Training of arc fault detection model, arc fault detection method and electronic device

Abstract

The present disclosure provides a training method of an arc fault detection model, an arc fault detection method and an electronic device, which can be applied to the technical field of arc detection. The method comprises: processing at least one arc detection data by using an attention module corresponding to a first arc fault detection model to obtain a frequency band importance degree corresponding to each of the at least one arc detection data, wherein the frequency band importance degree represents the importance of a frequency band corresponding to the arc detection data; determining at least one target arc detection data from the at least one arc detection data by using the at least one frequency band importance degree; and training a second arc fault detection model by using the at least one target arc detection data to obtain a target arc fault detection model.

Description

Technical Field

[0001] This disclosure relates to the field of arc detection technology, and in particular to the training of an arc fault detection model, an arc fault detection method, and an electronic device. Background Technology

[0002] Arcing faults are a threat to system safety. For example, in photovoltaic (PV) systems, the energy generated comes from solar energy, causing no environmental pollution during power generation and producing considerable output, making it an important component of distributed power systems. However, because PV systems can operate at voltages up to 800V, arcing faults become a significant safety threat. Arcing faults in PV systems can generate temperatures as high as 20,000K, easily leading to electrical fires.

[0003] In realizing the present invention, the inventors discovered at least the following problem: the accuracy of arc fault detection is low. Summary of the Invention

[0004] In view of the above problems, this disclosure provides training of an arc fault detection model, an arc fault detection method, and an electronic device.

[0005] The first aspect of this disclosure provides a method for training an arc fault detection model, comprising:

[0006] At least one arc detection data is processed using an attention module corresponding to the first arc fault detection model to obtain the frequency band importance corresponding to each of the at least one arc detection data, wherein the frequency band importance characterizes the importance of the frequency band corresponding to the arc detection data;

[0007] Using at least one frequency band importance, determine at least one target arc detection data point from at least one arc detection data point; and

[0008] A second arc fault detection model is trained using at least one target arc detection data to obtain the target arc fault detection model.

[0009] According to embodiments of this disclosure, the attention module includes a first pooling layer, a convolutional layer, and a first activation layer;

[0010] Specifically, at least one arc detection data is processed using an attention module corresponding to the first arc fault detection model to obtain the frequency band importance corresponding to each of the at least one arc detection data, including:

[0011] At least one arc detection data is processed using the first pooling layer corresponding to the first arc fault detection model to obtain first pooling data corresponding to each of the at least one arc detection data.

[0012] At least one first pooling data is processed using a convolutional layer corresponding to the first arc fault detection model to obtain at least one convolutional data; and

[0013] At least one convolutional data is processed using the first activation layer corresponding to the first arc fault detection model to obtain the frequency band importance corresponding to each of the at least one arc detection data.

[0014] According to embodiments of this disclosure, the attention module includes a second pooling layer, a third pooling layer, a first fully connected layer, and a second activation layer;

[0015] Specifically, at least one arc detection data is processed using an attention module corresponding to the first arc fault detection model to obtain the frequency band importance corresponding to each of the at least one arc detection data, including:

[0016] At least one arc detection data is processed using the second pooling layer corresponding to the first arc fault detection model to obtain second pooling data corresponding to each of the at least one arc detection data.

[0017] At least one arc detection data is processed using the third pooling layer corresponding to the first arc fault detection model to obtain the third pooling data corresponding to each of the at least one arc detection data.

[0018] Using the first fully connected layer corresponding to the first arc fault detection model, the second pooling data and third pooling data corresponding to each of at least one arc detection data are processed to obtain fully connected data corresponding to each of the at least one arc detection data; and

[0019] By using the second activation layer corresponding to the first arc fault detection model to process the fully connected data corresponding to each of the at least one arc detection data, the frequency band importance corresponding to each of the at least one arc detection data is obtained.

[0020] According to embodiments of this disclosure, the attention module includes a second fully connected layer.

[0021] According to embodiments of this disclosure, determining at least one target arc detection data from at least one arc detection data using at least one frequency band importance includes:

[0022] Based on the frequency band corresponding to each of the at least one arc detection data, an arc detection dataset corresponding to each of the at least one arc detection data is determined, wherein the arc detection dataset corresponds to the frequency band interval;

[0023] Based on the importance of at least one frequency band corresponding to each of at least one arc detection dataset, determine at least one target frequency band interval from at least one frequency band interval; and

[0024] The arc detection data in the arc detection dataset corresponding to each of at least one target frequency band interval is determined as the target arc detection data.

[0025] According to embodiments of this disclosure, determining at least one target frequency band interval from at least one frequency band interval based on at least one frequency band importance corresponding to each of at least one arc detection dataset includes:

[0026] Based on the importance of at least one frequency band corresponding to each of at least one arc detection dataset, determine the comprehensive frequency band importance corresponding to each of the at least one arc detection dataset; and

[0027] Based on the comprehensive frequency band importance corresponding to each of the at least one arc detection dataset, at least one target frequency band interval is determined from at least one frequency band interval.

[0028] According to embodiments of this disclosure, determining the comprehensive frequency band importance corresponding to each of the at least one arc detection dataset based on at least one frequency band importance corresponding to each of the at least one arc detection dataset includes:

[0029] For each arc detection dataset in at least one arc detection dataset,

[0030] Determine the statistical importance of the arc detection dataset based on the importance of at least one frequency band corresponding to the arc detection dataset; and

[0031] The statistical importance corresponding to the arc detection dataset is determined as the comprehensive frequency band importance corresponding to the arc detection dataset.

[0032] According to embodiments of this disclosure, the training method for the above-mentioned arc fault detection model further includes:

[0033] According to a preset time window, at least one initial arc detection data is divided to obtain at least one arc detection interval data corresponding to each of the at least one initial arc detection data.

[0034] Perform frequency domain transformation on at least one arc detection interval data corresponding to each of at least one initial arc detection data to obtain frequency domain arc detection data corresponding to each of the at least one initial arc detection data; and

[0035] At least one frequency domain arc detection data is normalized to obtain at least one arc detection data.

[0036] According to embodiments of this disclosure, the training method for the above-mentioned arc fault detection model further includes:

[0037] Visualize at least one target electric arc detection data to obtain a visualization result; and

[0038] Adjust at least one target arc detection data based on the visualization results.

[0039] A second aspect of this disclosure provides a method for detecting electric arc faults, comprising:

[0040] Acquire the data of the electric arc to be detected; and

[0041] Input the arc data to be detected into the target arc fault detection model and output the arc fault detection results;

[0042] The target arc fault detection mode is obtained by training an arc fault detection model according to an embodiment of the present disclosure.

