Fault arc detection method, system, and storage medium

Abstract

The application discloses a kind of fault arc detection method, system and storage medium, method includes the current waveform of the line to be detected acquisition;The current waveform is enhanced and is handled, and current enhancement signal is obtained;The current enhancement signal is input to the arc detection model that is trained in advance, and the arc fault detection result corresponding to the current waveform is obtained;The arc detection model includes sequentially connected input layer, feature extraction layer and output layer, the feature extraction layer includes a plurality of residual modules connected in turn, and the output of each residual module is connected with compression excitation module, the output of the residual module and the output of the compression excitation module are linearly superimposed, and then linearly superimposed with the input data of the residual module and output.The application can significantly improve the accuracy of fault arc detection.

Description

Technical Field

[0001] This invention relates to the field of arc detection technology, and specifically to a fault arc detection method, system, and storage medium. Background Technology

[0002] With the continuous development and widespread use of power electronic equipment, electricity utilization has become more convenient. However, due to the increasing number of electrical devices in low-voltage power distribution systems, the associated fire hazards have also greatly increased. Low-voltage power distribution systems use a variety of electrical equipment; if conductors are damaged, aged, or subjected to prolonged overload, fault arcs can be generated in the lines. The temperature at the center of these fault arcs can reach over 5000℃, easily causing electrical fires or even explosions. Electrical fires account for over 30% of all fires nationwide, and the main electrical fault sources causing these fires include short circuits, overloads, leakage current, and arcing faults. Currently, circuit breakers and fuses can effectively protect against short circuits and overloads, respectively. However, fault arcs are characterized by strong randomness, concealment, and complexity, and their detection accuracy remains a critical technical challenge.

[0003] Currently, most arc fault detectors on the market use a fixed threshold method to detect arc characteristics. When the detected characteristic exceeds the set threshold, an arc fault is considered to exist in the circuit. However, due to the inherent electromagnetic interference in the circuit and the variety of load types, it is difficult to find an optimal threshold that is universally suitable for various combinations of different load types.

[0004] In recent years, with the rise of artificial intelligence, some new detection methods, such as neural network models, have been widely used in the field of arc detection. However, these methods suffer from problems such as complex structures, high computational costs, and low recognition accuracy. For example, in related technologies, the patent application document CN115498956A proposes a photovoltaic array series arc fault diagnosis method. This method trains a deep convolutional generative adversarial network (DCGAN) using a dataset of two-dimensional feature images of the fault time-series signal. The DCGAN then uses data augmentation methods to expand the arc fault samples. Finally, a convolutional neural network (CNN) classification model is trained using the dataset of two-dimensional feature images and the augmented arc fault samples. This approach uses a generative adversarial network to generate additional data when the data sample size is limited, achieving data augmentation and expanding the dataset, rather than enhancing the current waveform itself.

[0005] The patent application CN114152844A proposes an arc detection method based on a genetic algorithm-optimized residual network model. This method constructs a composite residual network model using a preprocessed dataset and optimizes it using a genetic algorithm to achieve arc detection. However, the composite residual network model constructed in this scheme does not adequately learn the high-frequency distortion characteristics of the arc and is easily affected by the temporal randomness of arc generation. Summary of the Invention

[0006] The technical problem to be solved by this invention is how to improve the detection accuracy of fault arcs.

[0007] The present invention solves the above-mentioned technical problems through the following technical means:

[0008] In a first aspect, the present invention proposes a method for detecting fault arcs, the method comprising the following steps:

[0009] Acquire the current waveform of the circuit under test;

[0010] The current waveform is enhanced to obtain a current enhancement signal;

[0011] The current enhancement signal is input into a pre-trained arc detection model to obtain an arc fault detection result corresponding to the current waveform;

[0012] The arc detection model includes an input layer, a feature extraction layer, and an output layer connected in sequence. The feature extraction layer includes several residual modules connected in sequence, and the output of each residual module is connected to a compression excitation module. The output of the residual module and the output of the compression excitation module are linearly superimposed, and then linearly superimposed with the input data of the residual module for output.

