Fault data extraction method and device, storage medium and electronic device

Abstract

The invention discloses a fault data extraction method and device, a storage medium and an electronic device, and the method comprises the steps: obtaining current signals in a power system, carrying out the preprocessing of the current signals, and obtaining a target signal, and the current signals comprise a normal current signal and a fault current signal; performing short-time Fourier transform on the target signal, and extracting a first feature map; performing wavelet transform on the target signal, and extracting a second feature map; fusing the first feature map and the second feature map to obtain a third feature map; and inputting the third feature map into a deep convolutional network containing channel attention weighting processing for processing to obtain a plurality of fourth feature maps with different scales, and extracting current fault data in the power system based on the plurality of fourth feature maps with different scales. The problem that the series fault arc cannot be effectively diagnosed only by extracting a single feature is solved. Furthermore, arc fault features are extracted through a multi-feature fusion algorithm, and relatively high arc fault identification precision is realized.

Description

Technical Field

[0001] The present application relates to the technical field of power system fault detection, and in particular to a method and device for extracting fault data, a storage medium and an electronic device. Background Art

[0002] In recent years, with the continuous growth of energy demand, the application of renewable energy such as photovoltaics has become more and more extensive. In photovoltaic grid-connected systems, arc faults may cause the quality of photovoltaic power to deteriorate, or even cause equipment damage, seriously threatening the safety of the power grid and human life and property. Therefore, rapid identification of arc faults has important practical significance. At present, arc fault identification methods are mainly divided into time domain, frequency domain, time-frequency domain feature identification methods, and deep learning methods. However, the extraction methods of time domain features, frequency domain features, and time-frequency domain features are easily affected by noise, signal frequency, and signal amplitude when extracting features, resulting in inaccurate feature extraction results. The feature extraction accuracy of deep learning methods is generally not high, and the recognition accuracy is low. Therefore, there is an urgent need for an arc fault identification method that integrates time domain, frequency domain, and time-frequency domain features to improve the accuracy of arc fault identification.

[0003] With regard to the problem in related technologies that only extracting a single feature cannot effectively diagnose series fault arcs, no effective solution has been proposed so far.

[0004] Therefore, it is necessary to improve the related technology to overcome the above-mentioned defects in the related technology. Summary of the invention

[0005] The embodiments of the present application provide a method and device for extracting fault data, a storage medium, and an electronic device to at least solve the problem that only extracting a single feature cannot effectively diagnose a series fault arc.

[0006] According to one aspect of an embodiment of the present application, a method for extracting fault data is provided, including: acquiring a current signal in an electric power system, and preprocessing the current signal to obtain a target signal, wherein the current signal includes: a normal current signal and a fault current signal; performing a short-time Fourier transform on the target signal to extract a first feature map; and performing a wavelet transform on the target signal to extract a second feature map; fusing the first feature map and the second feature map to obtain a third feature map; inputting the third feature map into a deep convolutional network including a channel attention weighted processing for processing to obtain a plurality of fourth feature maps of different scales, and extracting current fault data in the electric power system based on the plurality of fourth feature maps of different scales.

[0007] In an exemplary embodiment, current fault data in an electric power system is extracted based on a plurality of fourth feature maps of different scales, including: sampling a plurality of fourth feature maps of different scales respectively to obtain a plurality of sub-feature maps of the same size; fusing a plurality of sub-feature maps of the same size to obtain a fifth feature map; extracting a reading area where signal fluctuations exist in the fifth feature map to obtain a target detection result; and extracting current fault data in the electric power system based on the target detection result.

[0008] In an exemplary embodiment, a reading area in the fifth characteristic map where signal fluctuations exist is extracted to obtain a target detection result, including: using the fifth characteristic map to perform predictive analysis to obtain a current fault data probability map and a distribution map of different fault thresholds corresponding to the fifth characteristic map; identifying sub-reading areas in the current fault data probability map and the distribution map; and summarizing all sub-reading areas to obtain a target detection result.

[0009] In an exemplary embodiment, current fault data in an electric power system is extracted based on a target detection result, including: intercepting a target signal segment from a current signal based on the target detection result; inputting the target signal segment into a feature extraction network to obtain a segment feature sequence; performing frame prediction on the segment feature sequence to obtain a reading recognition result; and determining the current fault data in the electric power system according to the reading recognition result.

[0010] In an exemplary embodiment, current fault data in an electric power system is determined based on a reading identification result, including: determining the size of multiple identification data corresponding to the reading identification result and a preset value; identifying current data corresponding to a first type of identification result in the reading identification result whose identification value is less than or equal to the preset value as normal operating data of the electric power system; and identifying current data corresponding to a second type of identification result in the reading identification result whose identification value is greater than the preset value as current fault data of the electric power system.

