Fault type identification method for power equipment and related equipment

Abstract

The embodiment of the invention discloses a fault type identification method for power equipment and related equipment, and the method comprises the steps: firstly obtaining an IMF component of an electric signal of each triboelectric sensor, calculating the effectiveness index of the IMF component, screening out the IMF component of which the effectiveness index is greater than a preset first threshold value, and carrying out the reconstruction, thereby removing the invalid information, such as noise, electromagnetic interference, etc. And carrying out power spectrum estimation on the reconstructed signal to obtain a center frequency, simultaneously obtaining a high-low frequency cross-band correlation coefficient, and extracting the characteristics of the triboelectric sensing signal from multiple dimensions. And data and labels used for training are constructed according to the features and fault types, so that the model can comprehensively and accurately learn different fault type features. And during identification, the to-be-identified data set is input into the trained model, so that the fault type can be accurately judged, and the accuracy and reliability of fault diagnosis are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, and in particular to a method and related equipment for identifying fault types in power equipment. Background Technology

[0002] High-voltage power equipment operates in complex environments with high voltage, high current, and strong electromagnetic fields for extended periods, making it prone to mechanical failures, equipment damage, and potentially large-scale power outages, resulting in severe economic losses and social impacts. Therefore, condition monitoring of high-voltage switchgear and other equipment is of paramount importance.

[0003] In the operation of power equipment, fault diagnosis is crucial. Monitoring methods, such as triboelectric / acoustic sensors, can convert vibration and acoustic information from equipment operation into electrical signals. Different fault types typically produce different characteristics in these signals. However, raw triboelectric vibration (or acoustic) signals collected in the field are often non-stationary, broadband, and susceptible to environmental noise and electromagnetic interference. Directly using raw electrical signals or simple features for fault type identification makes it difficult to reliably extract the effective components related to the fault, leading to inaccurate fault type predictions. Summary of the Invention

[0004] In view of this, the present invention provides a method for identifying fault types of power equipment and related equipment.

[0005] The specific technical solution of the first embodiment of the present invention is as follows: a method for identifying fault types of power equipment, the method comprising: acquiring a dataset of triboelectric sensor signals of power equipment; the dataset of triboelectric sensor signals includes n triboelectric sensor signals corresponding to different fault types; performing empirical mode decomposition on the triboelectric sensor signals to obtain k IMF components of the triboelectric sensor signals; calculating a validity index for each IMF component; the validity index is used to characterize the probability of the IMF component being valid; reconstructing the IMF components in the triboelectric sensor signals whose validity index is greater than a preset first threshold to obtain n reconstructed signals; performing power spectrum estimation on the reconstructed signals to obtain the center frequency of the reconstructed signals; and obtaining the high-low frequency of the triboelectric sensor signals. The high-low frequency crossband correlation coefficient is used to construct training data by taking the effectiveness index, center frequency, and high-low frequency crossband correlation coefficient corresponding to n triboelectric sensor electrical signals, using the fault type corresponding to the triboelectric sensor electrical signals as training labels, and training a preset fault type identification model using the training data and training labels to obtain a trained fault type identification model; the dataset to be identified for the triboelectric sensor electrical signals to be identified is obtained, the dataset to be identified includes the effectiveness index to be identified, the center frequency to be identified, and the high-low frequency crossband correlation coefficient to be identified for the triboelectric sensor electrical signals to be identified; the dataset to be identified is input into the trained fault type identification model to obtain the fault type corresponding to the triboelectric sensor electrical signals to be identified output by the trained fault type identification model.

[0006] Preferably, obtaining the high-low frequency crossband correlation coefficient of the triboelectric sensor electrical signal includes: filtering and decomposing the triboelectric sensor electrical signal to obtain a first envelope energy sequence of the high-frequency band sensitive to partial discharge and a second envelope energy sequence of the low-frequency band sensitive to structural vibration, as well as a first time average value corresponding to the first envelope energy sequence and a second time average value corresponding to the second envelope energy sequence; and calculating the high-low frequency crossband correlation coefficient of each triboelectric sensor electrical signal based on the first envelope energy sequence, the second envelope energy sequence, the first time average value, and the second time average value.

[0007] Preferably, the high-low frequency crossband correlation coefficient is obtained using the following formula:

[0008] in, The high-low frequency crossband correlation coefficient is given, and T is the total duration of the triboelectric sensor signal. The first envelope energy sequence, This is the average value over the first time period. The second envelope energy sequence, This is the second time average.

[0009] Preferably, the method further includes: obtaining the first root mean square value, high-frequency band spectral entropy, first mean and first standard deviation of the high-frequency instantaneous frequency of the first envelope energy sequence; obtaining the second root mean square value, low-frequency band spectral entropy, second mean and second standard deviation of the low-frequency instantaneous frequency of the second envelope energy sequence; obtaining the high-low frequency energy ratio based on the first root mean square value and the second root mean square value; constructing a discriminative feature vector based on the first root mean square value, the high-frequency band spectral entropy, the first mean, the first standard deviation, the second root mean square value, the low-frequency band spectral entropy, the second mean, the second standard deviation, and the discriminative feature vector; then, constructing the effectiveness index, the center frequency, and the high-low frequency cross-band correlation coefficient corresponding to the n triboelectric sensor electrical signals as training data includes: constructing the discriminative feature vector, the effectiveness index, the center frequency, and the high-low frequency cross-band correlation coefficient corresponding to the n triboelectric sensor electrical signals as training data; the dataset to be identified further includes the discriminative feature vector to be identified of the triboelectric sensor electrical signals to be identified.

[0010] Preferably, the calculation of the effectiveness index of each IMF component includes: obtaining the first energy, kurtosis, and second energy of the triboelectric sensor electrical signal corresponding to the IMF component, and obtaining the Pearson linear correlation coefficient between the IMF component and its corresponding triboelectric sensor electrical signal; and obtaining the effectiveness index of the IMF component based on the first energy, the kurtosis, the second energy, and the Pearson linear correlation coefficient.

[0011] Preferably, the effectiveness index is obtained by: obtaining the energy ratio between the first energy and the second energy; and weighting and summing the absolute values ​​of the energy ratio, the kurtosis, and the Pearson linear correlation coefficient to obtain the effectiveness index.

