Arc detection method and device

By performing frame-based window processing and calculation of the current signal and combining with the convolutional neural network model, the online rapid identification of arc faults is achieved, and the accuracy and speed of arc fault recognition in the existing technology is solved. , improves the safety of electrical equipment.

CN113970680BActive Publication Date: 2024-11-15BEIJING TRUTH WISDOM POWER TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202111237108.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-22
Publication Date
2024-11-15
Estimated Expiration
2041-10-22

AI Technical Summary

Technical Problem

The existing technology cannot accurately and quickly identify arc faults online, resulting in the occurrence and development of electrical fires, and poses safety hazards.

Method used

The current signal is used to frame and window the processing, calculate the Mel frequency cepspectral coefficient, and build a convolutional neural network model for identification and classification to determine whether an arc occurs in the circuit.

Benefits of technology

It realizes rapid and accurate online identification of arc faults, reduces the occurrence of electrical fires, and avoids personal casualties and economic losses.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113970680B_ABST
    Figure CN113970680B_ABST
Patent Text Reader

Abstract

The present application relates to the field of arc detection technology, and in particular to an arc detection method and device. The method comprises: collecting current signals, and performing frame segmentation and windowing processing on the collected current signals to obtain a plurality of frames of current signals; calculating the Mel-frequency cepstral coefficients of each frame of current signals, and constructing a Mel-frequency cepstral coefficient matrix based on the Mel-frequency cepstral coefficients of each frame of current signals; constructing a convolutional neural network model, and training the convolutional neural network model to obtain a trained convolutional neural network model; using the pre-trained convolutional neural network model to identify and classify the Mel-frequency cepstral coefficient matrix, and judging whether an arc occurs in the circuit based on the identification and classification results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of arc detection technology, and in particular to an arc detection method and device. Background Art

[0002] Electrical fires account for a high proportion of fire accidents in today's society, and arc faults are one of the important causes of electrical fires. Fault arcs are usually caused by aging and damage of insulation of lines and equipment, or poor electrical connections. When a fault arc occurs, the arc will release high temperature, which is extremely likely to cause a fire. The pressure, radiation, and arc root effect generated by the arc are not only easy to damage electrical equipment, but may also cause personal injury and serious economic losses. Arc faults are the main cause of electrical fire accidents. Therefore, online accurate and rapid identification of arc faults has important guiding significance for reducing electrical fires. Summary of the invention

[0003] The present application provides an arc detection method and device to solve the problem in the prior art that arc faults cannot be accurately and quickly identified online.

[0004] In a first aspect, the present application provides an arc detection method, the method comprising:

[0005] Collecting current signals, and performing frame division and windowing processing on the collected current signals to obtain a plurality of frame current signals;

[0006] Calculate the Mel-frequency cepstral coefficients of each frame of the current signal, and construct a Mel-frequency cepstral coefficient matrix according to the Mel-frequency cepstral coefficients of each frame of the current signal;

[0007] Constructing a convolutional neural network model, and training the convolutional neural network model to obtain a trained convolutional neural network model;

[0008] A pre-trained convolutional neural network model is used to identify and classify the Mel-frequency cepstral coefficient matrix, and whether an arc occurs in the circuit is determined based on the identification and classification results.

[0009] Optionally, before performing frame division and windowing processing on the collected current signal, the method further includes:

[0010] A high-pass filter is used to perform pre-emphasis processing on the collected current signal to obtain a pre-emphasized current signal;

[0011] The steps of performing frame division and windowing processing on the collected current signal to obtain a plurality of frame current signals include:

[0012] The pre-emphasized current signal is subjected to frame division and windowing processing to obtain a plurality of frame current signals.

[0013] Optionally, the step of calculating the Mel-frequency cepstral coefficients of each frame of the current signal includes:

[0014] A fast Fourier transform is performed on each frame of current signal, and a preset triangular bandpass filter is used to extract features from the transform results. A discrete cosine transform is performed on the features extracted by each filter to obtain the corresponding Mel-frequency cepstrum coefficients.

