Intelligent monitoring method and system for GIS circuit breaker

Abstract

The application discloses a kind of intelligent monitoring method and system of GIS circuit breaker, the present application is by the sensor of GIS circuit breaker The sensing signal collected by conversion into wave frequency vector and the wave frequency vector of exception is deleted, complete first noise reduction processing;Then using wavelet threshold denoising mode completes secondary noise reduction and carries out dimensional split to its data, obtains multi-source data;Using the construction of feature vector matrix, feature similarity comparison is carried out, further optimize noise, finally through network model completes the monitoring diagnosis of GIS circuit breaker;Solve the technical problem that in the prior art, due to excessive dependence on the collected sensing data itself data, affected by sensor accuracy, when sensing data is abnormal, it will appear misdiagnosis to the monitoring of GIS circuit breaker, interference monitoring diagnosis result, accuracy is low, realize the effective monitoring and rapid fault diagnosis of GIS circuit breaker, improve monitoring and diagnosis accuracy.

Description

Technical Field

[0001] This invention relates to the field of circuit breaker monitoring, and more particularly to an intelligent monitoring method and system for GIS circuit breakers. Background Technology

[0002] GIS circuit breakers are large pieces of equipment in GIS substations and are key components of high-voltage transmission equipment, transformer equipment, transmission lines, and control and protection systems. As an important part of the power system, GIS circuit breakers have advantages such as small footprint, good insulation performance, and high reliability, and are widely used in modern power distribution and transmission systems. However, due to long-term operation and environmental factors, GIS circuit breakers are prone to various potential faults. If these faults are not detected and addressed in a timely manner, they may lead to equipment damage and unstable operation of the power system. Therefore, in the engineering application of GIS circuit breakers, operational monitoring and fault diagnosis are key research directions.

[0003] Current monitoring methods mainly fall into two categories. One involves daily inspections of GIS circuit breakers and periodic recording of the number of operations of the circuit breaker and its operating mechanism, as well as checking the hydraulic oil, to conduct long-term, continuous online monitoring of the mechanical characteristics of the GIS circuit breaker. This allows for comparison and analysis of data trends to troubleshoot equipment failures. However, this traditional monitoring method falls far short of the accuracy and timeliness requirements for fault prediction, often only detecting faults some time after they occur, severely impacting the timeliness of accident handling and resulting in low accuracy. The other method involves installing acoustic fingerprint sensors on the circuit breaker operating mechanism to collect real-time acoustic fingerprint data. This data is then analyzed online in the background to monitor the GIS circuit breaker operating mechanism in real time and analyze and judge its status. However, this sensor data analysis relies heavily on the accuracy of the collected sensor data and is greatly affected by the sensors themselves. Abnormal sensor data can lead to misdiagnosis of the GIS circuit breaker, interfering with the monitoring and diagnostic results, and compromising accuracy. Summary of the Invention

[0004] This invention provides an intelligent monitoring method and system for GIS circuit breakers, enabling effective monitoring and rapid fault diagnosis of GIS circuit breakers, and improving the accuracy of monitoring and diagnosis.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide an intelligent monitoring method and system for GIS circuit breakers, comprising:

[0006] The sensor signals collected by the sensors of the currently monitored GIS circuit breaker are acquired, and the sensor signals are converted into wave frequency vectors and mapped onto a spatial plane. Wave frequency vectors with a spatial distance exceeding a preset distance value on the same horizontal coordinate are deleted to complete the first noise reduction process and obtain multi-source sensor monitoring signal data.

[0007] The multi-source sensor monitoring signal data is subjected to a second denoising process using wavelet threshold denoising to obtain denoised monitoring signal data. The denoised monitoring signal data is then dimensionally split according to different data attribute types to obtain multi-source data.

[0008] The multi-source data is clustered and identified using the K-means clustering algorithm to obtain the sensor data clustering results. Based on the order of identification time, the sensor data clustering results are subjected to a first feature vector extraction from near to far to obtain a first feature vector matrix. The sensor data clustering results are then subjected to a second feature vector extraction from far to near to obtain a second feature vector matrix.

[0009] After flipping the first feature vector matrix along its central axis, a third feature vector matrix is ​​obtained. The similarity value of each row and column between the third feature vector matrix and the second feature vector matrix is ​​calculated. When the similarity value is less than a preset similarity threshold, the row and column are deleted to obtain an optimized feature vector matrix.

[0010] The optimized feature vector matrix is ​​input into a pre-built CNN convolutional neural network model, which outputs the monitoring and diagnostic results of the currently monitored GIS circuit breaker.

[0011] As a preferred embodiment, the multi-source data includes position sensing data, current sensing data, pressure sensing data, temperature sensing data, humidity sensing data, noise sensing data, and gas state sensing data.

[0012] The step of clustering the multi-source data using the K-means clustering algorithm to obtain the sensor data clustering results is as follows:

[0013] A 7-dimensional vector composed of position sensing data, current sensing data, pressure sensing data, temperature sensing data, humidity sensing data, noise sensing data, and gas state sensing data is used as a category object.

[0014] Initialize cluster centers with a preset number of categories; each cluster center is a 7-dimensional vector composed of position sensing data, current sensing data, pressure sensing data, temperature sensing data, humidity sensing data, noise sensing data, and gas state sensing data; the preset number of categories is determined by empirical method or elbow method.

