A monitoring method and system for abnormal operation status of GIS circuit breaker

Abstract

The present invention provides a monitoring method and system for the abnormal operating state of a GIS circuit breaker, which relates to the technical field of circuit breaker identification, and includes: acquiring image time series data under the operating state of the GIS circuit breaker in real time and identifying the opening and closing state of the GIS circuit breaker; when it is identified that the GIS circuit breaker is in the closed state, acquiring sound signal time series data under the operating state of the GIS circuit breaker in real time; constructing a GIS circuit breaker sound feature recognition model; analyzing and identifying the extracted sound signal features; obtaining the operating state of the GIS circuit breaker; and realizing real-time monitoring of the operating state of the GIS circuit breaker and efficient, accurate and rapid identification of the abnormal state. The abnormal state and sudden opening state under the operating state of the GIS circuit breaker are discovered in time, and rapid alarm can be realized to reduce risk loss, thereby effectively reducing the risk of equipment failure and improving the reliability and safety of the power system.

Description

Technical Field

[0001] The present invention relates to the field of circuit breakers, and in particular to a monitoring method and system for abnormal operation status of a GIS circuit breaker. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Circuit breakers are key components of gas insulated switchgear (GIS) and can play a role in protection and control. Mechanical failure of circuit breakers in GIS equipment is an important reason that threatens its safe and stable operation, mainly including mechanical defects such as loose shaft pins, core jamming, and spring mechanism jamming. If GIS circuit breakers are not handled in time after a failure, it will seriously endanger the safety of power grid operation.

[0004] In the prior art, mechanical vibration, coil current, contact travel and sound signals are usually used for status detection. Mechanical vibration signals can be collected by contact sensors on the circuit breaker, and then the signals are analyzed and processed to realize abnormal diagnosis of the circuit breaker. However, this method requires the sensor to be physically connected to the circuit breaker shaft, which is difficult, costly, and prone to potential interference with the operation of the circuit breaker. The coil current and contact line signals are generated during the circuit breaker disconnection process, and status monitoring can only be achieved in the switching state. The acquisition of sound signals is not constrained by the sensor installation position, and non-contact recognition can be achieved, but it is easily interfered by on-site noise during detection, and the effectiveness of feature extraction needs to be improved. At the same time, the current diagnosis of abnormal operating status of GIS circuit breakers mostly uses traditional neural network algorithms for identification, and the accuracy of identification needs to be improved. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a monitoring method and system for the abnormal operating status of a GIS circuit breaker, which can realize real-time monitoring of the operating status of the GIS circuit breaker and efficient, accurate and rapid identification of abnormal status, timely discover abnormal status and sudden tripping conditions under the operating status of the GIS circuit breaker, and can realize rapid alarm and reduce risk losses, thereby effectively reducing the risk of equipment failure and improving the reliability and safety of the power system.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, a method for monitoring an abnormal operating state of a GIS circuit breaker is provided, comprising:

[0008] Real-time acquisition of image time series data of GIS circuit breaker in operation state and preprocessing;

[0009] Inputting the preprocessed image time series data into the GIS circuit breaker operation status image recognition model to obtain the opening and closing status of the GIS circuit breaker;

[0010] When it is identified that the GIS circuit breaker is in a closed state, the sound signal timing data of the GIS circuit breaker in the operating state is obtained in real time;

[0011] Preprocessing the sound signal time series data to extract sound signal features of the sound signal time series data;

[0012] Inputting the sound signal characteristics into a GIS circuit breaker sound characteristic recognition model to obtain the operating status of the GIS circuit breaker;

[0013] Determine whether the operating status of the GIS circuit breaker is abnormal. If it is abnormal, record the abnormal status information in time and send an alarm message. Use the real-time sound signal data of the GIS circuit breaker in the previous operating state and the sound signal data during the opening and closing process to correct the GIS circuit breaker sound feature recognition model.

