A method, system, device and medium for on-line monitoring of partial discharge of a generator

Abstract

The present application belongs to the field of power equipment state monitoring, and relates to a kind of generator partial discharge on-line monitoring method, system, equipment and medium, comprising the following process: obtaining four kinds of original signals of generator;Adaptive hierarchical denoising algorithm is used to extract partial discharge feature salient signal layer from original signal;Based on the feature salient signal layer of partial discharge, a feature quantity library is constructed, and the spectral barycenter and wavelet energy entropy of each type of original signal are extracted;The spectral barycenter and wavelet energy entropy of each type of original signal are weighted by fuzzy logic, and the primary fusion result of each type of original signal is obtained respectively;The primary fusion result of each type of original signal is deeply fused by using LSTM neural network to obtain the preliminary judgment result of partial discharge;The score of the preliminary judgment result of partial discharge is obtained based on the abnormality degree score of Mahalanobis distance, and if the score is greater than the decision threshold, partial discharge occurs.The generator partial discharge can be quickly and effectively judged, which is beneficial to the on-site maintenance personnel to respond to partial discharge.

Description

Technical Field

[0001] This invention belongs to the field of power equipment condition monitoring, and relates to a method, system, equipment and medium for online monitoring of partial discharge in generators. Background Technology

[0002] As the core power equipment of the power system, the insulation condition of generators directly affects the stable operation of the power grid. Approximately 42% of high-voltage generator failures originate from insulation degradation caused by partial discharge. Throughout the equipment's lifespan, factors such as manufacturing defects (e.g., air bubbles in epoxy resin casting), mechanical vibration stress during operation, material aging due to long-term temperature rise, and environmental moisture penetration gradually lead to various partial discharge phenomena. These include point discharge caused by metal burrs on the stator bar surface, surface discharge formed after the insulation layer becomes damp, air gap discharge caused by the infiltration of cooling media, and floating discharge caused by loose fasteners. Initially, these partial discharge phenomena are small in magnitude, but continuous discharge accelerates insulation carbonization, eventually leading to breakdown accidents. Currently, mainstream detection methods include ultrasonic methods, high-frequency methods, neutral point current methods, and transient ground voltage methods. Ultrasonic testing is widely used in online generator monitoring due to its strong resistance to electromagnetic interference, but it has significant limitations: insufficient reliability of single signals and susceptibility to mechanical vibration and noise interference; low efficiency of offline testing: the current IEC 60034-27 standard recommends offline partial discharge testing, which requires shutdown and high voltage application, taking an average of 8-12 hours per unit, seriously affecting power generation efficiency; difficulty in identifying defect types: existing technologies mostly rely on PRPD (Phase-Resolved Partial Discharge) spectrum analysis, requiring manual comparison of the phase distribution, amplitude statistics, and other characteristics of discharge pulses, which is highly dependent on the professional experience of operation and maintenance personnel. In addition, dynamic interference suppression and feature fusion decision-making under complex operating conditions remain technical challenges. Load fluctuations and temperature changes during generator operation can cause drift in the time-frequency characteristics of discharge signals, while existing threshold settings mostly use fixed values, which cannot adapt to environmental changes. Meanwhile, a single sensor is insufficient to fully characterize discharge features: ultrasonic signals reflect mechanical vibration effects, UHF (Ultra High Frequency) signals contain electromagnetic radiation information, and neutral point current can be correlated with discharge energy. The lack of multi-dimensional data fusion will lead to missed detection of key features. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, system, device and medium for online monitoring of generator partial discharge, which can quickly and effectively identify generator partial discharge, and help on-site maintenance personnel to react to partial discharge.

[0004] To achieve the above objectives, the present invention employs the following technical solution: A method for online monitoring of partial discharge in a generator includes the following steps: S1, acquire four types of raw signals from the generator, including ultrasonic signals, neutral point current signals, voltage phase signals and ultra-high frequency current signals; S2 uses an adaptive hierarchical denoising algorithm to extract the signal layer with significant partial discharge characteristics from the original signal; S3, construct a feature library based on the signal layer with significant partial discharge characteristics, and extract the spectral centroid and wavelet energy entropy of each type of original signal from the feature library; S4. The spectral centroid and wavelet energy entropy of each type of original signal are weighted by fuzzy logic to obtain the primary fusion result of each type of original signal. S5. The LSTM neural network is used to perform deep fusion of the primary fusion results of each type of original signal to obtain the preliminary judgment result of partial discharge. S6. The score of the preliminary judgment result of partial discharge is obtained based on the anomaly score of Mahalanobis distance. The judgment result of partial discharge is determined by the relationship between the score and the preset decision threshold. If the score is less than or equal to the decision threshold, no partial discharge has occurred. If the score is greater than the decision threshold, partial discharge has occurred.