[0043] A third aspect of this disclosure provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors perform the methods described above.

[0044] According to the training method, equipment, and arc fault detection method of the arc fault detection model provided in this disclosure, at least one target arc fault detection data can be determined from at least one arc fault detection data by using the frequency band importance corresponding to each of the at least one arc fault detection data. This can reduce the amount of data required to train the second arc fault detection model, thereby improving the training speed of the target arc fault detection model. Furthermore, by using the attention module to obtain the frequency band importance, at least one target arc fault detection data can be better determined based on the importance of the frequency band corresponding to the arc fault detection data, reducing the interference of useless arc fault detection data on the second arc fault detection model and improving the accuracy of the target arc fault detection model. Attached Figure Description

[0045] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0046] Figure 1 The diagram illustrates an application scenario of a training method for an arc fault detection model according to an embodiment of the present disclosure.

[0047] Figure 2 A flowchart illustrating a training method for an arc fault detection model according to an embodiment of the present disclosure is shown schematically.

[0048] Figure 3A A schematic diagram of the structure of a first arc fault detection model according to an embodiment of the present disclosure is shown.

[0049] Figure 3BThe diagram illustrates the training results of a first arc fault detection model according to an embodiment of the present disclosure.

[0050] Figure 4A A schematic diagram illustrating the frequency band of arc detection data according to an embodiment of the present disclosure is shown.

[0051] Figure 4B A schematic diagram illustrating the statistical importance of arc detection data in the presence of arc faults according to an embodiment of the present disclosure is shown.

[0052] Figure 4C A schematic diagram illustrating the statistical importance of arc detection data in the absence of arc faults according to an embodiment of the present disclosure is shown.

[0053] Figure 4D A schematic diagram illustrating the frequency band of target arc detection data according to an embodiment of the present disclosure is shown.

[0054] Figure 5 The diagram illustrates the training results of a target arc detection model according to an embodiment of the present disclosure.

[0055] Figure 6 A current signal waveform diagram according to an embodiment of the present disclosure is schematically shown;

[0056] Figure 7A The illustration schematically shows a frequency domain characteristic analysis spectrum of a current signal with a preset time window of 2ms according to an embodiment of the present disclosure;

[0057] Figure 7B The illustration schematically shows a frequency domain characteristic analysis spectrum of a current signal with a preset time window of 10ms according to an embodiment of the present disclosure;

[0058] Figure 7C The illustration schematically shows a frequency domain characteristic analysis spectrum of a current signal with a preset time window of 50ms according to an embodiment of the present disclosure;

[0059] Figure 8A A schematic diagram illustrating the visualization results of target arc detection data according to an embodiment of the present disclosure is shown.

[0060] Figure 8B A schematic diagram illustrating the initial visualization results of initial arc detection data according to an embodiment of the present disclosure is shown.

[0061] Figure 8C A schematic diagram illustrating the frequency domain visualization results of frequency domain arc detection data according to an embodiment of the present disclosure is shown.

[0062] Figure 8D This diagram schematically illustrates the current variation during the inverter startup process according to an embodiment of the present disclosure;

[0063] Figure 9 A flowchart illustrating an arc fault detection method according to an embodiment of the present disclosure is shown schematically.

[0064] Figure 10 This schematic diagram illustrates a structural block diagram of a training apparatus for an arc fault detection model according to an embodiment of the present disclosure;

[0065] Figure 11 A schematic diagram illustrating the structure of an arc fault detection device according to an embodiment of the present disclosure is shown; and

[0066] Figure 12 A block diagram of an electronic device suitable for implementing a training method for an arc fault detection model according to an embodiment of the present disclosure is shown schematically. Detailed Implementation

[0067] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0068] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0069] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0070] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).

[0071] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of data (including but not limited to user personal information) comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and they do not violate public order and good morals.

[0072] Taking a photovoltaic (PV) system as an example, fires caused by arc faults in PV systems pose a significant hazard. Therefore, the detection methods for arc faults are a key research focus. PV arc faults are classified into series arc faults, parallel arc faults, and grounding arc faults. Parallel and grounding arc faults, due to their larger currents, are easily tripped by overload protection circuit breakers. However, series arc faults, due to load limitations, generate currents smaller than normal currents, making them difficult to protect against by traditional overload protection circuit breakers and other conventional protection devices.

[0073] Threshold detection is a method for detecting electric arc faults. It relies on the time-frequency domain characteristics of the arc's voltage and current, and depends on a manually set fault arc identification threshold. Its advantages include simple algorithms and low-cost implementation. However, the range of variation of the time-frequency domain characteristics of the fault arc varies under different system operating conditions, making it difficult to accurately set the fault identification threshold based on engineering experience. In photovoltaic systems, inverters are power electronic devices that easily generate strong conducted interference in the bus current, and their frequency band overlaps with the arc current frequency band, affecting the accuracy of fault arc identification. To avoid the inconvenience caused by frequent malfunctions, the fault identification threshold needs to be increased, but this increases the probability of protection failure to operate.

[0074] In view of this, embodiments of the present disclosure provide a training method for an arc fault detection model, comprising: processing at least one arc detection data using an attention module corresponding to a first arc fault detection model to obtain frequency band importance corresponding to each of the at least one arc detection data, wherein the frequency band importance characterizes the importance of the frequency band corresponding to the arc detection data; determining at least one target arc detection data from the at least one arc detection data using the at least one frequency band importance; and training a second arc fault detection model using the at least one target arc detection data to obtain a target arc fault detection model.

[0075] By determining at least one target arc detection data point from at least one arc detection data point based on the frequency band importance corresponding to each of the at least one arc detection data point, the amount of data required to train the second arc fault detection model can be reduced, thereby improving the training speed of the target arc fault detection model. Furthermore, by using the attention module to obtain the frequency band importance, at least one target arc detection data point can be better determined based on the importance of the frequency band corresponding to the arc detection data, reducing the interference of useless arc detection data on the second arc fault detection model and improving the accuracy of the target arc fault detection model.

[0076] Figure 1 The diagram illustrates an application scenario of a training method for an arc fault detection model according to an embodiment of the present disclosure.

[0077] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0078] Users can interact with server 105 via network 104 using at least one of the first terminal device 101, second terminal device 102, and third terminal device 103 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, second terminal device 102, and third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0079] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0080] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0081] It should be noted that the training method for the arc fault detection model provided in this embodiment can generally be executed by server 105. Correspondingly, the training device for the arc fault detection model provided in this embodiment can generally be located in server 105. The training method for the arc fault detection model provided in this embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the training device for the arc fault detection model provided in this embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0082] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0083] Figure 2 A flowchart illustrating a training method for an arc fault detection model according to an embodiment of the present disclosure is shown.