[0011] In an exemplary embodiment, before the third feature map is input into a deep convolutional network including channel attention weighted processing for processing, the method further includes: determining a backbone subnetwork of a current convolutional network in a current power system; introducing a preset attention mechanism into the backbone subnetwork to obtain a target backbone subnetwork; and replacing the backbone subnetwork in the current convolutional network with the target backbone subnetwork to obtain a deep convolutional network.

[0012] According to another aspect of an embodiment of the present application, a fault data extraction device is also provided, including: an acquisition module, used to acquire a current signal in an electric power system, and preprocess the current signal to obtain a target signal, wherein the current signal includes: a normal current signal and a fault current signal; an extraction module, used to perform a short-time Fourier transform on the target signal to extract a first feature map; and perform a wavelet transform on the target signal to extract a second feature map; a fusion module, used to fuse the first feature map and the second feature map to obtain a third feature map; a processing module, used to input the third feature map into a deep convolutional network including channel attention weighted processing for processing to obtain a plurality of fourth feature maps of different scales, and extract current fault data in the electric power system based on the plurality of fourth feature maps of different scales.

[0013] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned fault data extraction method when running.

[0014] According to another aspect of the embodiments of the present application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the fault data extraction method through the computer program.

[0015] According to another aspect of the embodiments of the present application, a computer program product is provided, including a computer program, and the above-mentioned method for extracting fault data when the computer program is executed by a processor.

[0016] Through this application, the current signal in the power system is obtained, and the current signal is preprocessed to obtain the target signal, wherein the current signal includes: a normal current signal and a fault current signal; the target signal is subjected to a short-time Fourier transform to extract the first feature map; and the target signal is subjected to a wavelet transform to extract the second feature map; the first feature map and the second feature map are fused to obtain the third feature map; the third feature map is input into a deep convolutional network including a channel attention weighted processing for processing to obtain a plurality of fourth feature maps of different scales, and the current fault data in the power system is extracted based on the fourth feature maps of multiple different scales. The above technical solution solves the problem that only extracting a single feature cannot effectively diagnose the series fault arc. Furthermore, by establishing a multi-feature fusion algorithm in the time-frequency domain, the arc fault features are effectively extracted to achieve a higher arc fault recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0019] Figure 1 It is a hardware structure block diagram of a computer terminal of a method for extracting fault data in an embodiment of the present application;

[0020] Figure 2 is a flow chart of a method for extracting fault data according to an embodiment of the present application;

[0021] Figure 3 is a flow chart of a photovoltaic series arc fault identification method based on a lightweight neural network according to an embodiment of the present application;

[0022] Figure 4 is a time domain characteristic diagram of normal current and fault current according to an embodiment of the present application;

[0023] Figure 5 It is a flowchart of the improvement of the lightweight DBNet neural network model according to an embodiment of the present application;

[0024] Figure 6 is a channel weighting flow chart of a backbone network feature graph according to an embodiment of the present application;

[0025] Figure 7 It is a structural block diagram of fault data extraction according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] 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.

[0027] 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 sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. 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.

[0028] The method embodiments provided in the embodiments of the present application can be executed in a computer terminal, a mobile terminal or a similar computing device. Taking running on a computer terminal as an example, Figure 1 FIG. 1 is a hardware structure block diagram of a computer terminal of a method for extracting fault data in an embodiment of the present application. Figure 1 As shown, the computer 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 (Central Processing Unit, CPU) or a programmable logic device (Field Programmable Gate Array, FPGA)) and a memory 104 for storing data, wherein the above-mentioned computer terminal may also include a transmission device 106 and an input and 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 above-mentioned computer terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0029] 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 extracting fault data in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. 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 may be connected to the computer 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 combinations thereof.

[0030] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of a computer 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 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0031] In this embodiment, a method for extracting fault data is provided. Figure 2 is a flow chart of a method for extracting fault data according to an embodiment of the present application. Figure 2 As shown, the process includes the following steps S202-S208:

[0032] Step S202: acquiring a current signal in the power system, and preprocessing the current signal to obtain a target signal, wherein the current signal includes: a normal current signal and a fault current signal;

[0033] Step S204: performing short-time Fourier transform on the target signal to extract a first feature map; and performing wavelet transform on the target signal to extract a second feature map;

[0034] Step S206: fusing the first feature map and the second feature map to obtain a third feature map;

[0035] Step S208: Input the third feature map into a deep convolutional network including channel attention weighted processing for processing to obtain a plurality of fourth feature maps of different scales, and extract current fault data in the power system based on the plurality of fourth feature maps of different scales.