[0005] Preferably, after acquiring the four types of original signals from the generator, the method further includes: using a clock synchronization compensation method to perform time compensation on the four types of original signals, so that the time of each type of original signal is aligned.

[0006] Preferably, the specific process of extracting the signal layer with significant partial discharge characteristics from the original signal using the adaptive hierarchical denoising algorithm is as follows: filtering the original signal; adaptively selecting the signal layer with significant partial discharge characteristics from the filtered original signal based on the VMD decomposition layer number of the kurtosis index; and performing wavelet threshold denoising on the signal layer with significant partial discharge characteristics using a threshold function.

[0007] Preferably, after constructing the feature library, the spectral centroid and wavelet energy entropy of each type of original signal are extracted by analyzing the time-domain features, frequency-domain features, and time-frequency features of each type of original signal in the feature library.

[0008] Preferably, the specific process of fuzzy logic weighting of the spectral centroid and wavelet energy entropy of each type of original signal is as follows: determine the three-dimensional variables of each type of original signal; wherein, the three-dimensional variables include signal confidence, feature stability and historical matching degree; and perform fuzzy logic weighting on the spectral centroid and wavelet energy entropy of each type of original signal according to the interference effect of the three-dimensional variables on the original signal.

[0009] Preferably, the decision threshold is generated as follows: after setting the initial decision threshold, a historical database including various operating parameters is established. Based on the operating parameters in the historical database, the initial decision threshold is dynamically corrected through a Bayesian network to obtain the final decision threshold.

[0010] Preferably, after step S6, the method further includes: comparing the actual partial discharge state with the final judgment result of the partial discharge; when the judgment result of the partial discharge does not correspond to the actual partial discharge state, marking the current partial discharge judgment result as an abnormal data sample; and optimizing the decision threshold based on the incremental learning algorithm when any of the preset conditions are met. The preset conditions include: no abnormal data samples are generated in a certain quarter, a cumulative increase of 500 sets of abnormal data samples, and the number of times the partial discharge judgment result is manually corrected is greater than 10 times / month.

[0011] An online monitoring system for partial discharge of a generator, comprising: The signal acquisition module is used to acquire four types of raw signals from the generator, including ultrasonic signals, neutral point current signals, voltage phase signals, and ultra-high frequency current signals. The signal layer extraction module is used to extract the signal layer with significant partial discharge characteristics from the original signal using an adaptive hierarchical denoising algorithm. The feature extraction module is used to construct a feature library based on the significant signal layer of partial discharge features, and to extract the spectral centroid and wavelet energy entropy of each type of original signal from the feature library; The fuzzy logic weighting module is used to perform fuzzy logic weighting on the spectral centroid and wavelet energy entropy of each type of original signal to obtain the primary fusion result of each type of original signal. The preliminary judgment module is used to perform deep fusion of the primary fusion results of each type of original signal using an LSTM neural network to obtain the preliminary judgment result of partial discharge. The final judgment module is used to obtain a preliminary judgment result of partial discharge based on the anomaly score of Mahalanobis distance. The judgment result of partial discharge is determined by the relationship between the score and the preset decision threshold. If the score is less than or equal to the decision threshold, no partial discharge has occurred. If the score is greater than the decision threshold, partial discharge has occurred.

[0012] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the online monitoring method for partial discharge of a generator.

