Online monitoring and diagnosing system and method for mechanical instability of 500kV oil-immersed current limiting reactor

Abstract

The embodiment of the invention provides a 500kV oil-immersed current-limiting reactor mechanical instability online monitoring and diagnosing system and method, and the method comprises the steps: a high-precision synchronous collection unit synchronously collects multi-point vibration signals and current signals through high-precision time, and a signal processing unit carries out the signal processing and feature extraction of the multi-point vibration signals, and obtains vibration mode features; the current signal processing and feature extraction are carried out on the current signal to obtain a current feature, the correlation analysis unit establishes and continuously analyzes a dynamic correlation relationship between the vibration mode feature and the current feature in a time sequence, and the diagnosis and early warning unit analyzes a result based on the vibration mode feature, the current feature and the dynamic correlation relationship. And performing state evaluation, early warning and fault diagnosis on the mechanical stability of the reactor by adopting a preset diagnosis model. According to the method, the limitation of traditional single sensor monitoring is broken through, the diagnosis accuracy is improved, the false alarm rate and the missing report rate are reduced, the operation and maintenance management level of equipment is improved, and safe and stable operation of a power grid is guaranteed.

Description

Technical Field

[0001] This application relates to the field of power equipment condition monitoring and fault diagnosis technology, and in particular to an online monitoring and diagnosis system and method for mechanical instability of a 500kV oil-immersed current-limiting reactor. Background Technology

[0002] In modern high-voltage power systems, 500kV oil-immersed current-limiting reactors are core equipment for limiting short-circuit currents, and their operational stability directly affects the safety of the power grid. These reactors typically employ a large core with a large air gap design to withstand high electromagnetic stress, but this results in high-intensity inherent vibrations (such as resonance in the core, windings, and clamping components) far exceeding those of conventional equipment. During long-term operation, critical internal structural components (such as core laminations, clamping components, and winding clamping parts) are prone to loosening, displacement, or deformation due to electromagnetic forces, thermal expansion and contraction, and mechanical fatigue, leading to mechanical instability. Mechanical instability not only triggers electrical faults such as multi-point grounding of the core, partial discharge, and abnormal gas in the oil, but may also accelerate insulation aging through a fault chain effect, ultimately leading to equipment damage or power grid accidents.

[0003] Existing reactor condition monitoring technologies typically include vibration monitoring, dissolved gas analysis (DGA), partial discharge (PD) monitoring, and core grounding current monitoring. However, these technologies have limitations when applied to high-inherent-vibration, large-capacity equipment such as 500kV oil-immersed current-limiting reactors. Traditional vibration monitoring usually relies on monitoring the vibration amplitude or specific frequency components at a limited number of points outside the tank. For equipment with large inherent vibration amplitudes, this method struggles to effectively distinguish between high vibrations under normal operating conditions and abnormal vibration modes indicating internal mechanical instability, easily leading to false alarms or missed alarms. Furthermore, the oil immersion environment and the heavy tank attenuate and couple vibration signals from the outer surface, making it difficult to accurately reflect the true, localized vibration state and relative displacement of critical internal components. Traditional dissolved gas analysis has long cycles, making minute-level response difficult. While isolated monitoring of core grounding current or partial discharge signals can detect electrical anomalies, it is difficult to trace whether they are caused by mechanical instability, nor can it capture early warning signs where mechanical changes precede electrical anomalies.

[0004] Therefore, the above methods are insufficient to provide early, highly sensitive warnings and accurate diagnoses of mechanical instability faults in this type of first-of-its-kind equipment, and cannot meet the requirements for its safe, efficient operation and condition-based maintenance. Summary of the Invention

[0005] This application provides an online monitoring and diagnostic system and method for mechanical instability of a 500kV oil-immersed current-limiting reactor, which solves the problem that existing technologies are unable to provide early, highly sensitive warnings and accurate diagnoses of mechanical instability faults in reactor equipment, and thus cannot meet the requirements for safe, efficient operation and condition-based maintenance.

[0006] In a first aspect, embodiments of this application provide an online monitoring and diagnostic system for mechanical instability of a 500kV oil-immersed current-limiting reactor, comprising:

[0007] The sensor module and the monitoring host include a high-precision synchronous acquisition unit, a signal processing unit, a feature extraction unit, a correlation analysis unit, and a diagnostic and early warning unit.

[0008] The sensor module includes a vibration sensor and a wideband current sensor;

[0009] The high-precision synchronous acquisition unit is connected to the sensor module and is used to acquire multi-point vibration signals and current signals through the sensor module and send them to the signal processing unit.

[0010] The signal processing unit is connected to the high-precision synchronous acquisition unit and is used to preprocess the multi-point vibration signal and the current signal;

[0011] The feature extraction unit is connected to the signal processing unit and is used to extract vibration mode features from the processed multi-point vibration signal and extract current features from the processed current signal.

[0012] The correlation analysis unit is connected to the feature extraction unit and is used to establish and analyze the dynamic correlation between the vibration mode features and the current features;

[0013] The diagnostic early warning unit is connected to the correlation analysis unit and is used to perform state assessment, early warning and fault diagnosis of the mechanical stability of the reactor based on the vibration mode characteristics, the current characteristics and the dynamic correlation relationship.

[0014] In one possible implementation, the vibration sensor is a vibration acceleration sensor, and there are no fewer than 8 channels. They are arranged on the surface of the reactor's tank body at locations corresponding to the distribution of the core column, winding, or clamp, and are installed and fixed using a magnetic attraction method.

[0015] In one possible implementation, the broadband current sensor is a non-contact, open-close type broadband CT sensor, installed on the core grounding flat iron or clamp grounding flat iron of the reactor, for measuring current signals with a bandwidth ranging from power frequency to at least megahertz.

[0016] In one possible implementation, the system further includes a storage module and a communication interface, the storage module being used to store monitoring data, and the monitoring host communicating with external devices through the communication interface.

[0017] In one possible implementation, the monitoring host is installed in the local cabinet of the reactor and communicates with the substation backend system through the communication interface.

[0018] In one possible implementation, the signal processing unit preprocesses the multi-point vibration signal by filtering out low-frequency drift, power frequency interference, interference from cooling fans or oil pumps, and high-frequency random noise, and performs detrending, scaling, or normalization.

[0019] The signal processing unit preprocesses the current signal by including multi-level filtering to separate power frequency / harmonic components, avoiding specific interference frequency bands, and suppressing background noise.

[0020] In one possible implementation, the vibration mode characteristics include the vibration energy and distribution ratio of each measuring point in the key frequency band, the relative phase difference and coherence of signals from different measuring points, modal frequency, damping ratio, mode shape, vibration signal complexity, stability, correlation, energy similarity, effective energy ratio, and harmonic ratio.