[0036] The above steps obtain the current signal in the power system and pre-process the current signal to obtain the target signal, wherein the current signal includes: a normal current signal and a fault current signal; perform short-time Fourier transform on the target signal to extract the first feature map; and perform wavelet transform on the target signal to extract the second feature map; fuse the first feature map and the second feature map to obtain the third feature map; input the third feature map into a deep convolutional network containing channel attention weighted processing for processing to obtain a plurality of fourth feature maps of different scales, and extract the current fault data in the power system based on the fourth feature maps of multiple different scales. The above technical solution solves the problem that only extracting a single feature cannot effectively diagnose the series fault arc. Furthermore, by establishing a multi-feature fusion algorithm in the time-frequency domain, the arc fault features are effectively extracted to achieve a higher arc fault recognition accuracy.

[0037] In an exemplary embodiment, current fault data in an electric power system is extracted based on a plurality of fourth feature maps of different scales, including: sampling a plurality of fourth feature maps of different scales respectively to obtain a plurality of sub-feature maps of the same size; fusing a plurality of sub-feature maps of the same size to obtain a fifth feature map; extracting a reading area where signal fluctuations exist in the fifth feature map to obtain a target detection result; and extracting current fault data in the electric power system based on the target detection result.

[0038] It can be understood that in the above embodiment, the improved ShuffleNetV2 lightweight neural network is used to identify arc faults. After being processed by the ShuffleNetV2 network, the fourth feature maps {C1-C5} of different scales are obtained, and then the efficient channel attention (ECA) module is introduced to enhance the channel features of the feature maps {C2-C5} to obtain the feature maps {P2-P5}. Then, the feature maps {P2-P5} are sampled to the original Figure 1 / 4 of the sub-feature map size, and the sub-feature maps of the same size are spliced ​​together in the channel dimension to obtain the fused fifth feature map. The fused fifth feature map is processed to obtain the target detection result. Finally, based on the target detection result, the system accurately extracts the data fragments related to the arc fault from the original current signal. These data fragments contain specific information about the fault, such as the time of occurrence, duration and intensity, etc., which provide specific data basis for subsequent fault analysis and maintenance.

[0039] In an exemplary embodiment, a reading area in the fifth characteristic map where signal fluctuations exist is extracted to obtain a target detection result, including: using the fifth characteristic map to perform predictive analysis to obtain a current fault data probability map and a distribution map of different fault thresholds corresponding to the fifth characteristic map; identifying sub-reading areas in the current fault data probability map and the distribution map; and summarizing all sub-reading areas to obtain a target detection result.

[0040] It can be understood that in the above embodiment, the fifth feature map is predicted and analyzed, and the fifth feature map is input into the trained model for prediction to obtain the corresponding current fault data probability map and threshold map. The probability map represents the probability of each pixel corresponding to different categories, and the threshold map is used to determine the final binarization result. The DB module (DetectionBranch) is used to process the probability map and the threshold map to detect possible text areas. Finally, after the text area generation algorithm, the detection result of the reading area is obtained, that is, the sub-reading area in the current fault data probability map and the distribution map is identified, and all sub-reading areas are summarized to obtain the target detection result. Through the above steps, the reading area with signal fluctuations can be more accurately extracted, thereby achieving the purpose of target detection.

[0041] In an exemplary embodiment, current fault data in an electric power system is extracted based on a target detection result, including: intercepting a target signal segment from a current signal based on the target detection result; inputting the target signal segment into a feature extraction network to obtain a segment feature sequence; performing frame prediction on the segment feature sequence to obtain a reading recognition result; and determining the current fault data in the electric power system according to the reading recognition result.

[0042] Optionally, the improved ShuffleNetV2 combined with the DBNet model of the ECA module is used to detect the target of the current signal. The model outputs multiple reading areas suspected of arc faults, each of which is represented in the form of a time window, such as the 10th to 12th second, the 30th to 34th second, etc. Based on the target detection results, the reading areas suspected of arc faults in the current signal are cut into independent target signal segments, each of which is 2 seconds long and the sampling frequency remains unchanged at 100kHz. The signal segments processed in this way are easier to analyze and reduce the waste of computing resources. Each target signal segment is input into the feature extraction network, and the network outputs a segment feature sequence containing time series features. Subsequently, the segment feature sequence is predicted frame by frame, and each frame signal is evaluated whether it belongs to the arc fault state. The threshold is set to 0.9. If the probability of any frame exceeds the threshold, the frame is regarded as a fault frame. Finally, according to the frame prediction results, the number, location and duration of the fault frames in each target signal segment are analyzed to determine the severity of the arc fault. For example, in the target signal segment from 10 to 12 seconds, there are a large number of fault frames in the 1st second and 1.5 to 2 seconds, which indicates that the arc fault is more serious at these two time points and requires immediate attention. Through the above steps, not only can the arc fault be automatically detected, but also the characteristics of the fault can be quantified, providing strong technical support for real-time monitoring and fault prediction of power systems.

