High-voltage circuit breaker fault identification method and system

Abstract

The invention relates to the technical field of power equipment fault recognition, and discloses a high-voltage circuit breaker fault recognition method and system, and the method comprises the steps: obtaining the vibration data of a to-be-recognized circuit breaker in response to a fault recognition request, carrying out the preprocessing of the vibration data, inputting the preprocessed vibration data into a built target fault recognition model, and carrying out the recognition of a fault. The target fault recognition model comprises an information extraction layer and a decision layer, feature extraction is performed on the preprocessed vibration data through the information extraction layer to obtain a target feature vector, the target feature vector is input into the decision layer to perform fault recognition, and fault recognition information of the to-be-recognized circuit breaker is output. The technical problem of how to improve the fault identification accuracy of the high-voltage circuit breaker is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment fault identification, and in particular to a high-voltage circuit breaker fault identification method and system. Background Art

[0002] High-voltage circuit breakers occupy an extremely important position in the power system, playing a key role in switching, controlling or protecting in all aspects of power generation, transmission, distribution, power conversion and consumption. With the development of the power industry, the scale of the power system continues to expand, the voltage level continues to increase, and the performance and reliability requirements of high-voltage circuit breakers are becoming increasingly stringent. However, with the increasing complexity of the power system, traditional circuit breakers face many challenges. For example, under high voltage and high current conditions, the arc generated during breaking is difficult to effectively control, which can easily cause equipment damage and power failures.

[0003] At present, in the existing high-voltage circuit breaker fault identification technology, for traditional feature extraction methods, manual feature engineering is often required, which relies on the experience and knowledge of experts. When faced with complex failure modes of high-voltage circuit breakers, manually selected features may not be able to fully and accurately reflect the essential information of the fault. For example, when dealing with spring preload reduction failures, relying only on a few features such as vibration frequency selected by manual experience may miss some subtle features related to preload changes, thereby affecting the accuracy of fault identification. Summary of the invention

[0004] The present invention provides a high-voltage circuit breaker fault identification method and system, which solves the technical problem of how to improve the accuracy of high-voltage circuit breaker fault identification.

[0005] A first aspect of the present invention provides a high-voltage circuit breaker fault identification method, comprising:

[0006] In response to a fault identification request, obtaining vibration data of a circuit breaker to be identified, and preprocessing the vibration data;

[0007] Inputting the preprocessed vibration data into a built target fault identification model, wherein the target fault identification model includes an information extraction layer and a decision layer;

[0008] Performing feature extraction on the preprocessed vibration data through the information extraction layer to obtain a target feature vector;

[0009] The target feature vector is input into the decision layer for fault identification, and fault identification information of the circuit breaker to be identified is output.

[0010] Optionally, the information extraction layer includes an improved residual network module, a first layer normalization module, a feedforward network module, a second layer normalization module and a long short-term memory network module connected in sequence, and the step of extracting features from the preprocessed vibration data through the information extraction layer to obtain a target feature vector includes:

[0011] Performing feature extraction on the preprocessed vibration data through the improved residual network to obtain a residual feature vector;

[0012] Performing a normalization operation on the residual feature vector through the first-layer normalization module to obtain a first normalized vector;

[0013] constructing an initial feature vector using the first normalized vector and the preprocessed vibration data;

[0014] Performing feature extraction on the initial feature vector through the feedforward network module to obtain a feedforward vector;

[0015] Performing a normalization operation on the feedforward vector by the second-layer normalization module to obtain a second normalized vector;

[0016] constructing an intermediate feature vector using the second normalized vector and the initial feature vector;

[0017] The intermediate feature vector is subjected to feature extraction by the long short-term memory network module to obtain a target feature vector.

[0018] Optionally, the improved residual network includes a downscaling module and an upscaling module connected in sequence, the downscaling module includes a plurality of downscaling submodules, the upscaling module includes a plurality of upscaling submodules, and the step of extracting features from the preprocessed vibration data through the improved residual network to obtain a residual feature vector includes:

[0019] Performing a continuous downscaling operation on the preprocessed vibration data through a plurality of sequentially connected downscaling submodules to obtain a downscaled vector;

[0020] The downscaled vector is continuously upscaled by a plurality of upscaling submodules connected in sequence to obtain a residual feature vector.

[0021] Optionally, the downscaling submodule includes a residual unit and a strided convolution unit connected in sequence, the residual unit includes a plurality of residual blocks, and the step of performing a continuous downscaling operation on the preprocessed vibration data through the plurality of the downscaling submodules connected in sequence to obtain a downscaled vector includes:

[0022] Performing a continuous residual operation on the preprocessed vibration data through a plurality of sequentially connected residual blocks to obtain a first eigenvector;

[0023] A downsampling operation is performed on the first feature vector by the strided convolution unit to obtain a downscaled vector.

[0024] Optionally, the upscaling submodule includes a residual unit and a deconvolution unit connected in sequence, the residual unit includes a plurality of residual blocks, and the step of performing a continuous upscaling operation on the downscaled vector through the plurality of sequentially connected upscaling submodules to obtain a residual feature vector includes:

[0025] Performing a residual operation on the downscaled vector through a plurality of sequentially connected residual blocks to obtain a second eigenvector;

[0026] An upsampling operation is performed on the second feature vector by the deconvolution unit to obtain a residual feature vector.

[0027] Optionally, the residual block includes a first convolution unit, a rectified linear unit, and a second convolution unit connected in sequence, and the specific processing steps of the residual block include:

[0028] Performing a convolution operation on the input feature data by the first convolution unit to obtain first convolution data;

[0029] Performing a nonlinear transformation operation on the first convolution data by the rectified linear unit to obtain conversion data;

[0030] The second convolution unit performs a convolution operation on the conversion data to obtain output feature data.

[0031] Optionally, the decision layer includes a one-dimensional convolution module and an identification decision module connected in sequence, and the step of using the target feature vector to input the decision layer for fault identification and outputting the fault identification information of the circuit breaker to be identified includes:

[0032] Extracting features of the target feature vector in sequence through the one-dimensional convolution module to obtain a one-dimensional feature vector;

[0033] The one-dimensional feature vector is input into the identification decision module for fault identification, the fault type of the circuit breaker to be identified is generated, and fault identification information is generated.