[0013] Further, the enhancement processing of the current waveform to obtain the current enhancement signal includes:

[0014] The current waveform is enhanced using a parametric Wiener filter to obtain the enhanced current signal. The formula for the parametric Wiener filter is as follows:

[0015]

[0016] In the formula: P xx (ω k P is the power spectrum of the pure signal. nn (ω k ξ is the power spectrum of the noise signal. k It is the prior signal-to-noise ratio, and α and β are adjustable parameters.

[0017] Furthermore, the compression excitation module includes a channel attention mechanism network and a spatial attention mechanism network connected in sequence. The channel attention mechanism network adopts a global max pooling operation, and the spatial attention mechanism network adopts a global average pooling operation.

[0018] The output of the residual module is connected to the input of the channel attention mechanism network.

[0019] Furthermore, after inputting the current enhancement signal into a pre-trained arc detection model to obtain an arc fault detection result corresponding to the current waveform, the method further includes:

[0020] If the detection result indicates that an arc fault has occurred, then the fault count is incremented by one; otherwise, no counting operation is performed.

[0021] When the number of accumulated faults exceeds the set number, the line is disconnected via the disconnection device.

[0022] Furthermore, before inputting the current-enhancing signal into a pre-trained arc detection model to obtain the arc fault detection result corresponding to the current waveform, the method further includes:

[0023] Acquire the current signal from the fault arc test platform;

[0024] The current signal is enhanced using a parametric Wiener filter to obtain an enhanced dataset;

[0025] The arc detection model is trained using the augmented dataset to obtain a trained arc detection model.

[0026] Furthermore, the fault arc test platform includes a fault arc generating device, an AC power supply, and different types of loads connected in series. A switch is connected between the fault arc generating device and the loads, and a current acquisition device is connected between the AC power supply and the fault arc generating device.

[0027] The current acquisition device is used to acquire the current signal of the main circuit of the fault arc test platform. The current signal includes the main circuit current signal in the normal state and the main circuit current signal in the fault state.

[0028] Furthermore, after acquiring the current signal from the fault arc experimental platform, the method further includes:

[0029] The current signal is segmented according to the power frequency cycle to obtain an arc data set;

[0030] A high-pass filter is used to extract the circuit noise from the arc data in the arc dataset.

[0031] Based on the arc dataset and the circuit noise, a parametric Wiener filter is constructed.

[0032] Further, the step of training the arc detection model using the augmented dataset to obtain a trained arc detection model includes:

[0033] The augmented dataset is divided into a training dataset and a test dataset according to a certain ratio;

[0034] The training dataset is used as input to the arc detection model. The weights and biases of the convolutional and fully connected layers in the network are updated iteratively to obtain the set of parameter values ​​that minimize the loss function, which are then used as the model parameters. The loss function used is the cross-entropy loss function, expressed by the following formula:

[0035]

[0036] In the formula: m represents the batch size, n represents the data dimension, and P(x i，j Q(x) represents the true distribution of the input data. i，j ) represents the probability distribution of network predictions;

[0037] The arc detection model was tested using the test dataset, and the hyperparameters were adjusted until the test requirements were met.

[0038] Secondly, the present invention proposes a fault arc detection system, the system comprising:

[0039] The acquisition module is used to acquire the current waveform of the circuit under test;

[0040] An enhancement module is used to enhance the current waveform to obtain a current enhancement signal;

[0041] The detection module is used to input the current enhancement signal into a pre-trained arc detection model to obtain an arc fault detection result corresponding to the current waveform;

[0042] The arc detection model includes an input layer, a feature extraction layer, and an output layer connected in sequence. The feature extraction layer includes several residual modules connected in sequence, and the output of each residual module is connected to a compression excitation module. The output of the residual module and the output of the compression excitation module are linearly superimposed, and then linearly superimposed with the input data of the residual module for output.

[0043] Thirdly, the present invention proposes a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the fault arc detection method as described above.

[0044] The advantages of this invention are:

[0045] (1) The present invention first performs data enhancement processing on the current waveform of the line to be detected, and inputs the obtained current enhancement signal into the arc detection model for processing to obtain the corresponding arc fault detection result; by enhancing the current waveform, the influence of circuit noise caused by line aging, power grid pollution and electromagnetic leakage of electrical equipment on the fault arc characteristics is reduced, and the stability and accuracy of detection are improved; by using the feature self-extraction capability of convolutional neural network, it can meet the detection of multiple load types and expand the actual working conditions that the fault arc detection device can detect; by using residual connection to construct deep neural network, the gradient vanishing and gradient explosion phenomena of deep network are alleviated; by adding compression excitation module, the model can improve the extraction of fault arc distortion features without increasing the model complexity, and significantly improve the accuracy of fault arc detection.

