A series fault arc detection method, system, device and readable storage medium

Abstract

The present invention proposes a method, system, device and readable storage medium for detecting a series fault arc, which belongs to the technical field of fault arc detection. The method includes: iteratively training an autoencoder, and generating a denoising autoencoder by adding noise to the original input of the autoencoder during the training process; constructing a residual network model by introducing a residual network module based on a two-dimensional convolutional neural network; preprocessing the arc experimental data obtained from the experiment using a sliding window, and obtaining sample data; processing the sample data using a denoising autoencoder to obtain a corresponding feature grayscale image; performing pseudo-color processing on the feature grayscale image to generate a corresponding pseudo-color image; inputting the pseudo-color image into the residual network model for deep training to generate a series fault arc recognition model; obtaining arc data of electrical equipment in real time, and inputting it into the series fault arc recognition model to generate a series fault arc detection result.

Description

Technical Field

[0001] The present invention relates to the technical field of arc fault detection, and more particularly to a series arc fault detection method, system, device and readable storage medium. Background Art

[0002] Long-term operation of electrical equipment and lines is prone to insulation damage, line breakage, poor electrical contact and other prominent problems. In major fire accidents, fires caused by electrical faults are becoming more frequent, of which arc faults account for nearly 50%. Arc faults can be divided into parallel, series and grounding types. Parallel and grounding arc faults will generate large short-circuit currents or ground leakage currents, which can be prevented by circuit breakers and leakage protectors. However, series arc faults are difficult to be protected or detected by existing protection equipment because of their small fault current and complex and changeable fault waveforms affected by the load.

[0003] In the early detection of series arc faults, arc light, arc sound, electromagnetic radiation and other physical quantities generated by the arc were used to detect arc faults. However, due to the complexity of the environment in which the electrical lines are located, this method has great limitations and is mostly used for arc detection at fixed points such as switch cabinets and distribution boxes. In recent years, series fault arc detection has mostly used the time domain, frequency domain or time-frequency domain characteristics of the fault current signal, and combined with certain classification or machine learning algorithms to identify the fault arc. The detection methods of time domain information include mathematical statistical indicators such as zero rest time, current change rate, skewness, kurtosis and high-order cumulants.

[0004] The commonality of existing arc identification methods is to use existing mathematical methods to extract appearance features from normal and fault current waveforms, and analyze the differences between the two in one or more appearance features to detect series arc faults. Its advantage is that the classification process is more intuitive, but it also causes limitations in the detection method and cannot reveal the deeper fault characteristics of the arc. Summary of the invention

[0005] In view of the above problems, the purpose of the present invention is to provide a method, system, device and readable storage medium for detecting series fault arcs, which extracts the deep features of the series fault arc waveform through a denoising autoencoder, and then uses a deep residual network to learn and identify the fault features, thereby realizing preliminary detection of series fault arcs.

[0006] To achieve the above object, the present invention is implemented through the following technical solutions: A series fault arc detection method, comprising:

[0007] Iteratively train the autoencoder, and generate a denoising autoencoder by adding noise to the original input of the autoencoder during the training process;

[0008] Based on the two-dimensional convolutional neural network, a residual network model is constructed by introducing the residual network module;

[0009] Preprocessing the arc experimental data obtained from the experiment using a sliding window, and obtaining sample data; the sample data includes sample data with normal experimental results and sample data with arc faults;

[0010] The sample data is processed using a denoising autoencoder to obtain the corresponding feature grayscale image;

[0011] Perform pseudo-color processing on the characteristic grayscale image to generate a corresponding pseudo-color image;

[0012] The pseudo-color image is input into the residual network model for deep training to generate a series fault arc recognition model;

[0013] The arc data of the electrical equipment is acquired in real time and input into the series fault arc recognition model to generate the series fault arc detection results.

[0014] Further, the arc experimental data obtained by the experiment is preprocessed by using a sliding window, and sample data is obtained, including:

[0015] The arc experimental data obtained by the experiment are acquired in batches by using a sliding window to capture the local features of the arc experimental data;

[0016] The experimental data were normalized using maximum and minimum standardization.