[0034] Optionally, it also includes:

[0035] Performing fault simulation on the high-voltage circuit breaker to obtain fault simulation data of the high-voltage circuit breaker in different fault states;

[0036] Constructing a training set using all of the fault simulation data and benchmark simulation data of the high-voltage circuit breaker in a benchmark state;

[0037] Using the training set to input a preset initial fault recognition model for training, and outputting a training feature vector;

[0038] Calculating a target loss value between the training feature vector and a preset standard feature vector;

[0039] Comparing the target loss value with a preset standard loss value;

[0040] If the target loss value is less than the preset standard loss value, the training is stopped and a target fault recognition model is generated.

[0041] Optionally, it also includes:

[0042] If the target loss value is greater than or equal to the preset standard loss value, adjusting the model parameters of the preset initial fault identification model;

[0043] Jump to the step of using the training set to input a preset initial fault recognition model for training and outputting a training feature vector until the target loss value is less than the preset standard loss value.

[0044] A second aspect of the present invention provides a high-voltage circuit breaker fault identification system, comprising:

[0045] A response module, used for responding to a fault identification request, obtaining vibration data of a circuit breaker to be identified, and preprocessing the vibration data;

[0046] A data input module, used to input the pre-processed vibration data into a built target fault identification model, wherein the target fault identification model includes an information extraction layer and a decision layer;

[0047] An information extraction module, used for performing feature extraction on the preprocessed vibration data through the information extraction layer to obtain a target feature vector;

[0048] The fault identification module is used to use the target feature vector to input the decision layer to perform fault identification and output fault identification information of the circuit breaker to be identified.

[0049] It can be seen from the above technical solutions that the present invention has the following advantages:

[0050] In the present invention, in response to a fault identification request, vibration data of a circuit breaker to be identified is obtained and the vibration data is preprocessed, and the preprocessed vibration data is input into a constructed target fault identification model, the target fault identification model includes an information extraction layer and a decision layer, and the preprocessed vibration data is feature extracted through the information extraction layer to obtain a target feature vector, and the target feature vector is input into the decision layer for fault identification, and the fault identification information of the circuit breaker to be identified is output; the present invention utilizes the information extraction layer in the constructed target fault identification model to perform feature extraction on the vibration data of the circuit breaker to be identified, and can effectively extract a high-precision target feature vector from the vibration data, and then perform fault identification through the decision layer, thereby realizing accurate identification of high-voltage circuit breaker faults, and solving the technical problem of how to improve the accuracy of high-voltage circuit breaker fault identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0052] Figure 1 A flowchart of a method for identifying a high-voltage circuit breaker fault provided in Embodiment 1 of the present invention;

[0053] Figure 2 A flowchart of the steps of a high-voltage circuit breaker fault identification method provided in Embodiment 2 of the present invention;

[0054] Figure 3 It is a structural diagram of the target fault identification model;

[0055] Figure 4 It is a characteristic spectrum of vibration data of the high-voltage circuit breaker in a reference state;

[0056] Figure 5 It is a characteristic spectrum of vibration data of a high voltage circuit breaker in a low working voltage fault state;

[0057] Figure 6 It is a characteristic spectrum of vibration data of a high-voltage circuit breaker in a fault state of reducing the pre-tensioning force of a spring;

[0058] Figure 7 It is a characteristic spectrum of vibration data of a high-voltage circuit breaker in a loose bolt fault state;

[0059] Figure 8 It is a characteristic spectrum of vibration data of a high-voltage circuit breaker in a core bonding fault state;

[0060] Fig. 9 It is a schematic diagram of the structure of the original residual network;

[0061] Fig.10 Schematic diagram of the structure of the improved residual network;

[0062] Fig.11 Schematic diagram of the structure of the residual block;

[0063] Fig.12 It is the input layer sample distribution diagram of the target fault identification model;

[0064] Fig.13 It is the sample distribution diagram of the output layer of the target fault recognition model;

[0065] Fig.14 This is a structural block diagram of a high-voltage circuit breaker fault identification system provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0066] The embodiments of the present invention provide a high-voltage circuit breaker fault identification method and system, which are used to solve the technical problem of how to improve the accuracy of high-voltage circuit breaker fault identification.

[0067] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] The current detection signals of mechanical faults of high-voltage circuit breakers include coil current, contact travel and speed, mechanical vibration and sound during opening and closing operations. Vibration signals have unique advantages in fault diagnosis of high-voltage circuit breakers. Technicians can use the real-time and online monitoring characteristics of circuit breakers to capture internal operating abnormalities in a timely manner. In addition, the non-invasive nature of vibration signal acquisition does not require extensive system disassembly, minimizing the impact on the power system. Vibration signals provide comprehensive fault information, including aspects such as contact status and mechanical vibration, which helps to make a more comprehensive and accurate health assessment of circuit breakers. The response of vibration signals to various fault modes makes vibration signals an efficient and comprehensive diagnostic tool, providing reliable support for the stable operation of power systems.

[0069] The present invention simulates four common faults in the working mechanism of a high-voltage circuit breaker and compares them with the fault-free (reference state) working condition. The vibration signals associated with different types of faults are analyzed. On this basis, a circuit breaker fault identification method based on a target fault identification model of an improved deep residual network (DRSN) is proposed. This method can effectively extract features from vibration signals, thereby achieving accurate identification of faults in the working mechanism of a high-voltage circuit breaker. By processing the vibration signal using an improved deep residual neural network (DRSN), DRSN is successfully applied to noise reduction and key feature extraction, and a new high-voltage circuit breaker fault diagnosis scheme is introduced. This method is expected to improve the accuracy and efficiency of high-voltage circuit breaker fault diagnosis in practical applications, and provide strong support for the reliable operation of power systems.

[0070] See also Figure 1 , Figure 1 A flowchart of the steps of a high-voltage circuit breaker fault identification method provided in Embodiment 1 of the present invention.

[0071] The present invention provides a high-voltage circuit breaker fault identification method, comprising:

[0072] Step 101: respond to a fault identification request, obtain vibration data of a circuit breaker to be identified, and pre-process the vibration data.

[0073] The fault identification request refers to a request instruction for performing fault identification on a circuit breaker to be identified.

[0074] Vibration data refers to the signal corresponding to the vibration generated by various factors such as movement, collision, friction and electromagnetic force of internal mechanical parts during the operation process (such as opening and closing operations) of the circuit breaker to be identified.

[0075] Preprocessing refers to normalizing the vibration data to ensure that the vibration data features share a consistent scale so that the target fault identification model can converge quickly.