[0046] (2) The channel attention mechanism network adopts global max pooling operation to fully pay attention to the distortion characteristics of fault arc; the spatial attention mechanism network adopts global average pooling operation to reduce the impact of the randomness of fault arc generation time on the detection results.

[0047] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0048] Figure 1 This is a schematic flowchart of the fault arc detection method proposed in the embodiments of the present invention;

[0049] Figure 2 This is a schematic diagram of a local structure of the Arc-1DCNN arc detection model in an embodiment of the present invention;

[0050] Figure 3 This is a structural detail diagram illustrating the Arc-1DCNN arc detection model in this embodiment of the invention;

[0051] Figure 4 This is a schematic diagram of the overall structure of the Arc-1DCNN arc detection model in this embodiment of the invention;

[0052] Figure 5 This is a schematic diagram of the training process of the arc detection model proposed in the embodiments of the present invention;

[0053] Figure 6 This is a schematic diagram of the experimental platform circuit for generating fault arcs in an embodiment of the present invention;

[0054] Figure 7 This is a spectrum diagram of the noise from incandescent lamp malfunction and normal operation in an embodiment of the present invention;

[0055] Figure 8 This refers to the hyperparameter settings of the Arc-1DCNN arc detection model in this embodiment of the invention;

[0056] Figure 9 This is the training error variation curve of the Arc-1DCNN model in the embodiment of the present invention;

[0057] Figure 10 This is the training accuracy variation curve of the Arc-1DCNN model in the embodiment of the present invention;

[0058] Figure 11 This describes the performance of the Arc-1DCNN arc detection model in this embodiment of the invention for detecting fault arcs.

[0059] Figure 12 This is a schematic diagram of the fault arc detection system proposed in an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] like Figure 1 As shown, the first embodiment of the present invention proposes a fault arc detection method, the method comprising the following steps:

[0062] S10. Acquire the current waveform of the circuit under test;

[0063] It should be noted that in this embodiment, the current waveform of the circuit under test is collected at a fixed time interval. The time interval can be set to 1 second, or it can be set to other values ​​according to actual needs. No specific limitation is made here.

[0064] S20. The current waveform is enhanced to obtain a current enhancement signal;

[0065] It should be noted that by enhancing the current waveform, the influence of circuit noise caused by factors such as line aging, power grid pollution, and electromagnetic leakage from electrical equipment on the characteristics of the fault arc is reduced, thereby improving the stability and accuracy of the detection.

[0066] S30. Input the current enhancement signal into the pre-trained arc detection model to obtain the arc fault detection result corresponding to the current waveform;

[0067] The arc detection model includes an input layer, a feature extraction layer, and an output layer connected in sequence. The feature extraction layer includes several residual modules connected in sequence, and the output of each residual module is connected to a compression excitation module. The output of the residual module and the output of the compression excitation module are linearly superimposed, and then linearly superimposed with the input data of the residual module for output.

[0068] It should be noted that the arc detection model constructed in this embodiment can meet the detection of multiple load types by leveraging the feature self-extraction capability of convolutional neural networks, thus expanding the actual working condition range that the fault arc detection device can detect. The deep neural network is constructed using residual connections to alleviate the gradient vanishing and gradient explosion phenomena that occur in deep networks. By adding a compression excitation module, the model can improve the extraction of fault arc distortion features with almost no increase in model complexity, thus significantly improving the accuracy of fault arc detection.

[0069] In one embodiment, step S20: enhancing the current waveform to obtain a current enhancement signal includes:

[0070] The current waveform is enhanced using a parametric Wiener filter to obtain the enhanced current signal. The formula for the parametric Wiener filter is as follows:

[0071]

[0072] In the formula: P xx (ω k Pn is the power spectrum of the pure signal. n (ω k ξ is the power spectrum of the noise signal. k It is the prior signal-to-noise ratio, and α and β are adjustable parameters that can be flexibly set based on prior knowledge of the electric arc.