[0021] The current characteristics include the effective value of the power frequency circulating current, the effective value of higher harmonics, the harmonic content, as well as the amplitude, charge, pulse count per unit time, total discharge quantity, and PRPD and PRPS spectrum characteristics of the high-frequency partial discharge pulse.

[0022] In one possible implementation, dynamic association analysis employs time series cross-correlation analysis, Granger causality analysis, or machine learning-based dynamic association modeling.

[0023] Secondly, embodiments of this application provide an online monitoring and diagnostic method for mechanical instability of a 500kV oil-immersed current-limiting reactor, applied to the online monitoring and diagnostic system for mechanical instability of a 500kV oil-immersed current-limiting reactor described in any one of the first aspects, the method comprising:

[0024] The high-precision synchronous acquisition unit acquires multi-point vibration signals and current signals through high-precision time synchronization;

[0025] The signal processing unit performs signal processing and feature extraction on the multi-point vibration signal to obtain vibration mode features, which characterize the relative motion state of the key internal structure of the reactor.

[0026] The signal processing unit performs signal processing and feature extraction on the current signal to obtain current features, which characterize electrical abnormal states.

[0027] The correlation analysis unit establishes and continuously analyzes the dynamic correlation between the vibration mode characteristics and the current characteristics over time.

[0028] Based on the vibration mode characteristics, current characteristics, and dynamic correlation analysis results, the diagnostic and early warning unit uses a preset diagnostic model to perform state assessment, early warning, and fault diagnosis of the mechanical stability of the reactor.

[0029] Thirdly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the second aspect and / or various possible implementations of the second aspect.

[0030] Fourthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the second aspect and / or various possible implementations of the second aspect as described above.

[0031] The 500kV oil-immersed current-limiting reactor mechanical instability online monitoring and diagnosis system and method provided in this application includes a high-precision synchronous acquisition unit that acquires multi-point vibration and current signals synchronously. A signal processing unit performs signal processing and feature extraction on the multi-point vibration signals to obtain vibration mode features. Similarly, a signal processing unit performs signal processing and feature extraction on the current signals to obtain current features. A correlation analysis unit establishes and continuously analyzes the dynamic correlation between vibration mode features and current features over time. Based on the vibration mode features, current features, and the analysis results of the dynamic correlation, a diagnosis and early warning unit uses a preset diagnostic model to assess the mechanical stability of the reactor, provide early warnings, and diagnose faults. This method overcomes the limitations of traditional single-sensor monitoring, improves diagnostic accuracy, reduces false alarm and false negative rates, enhances equipment operation and maintenance management, and ensures the safe and stable operation of the power grid. Attached Figure Description

[0032] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0033] Figure 1 A schematic diagram of the structure of the 500kV oil-immersed current-limiting reactor mechanical instability online monitoring and diagnosis system provided in this application;

[0034] Figure 2 This is a schematic diagram of dynamic correlation analysis based on LSTM.

[0035] Figure 3 A schematic diagram of the OMA (Random Subspace Identification) algorithm for vibration mode feature extraction;

[0036] Figure 4 A schematic diagram illustrating the working principle of an SVM-based classification model for fault diagnosis;

[0037] Figure 5 A flowchart illustrating the online monitoring and diagnostic system for mechanical instability of the 500kV oil-immersed current-limiting reactor provided in this application;

[0038] Figure 6 A schematic diagram illustrating a specific example of the online monitoring and diagnosis method for mechanical instability of a 500kV oil-immersed current-limiting reactor provided in this application.

[0039] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0040] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0041] Oil-immersed current-limiting reactors are crucial equipment in modern power systems for limiting short-circuit currents. Especially for 500kV, high-capacity oil-immersed current-limiting reactors, their unique structural design (such as a large core air gap) and extremely high operating current load result in electromagnetic forces generated during operation far exceeding those of conventional parallel reactors, leading to significantly larger inherent vibration amplitudes. This high-intensity, continuous electromagnetic stress, along with factors such as thermal expansion and contraction during long-term operation, easily causes mechanical fatigue, loosening, or displacement of critical internal structural components, such as core laminations, clamps, windings, and clamping parts. Mechanical instability is one of the common and serious fault types in oil-immersed reactors.

[0042] Mechanical instability within the reactor's internal structure is not isolated; it often triggers a series of accompanying faults. For example, minute displacements or deformations of the core or clamps can alter the grounding circuit state, changing from a normal single-point grounding to multi-point grounding, creating harmful circulating currents, leading to localized overheating, and even burning out the grounding wire or accelerating insulation aging. Changes in mechanical stress can also cause distortions in the internal electric field distribution, generating localized high electric field strengths in specific areas, thereby inducing partial discharge and further accelerating insulation degradation. High-energy discharges caused by insulation degradation or partial discharges can then lead to an abnormal increase in the content of characteristic gases (such as acetylene) in the oil. These mechanical, electrical, and chemical phenomena are interconnected, forming a complex fault chain.

[0043] Existing reactor condition monitoring technologies typically include vibration monitoring, dissolved gas analysis (DGA), partial discharge (PD) monitoring, and core grounding current monitoring. However, these technologies have limitations when applied to high-inherent-vibration, large-capacity equipment such as 500kV oil-immersed current-limiting reactors. Traditional vibration monitoring usually relies on monitoring the vibration amplitude or specific frequency components at a limited number of points outside the tank. For equipment with large inherent vibration amplitudes, this method struggles to effectively distinguish between high vibrations under normal operating conditions and abnormal vibration modes indicating internal mechanical instability, easily leading to false alarms or missed alarms. Furthermore, the oil immersion environment and the heavy tank attenuate and couple vibration signals from the outer surface, making it difficult to accurately reflect the true, localized vibration state and relative displacement of critical internal components. Traditional dissolved gas analysis has long cycles, making minute-level response difficult. While isolated monitoring of core grounding current or partial discharge signals can detect electrical anomalies, it is difficult to trace whether they are caused by mechanical instability, nor can it capture early warning signs where mechanical changes precede electrical anomalies. Existing technologies lack the capability for high-precision synchronous acquisition and in-depth dynamic correlation analysis of state variables across different dimensions, particularly the ability to effectively integrate and correlate detailed internal mechanical state information (obtained through multi-point vibration mode analysis) with electrical anomaly information of the core grounding circuit (obtained through broadband current monitoring). Therefore, it is difficult to achieve early, highly sensitive warnings and accurate diagnosis of mechanical instability faults in this type of first-of-its-kind equipment, failing to meet its requirements for safe, efficient operation and condition-based maintenance.