[0084] like Figure 2 As shown, the training method for the arc fault detection model in this embodiment includes operations S210 to S230.

[0085] In operation S210, at least one arc detection data is processed using the attention module corresponding to the first arc fault detection model to obtain the frequency band importance corresponding to each of the at least one arc detection data, wherein the frequency band importance characterizes the importance of the frequency band corresponding to the arc detection data.

[0086] In operation S220, at least one target arc detection data is determined from at least one arc detection data using at least one frequency band importance.

[0087] In operation S230, a second arc fault detection model is trained using at least one target arc detection data to obtain the target arc fault detection model.

[0088] According to embodiments of this disclosure, at least one arc detection data may be data indicating the presence of an arc fault, or data indicating the absence of an arc fault.

[0089] According to embodiments of this disclosure, the first arc fault detection model may be a pre-trained model for detecting arc faults. The first arc fault detection model may be trained based on at least one arc detection data point.

[0090] According to embodiments of this disclosure, the first arc fault detection model may include an attention module. The attention module can determine which part of the input data the first arc fault detection model needs to focus on.

[0091] According to embodiments of this disclosure, the frequency band importance can be a numerical value, and a frequency band importance threshold can be set. Based on the comparison result between the frequency band importance and the frequency band importance threshold, at least one target arc detection data is determined from at least one arc detection data.

[0092] According to embodiments of this disclosure, the frequency band importance can be a level, and the frequency band importance level can be preset. Based on the frequency band importance level, at least one target arc detection data can be determined from at least one arc detection data.

[0093] According to embodiments of this disclosure, since at least one target arc detection data is determined from at least one arc detection data based on frequency band importance, at least one target arc detection data can be data that better reflects arc faults.

[0094] According to embodiments of this disclosure, the second arc fault detection model can be a model used to detect arc faults.

[0095] According to embodiments of this disclosure, the model structure of the second arc fault detection model can be the same as that of the first arc fault detection model, or it can be a different model structure from the first arc fault detection model.

[0096] According to embodiments of this disclosure, the attention module enables the first arc fault detection model to interpret the detection results of the arc detection data, that is, the recognition of arc features in the arc detection data by the first arc fault detection model can be determined according to the frequency band importance.

[0097] According to embodiments of this disclosure, at least one target arc detection data can be determined from at least one arc detection data by using the frequency band importance corresponding to each of the at least one arc detection data. This reduces the amount of data required to train the second arc fault detection model, thereby improving the training speed of the target arc fault detection model. Furthermore, by using an attention module to obtain the frequency band importance, at least one target arc detection data can be better determined based on the importance of the frequency band corresponding to the arc detection data, reducing the interference of useless arc detection data on the second arc fault detection model and improving the accuracy of the target arc fault detection model.

[0098] According to embodiments of this disclosure, the attention module includes a first pooling layer, a convolutional layer, and a first activation layer.

[0099] The process involves processing at least one arc detection data point using an attention module corresponding to the first arc fault detection model to obtain the frequency band importance corresponding to each of the at least one arc detection data point. This includes: processing the at least one arc detection data point using a first pooling layer corresponding to the first arc fault detection model to obtain first pooled data corresponding to each of the at least one arc detection data point; processing the at least one first pooled data point using a convolutional layer corresponding to the first arc fault detection model to obtain at least one convolutional data point; and processing the at least one convolutional data point using a first activation layer corresponding to the first arc fault detection model to obtain the frequency band importance corresponding to each of the at least one arc detection data point.

[0100] According to embodiments of this disclosure, the first pooling layer may be a global average pooling layer. The global average pooling layer can replace the fully connected layer in receiving arc detection data, and it can receive input feature maps of arbitrary size.

[0101] According to embodiments of this disclosure, at least one arc detection data can be converted into at least one arc input feature map corresponding to the at least one arc detection data, and the at least one arc input feature map can be input into a first pooling layer to obtain first pooling data corresponding to each of the at least one arc detection data.

[0102] According to embodiments of this disclosure, a convolutional layer can convolve at least one first pooling data to extract features from the first pooling data, thereby obtaining at least one convolutional data.

[0103] According to embodiments of this disclosure, the first activation layer may be an activation layer activated using the Sigmoid activation function.

[0104] According to embodiments of this disclosure, the attention module may have multiple channels. By processing at least one convolutional data through a first activation layer, the weights of each channel of the attention module can be obtained. The output data of the attention module can then be obtained based on the weights of each channel. Based on the output data of the attention module, the frequency band importance corresponding to each of the at least one arc detection data can be obtained.

[0105] According to embodiments of this disclosure, the attention module includes a second pooling layer, a third pooling layer, a first fully connected layer, and a second activation layer.

[0106] The process involves processing at least one arc detection data point using an attention module corresponding to the first arc fault detection model to obtain the frequency band importance corresponding to each of the at least one arc detection data point. This includes: processing the at least one arc detection data point using a second pooling layer corresponding to the first arc fault detection model to obtain second pooled data corresponding to each of the at least one arc detection data point; processing the at least one arc detection data point using a third pooling layer corresponding to the first arc fault detection model to obtain third pooled data corresponding to each of the at least one arc detection data point; processing the second pooled data and third pooled data corresponding to each of the at least one arc detection data point using a first fully connected layer corresponding to the first arc fault detection model to obtain fully connected data corresponding to each of the at least one arc detection data point; and processing the fully connected data corresponding to each of the at least one arc detection data point using a second activation layer corresponding to the first arc fault detection model to obtain the frequency band importance corresponding to each of the at least one arc detection data point.

[0107] According to embodiments of this disclosure, the second pooling layer may be a global average pooling layer, and the third pooling layer may be a global max pooling layer.

[0108] According to embodiments of this disclosure, the second pooling layer and the third pooling layer can process at least one arc detection data in parallel to obtain second pooling data and third pooling data, respectively. The size of at least one arc detection data can be H*W*C, and the number of channels in the second pooling layer and the third pooling layer can be 1*1*C. Different channels can extract different features.

[0109] According to embodiments of this disclosure, the second pooling data obtained by the second pooling layer and the third pooling data obtained by the third pooling layer can share a first fully connected layer, and the fully connected data can be obtained using the first fully connected layer.

[0110] According to embodiments of this disclosure, the second activation layer may be an activation layer activated by the Sigmoid activation function.