[0073] This embodiment performs time-frequency analysis on the original signal and uses Wiener filtering to reduce the impact of noise on the arc characteristics in order to achieve data augmentation. Using a parametric Wiener filter, parameters can be flexibly set according to prior knowledge of the arc. After parametric Wiener filtering, the noise inherent in the circuit can be suppressed without affecting the arc characteristics. This expands the difference between the fault current signal and the normal current signal, which can improve the detection accuracy of the subsequent network model.

[0074] In one embodiment, the compressed excitation module includes a channel attention mechanism network and a spatial attention mechanism network connected in sequence. The channel attention mechanism network employs global max pooling, and the spatial attention mechanism network employs global average pooling.

[0075] The output of the residual module is connected to the input of the channel attention mechanism network.

[0076] Furthermore, the channel attention mechanism network consists of several convolutional layers and global max pooling layers. Each convolutional layer is followed by an activation function, and ReLU and sigmoid activation functions are alternately followed by the convolutional layers.

[0077] The channel attention mechanism network consists of several convolutional layers and global average pooling layers. Each convolutional layer is followed by an activation function, and ReLU and sigmoid activation functions are alternately followed by the convolutional layers.

[0078] It should be noted that the residual network model proposed in this embodiment is specifically modified according to the task objective. While ensuring that the model has a certain level of complexity, some network layers are reduced to reduce the model complexity. Finally, the model proposed in this scheme is based on a convolutional neural network, uses a residual structure, and embeds channel and spatial compression excitation modules to ensure that the network can fully learn the distortion characteristics of the electric arc and reduce the influence of the temporal randomness of the electric arc generation on the model.

[0079] like Figures 2 to 4 As shown, the Arc-1DCNN arc detection model constructed in this embodiment specifically includes an input layer, a feature extraction layer, and an output layer. The input layer includes three convolutional layers (Convld) and a max pooling layer (MaxPoolld) connected in sequence. Each convolutional layer (Convld) is followed by a normalization layer (LN). The output feature size, kernel size, and number of kernels in each input layer are shown in [the diagram]. Figure 3 As shown in Module I.

[0080] The feature extraction layer includes a first residual module, a second residual module, a third residual module, and a fourth residual module connected in sequence. The residual modules are used to perform nonlinear transformation on the input data and then linearly superimpose it with the input data for output. The number of the first residual module, the second residual module, the third residual module, and the fourth residual module can be set according to actual needs, and this embodiment does not impose a specific limitation.

[0081] The first, second, third, and fourth residual modules each include three sequentially connected convolutional layers (Conv1d) and a compression activation module (SE-Block), with each Conv1d layer followed by a batch normalization (BN) layer. The output feature size, kernel size, and number of kernels for each layer in the first residual module are detailed below. Figure 3 As shown in Module II; the output feature size, kernel size, and number of kernels for each layer in the second residual module are shown in [reference needed]. Figure 3As shown in Module III; the output feature size, kernel size, and number of kernels for each layer in the third residual module are shown in [reference needed]. Figure 3 As shown in Module IV; the output feature size, kernel size, and number of kernels for each layer in the fourth residual module are shown in [reference needed]. Figure 3 As shown in module V.

[0082] The output layer includes an adaptive 1D pooling layer (AdaptiveAvgPool1d), a flatten layer, a fully connected linear layer, and an output layer (Softmax). See [link to parameters] for details. Figure 3 As shown in module VI.

[0083] Specifically, such as Figure 2 As shown, the compressed excitation module SE-Block includes a channel attention mechanism network and a spatial attention mechanism network connected in sequence. The output of the last convolutional layer in the residual module is connected to the channel attention mechanism network, and the output of the channel attention mechanism network is connected to the spatial attention mechanism network. The output of the spatial attention mechanism network is linearly superimposed with the last convolutional layer in the residual module and then linearly superimposed with the input data of the first convolutional layer in the residual module as the input of the next residual module.

