Weak discharge signal identification method and related equipment

Abstract

The invention discloses a weak discharge signal identification method and related equipment. Intelligent identification of weak discharge signals in a complex environment is realized through cooperation of multiple technologies. The broadband sensor is installed at a key node of a distribution network line, the problems that a traditional sensor is prone to being affected with damp and narrow in signal receiving range are solved, and weak signals can be captured. A three-stage cascade low-noise amplification circuit and a robust adaptive filtering algorithm are adopted, gain parameters are adjusted in real time based on an H-infinity control feedback mechanism, various noise interferences are effectively inhibited, and the defects of a traditional fixed threshold filtering algorithm are overcome. And inputting the adjusted signal into a detection model, forming a local and global feature combined capture capability by means of a feature pyramid network, an SE module, a bidirectional long-short term memory network and an attention mechanism, and then realizing fine-grained extraction and accurate classification of the signal by using a Softmax classifier in combination with a dynamic threshold strategy, thereby meeting the on-line monitoring requirements of the insulation state of the electrical equipment.

Description

Technical Field

[0001] The present application relates to the field of electrical detection technology, and more specifically, to a weak discharge signal recognition method and related equipment. Background Art

[0002] In the field of insulation condition monitoring for electrical equipment, discharge signal detection is a key method for determining the health of the equipment. However, in actual detection, discharge signals are often subject to multiple interferences such as pulse-type, narrowband, and broadband noise, making signal extraction difficult and detection accuracy low.

[0003] Current technologies primarily rely on traditional sensors for signal acquisition, which present significant limitations in complex environments such as high humidity and strong electromagnetic interference. First, traditional sensors are made of limited materials and have limited frequency coverage, making them susceptible to moisture failure in high-humidity environments and unable to effectively capture weak discharge signals within the 10kHz-1GHz frequency band. Second, signal processing often relies on fixed thresholds or simple filtering algorithms, making it difficult to distinguish between normal operating signals and abnormal discharge signals. This is especially true in strong electromagnetic environments such as industrial areas or substations, where noise can easily drown out weak discharge signals. Third, existing equipment lacks environmental adaptation mechanisms and is unable to adjust detection parameters in real time based on changes in humidity, electromagnetic interference, and other factors, resulting in a significant decrease in detection stability. Furthermore, traditional frequency domain analysis methods based on Fourier transforms lack real-time performance, making it difficult to meet the needs of online monitoring of weak discharge signals.

[0004] In view of the above problems, there is an urgent need to develop a solution that can realize weak discharge signal recognition in complex environments to improve the reliability of insulation status monitoring of electrical equipment. Summary of the Invention

[0005] The present application provides a weak discharge signal recognition method and related equipment, which uses wide-band sensor acquisition, a three-stage cascaded low-noise amplification circuit and a robust adaptive filtering algorithm combined with an H∞ control feedback mechanism to suppress noise, and a weak discharge signal detection model based on a feature pyramid network and a bidirectional long short-term memory network to perform feature extraction and classification, thereby solving the problem of low recognition accuracy of traditional technologies in high-humidity and strong electromagnetic interference scenarios, and meeting the needs of online monitoring of the insulation status of electrical equipment.

[0006] A method for identifying a weak discharge signal, comprising:

[0007] Obtain the original discharge signal captured by broadband sensors at key nodes of the distribution network;

[0008] The original discharge signal is subjected to gain processing and noise suppression processing in sequence through a three-stage cascaded low-noise amplifier circuit and a robust adaptive filtering algorithm to obtain an adjusted discharge signal, wherein the gain parameter is adjusted in real time based on a feedback mechanism constructed by H∞ control, and the robust adaptive filtering algorithm uses a Huber function, M estimation, or entropy-related measurement as an objective function;

[0009] The adjusted discharge signal is input into the weak discharge signal detection model, and the multi-level feature maps are fused using the feature pyramid network. The key channel features are enhanced through the SE module. The temporal features are extracted by combining the bidirectional long short-term memory network and the attention mechanism. The abnormal discharge signal is determined based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy.

[0010] Optionally, the process of performing gain processing on the original discharge signal by the three-stage cascaded low-noise amplifier circuit includes:

[0011] The first-stage amplifier uses GaAs HEMT transistors to construct a common-source input matching network to minimize the noise figure;

[0012] The second-stage amplifier achieves signal gain improvement through the inter-stage matching network;

[0013] The third stage amplifier performs output impedance matching and provides power gain.

[0014] Optionally, the total noise coefficient of the three-stage cascaded low-noise amplifier circuit satisfies:

[0015]

[0016] in, is the total noise figure, 、 、 are the noise coefficients of the first-stage amplifier, the second-stage amplifier, and the third-stage amplifier, respectively. 、 are the gains of the first-stage amplifier and the second-stage amplifier respectively.

