Direct-current arc fault identification method and device, storage medium and electronic device

Abstract

The invention discloses a DC arc fault identification method, and the method comprises the steps: obtaining a collected current parameter in a current time period; decomposing the current parameter to obtain a first fault feature vector of the current parameter; performing image conversion based on the current parameter to obtain a second fault feature vector of the current parameter; extracting a one-dimensional fault feature from the first fault feature vector, performing feature extraction on the second fault feature vector from multiple dimensions, and superposing feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature; and fusing the one-dimensional fault feature and the two-dimensional fault feature to obtain a fused fault feature, and performing DC arc fault recognition according to the fused fault feature, thereby improving the accuracy of DC arc fault recognition.

Description

Technical Field

[0001] The present application relates to the field of photovoltaic technology, and more specifically, to a method and device for identifying a DC arc fault, a storage medium, and an electronic device. Background Art

[0002] With the development of the photovoltaic industry, arc faults are a key factor restricting the promotion and safe use of DC transmission and distribution technology. Arc faults often occur due to factors such as aging and damage of insulation in electrical lines or equipment, loose electrical connections, humid air, and sharp increases in voltage and current.

[0003] The current types of arc faults include grounding arc faults, parallel arc faults, series arc faults, etc. Although a large number of algorithms have been developed in recent years to detect arc faults in lines, such as Fourier transform analysis and spectrum analysis, these algorithms not only have large computational complexity, but also have the problem of low current fault identification accuracy.

[0004] Therefore, how to improve the accuracy of DC arc fault identification has become a problem that needs to be solved urgently. Summary of the invention

[0005] The embodiments of the present application provide a DC arc fault identification method and device, a storage medium, and an electronic device to at least solve the problem of low current fault identification accuracy in the related art.

[0006] According to an embodiment of the present application, a DC arc fault identification method is provided, including: obtaining current parameters collected within a current time period; decomposing the current parameters to obtain a first fault feature vector of the current parameters; performing image conversion based on the current parameters to obtain a second fault feature vector of the current parameters; extracting a one-dimensional fault feature from the first fault feature vector, and performing feature extraction on the second fault feature vector from multiple dimensions, superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature; fusing the one-dimensional fault feature and the two-dimensional fault feature to obtain a fused fault feature, and performing DC arc fault identification according to the fused fault feature.

[0007] In an exemplary embodiment, the decomposing the current parameters to obtain the first fault characteristic vector of the current parameters includes: decomposing the current parameters to obtain multiple modal function components corresponding to the current parameters; selecting a modal function component that meets a preset correlation condition from the multiple modal function components to determine it as a target modal function component; and performing a Hilbert transform on the target modal function component to obtain the first fault characteristic vector of the current parameters.

[0008] In an exemplary embodiment, performing image conversion based on the current parameters to obtain a second fault feature vector of the current parameters includes: encoding the current parameters using a Gram angle field to obtain an initial two-dimensional image of the current parameters; adjusting the initial two-dimensional image to obtain a target two-dimensional image; and performing vector conversion based on the target two-dimensional image to obtain a second fault feature vector of the current parameters.

[0009] In an exemplary embodiment, the method of extracting a one-dimensional fault feature from the first fault feature vector, performing feature extraction on the second fault feature vector from multiple dimensions, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature includes: using a pre-trained fault identification model, extracting a one-dimensional fault feature from the first fault feature vector through a one-dimensional channel pair of the fault identification model, performing feature extraction on the second fault feature vector from multiple dimensions through a two-dimensional channel of the fault identification model, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature.

[0010] In an exemplary embodiment, the plurality of dimensions includes a horizontal dimension and a vertical dimension;

[0011] The method comprises: extracting features from the second fault feature vector in multiple dimensions through the two-dimensional channel of the fault identification model, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature, including: transforming the second fault feature vector in the horizontal dimension through the two-dimensional channel of the fault identification model to obtain a query matrix, a key-value matrix and a value matrix in the horizontal dimension; respectively determining the position coding of the query matrix, the key-value matrix and the value matrix in the horizontal dimension, and determining the position offset of the query matrix in the horizontal dimension, and determining the feature parameters of the key-value matrix in the horizontal dimension, and obtaining the key-value matrix in the horizontal dimension according to the position coding, the position offset and the feature parameters of the key-value matrix in the horizontal dimension. horizontal fault features; transform the second fault feature vector in the vertical dimension through the two-dimensional channel of the fault identification model to obtain the query matrix, key-value matrix and value matrix in the vertical dimension; respectively determine the position coding of the query matrix, the key-value matrix and the value matrix in the vertical dimension, determine the position offset of the query matrix in the vertical dimension, determine the characteristic parameters of the key-value matrix in the vertical dimension, and obtain the vertical fault features in the vertical dimension according to the position coding, the position offset and the characteristic parameters of the vertical dimension; obtain the two-dimensional fault features of the horizontal dimension and the vertical fault features in the vertical dimension through the two-dimensional channel of the fault identification model. In an exemplary embodiment,

[0012] In an exemplary embodiment, the two-dimensional fault features in the horizontal dimension and the vertical fault features in the vertical dimension are obtained through the two-dimensional channel of the fault identification model, including: splicing the horizontal fault features in the horizontal dimension and the vertical fault features in the vertical dimension to obtain an initial two-dimensional fault feature; determining a projection matrix of the initial two-dimensional fault feature; and multiplying the initial two-dimensional fault feature and the projection matrix to obtain the two-dimensional fault feature.

