Fault monitoring system, method, device and medium for low frequency on-load tap changer

Abstract

The application discloses a kind of low-frequency load tap-changer fault monitoring system, method, equipment and medium, it is related to power equipment state monitoring technical field, system includes: data acquisition module, for collecting the electrical parameter of low-frequency load tap-changer, mechanical state parameter and low-frequency working condition special parameter, and as raw data;Signal processing module is used to sequentially carry out pre-processing, noise reduction and feature extraction to raw data, obtain the feature set after processing;Data analysis module is used to carry out anomaly identification and fault prediction to the feature set after processing based on pre-trained machine learning model, and the analysis result containing fault type, fault probability and fault development cycle is output;Fault early warning module is used to generate graded early warning information according to analysis result and push.This application solves the problem that traditional load tap-changer monitoring system can only adapt to power frequency working condition and cannot identify low-frequency working condition specific fault hidden danger.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, and in particular to a fault monitoring system, method, device and medium for low-frequency load tap changers. Background Technology

[0002] Flexible low-frequency (20Hz and below) power transmission technology is a high-efficiency AC power transmission technology based on power electronics. It has significant technical and economic advantages in scenarios such as offshore wind power transmission and is one of the key technology directions for the construction of new power systems. To solve the problems of equipment voltage fluctuation, reactive power and harmonic characteristic optimization in low-frequency flexible power transmission systems, on-load tap changers need to be configured in low-frequency power transmission transformers.

[0003] Compared to traditional 50Hz / 60Hz on-load tap changers for transformers, low-frequency on-load tap changers of 20Hz and below have significant unique operating conditions: the arcing time and inter-taper circulating current time during tap switching are extended by about 2.5 times, which can easily lead to increased erosion of the arc-extinguishing contacts and overheating damage to the transition resistor; at the same time, the impact of low-frequency operating conditions on the switching action sequence, structural design and operating mechanism is not yet clear, and there are no precedents for its load switching test technology, making the development of high-current high-stability test equipment very difficult.

[0004] On-load tap changers are core components of transformer voltage regulation, and their reliability directly affects power system safety. Traditional monitoring methods rely on regular manual inspections, which suffer from low efficiency, high missed detection rates, and inability to capture potential faults in real time. Existing on-load tap changer monitoring systems are mostly designed for power frequency conditions, failing to consider specific parameters such as arcing and circulating current under low-frequency conditions. Furthermore, the monitoring models used have poor generalization capabilities and cannot adapt to the fault characteristics of low-frequency switches, making it difficult to meet the monitoring needs of low-frequency power transmission scenarios. Summary of the Invention

[0005] This invention provides a fault monitoring system, method, device, and medium for low-frequency on-load tap changers, which solves the problem that traditional on-load tap changer monitoring systems can only be adapted to power frequency conditions and cannot identify potential faults specific to low-frequency conditions.

[0006] In view of this, the first aspect of the present invention provides a fault monitoring system for a low-frequency load tap changer, the system comprising:

[0007] The data acquisition module is used to collect electrical parameters, mechanical status parameters, and low-frequency operating condition-specific parameters of the low-frequency on-load tap changer through several different types of sensors, and use them as raw data.

[0008] The signal processing module is used to combine the 3σ criterion, wavelet threshold denoising algorithm and adaptive notch filtering algorithm to preprocess, denoise and extract features from the original data in sequence to obtain the processed feature set.

[0009] The data analysis module is used to perform anomaly identification and fault prediction on the processed feature set based on a pre-trained improved LSTM-random forest hybrid machine learning model, and output analysis results including fault type, fault probability and fault development cycle.

[0010] The fault early warning module is used to generate and push graded early warning information based on the analysis results, wherein the graded early warning information is determined by the fault probability.

[0011] Optionally, it may also include: a data storage module and a human-computer interaction module;

[0012] The data storage module is used to store the original data, the processed feature set, the parameters of the machine learning model, and the hierarchical early warning information;

[0013] The human-computer interaction module is used to visualize the monitoring data and receive operation and maintenance instructions. The monitoring data includes: real-time operating parameter curves generated based on the original data, historical fault statistics reports generated based on the graded early warning information, and fault early warning trend charts generated based on the graded early warning information stored in the data storage module.

[0014] Optionally, the data acquisition module includes a high-precision current transformer, a high-precision voltage sensor, a temperature sensor, a vibration sensor, an arc light sensor, and a circulating current monitoring sensor;

[0015] The high-precision current transformer and high-precision voltage sensor have an acquisition frequency not lower than a preset threshold, and are used to monitor the real-time current and voltage when the low-frequency load tap changer is working.

[0016] The temperature sensor is a patch-type fiber optic temperature sensor, which is disposed on the surface of the arc-extinguishing contact and the transition resistor, and is used to collect the temperature of the arc-extinguishing contact and the transition resistor.

[0017] The vibration sensor is a triaxial acceleration vibration sensor, which is installed in the housing of the operating mechanism and is used to collect mechanical vibration signals during the operation of the low-frequency load tap changer.

[0018] The arc light sensor is used to collect the arc light intensity and arc duration when the low-frequency tap changer switches are switched.

[0019] The circulating current monitoring sensor is used to collect the peak value and duration of the circulating current between the tap positions of the low-frequency load tap changer.

[0020] Optionally, the signal processing module includes a preprocessing unit, a noise reduction unit, and a feature extraction unit connected in sequence;

[0021] The preprocessing unit is used to complete the missing data in the original data by linear interpolation and to remove outlier values ​​in the original data by the 3σ criterion to obtain the preprocessed data.

[0022] The noise reduction unit is used to eliminate environmental electromagnetic interference noise in the preprocessed data through a wavelet threshold denoising algorithm, and then filter out power frequency harmonic interference through an adaptive notch filter algorithm to obtain the noise-reduced data.

[0023] The feature extraction unit includes a time-domain feature extraction subunit, a frequency-domain feature extraction subunit, and a low-frequency operating condition-specific feature extraction subunit, which are respectively used to extract time-domain features, frequency-domain features, and low-frequency operating condition-specific features from the noise-reduced data.

[0024] The time-domain features include peak current and voltage, effective value of vibration acceleration, and average temperature; the frequency-domain features include harmonic content of current and voltage and dominant frequency of vibration signal; and the low-frequency operating condition-specific features include arc duration, peak circulating current, and temperature rise rate of transition resistance.

[0025] Optionally, the data analysis module includes a model training unit and a fault identification and prediction unit;

[0026] The model training unit is used to pre-train an improved LSTM-random forest hybrid machine learning model;

[0027] The fault identification and prediction unit is used to input the features extracted in real time by the feature extraction unit into the pre-trained improved LSTM-random forest hybrid machine learning model, and output the fault type, fault probability and fault development cycle.

[0028] Optionally, the training process of the improved LSTM-random forest hybrid machine learning model is as follows:

[0029] Using the feature set corresponding to historical fault data as training samples, the temporal correlation information of the training samples is extracted through the LSTM network. The temporal correlation information output by the LSTM network is then input into a random forest classifier to complete fault type classification and fault development trend prediction.

[0030] Optionally, the fault early warning module includes an early warning classification unit and an information push unit;

[0031] The early warning classification unit is used to classify the fault early warning level into three levels: Level I, Level II, and Level III. Level I corresponds to a potential hidden danger with a fault probability of 30%-50%, Level II corresponds to a moderate fault risk with a fault probability of 50%-80%, and Level III corresponds to an emergency fault risk with a fault probability of ≥80%.

[0032] The information push unit is used to push the graded early warning information through three methods: push via the operation and maintenance center platform, push via the mobile APP of operation and maintenance personnel, and push via on-site audible and visual alarms, and different fault early warning levels correspond to different push priorities.