[0015] Optionally, the steps of constructing a two-dimensional convolutional neural network model and training the two-dimensional convolutional neural network model to obtain a trained two-dimensional convolutional neural network model include:

[0016] Build a convolutional neural network model and initialize weight parameters;

[0017] The sample data is input into the convolutional neural network model, and the sample data is forward propagated through the convolution layer, pooling layer, and fully connected layer to obtain the output value;

[0018] Calculating the error between the output value and the target value;

[0019] If the error is greater than the expected value, the error is passed back to the convolutional neural network model, and the errors of the fully connected layer, the pooling layer, and the convolutional layer are obtained in turn. The weights of each layer of the convolutional neural network model are adjusted according to the errors of each layer, and the sample data is input into the convolutional neural network model. The sample data is forward propagated through the convolutional layer, the pooling layer, and the fully connected layer to obtain the output value.

[0020] If the error is less than or equal to the expected value, it is determined that the training of the convolutional neural network model is completed, and a trained convolutional neural network model is obtained.

[0021] Optionally,

[0022] The overall structure of the convolutional neural network includes: an input layer, five stacked convolutional layers and pooling layers, and a fully connected layer.

[0023] In a second aspect, the present application provides an arc detection device, the device comprising:

[0024] The acquisition unit is used to acquire the current signal and perform frame division and windowing processing on the acquired current signal to obtain a plurality of frame current signals;

[0025] A calculation unit, used for calculating the Mel-frequency cepstral coefficients of each frame of the current signal, and constructing a Mel-frequency cepstral coefficient matrix according to the Mel-frequency cepstral coefficients of each frame of the current signal;

[0026] A construction unit, used to construct a convolutional neural network model and train the convolutional neural network model to obtain a trained convolutional neural network model;

[0027] The judgment unit is used to use a pre-trained convolutional neural network model to identify and classify the Mel-frequency cepstral coefficient matrix, and judge whether an arc occurs in the circuit according to the identification and classification results.

[0028] Optionally, before performing frame division and windowing processing on the collected current signal, the device further includes:

[0029] A pre-processing unit, used for performing pre-emphasis processing on the collected current signal by using a high-pass filter to obtain a pre-emphasized current signal;

[0030] When the collected current signal is subjected to frame division and windowing processing to obtain a plurality of frame current signals, the collection unit is specifically used to

[0031] The pre-emphasized current signal is subjected to frame division and windowing processing to obtain a plurality of frame current signals.

[0032] Optionally, when calculating the Mel-frequency cepstrum coefficients of each frame of the current signal, the calculation unit is specifically used to:

[0033] A fast Fourier transform is performed on each frame of current signal, and a preset triangular bandpass filter is used to extract features from the transform results. A discrete cosine transform is performed on the features extracted by each filter to obtain the corresponding Mel-frequency cepstrum coefficients.

[0034] Optionally, when constructing a two-dimensional convolutional neural network model and training the two-dimensional convolutional neural network model to obtain a trained two-dimensional convolutional neural network model, the construction unit is specifically used to:

[0035] Build a convolutional neural network model and initialize weight parameters;

[0036] The sample data is input into the convolutional neural network model, and the sample data is forward propagated through the convolution layer, pooling layer, and fully connected layer to obtain the output value;

[0037] Calculating the error between the output value and the target value;

[0038] If the error is greater than the expected value, the error is passed back to the convolutional neural network model, and the errors of the fully connected layer, the pooling layer, and the convolutional layer are obtained in turn. The weights of each layer of the convolutional neural network model are adjusted according to the errors of each layer, and the sample data is input into the convolutional neural network model. The sample data is forward propagated through the convolutional layer, the pooling layer, and the fully connected layer to obtain the output value.

[0039] If the error is less than or equal to the expected value, it is determined that the training of the convolutional neural network model is completed, and a trained convolutional neural network model is obtained.

[0040] Optionally,

[0041] The overall structure of the convolutional neural network includes: an input layer, five stacked convolutional layers and pooling layers, and a fully connected layer.

[0042] In a third aspect, an embodiment of the present application provides an arc detection device, the device comprising:

[0043] A memory for storing program instructions;

[0044] The processor is used to call the program instructions stored in the memory, and execute the steps of the method as described in any one of the first aspects above according to the obtained program instructions.