[0017] Further, the iterative training of the autoencoder and generating a denoising autoencoder by adding noise to the original input of the autoencoder during the training process include:

[0018] According to the formula Through the encoder f EN maps the input sample X to the hidden layer and obtains its encoded feature value Z; where is the encoder activation function, W EN is the encoder weight matrix, b EN is the encoder bias;

[0019] After the encoding is completed, according to the formula Through the decoder f DE Reconstruct the data of the hidden layer encoding feature Z; where, is the activation function of the decoder ，W DE is the decoder weight matrix, b DE is the decoder bias;

[0020] Based on the autoencoder, random dropout regularization is added to its input layer or noise is superimposed on the input sample to serve as the input X′ of the autoencoder, and a denoising autoencoder is generated through iterative training;

[0021] The learning objective of the denoising autoencoder is set to minimize the deviation between the reconstructed output sample Y and the original input X without superimposed noise according to the following formula:

[0022] ;

[0023] Among them, X′ is the original input data with superimposed noise, N F is the noise superposition factor, X N It follows a standard normal distribution.

[0024] Furthermore, the method of processing the sample data using the denoising autoencoder to obtain the corresponding feature grayscale image includes:

[0025] Input sample data into the denoising autoencoder for training;

[0026] By setting the dimension of the hidden layer of the denoising autoencoder to 50, each sample data can obtain a 50-dimensional deep feature with a value between [0,1];

[0027] The sample data is expanded to the interval [0,255] to draw the characteristic grayscale images corresponding to normal situations and arc faults.

[0028] Further, the pseudo-color processing of the characteristic grayscale image to generate a corresponding pseudo-color image includes:

[0029] Input the characteristic grayscale image into the three primary color channels of red, green and blue respectively;

[0030] By synthesizing the RGB values ​​output by the three primary color channels of red, green, and blue, a pseudo-color image corresponding to the characteristic grayscale image is generated.

[0031] Furthermore, the residual network model is constructed by introducing a residual network module based on a two-dimensional convolutional neural network, including:

[0032] The inner product of the overlapping part of the input layer data information of the two-dimensional convolutional neural network and its two-dimensional convolution kernel is calculated by the following formula, and the characteristic output value is obtained through the nonlinear activation function:

[0033] ;

[0034] in, is the i-th feature map of the k-th layer, is the i-th convolution kernel matrix of the k-th layer, is the k-th layer output of the network, is the bias value of the kth convolutional layer; is the ReLU activation function, and its expression is as follows:

[0035] ;

[0036] The feature output value is pooled using the following formula:

[0037] ;

[0038] in, is the value of the kth pooling layer at (i, j); Represents pooling rules, including maximum pooling, random pooling, and average pooling; is the receptive field at (i, j); is the corresponding input at the receptive field (m, n);

[0039] The fully connected layer of the two-dimensional convolutional neural network is used to further extract the features of the pooling layer to output the model results, and the normalized exponential function is used to complete the multi-classification task of the data;

[0040] Based on the architecture of a two-dimensional convolutional neural network, a residual network module is introduced, and convolution blocks are skipped through shortcut connections to avoid the gradient vanishing phenomenon;

[0041] Through the identity mapping The input X of the two-dimensional convolutional neural network is short-circuited to the output Y of the parameter layer, and the parameter layer obtains the corresponding residual map .

[0042] Furthermore, the pseudo-color image is input into the residual network model for deep training to generate a series fault arc recognition model, including:

[0043] The pseudo-color image is input into the residual network model, and the RGB values ​​output by the red, green and blue primary color channels of the pseudo-color image are trained through deep learning through the residual network model to generate a series fault arc recognition model.

[0044] Correspondingly, the present invention also discloses a series fault arc detection system, comprising:

[0045] An encoder optimization module is configured to iteratively train the autoencoder and generate a denoising autoencoder by adding noise to the original input of the autoencoder during the training process;

[0046] A residual network model building module, configured to build a residual network model based on a two-dimensional convolutional neural network by introducing a residual network module;

[0047] A sample acquisition module is configured to pre-process the arc experimental data obtained from the experiment using a sliding window and acquire sample data;

[0048] A feature extraction module is configured to process the sample data using a denoising autoencoder to obtain a corresponding feature grayscale image;

[0049] A data processing module is configured to perform pseudo-color processing on the characteristic grayscale image to generate a corresponding pseudo-color image;

[0050] A model training module configured to input the pseudo-color image into the residual network model for deep training to generate a series fault arc recognition model;

[0051] The real-time identification module is configured to obtain arc data of the electrical equipment in real time and input the data into the series fault arc identification model to generate a series fault arc detection result.

