Method and device for identifying degradation state of lightning arrester, and computer program product

Abstract

The invention provides a lightning arrester degradation state identification method, an identification device and a computer program product. The method comprises the following steps: acquiring effective operation state data of a lightning arrester in real time; the effective operation state data are input into the lightning arrester degradation state recognition model, the current equipment state of the lightning arrester is obtained through output, the current equipment state comprises a current state, a voltage state and a lightning arrester state, the current state is normal current or abnormal current, the voltage state is normal voltage or abnormal voltage, and the lightning arrester state is the lightning arrester degradation state. The state of the lightning arrester is that the lightning arrester is degraded or normal; and under the condition that the lightning arrester state in the current equipment state is that the lightning arrester is degraded, prompting to execute lightning arrester maintenance measures. The problems that in the prior art, most researches are mainly based on simulation data, the set degradation working condition has large deviation from the actual condition, the lightning arrester degradation state recognition researched through an intelligent algorithm has large deviation from the actual working condition, and meanwhile the degradation state recognition precision is not high are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of arrester degradation identification, and in particular to a method for identifying the degradation state of an arrester, an identification device, a computer-readable storage medium and a computer program product. Background Art

[0002] During the operation of the arrester, the health of the arrester will change or even deteriorate due to various factors such as lightning strikes, short circuits, service life, and environmental changes. The deterioration state of the arrester will directly affect its safe and stable operation. Therefore, real-time monitoring and early warning of the deterioration state of the arrester is a key technology to improve equipment operation efficiency, extend equipment life, and reduce equipment maintenance costs. At present, the monitoring of the deterioration state of the arrester mostly relies on manual experience, and judgment is made based on the external operating conditions and internal state parameters of the arrester. It is not only labor-intensive, but also has low accuracy, and can no longer meet the needs of arrester status monitoring.

[0003] In recent years, with the development of deep learning technology, the arrester degradation state monitoring technology based on deep learning has been widely studied. In the arrester degradation state monitoring, the signal acquisition device collects the operating status data of the arrester through sensors, and then uses deep learning methods to extract and classify the collected data, so as to realize the recognition of the arrester degradation state. At present, the arrester degradation state monitoring based on deep learning mostly adopts the Convolutional Neural Network (CNN) model. For example, the power equipment degradation state monitoring method based on CNN can be divided into two categories: single-scale and multi-scale. The single-scale method uses a single-layer convolution kernel to perform multi-layer feature extraction on the signal, which can extract high-precision features of the signal, but lacks multi-scale feature extraction of the signal, which is prone to information loss. The multi-scale method uses a multi-scale convolution kernel to extract multi-scale features of the signal, which can extract multi-scale features of the signal, but this method cannot extract high-precision features of the signal, which is prone to inaccurate feature extraction.

[0004] It can be seen that the existing methods for monitoring the degradation state of lightning arresters based on deep learning have low recognition accuracy and recognition speed, and cannot cope with the characteristics of high signal sampling rate, multiple sampling points, complex features, and large amount of information in the degradation state monitoring of lightning arresters, which requires further research. At present, the recognition of the degradation state of lightning arresters studied by intelligent algorithms has a large deviation from the actual working conditions, and there is also the problem of low recognition accuracy of the degradation state. Most of the research is mainly based on simulation data, and the degradation conditions set have a large deviation from the actual ones, while there are few studies on recognition models based on measured data. The cost of data acquisition is high and the difficulty is great. Summary of the invention

[0005] The main purpose of the present application is to provide a method for identifying the degradation state of a lightning arrester, an identification device, a computer-readable storage medium and a computer program product, so as to at least solve the problem that most of the research in the prior art is mainly based on simulation data, the degradation conditions set are greatly deviated from the actual conditions, and the degradation state identification of the lightning arrester studied by intelligent algorithms is greatly deviated from the actual conditions, and there is also a problem that the accuracy of degradation state identification is not high.

[0006] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for identifying the degradation state of a lightning arrester is provided, comprising: acquiring effective operating state data of the lightning arrester in real time, wherein the effective operating state data at least includes effective voltage signals and current signals; inputting the effective operating state data into a lightning arrester degradation state identification model, so as to analyze and identify the lightning arrester degradation state identification model, and outputting the current device state of the lightning arrester, wherein the lightning arrester degradation state identification model comprises a data conversion layer, a feature extraction layer and an output layer, wherein the data conversion layer is used to convert the data signal into a graph; The image, the feature extraction layer is used to extract and classify features of the image, the output layer is used to output the equipment status of the arrester, the current equipment status includes current status, voltage status and arrester status, the current status is normal current or abnormal current, the voltage status is normal voltage or abnormal voltage, the arrester status is arrester deterioration or arrester normal; when the arrester status in the current equipment status is arrester deterioration, a prompt is given to perform arrester maintenance measures, and the arrester maintenance measures are at least replacement of the arrester or repair of the arrester.

[0007] Optionally, obtaining effective operating status data of a lightning arrester includes: obtaining the operating status data of the lightning arrester; and performing denoising processing on the operating status data using an overcomplete wavelet dictionary method to obtain the effective operating status data.

[0008] Optionally, before the effective operating status data is input into the arrester degradation state recognition model, the arrester degradation state recognition model is used for analysis and recognition, and the current device state of the arrester is obtained by output, the method further includes: constructing an initial arrester degradation state recognition model, the initial arrester degradation state recognition model is a predetermined model to be trained; a first input step, inputting multiple sets of training data into the data conversion layer of the initial arrester degradation state recognition model, so as to use the data conversion layer to perform normalization processing and polar coordinate transformation processing on the sample operating status data in turn, and generate two The invention relates to a method for generating a two-dimensional training image, wherein each group of the training data includes sample operation status data and a sample device status corresponding to the sample operation status data; a second input step, inputting the two-dimensional training image to the feature extraction layer of the initial arrester degradation state recognition model, so as to use a densely connected convolutional network combined with a multi-scale channel attention mechanism in the feature extraction layer to perform feature extraction and classification processing to generate training output data; a third input step, transmitting the training output data to the output layer of the initial arrester degradation state recognition model, so as to use the output layer to convert the training output into the arrester degradation state recognition model. The device state obtained by the device in the current training process is used to obtain the training device state; an analysis step is performed according to the training device state. Bandpass filtering is performed and slice analysis is performed in a sliding window manner to obtain a classification result of the training device state, and the classification result is a correct classification or an incorrect classification; a repetition step is performed, the first input step, the second input step, the third input step and the analysis step are repeated at least once in sequence until the classification result of the training device state corresponding to each group of the training data is obtained; a calculation step is performed according to the classification results of the training device state corresponding to all the training data to obtain the model accuracy; when the model accuracy is less than the set accuracy requirement, all parameters in the initial arrester degradation state identification model are adjusted, and the first input step, the second input step, the third input step, the analysis step and the calculation step are repeated at least once in sequence until the model accuracy is greater than or equal to the set accuracy requirement; when the model accuracy is greater than or equal to the set accuracy requirement, the trained initial arrester degradation state identification model is determined as the arrester degradation state identification model.

[0009] Optionally, multiple sets of training data are input into the data conversion layer of the initial arrester degradation state recognition model, so as to use the data conversion layer to perform normalization processing and polar coordinate transformation processing on the sample operation state data in sequence to generate a two-dimensional training image, including: normalizing the sample operation state data according to a first formula to obtain normalized operation state data, wherein the first formula is represents the normalized operating status data, x i represents the i-th sample running status data, maxX represents the maximum value of the sample running status data, and minX represents the minimum value of the sample running status data; all the normalized running status data are converted according to the second formula to obtain the polar coordinates corresponding to each normalized running status data, and the second formula is: θ i is the angular cosine value of the polar coordinate corresponding to the i-th normalized operating status data, represents the set of all the normalized running status data, r i is the radius of the polar coordinate corresponding to the i-th normalized operating status data, t i Represents x i The corresponding timestamp, M represents a normalization factor; the normalized running state data is reconstructed using a Gram angle and field algorithm to obtain the two-dimensional training image.

[0010] Optionally, the feature extraction layer is constructed by introducing the multi-scale channel attention mechanism into the densely connected convolutional network, and the two-dimensional training image is input into the feature extraction layer of the initial arrester degradation state recognition model, so as to perform feature extraction and classification processing in the feature extraction layer using the densely connected convolutional network combined with the multi-scale channel attention mechanism to generate training output data, including: using a two-dimensional discrete wavelet algorithm to decompose and reconstruct the two-dimensional training image to obtain a reconstructed two-dimensional training image; segmenting the reconstructed two-dimensional training image to obtain a plurality of segmented and reconstructed images; performing feature extraction on the plurality of segmented and reconstructed images through normalization operations, linear rectification operations, and pooling convolution operations in the feature extraction layer to obtain corresponding feature vectors; classifying and regressing the feature vectors through the fully connected layer in the feature extraction layer to generate the training output data.

[0011] Optionally, bandpass filtering is performed according to the training device state and slice analysis is performed in a sliding window manner to obtain a classification result of the training device state, including: dividing the state data corresponding to the training device state into multiple segmented data in sequence; obtaining a label value corresponding to each segmented data, wherein the sum of the label values ​​corresponding to all the segmented data is equal to n; when the first segmented data is greater than the second segmented data, the classification result is a correct classification, the first segmented data is the first segmented data, and the second segmented data is the remaining segmented data except the first segmented data; when the first segmented data is less than the second segmented data, the classification result is an incorrect classification.