[0033] A second aspect of the present invention provides a fault monitoring method for a low-frequency load tap changer, the method comprising:

[0034] Electrical parameters, mechanical status parameters, and low-frequency operating condition-specific parameters of low-frequency on-load tap changers are collected by several different types of sensors and used as raw data.

[0035] The original data is preprocessed, denoised, and feature extracted sequentially by combining the 3σ criterion, wavelet threshold denoising algorithm, and adaptive notch filtering algorithm to obtain the processed feature set.

[0036] An improved LSTM-random forest hybrid machine learning model based on pre-training is used to identify anomalies and predict faults in the processed feature set, and output analysis results including fault type, fault probability and fault development cycle.

[0037] Based on the analysis results, a graded early warning information is generated and pushed out, wherein the graded early warning information is determined by the failure probability.

[0038] A third aspect of the present invention provides a fault monitoring device for a low-frequency load tap changer, the device comprising a processor and a memory:

[0039] The memory is used to store program code and transmit the program code to the processor;

[0040] The processor is configured to execute the steps of the fault monitoring method for low-frequency load tap changers as described in the second aspect above, according to the instructions in the program code.

[0041] A fourth aspect of the present invention provides a computer-readable storage medium for storing program code for executing the fault monitoring method for a low-frequency load tap changer described in the second aspect above.

[0042] As can be seen from the above technical solutions, the present invention has the following advantages:

[0043] This invention provides a fault monitoring system for low-frequency on-load tap changers. 1) It is specifically adapted to the operating characteristics of low-frequency conditions at 20Hz and below, accurately capturing the arcing and circulating current parameters of the low-frequency on-load tap changer during tap switching. This solves the problem that traditional on-load tap changer monitoring systems can only adapt to power frequency conditions and cannot identify specific fault hazards under low-frequency conditions, filling the technical gap in specialized monitoring of low-frequency on-load tap changers in China and providing a dedicated monitoring solution for the safe operation of key components of low-frequency power transmission transformers; 2) This application integrates multiple types of sensors for electrical, temperature, mechanical vibration, and low-frequency specific parameters, comprehensively covering the operating status of the core components of the switch. It can monitor conventional electrical indicators such as current and voltage, as well as capture key status information such as arc-extinguishing contact temperature, operating mechanism vibration, and arcing duration, eliminating the drawbacks of manual inspection and single-parameter monitoring. 3) The monitoring blind spots are eliminated, ensuring comprehensive control over the operating status of switches; 4) The signal processing architecture with multiple algorithms is used to achieve efficient purification of monitoring data. At the same time, the improved hybrid machine learning model is used to mine the temporal correlation features of the data, which can accurately identify the unique fault types under low-frequency operating conditions and make advance predictions on the fault development trend. Compared with the qualitative analysis mode of traditional monitoring methods, the accuracy of fault identification is greatly improved, and the probability of fault omission and misjudgment is effectively reduced; 5) A graded early warning system is built based on the degree of fault risk and a differentiated information push strategy is configured. This allows maintenance personnel to quickly determine the urgency of the fault according to the early warning level and formulate targeted handling plans. This avoids the waste of maintenance resources caused by indiscriminate alarms and ensures that high-risk fault information is delivered in a timely manner, leaving sufficient time for fault handling and significantly reducing the possibility of the fault escalating into a major power accident. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the structure of a fault monitoring system for a low-frequency load tap changer provided in an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the signal processing module provided in an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of the structure of the data analysis module provided in an embodiment of the present invention;

[0048] Figure 4This is a schematic diagram of the fault early warning module provided in an embodiment of the present invention;

[0049] Figure 5 This is a schematic diagram of the structure of the human-computer interaction module provided in an embodiment of the present invention;

[0050] Figure 6 This is a flowchart illustrating a fault monitoring method for a low-frequency load tap changer provided in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0052] Example 1:

[0053] Please see Figure 1 The present invention provides a fault monitoring system for a low-frequency load tap changer, comprising: a data acquisition module 1, a signal processing module 2, a data analysis module 3, and a fault early warning module 4; wherein:

[0054] Data acquisition module 1 is used to collect electrical parameters, mechanical status parameters, and low-frequency operating condition-specific parameters of the low-frequency on-load tap changer through several different types of sensors, and use them as raw data.

[0055] In one embodiment, the data acquisition module 1 includes: a high-precision current transformer, a high-precision voltage sensor, a temperature sensor, a vibration sensor, an arc light sensor, and a circulating current monitoring sensor.

[0056] The high-precision current transformer and high-precision voltage sensor have an acquisition frequency of not less than 1kHz, which are used to monitor the real-time current and voltage when the low-frequency load tap changer is working.

[0057] The temperature sensor is a patch-type fiber optic temperature sensor, which is placed on the surface of the arc-extinguishing contact and the transition resistor to collect the temperature of the arc-extinguishing contact and the transition resistor.

[0058] It should be noted that the arc-extinguishing contact is responsible for extinguishing the arc during tap change and is a key guarantee for the electrical performance of the low-frequency tap changer; the transition resistor is used to suppress circulating current surges during tap change and protect the internal structure of the low-frequency tap changer. The state of both directly determines the operational reliability of the low-frequency tap changer, especially under low-frequency conditions (20Hz and below), where the arcing time and circulating current time during tap change are extended by approximately 2.5 times, easily leading to accelerated arc-extinguishing contact erosion and overheating damage to the transition resistor. Therefore, these two components are high-risk areas for faults under low-frequency conditions. Understandably, arc-extinguishing contact erosion and transition resistor overheating are unique fault types under low-frequency conditions, and abnormal temperature rise is a direct precursor to these faults (e.g., excessively high arc-extinguishing contact temperature leading to erosion, and excessively high transition resistor temperature leading to overheating damage). Temperature monitoring allows for real-time monitoring of potential hazards. Furthermore, temperature data is a core input for feature extraction in signal processing: the average temperature in the time domain features; and the transition resistor temperature rise rate in the low-frequency condition-specific features.

[0059] These features are used for fault identification and prediction in the improved LSTM-random forest hybrid model, which can accurately output fault type (such as contact erosion, overheating of transition resistance), probability and development cycle.

[0060] The vibration sensor is a triaxial accelerometer vibration sensor, which is installed in the housing of the operating mechanism to collect mechanical vibration signals during the operation of the low-frequency load tap changer;

[0061] It should be noted that the operating mechanism is the core mechanical component of the low-frequency load tap changer, responsible for driving the switch to complete the electrical connection switching action. The signals collected by the vibration sensor installed in the housing of the operating mechanism directly reflect the mechanical state during the switch operation (such as the accuracy of the action timing, the degree of wear of internal components, structural stability, etc.), and are a direct reflection of the health of the switch mechanical system.

[0062] An arc light sensor is used to collect the intensity and duration of the arc light during the switching of a low-frequency tap changer.

[0063] It should be noted that by collecting the arc intensity and duration during low-frequency tap changer position switching, the intensity and energy release of the arc discharge during the switching process can be effectively reflected. When faults such as poor contact, deterioration of the arc-extinguishing chamber performance, or insulation aging occur inside the switch, the arc intensity will significantly increase, and the arc duration will correspondingly prolong. Therefore, these parameters collected by the arc light sensor are key evidence for evaluating the switch's arc-extinguishing performance and electrical contact status, and can promptly detect potential fault risks caused by arc anomalies, providing important characteristic information for subsequent fault diagnosis.

[0064] The circulating current monitoring sensor is used to collect the peak value and duration of the circulating current between the tap positions of a low-frequency load tap changer.

