid for condition monitoring.

[0034] In an optional embodiment, when a set of prediction results meet the preset conditions, triggering a partial discharge early warning signal includes: when a set of prediction results indicate that the discharge types corresponding to the preprocessed signals of consecutive preset frames exist in multiple frames of preprocessed signals and are the preset discharge types, and the probability values of the preset discharge types are all greater than or equal to the preset probability threshold, triggering the partial discharge early warning signal.

[0035] In the above embodiments, when a set of prediction results indicates that the discharge types corresponding to the preprocessing signals with a continuous preset number of frames in multiple frames of preprocessing signals are preset discharge types and the probability values are all greater than or equal to a preset probability threshold, a partial discharge warning signal is triggered, which can improve the anti-interference ability, sensitivity, and reliability of the local discharge detection of the switch cabinet. Moreover, by training the model based on the historical sample data set and combining the comprehensive judgment of multiple frames of signals for warning, the accuracy and timeliness of the fault warning are improved. Through the dual judgment mechanism of multi-frame continuity and discharge type recognition probability, the false alarm problem caused by single-frame misjudgment or noise interference in traditional methods is solved, and the accuracy, stability, and intelligent level of the warning system are significantly improved.

[0036] When the preprocessing signals of multiple consecutive frames (preset number of frames) are all recognized as the same preset discharge type (such as corona discharge, surface discharge, etc.), if the preset number of frames is 3 frames (or other number of frames), to avoid false alarms caused by single-frame misjudgment, the probability value corresponding to each frame needs to be ≥ the preset threshold (such as 90%, or 85%, or other), ensuring a high confidence level of the prediction result and further excluding the interference of noise or interference signals; through dual-condition triggering, only when the types of consecutive frames are the same and the probabilities all meet the standards, that is, only when these two conditions are met simultaneously, will the partial discharge warning signal be triggered. This method improves the accuracy and reliability of the warning signal by combining the dual constraints of time and probability. The traditional threshold method or single detection is prone to false triggering of alarms due to instantaneous interference (such as switch operation noise). In this embodiment, false alarms are reduced through consistent judgment of multiple consecutive frames. In this embodiment, by setting the dual conditions of the number of consecutive frames and the probability threshold, a warning is triggered only when multiple consecutive signal frames all meet the high-probability partial discharge characteristics, effectively avoiding false alarms caused by accidental noise or interference signals and improving the accuracy of the warning signal; introducing the probability threshold ensures that the prediction result has a high credibility and further enhances the reliability of the warning signal; by evaluating the discharge type and probability of consecutive signal frames, the characteristics of partial discharge events can be captured more comprehensively, improving the recognition ability of partial discharge events. Especially in complex working conditions, the occurrence of partial discharge events can be judged more accurately. In addition, the preset number of frames and the probability threshold can be flexibly adjusted according to the actual application scenario, making the system applicable to different device types and operating environments; by integrating the AI recognition results with the dynamic judgment logic, intelligent perception and adaptive warning of partial discharge events are realized, meeting the development needs of the smart grid.

[0037] In an alternative embodiment, feature extraction is performed on each preprocessed signal in a multi-frame preprocessed signal to obtain multiple local features, and the discharge type and the probability value of the discharge type corresponding to each local feature are predicted respectively to obtain a set of prediction results, including: the target frame preprocessed signal is any one of the multi-frame preprocessed signals, and the target prediction result is obtained by processing the target frame preprocessed signal in the following manner: the target neural network model is used to perform feature extraction on the target frame preprocessed signal to obtain a target local feature, and classification and recognition are performed based on the target local feature to obtain a target prediction result, where the target prediction result includes the target discharge type and the probability value of the target discharge type, the target discharge type is used to represent the discharge type corresponding to the target local feature predicted by the target neural network model, and the probability value of the target discharge type is used to represent the probability that the discharge type of the target local feature predicted by the target neural network model is the target discharge type.

[0038] In the above embodiment, local features can be accurately extracted from each preprocessed signal in the multi-frame preprocessed signal, the discharge type and the corresponding probability value can be accurately predicted for each local feature, a target prediction result including the target discharge type and the probability value can be obtained for the target frame preprocessed signal, the accuracy and reliability of the local discharge type recognition of the switch cabinet can be improved, and further, it is helpful to more timely and accurately warn of the local discharge situation, improve the maintenance efficiency of power equipment, and reduce the equipment failure risk.

[0039] For each frame of preprocessed signal, such as the target frame preprocessed signal, feature extraction and classification recognition are performed through the target neural network model to obtain a target prediction result including the target discharge type and its probability value. The above set of prediction results includes this target prediction result. Specifically, the target neural network model is used to process the target frame preprocessed signal to extract target local features that can reflect the characteristics of partial discharge, such as time-frequency domain features, pulse waveform features, etc. Based on the extracted target local features, classification recognition is performed through the neural network model to predict the discharge type (target discharge type) corresponding to this frame of signal and its probability value, generating a target prediction result, which contains the target discharge type and the corresponding probability value, and is used for subsequent analysis and early warning judgment to achieve the purpose of intelligent diagnosis. In related technologies, partial discharge monitoring mostly relies on the analysis of the overall signal or single-threshold judgment, lacking detailed analysis of each frame of signal. In this embodiment, by processing the preprocessed signal frame by frame, the characteristics of each frame of signal can be analyzed more meticulously, so as to more accurately identify partial discharge events. By performing feature extraction and classification recognition on each frame of preprocessed signal, the characteristics of partial discharge can be more accurately identified, improving the detection accuracy. By introducing probability values, a quantitative evaluation is provided for the prediction results, enabling operation and maintenance personnel to more intuitively understand the reliability of the prediction results, and thus more accurately judge the occurrence of partial discharge events. The target neural network model can automatically learn and recognize the complex features of the signal, perform frame-by-frame analysis on each frame of signal, effectively process complex signals, and improve the accuracy and reliability of detection. The entire feature extraction and classification recognition process is automatically completed by the neural network model without manual intervention, realizing the intelligence of partial discharge detection and meeting the requirements of the smart grid for automated monitoring.