[0111] According to embodiments of this disclosure, the fully connected data can obtain the output data of the attention module through the second activation layer, thereby obtaining the frequency band importance corresponding to each of at least one arc detection data based on the output data.

[0112] According to embodiments of this disclosure, the attention module includes a second fully connected layer.

[0113] According to embodiments of this disclosure, the attention module is trained using a first arc fault detection model, and the training process of the first arc fault detection model is described below.

[0114] Figure 3A A schematic diagram of the structure of a first arc fault detection model according to an embodiment of the present disclosure is shown.

[0115] As shown in Figure 3, the first arc fault detection model may include an attention module, four one-dimensional convolutional layers and four max pooling layers followed by three third fully connected layers.

[0116] According to embodiments of this disclosure, the attention module can be any of the structures described in the foregoing embodiments. For example, the attention module can be a structure including a second fully connected layer, and the attention module can also include a product layer. At least one arc fault detection data can obtain the weight of each input channel through the second fully connected layer, and then pass through the product layer to multiply the original input data with the weights obtained by the fully connected layer to obtain the output data of the attention module. The output data of the attention module is transmitted to a one-dimensional convolutional layer of the first arc fault detection model. Each one-dimensional convolutional layer is followed by a ReLU activation operation and then a max pooling layer.

[0117] As shown in Figure 3, the input size of at least one arc fault detection data in the attention module can be 1000x1. The first and third one-dimensional convolutional layers can each have 96 filters, the second one-dimensional convolutional layer can have 128 filters, and the fourth one-dimensional convolutional layer can have 64 filters. The filter size of each one-dimensional convolutional layer is 5*1. The size of the max pooling layer is 2*1 with a stride of 1, which effectively reduces the size of the feature map and improves computational efficiency. After the last max pooling layer, a flattening operation is performed to reduce the dimensionality of the data. After the flattening operation, three fully connected layers with 64, 32, and 2 neurons respectively are connected. The output layer is a classification layer, which can include two neurons, each corresponding to the running state of at least one arc fault detection data. The running state can be either the presence or absence of an arc fault. Based on the final output, the softmax function can be used to convert the scores into probabilities that sum to 1, and the class with the higher probability is taken as the final classification state.

[0118] According to embodiments of this disclosure, the training process of the first arc fault detection model can be performed using the following formulas (1) to (6).

[0119] E=ωx+b (1)

[0120]

[0121] y = Wx (3)

[0122]

[0123] ReLU(x) = Max(0,x) (5)

[0124]

[0125] Where x represents at least one arc detection data point, E represents the output data of the second fully connected layer, ω represents the weight of at least the arc detection data point in the attention module, b represents the bias matrix, and e i Let represent the output data of the i-th arc detection data out of T arc detection data in the second fully connected layer, W represent the frequency band importance of the i-th arc detection data, y represent the output data of the attention module, f is the convolution vector of the one-dimensional convolutional layer, k represents the convolution kernel of the one-dimensional convolutional layer, j represents the j-th convolution kernel in r convolution kernels, and N represents the total number of training data. Indicates the prediction result, y j This represents the output of the first arc detection model.

[0126] According to embodiments of this disclosure, a first arc fault detection model can extract arc features using one-dimensional convolution. The feature extraction layer, composed of multiple one-dimensional convolutional layers, optimizes network parameters through multiple forward and backward propagations. During forward propagation, input features pass through convolutional layers, and output features are generated by the activation function of each layer. During backward propagation, the error between the output and the training labels is calculated using an error function, and the error function is propagated back to each layer. Finally, the network parameters are updated using a gradient descent formula.

[0127] Figure 3B The diagram illustrates the training results of a first arc fault detection model according to an embodiment of the present disclosure.

[0128] like Figure 3B As shown, the accuracy of the validation set is close to that of the training set, indicating that the first arc fault detection model did not overfit. Table 1 shows the confusion matrix of the first arc fault detection model. To evaluate the model performance, the accuracy, recall, and precision were calculated based on the confusion matrix in Table 1. The first arc fault detection model achieved an accuracy of 97.04%, a recall of 98.03%, and a precision of 96.00%, with both accuracy and recall exceeding 96%.

[0129] Table 1

[0130]

[0131] According to embodiments of this disclosure, by obtaining the frequency band importance corresponding to each of at least one arc detection data through an attention module, the features of the arc in the arc detection data can be effectively extracted.

[0132] According to embodiments of this disclosure, determining at least one target arc detection data from at least one arc detection data using at least one frequency band importance includes: determining an arc detection dataset corresponding to each of the at least one arc detection data based on the frequency band corresponding to each of the at least one arc detection data, wherein the arc detection dataset corresponds to a frequency band interval; determining at least one target frequency band interval from the at least one frequency band interval based on the at least one frequency band importance corresponding to each of the at least one arc detection dataset; and determining the arc detection data in the arc detection dataset corresponding to each of the at least one target frequency band interval as the target arc detection data.

[0133] According to embodiments of this disclosure, the frequency band interval can be a frequency domain interval pre-divided based on the frequency range of at least one arc detection data. For example, the frequency range of the arc detection data can be 10kHz to 100kHz, and the frequency band interval can be 9 intervals in 10kHz increments, or it can be 17 frequency band intervals in 5kHz increments.

[0134] According to embodiments of this disclosure, the frequency band corresponding to each of the at least one arc detection data can be determined based on the frequency of the at least one arc detection data. For example, if the frequency of the arc detection data is 16 kHz, the frequency band of the arc detection data can be determined to be 10 kHz to 20 kHz or 15 kHz to 25 kHz.

[0135] According to embodiments of this disclosure, based on the frequency band corresponding to each of the at least one arc detection data, a comparison can be made with a frequency band interval, and the arc detection dataset corresponding to the frequency band interval that is the same as the frequency band corresponding to each of the at least one arc detection data is determined as the arc detection dataset corresponding to each of the at least one arc detection data.

[0136] According to embodiments of this disclosure, frequency band intervals are sorted according to the frequency band importance corresponding to the arc detection data in the arc detection dataset to determine target frequency band intervals, thereby determining the arc detection data in the arc detection dataset corresponding to at least one target frequency band interval as target arc detection data.

[0137] According to embodiments of this disclosure, by determining at least one target frequency band interval through the frequency band interval corresponding to the arc detection dataset, the arc detection data can be divided more finely, making it easier to determine the target arc detection data.