[0043] In an exemplary embodiment, current fault data in an electric power system is determined based on a reading identification result, including: determining the size of multiple identification data corresponding to the reading identification result and a preset value; identifying current data corresponding to a first type of identification result in the reading identification result whose identification value is less than or equal to the preset value as normal operating data of the electric power system; and identifying current data corresponding to a second type of identification result in the reading identification result whose identification value is greater than the preset value as current fault data of the electric power system.

[0044] Optionally, in an electric power system, current data is obtained and identified through reading recognition technology, and the identification result is compared with a preset value. If the identification value is less than or equal to the preset value, it is identified as normal operation data; if the identification value is greater than the preset value, it is identified as current fault data. For example, the reading recognition results are: 100A, 150A, 80A, 200A, 120A, and the preset value is 130A. According to the comparison, 150A and 200A are current fault data, corresponding to the second type of identification results respectively; 100A, 80A, and 120A are normal operation data, corresponding to the first type of identification results. In this way, the current fault data in the system can be clearly distinguished, and targeted processing and maintenance can be carried out.

[0045] In an exemplary embodiment, before the third feature map is input into a deep convolutional network including channel attention weighted processing for processing, the method further includes: determining a backbone subnetwork of a current convolutional network in a current power system; introducing a preset attention mechanism into the backbone subnetwork to obtain a target backbone subnetwork; and replacing the backbone subnetwork in the current convolutional network with the target backbone subnetwork to obtain a deep convolutional network.

[0046] Optionally, if a deep convolutional network is to be designed to process image data in the power system, first determine that the main subnetwork in the current power system is ShuffleNetV2, which is a lightweight network structure that can reduce the consumption of computing resources while maintaining performance. Then introduce a preset attention mechanism for the main subnetwork, and weight the feature map according to the importance of different channels in the power system data, so that the network pays more attention to important features and improves the representation ability of the network. Finally, the main subnetwork in the current convolutional network is replaced with the target main subnetwork with an attention mechanism, and the obtained deep convolutional network can better adapt to the characteristics of the power system data and improve the accuracy and efficiency of image processing. Through these steps, a deep convolutional network that is more suitable for processing power system data can be designed, which can achieve better results in practical applications.

[0047] Obviously, the embodiments described above are only some embodiments of the present application, not all embodiments. In order to better understand the above method, the above process is described below in conjunction with the embodiments, but it is not intended to limit the technical solutions of the embodiments of the present application, specifically:

[0048] The optional embodiment of the present application provides a photovoltaic series arc fault identification method based on a lightweight neural network to solve the problem that only extracting a single feature cannot effectively diagnose series fault arcs. This method fuses the time and frequency spectrum generated by wavelet transform and short-time Fourier transform, improves the lightweight ShuffleNetV2, replaces the original ResNet network with more parameters and larger calculation amount, and significantly reduces the model parameters and calculation amount while sacrificing a small amount of accuracy; then, an efficient channel attention module is introduced to perform channel feature weighting on the feature maps of different scales generated by the backbone network, and guides the network to focus on more important features according to the weights, and suppress unimportant features, thereby effectively improving the detection accuracy of the reading area.

[0049] Optional, Figure 3 This is a flow chart of a photovoltaic series arc fault identification method based on a lightweight neural network according to an embodiment of the present application; specifically comprising the following steps:

[0050] Step 1: Current signal acquisition, collecting the normal current and fault current of the signal.

[0051] Optionally, arc fault detection is generally analyzed based on the current waveform, and the methods are divided into time domain, frequency domain and time-frequency domain detection methods. Common time domain feature extraction parameters include: waveform peak, peak time, waveform duration, waveform RMS value, etc. Figure 4 As shown, Figure 4 It is a time domain characteristic diagram of normal current and fault current according to the embodiment of the present application. Common frequency domain characteristic parameters include spectrum peak, spectrum bandwidth, spectrum energy, etc. Common frequency domain analysis methods include Fourier transform and spectrum analysis method. Common time-frequency analysis methods include short-time Fourier transform and wavelet transform.