[0014] In a second aspect, a monitoring system for abnormal operation status of a GIS circuit breaker is provided, comprising:

[0015] The image data acquisition module is configured to acquire image time series data of the GIS circuit breaker in real time under the operating state;

[0016] The image detection module is configured to preprocess the image time series data; input the preprocessed image time series data into the GIS circuit breaker operation status image recognition model to obtain the opening and closing status of the GIS circuit breaker;

[0017] The sound recognition module is configured to obtain the sound signal timing data of the GIS circuit breaker in the operating state in real time when it is recognized that the GIS circuit breaker is in the closed state;

[0018] Preprocessing the sound signal time series data to extract the sound signal characteristics under the operating state of the GIS circuit breaker;

[0019] Inputting the sound signal characteristics into a GIS circuit breaker sound characteristic recognition model to obtain the operating status of the GIS circuit breaker;

[0020] The result generating module is configured to determine whether the operating state of the GIS circuit breaker is abnormal. If it is abnormal, the abnormal state information is recorded in time and an alarm message is sent. The sound feature recognition model of the GIS circuit breaker is corrected by using the sound signal data of the GIS circuit breaker in the previous operating state and the sound signal data during the opening and closing process acquired in real time.

[0021] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0022] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0023] One or more of the above technical solutions have the following beneficial effects:

[0024] The present invention provides a monitoring method for the abnormal operation state of a GIS circuit breaker, which adopts non-contact detection methods such as image acquisition equipment and voiceprint sensors, and combines image features and sound signal features for joint identification to realize the detection of the abnormal operation state of the GIS circuit breaker.

[0025] The present invention determines the opening and closing states of the GIS circuit breaker based on the surface features of the GIS, so as to identify the abnormal operating conditions of the circuit breaker in the closed state, and obtain the abnormal condition information during the sudden opening when no abnormality is identified during the normal operation, and can give an alarm in time, thus reducing the loss caused by continuing to collect invalid sound signals without detecting the sudden opening, thereby effectively reducing the risk of equipment failure, improving the reliability and safety of the power system, and providing more reliable and important characteristic information for subsequent monitoring.

[0026] The present invention uses image recognition for situations where image features are obvious, such as the overall state of opening and closing, and uses sound signal feature recognition for situations where sound signal features are obvious in the closing state. Different methods are used for specific problems, thereby improving the efficiency and accuracy of overall monitoring.

[0027] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0029] Figure 1 This is a flow chart of a method for monitoring abnormal operating status of a GIS circuit breaker according to Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0030] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0031] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0032] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0033] Embodiment 1

[0034] This embodiment discloses a method for monitoring the abnormal operation state of a GIS circuit breaker, comprising: acquiring image time series data of the operation state of the GIS circuit breaker in real time through an image acquisition device;

[0035] Preprocessing the image time series data of the GIS circuit breaker under the operating state;

[0036] Inputting the preprocessed image time series data into the GIS circuit breaker operation status image recognition model to obtain the opening and closing status of the GIS circuit breaker;

[0037] When it is identified that the GIS circuit breaker is in the closed state, the sound signal timing data of the GIS circuit breaker in the operating state is obtained in real time through the voiceprint sensor;

[0038] Preprocessing the sound signal time series data of the GIS circuit breaker in the operating state;

[0039] Extract the sound signal characteristics of GIS circuit breaker under operating status;

[0040] Construct a GIS circuit breaker sound feature recognition model, analyze and identify the extracted sound signal features under the GIS circuit breaker operating state, and obtain the GIS circuit breaker operating state;

[0041] Determine whether the operating state of the GIS circuit breaker is abnormal; if it is abnormal, record the abnormal state information in time and send alarm information;

[0042] Specifically, when the GIS circuit breaker is identified as being in the open state, the abnormal state information is recorded in time and an alarm message is sent; the sound signal data of the GIS circuit breaker in the previous operating state and the sound signal data in the opening and closing process collected in real time by the voiceprint sensor are saved and analyzed, and then saved to the training set to further correct the GIS circuit breaker sound feature recognition model.

[0043] In some embodiments, image data of the GIS circuit breaker in operation is collected in real time by an image acquisition device, and the image data is preprocessed by deduplication, brightening, contrast enhancement, image noise reduction, etc. to ensure data quality, enhance relevant image information features, and retain edge features.

[0044] Remove the image data of GIS circuit breaker operation status acquired repeatedly;

[0045] Improve the image brightness and contrast when the GIS circuit breaker is in operation;

[0046] The image is smoothed by Gaussian bilateral filtering to reduce image noise and avoid blurring the edges as much as possible to preserve the edge information in the image.

[0047] Gaussian bilateral filtering is a nonlinear filtering method that combines image spatial proximity and pixel value similarity, that is, it considers both spatial information and grayscale similarity at the same time, and can remove image noise while retaining edge information.