[0012] Preferably, when training the preset fault type identification model using the training data and the training labels, a discriminative coupling loss function is used to simultaneously compress the feature distance of training data of the same fault type and expand the feature distance of training data of different fault types.

[0013] The specific technical solution of the second embodiment of the present invention is as follows: a fault type identification system for power equipment, the system comprising: a dataset acquisition module, an empirical mode decomposition module, a validity index calculation module, a reconstruction module, a power spectrum estimation module, a cross-band correlation coefficient acquisition module, a model training module, a dataset acquisition module, and a fault type output module; the dataset acquisition module is used to acquire a dataset of triboelectric sensor signals of power equipment; the triboelectric sensor signal dataset includes n triboelectric sensor signals corresponding to different fault types; the empirical mode decomposition module is used to perform empirical mode decomposition on the triboelectric sensor signals to obtain k IMF components of the triboelectric sensor signals; the validity index calculation module is used to calculate the validity index of each IMF component; the validity index is used to characterize the probability of the IMF component being valid; the reconstruction module is used to reconstruct the IMF components in the triboelectric sensor signals whose validity index is greater than a preset first threshold to obtain n reconstructed signals; the power spectrum estimation module is used to analyze the reconstructed signals... The power spectrum is estimated to obtain the center frequency of the reconstructed signal; the cross-band correlation coefficient acquisition module is used to obtain the high-low frequency cross-band correlation coefficient of the triboelectric sensor signal; the model training module is used to construct training data from the effectiveness index, center frequency, and high-low frequency cross-band correlation coefficient corresponding to n triboelectric sensor signals, use the fault type corresponding to the triboelectric sensor signal as training label, and use the training data and training label to train a preset fault type identification model to obtain a trained fault type identification model; the dataset acquisition module is used to acquire the dataset to be identified from the triboelectric sensor signal, which includes the effectiveness index, center frequency, and high-low frequency cross-band correlation coefficient to be identified; the fault type output module is used to input the dataset to be identified into the trained fault type identification model to obtain the fault type corresponding to the triboelectric sensor signal output by the trained fault type identification model.

[0014] The specific technical solution of the third embodiment of the present invention is as follows: a fault type identification device for power equipment, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method as described in any one of the first embodiments of this application.

[0015] The specific technical solution of the fourth embodiment of the present invention is as follows: a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor performs the steps of the method as described in any one of the first embodiments of this application.

[0016] Implementing the embodiments of the present invention will have the following beneficial effects: This invention first acquires the IMF (Integrated Motion Filter) component of the electrical signal from each triboelectric sensor and calculates its effectiveness index. IMF components with effectiveness indices greater than a preset first threshold are then reconstructed. This process removes invalid information such as noise and electromagnetic interference, resulting in a higher-quality signal. The reconstructed signal is then subjected to power spectrum estimation to obtain the center frequency, and high-low frequency crossband correlation coefficients are also acquired, extracting features of the triboelectric sensing signal from multiple dimensions. These features, along with fault types, are used to construct training data and labels, enabling the model to comprehensively and accurately learn the features of different fault types. During identification, the dataset to be identified is input into the trained model, allowing for accurate fault type determination, significantly improving the accuracy and reliability of fault diagnosis. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the steps involved in a method for identifying fault types in power equipment. Figure 2 This is a schematic diagram of the structure of a triboacoustic sensor; Figure 3 A schematic diagram showing the electrical signal obtained by a triboelectric pressure sensor during the operation of power equipment. Figure 4 This is a front view of the overall structure of the triboacoustic sensor; Figure 5 This is a top view of the overall structure of the triboacoustic sensor; Figure 6 Bottom view of the overall structure of the triboacoustic sensor; Figure 7 This is a schematic diagram of the monitoring process of an intelligent acoustic online monitoring system. Figure 8 This is a schematic diagram of a drone-based electricity meter positioning system. Figure 9 This is a diagram of the internal structure of a computer device. The components are as follows: 201, Supporting material; 202, Upper electrode; 203, Positive electro-friction material; 204, Multilayer gradient damping membrane pad; 205, Negative electro-friction material; 206, Lower electrode; 207, Supporting material; 301, Dataset acquisition module; 302, Empirical mode decomposition module; 303, Effectiveness index calculation module; 304, Reconstruction module; 305, Power spectrum estimation module; 306, Cross-band correlation coefficient acquisition module; 307, Model training module; 308, Dataset acquisition module; 309, Fault type output module. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0020] The terms "first," "second," etc., used in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or modules is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other steps or modules inherent to such processes, methods, products, or apparatus.

[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] High-voltage power equipment operates in complex environments characterized by high voltage, high current, and strong electromagnetic fields, making it susceptible to mechanical failures, equipment damage, and potentially large-scale power outages, resulting in severe economic losses and social impact. Therefore, condition monitoring of high-voltage switchgear and similar equipment is of paramount importance.

[0023] In recent years, the intelligentization of power equipment has flourished in traditional manufacturing. The advent of intelligent vibration sensors has provided more information for equipment fault prediction, making vibration monitoring more precise. However, traditional monitoring methods typically rely on piezoelectric or capacitive sensors, most of which require external power supply, making them difficult to install and costly in high-voltage or enclosed environments. Furthermore, existing vibration monitoring systems are limited by various environmental issues, are susceptible to strong electromagnetic interference in power equipment applications, and exhibit weak system stability. In addition, regular maintenance of lines and sensors is required during daily operation, resulting in high subsequent maintenance costs.

[0024] Existing triboelectric nanogenerators operate on the principle of coupling triboelectric effect and electrostatic induction effect, converting weak mechanical energy in the environment into electrical energy. Simultaneously, the electrical signals generated in this process can also be used to reflect the physical parameters of mechanical motion, thus enabling the device to serve as a type of self-driven triboelectric sensor with significant advantages such as simple structure, widely available materials, and low cost.

[0025] Existing online vibration monitoring systems suffer from complex hardware circuits and poor anti-interference capabilities, resulting in significantly reduced monitoring effectiveness, and the sensors used are also expensive. Therefore, it is necessary to develop a new type of sensor to monitor the operating status of power equipment.