[0076] In the embodiment of the present invention, in response to a received request instruction for fault identification of a circuit breaker to be identified, the request instruction is read, vibration data generated by the circuit breaker to be identified during operation is acquired, and the vibration data is normalized.

[0077] Step 102: input the preprocessed vibration data into a built target fault identification model, where the target fault identification model includes an information extraction layer and a decision layer.

[0078] The target fault identification model refers to a model used to identify possible faults of the circuit breaker. It determines whether the circuit breaker is in a fault state and determines the type of fault by analyzing the circuit breaker vibration data.

[0079] The information extraction layer refers to an important component of the target fault identification model, which is used to extract features from the data input to the model and extract feature information related to the circuit breaker fault.

[0080] The decision layer refers to an important component of the target fault identification model, which is used to process the feature information provided by the information extraction layer and make judgments and decisions on the fault status of the circuit breaker.

[0081] In an embodiment of the present invention, the preprocessed vibration data is input into a constructed target fault identification model for processing.

[0082] Step 103: extract features from the preprocessed vibration data through the information extraction layer to obtain a target feature vector.

[0083] In the embodiment of the present invention, feature extraction is performed on the preprocessed vibration data through the information extraction layer to extract a target feature vector related to the circuit breaker fault.

[0084] Step 104: Use the target feature vector to input the decision layer to perform fault identification, and output fault identification information of the circuit breaker to be identified.

[0085] In the embodiment of the present invention, the target feature vector extracted by the information extraction layer is input into the decision layer for fault identification, thereby outputting the fault identification information of the circuit breaker to be identified.

[0086] In the present invention, in response to a fault identification request, vibration data of a circuit breaker to be identified is obtained and the vibration data is preprocessed, and the preprocessed vibration data is input into a constructed target fault identification model, the target fault identification model includes an information extraction layer and a decision layer, and the preprocessed vibration data is feature extracted through the information extraction layer to obtain a target feature vector, and the target feature vector is input into the decision layer for fault identification, and the fault identification information of the circuit breaker to be identified is output; the present invention utilizes the information extraction layer in the constructed target fault identification model to perform feature extraction on the vibration data of the circuit breaker to be identified, and can effectively extract a high-precision target feature vector from the vibration data, and then perform fault identification through the decision layer, thereby realizing accurate identification of high-voltage circuit breaker faults, and solving the technical problem of how to improve the accuracy of high-voltage circuit breaker fault identification.

[0087] See also Figure 2 , Figure 2 A flowchart of the steps of a high-voltage circuit breaker fault identification method provided in Embodiment 2 of the present invention.

[0088] The present invention provides a high-voltage circuit breaker fault identification method, comprising:

[0089] Step 201: In response to a fault identification request, vibration data of a circuit breaker to be identified is obtained.

[0090] In the embodiment of the present invention, the specific implementation process of step 201 is similar to step 101 and will not be repeated here.

[0091] Step 202: pre-process the vibration data.

[0092] In the embodiment of the present invention, in order to ensure that the features share a consistent scale and make the fault identification model converge quickly, the vibration signal is normalized as the first step of dividing the training and test samples of the fault identification model. The original data is processed using the man-max normalization method, and the expression is as follows:

[0093]

[0094] In the formula, represents the normalized vibration data, represents the original vibration data, represents the minimum value of the data sample, Indicates the maximum value of the data samples.

[0095] It is worth mentioning that before step 203, the following steps are also included:

[0096] A1. Perform fault simulation on the high-voltage circuit breaker to obtain fault simulation data of the high-voltage circuit breaker in different fault states.

[0097] In the embodiment of the present invention, by simulating the typical fault of the operating mechanism of the high-voltage circuit breaker, a 126kV SF6 porcelain column circuit breaker is selected. The operating mechanism of the circuit breaker is a spring operating mechanism, which stores energy by compressing the spring. When the circuit breaker needs to be opened or closed, the stored spring energy provides sufficient force to help overcome the electrical attraction or mechanical friction between the switch contacts. This will ensure reliable switch operation and complete the opening and closing actions.

[0098] The vibration signal during the operation of the circuit breaker is measured by an acceleration sensor. The present invention selects a piezoelectric accelerometer CA-YD-180 with a maximum impact acceleration of 1000g. The sensor is easy to use and has the advantages of passive operation, small size, wide frequency response, high sensitivity, strong housing, and strong anti-interference. An oscilloscope is used to collect vibration signals.

[0099] It should be noted that the present invention simulates four typical faults, namely, too low operating voltage, reduced spring preload, loose bolts and iron core adhesion.

[0100] Low working voltage is a common fault in the operation of circuit breakers. When the fault is serious, it will cause the circuit breaker to open and close, resulting in significant losses. According to the "Preventive Test Procedures for Power Equipment", the minimum working voltage of the electromagnetic coil used to open and close the circuit breaker should be between 30% and 65% of the rated working voltage. Considering the actual working characteristics of the circuit breaker, a voltage of 110 volts is set to simulate the low working voltage fault.

[0101] The magnitude of the spring force directly affects the opening and closing speed. Long-term operation will cause fatigue damage to the closing spring, reduce energy release, and seriously affect the interruption performance of the circuit breaker. The present invention realizes the simulation of the spring preload force by adjusting the screw length through the nut.

[0102] In addition, loose bolts in the circuit breaker operating mechanism represent a potential fault that may adversely affect the normal operation and safety of the equipment. Loose bolts can cause the mechanism components to be misaligned, resulting in unstable or uncontrolled conditions during operation. In addition, loose bolts can cause mechanical vibrations that affect the overall performance of the operating mechanism. Timely detection and repair of this fault is key to ensure reliable operation of the circuit breaker. In the experiment, this fault was simulated by loosening the tightening bolts.

[0103] If the circuit breaker operating mechanism box is not tightly sealed, dust and other debris may enter the box, which may cause dirt to form between the iron core and the copper sleeve. In addition, factors such as insufficient lubricating oil and rust on the iron core can also cause the iron core rod to stick, affecting the normal opening and closing of the circuit breaker. In severe cases, it may cause the iron core to be unable to disengage from the locking device, resulting in continuous power supply of the opening and closing coils, overheating, and even burning. In the experiment, double-sided tape was wrapped around the iron core to increase the damping of the iron core and qualitatively simulate the failure of the opening and closing coil iron core.