[0084] It should be noted that the Arc-1DCNN arc detection model constructed in this embodiment is based on a convolutional neural network. It uses residual connections to alleviate the vanishing and exploding gradient phenomena that occur in deep networks, and uses one-dimensional convolutional kernels to extract abstract features at different levels of the data. It mainly consists of an input layer, alternating convolutional and pooling layers, fully connected layers, alternating batch normalization and layer normalization layers, and an output layer. The input layer is used to preprocess the input data. The convolutional and pooling layers extract arc features. The fully connected layers, to some extent, preserve the model's complexity, mapping the learned features to the sample label space. The batch and layer normalization layers accelerate model convergence and improve the model's generalization performance. The output layer outputs the Arc-1DCNN's prediction of the input sample type.

[0085] The channel attention mechanism network employs global max pooling to fully consider the distortion characteristics of the fault arc; the spatial attention mechanism network employs global average pooling to reduce the impact of the randomness of the fault arc generation time on the detection results.

[0086] It should be noted that the selection of network layers and the order in which they are placed are not unique, and other similar network stacks are considered as supplements to this invention.

[0087] In one embodiment, after step S30: inputting the current enhancement signal to a pre-trained arc detection model to obtain an arc fault detection result corresponding to the current waveform, the method further includes the following steps:

[0088] If the detection result indicates that an arc fault has occurred, then the fault count is incremented by one; otherwise, no counting operation is performed.

[0089] When the number of accumulated faults exceeds the set number, the line is disconnected via the disconnection device.

[0090] It should be noted that the Arc-1DCNN model calculates the category label corresponding to the current data to determine whether an arc fault has occurred. If a fault occurs, the fault count is incremented by one; if no fault occurs, no operation is performed; when the accumulated fault count is greater than 14, the line is disconnected through the disconnection device.

[0091] This embodiment overcomes the difficulties of traditional detection methods that require subjective selection of arc characteristic quantities and characteristic thresholds. It eliminates the need for manual selection of arc characteristic quantities and the setting of characteristic thresholds, and is unaffected by the magnitude of the load current. Furthermore, it uses the main circuit current as input, which is easier to obtain compared to other signals such as voltage, temperature, and electromagnetic quantities, facilitating signal acquisition under actual working conditions.

[0092] like Figure 5 As shown, before step S30: inputting the current enhancement signal into the pre-trained arc detection model to obtain the arc fault detection result corresponding to the current waveform, the method further includes the following steps:

[0093] S1. Acquire the current signal of the fault arc test platform;

[0094] S2. The current signal is enhanced using a parametric Wiener filter to obtain an enhanced dataset;

[0095] S3. Use the augmented dataset to train the arc detection model to obtain a trained arc detection model.

[0096] In one embodiment, such as Figure 6 As shown, the fault arc test platform includes a fault arc generating device, an AC power supply, and different types of loads connected in series. A switch is connected between the fault arc generating device and the loads, and a current acquisition device is connected between the AC power supply and the fault arc generating device.

[0097] The current acquisition device is used to acquire the current signal of the main circuit of the fault arc test platform. The current signal includes the main circuit current signal in the normal state and the main circuit current signal in the fault state.

[0098] The current acquisition device includes a current transformer, a sampling resistor, and a signal acquisition card. The output of the current transformer is also connected to an oscilloscope.

[0099] The fault arc generating device is used to simulate fault arcs in actual power lines; the current transformer is used to sense the current signal of the main circuit and, together with the sampling resistor and the acquisition card, to collect current data; the oscilloscope is used to display the current waveform during the experiment in real time; and the different types of loads are used to simulate line faults under different operating conditions.

[0100] In this implementation case, different types of loads include: incandescent lamps, air conditioners, computers, and combinations of incandescent lamps and air conditioners. These include resistive loads, inductive loads, and switching power supplies. There are a total of eight groups of data: normal data for resistive loads, fault data for resistive loads, normal data for inductive loads, fault data for inductive loads, normal data for switching power supply loads, fault data for switching power supply loads, normal data for both resistive and inductive loads, and fault data for both resistive and inductive loads.

[0101] In one embodiment, after step S1: acquiring the current signal of the fault arc experimental platform, the method further includes the following steps:

[0102] The current signal is segmented according to the power frequency cycle to obtain an arc data set;

[0103] A high-pass filter is used to extract the circuit noise from the arc data in the arc dataset.