[0052] Step 2: Analysis and extraction of arc fault characteristics. Short-time Fourier transform and wavelet transform are used to generate time and frequency spectrum respectively. The details are as follows:

[0053] (1) Use short-time Fourier transform to generate time and frequency spectrum. Specifically, it includes the following steps:

[0054] Step 1: Convert the original signal into an audio signal in WAV format and judge the result by the audio.

[0055] Step 2: Perform short-time Fourier transform on the data to obtain the time-resolved spectrum signal of the original signal.

[0056] Step 3: Use spectral subtraction to reduce noise. Use the mouse to select the area of ​​the original signal's time-resolved spectrum signal amplitude part that needs to be retained, and use the matrix mask to implement spectral subtraction. The mask matrix selects the area with the mouse, the inside of the area is 1, and the outside is 0. Multiply the mask matrix and the corresponding position elements of the original signal's time-resolved spectrum signal amplitude matrix to obtain the amplitude matrix after noise reduction.

[0057] Step 4: Reconstruct the frequency domain distribution of the complex signal using the phase of the short-time Fourier transform and the amplitude after noise reduction.

[0058] Step 5: Use the inverse short-time Fourier transform to reconstruct the filtered time domain signal.

[0059] Step 6: Convert the filtered signal into an audio signal in WAV format and determine the effect.

[0060] (2) Use wavelet transform to generate time and frequency spectrum. Specifically, the following steps are included:

[0061] Step 1: Select a suitable wavelet basis function. Different wavelet basis functions have different effects on signal analysis. Select a suitable wavelet basis function according to the characteristics and requirements of the signal.

[0062] Step 2: Wavelet decomposition: The signal to be analyzed is decomposed using the fast wavelet transform algorithm to obtain wavelet coefficients at different scales and translations.

[0063] Step 3: Select the appropriate decomposition level. Select the appropriate decomposition level according to the characteristics and requirements of the signal. The choice of the decomposition level determines the accuracy and resolution of the time-frequency analysis. A shallower decomposition level can provide a rougher time-frequency analysis result, while a deeper decomposition level can provide more detailed time-frequency information.

[0064] Step 4: Reconstruct the signal. According to the wavelet coefficients, perform wavelet reconstruction through inverse wavelet transform to obtain the result of time-frequency analysis.

[0065] Step 5: Analyze time-frequency characteristics. By performing time-frequency analysis on the reconstructed signal, we can obtain the energy distribution of the signal at different times and frequencies.

[0066] Step 3: Feature fusion. The time and frequency spectrum generated by wavelet transform and the time and frequency spectrum generated by Fourier transform are fused. Specifically, the result obtained by Fourier transform is multiplied point by point with the result obtained by wavelet transform.

[0067] Step 4: Improve the convolutional network. Improve the lightweight DBNet neural network model. Figure 5 As shown, Figure 5 It is a flowchart of the improvement of the lightweight DBNet neural network model according to the embodiment of the present application, which specifically includes the following steps:

[0068] Step 1: Input the feature fusion result in step 3 into the backbone network ShuffleNetV2 to obtain the feature map {C1-C5}.

[0069] Optionally, ShuffleNetV2 is a lightweight neural network model for image classification and object detection tasks. It is an upgraded version of ShuffleNet, which further improves the performance and efficiency of the model by introducing a new network structure and design ideas. ShuffleNetV2 can improve the accuracy and generalization ability of the model to a certain extent while keeping the model lightweight, so it has broad application prospects in mobile devices and edge computing.

[0070] Step 2: Introduce the efficient channel attention ECA module to perform channel weighting on the feature map {C2-C5} output by the backbone network, and after fusion, obtain the new feature map {P2-P5}.

[0071] Step 3: Sample 1 / 4 of the original image, then splice it in the channel dimension, predict the spliced ​​feature maps separately, and obtain the probability map and threshold map.

[0072] Step 4: Process the probability map and threshold map in step 3 through the DB module and text region generation algorithm to obtain the detection result of the reading area.

[0073] Optionally, the DB module will preprocess the input image, including grayscale, binarization, filtering, etc., in order to extract the text area in the image. Among them, binarization is to represent the image by converting the pixel values ​​of the image into binary values ​​(usually 0 and 255). This technology is usually used to simplify the image, reduce the details and noise in the image, and make the image easier to process and analyze.