[0026] To meet the needs of existing technologies, this invention employs a novel triboacoustic sensor to provide a low-cost, high-sensitivity wireless acoustic online monitoring system capable of fault identification and early warning. It solves the problems of difficult and costly installation and complex wiring associated with traditional vibration sensors, and overcomes the limitations of frequency sensing in traditional vibration sensors. By analyzing the acoustic characteristics of power equipment during operation, it can reflect the operating status of the power equipment in real time, thereby achieving fault identification and early warning.

[0027] Please see Figure 1 The above is a flowchart of the steps of a fault type identification method for power equipment in the first embodiment of this application, which aims to improve the accuracy and reliability of fault diagnosis. The method includes: Step 101: Obtain the triboelectric sensor signal dataset of the power equipment; the triboelectric sensor signal dataset includes triboelectric sensor signals corresponding to n different fault types; Step 102: Perform empirical mode decomposition on the triboelectric sensor electrical signal to obtain k IMF components of the triboelectric sensor electrical signal; Step 103: Calculate the validity index for each IMF component; the validity index is used to characterize the probability that the IMF component is valid; Step 104: Reconstruct the IMF components in the triboelectric sensor electrical signal whose effectiveness index is greater than a preset first threshold to obtain n reconstructed signals; Step 105: Perform power spectrum estimation on the reconstructed signal to obtain the center frequency of the reconstructed signal; Step 106: Obtain the high-low frequency crossband correlation coefficient of the triboelectric sensor electrical signal; Step 107: Construct training data by using the effectiveness index, center frequency, and high-low frequency crossband correlation coefficient corresponding to the n triboelectric sensor electrical signals, use the fault type corresponding to the triboelectric sensor electrical signals as training labels, and use the training data and training labels to train the preset fault type identification model to obtain the trained fault type identification model. Step 108: Obtain the dataset to be identified for the triboelectric sensor electrical signal to be identified. The dataset to be identified includes the identification validity index, the identification center frequency, and the identification high-low frequency crossband correlation coefficient of the triboelectric sensor electrical signal to be identified. Step 109: Input the dataset to be identified into the trained fault type identification model to obtain the fault type corresponding to the electrical signal of the triboelectric sensor to be identified output by the trained fault type identification model.

[0028] Specifically, triboelectric vibration sensors are placed on the surface of power equipment to acquire a dataset of triboelectric sensor electrical signals (hereinafter referred to as electrical signals). This dataset covers electrical signals corresponding to n different fault types, such as partial discharge (including surface discharge, metallic discharge, and suspension discharge), mechanical loosening, etc. Empirical Mode Decomposition (EMD) is performed on each electrical signal to obtain k Intrinsic Functional Factor (IMF) components. A validity index for each IMF component is calculated; this index measures the probability of the IMF component being valid. A preset first threshold is set. (e.g., 0.5) IMF components with an effectiveness index greater than this threshold are reconstructed to obtain n reconstructed signals. Power spectrum estimation is performed on the reconstructed signals to determine their center frequencies. Simultaneously, the high-low frequency cross-band correlation coefficient for each electrical signal is calculated. The effectiveness index, center frequency, and high-low frequency cross-band correlation coefficient corresponding to the n electrical signals are used to construct training data. The actual fault types corresponding to the electrical signals are used as training labels. This training data and labels are used to train a preset fault type identification model. After multiple iterations and optimizations, the trained fault type identification model is obtained. The dataset to be identified for the triboelectric sensor electrical signals is obtained. This dataset contains the effectiveness index, center frequency, and high-low frequency cross-band correlation coefficient to be identified. The dataset to be identified is input into the trained fault type identification model. After calculation and analysis, the model outputs the fault type corresponding to the triboelectric sensor electrical signal to be identified, thus completing the accurate identification of power equipment faults.

[0029] The method in this embodiment first acquires the IMF component of the electrical signal of each triboelectric sensor and calculates its effectiveness index. IMF components with effectiveness indices greater than a preset first threshold are selected for reconstruction. This process removes invalid information such as noise and electromagnetic interference, resulting in a higher-quality signal. Power spectrum estimation is performed on the reconstructed signal to obtain the center frequency, and high-low frequency cross-band correlation coefficients are also obtained, extracting features of the triboelectric sensing signal from multiple dimensions. These features are then combined with fault types to construct training data and labels, enabling the model to comprehensively and accurately learn the features of different fault types. During identification, the dataset to be identified is input into the trained model to accurately determine the fault type, greatly improving the accuracy and reliability of fault diagnosis. Through the above method based on the IMF effectiveness index… Adaptive center frequency Bandpass filtering and crossband correlation coefficient The feature construction method has formed a dedicated preprocessing and feature extraction system for triboelectric vibration signals of power equipment, which can significantly improve the distinguishability and identification accuracy of partial discharge and mechanical faults in complex power field environments.

[0030] In a specific embodiment, obtaining the high-low frequency cross-band correlation coefficient of the triboelectric sensor electrical signal includes: filtering and decomposing the triboelectric sensor electrical signal to obtain a first envelope energy sequence of the high-frequency band sensitive to partial discharge and a second envelope energy sequence of the low-frequency band sensitive to structural vibration, as well as a first time average value corresponding to the first envelope energy sequence and a second time average value corresponding to the second envelope energy sequence; and calculating the high-low frequency cross-band correlation coefficient of each triboelectric sensor electrical signal based on the first envelope energy sequence, the second envelope energy sequence, the first time average value, and the second time average value. Specifically, by obtaining the high-low frequency cross-band correlation coefficient, the correlation between the high-frequency and low-frequency characteristics of the electrical signal can be comprehensively reflected. Using it for fault type identification can enrich the feature dimensions, enabling the model to capture more comprehensive electrical signal feature information and improve the accuracy and reliability of fault identification. In a specific embodiment, the high-low frequency crossband correlation coefficient is obtained using the following formula:

[0031] in, The high-low frequency crossband correlation coefficient is given, and T is the total duration of the triboelectric sensor signal. The first envelope energy sequence, This is the average value over the first time period. The second envelope energy sequence, This is the second time average.