[0040] In this embodiment, by utilizing the powerful feature learning ability of the neural network, target local features are automatically extracted from the target frame preprocessed signal. These features may include the time-domain features of the signal (such as pulse width, rise time), frequency-domain features (such as frequency distribution, energy spectrum), or time-frequency domain combined features (such as the phase distribution characteristics of partial discharge); based on the extracted target local features, the target neural network model predicts the discharge type through its internal classification layer (such as fully connected layer, softmax layer), and outputs a target prediction result, including the target discharge type (such as specific types like corona discharge, internal discharge, etc.) and the probability value of the target discharge type. This probability value reflects the confidence level of the model in the prediction result. By processing multiple frames of signals one by one, a set of prediction results covering multiple local feature predictions is obtained, providing a data basis for subsequent early warning judgment.

[0041] In an optional embodiment, a target neural network model is used to extract features from the target frame preprocessing signal to obtain target local features, and further, classification and recognition are performed based on the target local features to obtain a target prediction result, including: The target neural network model includes an input layer, a convolutional layer, an activation function layer, a pooling layer, and a fully connected layer. The input layer receives the partial discharge signal and preprocesses the partial discharge signal to obtain multiple frames of preprocessing signals. The convolutional layer extracts the target local features through a sliding convolutional kernel; the activation function layer introduces non-linearity, the pooling layer reduces the data dimension, and the fully connected layer maps the extracted features to the classification result to obtain the target prediction result.

[0042] In the above embodiment, the input layer of the target neural network model receives the partial discharge signal for preprocessing to obtain multiple frames of preprocessing signals. The convolutional layer extracts the target local features through a sliding convolutional kernel. The activation function layer introduces non-linearity, the pooling layer reduces the data dimension, and the fully connected layer maps the extracted features to the classification result to obtain the target prediction result, which can improve the anti-interference ability and sensitivity of the partial discharge detection of the switch cabinet, accurately identify the partial discharge signal and the corresponding discharge type and probability value, and improve the accuracy and timeliness of fault warning. The entire feature extraction and classification and recognition process is automatically completed by the neural network model without manual intervention, realizing the intelligence of partial discharge detection and meeting the requirements of the smart grid for automatic monitoring.

[0043] By constructing a target neural network model including an input layer, a convolutional layer, an activation function layer, a pooling layer, and a fully connected layer, the target frame preprocessing signal is processed layer by layer. Input layer: As the data entry, it receives the partial discharge signal, preprocesses the partial discharge signal to obtain multiple frames of preprocessing signals, and provides a data basis for subsequent processing. Convolutional layer: Using a sliding convolutional kernel to perform convolution operations on each frame of preprocessing signal data (including the target frame preprocessing signal), automatically extracting local features in the signal through the weight parameters of the convolutional kernel, such as specific waveform patterns and frequency features of the partial discharge signal. The convolution operation can effectively capture the local correlation of the signal and extract key features related to the discharge type. Activation function layer: After the convolutional layer extracts features, introduce non-linear transformations (such as ReLU, Sigmoid, etc. functions), break the linear relationship of the data, enable the neural network to learn and express more complex non-linear features, and enhance the fitting ability of the model to the complex features of the partial discharge signal. Pooling layer: Perform dimensionality reduction processing on the data output by the activation function layer. By means of max pooling or average pooling, etc., while retaining key features, reduce the amount of data, reduce the computational complexity of the model, improve the generalization ability of the model, avoid overfitting phenomena, and at the same time have a certain robustness to small offsets and deformations of the signal. Fully connected layer: Integrate the features output by the pooling layer, map the extracted features to different discharge type categories through weight connections, output the probability value corresponding to each category, and finally obtain the target prediction result including the target discharge type and probability value, realizing the classification and recognition of the partial discharge type. The partial discharge detection methods in the related technologies are difficult to effectively extract the key features in complex discharge signals, especially the feature extraction effect on weak or interfered signals is poor. In this embodiment, the convolutional layer automatically learns features through the convolutional kernel, can adaptively extract the effective features of various partial discharge signals, and solves the problem of insufficient feature extraction in traditional methods. The activation function layer introduces non-linearity, enabling the model to learn more complex patterns and rules in the signal, overcoming the limitations of linear models, and enhancing the expression ability of the model to the features of the partial discharge signal. The pooling layer reduces the amount of data calculation through dimensionality reduction processing, improves the processing efficiency, and at the same time enhances the generalization ability of the model. The fully connected layer effectively integrates and classifies the features, avoiding classification errors caused by improper feature processing, and solving the problems of poor data processing and classification effects in traditional methods.

[0044] In an optional embodiment, the target neural network model is used to preprocess the partial discharge signal to obtain multiple frames of preprocessing signals, including: using the target neural network model to perform normalization processing on the partial discharge signal to obtain a normalized discharge signal, where the high-frequency pulse current signal is a one-dimensional time series signal; performing frame splitting on the normalized discharge signal to obtain multiple frames of preprocessing signals, where each frame of preprocessing signal in the multiple frames of preprocessing signals is a signal of a preset length.

[0045] In the above embodiments, the target neural network model is used to perform normalization processing and frame segmentation on the partial discharge signal to obtain multi-frame preprocessed signals of a preset length, which is beneficial for subsequent feature extraction, classification and recognition, and prediction of the discharge type and probability value. The signal frames after normalization and frame segmentation have a unified numerical range and a fixed length, which can better adapt to the input requirements of the neural network model, improve the training efficiency and stability of the model, provide a better data basis for subsequent feature extraction and classification and recognition, and thus improve the performance of the entire partial discharge warning system, including detection accuracy, reliability, and real-time performance.