[0015] Calculate new cluster centers based on the category objects and the current cluster centers;

[0016] The clustering process iterates repeatedly to calculate new cluster centers based on the category objects and the current cluster centers. The clustering process continues to iterate until the clustering convergence condition is met. The clustering convergence condition is met when the distance between the new cluster center obtained in the iteration and the previous cluster center is less than a preset distance threshold.

[0017] As a preferred approach, new cluster centers are calculated based on the category objects and the current cluster centers, specifically as follows:

[0018] Based on the distance between the category object and each current cluster center, each category object is assigned to the category of the nearest cluster center;

[0019] The mean of all objects in each class is used as the new cluster center.

[0020] As a preferred embodiment, the steps of performing a first feature vector extraction on the clustering results of the sensor data from near to far according to the chronological order of identification, to obtain a first feature vector matrix, and performing a second feature vector extraction on the clustering results of the sensor data from far to near, to obtain a second feature vector matrix, specifically include:

[0021] Based on the clustering results of the sensor data, extract the cluster center for each category;

[0022] Calculate the Euclidean distance between each category object data and its corresponding cluster center in each category, extract the category object data whose Euclidean distance values ​​fall within a preset range, and arrange them in chronological order of recognition time;

[0023] The sensor data corresponding to each category extracted from near to far are used to form a first feature vector matrix, and the sensor data corresponding to each category extracted from far to near are used to form a second feature vector matrix.

[0024] As a preferred embodiment, the monitoring and diagnostic results include normal operation, gas leakage, excessive moisture in the gas, switch failure, internal discharge of the GIS, and hydraulic mechanism failure; the pre-built CNN convolutional neural network model is obtained by training the neural network based on the sensor feature vector database, specifically as follows:

[0025] Collect normal and fault sensing signals from several GIS circuit breakers, and take each collected signal from the GIS circuit breaker as a multi-source sensing data set, take the multi-source sensing data set of several normal sensing signals as a normal signal set, and take the multi-source sensing data set of several fault sensing signals as a fault signal set.

[0026] Wavelet threshold denoising is used to denoise and normalize the multi-source sensor data groups of each normal signal set and each fault signal set to obtain normal denoised datasets and fault denoised datasets.

[0027] The normal noise reduction dataset and the fault noise reduction dataset are superimposed, and the normal label and the diagnostic fault type label corresponding to each multi-source sensor data group are labeled to form a multi-source sensor signal database.

[0028] Each multi-source sensor data group in the multi-source sensor signal database is clustered using the K-means clustering algorithm to obtain the sensor data clustering result for each multi-source sensor data group. The feature vectors of the sensor data clustering results for each multi-source sensor data group are then extracted to obtain the feature vector matrix for each multi-source sensor data group. Based on the feature vector matrix of each multi-source sensor data group in the multi-source sensor signal database, a sensor feature vector database is obtained.

[0029] According to a preset ratio, the multi-source sensor signal database is divided into a test set and a training set. Based on the training set and the test set, the CNN convolutional neural network model is trained, and the error rate of the current training model is calculated.

[0030] Determine if the current model error rate is less than the preset error rate threshold. If it is not less than the preset error rate threshold, train the CNN convolutional neural network model based on the training set and the test set, and calculate the current training model error rate. Stop training when the current training model error rate is less than the preset error rate threshold, and obtain the pre-built CNN convolutional neural network model.

[0031] As a preferred approach, the CNN convolutional neural network model is trained based on the training and test sets, and the error rate of the current trained model is calculated, specifically as follows:

[0032] The feature vector matrix of each multi-source sensor data group is used as the input sample, and the label corresponding to each multi-source sensor data group is used as the output sample.

[0033] Select a corresponding number of training sample data from the training set according to the preset batch size, and simultaneously input the input and output samples of the training sample data into the CNN convolutional neural network model for training; wherein, the preset batch size is the number of training samples required for one training session.

[0034] Input the input samples from the test set into the currently trained CNN convolutional neural network model to obtain the current model output. Compare the current model output with the output samples corresponding to the current input samples to obtain the error rate of the current trained model.

[0035] As a preferred embodiment, the multi-source sensor monitoring signal data is subjected to a second denoising process using wavelet threshold denoising to obtain denoised monitoring signal data. Furthermore, the denoised monitoring signal data is dimensionally split according to different data attribute types to obtain multi-source data, specifically as follows:

[0036] Extract each sensor signal from the multi-source sensor monitoring signal data as the current processing signal;

[0037] Using the sym4 wavelet, the current signal is decomposed into four levels of wavelets to obtain the scaled wavelet decomposition coefficients; where the scaled wavelet decomposition coefficients include scale coefficients and wavelet decomposition coefficients.

[0038] Based on a preset threshold function, the wavelet decomposition coefficients are denoised to obtain the processed wavelet coefficients.

[0039] The processed wavelet coefficients and scaling coefficients are reconstructed to obtain the denoised signal of the current processed signal;

[0040] The denoised signal of all sensor signals in the multi-source sensor monitoring signal data is used as the multi-source signal denoised data. The multi-source signal denoised data is then normalized to obtain the multi-source data.