[0044] To address the aforementioned issues, this application provides an online monitoring and diagnostic system and method for mechanical instability of a 500kV oil-immersed current-limiting reactor. This solves the problem that existing technologies struggle to provide early, highly sensitive warnings and accurate diagnoses of mechanical instability faults, failing to meet the requirements for safe, efficient operation and condition-based maintenance. Specifically, existing technologies for monitoring the mechanical stability of oil-immersed current-limiting reactors primarily rely on single-dimensional discrete monitoring methods: judging the equipment status by the vibration amplitude or specific frequency components at limited measuring points on the tank surface. However, such methods struggle to distinguish between high inherent vibrations and abnormal vibration modes, and the oil immersion environment significantly attenuates and couples vibration signals, leading to distortion in the perception of the internal structural state. Alternatively, they may assess the degree of insulation degradation by detecting the content of characteristic gases such as acetylene and hydrogen in the oil. However, this method suffers from long detection cycles (typically on the order of hours) and an inability to reflect early changes in mechanical instability in real time. Another approach involves detecting partial discharge signals using high-frequency current sensors, but isolated monitoring makes it difficult to determine whether the discharge is caused by mechanical instability, and it lacks sensitivity to early, weak discharges. Alternatively, the grounding status of the iron core can be determined by the amplitude of the power frequency circulating current, but this method cannot capture harmonic component changes or high-frequency abnormal signals caused by mechanical instability. The above methods lack the ability to perform high-precision synchronous acquisition and dynamic correlation analysis of multi-source state variables, failing to achieve early, high-sensitivity warnings for mechanical instability faults. For example, traditional methods struggle to identify the temporal dependency between mechanical state changes and electrical anomalies, leading to high false alarm and false negative rates; furthermore, under high natural vibration backgrounds, they cannot effectively extract vibration mode features characterizing the relative motion of the internal structure, making it difficult to distinguish between normal operation and mechanical instability. Considering these problems, the inventors investigated whether high-precision synchronous acquisition of multi-point vibration signals and broadband core grounding current signals could be used, combined with vibration mode feature extraction and current feature analysis, to establish dynamic correlations and achieve early warning and accurate diagnosis of mechanical instability based on machine learning models. This breaks through the limitations of traditional single-dimensional monitoring, solving the problem of collaborative judgment of mechanical state and electrical anomalies under high natural vibration backgrounds through deep fusion and dynamic correlation analysis of multi-source data, achieving early, high-sensitivity warnings for mechanical instability faults. Based on this, the solution proposed in this application is presented.

[0045] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0046] Figure 1 This is a schematic diagram of the 500kV oil-immersed current-limiting reactor mechanical instability online monitoring and diagnostic system provided in this application, as shown below. Figure 1As shown, the system acquires multi-dimensional state signals through sensors (including vibration sensor 10 and wideband current sensor 20), performs digital processing via a high-precision synchronous acquisition unit 30, and then processes, analyzes, and infers the data step by step through a signal processing unit 40, a feature extraction unit 50, a correlation analysis unit 60, and a diagnostic early warning unit 70. Finally, the system outputs diagnostic results and interacts with external systems through a data storage and communication interface 80. These processing units are typically integrated into a single monitoring unit (IED) 90.

[0047] The vibration sensor 10 employs a high-sensitivity IEPE or MEMS piezoelectric vibration acceleration sensor with a sensitivity of 100mV / g. Its measurement range covers 0.05g to 10g, and it can measure instantaneous vibration impacts up to 50g, meeting the vibration measurement requirements of the 500kV oil-immersed current-limiting reactor under high current or abnormal operating conditions. Based on the reactor's structural characteristics and monitoring focus, at least eight, preferably eight to twelve, vibration sensors are arranged on the reactor tank surface corresponding to key stress areas such as the core column, windings, and clamps. The arrangement strategy employs a matrix or focused area coverage, with sensors placed at different heights on the side of the tank corresponding to each core column or winding (e.g., at 1 / 4 and 3 / 4 of the tank height) and at the top or bottom. The sensors are fixed using a non-invasive magnetic method, with the magnetic base having a suction force of no less than 50N to ensure safe installation when the reactor is energized. The sensor signal cables use special shielded cables that meet the requirements of the substation environment, such as double-shielded cables with stainless steel armor protection, which reliably achieve signal shielding against strong electromagnetic interference in substations and protection against damage during cable laying.

[0048] The wideband current sensor 20 is a non-contact, openable wideband current transformer (CT) sensor, installed on the reactor's core grounding flat iron or clamp grounding flat iron. Designed based on the zero-flux principle, this sensor can be installed online without altering the original single-point grounding loop structure, thus not affecting equipment safety or surge current discharge capacity. The sensor features optimized winding parameters and core material, achieving wideband response characteristics, capable of measuring current signals from power frequency (50 / 60Hz) to at least 10MHz bandwidth. The measurement range covers a weak 1mA circulating current during normal operation to potentially high currents of 50A or even higher during faults. The measurement error meets high accuracy requirements; within the 1mA-50A range, the maximum permissible error is ≤±(2% of the reading + 1mA), and the resolution is better than 0.1mA. The sensor's output signal is transmitted to the high-precision synchronous acquisition unit 30 via a dedicated shielded cable.

[0049] A high-precision synchronous acquisition unit 30 connects multiple vibration sensors 10 and a wideband current sensor 20. This unit is equipped with multiple analog-to-digital converter (ADC) channels, the number of which matches the number of sensors (at least 8 vibration channels + 1 current channel). All ADC channels share a single high-precision clock source (a synchronization clock signal provided by a GPS or NTP server, with synchronization accuracy better than microseconds), enabling parallel synchronous sampling of all sensor signals. Strict time synchronization is fundamental for subsequent dynamic correlation analysis between signals. The sampling rate is determined based on the highest frequency requirement of the monitored signal; the vibration signal sampling rate can be 10kHz, 100MHz, or higher. The acquired digitized data is timestamped with high precision and stored in a buffer. It should possess the functions of parallel synchronous acquisition of signals from all monitoring channels and vibration-triggered waveform recording.

[0050] The signal processing unit 40 is connected to the high-precision synchronous acquisition unit 30 to preprocess the raw digitized signals in the buffer. For vibration signals, digital filters are used to remove low-frequency drift, power frequency interference, specific frequency interference caused by the equipment's own cooling fan or oil pump, and high-frequency random noise, performing detrending and possibly signal scaling or normalization. For broadband current signals, multi-stage filtering is performed, using digital bandpass filters to separate power frequency / harmonic components, and background noise suppression is applied. Preprocessing aims to remove interference, improve signal quality, and provide clean data for subsequent feature extraction. It should have automatic acquisition, signal conditioning, analog-to-digital conversion, and data preprocessing functions for monitoring equipment status parameters.

[0051] The feature extraction unit 50 is connected to the signal processing unit 40, and extracts a sequence of diagnostically significant feature parameters from the processed multi-point vibration signal and broadband core grounding current signal.

[0052] Processed multi-point vibration time-domain signal (in (where N is the number of measurement points and t is time) to perform time-domain, frequency-domain, or time-frequency analysis.