[0052] Accordingly, the present invention discloses a series fault arc detection device, comprising:

[0053] A memory, used for storing a series fault arc detection program;

[0054] A processor is used to implement the steps of the series fault arc detection method as described in any one of the above when executing the series fault arc detection program.

[0055] Accordingly, the present invention discloses a readable storage medium, on which a series fault arc detection program is stored. When the series fault arc detection program is executed by a processor, the steps of the series fault arc detection method as described in any one of the above items are implemented.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. The present invention uses a sliding window to obtain batch samples from normal and fault arc signals obtained in experiments, and normalizes them, which effectively prevents the denoising autoencoder and residual network model from generating feature bias during learning and training.

[0058] 2. The present invention utilizes the strong feature extraction capability of the denoising autoencoder to extract high-dimensional abstract features from normal and fault arc current signals, and no longer utilizes the time-frequency domain features of the fault arc current. The feature values ​​obtained by the denoising autoencoder are converted into grayscale images, the grayscale images are processed in pseudo-color, and finally the three-channel images are input into the residual network model training to improve its resolution capability.

[0059] 3. The present invention extracts the deep-level features of the series fault arc waveform through a denoising autoencoder, and then uses a deep residual network to learn and identify the fault features to achieve preliminary detection of series fault arcs. The present invention can extract deep-level fault features in the fault arc in addition to the surface features, overcoming the limitations of traditional series fault arc detection methods and the problem of low detection accuracy.

[0060] 4. The present invention can solve the problem that series arc faults are difficult to detect and the learning degradation problem caused by conventional deep learning, improve the detection rate of series arc faults, and also have a high detection rate for loads that are not involved in model training, with strong generalization ability.

[0061] It can be seen that compared with the prior art, the present invention has outstanding substantive features and significant progress, and the beneficial effects of its implementation are also obvious. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0063] Figure 1 It is a method flow chart of a specific implementation mode of the present invention.

[0064] Figure 2 It is a structural principle diagram of a denoising autoencoder according to a specific embodiment of the present invention.

[0065] Figure 3 It is a schematic diagram of the structure of a two-dimensional convolutional neural network according to a specific embodiment of the present invention.

[0066] Figure 4 It is a schematic diagram of the residual block structure of the residual network model according to a specific implementation manner of the present invention.

[0067] Figure 5 It is a characteristic grayscale image of the load when the load works normally according to the specific embodiment of the present invention.

[0068] Figure 6 It is a characteristic grayscale image when a series arc fault is sent according to a specific embodiment of the present invention.

[0069] Figure 7 yes Figure 5 Corresponding pseudo-color image.

[0070] Figure 8 yes Figure 6 Corresponding pseudo-color image.

[0071] Fig. 9 It is a schematic diagram of a series arc experimental circuit according to a specific implementation mode of the present invention.

[0072] Fig.10 It is a schematic diagram of the error rate of ResNet on the training set according to a specific embodiment of the present invention.

[0073] Fig.11 It is a schematic diagram of the error rate of ResNet on the test set according to a specific embodiment of the present invention.

[0074] Fig.12 It is a system structure diagram of a specific implementation mode of the present invention.

[0075] In the figure, 1. Encoder optimization module; 2. Residual network model construction module; 3. Sample acquisition module; 4. Feature extraction module; 5. Data processing module; 6. Model training module; 7. Real-time recognition module. DETAILED DESCRIPTION

[0076] See also Figure 1 As shown, this embodiment provides a series fault arc detection method, comprising the following steps:

[0077] S1: Iteratively train the autoencoder and generate a denoising autoencoder by adding noise to the original input of the autoencoder during the training process.