[0012] Optionally, using a two-dimensional discrete wavelet algorithm to decompose and reconstruct the two-dimensional training image to obtain a reconstructed two-dimensional training image includes: decomposing the two-dimensional training image according to a decomposition algorithm to obtain multiple image signal components, wherein the image signal components include a horizontal component, a vertical component and a high-frequency component, and the expression of the decomposition algorithm is: represents the approximation coefficient of the two-dimensional training image at the jth layer, represents the horizontal component of the two-dimensional training image at the jth layer, represents the vertical component of the two-dimensional training image at the jth layer, represents the high frequency component of the two-dimensional training image at the jth layer, Indicates that h l-2k 、h n-2m Both represent the coefficients of the wavelet low-pass filter used in the corresponding wavelet decomposition process, g l-2k , g n-2m Both represent the coefficients of the high-pass filter, k and m represent the position coordinates in the structure of the two-dimensional training image, k represents the row index, m represents the column index, 1 and n both represent the index filter coefficients in the wavelet transform; reconstructing the image signal components using a reconstruction algorithm to obtain the reconstructed two-dimensional training image, the expression of the reconstruction algorithm is: is the reconstructed data point, representing the approximate coefficient at the j+1th layer of the wavelet transform.

[0013] According to another aspect of the present application, a device for identifying the degradation state of a lightning arrester is provided, the device comprising: an acquisition unit, for acquiring effective operating state data of the lightning arrester in real time, the effective operating state data comprising at least effective voltage signals and current signals; an input-output unit, for inputting the effective operating state data into a lightning arrester degradation state identification model, so as to analyze and identify the lightning arrester degradation state identification model, and output the current device state of the lightning arrester, the lightning arrester degradation state identification model comprising a data conversion layer, a feature extraction layer and an output layer, the data conversion layer being used to convert the data signal into a feature extraction layer; The present invention relates to an image, wherein the feature extraction layer is used to extract features from and classify the image, and the output layer is used to output the device status of the arrester, wherein the current device status includes a current status, a voltage status and a lightning arrester status, wherein the current status is a normal current or an abnormal current, the voltage status is a normal voltage or an abnormal voltage, and the lightning arrester status is a degraded lightning arrester or a normal lightning arrester; and a prompting unit is used to prompt the execution of lightning arrester maintenance measures when the lightning arrester status in the current device status is that the lightning arrester is degraded, and the lightning arrester maintenance measures are at least replacement of the lightning arrester or repair of the lightning arrester.

[0014] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods described.

[0015] According to another aspect of the present application, a computer program product is provided, comprising computer instructions, wherein when the computer instructions are executed by a processor, any one of the methods described above is implemented.

[0016] Applying the technical solution of the present application, in the method for identifying the degradation state of the arrester, the arrester degradation state identification model designed by the present application is divided into a data conversion layer, a feature extraction layer and an output layer. The data conversion layer is used to transform the signal in the effective operation state data into an image, and then the feature extraction layer is used to complete the feature extraction and classification of the image data, and finally the equipment state of the arrester is output through the output layer. For the arrester in operation, signal collection is performed once every set time period, and the trained arrester degradation state identification model is input for identification to judge the equipment state of the arrester in operation. The present application solves the problem that most of the research in the prior art is mainly based on simulation data, the degradation condition set is greatly deviated from the actual condition, and the identification of the degradation state of the arrester studied by the intelligent algorithm is greatly deviated from the actual condition, and there is also a problem that the accuracy of degradation state identification is not high. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A hardware structure block diagram of a mobile terminal for performing a method for identifying a degradation state of a lightning arrester provided in an embodiment of the present application is shown;

[0018] Figure 2 A schematic flow chart of a method for identifying a degradation state of a lightning arrester provided according to an embodiment of the present application is shown;

[0019] Figure 3 A diagram showing a training process of a lightning arrester degradation state recognition model provided according to an embodiment of the present application is shown;

[0020] Figure 4 A lightning arrester voltage signal diagram provided according to an embodiment of the present application is shown;

[0021] Figure 5 A lightning arrester current signal diagram provided according to an embodiment of the present application is shown;

[0022] Figure 6 A GASF characteristic diagram of a voltage signal provided according to an embodiment of the present application is shown;

[0023] Figure 7 A GASF characteristic diagram of a current signal provided according to an embodiment of the present application is shown;

[0024] Figure 8 A schematic diagram of an encoding process of a Gram angle and field algorithm provided according to an embodiment of the present application is shown;

[0025] Fig. 9 A structural block diagram of a device for identifying a degradation state of a lightning arrester provided according to an embodiment of the present application is shown.

[0026] The above drawings include the following reference numerals:

[0027] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION

[0028] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0029] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] As introduced in the background technology, there are deep learning-based methods for monitoring the degradation state of lightning arresters in the prior art, which have low recognition accuracy and recognition speed, and are unable to cope with the characteristics of high signal sampling rate, multiple sampling points, complex features, and large amount of information in the degradation state monitoring of lightning arresters. In order to solve the problem that most of the research in the prior art is mainly based on simulation data, the degradation conditions set are greatly deviated from the actual conditions, and the degradation state identification of lightning arresters studied through intelligent algorithms is greatly deviated from the actual conditions, and there is a problem of low accuracy in degradation state identification, the embodiments of the present application provide a method for identifying the degradation state of a lightning arrester, an identification device, a computer-readable storage medium, and a computer program product.

[0032] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0033] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 1 is a hardware structure block diagram of a mobile terminal of a method for identifying a degradation state of a lightning arrester according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0034] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the method for identifying the degradation state of the lightning arrester in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The above-mentioned specific examples of the network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0035] In this embodiment, a method for identifying the degradation state of a lightning arrester running on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0036] Figure 2 FIG. 1 is a flow chart of a method for identifying a degradation state of a lightning arrester according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0037] Step S201, obtaining effective operation status data of the arrester in real time, wherein the effective operation status data at least includes effective voltage signals and current signals.

[0038] Specifically, it is necessary to monitor and record the operating status data of the arrester in real time through a signal acquisition device, and preprocess the operating status data to ensure effectiveness, so as to obtain the above-mentioned effective operating status data. The preprocessing operation may include steps such as noise removal, filtering and normalization to improve data quality and reduce errors in model analysis. The preprocessed data will be clearer, which will help the model to more accurately identify the state of the arrester. These operating status data mainly include the voltage signal and current signal of the arrester. Voltage and current are key parameters for evaluating the state of the arrester.

[0039] Step S202, input the above-mentioned effective operating status data into the arrester degradation state recognition model, so as to use the above-mentioned arrester degradation state recognition model for analysis and recognition, and output the current equipment state of the above-mentioned arrester. The above-mentioned arrester degradation state recognition model includes a data conversion layer, a feature extraction layer and an output layer. The above-mentioned data conversion layer is used to convert the data signal into an image, the above-mentioned feature extraction layer is used to extract and classify the features of the above-mentioned image, and the above-mentioned output layer is used to output the equipment state of the above-mentioned arrester. The above-mentioned current equipment state includes current state, voltage state and arrester state. The above-mentioned current state is normal current or abnormal current, the above-mentioned voltage state is normal voltage or abnormal voltage, and the above-mentioned arrester state is arrester degradation or arrester normal.

[0040] Specifically, the pre-processed voltage and current signal data (i.e., the above-mentioned effective operating state data) are input into the arrester degradation state recognition model. The model consists of three main parts: data conversion layer, feature extraction layer, and output layer. The main task of the data conversion layer is to convert the one-dimensional time series data (voltage and current signals) into two-dimensional image data. This step is implemented by GASF (Gram Angle Sum Field Algorithm), which encodes the time series data into a two-dimensional image containing time information, providing input data for subsequent feature extraction and classification. The feature extraction layer uses the MSCAM-DenseNet network model to extract and classify the converted image. MSCAM (multi-scale channel attention mechanism) can adaptively adjust the weights of each layer in the convolutional neural network, increase the weight of useful information, and weaken the weight of useless features. DenseNet (densely connected convolutional network) uses its special structure to connect each layer in the network to all previous layers, so that features can be effectively utilized, the number of parameters can be reduced, and performance can be improved. The output layer is responsible for outputting the current device status of the arrester, including current status, voltage status, and arrester status. The current state can be normal current or abnormal current, the voltage state can be normal voltage or abnormal voltage, and the arrester state can be arrester degraded or arrester normal. The model will output the current device state of the arrester based on the input data and extracted features. This state includes whether the current is normal, whether the voltage is normal, and whether the arrester is degraded.

[0041] Step S203, when the arrester state in the current device state is that the arrester is deteriorated, prompting to perform arrester maintenance measures, the arrester maintenance measures at least being replacement of the arrester or repair of the arrester.

[0042] Specifically, when the model identifies that the arrester is in a degraded state, the system will prompt the implementation of maintenance measures. These measures include at least replacing the arrester or repairing the arrester. This is a key link in ensuring the safe operation of the power system and can reduce power accidents caused by arrester failures. Through the above steps, the degraded state of the arrester can be identified in real time and maintained in a timely manner, thereby ensuring the stability and safety of the power system.

[0043] In this embodiment, first, effective operation status data of the arrester is acquired in real time, and the effective operation status data at least includes effective voltage signal and current signal; then, the effective operation status data is input into the arrester degradation status recognition model, so as to analyze and recognize by using the arrester degradation status recognition model, and the current device status of the arrester is obtained by output, and the arrester degradation status recognition model includes a data conversion layer, a feature extraction layer and an output layer, the data conversion layer is used to convert the data signal into an image, the feature extraction layer is used to extract and classify the image, and the output layer is used to output the device status of the arrester, the current device status includes current status, voltage status and arrester status, the current status is normal current or abnormal current, the voltage status is normal voltage or abnormal voltage, and the arrester status is arrester degradation or arrester normal; finally, when the arrester status is the arrester degradation in the current device status, a prompt is given to perform arrester maintenance measures, and the arrester maintenance measures are at least replacement of the arrester or repair of the arrester. The arrester degradation state recognition model designed in this application is divided into a data conversion layer, a feature extraction layer and an output layer. The data conversion layer is used to transform the signal in the effective operation state data into an image, and then the feature extraction layer is used to complete the feature extraction and classification of the image data, and finally the equipment state of the arrester is output through the output layer. For the arrester in operation, signal collection is performed once every set time period, and the trained arrester degradation state recognition model is input for recognition to judge the equipment state of the arrester in operation. This application solves the problem that most of the research in the prior art is mainly based on simulation data, and the degradation working conditions set are greatly deviated from the actual conditions, and the identification of the arrester degradation state through intelligent algorithm has a large deviation from the actual working conditions, and there is also a problem that the accuracy of degradation state recognition is not high.