[0053] For each measurement point signal During a time window Perform a Fast Fourier Transform (FFT) within the internal array to calculate the one-sided power spectral density (PSD). PSD reflects the distribution of signal energy with frequency. It is extracted from the key frequency range of the reactor (such as the power frequency harmonics). And the natural frequency range of the core, windings, and clamps determined through theoretical calculations or experiments. Vibrational energy within) Or amplitude. Calculate the distribution pattern and proportion of vibration energy among various measuring points. .

[0054] Calculate signals from any two or more measuring points and Cross-power spectral density at a specific frequency f Extracting the relative phase difference and coherence Coherence This reflects the frequency difference between the two measurement points. The degree of linear correlation at different locations is indicated by a value close to 1, suggesting a high degree of correlation. These characteristics reflect the synchronicity, coordination, or relative motion trends of vibrations at different locations.

[0055] Using synchronous multi-point vibration signals Operational mode analysis (OMA) algorithms, such as random subspace identification (SSI), are employed.

[0056] Figure 3 The diagram illustrates the OMA (Random Subspace Identification) algorithm for vibration mode feature extraction, as shown below. Figure 3 As shown, the N-channel synchronous vibration response data collected at time t are sliced ​​according to time to form a sequence. Construct the Hankel matrix. in Or output the covariance matrix The block Hankel matrix is ​​formed.

[0057] For the Hankel matrix Perform singular value decomposition: ,in It is a submatrix of a unitary matrix. It contains non-zero singular values diagonal matrix, It is the system order.

[0058] The effective system order is determined by the magnitude of the singular values ​​(usually through a singular value decay plot). Select a subspace containing the main system dynamics. Estimate the discrete-time state-space model matrix of the system using projection or least squares methods. For example, the observable matrix can be estimated using the results of SVD decomposition. and controllable matrix ,but ,in It is after removing the first row .

[0059] For discrete-time state-space matrix Perform eigenvalue decomposition: ,in It contains eigenvalues diagonal matrix, It is a matrix containing discrete-time mode vectors. Each eigenvalue... Corresponding to a mode. From Continuous-time modal frequencies can be extracted from it. And modal damping ratio ,in This refers to the sampling interval. The corresponding continuous-time mode shape vector. From Obtained from.

[0060] Calculate characteristic quantities that reflect the overall vibration state, such as vibration signal complexity (Shannon entropy based on spectral amplitude distribution). For frequency The percentage of energy at a given point), vibration signal stability (the volatility of the quantized time-domain signal envelope or the instantaneous rate of energy change, for example, based on the Teager-Kaiser energy operator, instantaneous energy). Stability and The variance or rate of change is related to the vibration signal (e.g., calculating the correlation between key measuring points). Pearson correlation coefficient of the signal within a time window W: (rolling mean or peak value), vibration energy similarity (comparing the cosine similarity of multi-point vibration energy distribution vectors under different time windows or different loads). E is the energy vector in a predefined frequency band. ), effective energy ratio (the proportion of power frequency and harmonic frequency energy to total energy) Harmonic ratio (the ratio of odd-order harmonic energy to even-order harmonic energy) These characteristics, through quantification of spectral distribution, time-domain signal fluctuations, and inter-measurement synchronization, provide a more comprehensive description of the reactor's mechanical vibration state and potential anomalies. The system should possess time-domain and frequency-domain vibration signal analysis capabilities, and should output the vibration displacement and acceleration amplitude corresponding to each frequency component. The vibration displacement amplitude (peak-to-peak value) can be calculated from the acceleration spectrum using a double integral. .

[0061] Processed broadband core grounding current signal Perform the analysis.

[0062] Fourier analysis was performed on the power frequency / harmonic components to extract the effective value of the power frequency circulating current. and the effective value of specific higher harmonic currents (e.g., 3rd, 5th, etc.) Calculate harmonic content Monitor the absolute and relative amounts of these components.

[0063] For high-frequency PD pulse components, pulse detection was performed using the Teager-Kaiser energy operator (TKEO).

[0064] To address the characteristics of high-frequency PD pulses—high instantaneous energy and short duration—the Teager-Kaiser Energy Operator (TKEO) is used to highlight the pulse component in the signal. For discrete-time signals... The discrete form of TKEO is:

[0065]

[0066] TKEO can instantly track the energy of a signal, is sensitive to changes in the amplitude and frequency of the signal, and has a certain degree of suppression effect on narrowband Gaussian noise.

[0067] Among them, the PD pulse usually manifests as a brief high-frequency oscillation, and its TKEO value is much higher than the background noise and continuous interference at the moment the pulse occurs.

[0068] Pulse recognition steps:

[0069] The processed high-frequency current signal Calculate its TKEO sequence .

[0070] For energy sequence Perform smoothing processes, such as using moving averages or low-pass filters, to reduce noise-induced glitches.

[0071] Set an energy threshold This threshold can be determined by analyzing the statistical properties of the TKEO sequence under normal operating conditions (e.g., the mean plus a certain number of standard deviations) or experimentally. The energy sequence E[n] is detected if it continuously exceeds the threshold. The time period. The time corresponding to the energy peak within each time period exceeding the threshold is marked as the potential PD pulse occurrence time. To avoid detecting the same pulse multiple times, a short suppression window can be set after a pulse is detected.

[0072] Once the PD pulse occurrence time is detected, the waveform segment of the pulse can be extracted, and the pulse amplitude can be calculated. , charge (in (Sampling interval). Statistical pulse count and total discharge rate per unit time. Maximum discharge amount Based on the pulse occurrence time Phase relationship with power frequency voltage (or current) We constructed and analyzed PRPD and PRPS maps.

[0073] High-level features such as spectrum shape, distribution center, and dispersion are extracted. For multi-source PD separation, feature vectors for each pulse can be extracted (e.g., amplitude, duration, bandwidth, wavelet decomposition-based coefficients, or TKEO envelope features). K-means clustering is then used in the feature space (e.g., the space after PCA dimensionality reduction) to cluster the pulse feature vectors, classifying PD pulse signals from different discharge sources or different types. The goal of the K-means algorithm is to minimize the sum of squares within each cluster. Where K is the number of clusters, It is the j-th cluster. It is the center of the j-th cluster. It should have high-frequency partial discharge monitoring capabilities, providing PRPD and PRPS spectra. It should also have the function of clustering partial discharge signals with two or more characteristic parameters to achieve multi-source partial discharge separation.

[0074] The correlation analysis unit 60 is connected to the feature extraction unit 50 and receives the extracted synchronous and time-aligned vibration mode feature sequence. and current characteristic sequence This unit executes a dynamic correlation analysis algorithm to uncover the inherent connections and temporal dependencies between the two states, thereby determining whether the mechanical and electrical states are abnormally coordinated, and which state's change may precede the other.

[0075] Machine learning association analysis based on sequence patterns (e.g., LSTM-based): Using a trained LSTM-based sequence model to identify abnormal temporal association patterns between vibration features and current features.