[0048] In some embodiments, the preprocessed image is input into a GIS circuit breaker operation status image recognition model to obtain surface features and identify the opening and closing status of the GIS circuit breaker, specifically:

[0049] The image data of the opening and closing states of GIS circuit breakers are obtained, randomly extracted using a random function, and divided into training sets and test sets according to a certain ratio.

[0050] The image data in the training set are labeled and preprocessed according to the open and closed states.

[0051] The GRU neural network is used to construct a GIS circuit breaker operating status image recognition model, extract the image data features of the opening and closing status, input the image data features in the training set into the GRU neural network input layer, and take the opening and closing status of the GIS circuit breaker as the model decision output layer. The GRU neural network is mainly composed of four layers: input, hidden, output and decision. The first layer is the input layer of the model, which is mainly responsible for the preprocessing and labeling of the input data; the second layer is the hidden layer, which is mainly responsible for the learning of the input data; the third layer is the model output layer, which is mainly responsible for transmitting the data output by the hidden layer to each classification for the final classification and recognition; the fourth layer is the model decision layer, which is mainly responsible for making result decisions on the classified data output by the output layer according to relevant specifications, so as to obtain the final classification and recognition results, thereby constructing a GIS circuit breaker operating status image recognition model.

[0052] Specifically, there are only two gate structures in the GRU neural network model: update gate G and reset gate C. The smaller the value of update gate G, the less data the neurons in the constructed GRU neural network retain; when reset gate C is 0, the data transmitted by the neurons in the constructed GRU neural network is unavailable. Therefore, in order to ensure the recognition effect of abnormal operation status of GIS circuit breakers, it is necessary to determine the values ​​of update gate G and reset gate C in the GRU neural network according to the following two formulas.

[0053]

[0054]

[0055] In the formula, ε represents the activation function of the GRU model. As an embodiment, the present disclosure adopts the sigmoid function; ω G ,ω C Respectively represent the update gate in the GRU model structure G With reset gate C The weight of y i -1 Represents the output value of the previous neuron in the GRU model; x i Indicates the first i The input value of each neuron. Based on the above formula, the calculation formula for neuron output in the GRU network recognition model is established.

[0056]

[0057]

[0058] In the formula, y Represents the output result of the pending neuron in the GRU model; y i Indicates the first i The output result of each neuron. When using the GRU network to establish a recognition model, not only can the input and output of the current neuron be analyzed, but also an in-depth analysis can be performed based on the previous neuron state of the current neuron. Therefore, when using the GRU network recognition model to detect the GIS operating status, there is no need to be limited to a certain neuron node in the GRU network. The overall structure of the network can be used for abnormal identification, thereby ensuring the recognition effect. The specific process of the GIS abnormal operating status identification method based on machine learning is as follows: the abnormal feature data of GIS equipment collected and fused by the sensor is input into the input layer of the GRU network recognition model. After the model is learned and trained, the classification recognition result can be output, thereby completing the identification of the abnormal operating status of the GIS.

[0059] That is, the image data in the test set is input into the constructed GIS circuit breaker operation status image recognition model to identify the opening and closing status of the GIS circuit breaker.

[0060] In some embodiments, when the GIS circuit breaker is identified as being in a closed state, the sound signal data of the GIS circuit breaker in the operating state is acquired in real time through a voiceprint sensor; the acquired sound signal data is a time domain waveform containing rich information.

[0061] The sound signal data is preprocessed to improve the sound data quality and the accuracy of feature extraction. Specifically, the preprocessing of the sound signal data mainly includes: noise removal, echo interference reduction, gain adjustment and normalization processing.

[0062] The signal denoising algorithm based on wavelet transform decomposes the signal into multiple wavelet coefficients of different scales, and then removes the noise by threshold processing the wavelet coefficients. This can effectively remove the noise in the signal and retain the main features of the signal.

[0063] An adaptive filtering algorithm is used to reduce the impact of echo on sound data; by establishing a model similar to the echo path and adjusting the filter coefficients by minimizing the root mean square error of the error signal, the echo can be effectively suppressed.

[0064] Gain adjustment and normalization are performed according to the differences in sound intensity in the collection environment to ensure that the volume levels of different samples are consistent, making it easier to compare and analyze data at the same scale.

