Arc fault detection method and device

By encoding the timing current data and extracting the feature of deep learning models, the problem of insufficient accuracy of traditional arc fault detection methods in complex arc fault detection is solved, and higher arc fault detection accuracy and electrical system safety are achieved.

CN118914773APending Publication Date: 2024-11-08GUANGDONG POWER GRID CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410989283.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Traditional arc fault detection methods have low accuracy when dealing with complex arc faults, making it difficult to effectively detect arc faults in circuits, especially in the face of the diversity and complexity of arc faults.

Method used

By collecting currents of the target circuit within the target time period, time sequence current data is obtained and encoded to generate current feature images, input them into the arc fault detection model, and feature extraction is performed using deep learning technology, especially convolutional neural networks. and fault detection.

Benefits of technology

It significantly improves the accuracy and robustness of arc fault detection, and is suitable for industrial electrical systems, home electrical equipment and transportation, etc., and improves the safety and reliability of electrical systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118914773A_ABST
    Figure CN118914773A_ABST
Patent Text Reader

Abstract

The embodiment of the invention discloses an arc fault detection method and device. The method comprises the following steps: in a target time period, carrying out current acquisition on a target circuit to obtain time sequence current data; encoding the time sequence current data to obtain a current feature image; and inputting the current feature image into an arc fault detection model, and determining an arc fault detection result of the target circuit in the target time period according to a model output result of the arc fault detection model. According to the technical scheme of the embodiment of the invention, the accuracy of arc fault detection is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of electrical fault detection, and in particular to an arc fault detection method and device. Background Art

[0002] Arc faults are a common and dangerous type of fault in circuits, which may cause serious safety hazards such as fire or circuit damage. As the requirements for electrical safety in industrial electrical systems, household electrical equipment, and transportation continue to increase, accurately detecting arc faults in circuits is becoming more and more important.

[0003] Traditional arc fault detection methods usually rely on manually set features and simple threshold judgments. However, due to the complex and changeable current waveforms in the circuit, these methods often perform poorly and have low accuracy when dealing with complex arc faults. Especially when faced with the diversity and complexity of arc faults, the limitations of traditional methods are more obvious. Therefore, it is currently difficult to accurately detect arc faults using traditional methods, and this problem needs to be solved urgently. Summary of the invention

[0004] The embodiments of the present invention provide an arc fault detection method and device, which improve the accuracy of arc fault detection.

[0005] According to one aspect of the present invention, there is provided an arc fault detection method, which may include:

[0006] During the target time period, current is collected for the target circuit to obtain time-series current data;

[0007] Encode the time series current data to obtain a current characteristic image;

[0008] The current characteristic image is input into the arc fault detection model, and the arc fault detection result of the target circuit within the target time period is determined according to the model output result of the arc fault detection model.

[0009] According to another aspect of the present invention, there is provided an arc fault detection device, which may include:

[0010] A time-series current data acquisition module is used to collect current from a target circuit within a target time period to obtain time-series current data;

[0011] A current characteristic image obtaining module is used to encode the time series current data to obtain a current characteristic image;

[0012] The arc fault detection result determination module is used to input the current characteristic image into the arc fault detection model, and determine the arc fault detection result of the target circuit within the target time period according to the model output result of the arc fault detection model.

[0013] The technical solution of the embodiment of the present invention is to collect current from the target circuit within the target time period to obtain time-series current data. The current data in the form of time series is convenient for subsequent encoding to obtain a current characteristic image; the time-series current data is encoded to obtain a current characteristic image. By encoding the time-series current data into a current characteristic image, it is beneficial for the image recognition capability of the subsequent arc fault detection model to extract features from the current characteristic image; the current characteristic image is input into the arc fault detection model, and the arc fault detection result of the target circuit within the target time period is determined according to the model output result of the arc fault detection model. The current characteristic image is processed by the arc fault detection model to improve the accuracy of arc fault detection. The above technical solution extracts features from the time-series current data encoded as a current characteristic image by utilizing the image feature extraction capability of the arc fault detection model, and then performs arc fault detection according to the extracted features, thereby improving the accuracy of arc fault detection.

[0014] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 is a flow chart of an arc fault detection method provided according to an embodiment of the present invention;

[0017] Figure 2 is a flow chart of another arc fault detection method provided according to an embodiment of the present invention;

[0018] Figure 3 is a flow chart of determining a current characteristic image in an arc fault detection method provided in an embodiment of the present invention;

[0019] Figure 4 is a flow chart of another arc fault detection method provided according to an embodiment of the present invention;

[0020] Figure 5 is a schematic diagram of data flow of an arc fault detection model in another arc fault detection method provided according to an embodiment of the present invention;

[0021] Figure 6 is a flow chart of an optional example of another arc fault detection method provided according to an embodiment of the present invention;

[0022] Figure 7 is a structural block diagram of an arc fault detection device provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. The situations of "target", "original", etc. are similar and will not be repeated here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] Before introducing the embodiments of the present invention, an exemplary description is given of the implementation process of the current arc fault detection scheme and the reasons why it is difficult to accurately perform arc fault detection through traditional methods, so as to better understand why the arc fault detection scheme proposed in the embodiments of the present invention can improve the accuracy of arc fault detection.

[0026] For example, the rapid development of deep learning technology has provided new solutions for fault detection. Deep learning models, especially convolutional neural networks (CNN), have achieved remarkable success in image recognition and classification tasks. CNN has powerful feature extraction capabilities and can automatically learn effective feature representations from data. However, time series current data is usually one-dimensional time series data, and there are certain difficulties in directly applying CNN. Therefore, accurate arc fault detection is performed using traditional methods.

[0027] In view of this, the embodiment of the present invention utilizes the image feature extraction capability of the arc fault detection model to extract features from the time series current data encoded as the current feature image, and then performs arc fault detection based on the extracted features, thereby improving the accuracy of arc fault detection. This will be described in detail below.