[0065] It should be noted that "taper" refers to the different operating positions (or graded positions) of the low-frequency on-load tap changer used to adjust the transformer output voltage. The switch adjusts the voltage by changing the winding turns ratio. The circulating current between taps is the current parameter generated during the switching process. Understandably, the circulating current monitoring sensor collects the peak value and duration of the circulating current during tap switching to provide data support for determining whether the transition state during switch switching is normal. For example, when the circulating current peak value is abnormally large or the duration is too long, it may indicate potential faults such as poor contact of the internal contacts, aging of the transition resistance, or deterioration of insulation performance. Real-time monitoring and analysis of these parameters can help detect potential faults in a timely manner and prevent further escalation of the fault.

[0066] Understandably, data acquisition module 1 comprehensively collects key parameters of the equipment's operating status through a variety of dedicated sensors: specifically, these include a high-precision current / voltage sensor (collecting data at a frequency of no less than 1kHz, monitoring real-time current and voltage), a patch-type fiber optic temperature sensor (deployed on the surface of the arc-extinguishing contact and transition resistor, collecting temperature to monitor for potential ablation or overheating hazards; its data provides input for the average temperature and time-domain specific features in feature extraction), a triaxial acceleration vibration sensor (deployed on the operating mechanism housing, collecting mechanical vibration signals to reflect mechanical conditions such as action sequence and component wear), an arc light sensor (collecting the intensity and duration of the arc light during gear switching to assess arc extinguishing performance and electrical contact status), and a circulating current monitoring sensor (collecting the peak value and duration of the circulating current between gears to determine whether the switching transition state is normal). These multi-dimensional data collectively constitute the raw monitoring data, specifically adapted to the characteristics of prolonged arcing and circulating current time under low-frequency operating conditions of 20Hz and below.

[0067] Signal processing module 2 is used to combine the 3σ criterion, wavelet threshold denoising algorithm and adaptive notch filtering algorithm to preprocess, denoise and extract features from the original data in sequence to obtain the processed feature set;

[0068] Please see Figure 2 In one embodiment, the signal processing module 2 includes a preprocessing unit 201, a noise reduction unit 202, and a feature extraction unit 203 connected in sequence.

[0069] The preprocessing unit 201 is used to complete the missing data in the original data by linear interpolation and remove outliers in the original data by the 3σ criterion to obtain the preprocessed data.

[0070] It should be noted that linear interpolation is a method used to complete missing values ​​in a data sequence. The specific strategy for completing missing data using linear interpolation in this invention is as follows: For the missing parameter values ​​in the acquisition sequences of each sensor in data acquisition module 1... If adjacent valid data points are known and ,and The formula for calculating missing values ​​is: The original acquisition sequence is traversed, and the above interpolation calculation is performed on the data points that are determined to be missing to complete the time series of parameters such as current, voltage, and temperature, thus ensuring data integrity.

[0071] The 3σ criterion is an outlier detection method based on the statistical properties of data. The specific strategy for removing outliers using the 3σ criterion in this invention is as follows: first calculate the parameter sequence... mean and standard deviation The formula is: If data points satisfy If it is an outlier, it will be identified as an abnormal outlier and removed.

[0072] Understandably, the key function of the preprocessing unit 201 is to preprocess the raw data, using linear interpolation to fill in missing data and the 3σ criterion to remove outliers, ensuring data integrity and accuracy, ultimately obtaining preprocessed data. Specifically, the linear interpolation method addresses missing parameter values ​​in each sensor acquisition sequence by using known adjacent valid data points to calculate and fill in the missing values ​​using a linear formula. For example, if adjacent points (… )and( If the missing point x is obtained by..., then the y-value of the missing point x is obtained by... The calculation process iterates through the original sequence to complete the time series of parameters such as current, voltage, and temperature, ensuring data integrity. The 3σ criterion first calculates the mean μ and standard deviation σ of the parameter sequence. If a data point x satisfies |x-μ|>3σ, it is identified as an outlier and removed, thereby filtering out interference data in the original data.

[0073] The noise reduction unit 202 is used to eliminate environmental electromagnetic interference noise in the preprocessed data through wavelet threshold denoising algorithm, and then filter out power frequency harmonic interference through adaptive notch filtering algorithm to obtain the noise-reduced data.

[0074] It should be noted that wavelet thresholding denoising is an algorithm used to eliminate noise in signals. The wavelet thresholding denoising strategy of this invention is as follows: The db4 wavelet basis is selected, and the preprocessed data sequence is... implement Layer wavelet decomposition yields approximate coefficients. and detail coefficient The detail coefficients are processed using a soft thresholding function, the formula of which is: ,in For the threshold, For data length, The noise standard deviation is used; finally, the denoised sequence is obtained through wavelet reconstruction. The formula is: .

[0075] The adaptive notch filter strategy is as follows: For power frequency harmonic interference, a second-order notch filter is constructed, with the following transfer function: ,in, , This is the power frequency interference frequency. The sampling frequency; The notch bandwidth adjustment factor is used; the LMS algorithm is employed to adaptively update the filter coefficients, and the weight update formula is as follows: ,in, Step size factor For filtering error, For input data.

[0076] Understandably, the key function of the noise reduction unit 202 is to denoise the preprocessed data. It eliminates two main types of noise interference through a combined algorithm of "wavelet threshold denoising + adaptive notch filtering," ultimately obtaining clean monitoring data. Specifically, the wavelet threshold denoising stage uses the db4 wavelet basis to perform multi-level decomposition of the data sequence, processes the detail coefficients through a soft thresholding function (the threshold calculation formula involves data length and noise standard deviation), and then eliminates environmental electromagnetic interference noise through wavelet reconstruction. The adaptive notch filtering stage targets power frequency harmonic interference, constructing a second-order notch filter (the transfer function includes the power frequency, sampling frequency, and bandwidth adjustment coefficients), and uses the LMS algorithm to adaptively update the filter coefficients (weight updates depend on the step size factor, filtering error, and input data), ultimately filtering out power frequency harmonic interference.

[0077] The feature extraction unit 203 includes: a time-domain feature extraction subunit 2031, a frequency-domain feature extraction subunit 2032, and a low-frequency operating condition-specific feature extraction subunit 2033, which are used to extract time-domain features, frequency-domain features, and low-frequency operating condition-specific features from the denoised data, respectively.

[0078] Among them, the time-domain characteristics include peak current and voltage, effective value of vibration acceleration and average temperature, the frequency-domain characteristics include harmonic content of current and voltage and dominant frequency of vibration signal, and the low-frequency operating condition-specific characteristics include arc duration, peak circulating current and transition resistance temperature rise rate.

[0079] The specific calculation strategy for time-domain characteristics is as follows: calculate the peak current and voltage. RMS value of vibration acceleration Average temperature .

[0080] Frequency domain characteristics: The time-domain signal is transformed to the frequency domain using the Fast Fourier Transform (FFT), as shown in the formula: Extracting current / voltage harmonic content ,in, for Subharmonic amplitude; dominant frequency of vibration signal .

[0081] Low-frequency specific feature: Calculating the duration of arcing Peak circulation ; Transition resistance temperature rise rate .

[0082] The above 18-dimensional features are extracted from the full dataset after noise reduction to form a feature set. This provides quantitative input for subsequent fault identification. It should be noted that the 18-dimensional features include: 1. Time-domain features (4 dimensions), including peak current, peak voltage, effective value of vibration acceleration, and average temperature, covering the basic time-series features of electrical parameters, mechanical vibration, and temperature status; 2. Frequency-domain features (11 dimensions), extracted through Fast Fourier Transform (FFT): current / voltage harmonic content (such as the amplitude of the 1st to 5th harmonics, a total of 10 dimensions);

[0083] Vibration signal main frequency (1-dimensional) reflects the frequency characteristics of mechanical vibration; 3. Low-frequency working condition exclusive features (3-dimensional) are designed for low-frequency scenarios, including arc duration, circulating current peak value, and transition resistance temperature rise rate, to capture the unique fault hazards of gear switching at low frequencies. The above three types of features total 18 dimensions.