[0045] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable the computer to execute the steps of the method described in any one of the first aspects above.

[0046] In summary, the arc detection method provided in the embodiment of the present application collects current signals, and performs frame and window processing on the collected current signals to obtain several frames of current signals; calculates the Mel-frequency cepstral coefficients of each frame of the current signal, and constructs a Mel-frequency cepstral coefficient matrix based on the Mel-frequency cepstral coefficients of each frame of the current signal; constructs a convolutional neural network model, and trains the convolutional neural network model to obtain a trained convolutional neural network model; uses the pre-trained convolutional neural network model to identify and classify the Mel-frequency cepstral coefficient matrix, and determines whether an arc occurs in the circuit based on the identification and classification results.

[0047] By adopting the arc detection method provided in the embodiment of the present application, the Mel-frequency cepstral coefficient matrix features of the current signal are collected and extracted, and arc faults are identified online based on a convolutional neural network, so that it is possible to detect whether an arc is generated in the circuit online, quickly and accurately. This can reduce electrical fires, avoid personal injury and death accidents, and avoid electrical fires caused by arcs, thereby causing significant economic losses to enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments of the present application or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings of the embodiments of the present application.

[0049] Figure 1 A detailed flow chart of an arc detection method provided in an embodiment of the present application;

[0050] Figure 2 A schematic diagram of a MFCC extraction process provided in an embodiment of the present application;

[0051] Figure 3 A structural schematic diagram of an arc detection device is provided for an embodiment of the present application;

[0052] Figure 4 A schematic diagram of the structure of another arc detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, rather than limiting the present application. The singular forms of "a", "said" and "the" used in the present application and claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to any or all possible combinations of one or more associated listed items.

[0054] It should be understood that, although the terms first, second, third, etc. may be used to describe various information in the embodiments of the present application, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, in addition, the word "if" used may be interpreted as "at..." or "when..." or "in response to determination".

[0055] For example, see Figure 1 , which is a detailed flow chart of an arc detection method provided in an embodiment of the present application, the method comprises the following steps:

[0056] Step 100: Collect current signals, and perform frame division and windowing processing on the collected current signals to obtain a plurality of frames of current signals.

[0057] In the embodiment of the present application, after the current signal is collected and before the collected current signal is subjected to frame segmentation and windowing processing, the arc detection method may further include the following steps:

[0058] A high-pass filter is used to perform pre-emphasis processing on the collected current signal to obtain a pre-emphasized current signal.

[0059] Then, when performing frame division and windowing processing on the collected current signal to obtain several frames of current signals, a better implementation method is:

[0060] The pre-emphasized current signal is subjected to frame division and windowing processing to obtain a plurality of frame current signals.

[0061] Step 110: Calculate the Mel-frequency cepstral coefficients of each frame of the current signal, and construct a Mel-frequency cepstral coefficient matrix according to the Mel-frequency cepstral coefficients of each frame of the current signal.

[0062] Specifically, in the embodiment of the present application, when calculating the Mel-frequency cepstrum coefficients of each frame of the current signal, a preferred implementation method is:

[0063] A fast Fourier transform is performed on each frame of current signal, and a preset triangular bandpass filter is used to extract features from the transform results. A discrete cosine transform is performed on the features extracted by each filter to obtain the corresponding Mel-frequency cepstrum coefficients.

[0064] Step 120: construct a convolutional neural network model, and train the convolutional neural network model to obtain a trained convolutional neural network model.

[0065] In an embodiment of the present application, when constructing a two-dimensional convolutional neural network model and training the two-dimensional convolutional neural network model to obtain a trained two-dimensional convolutional neural network model, a preferred implementation method is:

[0066] Step 1: Build a convolutional neural network model and initialize the weight parameters.

[0067] Step 2: Input the sample data into the convolutional neural network model. The sample data is forward propagated through the convolutional layer, pooling layer, and fully connected layer to obtain the output value.

[0068] Step 3: Calculate the error between the output value and the target value.

[0069] Step 4: If the error is greater than the expected value, the error is passed back to the convolutional neural network model, and the errors of the fully connected layer, pooling layer, and convolutional layer are calculated in turn. The weights of each layer of the convolutional neural network model are adjusted according to the errors of each layer, and steps 2 and 3 are executed.