[0028] Figure 1 1 is a flow chart of an arc fault detection method provided in an embodiment of the present invention. This embodiment is applicable to arc fault detection, and is particularly applicable to arc fault detection in a series circuit. The method can be executed by an arc fault detection device provided in an embodiment of the present invention, which can be implemented in software and / or hardware, and can be integrated in an electronic device, which can be various user terminals or servers.

[0029] See also Figure 1 The method of the embodiment of the present invention specifically includes the following steps:

[0030] S110 . Within a target time period, current is collected for a target circuit to obtain sequential current data.

[0031] Among them, the target time period is the time period for current collection for the target circuit, that is, within the target time period, current collection will continue or be performed at the target sampling rate; the target time period can be determined according to the current collection cycle, that is, the target time period can be the time period corresponding to the current collection cycle at the current time, and so on. The target circuit is the circuit that requires arc fault detection. Time series current data is data composed of current values ​​expressed in time series form; time series current data can be one-dimensional data.

[0032] It can be understood that, since the current is collected for the target circuit within the target time period, the collected current values ​​are arranged in a time series format, and thus the obtained current values ​​can be used as time series current data.

[0033] S120 , encoding the time series current data to obtain a current characteristic image.

[0034] The current characteristic image is an image obtained by encoding the time series current data from which features can be extracted.

[0035] In the embodiment of the present invention, for example, the time series current data may be encoded according to Gramian Angular Summation Fields (GASF) to obtain a current characteristic image, so as to amplify the inherent characteristics of the arc fault when an arc fault exists in the target circuit. In the embodiment of the present invention, the method of encoding the time series current data to obtain the current characteristic image is not specifically limited.

[0036] S130, inputting the current characteristic image into an arc fault detection model, and determining an arc fault detection result of a target circuit within a target time period according to a model output result of the arc fault detection model.

[0037] The arc fault detection model is a model for arc fault detection; the arc fault model may be, for example, an AlexNet model; in the embodiment of the present invention, the structure and parameters of the arc fault detection model are not specifically limited. The model output result is the output result of the arc fault detection model. The arc fault detection result is a result that can characterize whether there is an arc fault in the target circuit; the arc fault detection result includes at least one of whether there is an arc fault and the type of arc fault.

[0038] In the embodiment of the present invention, there is no specific limitation on the method of inputting the current characteristic image into the arc fault detection model and determining the arc fault detection result of the target circuit within the target time period according to the model output result of the arc fault detection model.

[0039] For example, considering that the activation function ReLU is a nonlinear activation function, the calculation is relatively simple, that is, if the input is greater than 0, the input value is directly returned, and if the input value is 0 or less, the value 0 is returned. However, the results of ReLU are all positive values, without negative values, and forced sparse processing will cover up too many features, resulting in the arc fault detection model being unable to learn effective features. Therefore, the LeakyReLU activation function can be used. The original detection model is trained, where α is a small constant, usually around 0.01, which retains some values ​​of the negative axis, so that all information of the negative axis is not lost. Considering that L2 regularization, also known as weight decay, is a common technique in machine learning to control model complexity and reduce overfitting, by adding L2 regularization terms to the original detection model, the role of the parameter part that better represents the data distribution can be highlighted, so that the parameter part of the arc fault detection model that is prone to overfitting is closer to 0. The L2 regularization formula can be used It means that the sum of the squares of the weight parameters added to the original loss function, where E in is the training sample error without regularization, λ is the regularization parameter, is the L2 norm. Considering that the gradient descent algorithm requires a learning rate to be specified as a control factor for the weight update step size, and at the beginning of training, the learning step size is expected to be as large as possible within a tolerable range. As the number of training rounds increases, in order to avoid the gradient descent from being unable to converge to the global optimum or even diverging, the learning step size is required to become smaller and smaller, gradually approaching the optimum, or at least oscillating back and forth within the optimal range. Therefore, the learning rate can be adjusted at equal intervals, and it can be adjusted to α for each certain number of training times the original number of times, so that the arc fault detection model can better converge to the optimal solution. Considering that in the process of model training and parameter tuning, the training sample set is used to train the original detection model and fit the parameters, while the validation sample set is used to adjust the hyperparameters of the original detection model, a part of the training sample set can be determined from multiple training sample sets as the validation sample set. The original detection model learns the data features in the training sample set and updates the parameters through forward propagation and back propagation to minimize the loss function value. At the same time, at the end of each training cycle, the model will be evaluated on the validation sample set, and the loss and other performance indicators will be calculated to monitor the generalization ability of the model to prevent overfitting or underfitting. According to the performance of the validation sample set, hyperparameters such as learning rate and batch size are adjusted, and the model parameters with the best performance on the validation sample set are continuously saved until the end of training. Finally, the model with the smallest loss on the validation sample set is the optimal network model, that is, the arc fault detection model, which is expected to achieve the best performance in the test sample set or actual application. The loss and accuracy of the arc fault detection model on multiple training sample sets can also be calculated. The accuracy measures the classification ability of the model, and the loss measures the prediction error of the model. For example, ten independent experiments can be conducted, the results of each experiment can be recorded, and the average value can be taken as the final evaluation result to ensure the reliability and stability of the results. A variety of evaluation indicators can be used, including accuracy, precision, recall rate and F1 score, to comprehensively evaluate the performance of the arc fault detection model, analyze the performance of the arc fault detection model under different fault types and different working conditions, and ensure the robustness and generalization ability of the arc fault detection model.

[0040] In an embodiment of the present invention, by utilizing the image feature extraction capability of the arc fault detection model, feature extraction is performed on the time series current data encoded as a current feature image, and then arc fault detection is performed based on the extracted features, thereby significantly improving the detection accuracy and robustness of arc fault detection. The method is suitable for various scenarios such as industrial electrical systems, household electrical equipment, and transportation, thereby significantly improving the safety and reliability of electrical systems.