[0084] Understandably, the feature extraction unit 203 extracts key features from the denoised data through three sub-units: time domain, frequency domain, and low-frequency operating condition, forming an 18-dimensional feature set to provide quantitative input for fault identification. Among them, the time domain features (4 dimensions) include peak current and voltage, effective value of vibration acceleration, and average temperature, reflecting the basic time-series characteristics of electrical, mechanical, and temperature; the frequency domain features (11 dimensions) convert the time domain signal to the frequency domain through FFT, extracting the 1st to 5th harmonic content of current / voltage (10 dimensions) and the main frequency of vibration signal (1 dimension), reflecting frequency characteristics; the low-frequency operating condition-specific features (3 dimensions) are designed for low-frequency scenarios, including arc duration, circulating current peak value, and transition resistance temperature rise rate, capturing unique fault hazards during gear shifting.

[0085] Data analysis module 3 is used to perform anomaly identification and fault prediction on the processed feature set based on a pre-trained improved LSTM-random forest hybrid machine learning model, and output analysis results including fault type, fault probability and fault development cycle.

[0086] Please see Figure 3 In one embodiment, the data analysis module 3 includes a model training unit 301 and a fault identification and prediction unit 302;

[0087] Model training unit 301 is used to pre-train an improved LSTM-random forest hybrid machine learning model;

[0088] The fault identification and prediction unit 302 is used to input the features extracted in real time by the feature extraction unit into the pre-trained improved LSTM-random forest hybrid machine learning model, and output the fault type, fault probability and fault development cycle.

[0089] The training process of the improved LSTM-random forest hybrid machine learning model is as follows: using the feature set corresponding to historical fault data as training samples, extracting the temporal correlation information of the training samples through the LSTM network, and then inputting the temporal correlation information output by the LSTM network into the random forest classifier to complete the fault type classification and fault development trend prediction.

[0090] It should be noted that the improved LSTM-random forest hybrid machine learning model of this invention achieves temporal feature mining and accurate fault classification through mathematical formulas. The specific strategy is as follows:

[0091] LSTM network (temporal feature extraction, i.e., extracting temporal correlation information of training samples).

[0092] LSTM networks achieve the memorization and transmission of temporal features through forget gates, input gates, cell states, and output gates. The calculation formulas for each core unit are as follows:

[0093] Forget gate: controls the degree to which historical cell states are preserved, and the formula is: .

[0094] Input gate: Determines the input weights of new information, using the following formula: , .

[0095] Cell state update: This combines historical states with new information, and the formula is as follows: .

[0096] Output gate: Generates the hidden state at the current moment, with the following formula: , .

[0097] in, , , , This is the weight matrix; , , , For bias vectors, It is the Sigmoid activation function. It is the hyperbolic tangent activation function; for Input features at all times, for The state is hidden at all times, which is the time sequence feature.

[0098] feature set Reconstructed into a time-series input based on the time dimension. An LSTM network with one input layer (18 dimensions), two hidden layers (128 neurons each), and one output layer was constructed to output temporal correlation features. To explore the temporal coupling relationship between arc duration and contact temperature.

[0099] Random forest classifier (fault identification and prediction).

[0100] Decision tree node splitting: using the Gini coefficient The formula for measuring node purity is: ,in For the first node Based on the proportion of samples with the same type of fault, the feature with the smallest Gini coefficient is selected to complete the node split.

[0101] Multi-Decision Tree Voting: Construction A decision tree, for the temporal features of the LSTM output. The classification process is performed, and the final fault type is determined by majority vote, using the following formula: ,in For indicator functions (classification results are) (Take 1 if the condition is met, otherwise take 0), and output the failure probability. .

[0102] It should be noted that LSTM (Long Short-Term Memory) is a special type of recurrent neural network (RNN) used in the improved LSTM-random forest hybrid machine learning model of this invention to extract temporal correlation information from training samples. Its core functionality involves the memory and transmission of temporal features through forget gates, input gates, cell states, and output gates: the forget gate controls the retention of historical cell states, the input gate determines the input weights for new information, cell state updates fuse historical states with new information, and the output gate generates the hidden state at the current moment. In specific applications, the model constructs an LSTM network containing an input layer (18 dimensions), two hidden layers (128 neurons each), and one output layer. This reconstructs the feature set into temporal inputs along the time dimension, uncovering temporal coupling relationships such as arc duration and contact temperature, providing key temporal feature support for the subsequent random forest classifier.

[0103] Application method: Based on the timing characteristics corresponding to historical fault data (including five types of faults such as contact erosion, overheating of transition resistance, and mechanical jamming). To train the model, a random forest classifier with 50 decision trees is constructed. Real-time time series features are input into the model, and the model outputs the fault type, fault probability, and fault development cycle.

[0104] Understandably, the key function of data analysis module 3 is to perform anomaly identification and fault prediction on the processed feature set based on a pre-trained improved LSTM-random forest hybrid machine learning model, outputting analysis results including fault type, fault probability, and fault development cycle. This module includes a model training unit 301 and a fault identification and prediction unit 302. The model training unit is responsible for pre-training the hybrid model, while the fault identification and prediction unit inputs the real-time extracted features into the pre-trained model to output results. The model training process is as follows: using the feature set corresponding to historical fault data as training samples, the data is processed through an LSTM network (containing core units such as forget gates and input gates for extracting temporal correlation information), and the output temporal features are input into a random forest classifier (which achieves classification through Gini coefficient split nodes and multi-decision tree voting), ultimately completing fault type classification and development trend prediction.

[0105] The fault warning module 4 is used to generate and push graded warning information based on the analysis results, wherein the graded warning information is determined by the fault probability.

[0106] Please see Figure 4 In one embodiment, the fault warning module includes a warning classification unit 401 and an information push unit 402;

[0107] The early warning classification unit 401 is used to classify the fault early warning level into three levels: Level I, Level II and Level III. Level I corresponds to a potential hidden danger with a fault probability of 30%-50%, Level II corresponds to a moderate fault risk with a fault probability of 50%-80%, and Level III corresponds to an emergency fault risk with a fault probability of ≥80%.

[0108] The information push unit 402 is used to push graded early warning information through three methods: push via the operation and maintenance center platform, push via the mobile APP of operation and maintenance personnel, and push via on-site audible and visual alarms. Different fault early warning levels correspond to different push priorities.

[0109] Understandably, the key function of the fault early warning module 4 is to generate and push graded early warning information based on the analysis results of fault type, probability, and development cycle output by the data analysis module. The graded early warning information is determined by the fault probability. This module includes an early warning grading unit 401 and an information push unit 402: the early warning grading unit 401 divides the early warning level into Level I (fault probability 30%-50%, corresponding to potential hidden dangers), Level II (50%-80%, corresponding to moderate fault risk), and Level III (≥80%, corresponding to emergency fault risk); the information push unit 402 pushes graded early warning information through three methods: the operation and maintenance center platform, the operation and maintenance personnel's mobile APP, and on-site audible and visual alarms. Different early warning levels correspond to different push priorities to ensure that high-risk fault information reaches operation and maintenance personnel first.

[0110] Furthermore, in one embodiment, the fault monitoring system for the low-frequency load tap changer of the present invention further includes: a human-machine interaction module 5 and a data storage module 6; see also Figure 5 .