[0041] To address the same technical problem, this invention also provides an intelligent monitoring system for GIS circuit breakers, comprising: a multi-source data acquisition module, a data preprocessing module, a clustering feature extraction module, a feature noise optimization module, and an intelligent diagnosis module.

[0042] Among them, the multi-source data acquisition module is used to acquire the sensor signals collected by the sensors of the currently monitored GIS circuit breaker, convert the sensor signals into wave frequency vectors and map them on the spatial plane, delete wave frequency vectors whose spatial distance on the same horizontal coordinate exceeds a preset distance value, complete the first noise reduction process, and obtain multi-source sensor monitoring signal data.

[0043] The data preprocessing module is used to perform a second noise reduction process on the multi-source sensor monitoring signal data using wavelet threshold denoising to obtain noise-reduced monitoring signal data, and to perform dimensional splitting on the noise-reduced monitoring signal data according to different data attribute types to obtain multi-source data.

[0044] The clustering feature extraction module is used to cluster the multi-source data using the K-means clustering algorithm to obtain the sensor data clustering results. Based on the order of recognition time, the sensor data clustering results are subjected to a first feature vector extraction from near to far to obtain a first feature vector matrix. The sensor data clustering results are subjected to a second feature vector extraction from far to near to obtain a second feature vector matrix.

[0045] The feature noise optimization module is used to flip the first feature vector matrix along its central axis to obtain a third feature vector matrix, calculate the similarity value of each row and column between the third feature vector matrix and the second feature vector matrix, and delete the row and column when the similarity value is less than a preset similarity threshold to obtain an optimized feature vector matrix.

[0046] The intelligent diagnostic module is used to input the optimized feature vector matrix into a pre-built CNN convolutional neural network model and output the monitoring and diagnostic results of the currently monitored GIS circuit breaker.

[0047] To address the same technical problem, embodiments of the present invention also provide a computer device, including a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, it implements an intelligent monitoring method for GIS circuit breakers.

[0048] To address the same technical problem, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements an intelligent monitoring method for GIS circuit breakers.

[0049] By implementing this embodiment of the invention, the sensor signals collected by the sensors of the GIS circuit breaker are converted into wave frequency vectors and abnormal wave frequency vectors are deleted to complete the first noise reduction process. Then, a second noise reduction is performed using wavelet threshold denoising, and the data is dimensionally split to obtain multi-source data. By constructing a feature vector matrix, feature similarity comparison is performed to further optimize the noise. Finally, the monitoring and diagnosis of the GIS circuit breaker is completed through a network model. This invention solves the technical problems in the prior art, which rely too much on the collected sensor data itself, are greatly affected by the accuracy of the sensors, and cause misdiagnosis of the GIS circuit breaker when the sensor data is abnormal, thus interfering with the monitoring and diagnosis results and resulting in low accuracy. This invention achieves effective monitoring and rapid fault diagnosis of GIS circuit breakers, improving the accuracy of monitoring and diagnosis.

[0050] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0051] Figure 1 : A flowchart illustrating an embodiment of the intelligent monitoring method for GIS circuit breakers provided by the present invention;

[0052] Figure 2 : A schematic diagram of an embodiment of the intelligent monitoring system for GIS circuit breakers provided by the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Example 1

[0055] Please refer to Figure 1 This is a flowchart illustrating an intelligent monitoring method for a GIS circuit breaker according to an embodiment of the present invention. The intelligent monitoring method includes steps 101 to 104, each step being as follows:

[0056] Step 101: Acquire the sensor signals collected by the sensors of the currently monitored GIS circuit breaker, convert the sensor signals into frequency vectors mapped onto a spatial plane, delete frequency vectors whose spatial distance on the same horizontal coordinate exceeds a preset distance value, complete the first noise reduction process, and obtain multi-source sensor monitoring signal data. To further standardize the monitoring environment, before acquiring the sensor signals collected by the sensors of the currently monitored GIS circuit breaker, the process further includes: acquiring the current sensor signals collected by the sensors of the currently monitored GIS circuit breaker in real time, and determining whether the current sensor signals meet the status monitoring conditions. If they do, based on a preset time, capture the sensor signals collected by the sensors of the currently monitored GIS circuit breaker to obtain multi-source sensor monitoring signal data; wherein, the status monitoring condition is that the current sensor signal of the sensor is greater than a preset sensor signal threshold.

[0057] In this embodiment, based on the research on the structural characteristics and fault causes of GIS circuit breakers in the technical project, the main monitored state quantities are determined to include location data, current data, pressure data, temperature data, humidity data, noise data, and gas state data. Multi-source sensor monitoring signal data is acquired using existing sensors and other tools. The sensors monitoring the GIS circuit breaker include position sensors, current sensors, pressure sensors, temperature sensors, humidity sensors, noise sensors, and gas state sensors. When the current sensor signal does not meet the state monitoring conditions, it indicates that the current state of the GIS circuit breaker is normal, and no further diagnostic analysis is required. Only when the state monitoring conditions are met does it indicate that the current state of the GIS circuit breaker is abnormal, and only then is it necessary to proceed to the next step of capturing sensor data within a preset time period. By fixing the sensor data capture time and unifying the signal source time period, subsequent diagnostic analysis is ensured only when an abnormal state occurs, improving monitoring efficiency and achieving real-time monitoring of the GIS circuit breaker.