[0032] Specifically, this embodiment utilizes wavelet packet decomposition, envelope demodulation, and higher-order spectral analysis methods to construct cross-band correlation features reflecting the coupling relationship between high-frequency partial discharge impacts and low-frequency vibrations of the equipment structure. The envelope energy sequence of the sensitive high-frequency band for partial discharge is obtained through wavelet packet or filter decomposition. The structural vibration-sensitive low-frequency envelope energy sequence is Their time averages are respectively , . Characterizing the synchronous change relationship between high-frequency partial discharge impact energy and low-frequency vibration energy of equipment structure: When partial discharge generates impact excitation, the high-frequency triboelectric signal and the low-frequency structural vibration exhibit significant energy coupling, making... Increase; while simple mechanical loosening or background vibration exhibit different statistical patterns. This feature has high discriminative power in the electrical signal processing of this embodiment and can be combined with conventional features such as spectral entropy and instantaneous frequency statistics to form a discriminative feature vector.

[0033] In a specific embodiment, the method further includes: obtaining the first root mean square value, high-frequency band spectral entropy, first mean and first standard deviation of the high-frequency instantaneous frequency of the first envelope energy sequence; obtaining the second root mean square value, low-frequency band spectral entropy, second mean and second standard deviation of the low-frequency instantaneous frequency of the second envelope energy sequence; obtaining the high-low frequency energy ratio based on the first root mean square value and the second root mean square value; constructing a discriminative feature vector based on the first root mean square value, the high-frequency band spectral entropy, the first mean, the first standard deviation, the second root mean square value, the low-frequency band spectral entropy, the second mean, the second standard deviation, and the discriminative feature vector; then, constructing the effectiveness index, the center frequency, and the high-low frequency cross-band correlation coefficient corresponding to the n triboelectric sensor electrical signals as training data includes: constructing the discriminative feature vector, the effectiveness index, the center frequency, and the high-low frequency cross-band correlation coefficient corresponding to the n triboelectric sensor electrical signals as training data; the dataset to be identified further includes the discriminative feature vector to be identified of the triboelectric sensor electrical signals to be identified.

[0034] Specifically, the high-low frequency crossband correlation coefficient It is not used as a standalone criterion, but rather together with various parameters characterizing the statistical properties of high-frequency partial discharge impact and low-frequency structural vibration to form a multidimensional discriminative feature vector for fault identification: preferably, for the high-frequency envelope energy sequence Calculate its root mean square value High-frequency band spectral entropy Mean of high-frequency instantaneous frequency and standard deviation ; for low-frequency envelope energy sequences Calculate its root mean square value Low-frequency band spectral entropy Mean of low-frequency instantaneous frequency and standard deviation And define the high-frequency energy ratio as Then, a discriminative feature vector can be constructed. During the offline training phase, electrical signals from power equipment under typical operating conditions, including normal operation, surface discharge, metal discharge, suspension discharge, and mechanical loosening, are collected. Feature vectors are extracted for each time window. The signal is then fed into a classifier (which can be a support vector machine, random forest, or a deep learning model compatible with this embodiment) in the signal analysis and recognition module. By minimizing the classification error or cross-entropy loss, the optimal decision boundary in the feature space for different operating conditions is learned. During the online monitoring phase, feature vectors are extracted from the real-time acquired electrical signals according to the same process. The system takes a pre-trained classification model as input, outputs the operating condition category and its confidence level corresponding to the current time window, and transmits the results to the intelligent online acoustic monitoring system for display and recording. This embodiment is based on... and Set an adaptive warning threshold when When the high-frequency discharge impact and low-frequency structural vibration are significantly enhanced together with the spectral entropy and instantaneous frequency statistics within several continuous monitoring windows, an early warning is triggered and an alarm signal is issued, thereby realizing the engineering application and quantitative utilization of the high-low frequency cross-band correlation coefficient in the identification of partial discharge and mechanical faults in power equipment.

[0035] In a specific embodiment, calculating the effectiveness index of each IMF component includes: obtaining the first energy, kurtosis, and second energy of the triboelectric sensor electrical signal corresponding to the IMF component, and obtaining the Pearson linear correlation coefficient between the IMF component and its corresponding triboelectric sensor electrical signal; and obtaining the effectiveness index of the IMF component based on the first energy, the kurtosis, the second energy, and the Pearson linear correlation coefficient.

[0036] Specifically, for the non-stationary, broadband electrical signals output by triboelectric vibration sensors in power equipment monitoring scenarios, targeted preprocessing and feature extraction algorithms were designed in the signal analysis and recognition module to improve the accuracy of identifying typical faults such as partial discharge and mechanical loosening. First, the preprocessing module performs empirical mode decomposition (EMD) on the acquired raw electrical signals and proposes an intrinsic mode function (IMF) validity discrimination function based on the fault characteristics of power equipment. Let k IMF components be obtained by performing EMD on the raw electrical signal x(t). Its energy is The total energy of the original electrical signal is , The kurtosis is The correlation coefficient between the Pearson linear correlation coefficient and the original electrical signal is ( ). ).

[0037] In a specific embodiment, the effectiveness index is obtained as follows: the energy ratio between the first energy and the second energy is obtained; the absolute values ​​of the energy ratio, the kurtosis, and the Pearson linear correlation coefficient are weighted and summed to obtain the effectiveness index.

[0038] Specifically, the effectiveness index is obtained using the following formula:

[0039] in, The effectiveness index is... For the first energy, For the second energy, For the kurtosis, The absolute value of the Pearson linear correlation coefficient. , and All are preset coefficients.

[0040] in , and These are weighting coefficients set based on experience with power equipment operating conditions. Only when... ≥Preset first threshold At that time, the IMF component was identified as an effective component related to partial discharge impact or mechanical vibration and participated in the reconstruction. By simultaneously considering energy proportion, impulsivity (kurtosis), and correlation, the above indicators are more suitable for the complex scenario of partial discharge pulses and structural vibration superimposed on power equipment than the existing single-indicator screening method.