[0111] The human-computer interaction module 5 is used to visualize the monitoring data and receive operation and maintenance instructions. The monitoring data includes: real-time operating parameter curves generated based on raw data, historical fault statistical reports generated based on hierarchical early warning information, and fault early warning trend charts generated based on hierarchical early warning information stored in the data storage module.

[0112] It should be noted that, specifically, the human-computer interaction module 5 includes a data visualization unit 501 and an instruction interaction unit 502; the data visualization unit 501 can display the real-time operating parameter curves of the switch, historical fault statistics reports, and fault warning trend charts; the instruction interaction unit 502 supports maintenance personnel to issue model parameter update instructions, warning threshold adjustment instructions, and data export instructions.

[0113] Understandably, the key function of the human-machine interaction module 5 is to realize the visualization of monitoring data and the two-way interaction of operation and maintenance commands, specifically including a data visualization unit 501 and a command interaction unit 502. The data visualization unit 501 can display real-time operating parameter curves generated based on raw data, historical fault statistical reports generated based on hierarchical early warning information, and fault early warning trend charts generated based on hierarchical early warning information stored in the data storage module, intuitively presenting the equipment's operating status and fault history. The command interaction unit 502 supports operation and maintenance personnel in issuing model parameter update commands, early warning threshold adjustment commands, and data export commands, enabling dynamic configuration and data management of the system.

[0114] Data storage module 6 is used to store raw data, processed feature sets, parameters of machine learning models, and hierarchical early warning information;

[0115] It should be noted that the data storage module 6 adopts a dual storage architecture of edge storage and cloud storage. The edge storage stores the real-time collected data of the past 7 days, such as raw data; the cloud storage stores the historical full data, such as processed feature sets, and model training data, such as parameters of machine learning models; and the dual-end data supports bidirectional synchronization and breakpoint resume.

[0116] Understandably, the key function of data storage module 6 is to store critical data during system operation, including raw data, processed feature sets, machine learning model parameters, and tiered early warning information. This module employs a dual-storage architecture combining edge and cloud storage: the edge stores the most recent seven days of real-time collected data (such as raw data) to ensure rapid access to real-time information; the cloud stores historical full data (such as processed feature sets) and model training data (such as machine learning model parameters) to meet the needs of long-term data archiving and model iteration. Simultaneously, both ends support bidirectional synchronization and breakpoint resume functionality, ensuring data integrity and traceability.

[0117] This invention provides a fault monitoring system for low-frequency on-load tap changers. 1) It is specifically adapted to the operating characteristics of low-frequency conditions at 20Hz and below, accurately capturing the arcing and circulating current parameters of the low-frequency on-load tap changer during tap switching. This solves the problem that traditional on-load tap changer monitoring systems can only adapt to power frequency conditions and cannot identify specific fault hazards under low-frequency conditions, filling the technical gap in specialized monitoring of low-frequency on-load tap changers in China and providing a dedicated monitoring solution for the safe operation of key components of low-frequency power transmission transformers; 2) This application integrates multiple types of sensors for electrical, temperature, mechanical vibration, and low-frequency specific parameters, comprehensively covering the operating status of the core components of the switch. It can monitor conventional electrical indicators such as current and voltage, as well as capture key status information such as arc-extinguishing contact temperature, operating mechanism vibration, and arcing duration, eliminating the drawbacks of manual inspection and single-parameter monitoring. 3) The monitoring blind spots are eliminated, ensuring comprehensive control over the operating status of switches; 4) The signal processing architecture with multiple algorithms is used to achieve efficient purification of monitoring data. At the same time, the improved hybrid machine learning model is used to mine the temporal correlation features of the data, which can accurately identify the unique fault types under low-frequency operating conditions and make advance predictions on the fault development trend. Compared with the qualitative analysis mode of traditional monitoring methods, the accuracy of fault identification is greatly improved, and the probability of fault omission and misjudgment is effectively reduced; 5) A graded early warning system is built based on the degree of fault risk and a differentiated information push strategy is configured. This allows maintenance personnel to quickly determine the urgency of the fault according to the early warning level and formulate targeted handling plans. This avoids the waste of maintenance resources caused by indiscriminate alarms and ensures that high-risk fault information is delivered in a timely manner, leaving sufficient time for fault handling and significantly reducing the possibility of the fault escalating into a major power accident.

[0118] Example 2:

[0119] Please see Figure 6 The present invention provides a fault monitoring method for a low-frequency load tap changer, which is applied to a fault monitoring system for a low-frequency load tap changer. The method includes:

[0120] Step 201: Collect electrical parameters, mechanical status parameters, and low-frequency operating condition-specific parameters of the low-frequency on-load tap changer using several different types of sensors, and use them as raw data;

[0121] It should be noted that by configuring various specialized sensors, key status data of low-frequency on-load tap changers during operation are comprehensively collected, providing a foundation for subsequent fault monitoring. For example, the system deploys different types of equipment, such as high-precision current / voltage sensors, patch-type fiber optic temperature sensors, triaxial acceleration vibration sensors, arc light sensors, and circulating current monitoring sensors. These devices collect real-time current / voltage data reflecting the electrical performance of the switch, vibration signals reflecting the mechanical structure status, and unique parameters specific to low-frequency operating conditions, such as arc light intensity / duration and peak / duration of circulating current between taps. These multi-dimensional data together constitute the raw monitoring data, providing comprehensive input for subsequent signal processing and fault analysis.

[0122] Step 202: Combine the 3σ criterion, wavelet threshold denoising algorithm and adaptive notch filtering algorithm to preprocess, denoise and extract features from the original data in sequence to obtain the processed feature set;

[0123] It should be noted that, firstly, outliers are removed and missing data is filled in using the 3σ criterion to complete the preprocessing. Then, wavelet threshold denoising algorithm is used to eliminate environmental electromagnetic interference. Next, an adaptive notch filter algorithm is used to filter out power frequency harmonic interference to achieve noise reduction. Finally, time domain, frequency domain and low-frequency operating condition specific features are extracted from the denoised data to form the processed feature set.

[0124] Step 203: Based on the pre-trained improved LSTM-random forest hybrid machine learning model, perform anomaly identification and fault prediction on the processed feature set, and output analysis results including fault type, fault probability and fault development cycle.

[0125] It should be noted that, firstly, the improved LSTM-random forest hybrid machine learning model has been pre-trained using historical fault data. Its structure combines the ability of LSTM networks to capture time-series data with the classification advantages of random forests. Next, the feature set after preprocessing, noise reduction, and feature extraction is input into the model. The LSTM network is responsible for extracting the time-series correlation information (such as the change pattern of parameters over time) from the feature set. This information is then input into the random forest classifier, ultimately achieving anomaly identification and fault prediction for low-frequency on-load tap changers, outputting specific fault types (such as contact erosion, overheating of transition resistance, etc.), the probability of fault occurrence, and the development cycle from potential hazards to actual occurrence of the fault.

[0126] Step 204: Generate and push graded early warning information based on the analysis results, wherein the graded early warning information is determined by the fault probability.

[0127] It should be noted that, firstly, the warning level is determined based on the probability of failure. For example, it is divided into Level I (30%-50% probability of failure, corresponding to potential hidden dangers), Level II (50%-80%, corresponding to moderate failure risk), and Level III (≥80%, corresponding to emergency failure risk). Subsequently, the graded warning information is pushed through three methods: the operation and maintenance center platform, the operation and maintenance personnel's mobile APP, and on-site audible and visual alarms. Different warning levels correspond to different push priorities to ensure that high-risk failure information can reach the operation and maintenance personnel first, so as to formulate targeted handling plans.