[0013] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the online monitoring method for partial discharge of a generator.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention employs an adaptive hierarchical denoising algorithm to separate high-frequency noise and effective discharge pulses in a signal, solving the problems of discontinuous hard thresholding and constant deviation in soft thresholding. Preliminary fusion is achieved through fuzzy logic weighting, followed by deep fusion using an LSTM (Long Short-Term Memory) neural network. Finally, anomaly scoring based on Mahalanobis distance is performed to obtain the final judgment result. This facilitates rapid and effective identification of generator partial discharge by on-site maintenance personnel, enabling them to react more quickly to partial discharges. Attached Figure Description

[0015] Figure 1 This is a flowchart of the online monitoring process for partial discharge of a generator according to Embodiment 1 of the present invention. Detailed Implementation

[0016] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0017] Example 1: like Figure 1 As shown, this embodiment provides a method for online monitoring of partial discharge in a generator, including the following process: S1 acquires four types of raw signals from the generator in real time: ultrasonic signal, neutral point current signal, voltage phase signal, and ultra-high frequency current signal.

[0018] S2 uses an adaptive hierarchical denoising algorithm to extract a signal layer with significant partial discharge characteristics from the original signal.

[0019] S3 constructs a feature library based on the signal layer with significant partial discharge characteristics, and extracts the spectral centroid and wavelet energy entropy of each type of original signal from the feature library.

[0020] S4. The spectral centroid and wavelet energy entropy of each type of original signal are weighted by fuzzy logic to obtain the primary fusion result of each type of original signal.

[0021] S5 uses an LSTM neural network to perform deep fusion of the primary fusion results of each type of original signal to obtain preliminary judgment results of partial discharge.

[0022] S6. The score of the preliminary judgment result of partial discharge is obtained based on the anomaly score of Mahalanobis distance. The judgment result of partial discharge is determined by the relationship between the score and the preset decision threshold. If the score is less than or equal to the decision threshold, no partial discharge has occurred. If the score is greater than the decision threshold, partial discharge has occurred.

[0023] Example 2: This embodiment provides a method for online monitoring of partial discharge in a generator, including the following process: I. Continuous monitoring is carried out using a multi-dimensional online monitoring system.

[0024] The multi-dimensional online monitoring system uses a GPS (Global Positioning System) clock to achieve μs-level time alignment and synchronously collects the following four types of signals as raw signals: 1. Ultrasonic signals generated by partial discharge are collected using an ultrasonic sensor with an operating frequency band of 40-300kHz; 2. Neutral point current signals are collected using a Rogowski coil with an accuracy of at least ±0.5%; 3. Voltage phase signals are collected using a capacitive voltage divider; 4. Ultra-high frequency current signals are collected using a UHF sensor with an operating frequency band of 300MHz-3GHz. The system is powered by electromagnetic induction to achieve 24 / 7 uninterrupted monitoring.

[0025] When the multi-dimensional online monitoring system performs synchronous data acquisition, a clock synchronization compensation method is used to compensate for the time of the acquired signals, ensuring time alignment of data from multiple sensors and accurately associating the multi-dimensional characteristics of the same discharge event. The specific method is as follows:

[0026] Where t correct Indicates the corrected timestamp, Δ cable This indicates the difference in cable transmission delay (such as the difference in wiring length between a UHF sensor and a current transformer, with synchronization accuracy <100ns to ensure accurate phase resolution), t local t represents the raw timestamp recorded by the local sensor. master The time base of the master clock, t slave This indicates the timestamp reported by the subordinate sensor. This represents half the compensation for the time difference between master and slave.

[0027] II. Adaptive hierarchical denoising algorithm.

[0028] An adaptive hierarchical denoising algorithm is used to extract the signal layer with significant partial discharge characteristics from the original signal. The adaptive hierarchical denoising algorithm includes the following steps: (The algorithm employs improved wavelet packet transform combined with variational mode decomposition.) Step 1: Implement hardware filtering of the original signal using a bandpass filter bank; Step 2: Adaptively select the signal layer with significant partial discharge characteristics based on the VMD (Variational Mode Decomposition) decomposition layer number according to the kurtosis index: The original signal... Decomposed into K modal components with different center frequencies ωk It effectively separates high-frequency noise from effective discharge pulses. By comparing the signal decomposition with that under normal conditions, if a significant pulse change occurs, it is considered that a partial discharge has occurred in that layer, and the signal of that layer is extracted. In step 2, the adaptive selection of the VMD decomposition layer number based on the kurtosis index is as follows:

[0029]

[0030] Modal components ω is the k-th modal component, representing partial discharge components of different frequencies in the signal (such as high-frequency pulses of tip discharge or low-frequency oscillations of surface discharge). k It is the center frequency of the kth mode, which identifies the main frequency band of the discharge pulse and is used to distinguish the discharge type (such as air gap discharge concentrated in 100–300kHz). K is the original signal, and K is the adaptive decomposition level, which is dynamically adjusted according to the signal kurtosis index to ensure efficient separation of noise and effective signal under complex working conditions. K∈[3,8].