[0076] Model training process:

[0077] Collect multi-dimensional synchronous monitoring data containing historical data of normal operation and various known faults (such as core loosening, winding deformation, associated circulating current, and associated partial discharge). Perform preprocessing and feature extraction on the data to form a multi-dimensional feature sequence with precise timestamps. Based on fault records, expert judgments, or other auxiliary detection results, the time periods of historical data are labeled, such as which time periods correspond to "normal correlation", "vibration-led current abnormal correlation (V→C)", "current-led vibration abnormal correlation (C→V)", "synchronization abnormal correlation", "correlation caused by external interference", or more specific correlation types (such as "correlation mode of increased circulating current caused by mechanical loosening").

[0078] The labeled feature sequences are divided into fixed-length time windows (e.g., each segment contains past data). Each time step has an interval of [number] time steps. The data is divided into training samples. Each sample includes the input sequence (within the time window). Total length ) and corresponding label categories The dataset is divided into training, validation, and test sets. Figure 2 This is a schematic diagram of dynamic correlation analysis based on LSTM, such as... Figure 2 Construct an LSTM network model. For example, the input layer receives data containing L time steps, each time step containing data of dimension L. eigenvectors The sequence. The network contains one or more LSTM layers, each containing H LSTM units. The core computation of an LSTM unit includes the input gate. Forgotten Gate Output gate and unit state The update of the LSTM layer output (e.g., the hidden state at the last time step). The outputs (or the outputs at all time steps) can be connected to fully connected (dense) layers, and finally to a classification layer (e.g., using a softmax activation function), which outputs the probabilities of belonging to different association pattern categories. , where M is the number of associated pattern categories.

[0079] Supervised training of the LSTM model is performed using the training set. An appropriate loss function is chosen (e.g., for classification tasks, the cross-entropy loss function is used). y is the one-hot vector of the true label. (This is the probability vector predicted by the model). Choose a suitable optimizer (e.g., the Adam optimizer) and compute the gradient of the loss function with respect to the network weights using the backpropagation algorithm. Iteratively update the network weights to minimize the loss function: ,in The learning rate is used. Model performance is monitored on the validation set, hyperparameters are tuned, and overfitting is prevented (e.g., using Early Stopping, Dropout, etc.). The training objective is to enable the model to accurately identify which dynamic association pattern the input feature sequence belongs to. The generalization ability of the trained model is evaluated using the test set (e.g., calculating classification accuracy, precision, recall, and F1 score). Once performance requirements are met, the trained LSTM model (including network structure and weight parameters) is deployed to the association analysis unit 60 of the monitoring host 90.

[0080] During online operation, the correlation analysis unit 60 acquires the multi-dimensional feature sequence output by the feature extraction unit 50 in real time. At each new time step t, construct a sliding time window sequence containing data from the most recent L time steps. ,in The sliding time window sequence As input, it is fed into the deployed LSTM model. The model performs forward computation in real time, outputting the probability or score vector of the current time window data belonging to different preset association pattern categories:

[0081] (A2)

[0082] For example, output a vector ,in This represents the probability that the data sequence exhibits the m-th correlation pattern within the time window ending at time t. This probability vector, as the real-time output of dynamic correlation analysis, provides a quantitative assessment of whether the current mechanical and electrical states are abnormally linked and the type of such abnormality.

[0083] Other association analysis tools (e.g., time-series cross-correlation and Granger causality):

[0084] For the selected key vibration characteristics (For example, the lowest order modal frequency extracted by OMA) (Changes over time) and key current characteristics C t (For example, the amplitude of the third harmonic of the core grounding current) (Varying over time), calculate the Pearson rolling cross-correlation function between the two within a fixed-length sliding time window W:

[0085]

[0086] Where V and C represent the selected time series of vibration and current characteristics, respectively. It is the mean within the window. Due to time lag. Real-time monitoring. With time t and lag Changes. For example, continuous tracking. Whether the peak value continues to increase indicates whether the vibration change stably leads the current change.

[0087] For a selected feature pair (V, C), a VAR model is built within a sliding time window:

[0088]

[0089] Where p is the model order. Granger causality tests are performed within a sliding time window. For example, the null hypothesis is tested. (i.e., V does not induce Granger causation C) vs. the Bezeitgeist hypothesis H1: H0 is not true (i.e., V induces Granger causation C). Calculate the residual sum of squares (RSS). u (Unconstrained Model) and RSSr (Constrained model). Calculate the F-statistic:

[0090]

[0091] Calculate the corresponding p-value. Monitor the p-value over time in real time. For example, if the p-value is detected to continuously decrease to a significant level (e.g., <0.05), it indicates that the vibration characteristics have begun to statistically significantly predict the current characteristics, providing evidence that mechanical changes lead to electrical changes. These auxiliary analysis results (such as rolling cross-correlation peaks, Granger causality p-values) can supplement the correlation analysis of machine learning models or be part of their output dimensions.

[0092] The diagnostic early warning unit 70 is connected to the correlation analysis unit 60 and the feature extraction unit 50. This unit is the decision center of the system. It receives the currently extracted vibration mode features. Current characteristics And the dynamic correlation index or pattern recognition results output by the correlation analysis unit 60 (e.g., the correlation pattern probability vector output by the LSTM model). Based on these inputs, a built-in diagnostic model is used to assess the mechanical stability of the reactor, determine the fault type, and provide early warnings.

[0093] The pre-set diagnostic model is an intelligent classification model that has been trained and validated in advance based on the equipment's historical operating data, typical fault samples, and expert knowledge. It is used to integrate multi-dimensional monitoring information and achieve automatic fault identification.

[0094] The model takes vibration mode characteristics, current characteristics, and the dynamic correlation between vibration and current as inputs. By learning the characteristic patterns of normal state, early anomalies, typical mechanical instability faults, and accompanying electrical anomalies, it establishes a mapping relationship from multi-dimensional characteristics to equipment health status. It can automatically determine the current operating status of the reactor, identify the fault type and severity, and output the corresponding confidence level.

[0095] When the model is running, it comprehensively judges whether the equipment is in a normal, alert, or abnormal state based on the real-time extracted features and correlation analysis results, locates the location and cause of the fault, and triggers graded early warnings according to preset thresholds, thereby achieving early, highly sensitive, and highly accurate diagnosis and early warning of reactor mechanical instability faults.

[0096] Build a fault diagnosis model, such as a multi-classifier based on support vector machine (SVM). Figure 4 This is a schematic diagram illustrating the working principle of an SVM-based classification model for fault diagnosis, as shown below. Figure 4As shown, the model training process involves using the same historical data as the correlation analysis model to ensure time alignment. For each time window or each sampling time t, a diagnostic input feature vector is constructed. It includes: vibration mode characteristics V at the current moment or within a time window. t (or its statistics, such as mean V, variance Var(V)), current characteristics C t (or its statistics, such as mean C, variance Var(C)), and the dynamic correlation index calculated by the correlation analysis unit 60 within the current time window (e.g., the correlation pattern probability vector output by the LSTM model). Rolling cross-correlation peak Granger causality p-value).