[0017] Optionally, when the robust adaptive filtering algorithm adopts the Huber function as the objective function, the objective function formula is:

[0018]

[0019] in, is the objective function, is the filtered error signal, The dynamic threshold is set based on the standard deviation of the ambient noise.

[0020] Optionally, the abnormal discharge signal is determined based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy, including:

[0021] Calculating an anomaly score based on the anomaly probability output by the Softmax classifier;

[0022] Performing temperature scaling calibration on the anomaly probability, and setting a dynamic threshold according to a quantile of the calibrated probability;

[0023] When the abnormality score exceeds the set dynamic threshold, it is determined to be an abnormal discharge signal.

[0024] Optionally, the H∞ control is implemented by minimizing the H∞ norm of the system transfer function, where the norm is defined as:

[0025]

[0026] in, is the norm, is the closed-loop transfer coefficient from the disturbance input w to the control output z, is the supremum for all frequencies, is the largest singular value of the matrix, and w is the angular frequency.

[0027] Optionally, the method further includes performing temperature scaling calibration on the abnormal probability output by the Softmax classifier, wherein the temperature scaling calibration formula is:

[0028]

[0029] in, is the temperature scaling value of the Softmax classifier class i, is the original logical value of the ith neuron of the Softmax classifier, T is the temperature parameter, and K is the total number of categories.

[0030] A weak discharge signal recognition device, comprising:

[0031] The original discharge capture unit is used to obtain the original discharge signal captured at the key node of the distribution network line using a broadband sensor;

[0032] a gain suppression processing unit, configured to sequentially perform gain processing and noise suppression processing on the original discharge signal through a three-stage cascaded low-noise amplifier circuit and a robust adaptive filtering algorithm to obtain an adjusted discharge signal, wherein the gain parameter is adjusted in real time based on a feedback mechanism constructed by H∞ control, and the robust adaptive filtering algorithm uses a Huber function, an M estimate, or an entropy-related metric as an objective function;

[0033] The signal model detection unit is used to input the adjusted discharge signal into the weak discharge signal detection model, use the feature pyramid network to fuse multi-level feature maps, enhance the key channel features through the SE module, combine the bidirectional long short-term memory network and the attention mechanism to extract the timing features, and determine the abnormal discharge signal based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy.

[0034] A weak discharge signal recognition device, comprising a memory and a processor;

[0035] The memory is used to store programs;

[0036] The processor is configured to execute the program to implement each step of the weak discharge signal recognition method as described in any one of the above items.

[0037] A readable storage medium stores a computer program thereon, wherein when the computer program is executed by a processor, each step of the weak discharge signal recognition method as described in any one of the above items is implemented.

[0038] It can be seen from the above technical solutions that the embodiments of the present application provide a method and related equipment for identifying weak discharge signals, which realize intelligent identification of weak discharge signals in complex environments through technologies such as wide-band sensor acquisition, multi-level noise suppression, adaptive feedback control, and deep learning feature recognition. Specifically, wide-band sensors are installed at key nodes of the distribution network line to solve the problems of traditional sensors being easily affected by moisture and having a narrow signal reception range in high-humidity environments, and can capture weak signals. A three-stage cascaded low-noise amplification circuit is combined with a robust adaptive filtering algorithm with Huber function, M estimation or entropy-related measurement as the objective function, and the gain parameters are adjusted in real time based on the H∞ control feedback mechanism to effectively suppress pulse-type, narrowband and other noise interference, overcoming the defect that the traditional fixed threshold filtering algorithm cannot distinguish between normal operation signals and abnormal discharge signals. The adjusted discharge signal is input into the weak discharge signal detection model, and the multi-level feature maps are fused with the help of the feature pyramid network. The key channel features are enhanced through the SE module, and the temporal features are extracted by combining the bidirectional long short-term memory network and the attention mechanism to form a joint capture capability of local fine-grained features and global semantics. The Softmax classifier is then used to output the abnormality probability and combined with the dynamic threshold adjustment strategy to achieve fine-grained feature extraction and accurate classification of the discharge signal, solving the problem of low signal recognition accuracy and easy misjudgment due to the lack of environmental adaptive mechanism in the existing technology. Through the collaboration of multiple technologies, this solution significantly improves the sensitivity, anti-interference ability and recognition accuracy of weak discharge signal detection in high humidity and strong electromagnetic interference scenarios, meeting the needs of online monitoring of the insulation status of electrical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0040] Figure 1 This is a flow chart of a method for identifying a weak discharge signal disclosed in an embodiment of the present application;

[0041] Figure 2 A schematic diagram of a weak discharge signal recognition device disclosed in an embodiment of the present application;

[0042] Figure 3 This is a hardware structure block diagram of a weak discharge signal recognition device disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0044] The present application can be used in a variety of general or special computing device environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multi-processor devices, and distributed computing environments including any of the above devices or devices.