[0046] Since the high-frequency pulse current signal is a one-dimensional time series signal, the amplitude range and distribution of its original data may vary greatly. Using the target neural network model to perform normalization processing on the partial discharge signal and map the signal data to a specific interval (such as [0, 1] or [-1, 1]) can unify the data scale, eliminate the influence caused by the amplitude difference between different signals, enable the neural network to learn features more efficiently during subsequent processing, and improve the stability of model training and prediction; the normalized discharge signal is segmented according to a fixed length to obtain multi-frame preprocessed signals, and each frame of signal has a preset length; frame segmentation converts the continuous time series signal into discrete data units with a fixed format. On the one hand, it is convenient for the neural network to perform parallel processing frame by frame, improving the calculation efficiency; on the other hand, different frames can reflect the characteristics of the partial discharge signal in different time segments, providing data support for subsequent analysis of the discharge law from the time dimension. Since the amplitude of the partial discharge signal may vary greatly due to factors such as equipment status and environmental conditions, directly processing the original signal may lead to inaccurate detection results. Normalization processing can eliminate this amplitude difference, make the signal have a unified numerical range, and thus improve the detection accuracy. By normalizing to unify the data scale, the neural network can more accurately learn the signal features, accelerate the model convergence speed, reduce the risk of overfitting, improve the generalization ability of the model to different partial discharge signals, and thus enhance the accuracy and stability of partial discharge type prediction; frame segmentation makes the signal data structured, facilitating parallel computing of the neural network and greatly improving the signal processing efficiency; at the same time, by analyzing the characteristics of different frames, the dynamic change law of the partial discharge signal in the time dimension can be mined, such as the fluctuation of the discharge intensity over time and the transition of the discharge type, providing rich information for more accurate fault warning and diagnosis; the signals after normalization and frame segmentation have a unified format, which can better adapt to different neural network models and analysis algorithms, facilitate the expansion and upgrade of the system, and improve the adaptability of the entire partial discharge monitoring system to complex power environments and diverse equipment operating conditions.

[0047] Suppose we are monitoring the partial discharge situation of a high-voltage switchgear. The high-frequency pulse current signal collected by a high-frequency current transformer (HFCT) is a one-dimensional time series signal. For subsequent feature extraction and analysis, we need to perform normalization processing and framing processing on this signal. Suppose the collected high-frequency pulse current signal is a time series signal with a length of 1000, and the specific values are as follows (only part of the data is listed here): [120, 150, 180, 200, 220, 250, 280, 300, 320, 350,..., 500] Step 1: Normalization processing. The purpose of normalization processing is to standardize the amplitude range of the signal to a fixed interval (for example, between 0 and 1); the normalization formula is as follows: normalized_value = (original_value - min_value) / (max_value - min_value); suppose the maximum value of the signal is 500 and the minimum value is 120. After performing normalization processing on the above signal, the normalized signal obtained is as follows: [0.00, 0.06, 0.12, 0.16, 0.20, 0.26, 0.32, 0.36, 0.40, 0.46,..., 1.00]; Step 2: Framing processing. Framing processing is to divide the normalized signal into multiple signal frames with a fixed length. Suppose we choose the length of each frame to be 100 and there is no overlap between frames. Then, the framed signal is as follows: Frame 1: [0.00, 0.06, 0.12, 0.16, 0.20, 0.26, 0.32, 0.36, 0.40, 0.46,..., 0.96]; Frame 2: [0.96, 1.00, 0.94, 0.88, 0.84, 0.80, 0.76, 0.72, 0.68, 0.64,..., 0.04]; Frame 3: [0.04, 0.08, 0.12, 0.16, 0.20, 0.24, 0.28, 0.32, 0.36, 0.40,..., 0.96]..., and so on until the entire normalized signal is divided up.

[0048] In an optional embodiment, the high-frequency pulse current signal is denoised to obtain a partial discharge signal, including: performing 5-layer wavelet decomposition on the high-frequency pulse current signal using the db4 wavelet basis to obtain a target decomposition signal; performing hierarchical threshold processing on a group of high-frequency coefficients in the target decomposition signal to remove high-frequency noise, obtaining processed coefficients, wherein the third to fifth layer high-frequency coefficients are retained in the processed coefficients as effective partial discharge features and the high-frequency noise components of the first to second layers are suppressed; performing wavelet reconstruction based on the processed coefficients to obtain a denoised partial discharge signal.

[0049] In the above embodiments, the db4 wavelet basis is used to perform 5-layer wavelet decomposition on the high-frequency pulsed current signal, the high-frequency coefficients are processed by hierarchical thresholding to remove high-frequency noise and retain effective partial discharge characteristics, and then wavelet reconstruction is performed. Through wavelet decomposition and threshold processing, high-frequency noise can be effectively removed, the partial discharge characteristics can be effectively retained, and noise can be suppressed, significantly improving the anti-interference ability of the signal.

[0050] The db4 wavelet basis is used to perform 5-layer wavelet decomposition on the high-frequency pulsed current signal, decomposing the signal into coefficients of different frequency components. Wavelet decomposition can decompose the signal into a low-frequency part and a high-frequency part. Hierarchical threshold processing is performed on the decomposed high-frequency coefficients. The decomposed high-frequency coefficients (layers 1-5) are analyzed to determine that the noise is mainly concentrated in the high-frequency coefficients of layers 1-2 (usually corresponding to high-frequency noise such as electromagnetic interference and white noise), while the high-frequency coefficients of layers 3-5 may contain effective partial discharge characteristics (such as high-frequency components of discharge pulses). By setting a threshold, the high-frequency coefficients of layers 1-2 are suppressed (such as setting to zero or shrinking), and the high-frequency coefficients of layers 3-5 are retained, thereby retaining the key characteristics of the partial discharge signal while removing noise. Specifically, the high-frequency coefficients of the third to fifth layers are retained. These high-frequency coefficients usually contain effective characteristics of partial discharge and are thus retained; the high-frequency noise components of the first to second layers are suppressed. These high-frequency coefficients usually mainly contain noise and are suppressed through threshold processing to remove high-frequency noise; wavelet reconstruction is performed based on the processed coefficients (the retained effective high-frequency coefficients and low-frequency coefficients) to obtain the denoised partial discharge signal. The partial discharge detection methods in the related art (such as earth current and ultrasonic waves) are vulnerable to noise such as environmental electromagnetic interference and mechanical vibration, resulting in a low signal-to-noise ratio and difficulty in accurately extracting partial discharge characteristics. In this embodiment, through wavelet hierarchical denoising, high-frequency noise (such as environmental interference in layers 1-2) is specifically suppressed, significantly improving the signal purity; by hierarchically retaining the high-frequency coefficients of layers 3-5, the loss of effective partial discharge characteristics is avoided, ensuring the accuracy of subsequent analysis; in this embodiment, through the parameter selection of the wavelet basis (db4) and 5-layer decomposition, combined with the hierarchical threshold strategy, it can flexibly adapt to the complex noise scenarios in the switchgear cabinet (such as changes in interference intensity under different working conditions). In this embodiment, through wavelet decomposition and threshold processing, high-frequency noise is effectively removed, the effective characteristics of partial discharge are retained, and the anti-interference ability of the signal is significantly improved. Hierarchical threshold processing can effectively distinguish partial discharge characteristics and noise, retain the high-frequency coefficients of the third to fifth layers as effective partial discharge characteristics, improve the recognizability of the partial discharge signal, and the denoised partial discharge signal is clearer and can more accurately reflect the true situation of partial discharge, thereby improving the accuracy of partial discharge detection; wavelet transform decomposes and processes the frequency components of the signal, and can adapt to noise environments of different frequencies and intensities, enabling the system to still operate stably under complex working conditions.