[0078] Specifically, for machine classification learning algorithms, if the input data contains unseparated noise, it may cause overfitting of the classification algorithm, resulting in a decrease in the classification effect and generalization ability of the algorithm. The reason is that the machine learning algorithm will also regard the noise contained in it as a classification feature of the data. The traditional autoencoder AE can achieve good results after iterative training, but it does not have the ability to reduce noise. Therefore, it is necessary to introduce a denoising autoencoder DAE to improve the generalization ability of the model.

[0079] In a specific embodiment, the AE is first trained by passing an encoder Map the input sample X to the hidden layer and obtain its encoded feature value Z, as shown in the following formula:

[0080] ;

[0081] in, is the encoder activation function, WEN is the encoder weight matrix, and bEN is the encoder bias.

[0082] After the encoding is completed, the decoder fDE is used to reconstruct the hidden layer encoding feature Z. As shown in the following formula:

[0083] ;

[0084] in, is the activation function of the decoder ，W DE is the decoder weight matrix, b DE is the decoder bias.

[0085] The principle of denoising autoencoder is as follows Figure 2 As shown in the figure, based on AE, random dropout regularization is added to the input layer or noise is superimposed on the input sample to serve as the input X′ of the traditional AE. This method chooses to implement DAE by superimposing noise on the original input, as shown in the following formula. The DAE learning goal is to minimize the deviation between the reconstructed output sample Y and the original input X without superimposed noise.

[0086] ;

[0087] Where X′ is the original input data with superimposed noise, N F is the noise superposition factor, X N It follows a standard normal distribution.

[0088] After the above processing, the features learned by DAE from the input of superimposed noise are highly robust and can retain more relevant information in the data latent space while filtering out irrelevant content such as noise.

[0089] S2: Based on the two-dimensional convolutional neural network, a residual network model is constructed by introducing the residual network module.

[0090] Among them, the two-dimensional convolutional neural network (CNN) is a typical method of deep learning, which contains two dimensions, height and width, and extracts image features through multi-layer convolution operations. The CNN structure is as follows: Figure 3 As shown in the figure, it consists of a convolutional layer, a pooling layer and a fully connected layer, and the convolutional layer is used to extract local features.

[0091] In a specific implementation, the inner product of the overlapping part of the input layer data information and the two-dimensional convolution kernel is first calculated, and then a nonlinear activation function is passed to obtain the characteristic output value, as shown in the following formula. The activation function usually uses a rectified linear unit (ReLU), which can clear some neuron outputs to speed up training.

[0092] ;

[0093] in, is the i-th feature map of the k-th layer, is the i-th convolution kernel matrix of the k-th layer, is the k-th layer output of the network, is the bias value of the kth convolutional layer; is the ReLU activation function, and its expression is as follows:

[0094] ;

[0095] The pooling layer is located after the convolution layer. It has two main functions: first, it reduces the dimensionality of the extracted high-dimensional features to improve the computational efficiency and avoid overfitting; second, it ensures the invariance of the features and improves the generalization ability of the model. The feature output values ​​are pooled using the following formula:

[0096] ;

[0097] In the formula, is the value of the kth pooling layer at (i, j); Represents pooling rules, including maximum pooling, random pooling, and average pooling; is the receptive field at (i, j); is the corresponding input at the receptive field (m, n).

[0098] After convolution and pooling processing, the features of the pooling layer are further extracted by the fully connected layer to output the model results, and finally the normalized exponential function (Softmax) is used to complete the multi-classification task of the data.

[0099] Based on the CNN architecture, the residual network model ResNet of this method draws on the cross-layer connection idea of ​​the high-speed network and introduces the residual network module (RBB). It skips the convolution block through the shortcut connection to avoid the gradient disappearance phenomenon, so that the input X of the neural network is mapped by the identity. Short-circuit connected to the output Y of the parameter layer, the parameter layer gets the corresponding residual map .

[0100] Among them, the residual block structure of the residual network model is as follows Figure 4 As shown in the figure, CONV represents the convolution layer; BN represents the de-mean normalization to reduce the difference between different features. The left and right figures are the standard residual block and the residual block with downsampling layer, respectively. Compared with CNN, the parameter layer learns the direct mapping between input and output. , learn the residual mapping between the two It can effectively reduce the difficulty of network learning and break the convention that the output of the nth layer of the traditional neural network can only be used as the input of the n+1th layer.