[0070] Step 5: If the error is less than or equal to the expected value, it is determined that the training of the convolutional neural network model is completed, and a trained convolutional neural network model is obtained.

[0071] Step 130: using a pre-trained convolutional neural network model to identify and classify the Mel-frequency cepstral coefficient matrix, and judging whether an arc occurs in the circuit according to the identification and classification results.

[0072] The arc detection method provided in the embodiment of the present application is described in detail below in conjunction with specific application scenarios.

[0073] In practical applications, in the process of realizing arc detection, the first step is to extract the features of the current signal. Preferably, the collected current signal can be represented by parameters with less redundant information. Then, the Mel-Frequency Cepstral Coefficients (MFCC) are used to extract the features (Mel-Frequency Cepstral Coefficient Matrix) of the collected current signal. For example, see Figure 2 , which is a flow chart of an MFCC extraction process provided in an embodiment of the present application.

[0074] The first step: collect current signals.

[0075] Step 2: Pre-emphasize the collected current signal.

[0076] Specifically, the sampled vibration signal S(z) is passed through a high-pass filter. As shown in formula (1):

[0077] H(z)=1-uz -1 (1)

[0078] Where H(z) is the Gaussian filter function, z is the input signal, and u is the weighting coefficient, which is between 0.9 and 1.0. The purpose of pre-emphasis is to enhance the high-frequency part of the signal, making the spectrum of the signal flat and maintaining it in the entire frequency band from low frequency to high frequency.

[0079] Step 3: Perform frame division and windowing processing on the current signal after pre-emphasis processing.

[0080] Specifically, N sampling points are first grouped into an observation unit, called a frame. The value of N is usually set to 256 or 512. In practical applications, electrical equipment uses a rated frequency of 50Hz, so the time unit covered is 20ms. In order to avoid signal loss between two adjacent frames, there is an overlapping area between two adjacent frames during sampling. The overlapping area contains M sampling points, and the value of M is usually set to 1 / 2 or 1 / 3 of N. In an embodiment of the present application, the signal sampling rate can be set to 48kHz, and the number of sampling points is set to 1024, then the corresponding sampling time is approximately 21ms. Each frame is then multiplied by a Hamming window to increase the continuity of the left and right ends of the frame.

[0081] S′(z)=S(z)×W(z)(2)

[0082] Among them, is the signal after S′(z) is framed, and W(z) is the Hamming window function.

[0083]

[0084] Different a values ​​will produce different Hamming windows. Generally, a is 0.46.

[0085] Step 4: Perform fast Fourier transform on each frame of current signal, and use a preset triangular bandpass filter to extract features from the transformation results.

[0086] Specifically, it is difficult to distinguish the differences between different signal characteristics by analyzing the changes of the signal in the time domain. In the process of digital signal processing, the signal is usually converted from the changes in the time domain to the energy distribution in the frequency domain to observe the characteristics of different signals. Therefore, after the signal is windowed, it needs to be subjected to a fast Fourier transform to obtain the energy distribution of the signal in the spectrum. The frequency spectrum of each frame after frame division and windowing is obtained by fast Fourier transform. And the power spectrum of the vibration signal is obtained by taking the modulus square of the frequency spectrum of each frame. Assume that the DFT of the current signal is:

[0087]

[0088] Where j is an imaginary unit and k is a natural integer.

[0089] Next, the power spectrum is passed through a group of Mel-scale filter groups, usually a series of triangular filters arranged in a cross-overlapping manner in the low-frequency region. Due to the nonlinear correspondence between the Hz-Mel frequency, there are more filters in the low-frequency region, so the low-order MFCC coefficients will weaken the spectrum information in the high-frequency region. Usually, the number of MFCC filters K is set to 9 or 13. In the embodiment of the present application, in order to ensure the calculation accuracy of the low, medium and high frequency bands on the one hand, and to cover the entire frequency domain on the other hand without missing any spectrum information, 26 filters are selected for extraction.