[0041] The technical solution of the embodiment of the present invention is to collect current from the target circuit within the target time period to obtain time-series current data. The current data in the form of time series is convenient for subsequent encoding to obtain a current characteristic image; the time-series current data is encoded to obtain a current characteristic image. By encoding the time-series current data into a current characteristic image, it is beneficial for the image recognition capability of the subsequent arc fault detection model to extract features from the current characteristic image; the current characteristic image is input into the arc fault detection model, and the arc fault detection result of the target circuit within the target time period is determined according to the model output result of the arc fault detection model. The current characteristic image is processed by the arc fault detection model to improve the accuracy of arc fault detection. The above technical solution extracts features from the time-series current data encoded as a current characteristic image by utilizing the image feature extraction capability of the arc fault detection model, and then performs arc fault detection according to the extracted features, thereby improving the accuracy of arc fault detection.

[0042] An optional technical solution, within a target time period, current is collected for a target circuit to obtain time-series current data, including: within the target time period, current is collected for the target circuit at a target sampling rate to obtain time-series current data; after obtaining the time-series current data, the method also includes: updating the target time period according to a preset current collection cycle and the target time period; segmenting the time-series current data according to a preset time window to obtain at least one segmented data; for each segmented data in at least one segmented data, determining the number of current values ​​of the segmented data, and determining the margin factor of the segmented data according to the segmented data and the number of current values; updating the target sampling rate according to the margin factor corresponding to the at least one segmented data, and returning to execute the step of collecting current from the target circuit at the target sampling rate within the target time period to obtain time-series current data.

[0043] Among them, the target sampling rate is the sampling rate for current acquisition for the target circuit within the target time period. The current acquisition cycle is the period for periodically acquiring current from the target circuit to obtain timing current data. The preset time window is a preset time window for segmenting the timing current data in time series, and the preset time window can also be understood as a preset time length. The segmented data is the data obtained by segmenting the timing current data. The number of current values ​​is the number of target timing current values ​​in the segmented data. The target timing current value is the current value in the timing current data. The margin factor is the ratio of the peak value to the square root amplitude of the target timing current value in the segmented data.

[0044] It should be noted that, when the first demand is to collect current for the target circuit at the target sampling rate to obtain timing current data, the target time period can be determined based on the preset current collection cycle, or based on the preset current collection cycle and the current time point. When the subsequent demand is to collect current for the target circuit at the target sampling rate to obtain timing current data, the target time period is updated based on the preset current collection cycle and the target time period.

[0045] In the embodiment of the present invention, the target time period obtained after updating the target time period may be the time period between the end time point corresponding to the target time period and the time point corresponding to the time length corresponding to the current acquisition cycle plus the end time point. For example, if the current acquisition cycle is 1 second as one cycle, the target time period is the time period corresponding to 8:00:00-8:00:01, and the target time period obtained after updating the target time period may be the time period corresponding to 8:00:01-8:00:02. In the embodiment of the present invention, there is no specific limitation on the method of updating the target time period according to the preset current acquisition cycle and the target time period.

[0046] In an embodiment of the present invention, each target time series current value across a preset time window in the time series current data can be segmented as a segmented data according to a preset time window to obtain at least one segmented data. In an embodiment of the present invention, the method of segmenting the time series current data according to the preset time window to obtain at least one segmented data is not specifically limited.

[0047] In the embodiment of the present invention, for each segmented data in the at least one segmented data, the number of target time series current values ​​in the segmented data can be determined and used as the number of current values. In the embodiment of the present invention, the method for determining the number of current values ​​of the segmented data is not specifically limited.

[0048] In the embodiment of the present invention, for example, the formula According to the segmented data and the number of current values, the margin factor of the segmented data is determined, wherein max(X) and min(X) respectively represent the maximum value and the minimum value of each target sequential current value in the segmented data X, N is the number of current values, and i is the i-th target sequential current value in X. In the embodiment of the present invention, the manner of determining the margin factor of the segmented data according to the segmented data and the number of current values ​​is not specifically limited.

[0049] It is understandable that, since the time series current data is segmented according to the preset time window to obtain at least one segmented data, there is also a temporal sequence between the at least one segmented data, that is, there is a time series relationship between the at least one segmented data, and the change of each margin factor can be determined according to the time series relationship between the at least one segmented data and the margin factor corresponding to the at least one segmented data, and the target sampling rate is updated according to the change. In the embodiment of the present invention, the method of updating the target sampling rate according to the margin factor corresponding to the at least one segmented data is not specifically limited.

[0050] The above technical scheme can update the target sampling rate according to the margin factor corresponding to at least one segmented data, so that when arc fault detection is required periodically, the target sampling rate in each cycle changes dynamically, so that high-quality timing current data can be collected according to the target sampling rate according to the current situation, thereby further improving the accuracy of arc fault detection.

[0051] Another optional technical solution is that the arc fault detection model is pre-trained through the following steps: obtaining sample timing current data of the sample circuit and the arc fault label of the sample timing current data, and determining a training sample set based on the sample timing current data and the arc fault label; based on multiple training sample sets, training the original detection model to obtain an arc fault detection model.

[0052] The sample circuit is the circuit for which the sample time series current data is collected. The sample time series current data is the data obtained by collecting and processing the current of the sample circuit; the time series sample data may be image data obtained by encoding the data obtained by collecting the current of the sample circuit. The arc fault label is a label that can reflect whether there is an arc fault in the sample circuit. The training sample set is a data set including at least one training sample for training an arc fault detection model. The original detection model is an untrained arc fault detection model.