[0101] Since ResNet has a high recognition rate at different network training depths, the arc recognition rate can reach 99.7% at an ultra-high-level training depth of 152 layers. And for loads that have not participated in ResNet network training, the recognition rate can also reach 97.6%, showing good generalization ability. Therefore, by constructing this residual network, the training depth of ResNet can reach hundreds or thousands of layers without network performance degradation.

[0102] S3: Use a sliding window to preprocess the arc experimental data obtained from the experiment and obtain sample data.

[0103] In a specific implementation, in order to prevent overfitting and decreased generalization ability of the learning model, the arc experimental data is processed using a sliding window, which can capture local features while retaining the time dependency of time series data while obtaining batch data. For neural networks and deep learning, if the value of a certain input dimension is high, it will cause the neural network to be overly biased towards such features during learning. Therefore, after obtaining the arc experimental data obtained from the experiment in batches, the maximum and minimum standardization is used to normalize the experimental data and use it as sample data.

[0104] The sample data include sample data with normal experimental results and sample data with arc faults.

[0105] S4: Use the denoising autoencoder to process the sample data to obtain the corresponding feature grayscale image.

[0106] In a specific implementation, first, the normalized sample data is input into the denoising autoencoder for training. By setting the dimension of the hidden layer of the denoising autoencoder to 50, each sample data can obtain a 50-dimensional deep feature with a value between [0,1]; the sample data is expanded to the interval [0,255] to draw the corresponding feature grayscale images of normal conditions and arc faults.

[0107] S5: Perform pseudo-color processing on the characteristic grayscale image to generate a corresponding pseudo-color image.

[0108] In a specific implementation, computer vision recognition is similar to the visual sense of the human eye, and the resolution ability of color images is much higher than that of grayscale images. In order to extract more arc fault features from the feature map, the grayscale image obtained is enhanced with pseudo color. Therefore, the feature grayscale image is first input into the three primary color channels of red, green, and blue (RGB) with different change characteristics, and then the RGB values ​​output by each channel are synthesized to obtain the pseudo color image corresponding to the feature grayscale image. Among them, the feature grayscale image and the corresponding pseudo color image are as follows Figure 5-Figure 8 shown.

[0109] S6: Input the pseudo-color image into the residual network model for deep training to generate a series fault arc recognition model.

[0110] In a specific implementation, the pseudo-color image is input into a residual network model, and the RGB values ​​output by the red, green and blue primary color channels of the pseudo-color image are subjected to deep learning training through the residual network model to generate a series fault arc recognition model.

[0111] S7: Acquire arc data of electrical equipment in real time and input it into a series fault arc recognition model to generate a series fault arc detection result.

[0112] In a specific implementation, the arc current signal of the electrical equipment refers to the data processing process of steps S3-S5, and the processed data is input into the series fault arc recognition model to generate a series fault arc detection result.

[0113] Based on the above series fault arc detection method, the present invention also discloses a series fault arc detection method, which is as follows:

[0114] 1. Construction of arc experimental device:

[0115] Arc test equipment Fig. 9 As shown in the figure, it is mainly composed of 220V AC power supply, current transformer, load, arc generator, data acquisition module and computer, etc. The arc generator is made according to the international UL1699 standard, including movable pointed copper electrode, fixed flat graphite electrode, insulating rod, metal slide rail base and stepper motor, etc.

[0116] The experimental process is as follows: after the power supply and the load to be tested are turned on, the stepper motor is controlled to move the copper electrode to generate a stable burning arc, the main current is collected by the current transformer, and the AD7606 high-precision 16-bit ADC and the STM32 microcontroller are used to complete the analog-to-digital conversion and data collection, and then the fault and normal current data are uploaded to the computer for further analysis. Considering the actual benefits, the sampling frequency of the current transformer should not be too high, while a low sampling frequency will cause the burial of fault information and bring difficulties to arc identification. After comprehensive consideration, 256 points are sampled per current cycle, that is, the sampling frequency is set to 12.8kHz. According to the current waveform characteristics of the load, it can be divided into two categories: linear and nonlinear loads. The current waveform of the linear load presents an ideal sine wave, and the current waveform of the nonlinear load is a non-sinusoidal wave with periodic distortion. Considering the parallel use of loads in practice and the different positions of the main and branch circuits where the fault arc is located, a variety of arc experimental schemes are designed.