[0104] A2. Construct a training set using all fault simulation data and benchmark simulation data of the high-voltage circuit breaker in a benchmark state.

[0105] In the embodiment of the present invention, five states of the high-voltage circuit breaker are considered, as shown in Table 1. For each state, 40 vibration signals are measured. Subsequently, 25 random samples are selected for training and the remaining samples are tested. The vibration signals of the high-voltage circuit breaker under different fault states are shown in Table 1. Figure 4-Figure 8 shown.

[0106] Table 1. Fault description of high voltage circuit breaker

[0107]

[0108] It should be noted that the vibration signals collected above are divided into training sets and test sets, of which 70% are used as training sets and 30% are used as test sets. The batch size is set to 50, and a data iterator is constructed to input data into the model in batches of appropriate size during training.

[0109] A3. Use the training set to input the preset initial fault recognition model for training and output the training feature vector.

[0110] The preset initial fault identification model refers to a basic model framework set at the beginning of the circuit breaker fault identification task. Figure 3 The basic model framework includes an information extraction layer and a decision layer. The information extraction layer uses the improved DRSN (Deep Residual neural network) deep residual neural network as the benchmark model of the fault recognition model. The information extraction layer is sequentially connected to the improved residual network module, the first layer normalization module, the feedforward network module, the second layer normalization module and the long short-term memory network module. The decision layer includes a one-dimensional convolution module and a recognition decision module that are sequentially connected.

[0111] It is worth mentioning that the improved DRSN (Deep Residual neural network) deep residual neural network is used as the benchmark model of the fault recognition model. The improved DRSN (Deep Residual neural network) is a network structure based on ResNet, which is improved by an attention mechanism - Squeeze and Excitation Network (SENet) and a soft threshold function. Fig. 9 , Fig. 9 This is a schematic diagram of the structure of the original residual network. By improving the original residual network, the improved DRSN (Deep Residual neural network) deep residual neural network is obtained. Please refer to Fig.10 , Fig.10 Schematic diagram of the structure of the improved residual network.

[0112] It should be noted that DRSN replaces the nonlinear layer of the SENet structure with a soft threshold function and integrates it into the ResNet network structure. The ResNet operation is introduced into the operation of the original residual network. It is formed by stacking 6 residual blocks and using the soft threshold setting of each feature channel in SENet to replace the weighted operation, so that it can retain more useful features from the data set. This enhances the feature learning ability of DRSN on complex data sets and ultimately forms a better improved residual network.

[0113] In the embodiment of the present invention, a training set is used to input a preset initial fault recognition model for training, and a training feature vector is output.

[0114] A4. Calculate the target loss value between the training feature vector and the preset standard feature vector.

[0115] The preset standard feature vector refers to a vector that is pre-set in circuit breaker fault identification and represents a typical feature combination under a normal state or a known fault state.

[0116] The target loss value refers to the value used to measure the difference between the training feature vector and the preset standard feature vector, reflecting the deviation between the model's predicted results and the expected results.

[0117] In an embodiment of the present invention, a target loss value between a training feature vector and a preset standard feature vector is calculated, wherein a mean square error is used as a loss function of the model, and the target loss value is calculated by the loss function.

[0118] A5. Compare the target loss value with the preset standard loss value.

[0119] The preset standard loss value refers to a reference standard of loss value set in advance during model training or evaluation. The preset standard loss value is a threshold used to determine whether the performance of the model meets expectations or to decide whether model training should be stopped.

[0120] In the embodiment of the present invention, the target loss value is compared with a preset standard loss value.

[0121] A6. If the target loss value is less than the preset standard loss value, the training is stopped and the target fault recognition model is generated.

[0122] In an embodiment of the present invention, if the target loss value is less than a preset standard loss value, the training is stopped and a target fault recognition model is generated.

[0123] A7. If the target loss value is greater than or equal to the preset standard loss value, the model parameters of the preset initial fault identification model are adjusted.

[0124] In an embodiment of the present invention, if the target loss value is greater than or equal to the preset standard loss value, the model parameters of the preset initial fault identification model are adjusted, wherein the model parameters include but are not limited to weights, biases, and activation function related parameters.

[0125] A8. Jump to the step of using the training set to input the preset initial fault recognition model for training and outputting the training feature vector until the target loss value is less than the preset standard loss value.

[0126] In the embodiment of the present invention, the step jumps to the step of using the training set to input the preset initial fault recognition model for training and outputting the training feature vector until the target loss value is less than the preset standard loss value, thereby generating a target fault recognition model.

[0127] Step 203: input the preprocessed vibration data into the constructed target fault identification model, which includes an information extraction layer and a decision layer.

[0128] In an embodiment of the present invention, the preprocessed vibration data is input into a constructed target fault identification model for processing, and the target fault identification model includes an information extraction layer and a decision layer.

[0129] Step 204: extract features from the preprocessed vibration data through the information extraction layer to obtain a target feature vector.

[0130] Further, the information extraction layer includes an improved residual network module, a first layer normalization module, a feedforward network module, a second layer normalization module and a long short-term memory network module connected in sequence, and step 204 may include the following sub-steps:

[0131] S11. Extract features of the preprocessed vibration data by improving the residual network to obtain a residual feature vector.

[0132] It should be noted that the feature extraction of the preprocessed vibration data is performed by improving the residual network, and the preprocessed vibration data is first downscaled and then upscaled. When processing large-scale data, the original high-resolution vibration data may contain too much detailed information, resulting in a huge amount of data. Therefore, data downscaling is required to speed up data processing and reduce the demand for computing resources. It is worth mentioning that since the vibration data of the circuit breaker often contains a lot of noise information, some high-frequency noise signals will be smoothed out during the data downscaling operation, thereby filtering the noise. After removing the interference of high-frequency noise, the low-frequency vibration characteristics related to the fault will be more obvious, which helps to highlight the key features in the vibration data. In the process of data upscaling, the model reconstructs the high-resolution data based on the downscaled data features, which helps to explore the internal structure and feature relationship of the vibration data. The model learns the feature conversion relationship of vibration data at different scales. This enables the model to more accurately identify faults when faced with newly collected vibration data of different resolutions, such as when the data resolution changes due to sensor replacement or adjustment of the acquisition device settings (such as replacing the oscilloscope used to collect vibration signals with other types of sensors mentioned in the above steps), thereby improving the generalization ability of the model.