[0044] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the method for identifying the degradation state of a lightning arrester of the present application will be described in detail below in conjunction with specific embodiments.

[0045] In order to improve the signal quality of the arrester operation data, in an optional implementation manner, the above step S201 includes:

[0046] Step S2011, obtaining the operating status data of the above-mentioned lightning arrester;

[0047] Step S2012: using an overcomplete wavelet dictionary method to perform denoising on the operating status data to obtain the valid operating status data.

[0048] In the above embodiment, when the arrester begins to deteriorate, some operating data of the arrester also begins to change, which is different from the operating data under normal conditions. In the process of identifying the degradation state of the arrester, it is necessary to collect the operating state data of the arrester, and use the collected voltage and current signals as the basic data for degradation identification. Process the arrester status information data, denoise the data with unclear signal characteristics, improve the signal quality, use an overcomplete wavelet dictionary to reconstruct and denoise the signal, and enhance the signal characteristics. The processed signal is expressed as follows: Y′=DX′, where Y′ represents the denoised current signal or voltage signal, D represents the overcomplete wavelet dictionary, and X′ represents the estimated tree coefficient vector.

[0049] In order to improve the generalization ability of the arrester degradation state recognition model, in an optional implementation manner, before the above step S202, the method further includes:

[0050] Step S301, constructing an initial arrester degradation state recognition model, wherein the initial arrester degradation state recognition model is a predetermined model to be trained;

[0051] Step S302, a first input step, inputting multiple sets of training data into the data conversion layer of the initial arrester degradation state recognition model, so as to use the data conversion layer to perform normalization processing and polar coordinate transformation processing on the sample operation state data in sequence to generate a two-dimensional training image, wherein each set of the training data includes sample operation state data and sample device states corresponding to the sample operation state data;

[0052] Step S303, a second input step, inputting the two-dimensional training image into the feature extraction layer of the initial arrester degradation state recognition model, so as to perform feature extraction and classification processing in the feature extraction layer by using a densely connected convolutional network combined with a multi-scale channel attention mechanism to generate training output data;

[0053] Step S304, a third input step, transmitting the training output data to the output layer of the initial arrester degradation state recognition model, so as to utilize the output layer to convert the training output into the device state of the arrester obtained in the current training process, and obtain the training device state;

[0054] Step S305, an analysis step, performing bandpass filtering according to the training device state and performing slice analysis in a sliding window manner to obtain a classification result of the training device state, the classification result being a correct classification or an incorrect classification;

[0055] Step S306, repeating the steps, sequentially repeating the first input step, the second input step, the third input step and the analysis step at least once, until a classification result of the training device state corresponding to each set of the training data is obtained;

[0056] Step S307, a calculation step, performing accuracy calculation according to the classification results of the training device states corresponding to all the training data to obtain the model accuracy;

[0057] Step S308, when the accuracy of the above model is less than the set accuracy requirement, adjust all parameters in the above initial arrester degradation state identification model, and repeat the above first input step, the above second input step, the above third input step, the above analysis step and the above calculation step at least once in sequence until the accuracy of the above model is greater than or equal to the above set accuracy requirement;

[0058] Step S309: when the accuracy of the model is greater than or equal to the set accuracy requirement, the trained initial arrester degradation state recognition model is determined as the arrester degradation state recognition model.

[0059] In the above embodiment, the classification problem is simplified according to the actual application requirements. The arrester states correspond to the normal state and the degraded state in use, thereby simplifying the machine learning problem into a classification and recognition problem of 6 different working conditions. Each working condition corresponds to a label value, as shown in Table 1.

[0060] Table 1 Definitions of 6 classification problems

[0061] Tag value Signal Status Arrester status 0 Normal voltage, normal current normal 1 Normal voltage, normal current Deterioration 2 Abnormal voltage, normal current normal 3 Abnormal voltage, normal current Deterioration 4 Normal voltage, abnormal current normal 5 Normal voltage, abnormal current Deterioration

[0062] Design an initial arrester degradation state recognition model, which includes a data conversion layer, a feature extraction layer and an output layer. The model parameters need to be adjusted through training data to improve its recognition accuracy. Multiple sets of training data are input into the data conversion layer of the initial arrester degradation state recognition model. The sample operation state data is normalized to make its numerical range uniform for model processing. The normalized data is transformed into polar coordinates to generate a two-dimensional training image. The generated two-dimensional training image is input into the feature extraction layer of the model. The densely connected convolutional network (DenseNet) combined with the multi-scale channel attention mechanism (MSCAM) is used for feature extraction and classification. Through the processing of the feature extraction layer, the training output data is generated. The MSCAM-DenseNet network sub-model uses the two-dimensional image generated by the GASF method as the input of the entire model, and then enters the feature extraction layer, passes through the convolution layer with a convolution kernel of 7×7, and then passes through the 3×3 maximum pooling layer to optimize the network parameters, and then the DenseNet connection. Add the MSCAM attention mechanism in each DenseNet for adaptive weight allocation to improve the model classification accuracy. The output layer is used to convert the training output data into a specific device state of the arrester, that is, any one of the six states in Table 1 above. The training device state is bandpass filtered to remove noise and irrelevant information. Slice analysis is performed in a sliding window manner to obtain the classification results of the training device state. The classification result is correct or incorrect. This is done until the classification result of the training device state corresponding to each set of training data is obtained. The overall classification accuracy calculation method of the defined model is: if a total of C signal data are collected, of which C are correctly classified by the model 0 C 1 C=C 0 +C 1 , then the model accuracy of the trained arrester degradation state recognition model classification is C 0 / C. Determine whether the model accuracy is less than the set accuracy requirement. If the model accuracy is insufficient, adjust all parameters in the model. Continue to repeat the above steps until the model accuracy is greater than or equal to the set accuracy requirement. When the model accuracy is greater than or equal to the set accuracy requirement, the trained model is determined as the arrester degradation state identification model. The training process is shown in the figure below. Figure 3 As shown in the figure, through continuous iterative training and optimization of model parameters, a high-accuracy arrester degradation state identification model is finally obtained.

[0063] It should be noted that the attention mechanism can adaptively adjust the weights of each layer in the convolutional neural network, increase the weight of useful information, and weaken the weight of useless features. On the one hand, it reduces the amount of calculation parameters, and on the other hand, it improves the recognition accuracy of the model. The present invention proposes an improved attention mechanism: Multi-scale Channel Attention Mechanism (MSCAM). The MSCAM attention mechanism simultaneously pays attention to feature information through two channels, local features and global features, and then fuses the features. The resulting features contain feature information of different scales, so that the extracted features are more comprehensive, thereby improving the recognition accuracy of the model. The multi-scale channel attention mechanism first processes the input data (such as image or sequence data) and extracts features through filters of different scales (such as convolution kernels). This can be achieved by using convolution kernels of different sizes to obtain visual details at different levels, or by applying sliding window techniques of different window sizes in sequence data. The extracted multi-scale features will enter a specially designed network module, which will evaluate the importance of each feature channel and scale. Through the learned weights, the network can determine which features are more important and should be emphasized, and which features are less important and can be suppressed. The weights are automatically learned through backpropagation during training, and usually depend on the loss function of the specific task. The attention weight reflects the contribution of each feature in performing a specific task. In this way, the model can focus on those features that have the greatest impact on the predicted output. The weighted feature map is processed again by the subsequent parts of the network to finally generate the output required for the task, such as classification labels, detection boxes, etc. Introducing MSCAM into the densely connected convolutional network enables the original DenseNet to have multi-scale feature recognition capabilities. For a densely connected convolutional network, each layer repeatedly implements the nonlinear transformation H t (·), where t represents the level index. t (·) The function includes normalization (BN), linear rectification (ReLU), pooling or convolution (Conv) and other processes. The output of the "tth layer" is represented as X t , the traditional feedforward neural network connects the output of the tth layer as input to the t+1th layer, resulting in the following layer transformation: X t =H t (X t-1 ), by combining the identity function to jump out of the nonlinear transformation: X t =H t (X t-1 )+X t-1 , therefore, the tth layer obtains the feature maps [X 0 , X 1 , …, X t-1 ], and takes it as input: X t =Ht ([X 0 , X 1 , …, X t-1 ]). Among them, [X 0 , X 1 , …, X t-1 ] represents the concatenation 0, 1, …, t-1 of feature maps produced in layer 0.

[0064] In order to retain the time information in the original signal data, in an optional implementation manner, the above step S302 includes:

[0065] Step S3021, normalize the sample running status data according to the first formula to obtain normalized running status data. The first formula is: represents the above normalized running status data, x i represents the i-th sample running status data, maxX represents the maximum value among the sample running status data, and minX represents the minimum value among the sample running status data;

[0066] Step S3022: All the normalized running status data are converted according to the second formula to obtain the polar coordinates corresponding to each of the normalized running status data. The second formula is: θ i is the angular cosine value of the polar coordinates corresponding to the i-th normalized operating status data, represents the set of all the above normalized running status data, r i is the radius of the polar coordinates corresponding to the i-th normalized operating status data, t i Represents x i The corresponding timestamp, M represents the normalization factor;

[0067] Step S3023: reconstruct the normalized running state data using the Gram angle and field algorithm to obtain the two-dimensional training image.