[0138] According to embodiments of this disclosure, determining at least one target frequency band interval from at least one frequency band interval based on at least one frequency band importance corresponding to each of at least one arc detection dataset includes: determining a comprehensive frequency band importance corresponding to each of the at least one arc detection dataset based on the at least one frequency band importance corresponding to each of the at least one arc detection dataset; and determining at least one target frequency band interval from at least one frequency band interval based on the comprehensive frequency band importance corresponding to each of the at least one arc detection dataset.

[0139] According to embodiments of this disclosure, the comprehensive frequency band importance can characterize the importance of the frequency band interval corresponding to the arc detection dataset.

[0140] According to embodiments of this disclosure, the overall frequency band importance can be determined based on at least one arc detection data in the arc detection dataset.

[0141] According to embodiments of this disclosure, the importance of a frequency band interval corresponding to an arc detection dataset can be determined based on the magnitude of the overall frequency band importance, and at least one target frequency band interval can be determined based on the importance of the frequency band interval.

[0142] According to embodiments of this disclosure, determining the comprehensive frequency band importance corresponding to each of the at least one arc detection dataset based on the frequency band importance corresponding to each of the at least one arc detection dataset may include: for each of the at least one arc detection datasets: determining the statistical importance corresponding to the arc detection dataset based on the frequency band importance corresponding to the arc detection dataset; and determining the statistical importance corresponding to the arc detection dataset as the comprehensive frequency band importance corresponding to the arc detection dataset.

[0143] According to embodiments of this disclosure, the sum of the importance scores of all frequency bands corresponding to the arc detection dataset can be calculated, and this sum can be determined as the statistical importance corresponding to the arc detection dataset. Alternatively, the average importance score of all frequency bands corresponding to the arc detection dataset can be calculated, and this average score can be determined as the statistical importance corresponding to the arc detection dataset. Furthermore, the maximum value of the importance score of all frequency bands corresponding to the arc detection dataset can be calculated, and this maximum value can be determined as the statistical importance corresponding to the arc detection dataset.

[0144] Figure 4A A schematic diagram of the frequency band of arc detection data according to an embodiment of the present disclosure is shown.

[0145] Figure 4B A schematic diagram illustrating the statistical importance of arc detection data in the presence of arc faults according to an embodiment of the present disclosure is shown.

[0146] Figure 4CA schematic diagram illustrating the statistical importance of arc detection data in the absence of arc faults according to an embodiment of the present disclosure is shown.

[0147] Figure 4D A schematic diagram illustrating the frequency band of target arc detection data according to an embodiment of the present disclosure is shown.

[0148] like Figure 4A As shown, the frequency range of the arc detection data spectrum is 3kHz to 125kHz. However, in the high-frequency energy spectrum, there is a high degree of overlap between the spectra of arc detection data with and without arc faults. For example, the high-frequency band can include frequencies beyond 30kHz. The overlapping frequency bands do not contribute positively to arc fault identification and may even reduce the accuracy of the target arc fault detection model. Therefore, it is necessary to scientifically select the target arc detection data, reduce the model width, and improve the accuracy of arc fault identification.

[0149] Such as 4B and Figure 4C As shown, the frequency range of the arc detection data spectrum is 3kHz to 125kHz. Using a frequency band interval length of 10kHz, the importance of each frequency band within that interval is summed to obtain the statistical importance. To extract key feature frequency bands more comprehensively, the 3-125kHz frequency interval can be divided using an overlap sampling method with a step size of 5kHz, ultimately resulting in 24 frequency band intervals.

[0150] like Figure 4B As shown, when identifying arc detection data with arc faults, the statistical importance of the 8-23kHz frequency band is more significant. Figure 4C As shown, when identifying arc detection data without arc faults, the statistical importance of the 8-18kHz frequency band is relatively significant, as is the statistical importance of the 28-38kHz frequency band. Furthermore, arc detection data across the entire 3-125kHz frequency band can play a positive role in training for identifying arc faults without arc faults.

[0151] Therefore, the 8-18kHz frequency band corresponding to the arc detection data with arc faults, the 8-18kHz frequency band corresponding to the arc detection data without arc faults, and the 28-38kHz frequency band can be defined as target frequency bands. The arc detection data in the arc detection dataset corresponding to at least one target frequency band is defined as the target arc detection data. The joint spectrum of the target arc detection data is as follows: Figure 4D As shown.

[0152] Figure 5 The diagram illustrates the training results of a target arc detection model according to an embodiment of the present disclosure.

[0153] like Figure 5 As shown, the accuracy of the target arc fault detection model has been improved to 99.89%. It can be seen that the training accuracy and validation accuracy of the target arc fault detection model increase faster. This is because the target arc fault detection model needs to learn less content, and the joint frequency band can better represent the arc fault.

[0154] Table 2

[0155]

[0156]

[0157] According to the experimental results of this disclosure, training the second arc fault detection model based on the target arc detection data determined by the attention module can significantly improve the accuracy of arc fault identification and reduce the false alarm rate.

[0158] According to embodiments of this disclosure, the training method for the above-mentioned arc fault detection model further includes: segmenting at least one initial arc detection data according to a preset time window to obtain at least one arc detection interval data corresponding to each of the at least one initial arc detection data; performing frequency domain transformation on the at least one arc detection interval data corresponding to each of the at least one initial arc detection data to obtain frequency domain arc detection data corresponding to each of the at least one initial arc detection data; and normalizing the at least one frequency domain arc detection data to obtain at least one arc detection data.

[0159] Figure 6 A current signal waveform diagram according to an embodiment of the present disclosure is illustrated schematically.

[0160] like Figure 6 As shown, compared with AC systems, photovoltaic systems exhibit strong current randomness, lack periodicity, have smaller amplitudes, and may even resemble current waveforms under certain interference conditions. The current signal of an arc fault is difficult to distinguish from a normal current signal based solely on time-domain characteristics such as the original current. Directly using time-domain waveforms as identification features can easily lead to misjudgments. Therefore, frequency-domain characteristic analysis is performed on the DC arc fault current.

[0161] If a Fourier transform is directly performed on the raw DC arc fault signal acquired by the oscilloscope, a lot of tiny current change information will be lost, resulting in an inaccurate spectrum. Therefore, before performing feature analysis on the fault arc current, it is necessary to determine a reasonably sized time window to segment the data and obtain at least one arc detection interval data corresponding to at least one initial arc detection data.

[0162] Figure 7A The illustration shows a frequency domain characteristic analysis spectrum of a current signal with a preset time window of 2ms according to an embodiment of the present disclosure.