[0031] Step 3: By improving the threshold function, wavelet threshold denoising is performed on the signal layer with significant partial discharge characteristics to solve the problems of discontinuity of traditional hard threshold and constant deviation of soft threshold, so as to improve the signal-to-noise ratio. The threshold function with smooth transition suppresses noise and preserves the steep edge characteristics of the discharge pulse.

[0032] In step 3, the improved threshold function is implemented to achieve wavelet threshold denoising as follows:

[0033] in α is the threshold function, x is the wavelet coefficient, α is the smoothing factor, which controls the slope of the transition region to avoid signal distortion caused by traditional hard thresholding or energy loss of soft thresholding. The wavelet threshold smoothing factor α∈[3,7], and e is the natural constant.

[0034]

[0035] Where σ is the noise standard deviation, N is the signal length, the threshold τ is adaptively adjusted according to the noise standard deviation σ and the signal length N to ensure effective noise reduction in strong noise environments, and ln is the logarithm.

[0036] III. Multi-source feature fusion decision model.

[0037] A multi-source feature fusion decision model is used to perform four types of signal fusion judgment on the partial discharge salient signal layer to obtain the partial discharge judgment result. The multi-source feature fusion decision model includes the following steps: Step 1: Construct a feature library based on the significant partial discharge signal layer. By analyzing the time-domain, frequency-domain, and time-frequency features of the four types of signals in the feature library, extract the spectral centroid and wavelet energy entropy of the four types of signals.

[0038] In step 1, the extraction of spectral centroid and wavelet energy entropy features is as follows:

[0039] SC (Spectral Center) represents the center frequency index of the signal energy distribution in the frequency domain, and is used to quantify the "center position" of the signal energy in the spectrum. f represents the signal spectrum. k The frequency component k is represented by the centroid of the spectrum, which characterizes the center frequency of the energy distribution of the partial discharge signal. The discharge type is distinguished by the difference between the high-frequency centroid and the low-frequency centroid. For example, the SC value is usually higher for tip discharge due to the steep pulse, while the SC value is usually lower for surface discharge due to oscillation attenuation.

[0040]

[0041] J represents the decomposition level, E represents the wavelet energy entropy, and E j E represents the frequency band energy corresponding to the wavelet coefficients of the j-th layer. total It represents the total energy of the signal across all frequency bands, i.e., the global energy benchmark. It quantifies the complexity of the energy distribution of the signal in different frequency bands, reflects the randomness of the discharge mode, and distinguishes different discharge types by high and low entropy values. For example, air gap discharge has a high entropy value because random bubble breakdown results in multi-frequency band energy dispersion; while suspension discharge has a low entropy value because periodic arcs are concentrated in a specific frequency band.

[0042] Step 2 involves fuzzy logic weighting of the spectral centroid and wavelet energy entropy of each type of original signal to obtain the primary fusion result for each type of original signal. The fuzzy logic weighting uses three-dimensional variables for each type of original signal: signal confidence (calculated based on signal-to-noise ratio, reflecting signal quality; for example, the confidence of UHF signals decreases under high electromagnetic interference scenarios), feature stability (assessing the volatility of feature values ​​through sliding window variance analysis to eliminate transient interference), and historical matching degree (comparing typical discharge patterns in the historical database to identify abnormal features). By comprehensively considering the interference effects of signal confidence, feature stability, and historical matching degree on the original signal, fuzzy logic weighting is applied to the spectral centroid and wavelet energy entropy of each type of original signal to achieve dynamic weight allocation and thus optimize decision-making. In step 2, the fuzzy logic weighted fusion is performed as follows:

[0043] W i The fusion weights for the i-th type of signal are dynamically adjusted to determine the contribution of each signal to the final decision. μC i Let μS be the membership function for the signal confidence level. i For characteristic stability membership function, μH i Here, (x) represents the historical matching degree membership function, which is used to dynamically evaluate signal characteristics or system state.

[0044] Step 3: Use an LSTM neural network to perform deep fusion of the primary fusion results of different signals to obtain preliminary judgment results of partial discharge. The LSTM neural network is used to capture the temporal correlation of partial discharge signals (such as the periodicity and suddenness of discharge pulses) and improve the prediction ability of insulation defect evolution.