[0068] In the above embodiment, the Gramian Angular Field (GASF) algorithm is used to encode the collected signal data while retaining the timing information of the signal. Through signal conversion, the accuracy and performance of downstream tasks are improved. The principle of the GASF image encoding method is: using the Gramian matrix to calculate the linear correlation relationship of a set of vectors, encode the time series into a two-dimensional image and retain the time information contained in the time series. The collected lightning arrester signal is a typical one-dimensional time series. Usually in a rectangular coordinate system, the horizontal axis represents time, and the vertical axis represents the voltage or current signal corresponding to each time point, such as Figure 4 and Figure 5 As shown, Figure 4 is the arrester voltage signal diagram, Figure 5 The signal is converted into graphic data by GASF, such as Figure 6 and Figure 7 As shown, Figure 6 is the GASF characteristic diagram of the voltage signal, Figure 7 is the GASF characteristic diagram of the current signal. The implementation process is as follows Figure 8 As shown in step 1, assume that the sample running status data after preprocessing is X = {x 1 , x 2 , x 3 , ..., x n-1 , x n}, first normalize it using the first formula and scale it to [-1, 1]. Step 2, convert the scaled signal in step 1 into polar coordinates. Pay attention to two variables: the value of the signal and its corresponding timestamp. These two variables can be represented by the angle and radius of the polar coordinates, respectively. Encode on polar coordinates, encode the scaled signal value with the angle cosine, and the value is between [0, π]. Encode its corresponding timestamp into the radius, the formula is as follows: Divide the interval [0, 1] into several parts, then discard 0 to associate the remaining points with the time series data, and let each point fall into a unit circle in polar coordinates. As time increases, the collected signal sequence will also be continuously distorted within this unit circle at different angles and radii. Step three, the signal time series is transformed into polar coordinates in step two, so that it also contains time information, so the GASF algorithm can be used to reconstruct the signal time series. The expression is as follows: It is to calculate the cosine value between the angles of the midpoints of the polar coordinates to generate a matrix or image, that is, the above two-dimensional training image. The obtained GASF feature map (that is, the above two-dimensional training image) can be directly used as the input data of the feature extraction layer. It can be seen from the formula that as time increases, the encoding position moves from the upper left corner of the matrix to the lower right corner, so the time dimension is encoded into the geometric structure of the matrix through GASF.

[0069] In order to ensure effective classification and regression prediction of the device status of the arrester, in an optional implementation manner, the above step S303 includes:

[0070] Step S3031, using a two-dimensional discrete wavelet algorithm to decompose and reconstruct the two-dimensional training image to obtain a reconstructed two-dimensional training image;

[0071] Step S3032, segmenting the reconstructed two-dimensional training image to obtain a plurality of segmented and reconstructed images;

[0072] Step S3033, performing feature extraction on the plurality of segmentation and reconstruction images through the normalization operation, linear rectification operation, and pooling convolution operation in the feature extraction layer to obtain corresponding feature vectors;

[0073] Step S3034, classifying and regressing the feature vector through the fully connected layer in the feature extraction layer to generate the training output data.

[0074] In the above embodiment, a two-dimensional discrete wavelet algorithm is used to simultaneously perform a one-dimensional wavelet transform on the rows and columns of a two-dimensional training image using a one-dimensional biorthogonal filter group to obtain the above-mentioned reconstructed two-dimensional training image, which contains the main feature information of the original image. According to the characteristics and structure of the image, an appropriate image segmentation algorithm (such as threshold-based segmentation, region growing segmentation, cluster segmentation, etc.) is used to segment the image to obtain multiple segmented reconstructed images, each of which represents a part of the feature area in the image. The pixel values ​​of the segmented and reconstructed images are normalized to reduce the scale difference between different features and improve the generalization ability of the model. The normalized feature map, in which each pixel value is adjusted to the same scale. The normalized feature map is nonlinearly transformed by the ReLU activation function to introduce nonlinear features and enhance the expression ability of the model. The feature map after ReLU activation, in which negative values ​​are set to 0 and positive values ​​remain unchanged. The convolution kernel is used to perform a convolution operation on the feature map after ReLU activation to extract local features. Perform pooling operations on the convolutional feature map, such as maximum pooling or average pooling, to reduce the spatial dimension of the feature map and extract more abstract features. After pooling and convolution operations, the feature vectors contain the key information of the image. The feature vectors output by the feature extraction layer are input to the fully connected layer for further feature integration and classification decisions. In the fully connected layer, the feature vectors are linearly combined by training weights and biases, and then classified by activation functions (such as Softmax function), or regression prediction is performed by linear regression. Generate classification labels or regression output data, which represent the final prediction results of the model for the input image, enabling the model to extract key features from the original image and perform effective classification and regression predictions.

[0075] In order to improve the accuracy of the arrester degradation state recognition model, in an optional implementation manner, the above step S305 includes:

[0076] Step S3051, dividing the state data corresponding to the state of the training device into a plurality of segmented data in sequence;

[0077] Step S3052, obtaining a label value corresponding to each of the segmented data, wherein the sum of the label values ​​corresponding to all the segmented data is equal to n;

[0078] Step S3053, when the first segmented data is larger than the second segmented data, the classification result is correct, the first segmented data is the first segmented data, and the second segmented data is the remaining segmented data except the first segmented data;

[0079] Step S3054: in the case where the first segmented data is smaller than the second segmented data, the classification result is a classification error.

[0080] In the above embodiment, the collected data is analyzed and classified, 70% of the data is selected as the training set, 30% of the data is selected as the test set, and the test data is further analyzed according to different operating conditions. The training set data is bandpass filtered and sliced ​​in a sliding window manner. The sliced ​​data is divided into a training set and a validation set according to a ratio of 0.8 and 0.2 respectively, and the trained model is classified on the complete training set data. The method for determining whether the model classification is correct is defined as follows: If a piece of data with a true label value of 0 (normal voltage signal, normal current signal, non-degraded lightning arrester) is collected and divided into n segments by a sliding window, the model divides m of the n segments into n segments. 0 , m 1 , m 2 , m 3 , m 4 , m 5 The segment data are predicted to have label values ​​of 0, 1, 2, 3, 4, and 5, respectively, where m 0 +m 1 +m 2 +m 3 +m 4 +m 5 =n. If m 0 >m i , i=1,2,3,4,5, then the classification result of the vibration data is correct, otherwise the classification result is wrong.

[0081] In order to extract the useful information of the channel to the maximum extent, in an optional implementation manner, the above step S3031 includes:

[0082] Step S30311, decompose the two-dimensional training image according to the decomposition algorithm to obtain multiple image signal components, the image signal components include horizontal components, vertical components and high-frequency components, the expression of the decomposition algorithm is: represents the approximate coefficient of the above two-dimensional training image at the jth layer, represents the horizontal component of the above two-dimensional training image at the jth layer, represents the vertical component of the above two-dimensional training image at the jth layer, represents the high-frequency component of the above two-dimensional training image at the jth layer, Indicates that h l-2k 、h n-2m Both represent the coefficients of the wavelet low-pass filter used in the corresponding wavelet decomposition process, g l-2k , g n-2m Both represent the coefficients of the high-pass filter, k and m represent the position coordinates in the structure of the above two-dimensional training image, k represents the row index, m represents the column index, and 1 and n both represent the index filter coefficients in the wavelet transform;

[0083] Step S30312, reconstructing the above two-dimensional training image using a reconstruction algorithm according to the above image signal components, the expression of the above reconstruction algorithm is: is the reconstructed data point, representing the approximate coefficient at the j+1th layer of the wavelet transform.

[0084] In the above embodiment, in order to extract the useful information in each channel to the maximum extent, the idea of ​​2D-Discrete Wavelet Transform (2D-DWT) is adopted to fuse the beneficial information of different channels to improve the utilization rate of two-dimensional feature information. The 2D-DWT method uses multiple wavelet filters for decomposition and reconstruction. The decomposition algorithm formula is shown in the following formula. Among them, k and m represent the position coordinates in the two-dimensional data structure, k represents the row index, and m represents the column index. 1 and n are indices, which are usually used to index filter coefficients in wavelet transform.

[0085] hl- 2 k, hn- 2m Represents the coefficients of the wavelet low-pass filter used in the corresponding wavelet decomposition process. The coefficients are used to calculate the approximate coefficients g l-2k , g n-2m Represents the coefficients of the high-pass filter, used to calculate the detail coefficients The reconstruction algorithm formula is shown as follows: is the reconstructed data point, representing the approximate coefficients at the j+1th layer of the wavelet transform. Using two low pass filters h t-2k and h n-2m To process the approximate coefficients of the jth layer to obtain the approximate coefficients of the higher layer, this part retains the low-frequency information of the data. The low-pass filter and the high-pass filter are used together to process the horizontal detail coefficients of the j-th layer to extract and reconstruct the high-frequency information in the horizontal direction. is a high-pass filter and a low-pass filter, which are used to process the vertical detail coefficients of the j-th layer in order to extract high-frequency information in the vertical direction. Two high-pass filters are used to process the diagonal detail coefficients of the jth layer to obtain the detail information in the diagonal direction. From the reconstruction formula and the decomposition formula, it can be seen that the essence of the 2D-DWT method is actually to use a one-dimensional biorthogonal filter bank to perform a one-dimensional wavelet transform on the rows and columns of the two-dimensional signal at the same time.