[0051] In practical applications, the db4 wavelet basis is used to perform five-layer wavelet decomposition on the high-frequency pulse current signal, obtaining a set of low-frequency approximation coefficients (A5) and five sets of high-frequency detail coefficients (D1 to D5). Hierarchical threshold processing is performed on a set of high-frequency coefficients in the target decomposed signal to remove high-frequency noise. Specifically, the high-frequency coefficients of the third to fifth layers are retained as effective partial discharge characteristics, while the high-frequency noise components of the first to second layers are suppressed. Among the processed coefficients, A5 is the low-frequency approximation coefficient without any modification, D3, D4, and D5 can be the high-frequency detail coefficients without any modification or with slight adjustment, and D1 and D2 may be set to zero or close to zero after threshold processing (if soft threshold processing is used, these coefficients will be compressed rather than completely set to zero). Based on the processed coefficients, the next step is to reconstruct the original signal through inverse wavelet transform. The reconstruction process is roughly as follows: Prepare the coefficients to ensure that all required coefficients (A5, D3, D4, D5) are ready. For D1 and D2, if they are directly set to zero, all corresponding array elements are set to 0; if soft threshold processing is used, update according to the processed results. Perform the inverse wavelet transform, using the same db4 wavelet basis function, combining the prepared coefficients (including A5, the processed D3, D4, D5, and the suppressed D1, D2), and reconstruct the signal through the inverse wavelet transform algorithm.

[0052] In an optional embodiment, before preprocessing the partial discharge signal using the target neural network model to obtain multiple frames of preprocessed signals, the above method further includes: obtaining a historical sample data set, which includes partial discharge signal sample data of different discharge types and corresponding actual sample results, where the actual sample results are used to represent the results of the actual discharge types corresponding to the partial discharge signal sample data; training the original neural network model using the historical sample data set until the loss value between the predicted sample results output by the original neural network model and the pre-determined actual sample results meets the preset convergence condition, ending the training, and determining the original neural network model at the end of the training as the target neural network model.

[0053] In the above embodiment, obtaining a historical sample data set containing partial discharge signal sample data of different discharge types and corresponding actual sample results and training the original neural network model can enable the model to learn the characteristic laws of different discharge types; when the loss value between the predicted sample results and the actual sample results meets the convergence condition during training and the target neural network model is determined, it can ensure that the target neural network model has high accuracy and reliability, and thus be more accurate in subsequent processing and predicting the discharge type of the partial discharge signal, improving the accuracy and timeliness of the partial discharge warning of the switchgear and solving the problems of the existing system lacking the analysis and utilization of historical data and inaccurate and untimely fault warning.

[0054] In this embodiment, the original neural network model is trained through supervised learning using a historical sample dataset to optimize the model parameters so that it can accurately predict the partial discharge type. The specific steps are as follows: Obtain the historical sample dataset, which contains the partial discharge signal sample data of different discharge types and their corresponding actual sample results (i.e., the true discharge type); Model training, use the historical sample dataset to train the original neural network model. During the training process, the model will output a predicted sample result based on the input sample data and calculate the loss value (such as cross-entropy loss) between the predicted result and the actual sample result; Optimize the model parameters, adjust the model parameters through an optimization algorithm (such as gradient descent) to minimize the loss value. The training process will continue until the loss value meets the preset convergence condition (such as the loss value no longer significantly decreases or reaches the set number of iterations); Determine the target neural network model. When the training is completed, the trained original neural network model is determined as the target neural network model for subsequent preprocessing and classification recognition of partial discharge signals. In this embodiment, the neural network model is trained through the historical sample dataset, and the model can learn the characteristics of different discharge types, thereby improving the prediction accuracy of the partial discharge type; Using sample data containing multiple discharge types for training enables the model to adapt to different types of partial discharge signals and enhances the generalization ability of the model; Through supervised learning and loss value evaluation, the model parameters are optimized to ensure that the model is continuously improved during the training process and the overall performance of the model is improved; The optimized neural network model can more accurately identify the partial discharge type, reduce the possibility of false alarms and missed alarms, and thus improve the reliability of the partial discharge warning system.

[0055] In an alternative embodiment, the high-voltage electrode and the housing of the target switchgear are respectively connected to the high-frequency pulse current sampling circuit. The high-frequency pulse current sampling circuit includes a coupling capacitor, a detection impedance, a calibration capacitor, a calibration pulse generator, and an AC power supply. Among them, the first end of the coupling capacitor is connected to the high-voltage electrode of the target switchgear, the second end of the coupling capacitor is connected to the ground terminal through the detection impedance, the housing of the target switchgear is connected to the ground terminal, the first end and the second end of the coupling capacitor are respectively connected to both ends of the AC power supply, the calibration capacitor and the calibration pulse generator are connected in series and then connected in parallel to both ends of the AC power supply, the HFCT is sleeved in the grounding loop, and the coupling capacitor, the detection impedance, and the target switchgear form a grounding loop. The coupling capacitor is used to capture the voltage signal generated by the partial discharge inside the target switchgear.

[0056] In the above embodiment, the high-voltage electrode and the outer shell of the target switchgear are respectively connected to the high-frequency pulse current sampling circuit. The coupling capacitor in this circuit can capture the voltage signal generated by partial discharge inside the target switchgear. Cooperating with components such as the detection impedance, calibration capacitor, calibration pulse generator, and AC power supply, the HFCT can be sleeved on the grounding loop composed of the coupling capacitor, detection impedance, and target switchgear to collect high-frequency pulse current signals, which is conducive to accurately monitoring and warning the partial discharge situation of the switchgear subsequently.