[0128] In one embodiment, step 204 is followed by:

[0129] Step 205: Store the original data, the processed feature set, the parameters of the machine learning model, and the hierarchical early warning information;

[0130] It should be noted that four types of key data are persistently stored: raw data collected by sensors, feature sets extracted after signal processing, parameters of pre-trained machine learning models, and tiered early warning information generated by the fault warning module. This storage mechanism not only ensures data integrity and traceability but also provides data support for subsequent model iteration and optimization, historical fault analysis, and system operation status retrospective. Furthermore, it employs a dual-storage architecture (edge ​​storage for recent real-time data and cloud storage for all historical data), balancing data access efficiency with long-term archiving requirements.

[0131] Step 206: Visualize the monitoring data and receive operation and maintenance instructions, wherein the monitoring data includes: real-time operating parameter curves generated based on the original data, historical fault statistics reports generated based on the graded early warning information, and fault early warning trend charts generated based on the graded early warning information stored in the data storage module.

[0132] It should be noted that this step presents monitoring data and receives operation and maintenance instructions in a visual manner: on the one hand, it transforms raw data into real-time operating parameter curves to intuitively display the current status of the equipment, generates historical fault statistical reports based on hierarchical early warning information to trace fault patterns, and generates fault early warning trend charts in combination with early warning data to predict risk changes; on the other hand, it also supports operation and maintenance personnel to issue instructions such as model parameter updates, early warning threshold adjustments, and data export, realizing two-way human-machine interaction, which not only meets the needs of visual monitoring of equipment operating status, but also provides an operation entry point for operation and maintenance decisions.

[0133] In one embodiment, step 201 includes:

[0134] Step 2011: Monitor the real-time current and voltage of the low-frequency load tap changer during operation;

[0135] Step 2012: Collect the temperatures of the arc-extinguishing contact and the transition resistor;

[0136] Step 2013: Collect mechanical vibration signals during the operation of the low-frequency load tap changer;

[0137] Step 2014: Collect the arc intensity and arc duration during low-frequency load tap changer position switching;

[0138] Step 2015: Collect the peak value and duration of the circulating current between the tap positions of the low-frequency load tap changer.

[0139] It should be noted that step 201 comprehensively collects key parameters of the equipment's operating status through five sub-steps: First, it monitors the real-time current and voltage during switch operation to obtain basic electrical performance data; second, it collects the temperature of the arc-extinguishing contact and transition resistor, as these two components are high-risk areas for failure under low-frequency conditions, and temperature data can directly reflect potential ablation or overheating hazards; next, it collects mechanical vibration signals during operation to assess the mechanical condition of the operating mechanism (such as action sequence, component wear, etc.); simultaneously, it collects the arc intensity and duration during gear switching to capture the prolonged arc characteristics at low frequencies to determine arc-extinguishing performance; finally, it collects the peak value and duration of the circulating current between gears to monitor the circulating current impact during switching. These multi-dimensional data together constitute the raw monitoring data.

[0140] In one embodiment, step 202 includes:

[0141] Step 2021: Complete the missing data in the original data by linear interpolation and remove outliers from the original data by the 3σ criterion to obtain the preprocessed data;

[0142] Step 2022: After eliminating environmental electromagnetic interference noise in the preprocessed data using the wavelet threshold denoising algorithm, the power frequency harmonic interference is filtered out using the adaptive notch filter algorithm to obtain the denoised data.

[0143] Step 2023: Extract time-domain features, frequency-domain features, and low-frequency operating condition-specific features from the noise-reduced data;

[0144] The time-domain features include peak current and voltage, effective value of vibration acceleration, and average temperature; the frequency-domain features include harmonic content of current and voltage and dominant frequency of vibration signal; and the low-frequency operating condition-specific features include arc duration, peak circulating current, and temperature rise rate of transition resistance.

[0145] It should be noted that step 202 achieves the purification and feature extraction of the original data through three sub-steps: First, in the preprocessing stage, missing values ​​in the original data are filled in using linear interpolation, and outliers are removed using the 3σ criterion to ensure data integrity and accuracy, resulting in preprocessed data; next, in the denoising stage, environmental electromagnetic interference noise in the preprocessed data is first eliminated using a wavelet threshold denoising algorithm, and then power frequency harmonic interference is filtered out using an adaptive notch filter algorithm, further improving data quality, resulting in denoised data; finally, in the feature extraction stage, three types of key features are extracted from the denoised data, including time-domain features reflecting basic time-series characteristics (peak current and voltage, effective value of vibration acceleration, and average temperature), frequency-domain features reflecting frequency characteristics (harmonic content of current and voltage, dominant frequency of vibration signal), and exclusive features designed for low-frequency scenarios (arc duration, peak circulating current, and temperature rise rate of transition resistance), forming a feature set for fault identification.

[0146] In one embodiment, step 203 includes:

[0147] Step 2031: Pre-train the improved LSTM-random forest hybrid machine learning model;

[0148] Step 2032: Input the features extracted in real time by the feature extraction unit into the pre-trained improved LSTM-random forest hybrid machine learning model, and output the fault type, fault probability and fault development cycle.

[0149] The training process of the improved LSTM-random forest hybrid machine learning model is as follows: using the feature set corresponding to historical fault data as training samples, extracting the temporal correlation information of the training samples through the LSTM network, and then inputting the temporal correlation information output by the LSTM network into the random forest classifier to complete the fault type classification and fault development trend prediction.

[0150] It should be noted that step 203 includes two key steps: model training and fault prediction. First, step 2031 provides the algorithmic foundation for fault identification by pre-training an improved LSTM-random forest hybrid machine learning model. Next, step 2032 inputs the real-time extracted features into the pre-trained model, outputting the fault type, fault probability, and fault development cycle. The model training process specifically involves using the feature set corresponding to historical fault data as training samples. First, the LSTM network extracts the temporal correlation information of the samples (such as the changing patterns of parameters over time). Then, this information is input into a random forest classifier, ultimately completing fault type classification and fault development trend prediction, achieving accurate identification and early warning of faults specific to low-frequency operating conditions.

[0151] In one embodiment, step 204 includes:

[0152] Step 2041: Divide the fault warning level into three levels: Level I, Level II and Level III. Level I corresponds to a potential hazard with a fault probability of 30%-50%, Level II corresponds to a moderate fault risk with a fault probability of 50%-80%, and Level III corresponds to an emergency fault risk with a fault probability of ≥80%.

[0153] Step 2042: Push the graded early warning information through three methods: push from the operation and maintenance center platform, push from the operation and maintenance personnel's mobile APP, and push from on-site audible and visual alarms, with different push priorities corresponding to different fault early warning levels.

[0154] It should be noted that step 204 constructs a tiered early warning and push mechanism through two sub-steps: First, step 2041 divides the fault early warning level into Level I, Level II, and Level III. Level I corresponds to a potential hidden danger with a fault probability of 30%-50%, Level II corresponds to a moderate fault risk of 50%-80%, and Level III corresponds to an emergency fault risk of ≥80%, thus achieving risk quantification and tiering based on fault probability. Next, step 2042 pushes tiered early warning information through three methods: push from the operation and maintenance center platform, push from the operation and maintenance personnel's mobile APP, and on-site audible and visual alarms. Different early warning levels correspond to different push priorities, ensuring that high-risk fault information reaches operation and maintenance personnel first, so as to formulate targeted handling plans. This avoids the waste of resources caused by indiscriminate alarms and ensures timely response to emergency faults.