[0074] Optionally, use the improved lightweight DBNet as the convolutional network. There are two main improvements:

[0075] Improvement 1: Introduce ShuffleNetV2, a lightweight network with a more balanced model accuracy and complexity, as the backbone network ( Figure 5 C1 to C5 in the traditional DBNet network model). In the traditional DBNet network model, the original network uses the RestNet network. In comparison, ResNet has more parameters and a larger amount of calculation, while the ShuffleNetV2 model has a more balanced accuracy and complexity. As shown in Table 1, Table 1 is a table of recognition accuracy results of the ShuffleNetV2 model and the RestNet model as the DBNet backbone network according to the embodiment of the present application. It can be seen from Table 1 that when DBNet uses ResNet18 or ResNet50 as the backbone network, the accuracy of the model has reached a relatively high level, but the number of parameters and the amount of calculation are large, which is not conducive to the practical application of the model. Compared with the baseline model of ResNet, although the F1 value of the ShuffleNetV2 model has decreased, the number of parameters has been reduced to 15.4% of the original, and the amount of calculation has been reduced to 67.4% of the original, which significantly reduces the complexity of the model. After comprehensively weighing the accuracy and complexity of the reading detection model constructed with different backbone networks, it was decided to use ShuffleNetV2 as the backbone network.

[0076] Table 1

[0077] P R F1 Paras FLOPs ResNet18 0.9926 0.9803 0.9814 12.26 42.64 ResNet50 0.9840 0.9840 0.9840 24.78 62.07 ShuffleNetV2 0.9788 0.9721 0.9754 1.89 28.74

[0078] Optionally, the improved backbone network architecture is shown in Table 2, which is a table of the improved backbone network architecture of the lightweight DBNet neural network model according to an embodiment of the present application.

[0079] Table 2

[0080]

[0081] Optionally, the ShuffleNetV2 model has two operations, as follows:

[0082] Operation 1: Use deep convolution to replace ordinary convolution. Deep convolution can achieve similar accuracy to ordinary convolution with fewer parameters and calculations, so it is said here that "a small amount of accuracy is sacrificed while the number of model parameters and calculations are greatly reduced."

[0083] Operation 2: Channel shuffling is added, which divides the channels into several groups and then shuffles the channel information of different groups together, increasing the exchange of feature information between different groups, thereby enhancing the feature representation ability of the model. The channel shuffling operation can help the model increase the interaction and information flow between different channel features without increasing the amount of calculation. In the ShuffleNetV2 network, the channel shuffling operation is applied to the basic unit to promote feature interaction between different branches.

[0084] Improvement 2: In order to improve the network's ability to utilize feature maps at all levels, an efficient channel attention module is introduced into the feature pyramid network to weight the channel features of feature maps of different scales generated by the backbone network. According to the weights, the network is guided to focus on more important features and suppress unimportant features, thereby effectively improving the detection accuracy of the reading area. Figure 6 As shown, Figure 6 This is a flow chart of channel weighting of the backbone network feature graph according to an embodiment of the present application. Specifically, it includes the following steps:

[0085] Step 1: Input feature map, assuming the dimension is H×W×C.

[0086] Step 2: In the spatial dimension, use the global average pooling layer to compress the input feature map to obtain a 1×1×C feature map.

[0087] Step 3: For the compressed feature map, learn the feature information between different channels through one-dimensional convolution with a convolution kernel size of k. The kernel size k represents the local coverage of cross-channel interaction. An adaptive function is designed to determine the kernel size, and the function expression is: Where C is the channel dimension. γ and b are used to change the ratio between the number of channels C and the number of convolution kernels.

[0088] Step 4: Obtain channel attention information through the Sigmoid activation function.

[0089] Step 5: Multiply the channel attention feature map and the original input feature map channel by channel, and output the feature map weighted by the channel attention.

[0090] It should be noted that when performing wavelet transform and short-time Fourier transform, it is necessary to set the parameters reasonably to ensure that the generated time-frequency spectrum has good time-frequency resolution and spectral characteristics. When fusing the time-frequency spectrum, it is necessary to consider how to effectively combine the information of the two transforms to extract useful feature information to the greatest extent. When improving ShuffleNetV2, it is necessary to balance model accuracy and computational performance to ensure that the model accuracy is not excessively sacrificed while reducing the number of model parameters and the amount of computation. When introducing an efficient channel attention module to weight the channel features of the feature maps of different scales generated by the backbone network, ensure that the channel attention module can effectively calculate the weight of each channel so that the network can better focus on important features.