[0117] 2. DAE feature extraction:

[0118] The loads are numbered, and the normal and fault current signals are collected for 5 power frequency cycles respectively. According to the sliding window data processing method, 1025 samples of 1280-256+1 can be obtained for the normal and fault conditions of each load. The normal and fault conditions correspond to 5125 samples respectively, and the total number of samples is 10250. The normalized sample data is input into DAE for training, and the dimension of the hidden layer of DAE is set to 50. Each sample can obtain a 50-dimensional deep feature with a value between [0,1]. Expand it to the interval [0,255] to obtain the corresponding feature grayscale image. The grayscale image is input into the red, green, and blue (RGB) three primary color channels with different change characteristics, and then the RGB values ​​output by each channel are synthesized to obtain the pseudo-color image corresponding to the grayscale image.

[0119] 3. ResNet network training:

[0120] After using the previous data processing method and DAE to extract the normal and fault arc features, a total of 10,250 samples were obtained, of which 8,000 samples were used as the training set of ResNet and 2,250 samples were used as the test set. Each sample is a 1×50-dimensional RGB three-primary color image, which is cut and reorganized into 5×10 dimensions, and then the 5×10-dimensional graphics of the three RGB channels are used as the input of ResNet. The network learning depth of ResNet is set to 18, 50, and 152 layers respectively (referring to the convolutional layer and the fully connected layer, excluding the pooling layer). The batch sample size (BatchSize) of the dataset used for training is set to 50. At different training depths of 18, 50, and 152, the error rate of ResNet on the training set is as follows Fig.10 As shown, the error rate of ResNet on the test set is Fig.11 shown.

[0121] It can be seen that the series fault arc detection method disclosed in the present invention is a series fault arc detection method based on denoising autoencoder (DAE) and deep residual network (ResNet), which is characterized in that firstly, a large number of training samples are obtained from the experimental data by using sliding window and data normalization method, and then the fault and normal samples collected in the experiment are processed by DAE to obtain high-dimensional abstract feature data of each sample; secondly, the obtained features are converted into grayscale images and then processed by pseudo-color, and the single-channel grayscale image is mapped to the color value of three channels. Through ResNet, a large number of RGB values ​​of the three channels are deep-learned and trained, and the obtained training model can realize the preliminary recognition of series fault arcs. This method can solve the problem that series arc faults are difficult to detect and the learning degradation problem caused by conventional deep learning, improve the detection rate of series arc faults, and have a high detection rate for loads that do not participate in model training, and have strong generalization ability.

[0122] See also Fig.12 As shown, the present invention also discloses a series fault arc detection system, including: an encoder optimization module 1, a residual network model construction module 2, a sample acquisition module 3, a feature extraction module 4, a data processing module 5, a model training module 6 and a real-time recognition module 7.

[0123] The encoder optimization module 1 is configured to iteratively train the autoencoder and generate a denoising autoencoder by adding noise to the original input of the autoencoder during the training process.

[0124] The residual network model construction module 2 is configured to construct a residual network model based on a two-dimensional convolutional neural network by introducing a residual network module.

[0125] The sample acquisition module 3 is configured to pre-process the arc experimental data obtained from the experiment by using a sliding window and acquire sample data.

[0126] The feature extraction module 4 is configured to process the sample data using a denoising autoencoder to obtain a corresponding feature grayscale image.

[0127] The data processing module 5 is configured to perform pseudo-color processing on the characteristic grayscale image to generate a corresponding pseudo-color image.

[0128] The model training module 6 is configured to input the pseudo-color image into the residual network model for deep training to generate a series fault arc recognition model.

[0129] The real-time identification module 7 is configured to obtain arc data of the electrical equipment in real time, and input the data into the series fault arc identification model to generate a series fault arc detection result.

[0130] The specific implementation of the series fault arc detection system of this embodiment is basically the same as the specific implementation of the series fault arc detection method described above, and will not be repeated here.