[0097] For example,

[0098] The historical data is labeled with time periods, and the overall device status category corresponding to each sample is marked. For example, “normal”, “caution”, “abnormal”, or more specific fault types (e.g., “normal operation”, “slight loosening of the iron core”, “severe loosening of the iron core with circulating current”, “winding deformation with partial discharge”, etc.).

[0099] The labeled feature vector D t and their corresponding category labels Constructing the training dataset The dataset is divided into training, validation, and test sets.

[0100] Construct an SVM classifier. The input is a D-dimensional vector. D eigenvectors D t The model classifies samples by finding the optimal separating hyperplane in a high-dimensional feature space. For binary classification problems, the learning objective is to minimize the following objective function:

[0101]

[0102] in It is the input vector x i Kernel functions mapped to high-dimensional space (e.g., radial basis function (RBF) kernels) The class label is (±1), and C>0 is the penalty factor. These are slack variables. For multi-class classification problems, one-to-many or one-to-one strategies can be used.

[0103] The SVM model is trained under supervision using the training set. The optimal parameters w and b (or the corresponding support vectors and Lagrange multipliers) are found by solving the convex quadratic programming problem described above. Cross-validation and parameter tuning (e.g., adjusting C and gamma parameters) are then performed on the validation set.

[0104] Use the test set to evaluate the generalization ability of the trained model (e.g., calculate classification accuracy). Precision = Recall = , (Including ROC curve and AUC value). After meeting the performance requirements, the trained SVM model (including support vectors, kernel function parameters, bias terms, etc.) is deployed to the diagnostic and early warning unit 70 of the monitoring host 90.

[0105] During online operation, the diagnostic warning unit 70 acquires the real-time extracted feature vector D. t (Includes vibration characteristics, current characteristics, and real-time correlation analysis results). D t The input is fed into the deployed SVM model. The model outputs a classification result y. diag,t This indicates the current state category or fault type of the device (e.g., "Note: Slightly loose iron core"). The model can also output a confidence score or probability P for each category. j |D t ).

[0106] Equipment status is assessed based on the state category or probability output by the diagnostic model. For example, if the SVM model outputs an "abnormal" category probability P(Abnormal|D)... t If the percentage exceeds a preset threshold (≥50%), the device is determined to be in an abnormal state. The diagnostic results are mapped to three levels: Normal, Attention, and Abnormal, for example:

[0107] Normal state probability Attention state probability P(Attention) ≥ 0.5 and The probability of an abnormal state, P(Abnormal) ≥ 0.5, and ,in For margin.

[0108] When the equipment status is assessed as "Caution" or "Abnormal," the diagnostic model further outputs the specific fault type (if the model has been trained with detailed classification). The diagnostic model can preliminarily determine the specific location or area of ​​the fault based on the location of the specific sensor measurement points involved in the diagnosis (e.g., in which measurement point areas the abnormal vibration characteristics mainly appear, or in which measurement points the maximum amplitude corresponding to the abnormal mode shape appears) or the mode shape distribution extracted by OMA.

[0109] Based on the diagnosed fault type, severity (e.g., confidence score of the classification model, or degree of deviation from the normal baseline), and correlation analysis results (e.g., types of identified abnormal correlation patterns), the system generates tiered early warning information and triggers corresponding alarms. Following the alarm strategy diagram (as shown in the project documentation), different types of alarms are triggered, including:

[0110] When the vibration acceleration amplitude of multiple measuring points (≥4 configurable measuring points) increases rapidly in a short period of time Exceeding the preset mutation threshold Triggered at time.

[0111] Triggered when multiple vibration characteristics combine to meet specific abnormal conditions, such as the average complexity of vibration signals from ≥3 adjacent measuring points. And the average stability of vibration signals And at this time, the amplitude of vibration acceleration And effective energy percentage And harmonic ratio Triggered when (for conditions where external interference is excluded). These threshold parameters can be set and adjusted according to equipment type and operating experience.

[0112] When the diagnostic warning unit 70 determines, based on comprehensive analysis, that the reactor is in a "caution" or "abnormal" state, and the corresponding probability exceeds a preset threshold (≥50% or ≥60%), it is triggered. The alarm level in the abnormal state is higher than that in the caution state.

[0113] The early warning information includes a fault description, suggested handling measures, and urgency level, and is sent to the station control backend or maintenance personnel's terminal via communication interface 80. It should have online monitoring and alarm functions, issuing alarm signals for various abnormal states, and the alarm limits should be modifiable.

[0114] The data storage interface 80 is used to reliably store high-precision synchronously acquired raw time-domain waveform data, processed feature data sequences, dynamic correlation analysis results, diagnostic conclusions, status assessment reports, early warning information, and alarm records in the industrial-grade non-volatile memory inside the monitoring host 90, or upload them to a remote database server via the communication interface. It should meet long-term storage requirements, such as storing at least one year's worth of raw monitoring data and at least ten years' worth of feature values ​​and diagnostic history, ensuring the security of recorded data and preventing loss of recorded data due to power interruption or drops. The communication interface 80 provides a standard Ethernet interface, supporting commonly used communication protocols in the power industry, such as DL / T 860 (IEC61850), IEC 60870-5-104, Modbus, etc., to enable bidirectional communication with substation control systems, integrated processing units, or remote monitoring centers. Communication functions include, but are not limited to: periodic uploading of monitoring data, active uploading of alarms and events, remote data querying, historical data export, remote configuration parameter and diagnostic model updates, and transmission of graph files (such as PRPD / PRPS graphs). The monitoring host 90 also provides web service functionality. Through the built-in web server, maintenance personnel can access the monitoring host via a web browser to view real-time data, historical trends, graphs, diagnostic results, and alarm information, and perform on-site monitoring and maintenance.

[0115] The monitoring host (IED) 90 integrates a high-precision synchronous acquisition unit 30, a signal processing unit 40, a feature extraction unit 50, a correlation analysis unit 60, and a diagnostic and early warning unit 70 into one unit. The monitoring host 90 adopts an industrial-grade hardware design, conforming to the standard substation cabinet rack-mounted installation requirements (e.g., standard U-size industrial control chassis), and features a high-performance multi-core processor (e.g., embedded CPU or industrial control computer CPU), large-capacity memory, and storage space. Internally, it is equipped with a multi-channel high-speed data acquisition card to achieve high-speed, high-precision digitization of sensor signals. Some real-time-critical signal processing and feature extraction tasks (such as real-time detection of high-frequency PD pulses, PRPD spectrum construction, and fast FFT calculation) can be implemented in a field-programmable gate array (FPGA) or digital signal processor (DSP) to ensure processing speed and fidelity. The processor runs an embedded operating system and the monitoring and diagnostic software of this invention. The software adopts a modular design, with each functional module running in parallel. The monitoring host has its own operating status monitoring function and can automatically restart and recover in the event of an anomaly (e.g., communication interruption, storage anomaly). The system should have automatic monitoring of the device's operating status and automatic restart functions for system anomalies. Necessary electromagnetic interference prevention measures should be taken, the metal chassis should be reliably grounded, each module should be flexible to plug and unplug with reliable contact and good interchangeability, and heat-generating components should have ventilation and heat dissipation conditions.