[0013] In an exemplary embodiment, the step of training the fault identification model includes:

[0014] Acquire a training sample; the training sample includes historical current parameters;

[0015] Decomposing the historical current parameter to obtain a first sample fault feature vector of the historical current parameter;

[0016] Perform image conversion based on the historical current parameters to obtain a second sample fault feature vector of the historical current parameters; extract sample one-dimensional fault features from the first sample fault feature vector based on the neural network model to be trained, and perform feature extraction on the second sample arc fault feature vector from multiple dimensions, superimpose the sample feature extraction results of the multiple dimensions to obtain sample two-dimensional fault features; fuse the sample one-dimensional fault features and the sample two-dimensional fault features to obtain sample fusion fault features; perform DC arc fault identification based on the sample fusion fault features based on the neural network model to be trained to obtain sample fault prediction results; compare the sample fault prediction results with a pre-set sample fault label to determine the loss, continue to train the neural network model to be trained according to the damage, and stop when the iteration condition is met to obtain a trained fault identification model.

[0017] According to another embodiment of the embodiment of the present application, a DC arc fault identification device is also provided, including: an acquisition module, used to acquire the current parameters collected in the current time period; a decomposition module, used to decompose the current parameters to obtain a first fault feature vector of the current parameters; a conversion module, used to perform image conversion based on the current parameters to obtain a second fault feature vector of the current parameters; an extraction module, used to extract a one-dimensional fault feature from the first fault feature vector, and perform feature extraction on the second fault feature vector from multiple dimensions, and superimpose the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature; an identification module, used to fuse the one-dimensional fault feature and the two-dimensional fault feature to obtain a fused fault feature, and perform DC arc fault identification according to the fused fault feature.

[0018] 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 method when running.

[0019] 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 method through the computer program.

[0020] According to another aspect of the embodiments of the present application, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0021] In an embodiment of the present application, the current parameters collected in the current time period are obtained, and a first fault feature vector of the current parameters is obtained by decomposition processing, and a second fault feature vector of the current parameters is obtained by image conversion. On this basis, a one-dimensional fault feature is extracted from the first fault feature vector, and feature extraction is performed on the second fault feature vector from multiple dimensions, and the feature extraction results of multiple dimensions are superimposed to obtain a two-dimensional fault feature, and then a fused fault feature is obtained by fusing the one-dimensional fault feature and the two-dimensional fault feature. DC arc fault identification is performed by fusing the fault features, which can solve the limitation of the incompleteness of a single feature in the related art when identifying DC arc faults, and can effectively improve the accuracy of DC arc fault identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0023] Figure 1 It is a hardware structure block diagram of a computer terminal of a DC arc fault identification method according to an embodiment of the present application;

[0024] Figure 2 is a flow chart of a DC arc fault identification method according to an embodiment of the present application;

[0025] Figure 3 is another flow chart of a DC arc fault identification method according to an embodiment of the present application;

[0026] Figure 4 is another flow chart of a DC arc fault identification method according to an embodiment of the present application;

[0027] Figure 5It is a structural block diagram of a DC arc fault identification device according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] 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.

[0029] 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 interchanged 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 variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices; "plurality" means two or more.

[0030] The method embodiments provided in the embodiments of the present application can be executed in a computer terminal or similar computing device or cloud platform or independent physical server or software platform, wherein the above software platform is run through one or more servers. Taking running on a computer terminal as an example, Figure 1 is a hardware structure block diagram of a DC arc fault identification method according to 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) processor 102 and memory 104 for storing data. In an exemplary embodiment, the computer terminal may also include a transmission device 106 for communication functions and an input and output device 108, wherein the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA. A person skilled in the art can understand 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 Equivalent functions or comparisons shown Figure 1 A different configuration with more features is shown.

[0031] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as computer programs corresponding to the methods in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, that is, the above-mentioned methods are 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 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.

[0032] 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.