[0163] Figure 7B The illustration shows a frequency domain characteristic analysis spectrum of a current signal with a preset time window of 10ms according to an embodiment of the present disclosure.

[0164] Figure 7C The illustration shows a frequency domain characteristic analysis spectrum of a current signal with a preset time window of 50ms according to an embodiment of the present disclosure.

[0165] like Figures 7A-7C As shown, to select an appropriate time window size, preset time windows of 2ms, 10ms, and 50ms were used to analyze the same 8A DC arc fault current signal. The frequency range of the DC arc fault signal spectrum is 1kHz to 120kHz. As can be seen from the figure, the smaller the window, the clearer the details displayed in the spectrum results, and the more obvious the characteristic differences between the current spectrum of the arc fault and the current spectrum of the normal current, which is more conducive to the identification of the arc fault detection model. However, too small a window also increases the real-time requirements of the arc fault detection model, placing higher demands on the hardware's computing power. Therefore, a 10ms time window can be selected to segment at least one initial arc detection data point.

[0166] According to embodiments of this disclosure, the Discrete Fourier Transform (DWT) can be used to discretize the DC arc fault current signal in the frequency domain, converting time-domain sampling into frequency-domain sampling, thereby providing the frequency domain characteristics of the DC arc fault signal. The DWT divides the entire DC arc fault signal into several equal-length small signals through a preset time window, thus obtaining sufficient detailed information about the signal. The DWT is computationally simple, has a low application threshold, and can effectively distinguish the characteristics of arc detection data. The processing result only contains frequency domain and amplitude information, requiring low information processing capabilities from the machine learning model. While the Short-Time Fourier Transform (SFT) and Wavelet Transform can also distinguish the characteristics of arc detection data, they are computationally more complex, have a higher application threshold, and are unsuitable for tasks with high real-time requirements such as arc fault detection.

[0167] According to embodiments of this disclosure, the current signals in at least one arc detection interval data corresponding to at least one initial arc detection data are discretized in the frequency domain after undergoing a discrete Fourier transform, thereby enabling the study of spectral characteristics. Assuming a current signal x(n) is a finite-length sequence of length N, the expression for x(n) is:

[0168]

[0169] Analyzing the current signal x(n) using the Discrete Fourier Transform, we find that X(k) is still a finite-length sequence of length N. The transformation process can be expressed as:

[0170]

[0171] According to embodiments of this disclosure, at least one frequency domain arc detection data is obtained by performing a Discrete Fourier Transform on at least one arc detection interval data corresponding to each of at least one initial arc detection data. Since arc characteristics are mainly after 3kHz and low-frequency harmonic interference exists in the frequency band below 3kHz, which can easily affect arc fault identification, at least one frequency domain arc detection data in the 3-125kHz range can be normalized to obtain at least one arc detection data. The normalization formula is as follows:

[0172]

[0173] Among them, X min X is the minimum value among at least one frequency domain arc detection data. max It is the maximum value in at least one frequency domain arc detection data.

[0174] According to embodiments of this disclosure, at least one arc detection data set can be randomly divided into a training set of 75%, a validation set of 10%, and a test set of 15% using a sample hold-out method.

[0175] According to embodiments of this disclosure, the training method for the above-mentioned arc fault detection model further includes: visualizing at least one target arc detection data to obtain a visualization result.

[0176] Adjust at least one target arc detection data based on the visualization results.

[0177] According to embodiments of this disclosure, at least one target arc detection data can be visualized using t-SNE, a dimensionality reduction technique in manifold learning, to obtain visualization results.

[0178] According to embodiments of this disclosure, at least one initial arc detection data can also be visualized to obtain an initial visualization result.

[0179] According to embodiments of this disclosure, at least one frequency domain arc detection data can also be visualized to obtain frequency domain visualization results.

[0180] Figure 8A A schematic diagram illustrating the visualization results of target arc detection data according to an embodiment of the present disclosure is shown.

[0181] Figure 8B A schematic diagram illustrating the initial visualization results of initial arc detection data according to an embodiment of the present disclosure is shown.

[0182] Figure 8CA schematic diagram illustrating the frequency domain visualization results of frequency domain arc detection data according to an embodiment of the present disclosure is shown.

[0183] like Figures 8A-8C As shown, in the visualization results of the target arc detection data, both the target arc detection data with arc faults and the normal target arc detection data exhibit large clusters. The initial visualization results of the initial arc detection data show the distribution of the initial arc detection data; due to its randomness, it is difficult to distinguish whether the initial arc detection data has an arc fault. The frequency domain visualization results of the frequency domain arc detection data show slight overlap but still do not exhibit good clustering.

[0184] According to embodiments of this disclosure, photovoltaic power plants are typically installed in open rooftop environments, where shading of the photovoltaic panels can affect fluctuations in the generated current. Current variations also occur during the inverter's startup process.

[0185] Figure 8D A schematic diagram illustrating the current changes during the inverter startup process according to an embodiment of the present disclosure is shown.

[0186] Figure 8D The current change process during inverter startup is shown.

[0187] Both shading and inverter startup can affect the detection of arc faults. A dataset of 60 samples was compiled for testing under both shading and inverter startup conditions. The performance of the interference immunity test is summarized in Table 3.

[0188] Table 3

[0189]

[0190] According to embodiments of this disclosure, in inverter startup interference testing, the unoptimized arc fault resulted in four false positives, causing the accuracy to drop to 93.3%. This is because both inverter startup and shading alter the current magnitude of the photovoltaic system. Transient changes in the time domain also affect frequency domain characteristics, causing interference features to overlap with arc features, leading to misjudgments by the arc fault detection model. In contrast, the current changes rapidly during inverter startup, and the power electronic devices in the inverter also generate switching noise that interferes with arc fault detection, making the unoptimized arc features less resistant to interference during inverter startup. This result verifies the effectiveness of the target arc fault detection model in interpretably extracting arc features.

[0191] Figure 9 A flowchart illustrating an arc fault detection method according to an embodiment of the present disclosure is shown schematically.

[0192] like Figure 9 As shown, the arc fault detection method includes operations S910 to S920.

[0193] The S910 is used to acquire data on the electric arc to be detected.

[0194] When operating the S920, the arc data to be detected is input into the target arc fault detection model, and the arc fault detection result is output.

[0195] According to embodiments of this disclosure, the target arc fault detection model is trained using the training method for arc fault detection models according to embodiments of this disclosure. By inputting the arc data to be detected into the target arc fault detection model, the arc fault detection result for the arc data to be detected can be output.

[0196] According to embodiments of this disclosure, the accuracy of arc fault detection can be improved by using the target arc fault detection model of this disclosure.