[0131] The present invention also discloses a series fault arc detection device, comprising a processor and a memory; wherein the processor implements the steps of the series fault arc detection method as described in any one of the above items when executing the series fault arc detection program stored in the memory.

[0132] Furthermore, the series fault arc detection device in this embodiment may also include:

[0133] The input interface is used to obtain the series fault arc detection program imported from the outside, and save the obtained series fault arc detection program to the memory, and can also be used to obtain various instructions and parameters transmitted by the external terminal device, and transmit them to the processor, so that the processor can use the above various instructions and parameters to carry out corresponding processing. In this embodiment, the input interface can specifically include but is not limited to a USB interface, a serial interface, a voice input interface, a fingerprint input interface, a hard disk reading interface, etc.

[0134] The output interface is used to output various data generated by the processor to the terminal device connected thereto, so that other terminal devices connected to the output interface can obtain various data generated by the processor. In this embodiment, the output interface may specifically include but is not limited to a USB interface, a serial interface, etc.

[0135] The communication unit is used to establish a remote communication connection between the series fault arc detection device and the external server so that the series fault arc detection device can mount the image file to the external server. In this embodiment, the communication unit may specifically include but is not limited to a remote communication unit based on wireless communication technology or wired communication technology.

[0136] The keyboard is used to obtain various parameter data or instructions input by the user by tapping the keycaps in real time.

[0137] A display is used to display relevant information of the series fault arc detection process in real time.

[0138] The mouse can be used to assist users in inputting data and simplify user operations.

[0139] The present invention also discloses a readable storage medium, wherein the readable storage medium includes a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable hard disk, a CD-ROM or any other form of storage medium known in the technical field. The readable storage medium stores a series fault arc detection program, and when the series fault arc detection program is executed by a processor, the steps of the series fault arc detection method described in any one of the above items are implemented.

[0140] In summary, the present invention extracts the deep features of the series fault arc waveform through a denoising autoencoder, and then uses a deep residual network to learn and identify the fault features to achieve preliminary detection of series fault arcs.

[0141] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. As for the method disclosed in the embodiment, since it corresponds to the system disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0142] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0143] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or units, which can be electrical, mechanical or other forms.

[0144] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0145] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, or each module may exist physically separately, or two or more modules may be integrated into one unit.

[0146] Similarly, each processing unit in each embodiment of the present invention may be integrated into one functional module, or each processing unit may exist physically, or two or more processing units may be integrated into one functional module.