[0086] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0087] The embodiment of the present application also provides a device for identifying the deterioration state of a lightning arrester. It should be noted that the device for identifying the deterioration state of a lightning arrester of the embodiment of the present application can be used to execute the identification method for the deterioration state of a lightning arrester provided by the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and those that have been described will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware is also possible and conceived.

[0088] The following is an introduction to the device for identifying the degradation state of a lightning arrester provided in an embodiment of the present application.

[0089] Fig. 9 is a structural block diagram of a device for identifying the degradation state of a lightning arrester according to an embodiment of the present application. Fig. 9 As shown, the device comprises:

[0090] The acquisition unit 10 is used to acquire effective operation status data of the arrester in real time, and the effective operation status data at least includes effective voltage signals and current signals.

[0091] Specifically, it is necessary to monitor and record the operating status data of the arrester in real time through a signal acquisition device, and preprocess the operating status data to ensure effectiveness, so as to obtain the above-mentioned effective operating status data. The preprocessing operation may include steps such as noise removal, filtering and normalization to improve data quality and reduce errors in model analysis. The preprocessed data will be clearer, which will help the model to more accurately identify the state of the arrester. These operating status data mainly include the voltage signal and current signal of the arrester. Voltage and current are key parameters for evaluating the state of the arrester.

[0092] The input-output unit 20 is used to input the above-mentioned effective operating status data into the arrester degradation state recognition model, so as to use the above-mentioned arrester degradation state recognition model for analysis and recognition, and output to obtain the current equipment state of the above-mentioned arrester. The above-mentioned arrester degradation state recognition model includes a data conversion layer, a feature extraction layer and an output layer. The above-mentioned data conversion layer is used to convert the data signal into an image, the above-mentioned feature extraction layer is used to extract and classify the features of the above-mentioned image, and the above-mentioned output layer is used to output the equipment state of the above-mentioned arrester. The above-mentioned current equipment state includes current state, voltage state and arrester state. The above-mentioned current state is normal current or abnormal current, the above-mentioned voltage state is normal voltage or abnormal voltage, and the above-mentioned arrester state is arrester degradation or arrester normal.

[0093] Specifically, the pre-processed voltage and current signal data (i.e., the above-mentioned effective operating state data) are input into the arrester degradation state recognition model. The model consists of three main parts: data conversion layer, feature extraction layer, and output layer. The main task of the data conversion layer is to convert the one-dimensional time series data (voltage and current signals) into two-dimensional image data. This step is implemented by GASF (Gram Angle Sum Field Algorithm), which encodes the time series data into a two-dimensional image containing time information, providing input data for subsequent feature extraction and classification. The feature extraction layer uses the MSCAM-DenseNet network model to extract and classify the converted image. MSCAM (multi-scale channel attention mechanism) can adaptively adjust the weights of each layer in the convolutional neural network, increase the weight of useful information, and weaken the weight of useless features. DenseNet (densely connected convolutional network) uses its special structure to connect each layer in the network to all previous layers, so that features can be effectively utilized, the number of parameters can be reduced, and performance can be improved. The output layer is responsible for outputting the current device status of the arrester, including current status, voltage status, and arrester status. The current state can be normal current or abnormal current, the voltage state can be normal voltage or abnormal voltage, and the arrester state can be arrester degraded or arrester normal. The model will output the current device state of the arrester based on the input data and extracted features. This state includes whether the current is normal, whether the voltage is normal, and whether the arrester is degraded.

[0094] The prompting unit 30 is used for prompting to perform a lightning arrester maintenance measure when the lightning arrester state in the current device state is that the lightning arrester is deteriorated. The lightning arrester maintenance measure is at least to replace the lightning arrester or repair the lightning arrester.

[0095] Specifically, when the model identifies that the arrester is in a degraded state, the system will prompt the implementation of maintenance measures. These measures include at least replacing the arrester or repairing the arrester. This is a key link in ensuring the safe operation of the power system and can reduce power accidents caused by arrester failures. Through the above steps, the degraded state of the arrester can be identified in real time and maintained in a timely manner, thereby ensuring the stability and safety of the power system.

[0096] In this embodiment, the acquisition unit is used to acquire the effective operation status data of the lightning arrester in real time, and the effective operation status data at least includes an effective voltage signal and a current signal; the input and output unit is used to input the effective operation status data into the lightning arrester degradation state recognition model, so as to analyze and identify the lightning arrester degradation state recognition model, and output the current device state of the lightning arrester. The lightning arrester degradation state recognition model includes a data conversion layer, a feature extraction layer and an output layer. The data conversion layer is used to convert the data signal into an image, and the feature extraction layer is used to extract and classify the image. The output layer is used to output the device state of the lightning arrester. The current device state includes a current state, a voltage state and a lightning arrester state. The current state is a normal current or an abnormal current, the voltage state is a normal voltage or an abnormal voltage, and the lightning arrester state is a lightning arrester degradation or a lightning arrester normal; the prompt unit is used to prompt the execution of the lightning arrester maintenance measures when the lightning arrester state is the above-mentioned lightning arrester degradation in the above-mentioned current device state. The above-mentioned lightning arrester maintenance measures are at least replacing the above-mentioned lightning arrester or repairing the above-mentioned lightning arrester. The arrester degradation state recognition model designed in this application is divided into a data conversion layer, a feature extraction layer and an output layer. The data conversion layer is used to transform the signal in the effective operation state data into an image, and then the feature extraction layer is used to complete the feature extraction and classification of the image data, and finally the equipment state of the arrester is output through the output layer. For the arrester in operation, signal collection is performed once every set time period, and the trained arrester degradation state recognition model is input for recognition to judge the equipment state of the arrester in operation. This application solves the problem that most of the research in the prior art is mainly based on simulation data, and the degradation working conditions set are greatly deviated from the actual conditions, and the identification of the arrester degradation state through intelligent algorithm has a large deviation from the actual working conditions, and there is also a problem that the accuracy of degradation state recognition is not high.

[0097] In order to improve the signal quality of the arrester operation data, in an optional implementation manner, the acquisition unit includes:

[0098] A first acquisition module is used to acquire the operating status data of the above-mentioned lightning arrester;

[0099] The first preprocessing module uses an overcomplete wavelet dictionary method to perform denoising on the operating status data to obtain the effective operating status data.

[0100] In the above embodiment, when the arrester begins to deteriorate, some operating data of the arrester also begins to change, which is different from the operating data under normal conditions. In the process of identifying the degradation state of the arrester, it is necessary to collect the operating state data of the arrester, and use the collected voltage and current signals as the basic data for degradation identification. Process the arrester status information data, denoise the data with unclear signal characteristics, improve the signal quality, use an overcomplete wavelet dictionary to reconstruct and denoise the signal, and enhance the signal characteristics. The processed signal is expressed as follows: Y′=DX′, where Y′ represents the denoised current signal or voltage signal, D represents the overcomplete wavelet dictionary, and X′ represents the estimated tree coefficient vector.

[0101] In order to improve the generalization ability of the arrester degradation state recognition model, in an optional implementation manner, the device further includes:

[0102] A construction unit, used for constructing an initial arrester degradation state recognition model before inputting the above-mentioned effective operation state data into the arrester degradation state recognition model, analyzing and identifying using the above-mentioned arrester degradation state recognition model, and outputting the current device state of the above-mentioned arrester, wherein the above-mentioned initial arrester degradation state recognition model is a predetermined model to be trained;

[0103] A first input unit is used to execute a first input step, input multiple groups of training data into the data conversion layer of the initial arrester degradation state recognition model, so as to use the data conversion layer to perform normalization processing and polar coordinate transformation processing on the sample operation state data in sequence to generate a two-dimensional training image, wherein each group of the training data includes sample operation state data and sample device states corresponding to the sample operation state data;

[0104] A second input unit is used to perform a second input step, inputting the two-dimensional training image into the feature extraction layer of the initial arrester degradation state recognition model, so as to perform feature extraction and classification processing in the feature extraction layer by using a densely connected convolutional network combined with a multi-scale channel attention mechanism to generate training output data;

[0105] A third input unit is used to perform a third input step, transmitting the training output data to the output layer of the initial arrester degradation state recognition model, so as to use the output layer to convert the training output into the device state of the arrester obtained in the current training process, and obtain the training device state;

[0106] An analysis unit, configured to execute the analysis step, perform bandpass filtering according to the training device state and perform slice analysis in a sliding window manner to obtain a classification result of the training device state, wherein the classification result is a correct classification or an incorrect classification;

[0107] A first repeating unit is used to execute a repeating step, and sequentially repeat the first input step, the second input step, the third input step and the analyzing step at least once, until a classification result of the training device state corresponding to each group of the training data is obtained;

[0108] A calculation unit, used to execute the calculation step, perform accuracy calculation according to the classification results of the training device states corresponding to all the training data, and obtain the model accuracy;

[0109] A second repeating unit is used to adjust all parameters in the initial arrester degradation state identification model when the accuracy of the model is less than the set accuracy requirement, and to repeat the first input step, the second input step, the third input step, the analysis step and the calculation step at least once in sequence until the accuracy of the model is greater than or equal to the set accuracy requirement;

[0110] The determination unit is used to determine the trained initial arrester degradation state identification model as the arrester degradation state identification model when the model accuracy is greater than or equal to the set accuracy requirement.

[0111] In the above embodiment, the classification problem is simplified according to the actual application requirements. The arrester states correspond to the normal state and the degraded state in use, thereby simplifying the machine learning problem into a classification and recognition problem of 6 different working conditions, as shown in Table 1.