[0065] In some embodiments, extracting a fault feature vector refers to extracting a certain numerical parameter that can accurately reflect the characteristics of a mechanical fault or the operating state of the equipment at this moment. A mechanical component failure of a circuit breaker can be represented by different feature vectors, and a feature vector can also reflect different mechanical component failures of circuit breakers. Such data are mixed and overlapped with each other, which creates many obstacles for the fault diagnosis of high-voltage circuit breakers. However, different feature vectors have different sensitivities to the mechanical component failures of circuit breakers. Find the fault feature vector that best reflects the change in the fault state as the input data for fault diagnosis. Obtain a feature vector from the sound signal under each operating state of the high-voltage circuit breaker, which can most accurately determine the operating state of the high-voltage circuit breaker at this time. The present disclosure adopts IMF energy entropy and envelope feature entropy as signal feature value extraction methods.

[0066] First, the IMF energy entropy feature extraction is performed, assuming that the circuit breaker operation state sound signal x t The ensemble empirical mode decomposition method is used to obtain nIMF components IMF1, IMF2, ..., IMFn and a residual component res. In practical applications, the residual component res is often ignored, so it is assumed that the energy sum of each IMF component is equal to the signal energy, that is,

[0067]

[0068] In the formula, E is the sound signal energy, E i Indicates i The energy of each IMF component. For the i The ratio of the energy of an IMF component to the total energy, so the IMF energy entropy can be written as,

[0069]

[0070] in, H Represents energy entropy.

[0071] Through the above steps, the IMF energy entropy characteristics are obtained in MATLAB to obtain the IMF energy entropy characteristics.

[0072] Furthermore, the envelope spectrum features are extracted. The signal envelope refers to the line between the maximum and minimum points of the signal. Research in the field of signal analysis shows that the envelope contains a lot of information that can reflect the operating status of high-voltage circuit breakers. The envelope of the sound signal is the product of the Hilbert transform. Based on the EEMD decomposition (Empirical Mode Decomposition), the envelope spectrum features are extracted using the Hilbert method.

[0073] Assumptions c ( t ) is a discrete real signal, then its Hilbert transform is defined as:

[0074]

[0075] Construct the parsing signal:

[0076]

[0077] Its amplitude function, that is, the signal envelope is:

[0078]

[0079] When the circuit breaker is working in different states, the time periods of the sound signal energy strength are different. The envelope characteristic entropy is used to represent the operating state of the equipment. Envelope characteristic entropy extraction is based on signal noise reduction and the following steps are performed:

[0080] 1) Obtain the marginal spectrum of each IMF for verification. If it meets the threshold, obtain the envelope; if it does not meet the threshold, return to the signal denoising step or re-collect the sound signal;

[0081] 2) The envelope of each IMF component is t Divided into l The energy contained in each envelope is:

[0082]

[0083] In the formula, i =1,2,......, l , k= 1,2,......, r , t 1 is the i The start time of the segment, t 2 is the i The termination time of the segment, Q k ( i ) is the k The first order IMF component i Segment energy.

[0084] 3) Energy normalization:

[0085]

[0086] 4) Then calculate the energy entropy:

[0087]

[0088] 5) The feature vector that finally constitutes the sound signal:

[0089]

[0090] in, For sound signal r The energy entropy of the order.

[0091] The present invention solves the envelope of the IMF component of each sound signal after modal decomposition through the above Hilbert transform. The characteristic entropy of the envelope essentially obtains the energy entropy of the envelope of the denoised signal and analyzes the energy spectrum of the denoised signal.

[0092] The present invention obtains the envelope of each IMF component in MATLAB for each layer of IMF components decomposed by the ensemble empirical mode decomposition (EEMD) method.

[0093] After the Hilbert transform of the IMF component, the envelope of the signal can effectively obtain the time-frequency characteristics of the signal in a non-stationary state. Envelope analysis eliminates the interference of unnecessary frequencies and effectively demodulates and extracts low-frequency signals, which is more suitable for fault feature extraction.

[0094] In order to extract the features of the sound signal, the energy of the envelope of each IMF component is obtained. i The envelope energy of each IMF component is:

[0095]

[0096] In the formula, f i ( t ) is EEMD i Envelope spectrum characteristics of IMF components after layer decomposition.