[0133] Further, the improved residual network includes a downscaling module and an upscaling module connected in sequence, the downscaling module includes a plurality of downscaling submodules, and the upscaling module includes a plurality of upscaling submodules;

[0134] It is worth mentioning that in the constructed improved residual network, through the connection method of the downscaling module and the upscaling module, as well as the skip connection structure of the internal multi-level residual block, an inter-layer residual skip connection is formed, so that the gradient can be directly propagated through the skip connection, avoiding excessive gradient attenuation or amplification in the multi-layer network, alleviating the problems of gradient explosion and degradation in the deep network, and making the model easier to train.

[0135] See also Fig.10 Taking four scales as an example, both the downscaling module and the upscaling module contain four scale sub-modules. The residual unit in each scale sub-module is formed by stacking six consecutive residual blocks. From the first scale sub-module to the fourth scale sub-module, the number of channels in each layer is 64, 128, 256, and 512, respectively.

[0136] The downscaling module includes a first downscaling submodule, a second downscaling submodule, a third downscaling submodule and a fourth downscaling submodule connected in sequence;

[0137] The upscaling module includes a fourth upscaling submodule, a third upscaling submodule, a second upscaling submodule and a first upscaling submodule connected in sequence;

[0138] The fourth downscaling submodule is connected to the fourth upscaling submodule;

[0139] The following is Fig.10 The basic steps to improve the residual network in :

[0140] Fig.10 Each rectangular module in the figure refers to a state at each scale. Specifically, starting from the leftmost (first from the left) red module (initial state), it goes through the 3×3 convolution and rectified linear unit process (the black arrows indicate: 3×3 Conv, ReLU) to become the second from the left green module state.

[0141] Then, based on the state of the second module from the left, it passes through 6 consecutive residual blocks containing 1×1×64 and 3×3×64 convolutional layers (the red arrows indicate: 6 Residual Blocks) to become the state of the third blue module from the left.

[0142] Based on the state of the third module on the left, it becomes the state of the fourth yellow module on the left through 2×2 strided convolution SConv (i.e. Skip Convolution) (i.e. the meaning indicated by the green arrow: 2×2 SConv).

[0143] The number 64 under the second, third and fourth boxes from the left refers to the 64 channels corresponding to the first scale (the number 64 under the second, third and fourth boxes from the right also refers to the 64 channels corresponding to the first scale, and the numbers under the boxes between the fourth and fourth boxes from the left and right are similar, 128, 256 and 512 correspond to 128, 256 and 512 channels at the second, third and fourth scales respectively, and the first on the right is an illustration of the final module state of this upscaling and downscaling).

[0144] In the fourth yellow module state on the left, the skip connection (i.e. Skip Connection) shown by the blue line can reach the fourth state on the right at the same scale.

[0145] Another case without skip connection is the process from the left four modules to the left five modules, which becomes the left five modules through 6 residual blocks (the red arrows indicate: 6 Residual Blocks), that is, from the first scale to the second scale.

[0146] Subsequently, the left fifth state is transformed into the left sixth state through a 2×2 strided convolution SConv (i.e., Skip Convolution) (i.e., the meaning indicated by the green arrow: 2×2 SConv).

[0147] Similarly, in the left six states, the skip connection shown by the blue line can lead to the right six states of the same scale.

[0148] Similarly, another case without skip connection is that the left six modules go through 6 residual blocks to the left seven modules, that is, from the second scale to the third scale.

[0149] The left seven state is transformed into the left eight state through 2×2 strided convolution SConv.

[0150] Similarly, in the left eight state, the right eight state of the same scale can be reached through the skip connection shown by the blue line.

[0151] In addition, in the left eight state, 6 residual blocks can be used to enter the left nine state, that is, from the third scale to the fourth scale.

[0152] In the left nine, i.e., fourth scale state, it changes to the right eight module state through 6 residual blocks.

[0153] Subsequently, in the state shown on the right eight, it changes to the right seven module state through a 2×2 deconvolution TConv scaling operation (i.e., the purple arrow means: 2×2 TConv).

[0154] The states shown in the modules from the right seventh module to the right sixth, right fifth, right fourth, right third and right second modules are all changed through the 2×2 deconvolution TConv scaling operation.

[0155] From the second right state, it is restored to the first right state through a 3×3 convolution and a rectified linear unit process (the black arrows indicate: 3×3Conv, ReLU).

[0156] S11 may include the following sub-steps:

[0157] S111 . Performing a continuous downscaling operation on the preprocessed vibration data through a plurality of sequentially connected downscaling submodules to obtain a downscaled vector.

[0158] Further, the downscaling submodule includes a residual unit and a strided convolution unit connected in sequence, the residual unit includes a plurality of residual blocks, and S111 may include the following sub-steps:

[0159] Specifically, the first downscaling submodule uses 6 consecutive residual blocks containing 1×1×64 and 3×3×64 convolutional layers, and uses 2×2 strided convolution (SConv) with a step size of 2 when reducing the dimension.

[0160] S1111 . Perform a continuous residual operation on the preprocessed vibration data through a plurality of sequentially connected residual blocks to obtain a first eigenvector.

[0161] S1112. Perform a downsampling operation on the first eigenvector through a strided convolution unit to obtain a downscaled vector.

[0162] The downsampling operation refers to the process of coarsely extracting data features, which is used to merge the vibration data features in a local area.

[0163] In an embodiment of the present invention, the preprocessed vibration data is input into a first downscaling submodule for continuous residual operation and downsampling operation, and then input into a second downscaling submodule, and so on. In the present invention, four downscaling submodules are provided, so four continuous residual operations and downsampling operations are repeatedly performed to obtain a downscaled vector with noise suppression and enhanced stability.

[0164] S112, performing a continuous upscaling operation on the downscaled vector through a plurality of sequentially connected upscaling submodules to obtain a residual feature vector.

[0165] Further, the upscaling submodule includes a residual unit and a deconvolution unit connected in sequence, the residual unit includes a plurality of residual blocks, and S112 may include the following sub-steps:

[0166] Specifically, the first upscaling submodule uses 6 consecutive residual blocks containing 1×1×64 and 3×3×64 convolutional layers, and uses 2×2 deconvolution (TConv) when upscaling.

[0167] S1121. Perform a residual operation on the downscaled vector through a plurality of sequentially connected residual blocks to obtain a second eigenvector.