[0112] Table 26 Classification Problem Definitions

[0113]

[0114]

[0115] Design an initial arrester degradation state recognition model, which includes a data conversion layer, a feature extraction layer and an output layer. The model parameters need to be adjusted through training data to improve its recognition accuracy. Multiple sets of training data are input into the data conversion layer of the initial arrester degradation state recognition model. The sample operation state data is normalized to make its numerical range uniform for model processing. The normalized data is transformed into polar coordinates to generate a two-dimensional training image. The generated two-dimensional training image is input into the feature extraction layer of the model. The densely connected convolutional network (DenseNet) combined with the multi-scale channel attention mechanism (MSCAM) is used for feature extraction and classification. Through the processing of the feature extraction layer, the training output data is generated. The MSCAM-DenseNet network sub-model uses the two-dimensional image generated by the GASF method as the input of the entire model, and then enters the feature extraction layer, passes through the convolution layer with a convolution kernel of 7×7, and then passes through the 3×3 maximum pooling layer to optimize the network parameters, and then the DenseNet connection. Add the MSCAM attention mechanism in each DenseNet for adaptive weight allocation to improve the model classification accuracy. The output layer is used to convert the training output data into a specific device state of the arrester, that is, any one of the six states in Table 1 above. The training device state is bandpass filtered to remove noise and irrelevant information. Slice analysis is performed in a sliding window manner to obtain the classification results of the training device state. The classification result is correct or incorrect. This is done until the classification result of the training device state corresponding to each set of training data is obtained. The overall classification accuracy calculation method of the defined model is: if a total of C signal data are collected, of which C are correctly classified by the model 0 C 1 C=C 0 +C 1 , then the model accuracy of the trained arrester degradation state recognition model classification is C 0 / C. Determine whether the model accuracy is less than the set accuracy requirement. If the model accuracy is insufficient, adjust all parameters in the model. Continue to repeat the above steps until the model accuracy is greater than or equal to the set accuracy requirement. When the model accuracy is greater than or equal to the set accuracy requirement, the trained model is determined as the arrester degradation state identification model. The training process is shown in the figure below. Figure 3 As shown in the figure, through continuous iterative training and optimization of model parameters, a high-accuracy arrester degradation state identification model is finally obtained.

[0116] It should be noted that the attention mechanism can adaptively adjust the weights of each layer in the convolutional neural network, increase the weight of useful information, and weaken the weight of useless features. On the one hand, it reduces the amount of calculation parameters, and on the other hand, it improves the recognition accuracy of the model. The present invention proposes an improved attention mechanism: Multi-scale Channel Attention Mechanism (MSCAM). The MSCAM attention mechanism simultaneously pays attention to feature information through two channels, local features and global features, and then fuses the features. The resulting features contain feature information of different scales, so that the extracted features are more comprehensive, thereby improving the recognition accuracy of the model. The multi-scale channel attention mechanism first processes the input data (such as image or sequence data) and extracts features through filters of different scales (such as convolution kernels). This can be achieved by using convolution kernels of different sizes to obtain visual details at different levels, or by applying sliding window techniques of different window sizes in sequence data. The extracted multi-scale features will enter a specially designed network module, which will evaluate the importance of each feature channel and scale. Through the learned weights, the network can determine which features are more important and should be emphasized, and which features are less important and can be suppressed. The weights are automatically learned through backpropagation during training, and usually depend on the loss function of the specific task. The attention weight reflects the contribution of each feature in performing a specific task. In this way, the model can focus on those features that have the greatest impact on the predicted output. The weighted feature map is processed again by the subsequent parts of the network to finally generate the output required for the task, such as classification labels, detection boxes, etc. Introducing MSCAM into the densely connected convolutional network enables the original DenseNet to have multi-scale feature recognition capabilities. For a densely connected convolutional network, each layer repeatedly implements the nonlinear transformation H t (·), where t represents the level index. t (·) The function includes normalization (BN), linear rectification (ReLU), pooling or convolution (Conv) and other processes. The output of the "tth layer" is represented as X t , the traditional feedforward neural network connects the output of the tth layer as input to the t+1th layer, resulting in the following layer transformation: X t =H t (X t-1 ), by combining the identity function to jump out of the nonlinear transformation: X t =H t (X t-1 )+X t-1 , therefore, the tth layer obtains the feature maps [X 0 , X 1 , …, X t-1 ], and takes it as input: X t =Ht ([X 0 , X 1 , …, X t-1 ]). Among them, [X 0 , X 1 , …, X t-1 ] represents the concatenation 0, 1, …, t-1 of feature maps produced in layer 0.

[0117] In order to retain the time information in the original signal data, in an optional implementation manner, the first input unit includes:

[0118] The second preprocessing module normalizes the sample operation status data according to the first formula to obtain normalized operation status data. The first formula is: represents the above normalized running status data, x i represents the i-th sample running status data, maxX represents the maximum value among the sample running status data, and minX represents the minimum value among the sample running status data;

[0119] The data conversion module converts all the above normalized operating status data according to the second formula to obtain the polar coordinates corresponding to each of the above normalized operating status data. The above second formula is: θ i is the angular cosine value of the polar coordinates corresponding to the i-th normalized operating status data, represents the set of all the above normalized running status data, r i is the radius of the polar coordinates corresponding to the i-th normalized operating status data, t i Represents x i The corresponding timestamp, M represents the normalization factor;

[0120] The reconstruction module reconstructs the normalized running state data using the Gram angle and field algorithm to obtain the two-dimensional training image.

[0121] In the above embodiment, the Gramian Angular Field (GASF) algorithm is used to encode the collected signal data while retaining the timing information of the signal. Through signal conversion, the accuracy and performance of downstream tasks are improved. The principle of the GASF image encoding method is: using the Gramian matrix to calculate the linear correlation relationship of a set of vectors, encode the time series into a two-dimensional image and retain the time information contained in the time series. The collected lightning arrester signal is a typical one-dimensional time series. Usually in a rectangular coordinate system, the horizontal axis represents time, and the vertical axis represents the voltage or current signal corresponding to each time point, such as Figure 4 and Figure 5 As shown, Figure 4 is the arrester voltage signal diagram, Figure 5 The signal is converted into graphic data by GASF, such as Figure 6 and Figure 7 As shown, Figure 6 is the GASF characteristic diagram of the voltage signal, Figure 7 is the GASF characteristic diagram of the current signal. The implementation process is as follows Figure 8 As shown in step 1, assume that the sample running status data after preprocessing is X = {x 1 , x 2 , x 3 , ..., x n-1 , x n}, first normalize it using the first formula and scale it to [-1, 1]. Step 2, convert the scaled signal in step 1 into polar coordinates. Pay attention to two variables: the value of the signal and its corresponding timestamp. These two variables can be represented by the angle and radius of the polar coordinates, respectively. Encode on polar coordinates, encode the scaled signal value with the angle cosine, and the value is between [0, π]. Encode its corresponding timestamp into the radius, the formula is as follows: Divide the interval [0, 1] into several parts, then discard 0 to associate the remaining points with the time series data, and let each point fall into a unit circle in polar coordinates. As time increases, the collected signal sequence will also be continuously distorted within this unit circle at different angles and radii. Step three, the signal time series is transformed into polar coordinates in step two, so that it also contains time information, so the GASF algorithm can be used to reconstruct the signal time series. The expression is as follows: It is to calculate the cosine value between the angles of the midpoints of the polar coordinates to generate a matrix or image, that is, the above two-dimensional training image. The obtained GASF feature map (that is, the above two-dimensional training image) can be directly used as the input data of the feature extraction layer. It can be seen from the formula that as time increases, the encoding position moves from the upper left corner of the matrix to the lower right corner, so the time dimension is encoded into the geometric structure of the matrix through GASF.

[0122] In order to ensure effective classification and regression prediction of the device status of the arrester, in an optional implementation manner, the second input unit includes:

[0123] A decomposition and reconstruction module uses a two-dimensional discrete wavelet algorithm to decompose and reconstruct the two-dimensional training image to obtain a reconstructed two-dimensional training image;

[0124] A segmentation module is used to segment the reconstructed two-dimensional training image to obtain a plurality of segmentation and reconstruction images;

[0125] A feature extraction module performs feature extraction on the plurality of segmentation and reconstruction images through normalization operation, linear rectification operation, and pooling convolution operation in the feature extraction layer to obtain corresponding feature vectors;

[0126] The classification and regression module classifies and regresses the feature vectors through the fully connected layer in the feature extraction layer to generate the training output data.

[0127] In the above embodiment, a two-dimensional discrete wavelet algorithm is used to simultaneously perform a one-dimensional wavelet transform on the rows and columns of a two-dimensional training image using a one-dimensional biorthogonal filter group to obtain the above-mentioned reconstructed two-dimensional training image, which contains the main feature information of the original image. According to the characteristics and structure of the image, an appropriate image segmentation algorithm (such as threshold-based segmentation, region growing segmentation, cluster segmentation, etc.) is used to segment the image to obtain multiple segmented reconstructed images, each of which represents a part of the feature area in the image. The pixel values ​​of the segmented and reconstructed images are normalized to reduce the scale difference between different features and improve the generalization ability of the model. The normalized feature map, in which each pixel value is adjusted to the same scale. The normalized feature map is nonlinearly transformed by the ReLU activation function to introduce nonlinear features and enhance the expression ability of the model. The feature map after ReLU activation, in which negative values ​​are set to 0 and positive values ​​remain unchanged. The convolution kernel is used to perform a convolution operation on the feature map after ReLU activation to extract local features. Perform pooling operations on the convolutional feature map, such as maximum pooling or average pooling, to reduce the spatial dimension of the feature map and extract more abstract features. After pooling and convolution operations, the feature vectors contain the key information of the image. The feature vectors output by the feature extraction layer are input to the fully connected layer for further feature integration and classification decisions. In the fully connected layer, the feature vectors are linearly combined by training weights and biases, and then classified by activation functions (such as Softmax function), or regression prediction is performed by linear regression. Generate classification labels or regression output data, which represent the final prediction results of the model for the input image, enabling the model to extract key features from the original image and perform effective classification and regression predictions.