[0090] According to formula (5):

[0091]

[0092] Where f is the maximum frequency of the vibration signal, from which the maximum Mel frequency f can be obtained mel , because within the Mel scale range, the center frequencies of each triangular filter are equally spaced linearly distributed. Therefore, the spacing between the center frequencies of two adjacent triangular filters can be calculated as

[0093]

[0094] Step 5: Perform discrete cosine transform on the Mel features corresponding to each frame of current signal.

[0095] Specifically, the logarithmic energy of each filter group output is calculated, and the calculation process is as shown in formula (7). m (h) is the frequency response of the m-th filter, and h is a discrete digital variable.

[0096]

[0097] The MFCC coefficients are then obtained through discrete cosine transform (DCT).

[0098]

[0099] Substitute the above logarithmic energy into discrete cosine transform to obtain the MFCC parameters of order L. In the embodiment of the present application, the value of L is 16.

[0100] In order to detect arc signals, the current signal generated by the circuit is collected during the experiment, and the current signal is divided into frames in units of 20ms. The Mel frequency cepstral coefficient (MFCC) of each frame signal is calculated, and the Mel frequency cepstral coefficient (MFCC) of all frames is formed into a (1872, 26) MFCC matrix, as shown below.

[0101]

[0102] Furthermore, the Mel-frequency cepstral coefficient feature matrix is ​​input into a pre-trained convolutional neural network model, and the convolutional neural network model recognizes and classifies the Mel-frequency cepstral coefficient matrix, and determines whether an arc occurs in the circuit based on the recognition and classification results.

[0103] In practical applications, convolutional neural networks (CNNs) have stronger adaptability to two-dimensional signals with complex background environments and unclear features. Convolutional neural networks are feedforward neural networks composed of multiple layers of convolutional layers, pooling layers, and fully connected layers. The convolutional layer perceives local features of different sizes of data through convolution kernels of different sizes, weights and averages the local features through the convolution kernel parameters, and then integrates the local information at a higher level to obtain the global signal features. The pooling layer downsamples the high-dimensional data and scales or reconstructs the sample size, which can better aggregate features and reduce the amount of calculation. Commonly used pooling methods include maximum pooling and average pooling. The fully connected layer weights all neuron nodes between layers to fuse local features into global features. Since the MFCC coefficients can characterize the characteristics of the current signal in the frequency domain, and the convolutional neural network can extract deeper features in the frequency domain of the current signal, the softmax classifier can map high-dimensional features to the probability of the category to which they belong. In an embodiment of the present application, based on the above principles, a classification model combining Mel-Frequency Cepstral Coefficients (MFCC) and a convolutional neural network is proposed, and the MFCC two-dimensional matrix composed of the collected current signal is used to classify whether an arc occurs in the circuit, and whether an arc fault occurs in the circuit is detected according to the classification result.

[0104] In an embodiment of the present application, an FD-CNN network is proposed based on the LD-CNN network. The FD-CNN network removes the second consecutive convolutional layer in the LD-CNN network, and directly enters the output of the first convolutional layer into the pooling layer for data dimensionality reduction. At the same time, in order to retain more information in the process of network deepening, in an embodiment of the present application, it is preferred to set the convolutional layer step size to 1, the maximum pooling layer step size to 2, and finally use a fully connected layer to replace the FeatureSum layer in the LD-CNN. For example, the overall structure of the FD-CNN network includes: an input layer, five stacked convolutional layers and pooling layers, and a fully connected layer. The convolution kernel uses a 3×3 matrix, and the activation function uses a relu function.

[0105] In the embodiment of the present application, the training process of the convolutional neural network model is as follows:

[0106] (1) Build a convolutional neural network model and initialize the weights;

[0107] (2) The input data is forward propagated through the convolutional layer, pooling layer, and fully connected layer to obtain the output value;

[0108] (3) Calculate the error between the network output value and the target value;

[0109] (4) When the error is greater than the expected value, the error is transmitted back to the network, and the errors of the network's fully connected layer, pooling layer, and convolutional layer are calculated in turn. When the error is less than or equal to the given expected value, the training ends and proceeds to step 6.

[0110] (5) Adjust the weights of each layer of the network according to the obtained error and enter step 2 again;

[0111] (6) After training, high-level feature data is output.