[0097] As an embodiment, feature selection is to select feature vectors that can better characterize the operating state of the circuit breaker, avoid the influence of irrelevant feature quantities, so as to improve the generalization ability of the diagnosis model, reduce the fault diagnosis time, and ensure the timeliness of diagnosis. Currently, the commonly used methods mainly include variance method, Pearson coefficient method, sensitivity analysis, regression analysis verification, recursive feature elimination (RFE) and SelectFromModel feature selection. In order to further improve the fault diagnosis rate, this paper needs to compare the two collected feature vectors and select a more suitable fault feature value. Since the variance selection method is simple and only targets the case of taking a value, the selection effect is relatively small; while RFE and SelectFromModel require the help of evaluators and converters, the present disclosure selects existing regression analysis verification and Pearson correlation analysis to verify that obtaining the envelope based on IMF and then obtaining the energy entropy can better characterize the operating state of the circuit breaker than directly obtaining the energy entropy of the IMF component.

[0098] The present disclosure uses the envelope characteristic entropy as a training parameter and a test parameter for subsequent fault diagnosis.

[0099] In some implementations, a GIS circuit breaker sound feature recognition model is constructed to analyze and recognize the extracted sound signal features of the GIS circuit breaker in the operating state to obtain the GIS circuit breaker operating state, specifically:

[0100] The envelope feature entropy is used as the subsequent input feature to form a sample set, and random feature selection is performed through random sampling. After sampling, the samples are put back to allow repetition.

[0101] The improved deep neural network is used as the basis for building the GIS circuit breaker sound feature recognition model. The original one fully connected layer is changed to a three-layer fully connected neural network structure, and multiple hidden layers are set to enhance the overall expression ability of the model. The deep neural network (DNN) is a multi-layer unsupervised neural network, and the output features of the previous layer are used as the input of the next layer for feature learning. Through layer-by-layer feature mapping, it learns better feature expressions for existing inputs. The deep neural network has multiple nonlinear mapping feature transformations, which can fit complex functions. Set the ReLU neuron activation function, and finally use batch normalization and Dropout to improve the generalization performance and robustness of the model.

[0102] The envelope characteristic entropy is input into the constructed GIS circuit breaker sound feature recognition model for analysis and recognition to determine whether the operating status of the GIS circuit breaker is abnormal.

[0103] If it is identified as an abnormal state, the abnormal state information will be recorded in time and an alarm message will be sent;

[0104] If the recognition is normal, the image acquisition device and the voiceprint sensor are used to obtain the image time series data and the sound signal time series data and analyze and recognize them.

[0105] In some embodiments, when an abnormal operating state of a GIS circuit breaker is identified or when the GIS circuit breaker is identified to be in an open state, abnormal state information is recorded and an alarm message is sent.

[0106] The abnormal status information includes the time when the abnormality occurred, the model and code of the circuit breaker where the abnormality occurred, the address of the device to which it belongs, the abnormal status judgment, etc.

[0107] As an embodiment, it is determined whether the operating state of the GIS circuit breaker is an abnormal state. If it is an abnormal state, the abnormal state information is recorded in time and an alarm message is sent. The sound signal data of the GIS circuit breaker in the previous operating state and the sound signal data during the opening and closing process acquired in real time are used to continuously update and correct the sound feature recognition model of the GIS circuit breaker.

[0108] Embodiment 2

[0109] The purpose of this embodiment is to provide a monitoring system for abnormal operation status of a GIS circuit breaker, which specifically includes:

[0110] The image data acquisition module is configured to acquire image time series data of the GIS circuit breaker in real time under the operating state;

[0111] The image detection module is configured to preprocess the image time series data; input the preprocessed image time series data into the GIS circuit breaker operation status image recognition model to obtain the opening and closing status of the GIS circuit breaker;

[0112] The sound recognition module is configured to obtain the sound signal timing data of the GIS circuit breaker in the operating state in real time when it is recognized that the GIS circuit breaker is in the closed state;

[0113] Preprocessing the sound signal time series data to extract the sound signal characteristics under the operating state of the GIS circuit breaker;

[0114] Inputting the sound signal characteristics into a GIS circuit breaker sound characteristic recognition model to obtain the operating status of the GIS circuit breaker;

[0115] The result generating module is configured to determine whether the operating state of the GIS circuit breaker is abnormal. If it is abnormal, the abnormal state information is recorded in time and an alarm message is sent. The sound feature recognition model of the GIS circuit breaker is corrected by using the sound signal data of the GIS circuit breaker in the previous operating state and the sound signal data during the opening and closing process acquired in real time.