[0041] Furthermore, in order to overcome the weighting coefficient in the existing technology , and The problem that most of these are manually preset constants that are difficult to adaptively adjust according to the operating conditions of power equipment is addressed in this embodiment by proposing an adaptive weight coefficient calculation method based on "operating condition perception - performance evaluation - weight normalization," building upon the aforementioned IMF effectiveness indicators. Specifically, in the offline phase, labeled sample sets are constructed for various typical operating conditions of power equipment (including normal operation, different types of partial discharge, and mechanical loosening, etc.). For each type of sample, the energy percentage is individually applied to each candidate IMF component. , cliff and correlation coefficient The three single criteria were used to distinguish between fault and normal states, and the recognition accuracy of each criterion under the current operating conditions was statistically analyzed. , , Among them, the correlation coefficient Represents the original electrical signal With the k-th IMF component The absolute value of the Pearson linear correlation coefficient between the two components is used to measure the degree of consistency between the IMF component and the overall trend of the original electrical signal in the time domain waveform. The larger the value, the more correlated the components of the IMF component are with the main dynamic changes of the original electrical signal, thus effectively reflecting the structural impact of fault events on the electrical signal. Based on this, this embodiment defines the operating condition sensitivity coefficient of the energy index. Operating condition sensitivity coefficient of kurtosis index Pearson linear correlation coefficient index operating condition sensitivity coefficient The corresponding weight coefficients are automatically obtained through normalization operations: , ,

[0042] When the operating conditions of the equipment change (such as significant changes in load current, ambient noise intensity, or partial discharge pulse density), statistics can be re-statistically compiled on the new operating condition label or quasi-label sample. , , Update in real time according to the above formula , , and its normalization result, thereby achieving , and The system adaptively adjusts to changing operating conditions. By explicitly transforming maintenance personnel's experience-based judgments on "which indicators are more effective under which operating conditions" into quantitative evaluation and normalized calculation logic for performance identification, the weighting coefficients determined in this embodiment retain both power equipment operation and maintenance experience and can be automatically calculated and updated online in embedded processors or host computers. Compared to existing IMF screening methods that use fixed weights or trial-and-error weighting based on experience, this significantly improves performance. The indicators demonstrate adaptability and reliability of fault identification under different equipment and operating conditions.

[0043] In a specific embodiment, power spectrum estimation is performed on the reconstructed electrical signal, and the k-th frequency point is denoted as... The corresponding spectral value is Then, the centroid of the spectral energy is defined as the center frequency of the adaptive bandpass filter:

[0044] Among them, the center frequency The calculations are limited to the target frequency band where the triboelectric sensor is most sensitive to partial discharge and mechanical loosening responses. The filter passes through the frequency band... Centered on the signal source, and dynamically adjusting the bandwidth based on the current spectral distribution characteristics and estimated signal-to-noise ratio, this method automatically tracks the frequency region where electrical signal energy is concentrated under different equipment, installation locations, and fault types. Compared with traditional fixed-parameter bandpass filtering, this method has stronger adaptability under complex power field noise conditions.

[0045] In a specific embodiment, when training the preset fault type identification model using the training data and the training labels, a discriminative coupling loss function is used to simultaneously compress the feature distance of training data of the same fault type and expand the feature distance of training data of different fault types.

[0046] Specifically, addressing the problem that different types of partial discharges in power equipment (such as switchgear), including surface discharge, metallic discharge, and floating discharge, exhibit high similarity in the electrical signals output by triboelectric vibration sensors, making them difficult for traditional identification models to effectively distinguish, this embodiment proposes a "feature decomposition-discrimination coupling" framework based on the existing deep learning structure to enhance the separability of different partial discharge types at the model structure level. Specifically, the model first employs a parallel dual-encoder structure to map the input signal to the pulse morphology feature space and the fine-grained spectral feature space, respectively, thus obtaining... and The former is used to characterize the differences in waveform morphology of different partial discharges (such as the steep peaks of metal discharges and the tail echoes of levitation discharges), while the latter is used to characterize the slight differences in energy distribution of different partial discharges in local frequency bands. This represents the pulse morphology characteristics of the input signal, describing the differences in peak steepness, pulse duration, and other morphological features among different partial discharge types. This provides a fine-grained spectral characteristic representation of the input signal, used to capture the subtle differences in local frequency components, resonant components, and energy distribution among different partial discharge types. This is a pulse morphology encoding function used to extract the waveform morphology features of partial discharge events. This is a spectral coding function used to extract fine-grained differences in the frequency components and spectral structure of partial discharge signals. To further enhance the discriminative ability between different partial discharge types simultaneously in both feature spaces, this embodiment constructs a discriminative coupling loss function. This function allows the network to simultaneously compress the feature distance of samples of the same type and expand the interval between features of different types of partial discharges in the two spaces during training. Its form is as follows:

[0047] in , For the type of partial discharge corresponding to the sample, , Weights are used to adjust waveform and spectral characteristics, and the weight matrix... The design uses positive values ​​for the same type of partial discharge and negative values ​​for different types of partial discharge, i.e., when ,≠ hour =-y (y>1), thus explicitly "pushing away" the distance in feature space for the most easily confused partial discharge types, such as surface discharge and metallic discharge, during training. Through this structural improvement, the model can simultaneously capture key difference features between different partial discharge types from both the waveform and spectrum dimensions, and adaptively expand the feature interval of easily confused partial discharge types during training, enabling the triboelectric vibration sensor to achieve higher type discrimination and detection accuracy in the partial discharge feature identification of switchgear.

[0048] During model training, It will participate in backpropagation along with the classification loss, and is used for gradient updates of the encoder parameters. Because The algorithm assigns weights with opposite signs to different types of samples. When sample i and sample j belong to different partial discharge types (such as surface discharge and metallic discharge), their corresponding weights change. Since the value is negative, gradient descent will drive the network to adjust its parameters so that the two classes of samples are in the pulse shape feature space. With spectral feature space The Euclidean distance in the model continuously increases, thereby achieving the goal of "pushing away" the feature cluster centers of different partial discharge types. Conversely, when samples belong to the same partial discharge category, the model compresses their feature distance through positive weight terms, making the features within that category more concentrated. Through the above mechanism, the network can automatically form a more discriminative feature distribution structure after training, enabling the most easily confused surface discharge and metallic discharge categories to form mutually distant and clearly spaced feature subspaces in the joint feature space.

[0049] go through Constraining the features after "pushing away" and The data will be further input into the discriminator for category prediction. Since the feature space has been renormalized and different partial discharge (PD) types have clear separable boundaries in the space, the classifier can separate the categories with smaller linear or nonlinear mappings, thus significantly improving the accuracy and robustness of PD type identification. Finally, the model outputs the PD category probability distribution corresponding to each input signal, forming a robust PD type identification result. In this way, It not only actively shapes a feature space structure that is conducive to differentiation during the training phase, but also directly improves the reliability of feature-based classification decisions during the inference phase.