[0033] In this embodiment, a DC arc fault identification method is provided, which is applied to the above-mentioned computer device. Figure 2 : is a flow chart of a DC arc fault identification method according to an embodiment of the present application, the process includes the following steps:

[0034] Step S202, obtaining the current parameters collected in the current time period;

[0035] The current parameter is obtained by detecting the photovoltaic line, and specifically can be the current parameter detected within a period of time. The current parameter can include different types of currents, such as normal operating current, current in a resistive load arc fault state, current in a resistive-inductive load arc fault state, current in a resistive-capacitive load arc fault state, and current in a motor load arc fault state.

[0036] Step S204, decomposing the current parameter to obtain a first fault characteristic vector of the current parameter;

[0037] In an exemplary embodiment, the decomposing the current parameter to obtain the first fault characteristic vector of the current parameter includes: decomposing the current parameter to obtain a plurality of modal function components corresponding to the current parameter; selecting a modal function component that meets a preset correlation condition from the plurality of modal function components to determine it as a target modal function component; and performing a Hilbert transform on the target modal function component to obtain the first fault characteristic vector of the current parameter.

[0038] The correlation condition is a condition used to determine the degree of correlation between the modal function components and the original current parameters.

[0039] Specifically, the current parameters can be decomposed by the complete set empirical mode decomposition algorithm of adaptive noise to obtain multiple intrinsic mode function components, namely, IMF (Intrinsic Mode Function) components. Then, based on the correlation and variance contribution between each mode function component and the original signal, the IMF components with high contribution are selected for subsequent analysis and processing, wherein the IMF components with high contribution can be the components ranked first, namely, the target mode function components. It can be understood that for each current parameter, its corresponding multiple intrinsic mode function components can be determined. Then, the first fault feature vector is constructed by Hilbert transformation.

[0040] Step S206, performing image conversion based on the current parameter to obtain a second fault feature vector of the current parameter;

[0041] In an exemplary embodiment, performing image conversion based on the current parameters to obtain a second fault feature vector of the current parameters includes: encoding the current parameters using a Gram angle field to obtain an initial two-dimensional image of the current parameters; adjusting the initial two-dimensional image to obtain a target two-dimensional image; and performing vector conversion based on the target two-dimensional image to obtain a second fault feature vector of the current parameters.

[0042] It should be noted that in the process of image conversion, the current parameters, that is, the current signals of the current time period, are encoded through the Gram angle field to convert the original one-dimensional current signal into an initial two-dimensional image. Among them, the generated initial two-dimensional images inevitably contain auxiliary information such as text, coordinate axes, and scale bars. In order to optimize these initial two-dimensional images to meet the input specifications of the subsequent fault recognition model and reduce the time cost required to generate the initial two-dimensional images, the initial two-dimensional images are adjusted. These adjustments include adjusting the direction of the coordinate axis, completely removing non-essential information (such as text and scale bars) in the initial two-dimensional image, and performing efficient compression processing. After this series of processing, the output is a target two-dimensional image in a 64×64×3 format to meet specific analysis or recognition requirements.

[0043] In some embodiments, each current signal of the current time period specifically refers to a one-dimensional time series signal, and GAF ​​(Gramian Angular Field) encodes the one-dimensional time series signal and converts it into an initial two-dimensional image. The converted initial two-dimensional image can retain the complete information in the signal in the time dimension. GAF is a bijective mapping, so it will retain all the information of the original one-dimensional current signal. During the encoding process, the time corresponding to the original data point increases from the upper left corner to the lower right corner of the image as the position of the pixel point increases, retaining the time correlation. The value corresponding to the main diagonal is a special case, which is composed of the size and angle of the original signal after scaling, and can achieve approximate reconstruction of the time series.

[0044] Step S208, extracting a one-dimensional fault feature from the first fault feature vector, and performing feature extraction on the second fault feature vector from multiple dimensions, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature;

[0045] In an exemplary embodiment, the extracting of a one-dimensional fault feature from the first fault feature vector, performing feature extraction on the second fault feature vector from multiple dimensions, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature includes: using a pre-trained fault identification model, extracting a one-dimensional fault feature from the first fault feature vector through a one-dimensional channel pair of the fault identification model, performing feature extraction on the second fault feature vector from multiple dimensions through a two-dimensional channel of the fault identification model, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature.

[0046] Specifically, the feature extraction process is determined by a pre-trained fault recognition model. The fault recognition model is a pre-built dual-channel CNN (convolutional neural network model) model. The first fault feature vector is input into the corresponding one-dimensional channel in the fault model, and the second fault feature vector is input into the corresponding two-dimensional channel in the fault model, and then the obtained one-dimensional fault features and two-dimensional fault features are converged and fused.

[0047] In some embodiments, the one-dimensional channel uses a one-dimensional convolutional network feature extraction module to extract one-dimensional fault features, and its convolution kernel sizes are set to 5×1, 3×1 and 3×1 respectively to carry out feature extraction of one-dimensional fault features.