[0045] In step 3, the state update of the LSTM temporal feature modeling unit is as follows:

[0046]

[0047]

[0048]

[0049]

[0050]

[0051] f tσ represents the forgetting gate, controlling the degree of retention of historical memories (e.g., ignoring background noise during steady-state operation), and σ represents the Sigmoid activation function, which compresses the linear transformation result to [0, 1], controlling the degree of forgetting. f This represents the forget gate weight matrix for input x. t and the previous hidden state h t 1. Perform feature fusion. [h] t 1,x t The ] symbol represents the concatenation operation between the hidden state and the input: the hidden state h from the previous time step is concatenated. t 1 and the current input x t By concatenating the elements along the dimensions, a comprehensive feature vector is formed. f This represents the forget gate bias term, which adjusts the baseline value of the linear transformation, enhancing the flexibility of the multi-source feature fusion decision model; t This represents the input gate, which filters effective features related to discharge in the current input (such as the rise slope of an ultra-high frequency pulse), h t 1 indicates the hidden state of the previous time step, carrying temporal accumulation characteristics (such as the amplitude decay pattern of historical discharge pulses), x t W represents the current input feature vector, which may contain real-time sensor data (such as UHF signal amplitude and phase spectrum entropy). i The matrix represents a dynamic feature selection matrix, which learns the weight assignments for different features through training. i This indicates adaptive bias, controlling the gate activation threshold (e.g., setting a baseline noise level, allowing information to flow in only when the pulse energy exceeds the threshold). Represents the candidate memory at the current moment (the original update amount not merged with historical memory), tanh represents the numerical characteristics, i.e., the output range [-1, 1], the centrosymmetric gradient, and W. C Let b represent the weight matrix. C This indicates that the activation threshold is adjusted by an offset, x t h represents the sensor input characteristics at the current moment. t 1 represents the hidden state (short-term memory carrier) of the previous time step, [h t 1,x t [] indicates a concatenation operation, combining historical information with the current input; C t Indicates the current cell state, used to store accumulated information from the beginning of the sequence to the current time. (C) t 1 represents the historical unit state, i.e., the global temporal characteristics up to time t-1; o tThe output gate outputs the timing modeling results, representing the changing trend of discharge severity; W o Let b represent the weight matrix. o This represents the bias vector, where each neuron independently adjusts its activation threshold, h. t Represents the hidden state, ⊙ represents the Hadamard product, and tanh(C) represents the hidden state. t ) indicates that the cell state C t Compressed to the range [-1, 1], the number of hidden nodes in the LSTM layer is [32, 128].

[0052] Step 4, perform secondary judgment of partial discharge based on the anomaly score of Mahalanobis distance: the score of the preliminary judgment result of partial discharge is obtained based on the anomaly score of Mahalanobis distance to measure the degree of deviation of the current feature vector x (spectral centroid and wavelet energy entropy) from the historical normal state (mean μ and covariance Σ) (score). When the deviation exceeds the preset decision threshold, specifically: when the scores of 3 / 4 of the 4 types of signals exceed the decision threshold, it is considered that partial discharge has occurred, and the partial discharge judgment result is output.

[0053] In step 4, the anomaly score based on Mahalanobis distance is as follows:

[0054] Among them, D 2 Let Σ denote the squared generalized distance from a data point x to the center μ of the dataset in multidimensional space, expressed by the covariance matrix Σ. 1 The features are standardized and decorrelated, x∈R 4 Let μ represent the current feature vector, where μ ∈ R. 4 Represents the sliding window mean (historical 24 hours), Σ∈R 4×4 This represents the covariance matrix. μ and Σ are updated every 24 hours to adapt to changes in generator load, temperature, and other operating conditions.

[0055] IV. Decision Threshold Adjustment Mechanism.

[0056] A decision threshold adjustment mechanism is adopted to dynamically correct the decision threshold. The decision threshold adjustment mechanism includes the following steps: Step 1: Preset the initial decision threshold and establish a historical database, including operating parameters such as ambient temperature, ambient humidity, and load rate.

[0057] Step 2: Based on the operating parameters in the historical database, the initial decision threshold is dynamically adjusted using a Bayesian network to obtain the final decision threshold.