[0058] Step 102: Perform a second denoising process on the multi-source sensor monitoring signal data using wavelet threshold denoising to obtain denoised monitoring signal data, and then split the denoised monitoring signal data into multiple sources according to different data attribute types.

[0059] Optionally, step 102 specifically includes:

[0060] Extract each sensor signal from the multi-source sensor monitoring signal data as the current processing signal;

[0061] Using the sym4 wavelet, the current signal is decomposed into four levels of wavelets to obtain the scaled wavelet decomposition coefficients; where the scaled wavelet decomposition coefficients include scale coefficients and wavelet decomposition coefficients.

[0062] Based on a preset threshold function, the wavelet decomposition coefficients are denoised to obtain the processed wavelet coefficients; wherein the preset threshold function is specifically:

[0063]

[0064] In the formula, w j,k λ represents the processed wavelet coefficients; n is the preset denoising coefficient, which can take the value 2; j is the number of decomposition levels; and λ is the set threshold, the formula for which the threshold is set is as follows:

[0065]

[0066] N is the number of sampling points, and σ is the noise standard deviation.

[0067] The processed wavelet coefficients and scaling coefficients are reconstructed to obtain the denoised signal of the current processed signal;

[0068] The denoised signals of all sensor signals from the multi-source sensor monitoring data are used as the multi-source signal denoised data. The multi-source signal denoised data is then normalized to obtain multi-source data. The multi-source data includes position sensor data, current sensor data, pressure sensor data, temperature sensor data, humidity sensor data, noise sensor data, and gas state sensor data.

[0069] In this embodiment, the acquired multi-source sensor monitoring signal data may still contain a large amount of noise. To improve the accuracy of diagnosis, signal denoising is essential. Wavelet threshold denoising is employed, using wavelet decomposition to separate the high and low frequencies of the original signal and reconstruct them separately. This method is effective against abrupt noise and avoids abnormal noise affecting the diagnostic results.

[0070] Step 103: Cluster the multi-source data using the K-means clustering algorithm to obtain the sensor data clustering results. Based on the order of recognition time, extract the first feature vector from the sensor data clustering results from near to far to obtain the first feature vector matrix. Then, extract the second feature vector from the sensor data clustering results from far to near to obtain the second feature vector matrix.

[0071] In this embodiment, the K-means clustering algorithm and feature vector extraction are used to obtain a feature vector matrix. This feature vector matrix effectively represents multi-source data, facilitating diagnosis based on effective features, improving the diagnostic effect of the neural network, and accelerating the diagnostic efficiency.

[0072] Optionally, the multi-source data includes position sensing data, current sensing data, pressure sensing data, temperature sensing data, humidity sensing data, noise sensing data, and gas state sensing data.

[0073] The process of clustering multi-source data using the K-means clustering algorithm to obtain the clustering results of sensor data is as follows:

[0074] A 7-dimensional vector composed of position sensing data, current sensing data, pressure sensing data, temperature sensing data, humidity sensing data, noise sensing data, and gas state sensing data is used as a category object.

[0075] Initialize cluster centers with a preset number of categories; each cluster center is a 7-dimensional vector composed of position sensing data, current sensing data, pressure sensing data, temperature sensing data, humidity sensing data, noise sensing data, and gas state sensing data; the preset number of categories is determined by empirical method or elbow method.

[0076] Calculate new cluster centers based on the category objects and the current cluster centers;

[0077] The clustering process iterates repeatedly to calculate new cluster centers based on the category objects and the current cluster centers. The clustering process continues to iterate until the clustering convergence condition is met. The clustering convergence condition is met when the distance between the new cluster center obtained in the iteration and the previous cluster center is less than a preset distance threshold.

[0078] Optionally, new cluster centers can be calculated based on the category objects and the current cluster centers, specifically:

[0079] Based on the distance between the category object and the current cluster centers, each category object is assigned to the category of the nearest cluster center; the mean of all objects in each category is used as the new cluster center.

[0080] In this embodiment, the K-means clustering algorithm is an iterative clustering analysis algorithm. The data is pre-divided into K groups, and K objects are randomly selected as initial cluster centers. The distance between each object and each seed cluster center is calculated, and each object is assigned to the nearest cluster center. The cluster centers and the objects assigned to them represent a cluster. Each time a sample is assigned, the cluster centers are recalculated based on the existing objects in the cluster. This process is repeated until a termination condition is met. The K-means clustering algorithm can prune the tree based on the categories of a small number of known cluster samples to determine the classification of some samples. To overcome the inaccuracy of clustering with a small number of samples, the algorithm itself has an optimization iteration function. It iteratively corrects and prunes the already obtained clusters to determine the clustering of some samples, optimizing the unreasonable initial supervised learning sample classification. Since it only targets a small number of samples, it can reduce the overall clustering time complexity, achieving the goal of rapid diagnosis of GIS circuit breakers.