[0116] Figure 5This is a flowchart illustrating the online monitoring and diagnosis method for mechanical instability of a 500kV oil-immersed current-limiting reactor provided in this application. Figure 5 As shown, this method is applied to the 500kV oil-immersed current-limiting reactor mechanical instability online monitoring and diagnosis system in the aforementioned embodiments. The method includes:

[0117] S501: The high-precision synchronous acquisition unit acquires multi-point vibration signals and current signals through high-precision time synchronization.

[0118] S502: The signal processing unit performs signal processing and feature extraction on multi-point vibration signals to obtain vibration mode features.

[0119] S503: The signal processing unit performs signal processing and feature extraction on the current signal to obtain current features.

[0120] S504: The correlation analysis unit establishes and continuously analyzes the dynamic correlation between vibration mode characteristics and current characteristics over time.

[0121] S505: The diagnostic and early warning unit uses a preset diagnostic model to perform state assessment, early warning and fault diagnosis of the mechanical stability of the reactor based on the analysis results of vibration mode characteristics, current characteristics and dynamic correlation.

[0122] The specific implementation method of this method is the same as that of each unit in the aforementioned system, and will not be repeated here.

[0123] The online monitoring and diagnosis method for mechanical instability of a 500kV oil-immersed current-limiting reactor provided in this application includes a high-precision synchronous acquisition unit that acquires multi-point vibration and current signals synchronously. A signal processing unit performs signal processing and feature extraction on the multi-point vibration signals to obtain vibration mode features. Similarly, a signal processing unit performs signal processing and feature extraction on the current signals to obtain current features. A correlation analysis unit establishes and continuously analyzes the dynamic correlation between vibration mode features and current features over time. Finally, a diagnosis and early warning unit, based on the vibration mode features, current features, and the analysis results of the dynamic correlation, uses a preset diagnostic model to assess the mechanical stability of the reactor, provide early warnings, and diagnose faults. This method overcomes the limitations of traditional single-sensor monitoring, improves diagnostic accuracy, reduces false alarm and false negative rates, enhances equipment operation and maintenance management, and ensures the safe and stable operation of the power grid.

[0124] Figure 6 A schematic diagram illustrating a specific example of the online monitoring and diagnosis method for mechanical instability of a 500kV oil-immersed current-limiting reactor provided in this application is shown below. Figure 6As shown, the monitoring host 90 powers on and performs a hardware self-test to confirm that sensor connections, communication status, and storage status are normal. It synchronizes with the substation's master clock via NTP or IRIG-B protocol to ensure system clock synchronization. Monitoring configuration parameters and diagnostic model parameters are loaded from local storage or a remote server. If necessary, baseline data from historical health states or pre-trained model parameters are loaded.

[0125] The high-precision synchronous acquisition unit 30 continuously performs high-precision time-synchronous sampling on all connected vibration sensors 10 and wideband current sensors 20 according to a preset sampling rate and period, acquiring raw digital time-domain waveform data a. i (t k ), i g (t k The data is stored in a cache. Simultaneously, vibration signals are monitored, and trigger conditions are met (e.g., the increase in vibration acceleration amplitude |a)... i (t k )−a i (t k−Δt short) | Exceeds the preset threshold T shock Initiate the trigger recording function to record waveform data for a period of time before and after the event occurs.

[0126] The signal processing unit 40 reads the latest data segment from the buffer and performs preprocessing such as filtering and noise reduction. The feature extraction unit 50 extracts the current vibration mode feature vector V from the processed signal segment in real time or near real time. t and current eigenvector C t For example, within each sliding time window, the spectrum (FFT), energy distribution, relative phase / coherence, and advanced statistical features (complexity, stability, correlation, energy similarity, etc.) of the vibration signal are calculated. Periodically or triggered, the OMA algorithm is executed to extract modal parameters (frequency, damping, mode shape). Power frequency / harmonic features and high-frequency PD features (pulse statistics, spectral features) are separated and calculated from the broadband current signal. The extracted feature sequences are then stored.

[0127] The correlation analysis unit 60 reads data (including current and historical data) within a certain time window from the stored feature sequences and constructs a time series sequence sample X containing vibration and current features. t The sequence is then input into the trained LSTM model. In the model output, the probability or score vector of the current time window data belonging to different association pattern categories is determined. (As shown in Equation A2). Simultaneously, calculate the rolling cross-correlation function of key feature pairs (e.g., the lowest-order mode frequency and the amplitude of the third / fifth harmonics). Or the p-value for Granger causality. Update association status indicators or identify anomalous association patterns.

[0128] The diagnostic early warning unit 70 constructs a system containing the current feature value (V). t C t ) and correlation analysis results (e.g., LSTM probability vectors) Rolling cross-correlation peak Diagnostic input feature vector D (such as Granger p-value, etc.) t D t The input is fed into a trained SVM classifier, and the model outputs the probability that the current device belongs to each state category or fault type (e.g., [P(Normal), P(Attention), P(Abnormal), ...]). Based on a preset threshold, the category corresponding to the highest probability is taken as the diagnostic result, and a state assessment (normal, attention, abnormal) is performed. If the diagnostic result indicates an attention or abnormal state, further analysis of correlation patterns and feature anomalies is conducted to determine the fault type and root cause. According to the alarm strategy, corresponding hierarchical alarms are triggered (e.g., vibration mutation alarm, feature quantity combination alarm, evaluation anomaly alarm).

[0129] Real-time feature data, correlation analysis results, diagnostic conclusions, status assessments, and early warning information are sent to the station control system or remote monitoring center via communication interface 80 (supporting functions such as periodic uploading, call response, and proactive alarm transmission). All relevant data and events are recorded in local storage for historical querying and trend analysis.

[0130] The system returns to the aforementioned steps and continues online monitoring and diagnosis.

[0131] The stored historical data can be used for offline analysis, such as in-depth trend analysis, fault mode mining, and retraining and optimization of feature extraction algorithms, correlation models, and diagnostic models to improve the system's accuracy and ability to identify new fault modes. The accompanying software should have functions for trend analysis of monitored quantities, fault diagnosis, status evaluation, and alarms.