[0048] In other implementations, the two-dimensional channel uses Transformer (a deep learning model with self-attention mechanism), and the core module of Transformer is the self-attention mechanism.

[0049] In an exemplary embodiment, the plurality of dimensions include a horizontal dimension and a vertical dimension;

[0050] The method comprises: extracting features from the second fault feature vector in multiple dimensions through the two-dimensional channel of the fault identification model, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature, including: transforming the second fault feature vector in the horizontal dimension through the two-dimensional channel of the fault identification model to obtain a query matrix, a key-value matrix and a value matrix in the horizontal dimension; respectively determining the position coding of the query matrix, the key-value matrix and the value matrix in the horizontal dimension, and determining the position offset of the query matrix in the horizontal dimension, and determining the feature parameters of the key-value matrix in the horizontal dimension, and obtaining the key-value matrix in the horizontal dimension according to the position coding, the position offset and the feature parameters of the key-value matrix in the horizontal dimension. horizontal fault characteristics; transforming the second fault feature vector in the vertical dimension through the two-dimensional channel of the fault identification model to obtain a query matrix, a key-value matrix and a value matrix in the vertical dimension; respectively determining the position coding of the query matrix, the key-value matrix and the value matrix in the vertical dimension, as well as determining the position bias of the query matrix in the vertical dimension and determining the characteristic parameters of the key-value matrix in the vertical dimension, and obtaining the vertical fault characteristics in the vertical dimension according to the position coding, the position bias and the characteristic parameters of the vertical dimension; obtaining a two-dimensional fault characteristic from the horizontal fault characteristics in the horizontal dimension and the vertical fault characteristics in the vertical dimension through the two-dimensional channel of the fault identification model.

[0051] Specifically, through the two-dimensional channel of the fault recognition model, the two-dimensional self-attention mechanism is decomposed into two one-dimensional self-attentions in the horizontal and vertical directions, that is, feature extraction is carried out from the horizontal dimension and the vertical dimension respectively. During the extraction process, the position bias of the query matrix is ​​determined, the characteristic parameters of the key-value matrix are determined, the query matrix is ​​determined, and the fault characteristics are obtained.

[0052] In some embodiments, the self-attention mechanism expression is:

[0053]

[0054] Among them, y i is the extracted feature, w represents different dimensions, v is the location information, r v is the content information, r v 、r k 、r q Encode the position on V, K, and Q. And add the position offset q of Q T r q , and the characteristic information k of K T r k At the same time, the self-attention modules are superimposed and spliced, and their structure is as follows Figure 3 As shown, the learned weights of each self-attention module are different.

[0055] It can be understood that the two-dimensional channel of the fault identification model can include multiple self-attention mechanism modules. For each self-attention mechanism module, it is decomposed into two one-dimensional self-attentions along the horizontal direction and the vertical direction, and then the two-dimensional fault characteristics are determined.

[0056] In an exemplary embodiment, the two-dimensional fault features in the horizontal dimension and the vertical fault features in the vertical dimension are obtained through the two-dimensional channel of the fault identification model, including: splicing the horizontal fault features in the horizontal dimension and the vertical fault features in the vertical dimension to obtain an initial two-dimensional fault feature; determining a projection matrix of the initial two-dimensional fault feature; and multiplying the initial two-dimensional fault feature and the projection matrix to obtain the two-dimensional fault feature.

[0057] It should be noted that after obtaining the horizontal fault feature and the vertical fault feature in the vertical dimension, the horizontal fault feature and the vertical fault feature in the vertical dimension can be spliced, and the splicing result can be transformed with the projection matrix and then output to obtain the two-dimensional fault feature.

[0058] In an exemplary embodiment, the training step of the fault identification model includes: obtaining training samples; the training samples include historical current parameters; decomposing the historical current parameters to obtain a first sample fault feature vector of the historical current parameters; performing image conversion based on the historical current parameters to obtain a second sample fault feature vector of the historical current parameters; extracting sample one-dimensional fault features from the first sample fault feature vector based on the neural network model to be trained, and performing feature extraction on the second sample arc fault feature vector from multiple dimensions, superimposing the sample feature extraction results of the multiple dimensions to obtain sample two-dimensional fault features; fusing the sample one-dimensional fault features and the sample two-dimensional fault features to obtain sample fusion fault features; based on the neural network model to be trained, performing DC arc fault identification based on the sample fusion fault features to obtain sample fault prediction results; comparing the sample fault prediction results with pre-set sample fault labels to determine losses, and continuing to train the neural network model to be trained according to the damage until the iteration condition is met and stopping to obtain a trained fault identification model.

[0059] Step S210: fusing the one-dimensional fault feature and the two-dimensional fault feature to obtain a fused fault feature, and performing DC arc fault identification according to the fused fault feature.