[0128] In order to improve the accuracy of the arrester degradation state identification model, in an optional implementation manner, the above-mentioned analysis unit includes:

[0129] A division module divides the state data corresponding to the state of the training device into a plurality of segmented data in sequence;

[0130] A second acquisition module is configured to acquire a label value corresponding to each of the segmented data, wherein a sum of the label values ​​corresponding to all the segmented data is equal to n;

[0131] A first determination module, in the case where the first segmented data is larger than the second segmented data, the classification result is correct, the first segmented data is the first segmented data, and the second segmented data is the remaining segmented data except the first segmented data;

[0132] The second judgment module determines that, in the case where the first segmented data is smaller than the second segmented data, the classification result is a classification error.

[0133] In the above embodiment, the collected data is analyzed and classified, 70% of the data is selected as the training set, 30% of the data is selected as the test set, and the test data is further analyzed according to different operating conditions. The training set data is bandpass filtered and sliced ​​in a sliding window manner. The sliced ​​data is divided into a training set and a validation set according to a ratio of 0.8 and 0.2 respectively, and the trained model is classified on the complete training set data. The method for determining whether the model classification is correct is defined as follows: If a piece of data with a true label value of 0 (normal voltage signal, normal current signal, non-degraded lightning arrester) is collected and divided into n segments by a sliding window, the model divides m of the n segments into n segments. 0 , m 1 , m 2 , m 3 , m 4 , m 5 The segment data are predicted to have label values ​​of 0, 1, 2, 3, 4, and 5, respectively, where m 0 +m 1 +m 2 +m 3 +m 4 +m 5 =n. If m 0 >m i , i=1,2,3,4,5, then the classification result of the vibration data is correct, otherwise the classification result is wrong.

[0134] In order to extract the useful information of the channel to the maximum extent, in an optional implementation, the decomposition and reconstruction module includes:

[0135] The decomposition submodule decomposes the two-dimensional training image according to the decomposition algorithm to obtain multiple image signal components. The image signal components include horizontal components, vertical components and high-frequency components. The expression of the decomposition algorithm is: represents the approximate coefficient of the above two-dimensional training image at the jth layer, represents the horizontal component of the above two-dimensional training image at the jth layer, represents the vertical component of the above two-dimensional training image at the jth layer, represents the high-frequency component of the above two-dimensional training image at the jth layer, Indicates that h l-2k 、h n-2m Both represent the coefficients of the wavelet low-pass filter used in the corresponding wavelet decomposition process, g l-2k , g n-2m Both represent the coefficients of the high-pass filter, k and m represent the position coordinates in the structure of the above two-dimensional training image, k represents the row index, m represents the column index, and 1 and n both represent the index filter coefficients in the wavelet transform;

[0136] The reconstruction submodule reconstructs the above-mentioned two-dimensional training image using a reconstruction algorithm according to the above-mentioned multiple image signal components. The expression of the above-mentioned reconstruction algorithm is: is the reconstructed data point, representing the approximate coefficient of the j+1th layer of the wavelet transform

[0137] In the above embodiment, in order to extract the useful information in each channel to the maximum extent, the idea of ​​2D-Discrete Wavelet Transform (2D-DWT) is adopted to fuse the beneficial information of different channels to improve the utilization rate of two-dimensional feature information. The 2D-DWT method uses multiple wavelet filters for decomposition and reconstruction. The decomposition algorithm formula is shown as follows: Where k and m represent the position coordinates in the two-dimensional data structure, k represents the row index, and m represents the column index. 1 and n are indices, which are usually used to index filter coefficients in wavelet transform. l-2k 、h n-2m Represents the coefficients of the wavelet low-pass filter used in the corresponding wavelet decomposition process. The coefficients are used to calculate the approximate coefficients g l-2k , g n-2m Represents the coefficients of the high-pass filter, used to calculate the detail coefficients The reconstruction algorithm formula is shown as follows: is the reconstructed data point, representing the approximate coefficients at the j+1th layer of the wavelet transform. Using two low pass filters h t-2k and h n-2m To process the approximate coefficients of the jth layer to obtain the approximate coefficients of the higher layer, this part retains the low-frequency information of the data. The low-pass filter and the high-pass filter are used together to process the horizontal detail coefficients of the j-th layer to extract and reconstruct the high-frequency information in the horizontal direction. is a high-pass filter and a low-pass filter, which are used to process the vertical detail coefficients of the j-th layer in order to extract high-frequency information in the vertical direction. Two high-pass filters are used to process the diagonal detail coefficients of the jth layer to obtain the detail information in the diagonal direction. From the reconstruction formula and the decomposition formula, it can be seen that the essence of the 2D-DWT method is actually to use a one-dimensional biorthogonal filter bank to perform a one-dimensional wavelet transform on the rows and columns of the two-dimensional signal at the same time.

[0138] The above-mentioned device for identifying the degradation state of the lightning arrester includes a processor and a memory, and the above-mentioned units are stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize the corresponding functions. The above-mentioned modules are all located in the same processor; or, the above-mentioned modules are located in different processors in any combination.

[0139] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the kernel parameters can be adjusted to solve the problem that most of the research in the prior art is mainly based on simulation data, and the set degradation conditions have a large deviation from the actual conditions, and the degradation state identification of the arrester studied by the intelligent algorithm has a large deviation from the actual conditions, and the degradation state identification accuracy is not high.

[0140] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0141] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for identifying the degradation state of the lightning arrester.

[0142] An embodiment of the present invention provides a processor, and the processor is used to run a program, wherein the method for identifying the degradation state of the lightning arrester is executed when the program is run.

[0143] An embodiment of the present invention provides a lightning arrester management system, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, at least the following steps are implemented:

[0144] Step S201, obtaining effective operation status data of the arrester in real time, wherein the effective operation status data at least includes effective voltage signals and current signals;

[0145] Step S202, input the above-mentioned effective operation status data into the arrester degradation state recognition model, so as to use the above-mentioned arrester degradation state recognition model for analysis and recognition, and output to obtain the current device state of the above-mentioned arrester, the above-mentioned arrester degradation state recognition model includes a data conversion layer, a feature extraction layer and an output layer, the above-mentioned data conversion layer is used to convert the data signal into an image, the above-mentioned feature extraction layer is used to extract features and classify the above-mentioned image, and the above-mentioned output layer is used to output the device state of the above-mentioned arrester, the above-mentioned current device state includes current state, voltage state and arrester state, the above-mentioned current state is normal current or abnormal current, the above-mentioned voltage state is normal voltage or abnormal voltage, and the above-mentioned arrester state is arrester degradation or arrester normal;

[0146] Step S203, when the arrester state in the current device state is that the arrester is deteriorated, prompting to perform arrester maintenance measures, the arrester maintenance measures at least being replacement of the arrester or repair of the arrester.

[0147] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program for initializing at least the following method steps:

[0148] Step S201, obtaining effective operation status data of the arrester in real time, wherein the effective operation status data at least includes effective voltage signals and current signals;

[0149] Step S202, input the above-mentioned effective operation status data into the arrester degradation state recognition model, so as to use the above-mentioned arrester degradation state recognition model for analysis and recognition, and output to obtain the current device state of the above-mentioned arrester, the above-mentioned arrester degradation state recognition model includes a data conversion layer, a feature extraction layer and an output layer, the above-mentioned data conversion layer is used to convert the data signal into an image, the above-mentioned feature extraction layer is used to extract features and classify the above-mentioned image, and the above-mentioned output layer is used to output the device state of the above-mentioned arrester, the above-mentioned current device state includes current state, voltage state and arrester state, the above-mentioned current state is normal current or abnormal current, the above-mentioned voltage state is normal voltage or abnormal voltage, and the above-mentioned arrester state is arrester degradation or arrester normal;

[0150] Step S203, when the arrester state in the current device state is that the arrester is deteriorated, prompting to perform arrester maintenance measures, the arrester maintenance measures at least being replacement of the arrester or repair of the arrester.

[0151] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0152] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0153] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0154] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0155] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0156] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0157] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0158] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0159] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0160] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0161] 1) The method for identifying the deterioration state of a lightning arrester of the present application first obtains effective operating state data of the lightning arrester in real time, wherein the effective operating state data at least includes effective voltage signals and current signals; then, the effective operating state data is input into a lightning arrester deterioration state identification model, so as to analyze and identify the deterioration state of the lightning arrester using the lightning arrester deterioration state identification model, and output the current device state of the lightning arrester, wherein the lightning arrester deterioration state identification model includes a data conversion layer, a feature extraction layer and an output layer, wherein the data conversion layer is used to convert the data signal into an image, and the feature extraction layer is used to extract the current device state of the lightning arrester. It is used to extract features and classify the above-mentioned image, and the above-mentioned output layer is used to output the equipment status of the above-mentioned lightning arrester. The above-mentioned current equipment status includes current status, voltage status and lightning arrester status. The above-mentioned current status is normal current or abnormal current, the above-mentioned voltage status is normal voltage or abnormal voltage, and the above-mentioned lightning arrester status is that the lightning arrester is degraded or the lightning arrester is normal; finally, when the above-mentioned lightning arrester status is the above-mentioned lightning arrester deterioration in the above-mentioned current equipment status, it is prompted to perform lightning arrester maintenance measures, and the above-mentioned lightning arrester maintenance measures are at least to replace the above-mentioned lightning arrester or repair the above-mentioned lightning arrester. The lightning arrester degradation state recognition model designed in this application is divided into a data conversion layer, a feature extraction layer and an output layer. The data conversion layer is used to transform the signal in the effective operation status data into an image, and then the feature extraction layer is used to complete the feature extraction and classification of the image data, and finally the equipment status of the lightning arrester is output through the output layer. For the lightning arrester in operation, signal acquisition is performed every set time period, and the trained lightning arrester degradation state recognition model is input for identification to judge the equipment status of the lightning arrester in operation. The present application solves the problem that most of the research in the prior art is mainly based on simulation data, the degradation conditions set are greatly deviated from the actual conditions, and the degradation state identification of the lightning arrester studied through intelligent algorithms is greatly deviated from the actual conditions, and there is also a problem that the accuracy of degradation state identification is not high.