[0155] This invention provides a fault monitoring method for low-frequency on-load tap changers. 1) It is specifically adapted to the operating characteristics of low-frequency conditions at 20Hz and below, accurately capturing the arcing and circulating current parameters of the low-frequency on-load tap changer during tap switching. This solves the problem that traditional on-load tap changer monitoring systems can only adapt to power frequency conditions and cannot identify specific fault hazards under low-frequency conditions, filling the technical gap in domestic low-frequency on-load tap changer monitoring and providing a dedicated monitoring solution for the safe operation of key components of low-frequency power transmission transformers; 2) This application integrates multiple types of sensors for electrical, temperature, mechanical vibration, and low-frequency specific parameters, comprehensively covering the operating status of the core components of the switch. It can monitor conventional electrical indicators such as current and voltage, as well as capture key status information such as arc-extinguishing contact temperature, operating mechanism vibration, and arcing duration, eliminating the drawbacks of manual inspection and single-parameter monitoring. 3) The monitoring blind spots are eliminated, ensuring comprehensive control over the operating status of switches; 4) The signal processing architecture with multiple algorithms is used to achieve efficient purification of monitoring data. At the same time, the improved hybrid machine learning model is used to mine the temporal correlation features of the data, which can accurately identify the unique fault types under low-frequency operating conditions and make advance predictions on the fault development trend. Compared with the qualitative analysis mode of traditional monitoring methods, the accuracy of fault identification is greatly improved, and the probability of fault omission and misjudgment is effectively reduced; 5) A graded early warning system is built based on the degree of fault risk and a differentiated information push strategy is configured. This allows maintenance personnel to quickly determine the urgency of the fault according to the early warning level and formulate targeted handling plans. This avoids the waste of maintenance resources caused by indiscriminate alarms and ensures that high-risk fault information is delivered in a timely manner, leaving sufficient time for fault handling and significantly reducing the possibility of the fault escalating into a major power accident.

[0156] Example 3:

[0157] Embodiment 3 of the present invention also provides a fault monitoring device for a low-frequency load tap changer, the device comprising a processor and a memory:

[0158] The memory is used to store program code and transmit the program code to the processor;

[0159] The processor is used to execute the steps of the fault monitoring method for low-frequency load tap changers as described in the above method embodiments, according to the instructions in the program code.

[0160] It should be noted that the fault monitoring equipment for low-frequency load tap changers mainly consists of a processor and a memory:

[0161] Memory: Responsible for storing two types of key data: first, the program code that implements the fault monitoring function (including complete logic such as data acquisition, signal processing, model inference, and early warning generation); and second, dynamic data generated during operation (such as raw sensor data, preprocessed feature sets, machine learning model parameters, and tiered early warning information). It adopts a dual-storage architecture of edge and cloud. The edge stores the most recent 7 days of real-time data to ensure fast access, while the cloud stores historical full data and model training data, supporting bidirectional synchronization and breakpoint resume. Processor: As the computing core of the device, it receives the program code transmitted from the memory and executes instructions to drive the entire monitoring process.

[0162] Specific execution process:

[0163] The processor performs fault monitoring according to the instructions in the program code and the steps of the method embodiment, including:

[0164] Data acquisition phase: The control data acquisition module (including high-precision current / voltage sensors, fiber optic temperature sensors, vibration sensors, etc.) acquires the electrical parameters (current, voltage), mechanical state parameters (vibration signals), and low-frequency specific parameters (arc duration, circulating current peak value, etc.) of the low-frequency on-load tap changer. The raw data is temporarily stored in the memory.

[0165] Signal processing stage: The preprocessing algorithm (linear interpolation to complete missing data, 3σ criterion to remove outliers), the noise reduction algorithm (wavelet thresholding to eliminate electromagnetic interference, adaptive notch filtering to remove power frequency harmonics) and the feature extraction logic (extracting time-domain features such as current peak value, frequency-domain features such as harmonic content, and low-frequency specific features such as transition resistance temperature rise rate) are called from the memory. The processed feature set is then stored back into the memory.

[0166] Data analysis phase: The pre-trained improved LSTM-random forest hybrid model is loaded into memory, and the real-time feature set is input into the model. The LSTM network extracts temporal correlation information (such as the coupling relationship between arc duration and contact temperature), and the random forest classifier outputs the fault type (such as contact ablation, overheating of transition resistance), fault probability, and development cycle. The analysis results are written to memory in real time.

[0167] Fault warning stage: Based on the analysis results in the memory, graded warning information is generated according to the fault probability (Level I 30%-50%, Level II 50%-80%, Level III ≥80%), and pushed through the operation and maintenance platform, mobile APP, on-site audible and visual alarms, etc., and the push records are stored synchronously.

[0168] Through the above process, the equipment achieves fully automated monitoring from data collection to early warning push.

[0169] This invention provides a fault monitoring device for low-frequency on-load tap changers. 1) It is specifically adapted to the operating characteristics of low-frequency conditions at 20Hz and below, and can accurately capture the arcing and circulating current related parameters of low-frequency on-load tap changers during tap switching. This solves the problem that traditional on-load tap changer monitoring systems can only adapt to power frequency conditions and cannot identify specific fault hazards under low-frequency conditions, filling the technical gap in domestic low-frequency on-load tap changer monitoring and providing a dedicated monitoring solution for the safe operation of key components of low-frequency power transmission transformers; 2) This application integrates multiple types of sensors for electrical, temperature, mechanical vibration, and low-frequency specific parameters, which can comprehensively cover the operating status of the core components of the switch. It can monitor conventional electrical indicators such as current and voltage, and also capture key status information such as arc-extinguishing contact temperature, operating mechanism vibration, and arcing duration, eliminating the problems caused by manual inspection and single-parameter monitoring. 3) The monitoring blind spots are eliminated, ensuring comprehensive control over the operating status of switches; 4) The signal processing architecture with multiple algorithms is used to achieve efficient purification of monitoring data. At the same time, the improved hybrid machine learning model is used to mine the temporal correlation features of the data, which can accurately identify the unique fault types under low-frequency operating conditions and make advance predictions on the fault development trend. Compared with the qualitative analysis mode of traditional monitoring methods, the accuracy of fault identification is greatly improved, and the probability of fault omission and misjudgment is effectively reduced; 5) A graded early warning system is built based on the degree of fault risk and a differentiated information push strategy is configured. This allows maintenance personnel to quickly determine the urgency of the fault according to the early warning level and formulate targeted handling plans. This avoids the waste of maintenance resources caused by indiscriminate alarms and ensures that high-risk fault information is delivered in a timely manner, leaving sufficient time for fault handling and significantly reducing the possibility of the fault escalating into a major power accident.

[0170] Example 4:

[0171] Embodiment 4 of the present invention also provides a computer-readable storage medium for storing program code, which is used to execute the fault monitoring method for low-frequency load tap changers described in the above method embodiments.

[0172] It should be noted that the computer-readable storage medium is used to store program code, which contains complete logic for implementing a low-frequency load tap changer fault monitoring method. When loaded and executed by the processor, the program code can drive the device to complete steps such as data acquisition (collecting electrical parameters, mechanical state parameters and low-frequency operating condition-specific parameters through sensors), signal processing (preprocessing, noise reduction and feature extraction), data analysis (anomaly identification and fault prediction based on an improved LSTM-random forest hybrid model), and fault warning (generating and pushing hierarchical warning information), thereby realizing the fault monitoring function of the low-frequency load tap changer.