[0060] In an embodiment of the present application, the current parameters collected in the current time period are obtained, and a first fault feature vector of the current parameters is obtained by decomposition processing, and a second fault feature vector of the current parameters is obtained by image conversion. On this basis, a one-dimensional fault feature is extracted from the first fault feature vector, and feature extraction is performed on the second fault feature vector from multiple dimensions, and the feature extraction results of multiple dimensions are superimposed to obtain a two-dimensional fault feature, and then a fused fault feature is obtained by fusing the one-dimensional fault feature and the two-dimensional fault feature. DC arc fault identification is performed by fusing the fault features, which can solve the limitation of the incompleteness of a single feature in the related art when identifying DC arc faults, and can effectively improve the accuracy of DC arc fault identification.

[0061] In order to better understand the process of the above-mentioned DC arc fault identification method, the above-mentioned DC arc fault identification method is described below in combination with an optional embodiment, but it is not used to limit the technical solution of the embodiment of the present application.

[0062] At present, with the development of the photovoltaic industry, arc faults are a key factor restricting the promotion and safe use of DC transmission and distribution technology. Although a large number of algorithms have been produced in recent years to effectively detect arc faults in lines, the following problems still exist: restricted by the experimental conditions of DC arc fault simulation, the quality of the original sample directly affects the difficulty of subsequent feature extraction; the circuit load is relatively simple during the simulation experiment, which to a certain extent causes the arc fault recognition model to be not universal; the feature extraction method used is relatively simple, and most of them only realize arc fault diagnosis. The unsupervised recognition of arc faults under complex working conditions such as mixed or sudden loads needs further research.

[0063] The present invention proposes a DC arc fault recognition method based on a dual-stream convolutional neural network. By utilizing the automatic feature extraction capability of the convolutional neural network, and targeting the limitations of the single-channel convolutional neural network due to its single structure and incomplete input features, a method for arc fault recognition based on a dual-stream convolutional neural network is proposed, which fuses one-dimensional features and two-dimensional features. The results show that the algorithm can well identify arc faults.

[0064] The original current data is classified into normal operating current, resistive load arc fault status, inductive load arc fault status, resistive-capacitive load arc fault status and motor load arc fault status.

[0065] The current signal is decomposed by the complete set empirical mode decomposition algorithm of adaptive noise to obtain multiple intrinsic mode function components. Then, based on the correlation and variance contribution between each component and the original signal, the IMF components with high contribution are selected for subsequent analysis and processing. Then, the one-dimensional DC arc fault feature vector is constructed by Hilbert transform. Equal amounts of data are randomly selected from the original arc fault data pool according to each category of labels, and the original data set is obtained by division, and finally the processed one-dimensional arc fault data set is obtained.

[0066] In the process of converting a one-dimensional signal into a two-dimensional image, the images generated by any time-frequency extraction method inevitably contain auxiliary information such as text, coordinate axes, and scale bars. In order to optimize these images to meet the input specifications of the subsequent recognition model and reduce the time cost required for image generation, a detailed preprocessing step is performed on each sample. These preprocessing measures include adjusting the direction of the coordinate axis, completely removing non-essential information in the image (such as text and scale bars), and performing efficient compression processing. After this series of processing, the output is an image in 64×64×3 format to meet specific analysis or recognition needs.

[0067] The DC arc fault current signal is encoded by the Gram angle and field, the original one-dimensional current signal is converted into a two-dimensional image, and the original data set is processed to obtain a processed two-dimensional arc fault data set.

[0068] The Gram angular field encodes the one-dimensional time series signal and converts it into a two-dimensional image. The converted image can retain the complete information in the signal in the time dimension. GAF encoding is a bijective mapping, so it retains all the information of the original one-dimensional current signal. During the encoding process, the time corresponding to the original data point increases from the upper left corner to the lower right corner of the image as the pixel position increases, retaining the time correlation. The value corresponding to the main diagonal is a special case, which is composed of the scaled size and angle of the original signal, and can achieve approximate reconstruction of the time series.

[0069] The current signals under different loads are collected, and the samples are divided into the ratio of training set: validation set: test set = 7:2:1 to obtain the original data set. One-dimensional features and two-dimensional features are extracted from all samples in the original data set according to categories to form a one-dimensional data set and a two-dimensional data set. Among them, by decomposing the original data, several components corresponding to each sample are obtained, and then the Hilbert transform is performed and the envelope spectrum is obtained. Finally, they are combined in sequence to construct high-dimensional sample features to form a one-dimensional data set; and the original data is subjected to Gram angle and field to obtain the coding map to form a two-dimensional data set.