[0081] Optionally, the steps of performing a first feature vector extraction on the clustering results of the sensor data from near to far according to the order of recognition time to obtain a first feature vector matrix, and performing a second feature vector extraction on the clustering results of the sensor data from far to near to obtain a second feature vector matrix, specifically include:

[0082] Based on the clustering results of the sensor data, extract the cluster center for each category;

[0083] Calculate the Euclidean distance between each category object data and its corresponding cluster center in each category, extract the category object data whose Euclidean distance values ​​fall within a preset range, and arrange them in chronological order of recognition time;

[0084] The sensor data corresponding to each category extracted from near to far are used to form a first feature vector matrix, and the sensor data corresponding to each category extracted from far to near are used to form a second feature vector matrix. In this embodiment, using the symptoms corresponding to the fault to diagnose the fault is the most common and mature method. During the diagnosis process, various signals acquired during equipment operation are first analyzed, and various feature information in the signals is extracted to obtain the symptoms related to the fault. These symptoms are then used for fault diagnosis. Since there is no simple one-to-one correspondence between faults and various symptoms, using the feature vector matrix as symptoms for fault diagnosis can effectively improve the effectiveness and accuracy of the diagnosis.

[0085] Specifically, to reduce the noise generated after separating the high and low frequencies of the original signal, we need to filter out the features in the feature vector that are significantly affected by noise. We generate the first feature vector matrix in the order of feature recognition, and then generate the second feature vector matrix in reverse order. In subsequent steps, we only need to compare the similarity between the two matrices to accurately identify the noise that is significantly affected by noise.

[0086] Step 104: The feature noise optimization module is used to flip the first feature vector matrix along its central axis to obtain a third feature vector matrix, calculate the similarity value of each row and column between the third feature vector matrix and the second feature vector matrix, and delete the row and column when the similarity value is less than a preset similarity threshold to obtain an optimized feature vector matrix.

[0087] Specifically, since the first eigenvector matrix and the second eigenvector matrix are relative matrices that are flipped according to the chronological order, it is only necessary to flip one of the matrices along its central axis. Then, the similarity values ​​between the flipped matrix and the other matrix in each row and column are compared. When it is determined that the similarity value does not meet the requirements, it is regarded as noise and deleted, which ultimately optimizes the matrix.

[0088] Step 105: Input the optimized feature vector matrix into the pre-constructed CNN convolutional neural network model, and output the monitoring and diagnostic results of the currently monitored GIS circuit breaker. The monitoring and diagnostic results include normal operation, gas leakage, excessive moisture in the gas, switch failure, internal discharge of the GIS, and hydraulic mechanism failure.

[0089] In this embodiment, the main faults of the GIS circuit breaker are: 1. SF6 gas leakage. This type of fault usually occurs at the sealing surfaces, welding points, and pipe joints of the combined electrical appliances. The main cause is aging of the gaskets or pinholes in the welds. Therefore, a large amount of SF6 gas needs to be added to the GIS every year to ensure normal operating pressure. This type of fault is relatively common. 2. Excessive moisture in SF6 gas. This type of fault usually occurs when the SF6 gas in the circuit breaker chamber has excessive moisture content during operation. Faults caused by excessive moisture content in the SF6 gas are prone to flashover of insulators or other insulating components. The main reason for excessive moisture is that water seeps into the SF6 gas through leaks in the seals. 3. Switch faults, mainly involving gas breakdown of components such as circuit breakers, load switches, disconnect switches, or grounding switches. Also, the moving and stationary contacts may shift during closing, causing poor contact. 4. Internal discharge of the GIS. Due to manufacturing processes, some components inside the GIS are at a floating potential, leading to a local increase in electric field strength, which in turn generates corona discharge. The presence of metallic impurities in the GIS and air bubbles in the insulators can also lead to corona discharge or partial discharge. 5. Hydraulic mechanism malfunctions such as oil leakage or frequent pressurization are mostly caused by aging of the hydraulic mechanism seals, misalignment of the installation position, or nitrogen leakage from the pressure tank. These types of malfunctions are relatively common in GIS systems.

[0090] Optionally, a pre-built CNN convolutional neural network model can be obtained by training a neural network based on a sensor feature vector database, specifically as follows:

[0091] Collect normal and fault sensing signals from several GIS circuit breakers, and take each collected signal from the GIS circuit breaker as a multi-source sensing data set, take the multi-source sensing data set of several normal sensing signals as a normal signal set, and take the multi-source sensing data set of several fault sensing signals as a fault signal set.

[0092] Wavelet threshold denoising is used to denoise and normalize the multi-source sensor data groups of each normal signal set and each fault signal set to obtain normal denoised datasets and fault denoised datasets.

[0093] The normal noise reduction dataset and the fault noise reduction dataset are superimposed, and the normal label and the diagnostic fault type label corresponding to each multi-source sensor data group are labeled to form a multi-source sensor signal database.

[0094] Each multi-source sensor data group in the multi-source sensor signal database is clustered using the K-means clustering algorithm to obtain the sensor data clustering result for each multi-source sensor data group. The feature vectors of the sensor data clustering results for each multi-source sensor data group are then extracted to obtain the feature vector matrix for each multi-source sensor data group. Based on the feature vector matrix of each multi-source sensor data group in the multi-source sensor signal database, a sensor feature vector database is obtained.