[0053] It is understandable that since the number of sample timing current data and arc fault labels of the sample timing current data may be small, or the size or length of the sample timing current data does not meet the requirements for training the original detection model, the sample timing current data and arc fault labels can be processed by data enhancement or data cropping to obtain a set of training sample sets including at least one training sample to solve the above problems.

[0054] In an embodiment of the present invention, the sample time series current data can be subjected to at least one of the following processing operations: mirror flipping, rotation, and mirror flipping plus rotation, to obtain at least one processed data, and / or the sample time series current data can be cropped to obtain at least one processed data; for each of the at least one processed data obtained, the processed data and the arc fault label are used as a group of training samples; the at least one group of training samples obtained is used as a training sample set, so as to increase the diversity of training samples, so that the arc fault detection model can see more different image transformation forms during the training process, thereby improving the performance and generalization ability of the arc fault detection model in different scenarios. For example, the sample time series current data is subjected to mirror flipping, rotation, and mirror flipping and rotation operations, and the sample time series current data itself is added to obtain four processed data. For each of the four processed data, the processed data and the arc fault label are used as a group of training samples; the four groups of training samples are used as a training sample set. Specifically, the sample time series current data can be horizontally mirror flipped to obtain a processed data, the sample time series current data can be rotated 90 degrees to obtain a processed data, the sample time series current data can be mirror flipped and rotated to obtain a processed data, and the sample time series current data can be used as a processed data to obtain four processed data. After obtaining at least one processed data, the torchvision.transforms package of PyTorch can be used to normalize the at least one processed data. Specifically, the at least one processed data can be converted into a tensor and normalization can be applied. According to the normalization result obtained, at least one processed data is updated to help standardize the data, reduce the difference between the processed data, and make the training arc fault detection model more stable and effective. In the embodiment of the present invention, there is no specific limitation on the method of determining a training sample set according to the sample time series current data and the arc fault label.

[0055] In the embodiment of the present invention, there is no specific limitation on the method of training the original detection model based on multiple training sample sets to obtain the arc fault detection model.

[0056] In an embodiment of the present invention, a training sample set can be determined based on sample timing current data and arc fault labels, and then the original detection model can be trained based on multiple training sample sets to obtain an arc fault detection model. This can avoid the problem of low accuracy of arc fault detection by the arc fault detection model due to the fact that the number of sample timing current data and arc fault labels of the sample timing current data may be small, or the size or length of the sample timing current data does not meet the requirements for training the original detection model, thereby significantly improving the robustness of the arc fault detection model and the accuracy of arc fault detection.

[0057] Based on the above scheme, another optional technical scheme is to determine a training sample set according to sample timing current data and arc fault labels, including: determining at least one timing starting current value from the sample timing current data; for each timing starting current value of at least one timing starting current value, intercepting the sample timing current data according to the timing starting current value and a preset interception time length to obtain enhanced data, and using the enhanced data and arc fault labels as a group of training samples to obtain a training sample set based on the obtained multiple groups of training samples.

[0058] Among them, the time series starting point current value is the current value of the starting point of the enhanced data in the time series. The interception time length is the length of the intercepted enhanced data in time. The enhanced data is the data intercepted as the training data in the training sample. The enhanced data can also be understood as the data in the training sample obtained by data enhancement of the sample time series current data. The training sample is the sample used to train the original detection model.

[0059] It is understandable that in actual arc fault detection, due to the difficulty of data collection, it is often impossible to obtain a large enough amount of data to train a deep neural network. Therefore, there may be a problem that the number of sample time series current data and arc fault labels of the sample time series current data may be small. Therefore, data enhancement technology can be implemented by intercepting the sample time series current data to expand the existing data set to obtain more training samples.

[0060] In the embodiment of the present invention, at least one time-series starting current value can be randomly determined from the sample time-series current data. In the embodiment of the present invention, the method of determining at least one time-series starting current value from the sample time-series current data is not specifically limited.

[0061] In an embodiment of the present invention, at least one time series starting current value can be determined from the sample time series current data; for each time series starting current value in at least one time series starting current value, the time series starting current value is used as the starting point, data of the interception time length is intercepted from the sample time series current data, the intercepted data is used as enhanced data, and the enhanced data and the arc fault label are used as a group of training samples, so as to obtain a training sample set based on the obtained multiple groups of training samples, so that a large number of training samples can be generated from the limited sample time series current data, thereby significantly increasing the number of training samples, improving the robustness of the arc fault detection model under different fault conditions and measurement errors, and effectively reducing the risk of overfitting, so that the arc fault detection model can be better generalized to unseen data. In an embodiment of the present invention, the method of intercepting the sample time series current data according to the time series starting current value and the preset interception time length to obtain the enhanced data is not specifically limited.

[0062] In an embodiment of the present invention, when it is required to test the arc fault detection model obtained by training, in addition to determining the training samples, test samples are also required. To this end, at least one timing starting current value can be determined from the first part of the sample timing current data; for each timing starting current value in the at least one timing starting current value, the sample timing current data is intercepted and processed according to the timing starting current value and the preset interception time length to obtain enhanced data, and the enhanced data and the arc fault label are used as a group of training samples to obtain a training sample set based on the obtained multiple groups of training samples; at least one test starting current value can be determined from the second part of the sample timing current data; for each test starting current value in the at least one test starting current value, the sample timing current data is intercepted and processed according to the test starting current value and the preset interception time length to obtain test data, and the test data and the arc fault label are used as a group of test samples to obtain a test sample set based on the obtained multiple groups of test samples, so as to test the arc fault detection model according to the obtained multiple test sample sets.