[0045] Next, we will introduce the application scheme. This application proposes the following technical scheme, please see below for details.

[0046] Figure 1 This is a flow chart of a method for identifying a weak discharge signal disclosed in an embodiment of the present application.

[0047] like Figure 1 As shown, the method may include:

[0048] Step S1: Acquire an original discharge signal captured at a key node of a distribution network line using a broadband sensor.

[0049] Specifically, during the operation of distribution network lines, in order to effectively monitor potential weak discharge phenomena, wide-band sensors need to be precisely installed at key node locations of the lines. These key nodes usually include but are not limited to transformer outlets, cable joints, insulator strings, and other locations prone to partial discharge. Compared with traditional sensors, wide-band sensors have a wider signal reception frequency band range, which can effectively overcome the technical defects of being susceptible to moisture in high-humidity environments and having a narrow signal reception range. Through its built-in sensing element, it can sense the weak electromagnetic signals generated by partial discharge in the distribution network lines in real time and convert them into electrical signals for output. This output electrical signal is the original discharge signal, which contains the original characteristic information of the discharge phenomenon. At the same time, it is inevitably mixed with various noise signals in the environment and will serve as the basic data for subsequent processing.

[0050] Step S2: performing gain processing and noise suppression processing on the original discharge signal in sequence through a three-stage cascaded low-noise amplifier circuit and a robust adaptive filtering algorithm to obtain an adjusted discharge signal, wherein the gain parameter is adjusted in real time based on a feedback mechanism constructed by H∞ control, and the robust adaptive filtering algorithm uses a Huber function, M estimation, or entropy-related measurement as an objective function.

[0051] Specifically, the original discharge signal first enters a three-stage cascaded low-noise amplifier circuit. This circuit consists of three stages of amplifier units connected in series. Each stage of the amplifier unit uses a low-noise amplifier device. The purpose is to gradually amplify the original discharge signal while minimizing the introduction of its own noise, thereby improving the signal strength for subsequent processing. The gain parameters of each stage of the amplifier circuit are not fixed, but are adjusted in real time based on the feedback mechanism constructed based on H∞ control. H∞ control is a robust control strategy that constructs a feedback loop to monitor the characteristics of the output signal in real time. It dynamically adjusts the gain of each stage of the amplifier circuit according to the changes in the signal to ensure the best signal amplification effect under different input signal strengths and noise environments.

[0052] After amplification, the signal enters the robust adaptive filtering phase. This filtering algorithm uses the Huber function, M-estimation, or entropy-related metrics as its objective function. These objective functions are robust and can effectively address pulse-shaped, narrowband, and other noise interference in the signal. The algorithm continuously adjusts the filter parameters to minimize the error between the output signal and the objective function, thereby effectively suppressing noise. During this process, the robust adaptive filtering algorithm works in conjunction with a three-stage cascaded low-noise amplifier circuit to amplify and enhance the signal while suppressing and eliminating noise. The final output is an adjusted discharge signal with a high signal-to-noise ratio, laying the foundation for subsequent signal recognition.

[0053] The Huber function is a function that combines the advantages of the least squares method and the least absolute deviation method. When the data error is small, it is similar to the least squares method, can converge quickly and improve the estimation accuracy; when outliers (such as pulse noise) appear in the data, the penalty intensity of the Huber function will become relatively mild, avoiding the problem of excessive estimation deviation caused by the influence of outliers in the least squares method, thereby ensuring the robustness of the algorithm in complex noise environments. For example, in the monitoring of distribution network lines, strong electromagnetic interference pulses occasionally appear. The Huber function can prevent the filtering algorithm from being overly affected by these interferences and maintain accurate processing of the real discharge signal. When the robust adaptive filtering algorithm uses the Huber function as the objective function, the objective function formula is:

[0054]

[0055] in, is the objective function, is the filtered error signal, The dynamic threshold is set based on the standard deviation of the ambient noise.

[0056] M-estimation is a generalized maximum likelihood estimation method that, by introducing an appropriate weighting function, assigns less weight to data far from the center, thereby reducing the impact of outliers on parameter estimation. It offers significant flexibility, allowing weighting functions to be selected or designed based on different noise characteristics. When faced with narrowband noise, which is common in distribution lines, a properly designed M-estimation weighting function can significantly suppress interference from narrowband noise while preserving the effective characteristics of the discharge signal, ensuring stable signal processing.

[0057] Entropy-related metrics are based on information entropy theory and use signal uncertainty as a metric. During the filtering process, the goal is to minimize the information entropy of the output signal, making it as stable as possible and containing less noise. When complex and variable noise is present in the original discharge signal, filtering algorithms based on entropy-related metrics can dynamically adjust filter parameters. By reducing signal uncertainty, they can separate the mixed noise from the actual discharge signal, thereby improving the signal-to-noise ratio and providing high-quality data for subsequent signal recognition.