[0091] In summary, the optional embodiment of the present application analyzes the signal at different scales by wavelet transform and short-time Fourier transform. After fusion, the time-frequency characteristics of the signal can be captured more comprehensively, which improves the efficiency and accuracy of feature extraction. The fused time-frequency spectrum can better represent the characteristics of the signal, avoid repeated calculations and information redundancy, and reduce the computational complexity. The lightweight ShuffleNetV2 is improved to replace the original ResNet network with more parameters and larger calculations. The reduction of parameters and calculations can reduce the computing resources and memory usage required for the model during deployment, and improve the lightweight and deployment efficiency of the model. The introduction of the channel attention module can weight the channel features of feature maps at different scales, guide the network to focus on more important features according to the weights, and suppress unimportant features, thereby improving the importance of feature maps. Distinguishing ability. In addition, the channel attention module can help the network better learn the key features in the data, avoid paying too much attention to noise and unimportant features, and reduce the risk of overfitting of the model.

[0092] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0093] In this embodiment, a device for extracting fault data is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0094] Figure 7 : is a structural block diagram of a device for extracting fault data according to an embodiment of the present application, the device comprising:

[0095] An acquisition module 72 is used to acquire a current signal in the power system and preprocess the current signal to obtain a target signal, wherein the current signal includes: a normal current signal and a fault current signal;

[0096] An extraction module 74 is used to perform a short-time Fourier transform on the target signal to extract a first feature map; and perform a wavelet transform on the target signal to extract a second feature map;

[0097] A fusion module 76, configured to fuse the first feature map and the second feature map to obtain a third feature map;

[0098] The processing module 78 is used to input the third feature map into a deep convolutional network including channel attention weighted processing for processing to obtain a plurality of fourth feature maps of different scales, and extract the current fault data in the power system based on the plurality of fourth feature maps of different scales.

[0099] The above device obtains the current signal in the power system and preprocesses the current signal to obtain the target signal, wherein the current signal includes: a normal current signal and a fault current signal; performs a short-time Fourier transform on the target signal to extract the first feature map; and performs a wavelet transform on the target signal to extract the second feature map; fuses the first feature map and the second feature map to obtain the third feature map; inputs the third feature map into a deep convolutional network including channel attention weighted processing for processing to obtain a plurality of fourth feature maps of different scales, and extracts the current fault data in the power system based on the fourth feature maps of multiple different scales. The above technical solution solves the problem that only extracting a single feature cannot effectively diagnose the series fault arc. Furthermore, by establishing a multi-feature fusion algorithm in the time-frequency domain, the arc fault features are effectively extracted to achieve a higher arc fault recognition accuracy.

[0100] In an exemplary embodiment, the above-mentioned processing module is also used to sample multiple fourth feature maps of different scales respectively to obtain multiple sub-feature maps of the same size; fuse multiple sub-feature maps of the same size to obtain a fifth feature map; extract the reading area where there is signal fluctuation in the fifth feature map to obtain a target detection result; and extract current fault data in the power system based on the target detection result.

[0101] In an exemplary embodiment, the above-mentioned processing module uses the fifth characteristic map to perform predictive analysis to obtain a current fault data probability map corresponding to the fifth characteristic map and a distribution map of different fault thresholds; identifies sub-reading areas in the current fault data probability map and the distribution map; and summarizes all sub-reading areas to obtain target detection results.

[0102] In an exemplary embodiment, the above-mentioned processing module is also used to intercept a target signal segment from the current signal based on the target detection result; input the target signal segment into the feature extraction network to obtain a segment feature sequence; perform frame prediction on the segment feature sequence to obtain a reading recognition result; and determine the current fault data in the power system according to the reading recognition result.

[0103] In an exemplary embodiment, the above-mentioned processing module is also used to determine the size of multiple identification data corresponding to the reading identification result and the preset value; identify the current data corresponding to the first type of identification result in the reading identification result whose identification value is less than or equal to the preset value as the normal operating data of the power system; identify the current data corresponding to the second type of identification result in the reading identification result whose identification value is greater than the preset value as the current fault data of the power system.

[0104] In an exemplary embodiment, the above-mentioned device also includes: a determination module, which is used to determine the backbone subnetwork of the current convolutional network in the current power system before inputting the third feature map into the deep convolutional network including channel attention weighted processing for processing; introduce a preset attention mechanism into the backbone subnetwork to obtain a target backbone subnetwork; use the target backbone subnetwork to replace the backbone subnetwork in the current convolutional network to obtain a deep convolutional network.

[0105] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0106] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:

[0107] S1. Acquire a current signal in a power system, and preprocess the current signal to obtain a target signal, wherein the current signal includes: a normal current signal and a fault current signal;

[0108] S2, performing short-time Fourier transform on the target signal to extract a first feature map; and performing wavelet transform on the target signal to extract a second feature map;

[0109] S3, fusing the first feature map and the second feature map to obtain a third feature map;

[0110] S4. Input the third feature map into a deep convolutional network including channel attention weighted processing for processing to obtain a plurality of fourth feature maps of different scales, and extract current fault data in the power system based on the plurality of fourth feature maps of different scales.