[0173] This invention provides a computer-readable storage medium that: 1) is specifically adapted to the operating characteristics of low-frequency conditions at 20Hz and below, accurately capturing the arcing and circulating current parameters of low-frequency on-load tap changers during tap switching. This solves the problem that traditional on-load tap changer monitoring systems can only adapt to power frequency conditions and cannot identify specific fault hazards under low-frequency conditions, filling the technical gap in domestic monitoring of low-frequency on-load tap changers and providing a dedicated monitoring solution for the safe operation of key components of low-frequency power transmission transformers; 2) integrates multiple types of sensors for electrical, temperature, mechanical vibration, and low-frequency specific parameters, comprehensively covering the operating status of core switch components. It can monitor conventional electrical indicators such as current and voltage, as well as capture key status information such as arc-extinguishing contact temperature, operating mechanism vibration, and arcing duration, eliminating the monitoring difficulties caused by manual inspection and single-parameter monitoring. Blind spots ensure comprehensive control over the operating status of switches; 3) The signal processing architecture combining multiple algorithms achieves efficient purification of monitoring data. At the same time, relying on the improved hybrid machine learning model to mine the temporal correlation features of the data, it can accurately identify the unique fault types under low-frequency operating conditions and make advanced predictions on the development trend of faults. Compared with the qualitative analysis mode of traditional monitoring methods, it greatly improves the accuracy of fault identification and effectively reduces the probability of missed or misjudged faults; 4) A graded early warning system is built based on the degree of fault risk and a differentiated information push strategy is configured. This allows maintenance personnel to quickly determine the urgency of the fault according to the early warning level and formulate targeted handling plans. This avoids the waste of maintenance resources caused by indiscriminate alarms and ensures that high-risk fault information is delivered in a timely manner, leaving sufficient time for fault handling and significantly reducing the possibility of faults escalating into major power accidents.

[0174] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the method described above can be referred to the corresponding process in the aforementioned system embodiments, and will not be repeated here.

[0175] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and 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 through some interfaces, indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0176] 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.

[0177] Furthermore, 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. The integrated unit can be implemented in hardware or as a software functional unit.

[0178] If the integrated unit 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 the present invention, in essence, or the part that contributes to the prior art, or all or 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 the present 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.

[0179] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fault monitoring system for a low-frequency load tap changer, characterized in that, include: The data acquisition module is used to collect electrical parameters, mechanical status parameters, and low-frequency operating condition-specific parameters of the low-frequency on-load tap changer through several different types of sensors, and use them as raw data. The signal processing module is used to combine the 3σ criterion, wavelet threshold denoising algorithm and adaptive notch filtering algorithm to preprocess, denoise and extract features from the original data in sequence to obtain the processed feature set. The data analysis module is used to perform anomaly identification and fault prediction on the processed feature set based on a pre-trained improved LSTM-random forest hybrid machine learning model, and output analysis results including fault type, fault probability and fault development cycle. The fault early warning module is used to generate and push graded early warning information based on the analysis results, wherein the graded early warning information is determined by the fault probability.

2. The fault monitoring system for low-frequency load tap changers according to claim 1, characterized in that, Also includes: Data storage module and human-computer interaction module; The data storage module is used to store the original data, the processed feature set, the parameters of the machine learning model, and the hierarchical early warning information; The human-computer interaction module is used to visualize the monitoring data and receive operation and maintenance instructions. The monitoring data includes: real-time operating parameter curves generated based on the original data, historical fault statistics reports generated based on the graded early warning information, and fault early warning trend charts generated based on the graded early warning information stored in the data storage module.

3. The fault monitoring system for low-frequency load tap changers according to claim 1, characterized in that, The data acquisition module includes a high-precision current transformer, a high-precision voltage sensor, a temperature sensor, a vibration sensor, an arc light sensor, and a circulating current monitoring sensor. The high-precision current transformer and high-precision voltage sensor have an acquisition frequency not lower than a preset threshold, and are used to monitor the real-time current and voltage when the low-frequency load tap changer is working. The temperature sensor is a patch-type fiber optic temperature sensor, which is disposed on the surface of the arc-extinguishing contact and the transition resistor, and is used to collect the temperature of the arc-extinguishing contact and the transition resistor. The vibration sensor is a triaxial acceleration vibration sensor, which is installed in the housing of the operating mechanism and is used to collect mechanical vibration signals during the operation of the low-frequency load tap changer. The arc light sensor is used to collect the arc light intensity and arc duration when the low-frequency tap changer switches are switched. The circulating current monitoring sensor is used to collect the peak value and duration of the circulating current between the tap positions of the low-frequency load tap changer.

4. The fault monitoring system for low-frequency load tap changers according to claim 1, characterized in that, The signal processing module includes a preprocessing unit, a noise reduction unit, and a feature extraction unit connected in sequence. The preprocessing unit is used to complete the missing data in the original data by linear interpolation and to remove outlier values ​​in the original data by the 3σ criterion to obtain the preprocessed data. The noise reduction unit is used to eliminate environmental electromagnetic interference noise in the preprocessed data through a wavelet threshold denoising algorithm, and then filter out power frequency harmonic interference through an adaptive notch filter algorithm to obtain the noise-reduced data. The feature extraction unit includes a time-domain feature extraction subunit, a frequency-domain feature extraction subunit, and a low-frequency operating condition-specific feature extraction subunit, which are respectively used to extract time-domain features, frequency-domain features, and low-frequency operating condition-specific features from the noise-reduced data. The time-domain features include peak current and voltage, effective value of vibration acceleration, and average temperature; the frequency-domain features include harmonic content of current and voltage and dominant frequency of vibration signal; and the low-frequency operating condition-specific features include arc duration, peak circulating current, and temperature rise rate of transition resistance.

5. The fault monitoring system for a low-frequency load tap changer according to claim 4, characterized in that, The data analysis module includes a model training unit and a fault identification and prediction unit; The model training unit is used to pre-train an improved LSTM-random forest hybrid machine learning model; The fault identification and prediction unit is used to input the features extracted in real time by the feature extraction unit into the pre-trained improved LSTM-random forest hybrid machine learning model, and output the fault type, fault probability and fault development cycle.

6. The fault monitoring system for a low-frequency load tap changer according to claim 5, characterized in that, The training process of the improved LSTM-random forest hybrid machine learning model is as follows: Using the feature set corresponding to historical fault data as training samples, the temporal correlation information of the training samples is extracted through the LSTM network. The temporal correlation information output by the LSTM network is then input into a random forest classifier to complete fault type classification and fault development trend prediction.

7. The fault monitoring system for a low-frequency load tap changer according to claim 1, characterized in that, The fault early warning module includes an early warning classification unit and an information push unit; The early warning classification unit is used to classify the fault early warning level into three levels: Level I, Level II, and Level III. Level I corresponds to a potential hidden danger with a fault probability of 30%-50%, Level II corresponds to a moderate fault risk with a fault probability of 50%-80%, and Level III corresponds to an emergency fault risk with a fault probability of ≥80%. The information push unit is used to push the graded early warning information through three methods: push via the operation and maintenance center platform, push via the mobile APP of operation and maintenance personnel, and push via on-site audible and visual alarms, and different fault early warning levels correspond to different push priorities.

8. A fault monitoring method for a low-frequency load tap changer, characterized in that, include: Electrical parameters, mechanical status parameters, and low-frequency operating condition-specific parameters of low-frequency on-load tap changers are collected by several different types of sensors and used as raw data. The original data is preprocessed, denoised, and feature extracted sequentially by combining the 3σ criterion, wavelet threshold denoising algorithm, and adaptive notch filtering algorithm to obtain the processed feature set. An improved LSTM-random forest hybrid machine learning model based on pre-training is used to identify anomalies and predict faults in the processed feature set, and output analysis results including fault type, fault probability and fault development cycle. Based on the analysis results, a graded early warning information is generated and pushed out, wherein the graded early warning information is determined by the failure probability.

9. A fault monitoring device for a low-frequency load tap changer, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the fault monitoring method for the low-frequency load tap changer as described in claim 8 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for executing the fault monitoring method for the low-frequency load tap changer as described in claim 8.

Images