[0058] These three objective functions each have their own advantages. By applying them in the robust adaptive filtering algorithm, they can effectively deal with various types of noise interference such as pulse-type and narrowband noise in the distribution network lines, and significantly improve the processing effect of the original discharge signal.

[0059] The H∞ control is achieved by minimizing the H∞ norm of the system transfer function, which is defined as:

[0060]

[0061] in, is the norm, is the closed-loop transfer coefficient from the disturbance input w to the control output z, is the supremum for all frequencies, is the largest singular value of the matrix, and w is the angular frequency.

[0062] Step S3: Input the adjusted discharge signal into the weak discharge signal detection model, use the feature pyramid network to fuse multi-level feature maps, enhance the key channel features through the SE module, combine the bidirectional long short-term memory network and the attention mechanism to extract the temporal features, and determine the abnormal discharge signal based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy.

[0063] Specifically, the adjusted discharge signal obtained after preliminary processing is fed into the weak discharge signal detection model. This model first uses a feature pyramid network (FPN) to extract and fuse features from the input signal. The FPN constructs feature maps at different levels through both bottom-up and top-down approaches. Each level contains signal feature information at different scales. Through lateral connections, these multiple levels of feature maps are fused, enabling the model to capture multi-scale feature information, from local details to global structures.

[0064] Next, the SE (Squeeze-and-Excitation) module is introduced to process the fused feature map. The SE module compresses and excites the channel dimensions of the feature map, learning the interdependencies between different channels. This enhances the response to key channel features and suppresses irrelevant channel information, thereby highlighting important features related to weak discharge signals.

[0065] Subsequently, a bidirectional long short-term memory (BiLSTM) network combined with an attention mechanism is used to extract temporal features from the enhanced features. BiLSTM models the signal's time series from both the forward and reverse directions, fully capturing the temporal dependencies of signals and effectively addressing the long-distance dependency issues inherent in traditional recurrent neural networks. The attention mechanism dynamically assigns weights based on the importance of features at different time steps, focusing on temporal features that are more critical for identifying weak discharge signals.

[0066] Finally, the extracted features are input into a Softmax classifier, which outputs probabilities of the signal belonging to different categories, including the probability of abnormal discharge signals. Combined with a dynamic threshold adjustment strategy, the threshold for determining abnormalities is dynamically set based on different monitoring environments and historical data. When the abnormal probability output by the Softmax classifier exceeds the dynamic threshold, an abnormal discharge signal is determined, thus achieving accurate identification and judgment of weak discharge signals.

[0067] It can be seen from the above technical solutions that the embodiments of the present application provide a method and related equipment for identifying weak discharge signals, which realize intelligent identification of weak discharge signals in complex environments through technologies such as wide-band sensor acquisition, multi-level noise suppression, adaptive feedback control, and deep learning feature recognition. Specifically, wide-band sensors are installed at key nodes of the distribution network line to solve the problems of traditional sensors being easily affected by moisture and having a narrow signal reception range in high-humidity environments, and can capture weak signals. A three-stage cascaded low-noise amplification circuit is combined with a robust adaptive filtering algorithm with Huber function, M estimation or entropy-related measurement as the objective function, and the gain parameters are adjusted in real time based on the H∞ control feedback mechanism to effectively suppress pulse-type, narrowband and other noise interference, overcoming the defect that the traditional fixed threshold filtering algorithm cannot distinguish between normal operation signals and abnormal discharge signals. The adjusted discharge signal is input into the weak discharge signal detection model, and the multi-level feature maps are fused with the help of the feature pyramid network. The key channel features are enhanced through the SE module, and the temporal features are extracted by combining the bidirectional long short-term memory network and the attention mechanism to form a joint capture capability of local fine-grained features and global semantics. The Softmax classifier is then used to output the abnormality probability and combined with the dynamic threshold adjustment strategy to achieve fine-grained feature extraction and accurate classification of the discharge signal, solving the problem of low signal recognition accuracy and easy misjudgment due to the lack of environmental adaptive mechanism in the existing technology. Through the collaboration of multiple technologies, this solution significantly improves the sensitivity, anti-interference ability and recognition accuracy of weak discharge signal detection in high humidity and strong electromagnetic interference scenarios, meeting the needs of online monitoring of the insulation status of electrical equipment.

[0068] In some embodiments of the present application, the process of performing gain processing on the original discharge signal by the three-stage cascaded low-noise amplifier circuit in step S2 is introduced, which may specifically include:

[0069] Step S21: The first-stage amplifier uses GaAs HEMT transistors to construct a common-source input matching network to minimize the noise coefficient;

[0070] Step S22: The second-stage amplifier improves the signal gain through the inter-stage matching network;

[0071] Step S23: The third-stage amplifier performs output impedance matching and provides power gain.