[0104] Based on the arc dataset and the circuit noise, a parametric Wiener filter is constructed.

[0105] This embodiment collects experimental data from eight groups at a 1-second sampling interval. The collected data is then divided according to the power frequency cycle to construct the original arc dataset. Specifically, the collected current signal is segmented according to the power frequency cycle (0.02s) to obtain the arc dataset. The time-frequency analysis refers to using Fast Fourier Transform for current signals of different load types, combining the spectrum and time-domain waveforms to analyze the graphical differences between different load types.

[0106] By performing time-frequency analysis on experimental data from different groups, circuit noise was extracted using a high-pass filter. Figure 7 The display shows the frequency spectrum of noise during incandescent lamp malfunction and normal operation. Based on arc-related theoretical knowledge and extensive experimental verification, a parametric Wiener filter is constructed using appropriate parameters to enhance the original current signal.

[0107] In one embodiment, step S3, which involves training the arc detection model using the augmented dataset to obtain a trained arc detection model, specifically includes the following steps:

[0108] The augmented dataset is divided into a training dataset and a test dataset according to a certain ratio;

[0109] The training dataset is used as input to the arc detection model. The weights and biases of the convolutional and fully connected layers in the network are updated iteratively to obtain the set of parameter values ​​that minimize the loss function, which are then used as the model parameters. The loss function used is the cross-entropy loss function, expressed by the following formula:

[0110]

[0111] In the formula: m represents the batch size, n represents the data dimension, and P(x i，j Q(x) represents the true distribution of the input data. i，j ) represents the probability distribution of network predictions;

[0112] The arc detection model was tested using the test dataset, and the hyperparameters were adjusted until the test requirements were met.

[0113] Specifically, in this embodiment, an enhanced dataset is constructed using the enhanced signal and divided into a training set, a validation set, and a test set in a 7:1:2 ratio; the arc detection model is trained using the training set; the parameters of the trained arc detection model are adjusted using the validation set; and the performance of the trained arc detection model is tested using the test set.

[0114] The augmented dataset is used as the training input to the Arc-1DCNN model, with output categories of normal and faulty. The network model is trained as follows: The Arc-1DCNN model uses Kaiming to initialize weights and biases, and uses the cross-entropy loss function as the optimization objective function. The training and validation sets of the augmented dataset are input into the Arc-1DCNN in batches, and the weights and biases of the convolutional and fully connected layers in the network are updated through continuous iteration. Finally, the set of parameter values ​​that minimize the loss function is obtained. The entire Arc-1DCNN model includes the network structure and its parameters. Its training process flowchart is as follows. Figure 5 As shown. During training, hyperparameters need to be adjusted appropriately based on the network model's performance. Adjusting the corresponding hyperparameters refers to adjusting the learning rate, batch size, training iteration cycle, and weight decay coefficient before training begins. In this implementation example, the hyperparameter settings for the Arc-1DCNN model are as follows: Figure 8 As shown. The hyperparameter settings are gradually established during network training, but are not the only solution; other similar hyperparameter settings are considered extensions of this invention. The training error variation of the Arc-1DCNN model is shown below. Figure 9 As shown, the test results are as follows: Figure 10 As shown in the figure. Specifically, the performance of the trained arc detection model for arc detection is as follows: Figure 11 As shown.

[0115] This embodiment uses the back gradient propagation method to optimize network parameters. In the iterative process of finding the optimal parameters of the model, two different optimizers are used in different iteration cycles to achieve the goal of fast and stable model learning.

[0116] like Figure 12 As shown, the second embodiment of the present invention proposes a fault arc detection system, the system comprising:

[0117] Acquisition module 10 is used to acquire the current waveform of the circuit under test;

[0118] Enhancement module 20 is used to enhance the current waveform to obtain a current enhancement signal;

[0119] The detection module 30 is used to input the current enhancement signal into a pre-trained arc detection model to obtain an arc fault detection result corresponding to the current waveform;

[0120] The arc detection model includes an input layer, a feature extraction layer, and an output layer connected in sequence. The feature extraction layer includes several residual modules connected in sequence, and the output of each residual module is connected to a compression excitation module. The output of the residual module and the output of the compression excitation module are linearly superimposed, and then linearly superimposed with the input data of the residual module for output.