[0063] Exemplarily, at least one time series starting current value s can be randomly determined from a first part of a length L1 in the sample time series current data of a total length L according to a preset number of training samples N, and the time series starting current value s is within the interval [0, L1-1]; for each time series starting current value in the at least one time series starting current value, the sample time series current data is intercepted according to the time series starting current value and a preset interception time length l to obtain enhanced data, and the enhanced data and the arc fault label are used as a group of training samples to obtain a training sample based on the obtained multiple groups of training samples. This set; the number of test samples M can be preset, and at least one test starting current value is randomly determined from the second part of the sample timing current data with a length of L-L1; for each test starting current value of the at least one test starting current value, according to the test starting current value and the preset interception time length l, the sample timing current data is intercepted and processed to obtain test data, and the test data and the arc fault label are used as a group of test samples, so as to obtain a test sample set based on the obtained multiple groups of test samples, so as to test the arc fault detection model according to the obtained multiple test sample sets.

[0064] The technical solution of the embodiment of the present invention obtains enhanced data to obtain a group of training samples by intercepting and processing the sample time series current data, thereby realizing data enhancement of the sample data, and then obtaining more training samples to train the original detection model, avoiding overfitting of the arc fault detection model, and improving the robustness of the arc fault detection model, so that the arc fault detection accuracy of the trained arc fault detection model is higher.

[0065] Figure 2 It is a flow chart of another arc fault detection method provided in an embodiment of the present invention. This embodiment is optimized based on the above-mentioned technical solutions. In this embodiment, optionally, the time series current data is encoded to obtain a current characteristic image, including: for each target time series current value in the time series current data, the target time series current value is cosine encoded to obtain a current cosine value; according to the current cosine values ​​corresponding to each target time series current value, a current characteristic matrix is ​​determined, and according to the current characteristic matrix, a current characteristic image is determined. Among them, the explanations of the terms that are the same or corresponding to the above-mentioned embodiments are not repeated here.

[0066] See also Figure 2 The method of this embodiment may specifically include the following steps:

[0067] S210 . Within a target time period, current is collected for a target circuit to obtain sequential current data.

[0068] S220 . For each target time-series current value in the time-series current data, perform cosine encoding on the target time-series current value to obtain a current cosine value.

[0069] The current cosine value is a value obtained by performing cosine encoding on the target timing current value, that is, the cosine value of the target timing current value in polar coordinates.

[0070] For example, see Figure 3 , can be obtained by the formula The target time-series current value is cosine-encoded to obtain the current cosine value, and then the current characteristic image is determined according to the current cosine values ​​corresponding to each target time-series current value, where φ i is the cosine value of the current, In the embodiment of the present invention, the method of performing cosine encoding on the target timing current value to obtain the current cosine value is not specifically limited.

[0071] S230 , determining a current characteristic matrix according to current cosine values ​​corresponding to each target time-series current value, and determining a current characteristic image according to the current characteristic matrix.

[0072] The current characteristic matrix is ​​a matrix that can characterize the characteristics of time series current data.

[0073] It can be understood that the operation mode of the current characteristic matrix is ​​the inner product form, and the main diagonal of the current characteristic matrix can include the original direction and angle information of the time series current data. Therefore, the current characteristic matrix can characterize the characteristics of the time series current data.

[0074] In combination with the above examples, for example, the formula The current characteristic matrix is ​​determined in polar coordinates, where GASF is the current characteristic matrix, where the above formula can limit the value range of the current cosine value in the interval [-1,1] to the interval [0,π] by using the inverse trigonometric function under polar coordinate mapping, thereby obtaining the current characteristic matrix. It should be noted that since the current cosine value is unique, the inverse trigonometric function under polar coordinate mapping is also unique. In the embodiment of the present invention, the method of determining the current characteristic matrix according to the current cosine value corresponding to each target timing current value is not specifically limited.

[0075] In the embodiment of the present invention, the current characteristic image can be determined according to the current characteristic matrix to establish the absolute time relationship of the target time series current value in polar coordinates. In the embodiment of the present invention, the method of determining the current characteristic image according to the current characteristic matrix is ​​not specifically limited.

[0076] S240: Input the current characteristic image into an arc fault detection model, and determine the arc fault detection result of the target circuit within a target time period according to the model output result of the arc fault detection model.

[0077] The technical solution of the embodiment of the present invention performs cosine encoding on the target time-series current value for each target time-series current value in the time-series current data to obtain the current cosine value, so that a characteristic current feature matrix capable of characterizing the time-series current data can be determined through the current cosine value; a current feature matrix is ​​determined according to the current cosine values ​​corresponding to each target time-series current value, and a current feature image is determined according to the current feature matrix. The current feature image can be determined by determining the current feature matrix, so that a current feature diagram capable of characterizing the characteristics of the time-series current data can be obtained, thereby facilitating feature extraction of the time-series current data of the current feature image using an arc fault detection model with image feature extraction capability.

[0078] An optional technical solution, before cosine encoding the target timing current value to obtain the current cosine value, also includes: normalizing the target timing current value according to the timing current data; and updating the target timing current value according to the obtained normalization result.

[0079] The normalized result is a result obtained by normalizing the target timing current value.

[0080] In the embodiment of the present invention, the target sequential current value can be normalized according to the sequential current data. In combination with the above examples, for example, for each target sequential current value in the sequential current data, according to the sequential current data, the equation The target time series current value is normalized to obtain a normalized current value, and a normalization result is determined according to each normalized current value obtained, so that each target time series current value in the time series current data is scaled to the interval of [-1,1], wherein the time series current data X = {x1, x2, ..., x n}, that is, the time-series current data X includes n target time-series current values ​​x1, x2, ..., x n , x i is the target timing current value before updating, is the updated target sequential current value. In the embodiment of the present invention, the method of normalizing the target sequential current value according to the sequential current data is not specifically limited.

[0081] In the embodiment of the present invention, by normalizing the target time-series current value according to the time-series current data, the dimension effect of the target time-series current value can be eliminated, so as to facilitate the subsequent processing of the time-series current data.