[0072] Specifically, the first-stage amplifier uses GaAs HEMT transistors to construct a common-source input matching network to minimize noise figure. As the first stage of the amplifier, the original discharge signal enters the amplification circuit, and its performance plays a decisive role in the noise level and signal reception capability of the entire amplifier circuit. GaAs HEMTs (gallium arsenide high electron mobility transistors) were chosen as the core device due to their low noise, high electron mobility, and high cutoff frequency, effectively suppressing self-generated noise when amplifying weak signals. The input matching network uses a common-source structure. Through precise design of the gate and source bias circuits and matching inductor and capacitor parameters, the amplifier's input impedance matches the output impedance of the wideband sensor, minimizing signal reflection loss. Furthermore, the low noise characteristics of GaAs HEMT transistors are leveraged to minimize the noise figure of the input signal, laying a high-quality foundation for subsequent signal amplification. This ensures that the weak original discharge signal enters the amplifier circuit with minimal noise interference and fully preserves its signal characteristics.

[0073] The second-stage amplifier achieves signal gain improvement through the inter-stage matching network. Specifically, although the signal processed by the first-stage amplifier has a low noise level, the signal strength is still insufficient to meet the needs of subsequent processing. As the core link for improving signal gain, the second-stage amplifier further amplifies the signal through a carefully designed inter-stage matching network. The inter-stage matching network consists of a matching transformer, an impedance conversion circuit, and a bias circuit. Its function is to efficiently transmit the signal output by the first-stage amplifier to the second-stage amplifier, and adjust the impedance and phase of the signal during the transmission process to ensure distortion-free amplification of the signal. Inside the second-stage amplifier, high-gain amplifier devices and optimized amplification topology are used, combined with the synergistic effect of the inter-stage matching network to significantly increase the gain of the input signal, so that the signal strength reaches a level that can meet the requirements of complex signal processing algorithms, while maintaining a low noise level to avoid the synchronous enhancement of noise due to excessive amplification.

[0074] The third-stage amplifier performs output impedance matching and provides power gain. Specifically, the signal processed by the first two amplifier stages undergoes final optimization and output preparation in the third-stage amplifier. The primary task of the third-stage amplifier is output impedance matching. By designing an output matching network, the output impedance of the amplifier circuit is adapted to the input impedance of the subsequent robust adaptive filtering algorithm module, ensuring that the signal can be transmitted efficiently and without reflection to the next processing step. In addition, the third-stage amplifier must also provide a certain amount of power gain. By selecting power-type amplifier components and optimizing the amplifier circuit's power supply, heat dissipation design, and bias circuit, the signal output power is increased while ensuring signal quality, and the signal drive capability is enhanced. This allows the original discharge signal, processed by the three-stage cascaded low-noise amplifier circuit, to be input into the robust adaptive filtering algorithm module with a high signal-to-noise ratio, appropriate power, and matched impedance, providing a high-quality data foundation for subsequent noise suppression and signal adjustment.

[0075] The total noise coefficient of the three-stage cascaded low-noise amplifier circuit satisfies:

[0076]

[0077] in, is the total noise figure, 、 、 are the noise coefficients of the first-stage amplifier, the second-stage amplifier, and the third-stage amplifier, respectively. 、 are the gains of the first-stage amplifier and the second-stage amplifier respectively.

[0078] In some embodiments of the present application, the process of determining abnormal discharge signals based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy in step S3 is introduced, which may specifically include:

[0079] Step S31, calculating an anomaly score based on the anomaly probability output by the Softmax classifier;

[0080] Step S32: performing temperature scaling calibration on the abnormal probability, and setting a dynamic threshold according to the quantile of the calibrated probability;

[0081] Step S33: When the abnormal score exceeds the set dynamic threshold, it is determined to be an abnormal discharge signal.

[0082] Specifically, the anomaly score is calculated based on the anomaly probability output by the Softmax classifier. Specifically, the discharge signal features processed by the feature pyramid network, SE module, bidirectional long short-term memory network and attention mechanism are input into the Softmax classifier. The Softmax classifier maps the feature vector into a probability distribution, and outputs the probability values ​​of the signal belonging to different categories such as normal state and abnormal discharge state, wherein the probability value of the abnormal discharge state directly reflects the possibility that the current signal is abnormal. In order to more intuitively measure the degree of abnormality of the signal, an anomaly score is generated based on the anomaly probability output by the Softmax classifier in combination with specific calculation rules. The calculation rule can be flexibly designed according to actual needs, for example, the anomaly probability is directly used as the anomaly score, or the anomaly score is obtained by performing mathematical operations such as weighting and transformation on the anomaly probability. By calculating the anomaly score, the abstract probability value is converted into a more operational and comparable numerical indicator, which is convenient for subsequent comparison and analysis with the threshold.