[0070] Build a dual-channel CNN model and initialize the parameters. Input the training set and validation set data into the corresponding one-dimensional or two-dimensional channels for training. Then, aggregate and fuse the obtained one-dimensional and two-dimensional features. Select a batch size of 128 as the network model parameter. Add a pooling layer at the input. This can reduce the amount of calculation and shorten the calculation time. The convolution kernel size of the one-dimensional convolutional network feature extraction module is set to 5×1, 3×1, and 3×1 respectively to extract the features of DC arc faults. The two-dimensional data set convolutional network uses the Transformer model. The core module of the Transformer is the self-attention mechanism. The architecture diagram is as follows Figure 4 As shown: First, the input vector is converted into three different vectors and put into the query matrix Q, the key-value matrix K and the value matrix V. The similarity matrix is ​​obtained by performing a dot product of the transpose of the matrix Q and the matrix K. The larger the value in the similarity matrix, the higher the similarity. Then, the Softmax activation function is used to convert these values ​​into generalizations, and finally the weighted matrix is ​​obtained after multiplying them with the matrix V.

[0071] In order to improve computational efficiency, the attention module is improved by decomposing the two-dimensional self-attention mechanism into two one-dimensional self-attentions in the horizontal and vertical directions, and adding the position bias q of Q T r q And the characteristic information k of KT r k The single model structure has certain limitations. In order to extract the key features with distinguishing properties under different working conditions in the original current signal to the greatest extent, and give full play to the image processing and learning ability of the convolutional neural network, an improved two-stream convolutional neural network model is proposed to realize the organic combination of one-dimensional envelope spectrum and two-dimensional coding diagram, so as to classify and identify DC arc faults.

[0072] 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.

[0073] The present application also provides a structural block diagram of a DC arc fault identification device. Figure 5 is a structural block diagram of a DC arc fault identification device according to an embodiment of the present application; Figure 5 As shown, including:

[0074] An acquisition module 52 is used to acquire the current parameters collected in the current time period;

[0075] A decomposition module 54, configured to decompose the current parameter to obtain a first fault characteristic vector of the current parameter;

[0076] A conversion module 56, configured to perform image conversion based on the current parameter to obtain a second fault feature vector of the current parameter;

[0077] An extraction module 58, configured to extract a one-dimensional fault feature from the first fault feature vector, perform feature extraction on the second fault feature vector from multiple dimensions, and superimpose the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature;

[0078] The identification module 59 fuses the one-dimensional fault feature and the two-dimensional fault feature to obtain a fused fault feature, and performs DC arc fault identification according to the fused fault feature.

[0079] Through the above-mentioned device, the current parameters collected in the current time period are obtained, and the first fault feature vector of the current parameters is obtained by decomposition processing, and the second fault feature vector of the current parameters is obtained by image conversion. On this basis, one-dimensional fault features are extracted from the first fault feature vector, and features are extracted from the second fault feature vector from multiple dimensions, and the feature extraction results of multiple dimensions are superimposed to obtain two-dimensional fault features, and then the fused fault features are obtained by fusing the one-dimensional fault features and the two-dimensional fault features. DC arc faults are identified by fusing the fault features, which can solve the limitation of the incompleteness of a single feature in the related technology when identifying DC arc faults, and can effectively improve the accuracy of DC arc fault identification.

[0080] In an exemplary embodiment, the decomposition module 54 is also used to decompose the current parameter to obtain multiple modal function components corresponding to the current parameter; select a modal function component that meets a preset correlation condition from the multiple modal function components to determine it as a target modal function component; perform a Hilbert transform on the target modal function component to obtain a first fault characteristic vector of the current parameter.

[0081] In an exemplary embodiment, the conversion module 56 is further used to encode the current parameters using a Gram angle field to obtain an initial two-dimensional image of the current parameters; adjust the initial two-dimensional image to obtain a target two-dimensional image; and perform vector conversion based on the target two-dimensional image to obtain a second fault feature vector of the current parameters.

[0082] In an exemplary embodiment, the extraction module 58 is also used to use a pre-trained fault identification model to extract a one-dimensional fault feature from the first fault feature vector through the one-dimensional channel of the fault identification model, and to perform feature extraction on the second fault feature vector from multiple dimensions through the two-dimensional channel of the fault identification model, and to superimpose the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature.

[0083] In an exemplary embodiment, the multiple dimensions include a horizontal dimension and a vertical dimension; the extraction module 58 is further used to transform the second fault feature vector in the horizontal dimension through the two-dimensional channel of the fault identification model to obtain a query matrix, a key-value matrix and a value matrix in the horizontal dimension; respectively determine the position coding of the query matrix, the key-value matrix and the value matrix in the horizontal dimension, determine the position offset of the query matrix in the horizontal dimension, determine the characteristic parameters of the key-value matrix in the horizontal dimension, and obtain the horizontal fault feature in the horizontal dimension according to the position coding, the position offset and the characteristic parameters of the horizontal dimension; through the fault The two-dimensional channel of the identification model transforms the second fault feature vector in the vertical dimension to obtain the query matrix, key-value matrix and value matrix in the vertical dimension; respectively determines the position coding of the query matrix, the key-value matrix and the value matrix in the vertical dimension, determines the position offset of the query matrix in the vertical dimension, determines the characteristic parameters of the key-value matrix in the vertical dimension, and obtains the vertical fault feature in the vertical dimension according to the position coding, the position offset and the characteristic parameters of the vertical dimension; obtains the two-dimensional fault feature of the horizontal fault feature in the horizontal dimension and the vertical fault feature in the vertical dimension through the two-dimensional channel of the fault identification model.