[0082] Figure 4 It is a flowchart of another arc fault detection method provided in an embodiment of the present invention. This embodiment is optimized based on the above-mentioned technical solutions. In this embodiment, optionally, the arc fault detection model includes at least two target convolution layers, and the at least two target convolution layers include at least a first convolution layer and a second convolution layer; the current characteristic image is input into the arc fault detection model, and the arc fault detection result of the target circuit in the target time period is determined according to the model output result of the arc fault detection model, including: the current characteristic image is input into the arc fault detection model, so as to use the arc fault detection model to perform the following two steps to perform arc fault detection based on the current characteristic image: based on the first convolution layer, the current characteristic image is convoluted to obtain a first convolution result; based on the second convolution layer, the first convolution result is convoluted to obtain a second convolution result, and according to the second convolution result, the model output result of the arc fault detection model is determined to be output; according to the model output result, the arc fault detection result of the target circuit in the target time period is determined. Among them, the explanation of the terms that are the same or corresponding to the above-mentioned embodiments is not repeated here.

[0083] See also Figure 4 The method of this embodiment may specifically include the following steps:

[0084] S310 . Within a target time period, current is collected for a target circuit to obtain sequential current data.

[0085] S320 , encoding the time series current data to obtain a current characteristic image.

[0086] S330. Input the current characteristic image into an arc fault detection model, and use the arc fault detection model to execute steps S3301 to S3302 to perform arc fault detection based on the current characteristic image, wherein the arc fault detection model includes at least two target convolution layers, and the at least two target convolution layers include at least a first convolution layer and a second convolution layer.

[0087] The target convolution layer is the convolution layer in the arc fault detection model. The first convolution layer is the target convolution layer for the current feature image input. The second convolution layer is the target convolution layer for the first convolution result input.

[0088] S3301. Based on the first convolution layer, convolution processing is performed on the current characteristic image to obtain a first convolution result.

[0089] The first convolution result is a result obtained by convolving the current feature image based on the first convolution layer.

[0090] S3302. Based on the second convolution layer, perform convolution processing on the first convolution result to obtain a second convolution result, and determine the model output result of the arc fault detection model according to the second convolution result for output.

[0091] The second convolution result is a result obtained by convolutionally processing the first convolution result by the second convolution layer.

[0092] In the embodiment of the present invention, the second convolution result can be processed based on the convolution pooling submodule and / or the fully connected submodule and the like, and the model output result of the arc fault detection model is determined and outputted according to the processing result. In the embodiment of the present invention, the manner of determining the model output result of the arc fault detection model and outputting it according to the second convolution result is not specifically limited.

[0093] S340: Determine arc fault detection results of the target circuit within a target time period according to the model output results.

[0094] The technical solution of the embodiment of the present invention performs convolution processing on the current feature image through at least two target convolution layers to determine the model output result of the arc fault detection model for output, which can improve the convergence speed and ensure the accuracy of the arc fault detection model in arc fault detection.

[0095] An optional technical solution, the arc fault detection model also includes at least two convolution pooling sub-modules, and the at least two convolution pooling sub-modules include at least a first sub-module and a second sub-module; according to the second convolution result, the model output result of the arc fault detection model is determined for output, including: based on the first sub-module, the second convolution result is convolution pooling processed to obtain a first convolution pooling result; based on the second sub-module, the first convolution pooling result is convolution pooling processed to obtain a second convolution pooling result, and according to the second convolution pooling result, the model output result of the arc fault detection model is determined for output.

[0096] Among them, the convolution pooling submodule is a submodule in the arc fault detection model that includes a convolution layer and a pooling layer; in the embodiment of the present invention, the order of the convolution layer and the pooling layer in the convolution pooling submodule is not specifically limited. The first submodule is the convolution pooling submodule inputted by the second convolution result. The second submodule is the convolution pooling submodule inputted by the first convolution pooling result. The first convolution pooling result is the result obtained by the first submodule performing convolution pooling processing on the second convolution result. The second convolution pooling result is the result obtained by the second submodule performing convolution pooling processing on the first convolution pooling result.

[0097] In an embodiment of the present invention, the arc fault detection model may further include a flat layer, a fully connected module and a normalized index layer, and the fully connected module may include at least two fully connected submodules; the second convolution pooling result may be processed based on the flat layer to obtain a flat result; the flattening may be fully connected based on the fully connected module to obtain a fully connected result; the fully connected result may be processed based on the normalized index layer to obtain a normalized index result; and the model output result of the arc fault detection model may be determined and outputted according to the normalized index result. In an embodiment of the present invention, there is no specific limitation on the manner in which the model output result of the arc fault detection model may be determined and outputted according to the second convolution pooling result.

[0098] Optionally, the number of convolutional pooling submodules is five. It is understandable that the number of convolutional pooling submodules is five, which can increase the number of convolutional layers of the arc fault detection model to 7 and the number of pooling layers to 5, which can improve the feature extraction capability and generalization performance of the arc fault detection model, obtain better training results, and further improve the convergence speed of the arc fault detection model and ensure the accuracy of the arc fault detection model in arc fault detection.