[0083] The abnormal probability is temperature scaled and calibrated, and a dynamic threshold is set according to the quantile of the probability after calibration. Specifically, since the abnormal probability output by the Softmax classifier may be biased and affected by the distribution of model training data, environmental factors, etc., directly using the original probability for judgment may lead to misjudgment. Therefore, it is necessary to perform temperature scaling calibration on the abnormal probability. Temperature scaling adjusts the Softmax function by introducing a temperature parameter, changes the sharpness of the probability distribution, and makes the probability value more accurately reflect the actual abnormality of the signal. The abnormal probability output by the Softmax classifier is temperature scaled and calibrated, and the temperature scaling calibration formula is:

[0084]

[0085] in, is the temperature scaling value of the Softmax classifier class i, is the original logical value of the ith neuron of the Softmax classifier, T is the temperature parameter, and K is the total number of categories.

[0086] Adjusting the temperature parameters can make the calibrated probability distribution more reasonable. After calibration, the dynamic threshold is set based on the quantile of the calibrated probability. The quantile is selected based on the probability distribution characteristics of normal and abnormal signals in historical monitoring data. For example, the probability value corresponding to a higher quantile (such as the 95th quantile) is selected as the dynamic threshold to ensure that the threshold can adapt to changes in the signal probability distribution under different monitoring environments, effectively reducing the false positive rate while maintaining detection sensitivity.

[0087] When the abnormality score exceeds the set dynamic threshold, it is determined to be an abnormal discharge signal. Specifically, the abnormality score calculated in step S31 is compared with the dynamic threshold set in step S32. If the abnormality score is greater than the dynamic threshold, it indicates that there is a high possibility that the current discharge signal is abnormal and has exceeded the range of the normal signal probability distribution. At this time, the system determines that the signal is an abnormal discharge signal and triggers the corresponding alarm or recording mechanism so that the operation and maintenance personnel can promptly inspect and handle electrical equipment that may have faults; if the abnormality score is less than or equal to the dynamic threshold, it is considered that the current signal belongs to a discharge signal under normal operating conditions, and the system continues to monitor and analyze subsequent signals in real time. By comparing and determining the abnormality score with the dynamic threshold, accurate classification of the discharge signal is achieved, effectively solving the problem of low signal recognition accuracy and easy misjudgment in the existing technology.

[0088] A weak discharge signal recognition device provided in an embodiment of the present application is described below. The weak discharge signal recognition device described below and the weak discharge signal recognition method described above can correspond to each other.

[0089] See also Figure 2 , Figure 2 This is a schematic diagram of a weak discharge signal recognition device disclosed in an embodiment of the present application.

[0090] like Figure 2 As shown, the weak discharge signal recognition device may include:

[0091] The original discharge capture unit 110 is used to obtain the original discharge signal captured at the key node of the distribution network line using a broadband sensor;

[0092] a gain suppression processing unit 120 for performing gain processing and noise suppression processing on the original discharge signal through a three-stage cascaded low-noise amplifier circuit and a robust adaptive filtering algorithm to obtain an adjusted discharge signal, wherein the gain parameter is adjusted in real time based on a feedback mechanism constructed by H∞ control, and the robust adaptive filtering algorithm uses a Huber function, M estimation, or entropy-related metric as an objective function;

[0093] The signal model detection unit 130 is used to input the adjusted discharge signal into the weak discharge signal detection model, use the feature pyramid network to fuse multi-level feature maps, enhance the key channel features through the SE module, combine the bidirectional long short-term memory network and the attention mechanism to extract the timing features, and determine the abnormal discharge signal based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy.

[0094] It can be seen from the above technical solutions that the embodiments of the present application provide a method and related equipment for identifying weak discharge signals, which realize intelligent identification of weak discharge signals in complex environments through technologies such as wide-band sensor acquisition, multi-level noise suppression, adaptive feedback control, and deep learning feature recognition. Specifically, wide-band sensors are installed at key nodes of the distribution network line to solve the problems of traditional sensors being easily affected by moisture and having a narrow signal reception range in high-humidity environments, and can capture weak signals. A three-stage cascaded low-noise amplification circuit is combined with a robust adaptive filtering algorithm with Huber function, M estimation or entropy-related measurement as the objective function, and the gain parameters are adjusted in real time based on the H∞ control feedback mechanism to effectively suppress pulse-type, narrowband and other noise interference, overcoming the defect that the traditional fixed threshold filtering algorithm cannot distinguish between normal operation signals and abnormal discharge signals. The adjusted discharge signal is input into the weak discharge signal detection model, and the multi-level feature maps are fused with the help of the feature pyramid network. The key channel features are enhanced through the SE module, and the temporal features are extracted by combining the bidirectional long short-term memory network and the attention mechanism to form a joint capture capability of local fine-grained features and global semantics. The Softmax classifier is then used to output the abnormality probability and combined with the dynamic threshold adjustment strategy to achieve fine-grained feature extraction and accurate classification of the discharge signal, solving the problem of low signal recognition accuracy and easy misjudgment due to the lack of environmental adaptive mechanism in the existing technology. Through the collaboration of multiple technologies, this solution significantly improves the sensitivity, anti-interference ability and recognition accuracy of weak discharge signal detection in high humidity and strong electromagnetic interference scenarios, meeting the needs of online monitoring of the insulation status of electrical equipment.