[0058] In step 2, the Bayesian network dynamically adjusts the decision threshold by using P(T|L, N) to reduce false alarms caused by environmental changes, as detailed below:

[0059] P represents a probability function, the specific meaning and type of which vary depending on the combination of variables in the parentheses. T represents the decision threshold, L represents the load rate (%), N represents the noise baseline (dB), and E represents the ambient temperature and humidity.

[0060] V. Optimization of the self-learning model.

[0061] A self-learning model is used to optimize the decision threshold. This optimization includes establishing a closed-loop feedback mechanism and continuously optimizing the parameters of the multi-source feature fusion decision model using newly added data to adapt to generator insulation aging or new defect modes. The steps include: Step 1: Compare the actual partial discharge state with the partial discharge judgment result. When a false alarm is manually confirmed, that is, when the partial discharge judgment result does not correspond to the actual partial discharge state, the current partial discharge judgment result is automatically marked as an abnormal data sample.

[0062] Step 2: When any one of the following conditions is met: no abnormal data samples were generated in the first quarter, 500 new abnormal data sets were added cumulatively, and the number of manual corrections was greater than 10 times per month, the parameters (decision threshold) of the multi-source feature fusion decision model are optimized based on the incremental learning algorithm.

[0063] In step 2, the specific process of optimizing the parameters of the multi-source feature fusion decision model based on the incremental learning algorithm is as follows:

[0064] Where θ represents the parameters of the multi-source feature fusion decision model, θ t θ represents the parameter vector of the multi-source feature fusion decision model at the current time (step t). t+1 This represents the updated parameter vector, calculated based on the current gradient and historical parameter changes. This represents the loss function with respect to the parameter θ. t The gradient vector, η=0.001, represents the learning rate, λ=0.5, represents the momentum factor, which prevents the model from deviating from the learned general features due to short-term data fluctuations; L represents the loss function.

[0065] This embodiment improves the accuracy of multi-channel synchronization error and clock drift compensation by synchronously acquiring four types of signals and performing time compensation. A three-layer hardware-jointly improved wavelet packet transform and VMD (Variational Mode Decomposition) are employed to separate high-frequency noise and effective discharge pulses in the signal, solving the problems of discontinuity in traditional hard thresholding and constant deviation in soft thresholding. The extracted features are fused using a fuzzy logic weighting method to measure signal confidence, feature stability, and historical matching degree. Deep fusion is then achieved through LSTM (Long Short-Term Memory) time-series modeling, and finally, Mahalanobis distance is used for anomaly scoring. The decision threshold is revised based on Bayesian threshold judgment, and an early warning is triggered when the abnormal scores of 3 / 4 signal channels exceed the decision threshold. On this basis, adaptive incremental optimization learning is completed. When a false alarm is manually confirmed, abnormal data samples are automatically marked. When any one of the following conditions is met, a parameter optimization of the multi-source feature fusion decision model based on incremental learning algorithm is triggered. This not only makes it easier for on-site operation and maintenance personnel to quickly judge generator partial discharge, but also allows for continuous optimization to improve the accuracy of partial discharge identification.

[0066] Example 3: In this embodiment, an online monitoring system for partial discharge of a generator is provided. This online monitoring system for partial discharge of a generator can be used to implement the above-mentioned online monitoring method for partial discharge of a generator. Specifically, the online monitoring system for partial discharge of a generator includes a signal acquisition module, a signal layer extraction module, a feature extraction module, a fuzzy logic weighting module, a preliminary judgment module, and a final judgment module.

[0067] The signal acquisition module is used to acquire four types of raw signals from the generator, including ultrasonic signals, neutral point current signals, voltage phase signals, and ultra-high frequency current signals.

[0068] The signal layer extraction module is used to extract the signal layer with significant partial discharge characteristics from the original signal using an adaptive hierarchical denoising algorithm.

[0069] The feature extraction module is used to construct a feature library based on the significant signal layer of partial discharge characteristics, and to extract the spectral centroid and wavelet energy entropy of each type of original signal from the feature library.

[0070] The fuzzy logic weighting module is used to perform fuzzy logic weighting on the spectral centroid and wavelet energy entropy of each type of original signal to obtain the primary fusion result of each type of original signal.