[0084] In an exemplary embodiment, the extraction module 58 is further used to splice the horizontal fault features in the horizontal dimension and the vertical fault features in the vertical dimension to obtain an initial two-dimensional fault feature; determine a projection matrix of the initial two-dimensional fault feature; and multiply the initial two-dimensional fault feature and the projection matrix to obtain the two-dimensional fault feature.

[0085] In an exemplary embodiment, the extraction module 58 includes a model training module; the model training module is used to obtain training samples; the training samples include historical current parameters; the historical current parameters are decomposed to obtain a first sample fault feature vector of the historical current parameters; image conversion is performed based on the historical current parameters to obtain a second sample fault feature vector of the historical current parameters; based on the neural network model to be trained, a sample one-dimensional fault feature is extracted from the first sample fault feature vector, and feature extraction is performed on the second sample arc fault feature vector from multiple dimensions, and the sample feature extraction results of the multiple dimensions are superimposed to obtain a sample two-dimensional fault feature; the sample one-dimensional fault feature and the sample two-dimensional fault feature are fused to obtain a sample fusion fault feature; based on the neural network model to be trained, DC arc fault identification is performed based on the sample fusion fault feature to obtain a sample fault prediction result; the sample fault prediction result is compared with a pre-set sample fault label to determine the loss, and the neural network model to be trained is continued to be trained according to the damage until the iteration condition is met, and a trained fault identification model is obtained.

[0086] An embodiment of the present application further provides a storage medium, which includes a stored program, wherein the program executes any of the above methods when it is run.

[0087] Optionally, in this embodiment, the storage medium may be configured to store program codes for executing the following steps:

[0088] S1, obtaining the current parameters collected in the current time period;

[0089] S2, decomposing the current parameter to obtain a first fault characteristic vector of the current parameter;

[0090] S3, performing image conversion based on the current parameter to obtain a second fault feature vector of the current parameter;

[0091] S4, extracting a one-dimensional fault feature from the first fault feature vector, and performing feature extraction on the second fault feature vector from multiple dimensions, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature;

[0092] S5, fusing the one-dimensional fault feature and the two-dimensional fault feature to obtain a fused fault feature, and performing DC arc fault identification according to the fused fault feature.

[0093] 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.

[0094] Optionally, 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.

[0095] Optionally, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0096] S1, obtaining the current parameters collected in the current time period;

[0097] S2, decomposing the current parameter to obtain a first fault characteristic vector of the current parameter;

[0098] S3, performing image conversion based on the current parameter to obtain a second fault feature vector of the current parameter;

[0099] S4, extracting a one-dimensional fault feature from the first fault feature vector, and performing feature extraction on the second fault feature vector from multiple dimensions, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature;

[0100] S5, fusing the one-dimensional fault feature and the two-dimensional fault feature to obtain a fused fault feature, and performing DC arc fault identification according to the fused fault feature.

[0101] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: 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, and other media that can store program codes.

[0102] An embodiment of the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps in any one of the above method embodiments are implemented.

[0103] 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.

[0104] An embodiment of the present application also provides a computer program, which includes computer instructions stored in a computer-readable storage medium; a processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps in any one of the above method embodiments.

[0105] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.

[0106] 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, and optionally, 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.

[0107] 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 principles of the present application shall be included in the protection scope of the present application.

Claims

1. A DC arc fault identification method, characterized in that: include: Obtain the current parameters collected during the current time period; Decomposing the current parameter to obtain a first fault characteristic vector of the current parameter; Performing image conversion based on the current parameter to obtain a second fault feature vector of the current parameter; Extracting a one-dimensional fault feature from the first fault feature vector, performing feature extraction on the second fault feature vector from multiple dimensions, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature; The one-dimensional fault feature and the two-dimensional fault feature are fused to obtain a fused fault feature, and direct current arc fault identification is performed according to the fused fault feature.

2. The method according to claim 1, characterized in that: The decomposing the current parameter to obtain a first fault characteristic vector of the current parameter includes: Decomposing the current parameter to obtain a plurality of modal function components corresponding to the current parameter; Selecting a modal function component satisfying a preset correlation condition from the multiple modal function components and determining it as a target modal function component; A Hilbert transform is performed on the target modal function component to obtain a first fault characteristic vector of the current parameter.