[0116] Embodiment 3

[0117] The purpose of this embodiment is to provide a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the method for monitoring the abnormal operating state of a GIS circuit breaker are implemented.

[0118] Embodiment 4

[0119] The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method for monitoring the abnormal operating state of the GIS circuit breaker are implemented.

[0120] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0121] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for monitoring abnormal operation status of a GIS circuit breaker, characterized in that: Specifically include: Real-time acquisition of image time series data of GIS circuit breaker in operation state and preprocessing; Inputting the preprocessed image time series data into the GIS circuit breaker operation status image recognition model to obtain the opening and closing status of the GIS circuit breaker; The GRU neural network is used to build a GIS circuit breaker operating status image recognition model, and the image data features of the opening and closing status are extracted. The image data features in the training set are input into the GRU neural network input layer, and the GIS circuit breaker opening and closing status is used as the output layer. Specifically, there are only two gate structures in the GRU neural network model: update gate G and reset gate C. The values ​​of update gate G and reset gate C in the GRU neural network are determined according to the following formula: In the formula, ε Represents the activation function of the GRU model, using the sigmoid function; ω G ,ω C Respectively represent the update gate in the GRU model structure G With reset gate C The weight of y i -1 Represents the output value of the previous neuron in the GRU model; x i Indicates the first i The input value of each neuron is used to establish the neuron output calculation formula in the GRU network recognition model according to the above formula: In the formula, y Represents the output result of the pending neuron in the GRU model; y i Indicates the first i The output of each neuron; The specific process of the GIS abnormal operation status recognition method based on machine learning is as follows: the abnormal feature data of GIS equipment collected and fused by sensors is input into the input layer of the GRU network recognition model, and the classification recognition results are output after the model learning and training, so as to complete the recognition of the abnormal operation status of GIS; When it is identified that the GIS circuit breaker is in a closed state, the sound signal timing data of the GIS circuit breaker in the operating state is obtained in real time; Preprocessing the sound signal time series data to extract sound signal features of the sound signal time series data; The IMF energy entropy and envelope characteristic entropy are used as the method for extracting the characteristic value of the sound signal; the IMF energy entropy feature is extracted, and after the IMF component is Hilbert transformed, the envelope of the sound signal is obtained to obtain the time-frequency characteristics of the sound signal in a non-stationary state; the energy entropy of the envelope of the sound signal after denoising is obtained, and the energy spectrum of the sound signal after denoising is analyzed; The noise of the sound signal time series data under the operation state of the GIS circuit breaker is removed, specifically including: decomposing the signal into multiple wavelet coefficients of different scales through the signal denoising algorithm of wavelet transform, and then removing the noise by threshold processing the wavelet coefficients, which can effectively remove the noise in the signal and retain the main characteristics of the signal; Adaptive filtering algorithm is used to reduce the impact of echo on sound data; by building a model similar to the echo path and adjusting the filter coefficients by minimizing the root mean square error of the error signal; Perform IMF energy entropy feature extraction, assuming that the circuit breaker operation state sound signal x t The ensemble empirical mode decomposition method is used to obtain n IMF components IMF1, IMF2, ..., IMFn and a residual component res. In practical applications, the residual component res is ignored and the energy sum of each IMF component is considered to be equal to the signal energy: ； In the formula, E is the sound signal energy, E i Indicates i The energy of each IMF component; For the i The ratio of the energy of an IMF component to the total energy, so the IMF energy entropy can be written as, in, H represents energy entropy; Obtain the IMF energy entropy characteristics in MATLAB; Furthermore, the envelope spectrum features are extracted. The signal envelope refers to the line between the maximum and minimum points of the signal. The envelope of the sound signal is the product of the Hilbert transform. Based on the ensemble empirical mode decomposition method, the envelope spectrum features are extracted using the Hilbert method. Assumptions c ( t ) is a discrete real signal, then its Hilbert transform is defined as: Construct the parsing signal: Its amplitude function, that is, the signal envelope is: When the circuit breaker is working in different states, the time periods of the sound signal energy strength are different; the envelope characteristic entropy is used to represent the operating state of the equipment; the envelope characteristic entropy extraction is based on signal noise reduction, and the following steps are performed: 1) Obtain the marginal spectrum of each IMF for verification. If it meets the threshold, obtain the envelope; if it does not meet the threshold, return to the signal denoising step or re-collect the sound signal; 2) The envelope of each IMF component is t Divided into l The energy contained in each envelope is: In the formula, i =1,2,......, l , k= 1,2,......, r , t 1 is the i The start time of the segment, t 2 is the i The termination time of the segment, Q k ( i ) is the k The first order IMF component i Segment energy; 3) Energy normalization: 4) Then calculate the energy entropy: 5) The feature vector that finally constitutes the sound signal: in, For sound signal r Energy entropy of order; After the above Hilbert transform, the envelope of each IMF component of the sound signal after modal decomposition is solved. The characteristic entropy of the envelope essentially obtains the energy entropy of the envelope of the denoised signal and analyzes the energy spectrum of the denoised signal. Obtain the envelope of each IMF component in MATLAB after decomposition of each layer of IMF components by the ensemble empirical mode decomposition method; Obtain the envelope of the signal after Hilbert transform of the IMF component, obtain the time-frequency characteristics of the signal in the non-stationary state, compare the two acquired feature vectors, and select a more suitable fault feature value; Input the sound signal characteristics into the sound characteristic recognition model of the GIS circuit breaker to obtain the operating state of the GIS circuit breaker; determine whether the operating state of the GIS circuit breaker is an abnormal state, and if it is an abnormal state, record the abnormal state information in time and send an alarm message, and use the sound signal data of the GIS circuit breaker in the operating state at the previous moment and the sound signal data in the opening and closing process acquired in real time to correct the sound characteristic recognition model of the GIS circuit breaker; The GIS circuit breaker sound feature recognition model specifically includes: using an improved deep neural network as the basis for constructing the GIS circuit breaker sound feature recognition model, changing the original one fully connected layer to a three-layer fully connected neural network structure, and setting multiple hidden layers to enhance the overall expression ability of the model.