[0071] The preliminary judgment module is used to perform deep fusion of the primary fusion results of each type of original signal using an LSTM neural network to obtain the preliminary judgment result of partial discharge.

[0072] The final judgment module is used to obtain a preliminary judgment result of partial discharge based on the anomaly score of Mahalanobis distance. The judgment result of partial discharge is determined by the relationship between the score and the preset decision threshold. If the score is less than or equal to the decision threshold, no partial discharge has occurred; if the score is greater than the decision threshold, partial discharge has occurred.

[0073] Example 4: In this embodiment, a terminal device is provided, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment can be used in the operation of an online monitoring method for partial discharge of a generator, including: S1, acquiring four types of raw signals from the generator, including ultrasonic signals, neutral point current signals, voltage phase signals, and ultra-high frequency current signals; S2, extracting a signal layer with significant partial discharge characteristics from the raw signals using an adaptive hierarchical denoising algorithm; S3, based on... A feature library is constructed at the signal layer with significant partial discharge characteristics. The spectral centroid and wavelet energy entropy of each type of original signal are extracted from the feature library. In step S4, the spectral centroid and wavelet energy entropy of each type of original signal are weighted by fuzzy logic to obtain the primary fusion result of each type of original signal. In step S5, the primary fusion result of each type of original signal is deeply fused using an LSTM neural network to obtain the preliminary judgment result of partial discharge. In step S6, the anomaly score of the preliminary judgment result of partial discharge is obtained based on the Mahalanobis distance. The judgment result of partial discharge is determined by the relationship between the score and the preset decision threshold. If the score is less than or equal to the decision threshold, no partial discharge has occurred. If the score is greater than the decision threshold, partial discharge has occurred.

[0074] Example 5: This embodiment provides a computer-readable storage medium (Memory), which is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk storage device.

[0075] One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the online monitoring method for partial discharge of generators in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps: S1, acquiring four types of raw signals from the generator, including ultrasonic signals, neutral point current signals, voltage phase signals, and ultra-high frequency current signals; S2, extracting a layer of signals with significant partial discharge characteristics from the raw signals using an adaptive hierarchical denoising algorithm; S3, constructing a feature library based on the layer of signals with significant partial discharge characteristics, and extracting signals from the feature library. S4. Take the spectral centroid and wavelet energy entropy of each type of original signal; S5. Perform fuzzy logic weighting on the spectral centroid and wavelet energy entropy of each type of original signal to obtain the primary fusion result of each type of original signal; S6. Use an LSTM neural network to perform deep fusion on the primary fusion result of each type of original signal to obtain the preliminary judgment result of partial discharge; S7. Obtain the score of the preliminary judgment result of partial discharge based on the anomaly score of Mahalanobis distance, and determine the judgment result of partial discharge by the relationship between the score and the preset decision threshold; where, if the score is less than or equal to the decision threshold, no partial discharge has occurred, and if the score is greater than the decision threshold, partial discharge has occurred.

[0076] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0077] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0078] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0079] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0080] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0081] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

[0082] It should be understood that the above description is for illustrative purposes and not for limitation. Many embodiments and applications beyond the examples provided will become apparent to those skilled in the art upon reading the above description.

Claims

1. A method for online monitoring of partial discharge in a generator, characterized in that, The process includes the following: S1, acquire four types of raw signals from the generator, including ultrasonic signals, neutral point current signals, voltage phase signals and ultra-high frequency current signals; S2 uses an adaptive hierarchical denoising algorithm to extract the signal layer with significant partial discharge characteristics from the original signal; S3, construct a feature library based on the signal layer with significant partial discharge characteristics, and extract the spectral centroid and wavelet energy entropy of each type of original signal from the feature library; S4. The spectral centroid and wavelet energy entropy of each type of original signal are weighted by fuzzy logic to obtain the primary fusion result of each type of original signal. The specific process is as follows: Each type of original signal is given a three-dimensional variable; the three-dimensional variable includes signal confidence, feature stability, and historical matching degree. Based on the interference effect of the three-dimensional variables on the original signal, fuzzy logic weighting is applied to the spectral centroid and wavelet energy entropy of each type of original signal. S5. The LSTM neural network is used to perform deep fusion on the primary fusion results of each type of original signal to obtain the preliminary judgment result of partial discharge of each type of original signal. S6. Based on the preliminary judgment result of partial discharge for each type of original signal, calculate the anomaly score of Mahalanobis distance to obtain the score of the preliminary judgment result of partial discharge for each type of original signal. Determine the judgment result of partial discharge by the relationship between the score and the preset decision threshold. If the score is less than or equal to the decision threshold, no partial discharge has occurred. If the score is greater than the decision threshold, partial discharge has occurred. When the scores of 3 / 4 of the 4 types of signals exceed the decision threshold, partial discharge is considered to have occurred, and the partial discharge judgment result is output.