[0057] The high-frequency pulse current sampling circuit of this embodiment is used to collect the voltage signal generated by partial discharge inside the target switchgear. This sampling circuit includes key components such as a coupling capacitor, detection resistor, calibration capacitor, calibration pulse generator, and AC power supply. Through the collaborative action of these components, efficient collection and calibration of partial discharge signals are achieved. The coupling capacitor is connected in series between the high-voltage electrode and the grounding end, used to capture the transient voltage signal generated by partial discharge and convert it into a current signal that can be measured by the detection resistor. The HFCT is sleeved on the grounding loop to directly detect the partial discharge pulse current (introduced into the grounding loop by the coupling capacitor), avoiding direct contact with the high-voltage electrode; the calibration capacitor and the calibration pulse generator are connected in parallel across the AC power supply and can inject standard pulse signals for on-site calibration of the sensitivity and frequency response characteristics of the HFCT. When partial discharge occurs inside the target switchgear, transient voltage changes will be generated. These changes will be captured by the coupling capacitor and converted into current form, and then further processed by the detection impedance; the HFCT monitors the current fluctuations in the grounding loop caused by partial discharge and converts them into electrical signals that can be analyzed by subsequent electronic devices; the presence of the calibration capacitor and the calibration pulse generator ensures the accuracy and consistency of the system, allowing regular inspection and adjustment of the measurement accuracy.

[0058] In an alternative embodiment, the preset discharge types include at least one of the following: surface discharge type, internal discharge type, and corona discharge type; the output layer of the target neural network model includes three independent output nodes, corresponding to the probability of internal discharge type, the probability of surface discharge type, and the probability of corona discharge type respectively.

[0059] In the above embodiment, it is clear that the preset discharge types cover the surface discharge type, internal discharge type, and corona discharge type. When a set of prediction results meet specific conditions to trigger a partial discharge warning signal, it can more accurately conduct partial discharge warning for these common discharge types, improving the accuracy and pertinence of the partial discharge warning of the switchgear, thereby improving the maintenance efficiency of power equipment and reducing the equipment failure risk. The output layer of the target neural network model is set with three independent output nodes corresponding to the probability of internal discharge type, the probability of surface discharge type, and the probability of corona discharge type respectively, which can more clearly and specifically output the probabilities of different discharge types, helping to accurately judge the partial discharge situation of the switchgear and improving the accuracy of partial discharge warning.

[0060] The surface discharge type represents the discharge occurring along the surface of the insulator (such as the surface flashover caused by the pollution of the bus support insulator in the switchgear), also known as surface defect discharge; the internal discharge type represents the discharge caused by air gaps or defects inside the insulating medium (such as the internal bubble discharge of the GIS basin insulator), also known as internal cavity discharge; the corona discharge type represents the ionization caused by the local electric field concentration at the tip of the high-voltage electrode (such as the burr discharge of the high-voltage connection in the switchgear), also known as tip pulse discharge. The output layer sets three independent nodes, corresponding to the probability values of the three types of internal discharge, surface discharge, and corona discharge respectively (the value range is 0-1). The model learns the signal characteristics of different discharge types (such as waveform, frequency, energy distribution, etc.) in the historical samples, and establishes the mapping relationship between the input signal and the probability of each category; each output node converts the original value calculated by the model into a probability value through the Softmax activation function, reflecting the possibility that the input signal belongs to the corresponding discharge type. For example, if the probability of the internal discharge type is 0.8, it means that the model believes that the confidence of the signal belonging to the internal discharge is 80%; the system makes a comprehensive judgment based on the probability values output by each node. For example, when the probability of a certain type exceeds the preset threshold (such as 0.9), it is determined as the discharge of this type, and the corresponding early warning or diagnosis process is triggered.

[0061] Obviously, the above-described embodiments are only a part of the embodiments of the present application, rather than all embodiments. The following will specifically describe the present application in combination with specific embodiments.

[0062] The embodiment of the present application provides a method for early warning of partial discharge in a switchgear based on an artificial intelligence large model, which uses a new pulse current method under the D-CNN model relying on artificial intelligence algorithms to monitor the partial discharge of a distribution ring main switchgear, and will avoid the problems of weak anti-interference ability, low sensitivity, and low reliability of the ground wave / ultrasonic method and the ultra-high frequency method for 10kV distribution switchgears.

[0063] The on-line monitoring of partial discharge is mainly based on diagnostic data (spectrograms) rather than measurement data (numerical values). For diagnostic data, it is necessary to rely on big data and artificial intelligence technologies to train an automated fault early warning expert system.

[0064] With the rapid development of HFCT electronic devices, a high-frequency current transformer operating at 10 MHz is used to replace the existing method of capacitance coupling plus sampling resistors to capture high-frequency pulsed currents. Specifically, on the basis of the capacitance voltage coupling signal at the "lower outgoing line" of the high-voltage switchgear, a precision high-frequency pulsed current sensor is added. The collected current signal is decomposed by wavelet denoising in 5 layers to extract time-frequency features and statistical features (data cleaning, data analysis, iterative algorithms, data output), and then input into a pre-trained 1D-CNN model to output the probability of the discharge type and the mathematical quantization value of the discharge intensity. If the probability exceeds the threshold and persists for 3 frames, an early warning is triggered, which is displayed and output externally by the partial discharge early warning platform.

[0065] Wavelet denoising is a signal denoising method based on wavelet transform. By decomposing the signal into different frequency sub-bands and combining threshold processing to remove noise. db4 is one of the Daubechies wavelet series and has the following characteristics: Compact support: Limited length (filter length is 8), suitable for processing local mutation signals (such as partial discharge pulses); Vanishing moments: 4th-order vanishing moments, which can better represent the smooth part of the signal and suppress high-frequency noise; Orthogonality: There is no redundancy in the decomposed sub-bands, suitable for denoising and feature extraction.

[0066] For signals with a relatively high sampling rate (such as 10 MHz partial discharge data), 5-layer decomposition can effectively separate the main frequency components of noise and effective signals, as shown in Table 1.