[0050] To make the weight matrix in the above discriminative coupling loss function This approach can accurately reflect the "confusion level" between different partial discharge types, thereby enhancing the ability of the loss term to control the separability of the feature space. Instead of simply using fixed positive and negative constants as weights, this embodiment proposes an adaptive weight calculation method based on confusion matrix statistics. Specifically, on the baseline model trained using only classification loss, sample-by-sample inference is first performed on the labeled validation set to record the correspondence between the true and predicted categories, thus constructing a confusion count matrix. ,in Let represent the number of samples that are actually of class a but are misclassified as class b, and C be the total number of partial discharge types. Then, for any two distinct partial discharge categories a and b, their bidirectional confusion rate is defined as . ,in , , representing the total number of samples whose true label is class a or class b, respectively. Further, the confusion rates of all different class pairs are normalized to the maximum value to obtain... This is to quantitatively characterize the confusion intensity of different partial discharge category pairs. Based on the above statistics, this embodiment adjusts the weight matrix. Define it as follows:

[0051] in, >0 represents a preset category spacing amplification factor, used to enhance feature separation of highly confused category pairs. Through the above construction method, if two types of partial discharge (such as surface discharge and metallic discharge) are most easily confused in the baseline model, then its Larger, corresponding It has stronger negative weights, thus improving the discrimination coupling loss function. The model applies a greater spacing expansion to this category, automatically "pushing away" the distribution of these most easily confused categories in the pulse morphology feature space and fine-grained spectral feature space during training. By explicitly quantifying the "key differentiation pairs" from human experience and converting them into matrix weights that can be automatically updated by the computer, the adaptive weight matrix proposed in this embodiment not only improves the separability of different partial discharge types in the joint feature space, but also enhances the model's generalization ability under different devices and operating conditions, thereby further improving the accuracy and stability of partial discharge identification by triboelectric vibration sensors in complex power fields.

[0052] Figure 2This is a schematic diagram of a triboelectric acoustic sensor. The triboelectric pressure sensor has a layered structure, comprising: a support material 201, an upper electrode 202, a positively charged triboelectric material 203, a multilayer gradient damping film pad 204, a negatively charged triboelectric material 205, a lower electrode 206, and a support material 207. The support material 201 can be a sheet material such as acrylic or polycarbonate; the upper electrode 202 can be a conductive film such as copper, aluminum, or gold; the positively charged triboelectric material 203 can be a material such as Nylon or PET; the multilayer gradient damping film pad 204 is a dielectric material such as a PDMS / PVDF layer with gradually changing elastic modulus or thickness; the negatively charged triboelectric material 205 can be a dielectric material such as FEP, PTFE, or PVDF; and the lower electrode 206 can be a conductive film such as copper, aluminum, or gold. The porous structure on surfaces 201, 202, and 203 aims to reduce air damping, improve sensitivity, and broaden the sensor's frequency response. The multi-layered gradient damping membrane pad 204 provides space for thin-film vibration, while the acoustic impedance differences between layers allow the sensor to selectively respond to vibration signals of different frequencies. High-frequency signals (such as partial discharge) preferentially pass through the low-damping layer and generate an enhanced response on the vibrating membrane, while low-frequency signals (such as mechanical loosening) can effectively excite triboelectric output after being conducted through the high-damping layer. This multi-layered gradient damping membrane pad design significantly improves the sensor's signal-to-noise ratio within a specific frequency range, thereby enhancing the accuracy of subsequent feature extraction and fault identification.

[0053] The working principle of this triboacoustic sensor is as follows: when the sound of the device is transmitted to the surface of the device, the FEP film inside the sensor will vibrate and cause the contact to separate. Since the friction materials on the upper and lower layers of the sensor carry different charges, a high potential difference will be generated, which will drive the electrons in the external circuit to move in a specific direction, thereby generating an electrical signal.

[0054] Figure 3 This is a diagram showing the output electrical signal of a triboacoustic sensor in response to the sound of different devices operating. Figure 3 It can be seen that when there is no sound, the signal is the sensor's noise floor signal; when the equipment is running normally, there is a signal output compared to when there is no sound; the sound signal of metal discharge has a higher frequency and stronger signal than the signal during normal operation, which can distinguish the fault sound from the normal operation sound.

[0055] Figure 4 , Figure 5 , Figure 6The images show the front, top, and bottom views of the overall structure of the triboacoustic sensor. The sensor is placed in a power-function curved acoustic resonant cavity optimized according to the characteristic frequencies of the power equipment (e.g., 50-5000Hz), achieving frequency adaptive amplification. This aims to more effectively collect and concentrate the acoustic energy of the equipment onto the sensor diaphragm. Secondly, the resonant cavity acts as an acoustic impedance gradient, smoothly transmitting the acoustic signal to the sensor diaphragm surface, reducing sound wave reflection and improving acoustic energy transmission efficiency. Furthermore, the resonant cavity effectively suppresses interference from external environmental noise, resulting in a signal that better reflects the operating status of the equipment. Finite element acoustic simulation was used to design the resonant cavity cross-sectional function, increasing the sensor output voltage by more than 30%.

[0056] Figure 7 The workflow of the intelligent acoustic online monitoring system in this embodiment is as follows: During operation, the system collects sound signals from electrical equipment and converts these signals into electrical signals using a triboacoustic sensor. The resulting electrical signals are then transmitted wirelessly to devices such as mobile phones or computers via a wireless transmission module. This wireless transmission module utilizes a data acquisition card to collect signals and connects to the device via a wireless network. These signals are then analyzed. A short-time Fourier transform is used to convert the preprocessed time-domain signals into frequency-domain signals, and feature extraction is performed. Subsequently, a deep learning model is used to determine if the signal frequency is abnormal and to classify abnormal signals as faults. Finally, the monitoring system displays the fault type and issues an alarm signal. Furthermore, multiple self-powered sensor nodes are organized into a distributed wireless network, enabling group monitoring of electrical equipment without an external power source, significantly simplifying wiring and maintenance costs.