[0132] To verify the effectiveness of the method of this invention, a platform simulating reactor vibration and abnormal grounding loop current was built in a laboratory environment, or offline playback and analysis were performed using historical monitoring data of actual reactors. Different fault scenarios were simulated on the simulation platform or in historical data (e.g., loose core bolts causing a slight decrease in modal frequency accompanied by an increase in specific harmonics, local deformation of the winding causing changes in mode shape in specific areas accompanied by signs of partial discharge, etc.), and synchronous data was collected for processing and diagnosis.

[0133] The accuracy of the diagnostic method based on multi-point vibration mode and broadband core grounding current in this application in identifying mechanical instability faults is compared with that of traditional single vibration amplitude monitoring method and single grounding current monitoring method.

[0134] Table A1 Comparison of Fault Diagnosis Accuracy

[0135]

[0136] The comparison shows that the method of this application significantly improves the accuracy of fault identification by fusing multi-source information and performing correlation analysis.

[0137] The statistical method described in this application improves the lead time for detecting fault signs and issuing early warnings compared to traditional methods.

[0138] Table A2 Early Warning Time Lead

[0139]

[0140] Data shows that the method proposed in this application can detect early signs of failures more quickly, significantly improving the lead time for early warning and giving maintenance personnel valuable time to handle the situation.

[0141] Compare the false alarm rate and false negative rate of different methods.

[0142] Table A3 Comparison of False Alarm Rate and False Negative Rate

[0143]

[0144] The method in this application effectively reduces the false alarm rate by cross-validating and correlating multi-source data, while also improving the ability to detect early and subtle fault signs and reducing the false alarm rate.

[0145] During the simulated core loosening process, the changes in the lowest-order mode frequency and modal damping ratio, extracted by OMA, as well as the changes in the amplitude of the third harmonic of the core grounding current, were monitored, and the Granger causality relationship p-value between the two was analyzed. It was found that in the early stage of simulated loosening, slight changes occurred in the modal frequency and damping ratio. Simultaneously, the Granger causality test showed that the vibration characteristics began to significantly cause changes in the current harmonics (p-value changed from insignificant to significant, e.g., <0.05). At this point, traditional monitoring methods might not show obvious anomalies, while the method of this invention can capture the emergence of this correlation, achieving early warning.

[0146] In a scenario simulating a loose iron core that gradually leads to an increase in circulating current, the LSTM model is able to identify the mode shift from "normal correlation" to "vibration-leading current abnormal correlation," and the probability of the corresponding correlation mode in its output is significantly increased, providing strong evidence for diagnosis.

[0147] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the methods of the various embodiments described above.

[0148] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the methods of the above embodiments.

[0149] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0150] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0151] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0152] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0153] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0154] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0155] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0156] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. An online monitoring and diagnostic system for mechanical instability of a 500kV oil-immersed current-limiting reactor, characterized in that, include: The sensor module and the monitoring host include a high-precision synchronous acquisition unit, a signal processing unit, a feature extraction unit, a correlation analysis unit, and a diagnostic and early warning unit. The sensor module includes a vibration sensor and a wideband current sensor; The high-precision synchronous acquisition unit is connected to the sensor module and is used to acquire multi-point vibration signals and current signals through the sensor module and send them to the signal processing unit. The signal processing unit is connected to the high-precision synchronous acquisition unit and is used to preprocess the multi-point vibration signal and the current signal; The feature extraction unit is connected to the signal processing unit and is used to extract vibration mode features from the processed multi-point vibration signal and extract current features from the processed current signal. The correlation analysis unit is connected to the feature extraction unit and is used to establish and analyze the dynamic correlation between the vibration mode features and the current features; The diagnostic early warning unit is connected to the correlation analysis unit and is used to perform state assessment, early warning and fault diagnosis of the mechanical stability of the reactor based on the vibration mode characteristics, the current characteristics and the dynamic correlation relationship.

2. The system according to claim 1, characterized in that, The vibration sensors are vibration acceleration sensors, and there are no fewer than 8 channels. They are arranged on the surface of the reactor's tank body at the locations corresponding to the iron core column, winding, or clamp, and are installed and fixed using a magnetic attraction method.

3. The system according to claim 1, characterized in that, The broadband current sensor is a non-contact, open-close type broadband CT sensor, which is installed on the iron core grounding flat iron or clamp grounding flat iron of the reactor and is used to measure current signals with a bandwidth from power frequency to at least megahertz.

4. The system according to claim 1, characterized in that, The system also includes a storage module and a communication interface. The storage module is used to store monitoring data, and the monitoring host communicates with external devices through the communication interface.

5. The system according to claim 4, characterized in that, The monitoring host is installed in the local cabinet of the reactor and communicates with the substation backend system through the communication interface.

6. The system according to claim 1, characterized in that, The signal processing unit preprocesses the multi-point vibration signal by filtering out low-frequency drift, power frequency interference, interference from cooling fans or oil pumps, and high-frequency random noise, and performs detrending, scaling, or normalization. The signal processing unit preprocesses the current signal by including multi-level filtering to separate power frequency / harmonic components, avoiding specific interference frequency bands, and suppressing background noise.

7. The system according to claim 1, characterized in that, The vibration mode characteristics include the vibration energy and distribution ratio of each measuring point in the key frequency band, the relative phase difference and coherence of signals from different measuring points, modal frequency, damping ratio, mode shape, vibration signal complexity, stability, correlation, energy similarity, effective energy ratio and harmonic ratio. The current characteristics include the effective value of the power frequency circulating current, the effective value of higher harmonics, the harmonic content, and the amplitude, charge, pulse count per unit time, total discharge, PRPD spectrum, and PRPS spectrum characteristics of the high-frequency partial discharge pulse.

8. The system according to claim 1, characterized in that, Dynamic association analysis employs time series cross-correlation analysis, Granger causality analysis, or dynamic association modeling based on machine learning.

9. A method for online monitoring and diagnosis of mechanical instability in a 500kV oil-immersed current-limiting reactor, characterized in that, The method of the online monitoring and diagnostic system for mechanical instability of a 500kV oil-immersed current-limiting reactor, applicable to any one of claims 1 to 8, comprises: The high-precision synchronous acquisition unit acquires multi-point vibration signals and current signals through high-precision time synchronization; The signal processing unit performs signal processing and feature extraction on the multi-point vibration signal to obtain vibration mode features, which characterize the relative motion state of the key internal structure of the reactor. The signal processing unit performs signal processing and feature extraction on the current signal to obtain current features, which characterize electrical abnormal states. The correlation analysis unit establishes and continuously analyzes the dynamic correlation between the vibration mode characteristics and the current characteristics over time. Based on the vibration mode characteristics, current characteristics, and dynamic correlation analysis results, the diagnostic and early warning unit uses a preset diagnostic model to perform state assessment, early warning, and fault diagnosis of the mechanical stability of the reactor.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the online monitoring and diagnosis method for mechanical instability of a 500kV oil-immersed current-limiting reactor as described in claim 9.

Images