[0067] Table 1 Decomposition layer number Frequency range (Hz) Main component Layer 1 5 MHz to 10 MHz High-frequency noise (random interference) Layer 2 2.5 MHz to 5 MHz High-frequency noise and partial discharge pulses Layer 3 1.25 MHz to 2.5 MHz Main energy of discharge pulses Layer 4 625 kHz to 1.25 MHz Low-frequency noise and base interference Layer 5 0 to 625 kHz Power frequency interference and baseline drift Signal decomposition Using the db4 wavelet basis to perform 5-layer wavelet decomposition on the signal, obtaining the high-frequency coefficients (detail coefficients, denoted as D1, D2,..., D5) and low-frequency coefficients (approximate coefficients, denoted as A5) of each layer.

[0068] Mathematical representation: The original signal x = A5 + D5 + D4 + D3 + D2 + D1.

[0069] Threshold processing Universal Threshold: λ = σ√(2lnN); σ: Standard deviation of noise (can be estimated by the coefficients of the highest frequency sub-band D1), N: Signal length.

[0070] Soft threshold / Hard threshold: Soft threshold: sign(x)(|x| - λ) (smoother, suitable for retaining signal details); Hard threshold: x·I(|x| > λ) (sharper, may introduce artifacts); Layer-by-layer processing: Apply threshold processing to the high-frequency coefficients D1 - D5 to remove the sub-bands dominated by noise. Usually, only D3 - D5 (the main frequency region of discharge pulses) are retained, and D1 - D2 (high-frequency noise) are suppressed; Signal reconstruction Use the coefficients A5, D5′, D4′, D3′ after threshold processing for wavelet reconstruction to obtain the denoised signal.

[0071] The denoising method using the db4 wavelet basis + 5-layer decomposition can effectively remove high-frequency noise and reduce power frequency interference through hierarchical threshold processing, while retaining the key features of partial discharge pulses. In practical applications, the decomposition layer number and threshold strategy need to be adjusted according to the signal characteristics of the three major types of switchgear cabinets (internal cavity discharge, surface defect discharge, and tip pulse discharge), and the denoising effect is verified by combining domain knowledge.

[0072] Adopt a 1D-CNN (one-dimensional convolutional neural network) specifically for processing one-dimensional sequence data (such as time series, audio signals, sensor data), which is a deep learning model. Its core is to extract local features through convolution operations and is suitable for time-domain or frequency-domain signal analysis in switchgear partial discharge early warning.

[0073] The structure and working principle of the partial discharge analysis model are as follows: Input layer input data: one-dimensional time series (such as the pulse current signal collected by HFCT sensors), such as 10MHz sampling rate pulse current); Preprocessing: Normalization: Normalize the signal to [-1, 1] or [0, 1]; Frame processing: Cut the long signal into fixed-length segments (such as 1000 points / frame); Training and optimization Convolutional layer Extract local features (such as pulse rising edge, spectral spike) through a sliding convolutional kernel; Parameters: Convolutional kernel size: Determines the window length of feature extraction; Number of filters: Determines the number of feature maps extracted; Output formula: Output length = (input length - kernel_size + 2×paddingstride) / stride + 1; kernel_size represents the convolutional kernel size, paddingstride represents the one-sided padding step, and stride represents the sliding step of the convolutional kernel; Activation function: Introduce the non-linear ReLU (Rectified Linear Unit) to enhance the model's expressive ability; Formula: ReLU(x) = max(0, x); Pooling Layer Function: Reduce data dimension, retain important features, and improve computational efficiency; Types: Max Pooling (MaxPool1d), Average Pooling (AvgPool1d); Parameters: Pooling window size (e.g., 2), stride (usually the same as the window size); Fully Connected Layer (Dense) Function: Map the features extracted by the convolutional layer to classification or regression results; Output Dimension: Set according to the task (3 neurons corresponding to three major types of discharge); Original signal collected by the sensor (such as pulsed current with a sampling rate of 10 MHz); Give corresponding warnings according to the output probability distribution; Input: Preprocessed pulsed current signal (length 1000 points); Output: Discharge type probability (surface discharge, internal discharge, corona discharge).

[0074] The performance indicators are shown in Table 2: Table 2 Category Precision Recall F1-Score Surface discharge 96％ 94％ 95％ Internal discharge 92％ 93％ 92.5％ Corona discharge 89％ 90％ 89.5％ The 1D-CNN large model efficiently extracts the key features of partial discharge signals through convolution operations, and combines fully connected layers to achieve high-precision classification, which is an ideal choice for partial discharge warning of switchgear. In practical applications, the model structure needs to be adjusted according to the data characteristics, and the robustness can be further improved through data augmentation and hybrid model design.

[0075] Figure 2 This is a schematic diagram of a high-frequency pulsed current sampling circuit provided by an embodiment of the present application. C5 is a coupling voltage sensor (corresponding to the aforementioned coupling capacitor), HFCT is a high-frequency pulsed current transformer, R is a detection impedance, C1 is the equivalent capacitance of the insulator between the high-voltage electrode inside the switchgear and "ground", C3 is the equivalent capacitance of the air bubbles, surface defects at the internal cavity of the switchgear, and C2 is the equivalent capacitance of the remaining insulators connected in series with it. Figure 2 The medium-voltage ring main unit is equivalent to one of the aforementioned switchgears.

[0076] The classical analysis model of partial discharge is represented by a three-capacitance model. When partial discharge occurs inside an electrical equipment, the charges accumulated at both ends of the equivalent capacitance (Cg) at the defect will be neutralized, resulting in an equivalent charge jump in the intact part of the insulation (Cb) in series therewith, and forming a high-frequency pulse current flowing from the defect to the outside of the electrical equipment. When this high-frequency pulse current flows through the installation location of the coupling voltage sensor, it will be shunted by a high-frequency low-impedance loop composed of the coupling voltage sensor and the detection impedance, and transformed into a high-frequency pulse voltage at both ends of the detection impedance. By capturing the high-frequency pulse voltage at both ends of the detection impedance, the monitoring equipment can directly infer the amount of migrating charges caused by partial discharge.

[0077] The detection principle of the pulse current method belongs to direct detection, with high sensitivity, strong applicability, wide monitoring range, and is especially suitable for partial discharge detection of 10kV box-type enclosed switchgear.