[0111] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0112] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.

[0113] An embodiment of the present application further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps in any one of the above method embodiments are performed.

[0114] An embodiment of the present application further provides another computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0115] An embodiment of the present application further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0116] Optionally, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0117] S1. Acquire a current signal in a power system, and preprocess the current signal to obtain a target signal, wherein the current signal includes: a normal current signal and a fault current signal;

[0118] S2, performing short-time Fourier transform on the target signal to extract a first feature map; and performing wavelet transform on the target signal to extract a second feature map;

[0119] S3, fusing the first feature map and the second feature map to obtain a third feature map;

[0120] S4. Input the third feature map into a deep convolutional network including channel attention weighted processing for processing to obtain a plurality of fourth feature maps of different scales, and extract current fault data in the power system based on the plurality of fourth feature maps of different scales.

[0121] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0122] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.

[0123] Obviously, those skilled in the art should understand that the above modules or steps of the present application 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 from that herein, 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 application is not limited to any specific combination of hardware and software.

[0124] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for extracting fault data, characterized in that: include: Acquire a current signal in the power system, and preprocess the current signal to obtain a target signal, wherein the current signal includes: a normal current signal and a fault current signal; Performing a short-time Fourier transform on the target signal to extract a first characteristic graph; and performing a wavelet transform on the target signal to extract a second characteristic graph; fusing the first feature map and the second feature map to obtain a third feature map; The third feature map is input into a deep convolutional network including channel attention weighted processing for processing to obtain a plurality of fourth feature maps of different scales, and current fault data in the power system is extracted based on the plurality of fourth feature maps of different scales.

2. The method according to claim 1, characterized in that Extracting current fault data in the power system based on the fourth characteristic graphs of the plurality of different scales includes: Sampling the plurality of fourth feature maps of different scales respectively to obtain a plurality of sub-feature maps of the same size; Merging the multiple sub-feature maps of the same size to obtain a fifth feature map; Extracting a reading area in the fifth characteristic graph where signal fluctuation exists to obtain a target detection result; Current fault data in the power system is extracted based on the target detection result.

3. The method according to claim 2, characterized in that Extracting a reading area where signal fluctuation exists in the fifth characteristic graph to obtain a target detection result includes: Using the fifth characteristic graph to perform prediction analysis, obtaining a current fault data probability graph corresponding to the fifth characteristic graph and distribution graphs of different fault thresholds; identifying sub-reading regions in the current fault data probability map and the distribution map; All sub-read regions are summed up to obtain the target detection results.

4. The method according to claim 2, characterized in that: Extracting current fault data in the power system based on the target detection result includes: intercepting a target signal segment from the current signal based on the target detection result; Inputting the target signal segment into a feature extraction network to obtain a segment feature sequence; Performing frame prediction on the segment feature sequence to obtain a reading recognition result; Current fault data in the power system is determined based on the reading identification result.

5. The method according to claim 4, characterized in that Determining current fault data in the power system according to the reading identification result includes: Determine the magnitude of a plurality of identification data corresponding to the reading identification result and a preset value; Identify the current data corresponding to the first type of identification results in the reading identification results where the identification value is less than or equal to the preset value as normal operating data of the power system; The current data corresponding to the second type of identification results whose identification values ​​in the reading identification results are greater than the preset value are identified as current fault data of the power system.

6. The method according to claim 1, characterized in that Before inputting the third feature map into a deep convolutional network including channel attention weighted processing for processing, the method further includes: Determine the backbone subnetwork of the current convolutional network in the current power system; Introducing a preset attention mechanism into the backbone sub-network to obtain a target backbone sub-network; The target backbone subnetwork is used to replace the backbone subnetwork in the current convolutional network to obtain a deep convolutional network.

7. A device for extracting fault data, characterized in that: include: An acquisition module is used to acquire a current signal in the power system and preprocess the current signal to obtain a target signal, wherein the current signal includes: a normal current signal and a fault current signal; an extraction module is used to perform a short-time Fourier transform on the target signal to extract a first characteristic graph; and performing wavelet transform on the target signal to extract a second feature map; A fusion module is used to fuse the first feature map and the second feature map to obtain a third feature map; a processing module is used to input the third feature map into a deep convolutional network including channel attention weighted processing for processing to obtain a plurality of fourth feature maps of different scales, and extract current fault data in the power system based on the plurality of fourth feature maps of different scales.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the method according to any one of claims 1 to 6 when executed.

9. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 6 through the computer program.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images