[0099] For example, see Figure 5, the arc fault detection model can include a first convolutional layer and a second convolutional layer, five convolutional pooling submodules, a flat layer, two fully connected submodules and a softmax layer; the five convolutional pooling submodules include a convolutional layer and a pooling layer respectively, the first three convolutional pooling submodules have the pooling layer in front and the convolutional layer in the back, and the last two convolutional pooling submodules have the convolutional layer in front and the pooling layer in the back; the number of output nodes of the first convolutional layer is 48, the number of output nodes of the first convolutional layer is 128, the number of output nodes of the convolutional layers in the five convolutional pooling submodules are 192, 256, 256, 192, 128 and 48 respectively, the number of output nodes of the pooling layers in the five convolutional pooling submodules are 128, 192, 256, 128 and 48 respectively, the number of output nodes of the first fully connected submodule is 128, the number of output nodes of the second fully connected submodule is 64, and the number of output nodes of the softmax layer is 27; the size of the convolution kernel of all layers is 5×5, and the step size is 1. Compared with existing network structures such as the AlexNet network structure, the above technical solution has reduced convolution kernel size and the number of output nodes of the convolution layer, increased the number of convolution layers and pooling layers, reduced the number of output nodes of the fully connected layer, and removed the local response normalization layer. This not only improves the convergence speed of the arc fault detection model, but also ensures the training effect of the arc fault detection model.

[0100] In an embodiment of the present invention, the second convolution result is processed by at least two convolution pooling sub-modules to determine the model output result of the arc fault detection model for output, which can further improve the convergence speed and ensure the accuracy of the arc fault detection model in arc fault detection.

[0101] In order to better understand the technical solution of the above embodiment of the present invention, an optional example is provided here. Figure 6 , modeling to obtain the original detection model; training the original detection model and performing performance evaluation to obtain the arc fault detection model; obtaining time series current data; converting the time series current data into a current characteristic image; performing arc fault detection based on the current characteristic image and the arc fault detection model.

[0102] Figure 7 This is a structural block diagram of an arc fault detection device provided in an embodiment of the present invention, and the device is used to execute the arc fault detection method provided in any of the above embodiments. The device and the arc fault detection method of the above embodiments belong to the same inventive concept. For details not described in detail in the embodiment of the arc fault detection device, refer to the embodiment of the above arc fault detection method. Figure 7 The device may specifically include: a time series current data obtaining module 410, a current characteristic image obtaining module 420 and an arc fault detection result determining module 430.

[0103] The time-series current data obtaining module 410 is used to collect current from a target circuit within a target time period to obtain time-series current data;

[0104] A current characteristic image obtaining module 420 is used to encode the time series current data to obtain a current characteristic image;

[0105] The arc fault detection result determination module 430 is used to input the current characteristic image into the arc fault detection model, and determine the arc fault detection result of the target circuit within the target time period according to the model output result of the arc fault detection model.

[0106] Optionally, the current characteristic image obtaining module 420 includes:

[0107] A current cosine value obtaining submodule is used to perform cosine encoding on each target time-series current value in the time-series current data to obtain a current cosine value;

[0108] The current characteristic image determination submodule is used to determine the current characteristic matrix according to the current cosine values ​​corresponding to each target time series current value, and determine the current characteristic image according to the current characteristic matrix.

[0109] Optionally, based on the above device, the device may further include:

[0110] A normalization processing module is used to perform normalization processing on the target time-series current value according to the time-series current data before performing cosine encoding on the target time-series current value to obtain the current cosine value;

[0111] The target timing current value updating module is used to update the target timing current value according to the obtained normalization result.

[0112] Optionally, the arc fault detection model includes at least two target convolution layers, and the at least two target convolution layers include at least a first convolution layer and a second convolution layer;

[0113] The arc fault detection result determination module 430 includes:

[0114] The arc fault detection submodule is used to input the current characteristic image into the arc fault detection model, so as to use the arc fault detection model to perform arc fault detection based on the current characteristic image through the following two units:

[0115] A first convolution result obtaining unit, used for performing convolution processing on the current characteristic image based on the first convolution layer to obtain a first convolution result;

[0116] A model output result output unit, configured to perform convolution processing on the first convolution result based on the second convolution layer to obtain a second convolution result, and determine the model output result of the arc fault detection model according to the second convolution result for output;

[0117] The arc fault detection result determination submodule is used to determine the arc fault detection result of the target circuit within the target time period according to the model output result.

[0118] Optionally, based on the above device, the arc fault detection model further includes at least two convolutional pooling submodules, and the at least two convolutional pooling submodules include at least a first submodule and a second submodule;

[0119] Model output result output unit, including:

[0120] A first convolution pooling result obtaining subunit is used to perform convolution pooling processing on the second convolution result based on the first submodule to obtain a first convolution pooling result;

[0121] The model output result output subunit is used to perform convolution pooling processing on the first convolution pooling result based on the second submodule to obtain a second convolution pooling result, and determine the model output result of the arc fault detection model according to the second convolution pooling result for output.

[0122] Optionally, based on the above device, the number of convolutional pooling sub-modules is five.

[0123] Optionally, the time series current data obtaining module 410 includes:

[0124] The time-series current data acquisition submodule is used to collect current from the target circuit within the target time period and at the target sampling rate to obtain the time-series current data;

[0125] The device may also include:

[0126] A target time period updating module is used to update the target time period according to a preset current collection cycle and a target time period after obtaining the time series current data;

[0127] A segmentation data obtaining module, used for segmenting the time series current data according to a preset time window to obtain at least one segmentation data;

[0128] A margin factor determination module, for determining the number of current values ​​of each segmented data in at least one segmented data, and determining a margin factor of the segmented data according to the segmented data and the number of current values;

[0129] The module for executing the steps is returned, which is used to update the target sampling rate according to the margin factors corresponding to at least one segmented data, and return to the step of collecting current for the target circuit within the target time period and at the target sampling rate to obtain the timing current data.

[0130] Optionally, the device may also include the following modules to pre-train an arc fault detection model:

[0131] A training sample set determination module is used to obtain sample time-series current data of a sample circuit and arc fault labels of the sample time-series current data, and determine a training sample set according to the sample time-series current data and the arc fault labels;

[0132] The arc fault detection model obtaining module is used to train the original detection model based on multiple training sample sets to obtain the arc fault detection model.