[0057] The basic principle of the intelligent acoustic online monitoring system is as follows: A triboacoustic sensor is installed on the surface of the sound source equipment. The sound generated during operation is transmitted to the triboacoustic sensor, and the resulting electrical signal is transmitted to the signal analysis and recognition module. This module includes a preprocessing module and a feature extraction module. A deep learning model is used to classify sound signals, automatically identifying normal and fault sounds. For typical operating conditions such as partial discharge and mechanical loosening of power equipment, a specific database model is established based on acoustic signature spectral characteristics. Finally, the intelligent acoustic online monitoring system performs fault classification and early warning: displaying a visual representation of the sound signal and classification results. Based on historical acoustic signature signals of power equipment in the database, including normal acoustic signature signals and acoustic signature signals corresponding to different types of faults, the system automatically determines whether a fault has occurred and identifies the type of fault, such as partial discharge (including surface discharge, metallic discharge, and suspension discharge) and mechanical loosening. If a fault occurs, an alarm device is activated to alert maintenance personnel for repairs.

[0058] Compared with existing technologies, this invention has the following advantages: First, by introducing multi-layer gradient damping membrane pads and porous surface structures into the triboelectric vibration sensor, and combining this with the synergistic design of a power function surface acoustic resonator, the sensor achieves energy focusing of the target signal and suppression of unrelated noise within the typical frequency bands of partial discharge and mechanical vibration in power equipment. Under the same test conditions, the sensor output voltage amplitude is significantly improved, and the signal-to-noise ratio is markedly enhanced. Second, the proposed signal processing method based on intrinsic mode function effectiveness index, adaptive bandpass filtering, and high- and low-frequency energy cross-band correlation coefficients can stably extract fault-related feature components under strong background noise and changing operating conditions, improving the retention of useful information in the signal preprocessing stage. Third, by employing a dual-encoder structure and a deep learning recognition model with coupled discriminant loss, the complementary information of the triboelectric vibration signal in both the time and frequency domains is fully explored, significantly improving the ability to distinguish between various types of partial discharge and mechanical faults. This invention enables long-term online monitoring and refined fault identification of power equipment without the need for external power supply, and has promising engineering application prospects.

[0059] Specifically, the fault type identification model in this application is an existing prediction model, such as Convolutional Neural Network (CNN), Transformer model, Long Short-Term Memory Network, etc.

[0060] In a specific embodiment, please refer to Figure 8 The diagram below shows the structure of a fault type identification system for power equipment according to the second embodiment of this application. The system includes: a dataset acquisition module 301, an empirical mode decomposition module 302, an effectiveness index calculation module 303, a reconstruction module 304, a power spectrum estimation module 305, a cross-band correlation coefficient acquisition module 306, a model training module 307, a dataset acquisition module 308, and a fault type output module 309. The dataset acquisition module 301 is used to acquire the triboelectric sensor electrical signal dataset of the power equipment; the triboelectric sensor electrical signal dataset includes triboelectric sensor electrical signals corresponding to n different fault types; The empirical mode decomposition module 302 is used to perform empirical mode decomposition on the triboelectric sensor electrical signal to obtain k IMF components of the triboelectric sensor electrical signal; The validity index calculation module 303 is used to calculate the validity index of each IMF component; the validity index is used to characterize the probability that the IMF component is valid. The reconstruction module 304 is used to reconstruct the IMF components in the triboelectric sensor electrical signal whose effectiveness index is greater than a preset first threshold, and obtain n reconstructed signals. The power spectrum estimation module 305 is used to perform power spectrum estimation on the reconstructed signal to obtain the center frequency of the reconstructed signal. The cross-band correlation coefficient acquisition module 306 is used to acquire the high-low frequency cross-band correlation coefficient of the triboelectric sensor electrical signal; The model training module 307 is used to construct training data from the effectiveness index, center frequency, and high-low frequency crossband correlation coefficient corresponding to the n triboelectric sensor electrical signals, use the fault type corresponding to the triboelectric sensor electrical signals as training labels, and use the training data and the training labels to train the preset fault type identification model to obtain the trained fault type identification model. The dataset acquisition module 308 is used to acquire the dataset to be identified of the triboelectric sensor electrical signal to be identified. The dataset to be identified includes the identification validity index, the identification center frequency, and the identification high-low frequency crossband correlation coefficient of the triboelectric sensor electrical signal to be identified. The fault type output module 309 is used to input the dataset to be identified into the trained fault type identification model to obtain the fault type corresponding to the electrical signal of the triboelectric sensor to be identified output by the trained fault type identification model.

[0061] In this embodiment, the system first acquires the IMF component of the electrical signal from each triboelectric sensor and calculates its effectiveness index. IMF components with effectiveness indices greater than a preset first threshold are then reconstructed. This process removes invalid information such as noise and electromagnetic interference, resulting in a higher-quality signal. The reconstructed signal is then subjected to power spectrum estimation to obtain the center frequency, and high-low frequency crossband correlation coefficients are also acquired, extracting features of the triboelectric sensing signal from multiple dimensions. These features, along with fault types, are used to construct training data and labels, enabling the model to comprehensively and accurately learn the features of different fault types. During identification, the dataset to be identified is input into the trained model, allowing for accurate fault type determination, significantly improving the accuracy and reliability of fault diagnosis.

[0062] In a specific embodiment, the third embodiment of this application provides a fault type identification device for power equipment, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method as described in any of the first embodiments of this application.

[0063] In a specific embodiment, the fourth embodiment of this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method as described in any one of the first embodiments of this application.