3. The method according to claim 1, characterized in that: The performing image conversion based on the current parameter to obtain a second fault feature vector of the current parameter includes: Encoding the current parameters using a Gram angle field to obtain an initial two-dimensional image of the current parameters; Adjusting the initial two-dimensional image to obtain a target two-dimensional image; Vector conversion is performed based on the target two-dimensional image to obtain a second fault feature vector of the current parameter.

4. The method according to claim 1, characterized in that: The step of extracting a one-dimensional fault feature from the first fault feature vector, performing feature extraction on the second fault feature vector from multiple dimensions, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature includes: A pre-trained fault identification model is used to extract a one-dimensional fault feature from the first fault feature vector through the one-dimensional channel of the fault identification model, and feature extraction is performed on the second fault feature vector from multiple dimensions through the two-dimensional channel of the fault identification model, and the feature extraction results of the multiple dimensions are superimposed to obtain a two-dimensional fault feature.

5. The method according to claim 4, characterized in that The multiple dimensions include a horizontal dimension and a vertical dimension; The extracting features of the second fault feature vector from multiple dimensions through the two-dimensional channel of the fault identification model, and superimposing the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature, includes: The second fault feature vector is transformed in the horizontal dimension through the two-dimensional channel of the fault identification model to obtain a query matrix, a key-value matrix and a value matrix in the horizontal dimension; the position codes of the query matrix, the key-value matrix and the value matrix in the horizontal dimension are respectively determined, as well as the position offset of the query matrix in the horizontal dimension and the characteristic parameters of the key-value matrix in the horizontal dimension, and the horizontal fault feature in the horizontal dimension is obtained according to the position code, the position offset and the characteristic parameters in the horizontal dimension; The second fault feature vector is transformed in the vertical dimension through the two-dimensional channel of the fault identification model to obtain a query matrix, a key-value matrix and a value matrix in the vertical dimension; the position codes of the query matrix, the key-value matrix and the value matrix in the vertical dimension are determined respectively, as well as the position offset of the query matrix in the vertical dimension and the characteristic parameters of the key-value matrix in the vertical dimension are determined, and the vertical fault feature in the vertical dimension is obtained according to the position code, the position offset and the characteristic parameters in the vertical dimension; Through the two-dimensional channel of the fault identification model, the horizontal fault features in the horizontal dimension and the vertical fault features in the vertical dimension are used to obtain two-dimensional fault features.

6. The method according to claim 1, characterized in that The two-dimensional fault feature is obtained by using the two-dimensional channel of the fault identification model for the horizontal fault feature in the horizontal dimension and the vertical fault feature in the vertical dimension, including: splicing the horizontal fault features of the horizontal dimension and the vertical fault features of the vertical dimension to obtain an initial two-dimensional fault feature; Determining a projection matrix of the initial two-dimensional fault feature; The initial two-dimensional fault feature and the projection matrix are multiplied to obtain the two-dimensional fault feature.

7. The method according to claim 4, characterized in that The training step of the fault identification model includes: Acquire a training sample; the training sample includes historical current parameters; Decomposing the historical current parameter to obtain a first sample fault feature vector of the historical current parameter; Performing image conversion based on the historical current parameters to obtain a second sample fault feature vector of the historical current parameters; Extracting a sample one-dimensional fault feature from the first sample fault feature vector based on the neural network model to be trained, performing feature extraction on the second sample arc fault feature vector from multiple dimensions, and superimposing the sample feature extraction results of the multiple dimensions to obtain a sample two-dimensional fault feature; Fusion of the sample one-dimensional fault feature and the sample two-dimensional fault feature to obtain a sample fusion fault feature; Based on the neural network model to be trained, DC arc fault identification is performed based on the sample fusion fault characteristics to obtain a sample fault prediction result; The sample fault prediction result is compared with a preset sample fault label to determine the loss, and the neural network model to be trained is continuously trained according to the damage until an iteration condition is met, thereby obtaining a trained fault recognition model.

8. A DC arc fault identification device, characterized in that: include: An acquisition module is used to acquire the current parameters collected in the current time period; A decomposition module, used for decomposing the current parameter to obtain a first fault characteristic vector of the current parameter; A conversion module, used for performing image conversion based on the current parameter to obtain a second fault feature vector of the current parameter; An extraction module, used to extract a one-dimensional fault feature from the first fault feature vector, and perform feature extraction on the second fault feature vector from multiple dimensions, and superimpose the feature extraction results of the multiple dimensions to obtain a two-dimensional fault feature; The identification module is used to fuse the one-dimensional fault feature and the two-dimensional fault feature to obtain a fused fault feature, and perform DC arc fault identification according to the fused fault feature.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the method described in any one of claims 1 to 7 when executed.

10. 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 7 through the computer program.

Images