[0095] According to a preset ratio, the multi-source sensor signal database is divided into a test set and a training set. Based on the training set and the test set, the CNN convolutional neural network model is trained, and the error rate of the current training model is calculated.

[0096] Determine if the current model error rate is less than the preset error rate threshold. If it is not less than the preset error rate threshold, train the CNN convolutional neural network model based on the training set and the test set, and calculate the current training model error rate. Stop training when the current training model error rate is less than the preset error rate threshold, and obtain the pre-built CNN convolutional neural network model.

[0097] In this embodiment, if the current training model error rate is not less than the preset error rate threshold, it indicates that the accuracy of the CNN convolutional neural network model is not sufficient and training needs to continue. The model parameters of the CNN convolutional neural network model can be adjusted accordingly, training sample data in the training set can be reselected, and training can continue until the current training model error rate is less than the preset error rate threshold. Then, training is stopped, and the model parameters at the time of stopping training are the optimal model parameters of the CNN convolutional neural network model.

[0098] Optionally, the CNN convolutional neural network model is trained based on the training and test sets, and the error rate of the current trained model is calculated, specifically as follows:

[0099] The feature vector matrix of each multi-source sensor data group is used as the input sample, and the label corresponding to each multi-source sensor data group is used as the output sample.

[0100] Select a corresponding number of training sample data from the training set according to the preset batch size, and simultaneously input the input and output samples of the training sample data into the CNN convolutional neural network model for training; wherein, the preset batch size is the number of training samples required for one training session.

[0101] Input the input samples from the test set into the currently trained CNN convolutional neural network model to obtain the current model output. Compare the current model output with the output samples corresponding to the current input samples to obtain the error rate of the current trained model.

[0102] It should be noted that Convolutional Neural Networks (CNNs) are a type of feedforward neural network that includes convolutional computations and has a deep structure; they are one of the representative algorithms of deep learning. The structure of a CNN includes an input layer, hidden layers, and an output layer. Convolutional neural networks possess representation learning capabilities, enabling them to perform translation-invariant classification of input information according to their hierarchical structure.

[0103] By implementing this embodiment of the invention, the sensor signals collected by the sensors of the GIS circuit breaker are converted into wave frequency vectors and abnormal wave frequency vectors are deleted to complete the first noise reduction process. Then, a second noise reduction is performed using wavelet threshold denoising, and the data is dimensionally split to obtain multi-source data. By constructing a feature vector matrix, feature similarity comparison is performed to further optimize the noise. Finally, the monitoring and diagnosis of the GIS circuit breaker is completed through a network model. This invention solves the technical problems in the prior art, which rely too much on the collected sensor data itself, are greatly affected by the accuracy of the sensors, and cause misdiagnosis of the GIS circuit breaker when the sensor data is abnormal, thus interfering with the monitoring and diagnosis results and resulting in low accuracy. This invention achieves effective monitoring and rapid fault diagnosis of GIS circuit breakers, improving the accuracy of monitoring and diagnosis.

[0104] Example 2

[0105] Accordingly, see Figure 2 , Figure 2 This is a schematic diagram of a second embodiment of the intelligent monitoring system for GIS circuit breakers provided by the present invention. Figure 2 As shown, the intelligent monitoring system for GIS circuit breakers includes a multi-source data acquisition module 201, a data preprocessing module 202, a clustering feature extraction module 203, a feature noise optimization module 204, and an intelligent diagnosis module 205.

[0106] Among them, the multi-source data acquisition module 201 is used to acquire the sensor signals collected by the sensors of the currently monitored GIS circuit breaker, convert the sensor signals into wave frequency vectors and map them on the spatial plane, delete wave frequency vectors whose spatial distance on the same horizontal coordinate exceeds a preset distance value, complete the first noise reduction process, and obtain multi-source sensor monitoring signal data.

[0107] The data preprocessing module 202 is used to perform a second noise reduction process on the multi-source sensor monitoring signal data using wavelet threshold denoising to obtain noise-reduced monitoring signal data, and to perform dimensional splitting on the noise-reduced monitoring signal data according to different data attribute types to obtain multi-source data.

[0108] The clustering feature extraction module 203 is used to perform clustering and identification on the multi-source data using the K-means clustering algorithm to obtain the sensor data clustering results. According to the order of identification time, the sensor data clustering results are subjected to a first feature vector extraction from near to far to obtain a first feature vector matrix. The sensor data clustering results are subjected to a second feature vector extraction from far to near to obtain a second feature vector matrix.

[0109] The feature noise optimization module 204 is used to flip the first feature vector matrix along its central axis to obtain a third feature vector matrix, calculate the similarity value of each row and column between the third feature vector matrix and the second feature vector matrix, and delete the row and column when the similarity value is less than a preset similarity threshold to obtain an optimized feature vector matrix.

[0110] The intelligent diagnostic module 205 is used to input the optimized feature vector matrix into a pre-built CNN convolutional neural network model and output the monitoring and diagnostic results of the currently monitored GIS circuit breaker.