[0095] Optionally, the process of performing gain processing on the original discharge signal by the three-stage cascaded low-noise amplifier circuit includes:

[0096] The first-stage amplifier uses GaAs HEMT transistors to construct a common-source input matching network to minimize the noise figure;

[0097] The second-stage amplifier achieves signal gain improvement through the inter-stage matching network;

[0098] The third stage amplifier performs output impedance matching and provides power gain.

[0099] Optionally, the total noise coefficient of the three-stage cascaded low-noise amplifier circuit satisfies:

[0100]

[0101] in, is the total noise figure, 、 、 are the noise coefficients of the first-stage amplifier, the second-stage amplifier, and the third-stage amplifier, respectively. 、 are the gains of the first-stage amplifier and the second-stage amplifier respectively.

[0102] Optionally, when the robust adaptive filtering algorithm adopts the Huber function as the objective function, the objective function formula is:

[0103]

[0104] in, is the objective function, is the filtered error signal, The dynamic threshold is set based on the standard deviation of the ambient noise.

[0105] Optionally, the abnormal discharge signal is determined based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy, including:

[0106] Calculating an anomaly score based on the anomaly probability output by the Softmax classifier;

[0107] Performing temperature scaling calibration on the anomaly probability, and setting a dynamic threshold according to a quantile of the calibrated probability;

[0108] When the abnormality score exceeds the set dynamic threshold, it is determined to be an abnormal discharge signal.

[0109] Optionally, the H∞ control is implemented by minimizing the H∞ norm of the system transfer function, where the norm is defined as:

[0110]

[0111] in, is the norm, is the closed-loop transfer coefficient from the disturbance input w to the control output z, is the supremum for all frequencies, is the largest singular value of the matrix, and w is the angular frequency.

[0112] Optionally, the method further includes performing temperature scaling calibration on the abnormal probability output by the Softmax classifier, wherein the temperature scaling calibration formula is:

[0113]

[0114] in, is the temperature scaling value of the Softmax classifier class i, is the original logical value of the ith neuron of the Softmax classifier, T is the temperature parameter, and K is the total number of categories.

[0115] The weak discharge signal recognition device provided in the embodiment of the present application can be applied to a weak discharge signal recognition device. Figure 3 The hardware structure diagram of the weak discharge signal recognition device is shown. Figure 3 ,The hardware structure of the weak discharge signal recognition device may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;

[0116] In the embodiment of the present application, the number of the processor 1, the communication interface 2, the memory 3, and the communication bus 4 is at least one, and the processor 1, the communication interface 2, and the memory 3 communicate with each other through the communication bus 4;

[0117] The processor 1 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention;

[0118] The memory 3 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory;

[0119] The memory stores a program, and the processor can call the program stored in the memory, wherein the program is used to:

[0120] Obtain the original discharge signal captured by broadband sensors at key nodes of the distribution network;

[0121] The original discharge signal is subjected to gain processing and noise suppression processing in sequence through a three-stage cascaded low-noise amplifier circuit and a robust adaptive filtering algorithm to obtain an adjusted discharge signal, wherein the gain parameter is adjusted in real time based on a feedback mechanism constructed by H∞ control, and the robust adaptive filtering algorithm uses a Huber function, M estimation, or entropy-related measurement as an objective function;

[0122] The adjusted discharge signal is input into the weak discharge signal detection model, and the multi-level feature maps are fused using the feature pyramid network. The key channel features are enhanced through the SE module. The temporal features are extracted by combining the bidirectional long short-term memory network and the attention mechanism. The abnormal discharge signal is determined based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy.

[0123] Optionally, the refined functions and extended functions of the program may refer to the above description.