[0121] This embodiment enhances the current waveform signal, reducing the impact of circuit noise caused by line aging, power grid pollution, and electromagnetic leakage from electrical equipment on the characteristics of fault arcs, thus improving the stability and accuracy of detection. By leveraging the feature extraction capability of convolutional neural networks, it can meet the detection requirements of multiple load types, expanding the actual operating conditions that the fault arc detection device can detect. Using residual connections to construct a deep neural network alleviates the gradient vanishing and gradient exploding phenomena that occur in deep networks. By adding a compressed excitation module, the model's extraction of fault arc distortion features can be improved with almost no increase in model complexity, significantly improving the accuracy of fault arc detection.

[0122] In one embodiment, the enhancement module 20 is specifically used for:

[0123] The current waveform is enhanced using a parametric Wiener filter to obtain the enhanced current signal. The formula for the parametric Wiener filter is as follows:

[0124]

[0125] In the formula: P xx (ω kP is the power spectrum of the pure signal. nn (ω k ξ is the power spectrum of the noise signal. k It is the prior signal-to-noise ratio, and α and β are adjustable parameters.

[0126] In one embodiment, the compressed excitation module includes a channel attention mechanism network and a spatial attention mechanism network connected in sequence. The channel attention mechanism network employs global max pooling, and the spatial attention mechanism network employs global average pooling.

[0127] The output of the residual module is connected to the input of the channel attention mechanism network.

[0128] In one embodiment, the system further includes a fault handling module, configured to:

[0129] If the detection result indicates that an arc fault has occurred, then the fault count is incremented by one; otherwise, no counting operation is performed.

[0130] When the number of accumulated faults exceeds the set number, the line is disconnected via the disconnection device.

[0131] In one embodiment, the system further includes a model training module, specifically comprising:

[0132] The sample data acquisition unit is used to acquire the current signal of the fault arc test platform;

[0133] The signal enhancement unit is used to enhance the current signal using a parametric Wiener filter to obtain an enhanced dataset;

[0134] The training unit is used to train the arc detection model using the augmented dataset to obtain a trained arc detection model.

[0135] In one embodiment, the fault arc test platform includes a fault arc generating device, an AC power supply, and different types of loads connected in series. A switch is connected between the fault arc generating device and the loads, and a current acquisition device is connected between the AC power supply and the fault arc generating device.

[0136] The current acquisition device is used to acquire the current signal of the main circuit of the fault arc test platform. The current signal includes the main circuit current signal in the normal state and the main circuit current signal in the fault state.

[0137] In one embodiment, the system further includes a time-domain analysis module, specifically used for:

[0138] The current signal is segmented according to the power frequency cycle to obtain an arc data set;

[0139] A high-pass filter is used to extract the circuit noise from the arc data in the arc dataset.

[0140] Based on the arc dataset and the circuit noise, a parametric Wiener filter is constructed.

[0141] In one embodiment, the training unit is specifically used for:

[0142] The augmented dataset is divided into a training dataset and a test dataset according to a certain ratio;

[0143] The training dataset is used as input to the arc detection model. The weights and biases of the convolutional and fully connected layers in the network are updated iteratively to obtain the set of parameter values ​​that minimize the loss function, which are then used as the model parameters. The loss function used is the cross-entropy loss function, expressed by the following formula:

[0144]

[0145] In the formula: m represents the batch size, n represents the data dimension, and P(x i，j Q(x) represents the true distribution of the input data. i，j ) represents the probability distribution of network predictions;

[0146] The arc detection model was tested using the test dataset, and the hyperparameters were adjusted until the test requirements were met.

[0147] It should be noted that other embodiments or implementation methods of the fault arc detection system described in this invention can refer to the above-described method embodiments, and will not be repeated here.

[0148] Furthermore, the third embodiment of the present invention also proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the fault arc detection method as described in the first embodiment above.