[0064] Figure 9 An internal structural diagram of a computer device in one embodiment is shown. This computer device can specifically be a terminal or a server. See also... Figure 9 The computer device includes a processor, memory, etc., connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program. When executed by the processor, this computer program causes the processor to implement the method described in this embodiment. The internal memory may also store a computer program, which, when executed by the processor, causes the processor to perform the method described in this embodiment. Those skilled in the art will understand that... Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0065] The above embodiments merely illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for identifying fault types in power equipment, characterized in that, The method includes: Acquire a dataset of triboelectric sensor signals from power equipment; the dataset includes triboelectric sensor signals corresponding to n different fault types. Empirical mode decomposition (EMF) is performed on the triboelectric sensor electrical signal to obtain k IMF components of the triboelectric sensor electrical signal; Calculate the validity index for each IMF component; the validity index is used to characterize the probability that the IMF component is valid; The IMF components in the triboelectric sensor electrical signal whose effectiveness index is greater than a preset first threshold are reconstructed to obtain n reconstructed signals. Power spectrum estimation is performed on the reconstructed signal to obtain the center frequency of the reconstructed signal; Obtain the high-low frequency crossband correlation coefficient of the triboelectric sensor's electrical signal; The effectiveness index, center frequency, and high-low frequency crossband correlation coefficient corresponding to the n triboelectric sensor electrical signals are used to construct training data. The fault type corresponding to the triboelectric sensor electrical signals is used as training label. The training data and the training label are used to train the preset fault type identification model to obtain the trained fault type identification model. Obtain the dataset to be identified of the triboelectric sensor electrical signal to be identified, wherein the dataset to be identified includes the identification validity index, the identification center frequency, and the identification high-low frequency crossband correlation coefficient of the triboelectric sensor electrical signal to be identified; The dataset to be identified is input into the trained fault type identification model to obtain the fault type corresponding to the triboelectric sensor electrical signal output by the trained fault type identification model.

2. The fault type identification method for power equipment as described in claim 1, characterized in that, The acquisition of the high-low frequency crossband correlation coefficient of the triboelectric sensor electrical signal includes: The triboelectric sensor's electrical signal is filtered and decomposed to obtain a first envelope energy sequence of the high-frequency band sensitive to partial discharge and a second envelope energy sequence of the low-frequency band sensitive to structural vibration, as well as a first time average value corresponding to the first envelope energy sequence and a second time average value corresponding to the second envelope energy sequence; The high-low frequency crossband correlation coefficient of each triboelectric sensor signal is calculated based on the first envelope energy sequence, the second envelope energy sequence, the first time average value, and the second time average value.

3. The fault type identification method for power equipment as described in claim 2, characterized in that, The high-low frequency crossband correlation coefficient is obtained using the following formula: in, The high-low frequency crossband correlation coefficient is given, and T is the total duration of the triboelectric sensor signal. The first envelope energy sequence, This is the average value over the first time period. This is the second envelope energy sequence. This is the second time average.

4. The fault type identification method for power equipment as described in claim 2, characterized in that, The method further includes: Obtain the first root mean square value, high-frequency band spectral entropy, first mean and first standard deviation of the high-frequency instantaneous frequency of the first envelope energy sequence; Obtain the second root mean square value, low-frequency band spectral entropy, and second mean and second standard deviation of the low-frequency instantaneous frequency of the second envelope energy sequence; The high-frequency energy ratio is obtained based on the first root mean square value and the second root mean square value; Based on the first root mean square value, the high-frequency band spectral entropy, the first mean, the first standard deviation, the second root mean square value, the low-frequency band spectral entropy, the second mean, the second standard deviation, and the constructed discriminative feature vector; The step of constructing training data from the effectiveness index, center frequency, and high-low frequency crossband correlation coefficient corresponding to the n triboelectric sensor electrical signals includes: The discriminative feature vectors, effectiveness indexes, center frequencies, and high-low frequency crossband correlation coefficients corresponding to the n triboelectric sensor electrical signals are used to construct training data. The dataset to be identified also includes the discriminative feature vector of the triboelectric sensor electrical signal to be identified.

5. The fault type identification method for power equipment as described in claim 1, characterized in that, The validity indicators for calculating each IMF component include: The first energy, kurtosis, and second energy of the triboelectric sensor electrical signal corresponding to the IMF component are obtained, and the Pearson linear correlation coefficient between the IMF component and its corresponding triboelectric sensor electrical signal is obtained. The effectiveness index of the IMF component is obtained based on the first energy, the kurtosis, the second energy, and the Pearson linear correlation coefficient.

6. The fault type identification method for power equipment as described in claim 5, characterized in that, The effectiveness indicators were obtained in the following manner: Obtain the energy ratio between the first energy and the second energy; The effectiveness index is obtained by weighted summing of the absolute values ​​of the energy ratio, the kurtosis, and the Pearson linear correlation coefficient.

7. The fault type identification method for power equipment as described in claim 1, characterized in that, When training the preset fault type identification model using the training data and the training labels, the discriminative coupling loss function is used to simultaneously compress the feature distance of the training data of the same fault type and expand the feature distance of the training data of different fault types.

8. A fault type identification system for power equipment, characterized in that, The system includes: a dataset acquisition module, an empirical mode decomposition module, an effectiveness index calculation module, a reconstruction module, a power spectrum estimation module, a cross-band correlation coefficient acquisition module, a model training module, a dataset acquisition module, and a fault type output module; The dataset acquisition module is used to acquire the triboelectric sensor signal dataset of power equipment; the triboelectric sensor signal dataset includes triboelectric sensor signals corresponding to n different fault types; The empirical mode decomposition module is used to perform empirical mode decomposition on the triboelectric sensor electrical signal to obtain k IMF components of the triboelectric sensor electrical signal; The validity index calculation module is used to calculate the validity index of each IMF component; the validity index is used to characterize the probability that the IMF component is valid. The reconstruction module is used to reconstruct the IMF components in the triboelectric sensor electrical signal whose effectiveness index is greater than a preset first threshold, and obtain n reconstructed signals. The power spectrum estimation module is used to perform power spectrum estimation on the reconstructed signal to obtain the center frequency of the reconstructed signal. The cross-band correlation coefficient acquisition module is used to acquire the high-low frequency cross-band correlation coefficient of the triboelectric sensor electrical signal; The model training module is used to construct training data from the effectiveness index, center frequency, and high-low frequency crossband correlation coefficient corresponding to the n triboelectric sensor electrical signals, use the fault type corresponding to the triboelectric sensor electrical signals as training labels, and use the training data and the training labels to train the preset fault type identification model to obtain the trained fault type identification model. The dataset acquisition module is used to acquire the dataset to be identified of the triboelectric sensor electrical signal to be identified. The dataset to be identified includes the identification validity index, the identification center frequency, and the identification high-low frequency crossband correlation coefficient of the triboelectric sensor electrical signal to be identified. The fault type output module is used to input the dataset to be identified into the trained fault type identification model to obtain the fault type corresponding to the electrical signal of the triboelectric sensor to be identified output by the trained fault type identification model.

9. A fault type identification device for power equipment, comprising a memory and a processor, characterized in that, The memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

Images