[0147] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0148] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0149] The above is a detailed introduction to the series fault arc detection method, system, device and readable storage medium provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A method for detecting a series fault arc, characterized in that: include: Iteratively train the autoencoder, and generate a denoising autoencoder by adding noise to the original input of the autoencoder during the training process; Based on the two-dimensional convolutional neural network, a residual network model is constructed by introducing the residual network module; Preprocessing the arc experimental data obtained from the experiment using a sliding window, and obtaining sample data; the sample data includes sample data with normal experimental results and sample data with arc faults; The sample data is processed using a denoising autoencoder to obtain the corresponding feature grayscale image; Perform pseudo-color processing on the characteristic grayscale image to generate a corresponding pseudo-color image; The pseudo-color image is input into the residual network model for deep training to generate a series fault arc recognition model; Acquire arc data of electrical equipment in real time and input it into the series fault arc recognition model to generate series fault arc detection results; The iterative training of the autoencoder and generating a denoising autoencoder by adding noise to the original input of the autoencoder during the training process include: According to the formula Through the encoder Map the input sample X to the hidden layer to obtain its encoded feature value Z; where, is the encoder activation function, is the encoder weight matrix, is the encoder bias; After the encoding is completed, according to the formula Through the decoder Reconstruct the data of the hidden layer encoding feature Z; where, is the activation function of the decoder, is the decoder weight matrix, is the decoder bias; Based on the autoencoder, random dropout regularization is added to its input layer or noise is superimposed on the input sample to serve as the input X′ of the autoencoder, and a denoising autoencoder is generated through iterative training; The learning objective of the denoising autoencoder is set to minimize the deviation between the reconstructed output sample Y and the original input X without superimposed noise according to the following formula: ； Among them, X′ is the original input data with superimposed noise, N F is the noise superposition factor, X N Obey standard normal distribution; The method of processing the sample data by using the denoising autoencoder to obtain the corresponding feature grayscale image includes: Input sample data into the denoising autoencoder for training; By setting the dimension of the hidden layer of the denoising autoencoder to 50, each sample data can obtain a 50-dimensional deep feature with a value between [0,1]; Expand the 50-dimensional deep features to the interval [0,255] to draw the corresponding feature grayscale images under normal conditions and when an arc fault occurs; The step of performing pseudo-color processing on the characteristic grayscale image to generate a corresponding pseudo-color image includes: Input the characteristic grayscale image into the three primary color channels of red, green and blue respectively; By synthesizing the RGB values ​​output by the three primary color channels of red, green and blue, a pseudo-color image corresponding to the characteristic grayscale image is generated; The residual network model is constructed by introducing a residual network module based on a two-dimensional convolutional neural network, including: The inner product of the overlapping part of the input layer data information of the two-dimensional convolutional neural network and its two-dimensional convolution kernel is calculated by the following formula, and the characteristic output value is obtained through the nonlinear activation function: ； in, is the i-th feature map of the k-th layer, is the i-th convolution kernel matrix of the k-th layer, is the k-th layer output of the network, is the bias value of the kth convolutional layer; is the ReLU activation function, and its expression is as follows: ； The feature output value is pooled using the following formula: ； in, is the value of the kth pooling layer at (i, j); Represents pooling rules, including maximum pooling, random pooling, and average pooling; is the receptive field at (i, j); is the corresponding input at the receptive field (m, n); The fully connected layer of the two-dimensional convolutional neural network is used to further extract the features of the pooling layer to output the model results, and the normalized exponential function is used to complete the multi-classification task of the data; Based on the architecture of a two-dimensional convolutional neural network, a residual network module is introduced, and convolution blocks are skipped through shortcut connections to avoid the gradient vanishing phenomenon; Through the identity mapping The input X of the two-dimensional convolutional neural network is short-circuited to the output Y of the parameter layer, and the parameter layer obtains the corresponding residual map ; The method of preprocessing the arc experimental data obtained by the experiment using a sliding window and obtaining sample data includes: The arc experimental data obtained by the experiment are acquired in batches by using a sliding window to capture the local features of the arc experimental data; The experimental data were normalized using maximum and minimum standardization.

2. The method for detecting series fault arc according to claim 1, characterized in that: The method of inputting the pseudo-color image into the residual network model for deep training to generate a series fault arc recognition model includes: The pseudo-color image is input into the residual network model, and the RGB values ​​output by the red, green and blue primary color channels of the pseudo-color image are trained through deep learning through the residual network model to generate a series fault arc recognition model.

3. A series fault arc detection system, characterized in that: The system adopts the series fault arc detection method as described in any one of claims 1 to 2; The system comprises: An encoder optimization module is configured to iteratively train the autoencoder and generate a denoising autoencoder by adding noise to the original input of the autoencoder during the training process; A residual network model building module, configured to build a residual network model based on a two-dimensional convolutional neural network by introducing a residual network module; A sample acquisition module is configured to pre-process the arc experimental data obtained from the experiment using a sliding window and acquire sample data; A feature extraction module is configured to process the sample data using a denoising autoencoder to obtain a corresponding feature grayscale image; A data processing module is configured to perform pseudo-color processing on the characteristic grayscale image to generate a corresponding pseudo-color image; A model training module configured to input the pseudo-color image into the residual network model for deep training to generate a series fault arc recognition model; The real-time identification module is configured to obtain arc data of the electrical equipment in real time and input the data into the series fault arc identification model to generate a series fault arc detection result.

4. A series fault arc detection device, characterized in that: include: A memory, used for storing a series fault arc detection program; A processor is used to implement the steps of the series fault arc detection method as claimed in any one of claims 1 to 2 when executing the series fault arc detection program.

5. A readable storage medium, characterized in that: The readable storage medium stores a series fault arc detection program, and when the series fault arc detection program is executed by a processor, the steps of the series fault arc detection method according to any one of claims 1 to 2 are implemented.

Images