[0078] The equipment can be uniformly calibrated, facilitating quantitative analysis and other advantages.

[0079] Figure 3 This is the schematic diagram of the new partial discharge early warning for switchgear provided by the embodiments of this application. By collecting high-frequency pulse current signals through a partial discharge monitoring system, after data processing and storage, it is uploaded to a big data platform. Using an AI expert analysis system and machine learning model training for intelligent analysis, combined with the digital twin technology based on a large model, data visualization and early warning functions are finally realized on the fault early warning display platform, achieving intelligent monitoring and early warning of partial discharge faults. That is, after the on-line partial discharge monitoring system is connected to the digital twin system, with the help of the big data platform of the digital twin system, machine learning is carried out on the spectrograms, and the expert system is trained to realize automatic early warning. This system improves the accuracy and reliability of partial discharge monitoring of switchgear by integrating high-frequency pulse current sampling, data transmission, storage, processing, and AI analysis, effectively solving the problems existing in traditional monitoring methods.

[0080] The embodiments of this application have at least the following technical effects: Based on the traditional and extensive partial discharge monitoring principle of switchgear, it is improved and enhanced. High-frequency pulse current transformers are used to capture partial discharge signals, greatly improving the resistance to interference such as white noise and background noise. And the wavelet transform decomposition and convolutional model algorithm are introduced. Through learning the partial discharge model spectrograms under various working conditions, the success rate of partial discharge monitoring and early warning of distribution switchgear is greatly improved, and quantitative analysis can be carried out. By horizontally comparing and analyzing the partial discharge characteristics at different positions, the specific location of partial discharge can be located. For the partial discharge between the internal insulation material and the high-voltage electrode of the switchgear (including three categories: tip burr discharge, internal bubble cavity discharge, and surface defect creepage), more accurate judgment can be made, greatly realizing the digital application of switchgear condition monitoring.

[0081] The present application also provides a warning device for partial discharge of a switchgear cabinet, as Figure 4 shown, Figure 4 FIG. is a structural block diagram of a warning device for partial discharge of a switchgear cabinet provided by an embodiment of the present application. The device includes: An acquisition module 41, configured to acquire a high-frequency pulse current signal through a high-frequency pulse current transformer HFCT, where the high-frequency pulse current signal includes a partial discharge signal generated when a target switchgear cabinet has a partial discharge; A noise reduction module 42, configured to perform noise reduction processing on the high-frequency pulse current signal to obtain a partial discharge signal; A processing module 43, configured to preprocess the partial discharge signal by using a target neural network model to obtain multiple frames of preprocessed signals, extract features from each frame of the preprocessed signals in the multiple frames of preprocessed signals to obtain multiple local features, and respectively predict the discharge type and the probability value of the discharge type corresponding to each local feature to obtain a set of prediction results, where the target neural network model is trained by using a historical sample data set, and the historical sample data set includes local discharge signal sample data of different discharge types, and each prediction result includes the discharge type and the probability value of a corresponding local feature; A trigger module 44, configured to trigger a partial discharge warning signal when a set of prediction results meets a preset condition.

[0082] It should be noted that when the system provided in the above embodiment implements its functions, only the above division of each functional module is used as an example for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be elaborated here.

[0083] The present application also provides a computer-readable storage medium, in which instructions are stored, and when the instructions are executed, the method steps described in any one of the above are executed.

[0084] In an exemplary embodiment, the above computer-readable storage medium may include, but is not limited to: a USB flash drive, a read-only memory (ROM for short), a random access memory (RAM for short), a mobile hard disk, a magnetic disk, or an optical disc, etc., various media that can store computer programs.

[0085] The present application also discloses an electronic device. As Figure 5 shown, Figure 5It is a schematic structural diagram of an electronic device disclosed in an embodiment of the present application. The electronic device 500 may include: at least one processor 501, at least one communication bus 502, a user interface 503, at least one network interface 504, and a memory 505.

[0086] Among them, the communication bus 502 is used to realize the connection and communication between these components.

[0087] Among them, the user interface 503 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 503 may further include a standard wired interface and a wireless interface.

[0088] Among them, the network interface 504 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0089] Among them, the processor 501 may include one or more processing cores. The processor 501 connects various parts within the entire electronic device (such as a server) through various interfaces and lines. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 505, and by calling the data stored in the memory 505, it executes various functions of the server and processes data. Optionally, the processor 501 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 501 may integrate one or several combinations of a central processing unit (CPU), a graphics processing unit (GPU), and a modem, etc. Among them, the CPU mainly processes the operating system, user interface, and application programs, etc.; the GPU is responsible for the rendering and drawing of the content to be displayed on the display screen; the modem is used to process wireless communication. It can be understood that the above modem may not be integrated into the processor 501 and may be implemented separately by a single chip.

[0090] Among them, the memory 505 may include a Random Access Memory (RAM), or may also include a Read-Only Memory. Optionally, the memory 505 includes a non-transitory computer-readable storage medium. The memory 505 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 505 may include a program storage area and a data storage area. Among them, the program storage area can store instructions for implementing the operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-mentioned method embodiments, etc.; the data storage area can store the data involved in the above-mentioned method embodiments. Optionally, the memory 505 may also be at least one storage device located far from the aforementioned processor 501. Refer to Figure 5 , in the memory 505 as a computer storage medium, it may include an operating system, a network communication module, a user interface module, and an application program for a method of warning against partial discharge in a switchgear cabinet.

[0091] In Figure 5 In the electronic device 500 shown, the user interface 503 is mainly used to provide an input interface for the user and obtain the data input by the user; while the processor 501 can be used to call an application program for a method of warning against partial discharge in a switchgear cabinet stored in the memory 505. When executed by one or more processors 501, the electronic device 500 is caused to execute one or more of the methods as described in the above embodiments. It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present application is not limited by the described action sequence, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present application.

[0092] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0093] In several embodiments provided by the present application, it should be understood that the disclosed device or system can be implemented in other ways. For example, the device or system embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some service interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical or other form.

[0094] The foregoing are only exemplary embodiments of the present disclosure and should not be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure still fall within the scope covered by the present disclosure. Those skilled in the art will readily think of other implementation schemes of the present disclosure after considering the disclosure of the specification.