[0133] Optionally, based on the above device, the training sample set determination module includes:

[0134] A timing starting point current value determination submodule, used to determine at least one timing starting point current value from the sample timing current data;

[0135] The training sample set obtaining submodule is used to intercept the sample timing current data for each timing starting current value in at least one timing starting current value according to the timing starting current value and a preset interception time length to obtain enhanced data, and use the enhanced data and the arc fault label as a group of training samples to obtain a training sample set based on the obtained multiple groups of training samples.

[0136] The arc fault detection device provided by the embodiment of the present invention collects current from the target circuit within the target time period through the time series current data acquisition module to obtain the time series current data. The current data in the form of time series is convenient for subsequent encoding to obtain the current characteristic image; the time series current data is encoded through the current characteristic image acquisition module to obtain the current characteristic image. By encoding the time series current data into the current characteristic image, it is convenient for the image recognition ability of the subsequent arc fault detection model to extract the features of the current characteristic image; the current characteristic image is input into the arc fault detection model through the arc fault detection result determination module, and the arc fault detection result of the target circuit within the target time period is determined according to the model output result of the arc fault detection model. The current characteristic image is processed by the arc fault detection model to improve the accuracy of arc fault detection. The above device extracts features from the time series current data encoded as the current characteristic image by utilizing the image feature extraction ability of the arc fault detection model, and then performs arc fault detection according to the extracted features, thereby improving the accuracy of arc fault detection.

[0137] The arc fault detection device provided in the embodiment of the present invention can execute the arc fault detection method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0138] It is worth noting that in the above-mentioned embodiment of the arc fault detection device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

Claims

1. A method for detecting an arc fault, characterized in that: include: During the target time period, current is collected for the target circuit to obtain time-series current data; Encoding the time series current data to obtain a current characteristic image; The current characteristic image is input into an arc fault detection model, and the arc fault detection result of the target circuit within the target time period is determined according to the model output result of the arc fault detection model.

2. The method according to claim 1, characterized in that The encoding of the time series current data to obtain a current characteristic image includes: For each target time-series current value in the time-series current data, cosine encoding is performed on the target time-series current value to obtain a current cosine value; A current characteristic matrix is ​​determined according to the current cosine values ​​respectively corresponding to the target time-series current values, and a current characteristic image is determined according to the current characteristic matrix.

3. The method according to claim 2, characterized in that Before performing cosine encoding on the target time series current value to obtain the current cosine value, the method further includes: According to the time-series current data, normalizing the target time-series current value; According to the obtained normalization result, the target timing current value is updated.

4. The method according to claim 1, characterized in that The arc fault detection model includes at least two target convolutional layers, and the at least two target convolutional layers include at least a first convolutional layer and a second convolutional layer; The step of inputting the current characteristic image into an arc fault detection model and determining the arc fault detection result of the target circuit within the target time period according to the model output result of the arc fault detection model includes: The current characteristic image is input into an arc fault detection model, and the arc fault detection model is used to perform the following two steps to perform arc fault detection based on the current characteristic image: Based on the first convolution layer, convolution processing is performed on the current characteristic image to obtain a first convolution result; Based on the second convolution layer, the first convolution result is convolved to obtain a second convolution result, and according to the second convolution result, a model output result of the arc fault detection model is determined and outputted; The arc fault detection result of the target circuit within the target time period is determined according to the model output result.

5. The method according to claim 4, characterized in that The arc fault detection model further includes at least two convolutional pooling submodules, and the at least two convolutional pooling submodules include at least a first submodule and a second submodule; The step of determining, according to the second convolution result, a model output result of the arc fault detection model for outputting includes: Based on the first submodule, performing convolution pooling processing on the second convolution result to obtain a first convolution pooling result; Based on the second submodule, the first convolution pooling result is subjected to convolution pooling processing to obtain a second convolution pooling result, and according to the second convolution pooling result, a model output result of the arc fault detection model is determined and outputted.

6. The method according to claim 5, characterized in that The number of the convolutional pooling submodules is five.

7. The method according to claim 1, characterized in that The current is collected for the target circuit within the target time period to obtain the time series current data, including: In the target time period, according to the target sampling rate, current is collected for the target circuit to obtain time series current data; After obtaining the time series current data, the method further includes: According to the preset current collection cycle and the target time period, the target time period is updated; Segmenting the time series current data according to a preset time window to obtain at least one segmented data; For each of the at least one segmented data, determine the number of current values ​​of the segmented data, and determine a margin factor of the segmented data according to the segmented data and the number of current values; The target sampling rate is updated according to the margin factors respectively corresponding to the at least one segmented data, and the step of collecting current for the target circuit within the target time period and at the target sampling rate to obtain the time series current data is returned to be executed.

8. The method according to claim 1, characterized in that The arc fault detection model is pre-trained by the following steps: Acquire sample time-series current data of a sample circuit and arc fault labels of the sample time-series current data, and determine a training sample set according to the sample time-series current data and the arc fault labels; Based on multiple training sample sets, the original detection model is trained to obtain the arc fault detection model.

9. The method according to claim 8, characterized in that The step of determining a training sample set according to the sample time series current data and the arc fault label comprises: Determine at least one timing starting point current value from the sample timing current data; For each of the at least one timing starting point current value, the sample timing current data is intercepted and processed according to the timing starting point current value and a preset interception time length to obtain enhanced data, and the enhanced data and the arc fault label are used as a group of training samples to obtain a training sample set based on the obtained multiple groups of training samples.

10. An arc fault detection device, characterized in that: include: A time-series current data acquisition module is used to collect current from a target circuit within a target time period to obtain time-series current data; A current characteristic image obtaining module, used for encoding the time series current data to obtain a current characteristic image; The arc fault detection result determination module is used to input the current characteristic image into the arc fault detection model, and determine the arc fault detection result of the target circuit within the target time period according to the model output result of the arc fault detection model.