2. The method for monitoring abnormal operation status of a GIS circuit breaker according to claim 1, characterized in that: The GIS equipment fault image data preprocessing specifically includes: The image data is deduplicated, brightened, and contrast-enhanced, and image noise is reduced using the Gaussian bilateral filtering algorithm to ensure data quality, enhance relevant image information features, and retain edge features.

3. The method for monitoring abnormal operation status of a GIS circuit breaker according to claim 1, characterized in that: The sound signal time series data under the operation state of the GIS circuit breaker is preprocessed, specifically including: removing noise, reducing echo interference, adjusting gain and normalizing the sound signal time series data under the operation state of the GIS circuit breaker.

4. A monitoring system for abnormal operation status of GIS circuit breaker, characterized in that: The image data acquisition module is configured to acquire image time series data of the GIS circuit breaker in real time under the operating state; An image detection module is configured to preprocess the image time series data; Inputting the preprocessed image time series data into the GIS circuit breaker operation status image recognition model to obtain the opening and closing status of the GIS circuit breaker; The GRU neural network is used to build a GIS circuit breaker operating status image recognition model, and the image data features of the opening and closing status are extracted. The image data features in the training set are input into the GRU neural network input layer, and the GIS circuit breaker opening and closing status is used as the output layer. Specifically, there are only two gate structures in the GRU neural network model: update gate G and reset gate C. The values ​​of update gate G and reset gate C in the GRU neural network are determined according to the following formula: In the formula, ε Represents the activation function of the GRU model, using the sigmoid function; ω G ,ω C Respectively represent the update gate in the GRU model structure G With reset gate C The weight of y i -1 Represents the output value of the previous neuron in the GRU model; x i Indicates the first i The input value of each neuron is used to establish the neuron output calculation formula in the GRU network recognition model according to the above formula: In the formula, y Represents the output result of the pending neuron in the GRU model; y i Indicates the first i The output of each neuron; The specific process of the GIS abnormal operation status recognition method based on machine learning is as follows: the abnormal feature data of GIS equipment collected and fused by sensors is input into the input layer of the GRU network recognition model, and the classification recognition results are output after the model learning and training, so as to complete the recognition of the abnormal operation status of GIS; The sound recognition module is configured to obtain the sound signal timing data of the GIS circuit breaker in the operating state in real time when it is recognized that the GIS circuit breaker is in the closed state; Preprocessing the sound signal time series data to extract the sound signal characteristics under the operating state of the GIS circuit breaker; The IMF energy entropy and envelope characteristic entropy are used as the sound signal feature value extraction method; the IMF energy entropy feature extraction is performed, and after the IMF component is subjected to Hilbert transform, the envelope of the sound signal is obtained to obtain the time-frequency characteristics of the sound signal in a non-stationary state; the energy entropy of the envelope of the sound signal after noise reduction is obtained, and the energy spectrum of the sound signal after noise reduction is analyzed; the sound signal characteristics are input into the GIS circuit breaker sound feature recognition model to obtain the operating status of the GIS circuit breaker; The noise of the sound signal time series data under the operation state of the GIS circuit breaker is removed, specifically including: decomposing the signal into multiple wavelet coefficients of different scales through the signal denoising algorithm of wavelet transform, and then removing the noise by threshold processing the wavelet coefficients, which can effectively remove the noise in the signal and retain the main