2. The online monitoring method for partial discharge of a generator according to claim 1, characterized in that, After acquiring the four types of raw signals from the generator, the process also includes: using a clock synchronization compensation method to perform time compensation on the four types of raw signals, so that the time of each type of raw signal is aligned.

3. The online monitoring method for partial discharge of a generator according to claim 1, characterized in that, The specific process of extracting the signal layer with significant partial discharge characteristics from the original signal using the adaptive hierarchical denoising algorithm is as follows: The original signal is filtered; based on the VMD decomposition layer number of the kurtosis index, the signal layer with significant partial discharge characteristics is adaptively selected from the filtered original signal; wavelet threshold denoising is performed on the signal layer with significant partial discharge characteristics through a threshold function.

4. The online monitoring method for partial discharge of a generator according to claim 1, characterized in that, After constructing the feature library, the spectral centroid and wavelet energy entropy of each type of original signal are extracted by analyzing the time-domain, frequency-domain, and time-frequency characteristics of each type of original signal in the feature library.

5. The online monitoring method for partial discharge of a generator according to claim 1, characterized in that, The decision threshold is generated as follows: An initial decision threshold is preset, and a historical database including various operating parameters is established. Based on the operating parameters in the historical database, the initial decision threshold is dynamically corrected through a Bayesian network to obtain the final decision threshold.

6. The online monitoring method for partial discharge of a generator according to claim 1, characterized in that, After step S6, the following is also included: The actual partial discharge state is compared with the partial discharge judgment result. When the partial discharge judgment result does not correspond to the actual partial discharge state, the current partial discharge judgment result is marked as an abnormal data sample. When any of the preset conditions are met, the decision threshold is optimized based on the incremental learning algorithm; the preset conditions include: no abnormal data samples are generated in a certain quarter, 500 new abnormal data samples are added cumulatively, and the number of times the partial discharge judgment result is manually corrected is greater than 10 times / month.

7. A generator partial discharge online monitoring system, characterized in that, include: The signal acquisition module is used to acquire four types of raw signals from the generator, including ultrasonic signals, neutral point current signals, voltage phase signals, and ultra-high frequency current signals. The signal layer extraction module is used to extract the signal layer with significant partial discharge characteristics from the original signal using an adaptive hierarchical denoising algorithm. The feature extraction module is used to construct a feature library based on the significant signal layer of partial discharge features, and to extract the spectral centroid and wavelet energy entropy of each type of original signal from the feature library; The fuzzy logic weighting module is used to perform fuzzy logic weighting on the spectral centroid and wavelet energy entropy of each type of original signal to obtain the primary fusion result for each type of original signal. The specific process is as follows: Each type of original signal is given a three-dimensional variable; the three-dimensional variable includes signal confidence, feature stability, and historical matching degree. Based on the interference effect of the three-dimensional variables on the original signal, fuzzy logic weighting is applied to the spectral centroid and wavelet energy entropy of each type of original signal. The preliminary judgment module is used to perform deep fusion of the primary fusion results of each type of original signal using an LSTM neural network to obtain the preliminary judgment result of partial discharge for each type of original signal. The final judgment module is used to calculate the anomaly score of Mahalanobis distance based on the preliminary judgment result of partial discharge for each type of original signal, so as to obtain the score of the preliminary judgment result of partial discharge for each type of original signal. The judgment result of partial discharge is determined by the relationship between the score and the preset decision threshold. If the score is less than or equal to the decision threshold, no partial discharge has occurred; if the score is greater than the decision threshold, partial discharge has occurred. When the scores of 3 / 4 of the 4 types of signals exceed the decision threshold, partial discharge is considered to have occurred, and the partial discharge judgment result is output.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the online monitoring method for partial discharge of generators as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the online monitoring method for partial discharge of generators as described in any one of claims 1 to 6.

Images