[0149] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0150] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0151] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0152] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method of detecting a fault arc, characterized by, The method includes: Acquire the current waveform of the circuit under test; The current waveform is enhanced to obtain a current enhancement signal; The current enhancement signal is input into a pre-trained arc detection model to obtain an arc fault detection result corresponding to the current waveform; The arc detection model includes an input layer, a feature extraction layer, and an output layer connected in sequence. The feature extraction layer includes several residual modules connected in sequence, and the output of each residual module is connected to a compression excitation module. The output of the residual module and the output of the compression excitation module are linearly superimposed, and then linearly superimposed with the input data of the residual module for output. The compression excitation module includes a channel attention mechanism network and a spatial attention mechanism network connected in sequence. The channel attention mechanism network consists of several convolutional layers and a global max pooling layer, and the spatial attention mechanism network consists of several convolutional layers and a global average pooling layer.

2. The arc fault detection method of claim 1, wherein, The enhancement processing of the current waveform to obtain the current enhancement signal includes: The current waveform is enhanced using a parametric Wiener filter to obtain the enhanced current signal. The formula for the parametric Wiener filter is as follows: wherein: is the power spectrum of the pure signal, is the power spectrum of the noise signal, is the prior signal-to-noise ratio, is an adjustable parameter.

3. The arc fault detection method of claim 1, wherein, After inputting the current-enhancing signal into a pre-trained arc detection model to obtain an arc fault detection result corresponding to the current waveform, the method further includes: If the detection result indicates that an arc fault has occurred, then the fault count is incremented by one; otherwise, no counting operation is performed. When the number of accumulated faults exceeds the set number, the line is disconnected via the disconnection device.

4. The arc fault detection method of claim 1, wherein, Before inputting the current-enhancing signal into the pre-trained arc detection model to obtain the arc fault detection result corresponding to the current waveform, the method further includes: Acquire the current signal from the fault arc test platform; The current signal is enhanced using a parametric Wiener filter to obtain an enhanced dataset; The arc detection model is trained using the augmented dataset to obtain a trained arc detection model.

5. The arc fault detection method of claim 4, wherein, The fault arc test platform includes a fault arc generating device, an AC power supply, and different types of loads connected in series. A switch is connected between the fault arc generating device and the loads, and a current acquisition device is connected between the AC power supply and the fault arc generating device. The current acquisition device is used to acquire the current signal of the main circuit of the fault arc test platform. The current signal includes the main circuit current signal in the normal state and the main circuit current signal in the fault state.

6. The arc fault detection method of claim 4, wherein, After acquiring the current signal from the fault arc experimental platform, the method further includes: The current signal is segmented according to the power frequency cycle to obtain an arc data set; A high-pass filter is used to extract the circuit noise from the arc data in the arc dataset. Based on the arc dataset and the circuit noise, a parametric Wiener filter is constructed.

7. The arc fault detection method of claim 4, wherein, The step of training the arc detection model using the augmented dataset to obtain a trained arc detection model includes: The augmented dataset is divided into a training dataset and a test dataset according to a certain ratio; The training dataset is used as input to the arc detection model. The weights and biases of the convolutional and fully connected layers in the network are updated iteratively to obtain the set of parameter values ​​that minimize the loss function, which are then used as the model parameters. The loss function used is the cross-entropy loss function, expressed by the following formula: In the formula: m Indicates batch size. n Indicates data dimensions, This represents the true distribution of the input data. This represents the probability distribution of the network predictions; The arc detection model was tested using the test dataset, and the hyperparameters were adjusted until the test requirements were met.

8. A fault arc detection system characterized by, The system includes: The acquisition module is used to acquire the current waveform of the circuit under test; An enhancement module is used to enhance the current waveform to obtain a current enhancement signal; The detection module is used to input the current enhancement signal into a pre-trained arc detection model to obtain an arc fault detection result corresponding to the current waveform; The arc detection model includes an input layer, a feature extraction layer, and an output layer connected in sequence. The feature extraction layer includes several residual modules connected in sequence, and the output of each residual module is connected to a compression excitation module. The output of the residual module and the output of the compression excitation module are linearly superimposed, and then linearly superimposed with the input data of the residual module for output. The compression excitation module includes a channel attention mechanism network and a spatial attention mechanism network connected in sequence. The channel attention mechanism network consists of several convolutional layers and a global max pooling layer, and the spatial attention mechanism network consists of several convolutional layers and a global average pooling layer.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the fault arc detection method as described in any one of claims 1-7.

Images