[0197] Based on the training method of the above-mentioned arc fault detection model, this disclosure also provides a training device for the arc fault detection model. The following will be combined with... Figure 10 The device is described in detail.

[0198] Figure 10 A schematic block diagram of a training apparatus for an arc fault detection model according to an embodiment of the present disclosure is shown.

[0199] like Figure 10 As shown, the training device 1000 for the arc fault detection model in this embodiment includes a first obtaining module 1010, a determining module 1020, and a second obtaining module 1030.

[0200] The first obtaining module 1010 is used to process at least one arc detection data using an attention module corresponding to the first arc fault detection model, to obtain the frequency band importance corresponding to each of the at least one arc detection data, wherein the frequency band importance characterizes the importance of the frequency band corresponding to the arc detection data. In one embodiment, the first obtaining module 1010 can be used to perform the operation S210 described above, which will not be repeated here.

[0201] The determining module 1020 is used to determine at least one target arc detection data from at least one arc detection data using at least one frequency band importance. In one embodiment, the determining module 1020 may be used to perform the operation S220 described above, which will not be repeated here.

[0202] The second obtaining module 1030 is used to train a second arc fault detection model using at least one target arc detection data to obtain a target arc fault detection model. In one embodiment, the second obtaining module 1030 can be used to perform the operation S230 described above, which will not be repeated here.

[0203] According to embodiments of this disclosure, the attention module includes a first pooling layer, a convolutional layer, and a first activation layer;

[0204] The first obtaining module 1010, used to process at least one arc detection data using an attention module corresponding to the first arc fault detection model, and to obtain the frequency band importance corresponding to each of the at least one arc detection data, includes:

[0205] The first submodule is used to process at least one arc detection data using a first pooling layer corresponding to the first arc fault detection model, and to obtain first pooling data corresponding to each of the at least one arc detection data.

[0206] The second submodule is used to process at least one first pooling data using a convolutional layer corresponding to the first arc fault detection model to obtain at least one convolutional data; and

[0207] The third submodule is used to process at least one convolutional data using the first activation layer corresponding to the first arc fault detection model, and to obtain the frequency band importance corresponding to each of the at least one arc detection data.

[0208] According to embodiments of this disclosure, the attention module includes a second pooling layer, a third pooling layer, a first fully connected layer, and a second activation layer;

[0209] The first obtaining module 1010, used to process at least one arc detection data using an attention module corresponding to the first arc fault detection model, and to obtain the frequency band importance corresponding to each of the at least one arc detection data, includes:

[0210] The fourth submodule is used to process at least one arc detection data using the second pooling layer corresponding to the first arc fault detection model, and to obtain the second pooling data corresponding to each of the at least one arc detection data.

[0211] The fifth submodule is used to process at least one arc detection data using the third pooling layer corresponding to the first arc fault detection model, and to obtain the third pooling data corresponding to each of the at least one arc detection data.

[0212] The sixth submodule is used to process the second pooling data and third pooling data corresponding to each of the at least one arc detection data using the first fully connected layer corresponding to the first arc fault detection model, to obtain fully connected data corresponding to each of the at least one arc detection data; and

[0213] The seventh submodule is used to process the fully connected data corresponding to at least one arc detection data using the second activation layer corresponding to the first arc fault detection model, and to obtain the frequency band importance corresponding to at least one arc detection data.

[0214] According to embodiments of this disclosure, the attention module includes a second fully connected layer.

[0215] According to embodiments of this disclosure, a determining module 1020 for determining at least one target arc detection data from at least one arc detection data using at least one frequency band importance includes:

[0216] The first determining submodule is used to determine the arc detection dataset corresponding to each of the at least one arc detection data according to the frequency band corresponding to each of the at least one arc detection data, wherein the arc detection dataset corresponds to the frequency band interval;

[0217] The second determining submodule is configured to determine at least one target frequency band interval from at least one frequency band interval based on at least one frequency band importance corresponding to each of at least one arc detection dataset; and

[0218] The third determining submodule is used to determine the arc detection data in the arc detection dataset corresponding to at least one target frequency band interval as the target arc detection data.

[0219] According to embodiments of this disclosure, a second determining submodule for determining at least one target frequency band interval from at least one frequency band interval based on at least one frequency band importance corresponding to each of at least one arc detection dataset includes:

[0220] The first determining unit is configured to determine the comprehensive frequency band importance corresponding to each of the at least one arc detection dataset based on the frequency band importance corresponding to each of the at least one arc detection dataset; and

[0221] The second determining unit is used to determine at least one target frequency band interval from at least one frequency band interval based on the comprehensive frequency band importance corresponding to each of at least one arc detection dataset.

[0222] According to embodiments of this disclosure, a first determining unit for determining the comprehensive frequency band importance corresponding to each of the at least one arc detection dataset based on at least one frequency band importance corresponding to each of the at least one arc detection datasets includes:

[0223] The first determining subunit is configured for each arc detection dataset in at least one arc detection dataset.

[0224] The second determining subunit is used to determine the statistical importance corresponding to the arc detection dataset based on the importance of at least one frequency band corresponding to the arc detection dataset; and

[0225] The third determining subunit is used to determine the statistical importance corresponding to the arc detection dataset as the comprehensive frequency band importance corresponding to the arc detection dataset.

[0226] According to embodiments of this disclosure, the training apparatus for the above-mentioned arc fault detection model further includes:

[0227] The third module is used to divide at least one initial arc detection data according to a preset time window to obtain at least one arc detection interval data corresponding to each of the at least one initial arc detection data.

[0228] The fourth module is used to perform frequency domain transformation on at least one arc detection interval data corresponding to at least one initial arc detection data, to obtain frequency domain arc detection data corresponding to each of the at least one initial arc detection data; and

[0229] The fifth module is used to normalize at least one frequency domain arc detection data to obtain at least one arc detection data.

[0230] Figure 11 A schematic block diagram of an arc fault detection device according to an embodiment of the present disclosure is shown.

[0231] like Figure 11 As shown, the arc fault detection device 1100 of this embodiment includes an acquisition module 1110 and an output module 1120.

[0232] The acquisition module 1110 is used to acquire the data of the electric arc to be detected. In one embodiment, the acquisition module 1110 can be used to perform the operation S910 described above, which will not be repeated here.

[0233] The output module 1120 is used to input the arc data to be detected into the target arc fault detection model and output the arc fault detection result. The target arc fault detection model is trained using the training method for the arc fault detection model according to embodiments of this disclosure. In one embodiment, the output module 1120 can be used to perform the operation S920 described above, which will not be repeated here.