[0168] S1122, perform an upsampling operation on the second feature vector through a deconvolution unit to obtain a residual feature vector

[0169] In an embodiment of the present invention, the downscaled vector is input into the fourth upscaling submodule for continuous residual operation and upsampling operation, and then input into the third upscaling submodule, and so on. In the present invention, four upscaling submodules are provided, so four continuous residual operations and upsampling operations are repeatedly performed to obtain a high-dimensional residual feature vector with enhanced feature details.

[0170] For further information, see Fig.11 , Conv: the abbreviation of Convolution corresponding to convolution, ReLU: Rectified Linear Unit, the residual block in S111 and S112 includes a first convolution unit, a rectified linear unit and a second convolution unit connected in sequence, and the specific processing steps of the residual block include:

[0171] It should be noted that in S111, the input feature data is the pre-processed vibration data. Since it is composed of a plurality of residual blocks in cascade, the output feature data can be understood as the first feature vector output by the last residual block;

[0172] In S112, the input feature data is a downscaled vector, and since it is composed of a plurality of cascaded residual blocks, the output feature data can be understood as the second feature vector output by the last residual block.

[0173] In each residual block, the first convolution unit uses a layer of 1×1 convolution, the second convolution unit uses a layer of 3×3 convolution, and the rectified linear unit uses a layer of ReLU activation function.

[0174] a1. Perform a convolution operation on the input feature data through the first convolution unit to obtain first convolution data.

[0175] In the embodiment of the present invention, a convolution operation is performed on the input feature data through the first convolution unit to extract relatively preliminary vibration features, namely, the first convolution data.

[0176] a2. Perform a nonlinear transformation operation on the first convolution data through a rectified linear unit to obtain converted data.

[0177] In an embodiment of the present invention, a nonlinear transformation operation is performed on the first convolution data through a modified linear unit, and the ReLU function between the convolution units plays a role of nonlinear conversion. Weak features in the vibration data may be enhanced after passing through the ReLU function, while some useless features close to zero may be suppressed, which helps to improve the distinguishability of the features, so that the residual block can better handle the nonlinear relationship in the vibration data, thereby more accurately extracting features related to the vibration state.

[0178] a3. Perform a convolution operation on the converted data through the second convolution unit to obtain output feature data.

[0179] In the embodiment of the present invention, by performing a convolution operation on the converted data through the second convolution unit, more complex and abstract features can be further extracted based on the output of the first convolution unit, that is, output feature data.

[0180] S12. Perform a normalization operation on the residual feature vector through a first-layer normalization module to obtain a first normalized vector.

[0181] In the embodiment of the present invention, a normalization operation is performed on the residual feature vector through a first-layer normalization module to obtain a first normalized vector.

[0182] S13. Construct an initial feature vector using the first normalized vector and the preprocessed vibration data.

[0183] In the embodiment of the present invention, the first normalized vector and the preprocessed vibration data are used to construct an initial feature vector as the input of the feedforward network module.

[0184] S14. Extract features from the initial feature vector through a feedforward network module to obtain a feedforward vector.

[0185] In an embodiment of the present invention, the initial feature vector is passed through a two-layer feedforward network. In the first hidden layer, neurons learn local features in the vibration data, such as changes in vibration amplitude in a short period of time. As the data propagates forward in the network, the second hidden layer further combines and transforms the initial feature vector based on the output of the first hidden layer, thereby extracting a more complex feedforward vector that is more relevant to the vibration state.

[0186] S15. Perform a normalization operation on the feedforward vector through a second-layer normalization module to obtain a second normalized vector.

[0187] In the embodiment of the present invention, the feedforward vector is normalized by a second-layer normalization module to obtain a second normalized vector.

[0188] S16. Construct an intermediate feature vector using the second normalized vector and the initial feature vector.

[0189] In an embodiment of the present invention, the second normalized vector and the initial feature vector are used to construct an intermediate feature vector as an input of the long short-term memory network module.

[0190] S17. Extract features from the intermediate feature vector using a long short-term memory network module to obtain a target feature vector.

[0191] In an embodiment of the present invention, the intermediate feature vector is input into a long short-term memory network module to further extract a deep target feature vector.

[0192] Step 205: Use the target feature vector to input the decision layer to perform fault identification, and output fault identification information of the circuit breaker to be identified.

[0193] Furthermore, the decision layer includes a one-dimensional convolution module and a recognition decision module connected in sequence, and step 205 may include the following sub-steps:

[0194] S21. Extract features of the target feature vector in sequence through a one-dimensional convolution module to obtain a one-dimensional feature vector.

[0195] In an embodiment of the present invention, since the data processed by the improved residual network module is high-dimensional data, the information extraction layer outputs a high-dimensional target feature vector. The high-dimensional target feature vector is reduced in dimension by a one-dimensional convolution module to reduce the dimension of the data. At the same time, redundant information can be removed to improve the quality of the feature vector, thereby obtaining an accurate one-dimensional feature vector.

[0196] S22. Input the one-dimensional feature vector into the identification decision module for fault identification, generate the fault type of the circuit breaker to be identified, and generate fault identification information.

[0197] In an embodiment of the present invention, the recognition decision module includes a global average pooling (GAP) layer and a softmax layer. The one-dimensional feature vector is input into the recognition decision module for fault identification. The global average pooling (GAP) layer performs an average operation on the one-dimensional feature vector to capture the global characteristics of the data. Then, each element in the one-dimensional feature vector after the average operation is converted into a probability value between 0 and 1 through a softmax function. Each element represents the probability of the corresponding fault type, so that the fault type of the circuit breaker to be identified can be generated, and the fault identification information can be generated.

[0198] See also Figure 12-13,To better demonstrate the results of the target fault identification model for high-voltage circuit breakers, the present invention selects a subset of sample points from the data test set of high-voltage circuit breaker fault simulation. Then, principal component analysis (PCA) is used for dimensionality reduction. The sample points from the input layer and the output layer are then classified and mapped onto a two-dimensional plane.

[0199] It can be observed that in the input layer, the locations of the sampling points are randomly scattered throughout the space. In contrast, in the output layer, the data points corresponding to the above five different scenarios are separated, showing an organized spatial distribution. This shows that the high-voltage circuit breaker fault identification model is able to classify the sampling points under different fault types, which verifies the effectiveness of the model.