characteristics of the signal; Adaptive filtering algorithm is used to reduce the impact of echo on sound data; by building a model similar to the echo path and adjusting the filter coefficients by minimizing the root mean square error of the error signal; Perform IMF energy entropy feature extraction, assuming that the circuit breaker operation state sound signal x t The ensemble empirical mode decomposition method is used to obtain n IMF components IMF1, IMF2, ..., IMFn and a residual component res. In practical applications, the residual component res is ignored and the energy sum of each IMF component is considered to be equal to the signal energy: ； In the formula, E is the sound signal energy, E i Indicates i The energy of each IMF component; For the i The ratio of the energy of an IMF component to the total energy, so the IMF energy entropy can be written as, in, H represents energy entropy; Obtain the IMF energy entropy characteristics in MATLAB; Furthermore, the envelope spectrum features are extracted. The signal envelope refers to the line between the maximum and minimum points of the signal. The envelope of the sound signal is the product of the Hilbert transform. Based on the ensemble empirical mode decomposition method, the envelope spectrum features are extracted using the Hilbert method. Assumptions c ( t ) is a discrete real signal, then its Hilbert transform is defined as: Construct the parsing signal: Its amplitude function, that is, the signal envelope is: When the circuit breaker is working in different states, the time periods of the sound signal energy strength are different; the envelope characteristic entropy is used to represent the operating state of the equipment; the envelope characteristic entropy extraction is based on signal noise reduction, and the following steps are performed: 1) Obtain the marginal spectrum of each IMF for verification. If it meets the threshold, obtain the envelope; if it does not meet the threshold, return to the signal denoising step or re-collect the sound signal; 2) The envelope of each IMF component is t Divided into l The energy contained in each envelope is: In the formula, i =1,2,......, l , k= 1,2,......, r , t 1 is the i The start time of the segment, t 2 is the i The termination time of the segment, Q k ( i ) is the k The first order IMF component i Segment energy; 3) Energy normalization: 4) Then calculate the energy entropy: 5) The feature vector that finally constitutes the sound signal: in, For sound signal r Energy entropy of order; After the above Hilbert transform, the envelope of each IMF component of the sound signal after modal decomposition is solved. The characteristic entropy of the envelope essentially obtains the energy entropy of the envelope of the denoised signal and analyzes the energy spectrum of the denoised signal. Obtain the envelope of each IMF component in MATLAB after decomposition of each layer of IMF components by the ensemble empirical mode decomposition method; Obtain the envelope of the signal after Hilbert transform of the IMF component, obtain the time-frequency characteristics of the signal in the non-stationary state, compare the two acquired feature vectors, and select a more suitable fault feature value; The result generation module is configured to determine whether the operating state of the GIS circuit breaker is abnormal. If it is abnormal, the abnormal state information is recorded in time and an alarm message is sent. The sound signal data of the GIS circuit breaker in the previous operating state and the sound signal data in the opening and closing process obtained in real time are used to correct the sound feature recognition model of the GIS circuit breaker. The GIS circuit breaker sound feature recognition model specifically includes: using an improved deep neural network as the basis for constructing the GIS circuit breaker sound feature recognition model, changing the original one fully connected layer to a three-layer fully connected neural network structure, and setting multiple hidden layers to enhance the overall expression ability of the model.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for monitoring the abnormal operating state of a GIS circuit breaker as described in any one of claims 1 to 3 are implemented.

6. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the method for monitoring the abnormal operating state of a GIS circuit breaker according to any one of claims 1 to 3 are implemented.

Images