[0234] According to embodiments of this disclosure, any plurality of modules among the first obtaining module 1010, determining module 1020, and second obtaining module 1030 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of this disclosure, at least one of the first obtaining module 1010, determining module 1020, and second obtaining module 1030 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented by any other reasonable means of integrating or packaging circuitry, or implemented in any one of software, hardware, and firmware methods, or in a suitable combination of any of these. Alternatively, at least one of the first obtaining module 1010, determining module 1020, and second obtaining module 1030 can be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.

[0235] Figure 12 A block diagram of an electronic device suitable for implementing a training method for an arc fault detection model according to an embodiment of the present disclosure is shown schematically.

[0236] like Figure 12 As shown, an electronic device 1200 according to an embodiment of the present disclosure includes a processor 1201, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1202 or a program loaded from a storage portion 1208 into a random access memory (RAM) 1203. The processor 1201 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1201 may also include onboard memory for caching purposes. The processor 1201 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0237] RAM 1203 stores various programs and data required for the operation of electronic device 1200. Processor 1201, ROM 1202, and RAM 1203 are interconnected via bus 1204. Processor 1201 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 1202 and / or RAM 1203. It should be noted that the programs may also be stored in one or more memories other than ROM 1202 and RAM 1203. Processor 1201 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0238] According to embodiments of this disclosure, the electronic device 1200 may further include an input / output (I / O) interface 1205, which is also connected to the bus 1204. The electronic device 1200 may also include one or more of the following components connected to the input / output (I / O) interface 1205: an input section 1206 including a keyboard, mouse, etc.; an output section 1207 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1208 including a hard disk, etc.; and a communication section 1209 including a network interface card such as a LAN card, modem, etc. The communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to the input / output (I / O) interface 1205 as needed. A removable medium 1211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1210 as needed so that computer programs read from it can be installed into the storage section 1208 as needed.

[0239] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0240] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0241] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A method for training an arc fault detection model, comprising: processing at least one arc detection data using an attention module corresponding to a first arc fault detection model to obtain a frequency band importance degree corresponding to each of the at least one arc detection data, wherein the frequency band importance degree represents an importance degree of a frequency band corresponding to the arc detection data; determining at least one target arc detection data from the at least one arc detection data using the at least one frequency band importance degree; and training a second arc fault detection model using the at least one target arc detection data to obtain a target arc fault detection model; wherein the attention module comprises a first pooling layer, a convolution layer, and a first activation layer; wherein the processing at least one arc detection data using an attention module corresponding to a first arc fault detection model to obtain a frequency band importance degree corresponding to each of the at least one arc detection data comprises: processing the at least one arc detection data using a first pooling layer corresponding to the first arc fault detection model to obtain first pooling data corresponding to each of the at least one arc detection data; processing at least one of the first pooling data using a convolution layer corresponding to the first arc fault detection model to obtain at least one convolution data; and processing the at least one convolution data using a first activation layer corresponding to the first arc fault detection model to obtain a frequency band importance degree corresponding to each of the at least one arc detection data.

2. The method of claim 1, wherein, the attention module comprises a second pooling layer, a third pooling layer, a first full connection layer, and a second activation layer; wherein the processing at least one arc detection data using an attention module corresponding to a first arc fault detection model to obtain a frequency band importance degree corresponding to each of the at least one arc detection data comprises: processing the at least one arc detection data using a second pooling layer corresponding to the first arc fault detection model to obtain second pooling data corresponding to each of the at least one arc detection data; processing the at least one arc detection data using a third pooling layer corresponding to the first arc fault detection model to obtain third pooling data corresponding to each of the at least one arc detection data; processing the second pooling data and the third pooling data corresponding to each of the at least one arc detection data using a first full connection layer corresponding to the first arc fault detection model to obtain full connection data corresponding to each of the at least one arc detection data; and processing the full connection data corresponding to each of the at least one arc detection data using a second activation layer corresponding to the first arc fault detection model to obtain a frequency band importance degree corresponding to each of the at least one arc detection data.

3. The method of claim 1, wherein, the attention module comprises a second full connection layer.

4. The method of claim 1 or 2, wherein, the determining at least one target arc detection data from the at least one arc detection data using the at least one frequency band importance degree comprises: determining an arc detection data set corresponding to each of the at least one arc detection data according to a frequency band corresponding to each of the at least one arc detection data, wherein the arc detection data set corresponds to a frequency band interval; determine at least one target frequency band interval from the at least one frequency band interval according to at least one frequency band importance corresponding to each of the at least one arc detection data set; and determine arc detection data in the arc detection data set corresponding to each of the at least one target frequency band interval as the target arc detection data.

5. The method of claim 4, wherein, The determining at least one target frequency band interval from the at least one frequency band interval according to at least one frequency band importance corresponding to each of the at least one arc detection data set comprises: determine at least one target frequency band interval from the at least one frequency band interval according to at least one frequency band importance corresponding to each of the at least one arc detection data set; and determine at least one target frequency band interval from the at least one frequency band interval according to at least one frequency band importance corresponding to each of the at least one arc detection data set; and 6. The method of claim 3, wherein, The determining at least one target frequency band interval from the at least one frequency band interval according to at least one frequency band importance corresponding to each of the at least one arc detection data set comprises: for each of the at least one arc detection data set, determine a statistical importance corresponding to the arc detection data set according to at least one frequency band importance corresponding to the arc detection data set; and determine the statistical importance corresponding to the arc detection data set as the comprehensive frequency band importance corresponding to the arc detection data set.

7. The method of claim 1 or 2, further comprising: segmenting at least one initial arc detection data according to a preset time window to obtain at least one arc detection interval data corresponding to each of the at least one initial arc detection data; performing frequency domain transformation on the at least one arc detection interval data corresponding to each of the at least one initial arc detection data to obtain frequency domain arc detection data corresponding to each of the at least one initial arc detection data; and performing normalization processing on the at least one frequency domain arc detection data to obtain the at least one arc detection data.

8. The method of claim 1 or 2, further comprising: performing visualization processing on the at least one target arc detection data to obtain a visualization result; and adjusting the at least one target arc detection data according to the visualization result.

9. An arc fault detection method, comprising: obtaining to-be-detected arc data; and inputting the to-be-detected arc data into a target arc fault detection model to output an arc fault detection result; wherein the target arc fault detection model is trained by the method of any one of claims 1-8.

10. An electronic device, comprising: one or more processors; a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors perform the method of any one of claims 1-8. ​

Images