[0111] By implementing this embodiment of the invention, the sensor signals collected by the sensors of the GIS circuit breaker are converted into wave frequency vectors and abnormal wave frequency vectors are deleted to complete the first noise reduction process. Then, a second noise reduction is performed using wavelet threshold denoising, and the data is dimensionally split to obtain multi-source data. By constructing a feature vector matrix, feature similarity comparison is performed to further optimize the noise. Finally, the monitoring and diagnosis of the GIS circuit breaker is completed through a network model. This invention solves the technical problems in the prior art, which rely too much on the collected sensor data itself, are greatly affected by the accuracy of the sensors, and cause misdiagnosis of the GIS circuit breaker when the sensor data is abnormal, thus interfering with the monitoring and diagnosis results and resulting in low accuracy. This invention achieves effective monitoring and rapid fault diagnosis of GIS circuit breakers, improving the accuracy of monitoring and diagnosis.

[0112] In addition, this application also provides a computer device, which includes a processor and a memory. The memory is used to store a computer program, and when the computer program is executed by the processor, it implements the steps in any of the above method embodiments.

[0113] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above method embodiments.

[0114] The intelligent monitoring system for GIS circuit breakers described above can implement the intelligent monitoring method for GIS circuit breakers described in the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0115] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention in detail. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for intelligent monitoring of GIS circuit breakers, characterized in that, include: The sensor signals collected by the sensors of the currently monitored GIS circuit breaker are acquired, and the sensor signals are converted into wave frequency vectors and mapped onto a spatial plane. Wave frequency vectors with a spatial distance exceeding a preset distance value on the same horizontal coordinate are deleted to complete the first noise reduction process and obtain multi-source sensor monitoring signal data. The multi-source sensor monitoring signal data is subjected to a second denoising process using wavelet threshold denoising to obtain denoised monitoring signal data. The denoised monitoring signal data is then dimensionally split according to different data attribute types to obtain multi-source data. The multi-source data is clustered and identified using the K-means clustering algorithm to obtain the sensor data clustering results. Based on the order of identification time, the sensor data clustering results are subjected to a first feature vector extraction from near to far to obtain a first feature vector matrix. The sensor data clustering results are then subjected to a second feature vector extraction from far to near to obtain a second feature vector matrix. After flipping the first feature vector matrix along its central axis, a third feature vector matrix is ​​obtained. The similarity value of each row and column between the third feature vector matrix and the second feature vector matrix is ​​calculated. When the similarity value is less than a preset similarity threshold, the row and column are deleted to obtain an optimized feature vector matrix. The optimized feature vector matrix is ​​input into a pre-built CNN convolutional neural network model, which outputs the monitoring and diagnostic results of the currently monitored GIS circuit breaker.

2. The intelligent monitoring method for GIS circuit breakers as described in claim 1, characterized in that, The multi-source data includes position sensing data, current sensing data, pressure sensing data, temperature sensing data, humidity sensing data, noise sensing data, and gas state sensing data. The step of clustering the multi-source data using the K-means clustering algorithm to obtain the sensor data clustering results is as follows: The 7-dimensional vector composed of the position sensing data, the current sensing data, the pressure sensing data, the temperature sensing data, the humidity sensing data, the noise sensing data, and the gas state sensing data is used as a category object; Initialize the number of cluster centers for the preset number of categories; Each cluster center is a 7-dimensional vector composed of the location sensing data, the current sensing data, the pressure sensing data, the temperature sensing data, the humidity sensing data, the noise sensing data, and the gas state sensing data; the preset number of categories is determined by empirical method or elbow method. Calculate new cluster centers based on the category objects and the current cluster centers; The clustering process iterates repeatedly to calculate new cluster centers based on the category objects and the current cluster centers. The clustering process continues to iterate until the clustering convergence condition is met. The clustering convergence condition is met when the distance between the new cluster center obtained in the iteration and the previously obtained cluster center is less than a preset distance threshold.

3. The intelligent monitoring method for GIS circuit breakers as described in claim 2, characterized in that, The step of calculating new cluster centers based on the category objects and the current cluster centers is as follows: Based on the distance between the category object and each current cluster center, each category object is assigned to the category of the nearest cluster center; The mean of all objects in each class is used as the new cluster center.

4. The intelligent monitoring method for GIS circuit breakers as described in claim 2, characterized in that, The steps of performing a first feature vector extraction on the clustering results of the sensor data from near to far according to the order of recognition time to obtain a first feature vector matrix, and performing a second feature vector extraction on the clustering results of the sensor data from far to near to obtain a second feature vector matrix, are specifically as follows: Based on the clustering results of the sensor data, extract the cluster center for each category; Calculate the Euclidean distance between each category object data and its corresponding cluster center in each category, extract the category object data whose Euclidean distance values ​​fall within a preset range, and arrange them in chronological order of recognition time; The sensor data corresponding to each category extracted from near to far are used to form a first feature vector matrix, and the sensor data corresponding to each category extracted from far to near are used to form a second feature vector matrix.