[0162] 2) The device for identifying the deterioration state of the arrester of the present application, the designed arrester deterioration state identification model is divided into a data conversion layer, a feature extraction layer and an output layer. The data conversion layer is used to transform the signal in the effective operation state data into an image, and then the feature extraction layer is used to complete the feature extraction and classification of the image data, and finally the equipment state of the arrester is output through the output layer. For the arrester in operation, signal collection is performed once every set time period, and the trained arrester deterioration state identification model is input for identification to judge the equipment state of the arrester in operation. The present application solves the problem that most of the research in the prior art is mainly based on simulation data, the set deterioration working condition has a large deviation from the actual condition, and the identification of the deterioration state of the arrester studied by the intelligent algorithm has a large deviation from the actual working condition, and there is also a problem that the accuracy of the deterioration state identification is not high.

[0163] 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.

Claims

1. A method for identifying the degradation state of a lightning arrester, characterized in that: include: Acquire effective operation status data of the arrester in real time, wherein the effective operation status data at least includes effective voltage signal and current signal; The effective operation status data is input into the arrester degradation state recognition model, so as to use the arrester degradation state recognition model for analysis and recognition, and the current device state of the arrester is obtained by output. The arrester degradation state recognition model includes a data conversion layer, a feature extraction layer and an output layer. The data conversion layer is used to convert the data signal into an image, the feature extraction layer is used to extract and classify the image, and the output layer is used to output the device state of the arrester. The current device state includes a current state, a voltage state and an arrester state. The current state is a normal current or an abnormal current, the voltage state is a normal voltage or an abnormal voltage, and the arrester state is a degraded arrester or a normal arrester. When the arrester state in the current device state is that the arrester is deteriorated, a prompt is provided to perform arrester maintenance measures, and the arrester maintenance measures are at least replacement of the arrester or maintenance of the arrester.

2. The method according to claim 1, characterized in that Obtain effective operating status data of the arrester, including: Obtaining operating status data of the arrester; An overcomplete wavelet dictionary method is used to perform denoising on the operating status data to obtain the effective operating status data.

3. The method according to claim 1, characterized in that Before inputting the effective operation status data into the arrester degradation state identification model, analyzing and identifying using the arrester degradation state identification model, and outputting the current device state of the arrester, the method further includes: Constructing an initial arrester degradation state recognition model, wherein the initial arrester degradation state recognition model is a predetermined model to be trained; The first input step is to input multiple sets of training data into the data conversion layer of the initial arrester degradation state recognition model, so as to use the data conversion layer to perform normalization processing and polar coordinate transformation processing on the sample operation state data in sequence to generate a two-dimensional training image, wherein each set of the training data includes sample operation state data and a sample device state corresponding to the sample operation state data; The second input step is to input the two-dimensional training image into the feature extraction layer of the initial arrester degradation state recognition model, so as to perform feature extraction and classification processing in the feature extraction layer by using a densely connected convolutional network combined with a multi-scale channel attention mechanism to generate training output data; The third input step is to transmit the training output data to the output layer of the initial arrester degradation state recognition model, so as to use the output layer to convert the training output into the device state of the arrester obtained in the current training process to obtain the training device state; An analysis step, performing bandpass filtering according to the training device state and performing slice analysis in a sliding window manner to obtain a classification result of the training device state, wherein the classification result is a correct classification or an incorrect classification; Repeating step, repeating the first input step, the second input step, the third input step and the analysis step at least once in sequence, until a classification result of the training device state corresponding to each group of the training data is obtained; a calculation step, performing accuracy calculation according to the classification results of the training device states corresponding to all the training data to obtain the model accuracy; In the case where the model accuracy is less than the set accuracy requirement, adjusting all parameters in the initial arrester degradation state identification model, and sequentially repeating the first input step, the second input step, the third input step, the analysis step and the calculation step at least once until the model accuracy is greater than or equal to the set accuracy requirement; In a case where the model accuracy is greater than or equal to the set accuracy requirement, the trained initial arrester degradation state identification model is determined as the arrester degradation state identification model.

4. The method according to claim 3, characterized in that Inputting multiple sets of training data into the data conversion layer of the initial arrester degradation state recognition model, and using the data conversion layer to perform normalization processing and polar coordinate transformation processing on the sample operation state data in sequence to generate a two-dimensional training image, including: The sample running state data is normalized according to a first formula to obtain normalized running state data, wherein the first formula is: represents the normalized operating status data, x i represents the i-th sample running status data, maxX represents the maximum value of the sample running status data, and minX represents the minimum value of the sample running status data; All the normalized running state data are transformed according to the second formula to obtain the polar coordinates corresponding to each normalized running state data. The second formula is: θ i is the angular cosine value of the polar coordinate corresponding to the i-th normalized operating status data, represents the set of all the normalized running status data, r i is the radius of the polar coordinate corresponding to the i-th normalized operating status data, t i Represents x i The corresponding timestamp, M represents the normalization factor; The normalized operating state data is reconstructed using a Gram angle and field algorithm to obtain the two-dimensional training image.

5. The method according to claim 3, characterized in that: The feature extraction layer is constructed by introducing the multi-scale channel attention mechanism into the densely connected convolutional network, and the two-dimensional training image is input into the feature extraction layer of the initial arrester degradation state recognition model, so as to use the densely connected convolutional network combined with the multi-scale channel attention mechanism in the feature extraction layer to perform feature extraction and classification processing to generate training output data, including: Decomposing and reconstructing the two-dimensional training image using a two-dimensional discrete wavelet algorithm to obtain a reconstructed two-dimensional training image; Segmenting the reconstructed two-dimensional training image to obtain a plurality of segmented and reconstructed images; Performing feature extraction on the plurality of segmentation and reconstruction images through normalization operation, linear rectification operation, and pooling convolution operation in the feature extraction layer to obtain corresponding feature vectors; The feature vector is classified and regressed through the fully connected layer in the feature extraction layer to generate the training output data.

6. The method according to claim 3, characterized in that: Bandpass filtering is performed according to the training device state and slice analysis is performed in a sliding window manner to obtain a classification result of the training device state, including: Sequentially divide the state data corresponding to the state of the training device into a plurality of segmented data; Obtaining a label value corresponding to each of the segmented data, wherein a sum of the label values ​​corresponding to all the segmented data is equal to n; In the case where the first segmented data is larger than the second segmented data, the classification result is correct classification, the first segmented data is the first segmented data, and the second segmented data is the remaining segmented data except the first segmented data; In the case where the first segmented data is smaller than the second segmented data, the classification result is a classification error.

7. The method according to claim 5, characterized in that Decomposing and reconstructing the two-dimensional training image using a two-dimensional discrete wavelet algorithm to obtain a reconstructed two-dimensional training image includes: The two-dimensional training image is decomposed according to a decomposition algorithm to obtain multiple image signal components, wherein the image signal components include a horizontal component, a vertical component and a high-frequency component. The expression of the decomposition algorithm is: represents the approximation coefficient of the two-dimensional training image at the jth layer, represents the horizontal component of the two-dimensional training image at the jth layer, represents the vertical component of the two-dimensional training image at the jth layer, represents the high frequency component of the two-dimensional training image at the jth layer, Indicates that h l-2k 、h n-2m Both represent the coefficients of the wavelet low-pass filter used in the corresponding wavelet decomposition process, g l-2k , g n-2m Both represent coefficients of high-pass filters, k and m represent position coordinates in the structure of the two-dimensional training image, k represents a row index, m represents a column index, and l and n both represent index filter coefficients in wavelet transform; Reconstruct the image signal components using a reconstruction algorithm to obtain the reconstructed two-dimensional training image. The expression of the reconstruction algorithm is: is the reconstructed data point, representing the approximate coefficient at the j+1th layer of the wavelet transform.

8. A device for identifying the degradation state of a lightning arrester, characterized in that: The device comprises: An acquisition unit, used for acquiring effective operation status data of the arrester in real time, wherein the effective operation status data at least includes effective voltage signals and current signals; An input-output unit, used for inputting the effective operation status data into a lightning arrester degradation status recognition model, so as to use the lightning arrester degradation status recognition model for analysis and recognition, and outputting the current device status of the lightning arrester, wherein the lightning arrester degradation status recognition model comprises a data conversion layer, a feature extraction layer and an output layer, wherein the data conversion layer is used for converting data signals into images, the feature extraction layer is used for extracting features and classifying the images, and the output layer is used for outputting the device status of the lightning arrester, wherein the current device status comprises a current status, a voltage status and a lightning arrester status, wherein the current status is a normal current or an abnormal current, the voltage status is a normal voltage or an abnormal voltage, and the lightning arrester status is a lightning arrester degradation or a lightning arrester normal; A prompting unit is used to prompt the execution of a lightning arrester maintenance measure when the lightning arrester state in the current device state is that the lightning arrester is deteriorated. The lightning arrester maintenance measure is at least to replace the lightning arrester or repair the lightning arrester.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images