[0095] This application aims to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common general knowledge or conventional technical means in the technical field not recorded in the present disclosure.

Claims

1. A warning method for partial discharge in a switchgear, characterized in that, Including: Collecting high-frequency pulse current signals through a high-frequency pulse current transformer HFCT, where the high-frequency pulse current signals include partial discharge signals generated when a target switchgear cabinet has a partial discharge; Performing noise reduction processing on the high-frequency pulse current signals to obtain the partial discharge signals; Using a target neural network model to preprocess the partial discharge signals to obtain multiple frames of preprocessed signals, respectively extracting features from each frame of the preprocessed signals among the multiple frames of preprocessed signals to obtain multiple local features, and respectively predicting the discharge type and the probability value of the discharge type corresponding to each local feature to obtain a set of prediction results, where the target neural network model is trained using a historical sample data set, the historical sample data set includes partial discharge signal sample data of different discharge types, and each prediction result includes the discharge type and the probability value corresponding to a corresponding local feature; Triggering a partial discharge warning signal when the set of prediction results meets a preset condition.

2. The method according to claim 1, wherein Triggering a partial discharge warning signal when the set of prediction results meets a preset condition, including: Triggering the partial discharge warning signal when the set of prediction results indicates that the discharge types corresponding to preprocessed signals of a continuous preset number of frames in the multiple frames of preprocessed signals are preset discharge types and the probability values of the preset discharge types are all greater than or equal to a preset probability threshold.

3. The method according to claim 1, wherein Respectively extracting features from each frame of the preprocessed signals among the multiple frames of preprocessed signals to obtain multiple local features, and respectively predicting the discharge type and the probability value of the discharge type corresponding to each local feature to obtain a set of prediction results, including: The target frame preprocessed signal is any one of the multiple frames of preprocessed signals, and the target prediction result is obtained by processing the target frame preprocessed signal in the following manner: Using the target neural network model to extract features from the target frame preprocessed signal to obtain a target local feature, and performing classification and recognition based on the target local feature to obtain the target prediction result, where the target prediction result includes a target discharge type and the probability value of the target discharge type, the target discharge type is used to represent the discharge type corresponding to the target local feature predicted by the target neural network model, and the probability value of the target discharge type is used to represent the probability that the target neural network model predicts that the discharge type of the target local feature is the target discharge type.

4. The method according to claim 3, characterized in that, Using the target neural network model to extract features from the target frame preprocessed signal to obtain a target local feature, and performing classification and recognition based on the target local feature to obtain the target prediction result, including: The target neural network model includes an input layer, a convolutional layer, an activation function layer, a pooling layer, and a fully connected layer. The input layer receives the partial discharge signal and preprocesses the partial discharge signal to obtain multiple frames of preprocessed signals. The convolutional layer extracts the target local features through a sliding convolutional kernel. The activation function layer introduces non-linearity. The pooling layer reduces the data dimension. The fully connected layer maps the extracted features to a classification result to obtain the target prediction result.

5. The method according to claim 1, wherein Preprocessing the partial discharge signal using the target neural network model to obtain multiple frames of preprocessed signals, including: Normalizing the partial discharge signal using the target neural network model to obtain a normalized discharge signal, where the high-frequency pulse current signal is a one-dimensional time series signal; Framing the normalized discharge signal to obtain the multiple frames of preprocessed signals, where each frame of the preprocessed signals in the multiple frames of preprocessed signals is a signal of a preset length.

6. The method according to claim 1, characterized in that, Denosing the high-frequency pulse current signal to obtain the partial discharge signal, including: Performing 5-layer wavelet decomposition on the high-frequency pulse current signal using the db4 wavelet basis to obtain a target decomposed signal; Performing hierarchical threshold processing on a group of high-frequency coefficients in the target decomposed signal to remove high-frequency noise, obtaining processed coefficients, where the third to fifth layer high-frequency coefficients are retained as effective partial discharge features in the processed coefficients and the first to second layer high-frequency noise components are suppressed; Performing wavelet reconstruction based on the processed coefficients to obtain the denoised partial discharge signal.

7. The method according to claim 1, wherein The high-voltage electrode and the outer shell of the target switchgear are respectively connected to a high-frequency pulse current sampling circuit. The high-frequency pulse current sampling circuit includes a coupling capacitor, a detection impedance, a calibration capacitor, a calibration pulse generator, and an AC power supply. Among them, the first end of the coupling capacitor is connected to the high-voltage electrode of the target switchgear, the second end of the coupling capacitor is connected to the ground terminal through the detection impedance, the outer shell of the target switchgear is connected to the ground terminal, the first end and the second end of the coupling capacitor are respectively connected to both ends of the AC power supply, the calibration capacitor and the calibration pulse generator are connected in series and then connected in parallel to both ends of the AC power supply, the HFCT is sleeved in the grounding loop, the coupling capacitor, the detection impedance, and the target switchgear form the grounding loop, and the coupling capacitor is used to capture the voltage signal generated by partial discharge inside the target switchgear.

8. An early warning device for partial discharge of a switchgear, characterized in that, Including: An acquisition module for acquiring a high-frequency pulse current signal through a high-frequency pulse current transformer HFCT, where the high-frequency pulse current signal includes a partial discharge signal generated when partial discharge occurs in the target switchgear; A denoising module for denoising the high-frequency pulse current signal to obtain the partial discharge signal; A processing module, configured to preprocess the partial discharge signal by using a target neural network model to obtain multiple frames of preprocessed signals, extract features from each frame of the preprocessed signals among the multiple frames of preprocessed signals to obtain multiple local features, and respectively predict the discharge type and the probability value of the discharge type corresponding to each local feature to obtain a set of prediction results, wherein the target neural network model is trained by using a historical sample data set, the historical sample data set includes partial discharge signal sample data of different discharge types, and each prediction result includes the discharge type and the probability value of a corresponding local feature; A triggering module, configured to trigger a partial discharge warning signal when the set of prediction results meets a preset condition.

9. An electronic device, comprising a memory and a processor, wherein a computer program is stored on the memory, characterized in that, When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions, and when the instructions are executed, the method according to any one of claims 1 to 7 is executed.

Images