5. The intelligent monitoring method for GIS circuit breakers as described in claim 1, characterized in that, The monitoring and diagnostic results include normal operation, gas leakage, excessive moisture in gas, switch failure, internal discharge of GIS, and hydraulic mechanism failure; the pre-built CNN convolutional neural network model is obtained by training the neural network based on the sensor feature vector database, specifically: Collect normal and fault sensing signals from several GIS circuit breakers, and take each collected signal from the GIS circuit breaker as a multi-source sensing data group. Take the multi-source sensing data group of several normal sensing signals as a normal signal set, and take the multi-source sensing data group of several fault sensing signals as a fault signal set. The wavelet threshold denoising method is used to denoise and normalize the multi-source sensor data groups of each normal signal set and each fault signal set to obtain normal denoised datasets and fault denoised datasets. The normal noise reduction dataset and the fault noise reduction dataset are superimposed, and each multi-source sensor data group is labeled with a normal label and a diagnostic fault type label to form a multi-source sensor signal database. Each multi-source sensor data group in the multi-source sensor signal database is clustered and identified using the K-means clustering algorithm to obtain the sensor data clustering result for each multi-source sensor data group. Feature vectors are extracted from the sensor data clustering result for each multi-source sensor data group to obtain the feature vector matrix for each multi-source sensor data group. Based on the feature vector matrix of each multi-source sensor data group in the multi-source sensor signal database, the sensor feature vector database is obtained. According to a preset ratio, the multi-source sensor signal database is divided into a test set and a training set. Based on the training set and the test set, the CNN convolutional neural network model is trained, and the error rate of the current training model is calculated. Determine whether the error rate of the current training model is less than a preset error rate threshold. If it is not less than the preset error rate threshold, then train the CNN convolutional neural network model based on the training set and the test set, and calculate the error rate of the current training model until the error rate of the current training model is less than the preset error rate threshold, and stop training to obtain the pre-built CNN convolutional neural network model.

6. The intelligent monitoring method for GIS circuit breakers as described in claim 5, characterized in that, The process of training the CNN convolutional neural network model based on the training set and the test set, and calculating the error rate of the current training model, specifically involves: The feature vector matrix of each of the multi-source sensor data groups is used as the input sample, and the label corresponding to each of the multi-source sensor data groups is used as the output sample. Select a corresponding number of training sample data from the training set according to a preset batch size, and simultaneously input the input and output samples of the training sample data into the CNN convolutional neural network model for training; wherein, the preset batch size is the number of training samples required for one training session. The input samples from the test set are input into the currently trained CNN convolutional neural network model to obtain the current model output. The current model output is compared with the output samples corresponding to the current input samples to obtain the error rate of the currently trained model.

7. The intelligent monitoring method for GIS circuit breakers as described in claim 1, characterized in that, The second denoising process, which involves using wavelet threshold denoising to denoise the multi-source sensor monitoring signal data, yields denoised monitoring signal data. Furthermore, the denoised monitoring signal data is dimensionally split according to different data attribute types to obtain multi-source data. Specifically: Extract each sensor signal from the multi-source sensor monitoring signal data as the current processing signal; Using the sym4 wavelet, the current processed signal is decomposed into wavelet 4-level decomposition to obtain scale wavelet decomposition coefficients; wherein, the scale wavelet decomposition coefficients include scale coefficients and wavelet decomposition coefficients. Based on a preset threshold function, the wavelet decomposition coefficients are denoised to obtain the processed wavelet coefficients. The processed wavelet coefficients and the scaling coefficients are reconstructed to obtain the denoised signal of the current processed signal; The denoised signal of all sensor signals in the multi-source sensor monitoring signal data is taken as the multi-source signal denoised data. The multi-source signal denoised data is then normalized to obtain the multi-source data.

8. An intelligent monitoring system for GIS circuit breakers, characterized in that, include: The system includes a multi-source data acquisition module, a data preprocessing module, a clustering feature extraction module, a feature noise optimization module, and an intelligent diagnosis module. The multi-source data acquisition module is used to acquire the sensor signals collected by the sensors of the currently monitored GIS circuit breaker, convert the sensor signals into wave frequency vectors and map them onto the spatial plane, delete wave frequency vectors whose spatial distance on the same horizontal coordinate exceeds a preset distance value, complete the first noise reduction process, and obtain multi-source sensor monitoring signal data. The data preprocessing module is used to perform a second noise reduction process on the multi-source sensor monitoring signal data using wavelet threshold denoising to obtain noise-reduced monitoring signal data, and to perform dimensional splitting on the noise-reduced monitoring signal data according to different data attribute types to obtain multi-source data. The clustering feature extraction module is used to cluster the multi-source data using the K-means clustering algorithm to obtain the sensor data clustering results. Based on the order of recognition time, the sensor data clustering results are subjected to a first feature vector extraction from near to far to obtain a first feature vector matrix. The sensor data clustering results are subjected to a second feature vector extraction from far to near to obtain a second feature vector matrix. The feature noise optimization module is used to flip the first feature vector matrix along its central axis to obtain a third feature vector matrix, calculate the similarity value of each row and column between the third feature vector matrix and the second feature vector matrix, and delete the row and column when the similarity value is less than a preset similarity threshold to obtain an optimized feature vector matrix. The intelligent diagnostic module is used to input the optimized feature vector matrix into a pre-built CNN convolutional neural network model and output the monitoring and diagnostic results of the currently monitored GIS circuit breaker.

9. A computer device, characterized in that, It includes a processor and a memory, the memory being used to store a computer program, which, when executed by the processor, implements the intelligent monitoring method for GIS circuit breakers as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the intelligent monitoring method for GIS circuit breakers as described in any one of claims 1 to 7.

Images