[0200] In the present invention, in response to a fault identification request, vibration data of a circuit breaker to be identified is obtained and the vibration data is preprocessed, and the preprocessed vibration data is input into a constructed target fault identification model, the target fault identification model includes an information extraction layer and a decision layer, and the preprocessed vibration data is feature extracted through the information extraction layer to obtain a target feature vector, and the target feature vector is input into the decision layer for fault identification, and the fault identification information of the circuit breaker to be identified is output; the present invention utilizes the information extraction layer in the constructed target fault identification model to perform feature extraction on the vibration data of the circuit breaker to be identified, and can effectively extract a high-precision target feature vector from the vibration data, and then perform fault identification through the decision layer, thereby realizing accurate identification of high-voltage circuit breaker faults, and solving the technical problem of how to improve the accuracy of high-voltage circuit breaker fault identification.

[0201] See also Fig.14 , Fig.14 This is a structural block diagram of a high-voltage circuit breaker fault identification system provided in Embodiment 3 of the present invention.

[0202] The present invention provides a high-voltage circuit breaker fault identification system, comprising:

[0203] A response module 301 is used to respond to a fault identification request, obtain vibration data of a circuit breaker to be identified, and pre-process the vibration data;

[0204] A data input module 302 is used to input the pre-processed vibration data into a built target fault identification model, where the target fault identification model includes an information extraction layer and a decision layer;

[0205] The information extraction module 303 is used to extract features from the preprocessed vibration data through the information extraction layer to obtain a target feature vector;

[0206] The fault identification module 304 is used to use the target feature vector to input the decision layer to perform fault identification and output fault identification information of the circuit breaker to be identified.

[0207] Further, the information extraction layer includes an improved residual network module, a first layer normalization module, a feedforward network module, a second layer normalization module and a long short-term memory network module connected in sequence, and the information extraction module 303 includes:

[0208] The residual eigenvector submodule is used to extract features from the preprocessed vibration data by improving the residual network to obtain the residual eigenvector;

[0209] A first normalization vector submodule, used for performing a normalization operation on the residual feature vector through a first layer normalization module to obtain a first normalized vector;

[0210] An initial eigenvector submodule, used to construct an initial eigenvector using the first normalized vector and the preprocessed vibration data;

[0211] A feedforward vector module is used to extract features from the initial feature vector through a feedforward network module to obtain a feedforward vector;

[0212] A second normalization vector submodule, used for performing a normalization operation on the feedforward vector through a second layer normalization module to obtain a second normalized vector;

[0213] An intermediate eigenvector submodule, used for constructing an intermediate eigenvector using the second normalized vector and the initial eigenvector;

[0214] The target feature vector submodule is used to extract features from the intermediate feature vectors through the long short-term memory network module to obtain the target feature vector.

[0215] Furthermore, the improved residual network includes a downscaling module and an upscaling module connected in sequence, the downscaling module includes a plurality of downscaling submodules, the upscaling module includes a plurality of upscaling submodules, and the residual feature vector submodule includes:

[0216] A downscaling unit, used for performing a continuous downscaling operation on the preprocessed vibration data through a plurality of sequentially connected downscaling submodules to obtain a downscaled vector;

[0217] The upscaling unit is used to perform a continuous upscaling operation on the downscaled vector through a plurality of upscaling submodules connected in sequence to obtain a residual feature vector.

[0218] Furthermore, the downscaling submodule includes a residual unit and a strided convolution unit connected in sequence, the residual unit includes a plurality of residual blocks, and the downscaling unit includes:

[0219] A first eigenvector subunit, configured to perform a continuous residual operation on the preprocessed vibration data through a plurality of sequentially connected residual blocks to obtain a first eigenvector;

[0220] The downscaling vector subunit is used to perform a downsampling operation on the first eigenvector through a strided convolution unit to obtain a downscaled vector.

[0221] Furthermore, the upscaling submodule includes a residual unit and a deconvolution unit connected in sequence, the residual unit includes a plurality of residual blocks, and the upscaling unit includes:

[0222] A second eigenvector subunit, configured to perform a residual operation on the downscaled vector through a plurality of sequentially connected residual blocks to obtain a second eigenvector;

[0223] The up-sampling operation subunit is used to perform an up-sampling operation on the second feature vector through a deconvolution unit to obtain a residual feature vector.

[0224] Furthermore, the residual block includes a first convolution unit, a rectified linear unit, and a second convolution unit connected in sequence. The specific processing steps of the residual block include:

[0225] Performing a convolution operation on the input feature data through a first convolution unit to obtain first convolution data;

[0226] Performing a nonlinear transformation operation on the first convolution data through a rectified linear unit to obtain transformed data;

[0227] The conversion data is convolved by the second convolution unit to obtain output feature data.

[0228] Furthermore, the decision layer includes a one-dimensional convolution module and a recognition decision module connected in sequence, and the fault recognition module 304 includes:

[0229] A one-dimensional feature vector submodule is used to extract features of the target feature vector in sequence through a one-dimensional convolution module to obtain a one-dimensional feature vector;

[0230] The fault identification information submodule is used to input the one-dimensional feature vector into the identification decision module for fault identification, generate the fault type of the circuit breaker to be identified, and generate fault identification information.

[0231] Furthermore, it also includes:

[0232] A fault simulation data module is used to perform fault simulation on the high-voltage circuit breaker and obtain fault simulation data of the high-voltage circuit breaker in different fault states;

[0233] A training set module, used to construct a training set using all fault simulation data and reference simulation data of the high-voltage circuit breaker in a reference state;

[0234] A training feature vector module is used to use a training set to input a preset initial fault recognition model for training and output a training feature vector;

[0235] A target loss value module is used to calculate the target loss value between the training feature vector and the preset standard feature vector;

[0236] A comparison module, used for comparing the target loss value with a preset standard loss value;

[0237] The first output module is used to stop training and generate a target fault recognition model if the target loss value is less than a preset standard loss value.

[0238] Furthermore, it also includes:

[0239] A second output module, configured to adjust a model parameter of a preset initial fault identification model if the target loss value is greater than or equal to a preset standard loss value;

[0240] The jump module is used to jump to the step of using the training set to input the preset initial fault recognition model for training and outputting the training feature vector until the target loss value is less than the preset standard loss value.