[0124] The present application also provides a readable storage medium, which may store a program suitable for execution by a processor, wherein the program is used to:

[0125] Obtain the original discharge signal captured by broadband sensors at key nodes of the distribution network;

[0126] The original discharge signal is subjected to gain processing and noise suppression processing in sequence through a three-stage cascaded low-noise amplifier circuit and a robust adaptive filtering algorithm to obtain an adjusted discharge signal, wherein the gain parameter is adjusted in real time based on a feedback mechanism constructed by H∞ control, and the robust adaptive filtering algorithm uses a Huber function, M estimation, or entropy-related measurement as an objective function;

[0127] The adjusted discharge signal is input into the weak discharge signal detection model, and the multi-level feature maps are fused using the feature pyramid network. The key channel features are enhanced through the SE module. The temporal features are extracted by combining the bidirectional long short-term memory network and the attention mechanism. The abnormal discharge signal is determined based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy.

[0128] Optionally, the refined functions and extended functions of the program may refer to the above description.

[0129] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0130] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0131] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying a weak discharge signal, characterized in that: include: Obtain the original discharge signal captured by broadband sensors at key nodes of the distribution network; The original discharge signal is subjected to gain processing and noise suppression processing in sequence through a three-stage cascaded low-noise amplifier circuit and a robust adaptive filtering algorithm to obtain an adjusted discharge signal, wherein the gain parameter is adjusted in real time based on a feedback mechanism constructed by H∞ control, and the robust adaptive filtering algorithm uses a Huber function, M estimation, or entropy-related measurement as an objective function; The adjusted discharge signal is input into the weak discharge signal detection model, and the multi-level feature maps are fused using the feature pyramid network. The key channel features are enhanced through the SE module. The temporal features are extracted by combining the bidirectional long short-term memory network and the attention mechanism. The abnormal discharge signal is determined based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy.

2. The method according to claim 1, characterized in that The process of performing gain processing on the original discharge signal by the three-stage cascade low-noise amplifier circuit includes: The first-stage amplifier uses GaAs HEMT transistors to construct a common-source input matching network to minimize the noise figure; The second-stage amplifier achieves signal gain improvement through the inter-stage matching network; The third stage amplifier performs output impedance matching and provides power gain.

3. The method according to claim 2, characterized in that The total noise coefficient of the three-stage cascaded low-noise amplifier circuit satisfies: in, is the total noise figure, 、 、 are the noise coefficients of the first-stage amplifier, the second-stage amplifier, and the third-stage amplifier, respectively. 、 are the gains of the first-stage amplifier and the second-stage amplifier respectively.

4. The method according to claim 1, wherein When the robust adaptive filtering algorithm adopts the Huber function as the objective function, the objective function formula is: in, is the objective function, is the filtered error signal, The dynamic threshold is set based on the standard deviation of the ambient noise.

5. The method according to claim 1, wherein The abnormal discharge signal is determined based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy, including: Calculating an anomaly score based on the anomaly probability output by the Softmax classifier; Performing temperature scaling calibration on the anomaly probability, and setting a dynamic threshold according to a quantile of the calibrated probability; When the abnormality score exceeds the set dynamic threshold, it is determined to be an abnormal discharge signal.

6. The method according to claim 1, characterized in that The H∞ control is achieved by minimizing the H∞ norm of the system transfer function, which is defined as: in, is the norm, is the closed-loop transfer coefficient from the disturbance input w to the control output z, is the supremum for all frequencies, is the largest singular value of the matrix, and w is the angular frequency.

7. The method according to claim 1, characterized in that It also includes temperature scaling calibration of the abnormal probability output by the Softmax classifier, and the temperature scaling calibration formula is: in, is the temperature scaling value of the Softmax classifier class i, is the original logical value of the ith neuron of the Softmax classifier, T is the temperature parameter, and K is the total number of categories.

8. A weak discharge signal recognition device, characterized in that: include: The original discharge capture unit is used to obtain the original discharge signal captured at the key node of the distribution network line using a broadband sensor; a gain suppression processing unit, configured to sequentially perform gain processing and noise suppression processing on the original discharge signal through a three-stage cascaded low-noise amplifier circuit and a robust adaptive filtering algorithm to obtain an adjusted discharge signal, wherein the gain parameter is adjusted in real time based on a feedback mechanism constructed by H∞ control, and the robust adaptive filtering algorithm uses a Huber function, an M estimate, or an entropy-related metric as an objective function; The signal model detection unit is used to input the adjusted discharge signal into the weak discharge signal detection model, use the feature pyramid network to fuse multi-level feature maps, enhance the key channel features through the SE module, combine the bidirectional long short-term memory network and the attention mechanism to extract the timing features, and determine the abnormal discharge signal based on the abnormal probability output by the Softmax classifier and the dynamic threshold adjustment strategy.

9. A weak discharge signal recognition device, characterized in that: including memory and processor; The memory is used to store programs; The processor is configured to execute the program to implement each step of the weak discharge signal recognition method according to any one of claims 1 to 7.

10. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the weak discharge signal recognition method according to any one of claims 1 to 7 is implemented.

Images