[0241] In the present invention, in response to a fault identification request, vibration data of a circuit breaker to be identified is obtained and the vibration data is preprocessed, and the preprocessed vibration data is input into a constructed target fault identification model, the target fault identification model includes an information extraction layer and a decision layer, and the preprocessed vibration data is feature extracted through the information extraction layer to obtain a target feature vector, and the target feature vector is input into the decision layer for fault identification, and the fault identification information of the circuit breaker to be identified is output; the present invention utilizes the information extraction layer in the constructed target fault identification model to perform feature extraction on the vibration data of the circuit breaker to be identified, and can effectively extract a high-precision target feature vector from the vibration data, and then perform fault identification through the decision layer, thereby realizing accurate identification of high-voltage circuit breaker faults, and solving the technical problem of how to improve the accuracy of high-voltage circuit breaker fault identification.

[0242] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0243] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0244] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high voltage circuit breaker fault identification method, characterized in that: include: In response to a fault identification request, obtaining vibration data of a circuit breaker to be identified, and preprocessing the vibration data; Inputting the preprocessed vibration data into a built target fault identification model, wherein the target fault identification model includes an information extraction layer and a decision layer; Performing feature extraction on the preprocessed vibration data through the information extraction layer to obtain a target feature vector; The target feature vector is input into the decision layer for fault identification, and fault identification information of the circuit breaker to be identified is output.

2. The high voltage circuit breaker fault identification method according to claim 1, characterized in that: The information extraction layer includes an improved residual network module, a first layer normalization module, a feedforward network module, a second layer normalization module and a long short-term memory network module connected in sequence. The step of extracting features from the preprocessed vibration data through the information extraction layer to obtain a target feature vector includes: Performing feature extraction on the preprocessed vibration data through the improved residual network to obtain a residual feature vector; Performing a normalization operation on the residual feature vector through the first-layer normalization module to obtain a first normalized vector; constructing an initial feature vector using the first normalized vector and the preprocessed vibration data; Performing feature extraction on the initial feature vector through the feedforward network module to obtain a feedforward vector; Performing a normalization operation on the feedforward vector by the second-layer normalization module to obtain a second normalized vector; constructing an intermediate feature vector using the second normalized vector and the initial feature vector; The intermediate feature vector is subjected to feature extraction by the long short-term memory network module to obtain a target feature vector.

3. The high voltage circuit breaker fault identification method according to claim 2, characterized in that: The improved residual network includes a downscaling module and an upscaling module connected in sequence, the downscaling module includes a plurality of downscaling submodules, the upscaling module includes a plurality of upscaling submodules, and the step of extracting features from the preprocessed vibration data through the improved residual network to obtain a residual feature vector includes: Performing a continuous downscaling operation on the preprocessed vibration data through a plurality of sequentially connected downscaling submodules to obtain a downscaled vector; The downscaled vector is continuously upscaled by a plurality of upscaling submodules connected in sequence to obtain a residual feature vector.

4. The high-voltage circuit breaker fault identification method according to claim 3, characterized in that: The downscaling submodule includes a residual unit and a strided convolution unit connected in sequence, the residual unit includes a plurality of residual blocks, and the step of performing a continuous downscaling operation on the preprocessed vibration data through the plurality of the downscaling submodules connected in sequence to obtain a downscaled vector includes: Performing a continuous residual operation on the preprocessed vibration data through a plurality of sequentially connected residual blocks to obtain a first eigenvector; A downsampling operation is performed on the first feature vector by the strided convolution unit to obtain a downscaled vector.

5. The high voltage circuit breaker fault identification method according to claim 3, characterized in that: The upscaling submodule includes a residual unit and a deconvolution unit connected in sequence, the residual unit includes a plurality of residual blocks, and the step of performing a continuous upscaling operation on the downscaled vector through the plurality of the upscaling submodules connected in sequence to obtain a residual feature vector includes: Performing a residual operation on the downscaled vector through a plurality of sequentially connected residual blocks to obtain a second eigenvector; An upsampling operation is performed on the second feature vector by the deconvolution unit to obtain a residual feature vector.

6. The high voltage circuit breaker fault identification method according to claim 3 or 4, characterized in that: The residual block includes a first convolution unit, a rectified linear unit, and a second convolution unit connected in sequence. The specific processing steps of the residual block include: Performing a convolution operation on the input feature data by the first convolution unit to obtain first convolution data; Performing a nonlinear transformation operation on the first convolution data by the rectified linear unit to obtain conversion data; The second convolution unit performs a convolution operation on the conversion data to obtain output feature data.

7. The high voltage circuit breaker fault identification method according to claim 1, characterized in that: The decision layer includes a one-dimensional convolution module and an identification decision module connected in sequence. The step of using the target feature vector to input the decision layer for fault identification and outputting the fault identification information of the circuit breaker to be identified includes: Extracting features of the target feature vector in sequence through the one-dimensional convolution module to obtain a one-dimensional feature vector; The one-dimensional feature vector is input into the identification decision module for fault identification, the fault type of the circuit breaker to be identified is generated, and fault identification information is generated.

8. The high voltage circuit breaker fault identification method according to claim 1, characterized in that: Also includes: Performing fault simulation on the high-voltage circuit breaker to obtain fault simulation data of the high-voltage circuit breaker in different fault states; Constructing a training set using all of the fault simulation data and benchmark simulation data of the high-voltage circuit breaker in a benchmark state; Using the training set to input a preset initial fault recognition model for training, and outputting a training feature vector; Calculating a target loss value between the training feature vector and a preset standard feature vector; Comparing the target loss value with a preset standard loss value; If the target loss value is less than the preset standard loss value, the training is stopped and a target fault recognition model is generated.

9. The high voltage circuit breaker fault identification method according to claim 8, characterized in that: Also includes: If the target loss value is greater than or equal to the preset standard loss value, adjusting the model parameters of the preset initial fault identification model; Jump to the step of using the training set to input a preset initial fault recognition model for training and outputting a training feature vector until the target loss value is less than the preset standard loss value.

10. A high voltage circuit breaker fault identification system, characterized in that: include: A response module, used for responding to a fault identification request, obtaining vibration data of a circuit breaker to be identified, and preprocessing the vibration data; A data input module, used to input the pre-processed vibration data into a built target fault identification model, wherein the target fault identification model includes an information extraction layer and a decision layer; An information extraction module, used for performing feature extraction on the preprocessed vibration data through the information extraction layer to obtain a target feature vector; The fault identification module is used to use the target feature vector to input the decision layer to perform fault identification and output fault identification information of the circuit breaker to be identified.

Images