[0112] Based on the same inventive concept as the above method embodiment, for example, refer to Figure 3 FIG. 1 is a schematic diagram of the structure of an arc detection device provided in an embodiment of the present application, the device comprising:

[0113] The acquisition unit 30 is used to acquire the current signal and perform frame division and windowing processing on the acquired current signal to obtain a plurality of frames of current signals;

[0114] A calculation unit 31 is used to calculate the Mel-frequency cepstral coefficients of each frame of the current signal, and construct a Mel-frequency cepstral coefficient matrix according to the Mel-frequency cepstral coefficients of each frame of the current signal;

[0115] A construction unit 32 is used to construct a convolutional neural network model and train the convolutional neural network model to obtain a trained convolutional neural network model;

[0116] The judgment unit 33 is used to use a pre-trained convolutional neural network model to identify and classify the Mel-frequency cepstral coefficient matrix, and judge whether an arc occurs in the circuit according to the identification and classification results.

[0117] Optionally, before performing frame division and windowing processing on the collected current signal, the device further includes:

[0118] A pre-processing unit, used for performing pre-emphasis processing on the collected current signal by using a high-pass filter to obtain a pre-emphasized current signal;

[0119] When the collected current signal is subjected to frame division and windowing processing to obtain a plurality of frames of current signals, the collection unit 30 is specifically used to

[0120] The pre-emphasized current signal is subjected to frame division and windowing processing to obtain a plurality of frame current signals.

[0121] Optionally, when calculating the Mel-frequency cepstral coefficients of each frame of the current signal, the calculation unit 31 is specifically used to:

[0122] A fast Fourier transform is performed on each frame of current signal, and a preset triangular bandpass filter is used to extract features from the transform results. A discrete cosine transform is performed on the features extracted by each filter to obtain the corresponding Mel-frequency cepstrum coefficients.

[0123] Optionally, when constructing a two-dimensional convolutional neural network model and training the two-dimensional convolutional neural network model to obtain a trained two-dimensional convolutional neural network model, the construction unit 32 is specifically used to:

[0124] Build a convolutional neural network model and initialize weight parameters;

[0125] The sample data is input into the convolutional neural network model, and the sample data is forward propagated through the convolution layer, pooling layer, and fully connected layer to obtain the output value;

[0126] Calculating the error between the output value and the target value;

[0127] If the error is greater than the expected value, the error is passed back to the convolutional neural network model, and the errors of the fully connected layer, the pooling layer, and the convolutional layer are obtained in turn. The weights of each layer of the convolutional neural network model are adjusted according to the errors of each layer, and the sample data is input into the convolutional neural network model. The sample data is forward propagated through the convolutional layer, the pooling layer, and the fully connected layer to obtain the output value.

[0128] If the error is less than or equal to the expected value, it is determined that the training of the convolutional neural network model is completed, and a trained convolutional neural network model is obtained.

[0129] Optionally,

[0130] The overall structure of the convolutional neural network includes: an input layer, five stacked convolutional layers and pooling layers, and a fully connected layer.

[0131] The above units may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more digital singnal processors (DSP), or one or more field programmable gate arrays (FPGA). For another example, when a certain unit above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processors that can call program codes. For another example, these units may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0132] Furthermore, the arc detection device provided in the embodiment of the present application, from the hardware level, the hardware architecture diagram of the arc detection device can be seen in Figure 4 As shown, the arc detection device may include: a memory 40 and a processor 41,

[0133] The memory 40 is used to store program instructions; the processor 41 calls the program instructions stored in the memory 40 and executes the above method embodiment according to the obtained program instructions. The specific implementation method and technical effect are similar and will not be repeated here.

[0134] Optionally, the present application also provides an arc detection device, comprising at least one processing element (or chip) for executing the above method embodiment.

[0135] Optionally, the present application also provides a program product, such as a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are used to enable the computer to execute the above method embodiments.

[0136] Here, the machine-readable storage medium may be any electronic, magnetic, optical or other physical storage device that may contain or store information, such as executable instructions, data, etc. For example, the machine-readable storage medium may be: RAM (RadomAccess Memory), volatile memory, non-volatile memory, flash memory, storage drive (such as hard disk drive), solid state drive, any type of storage disk (such as CD, DVD, etc.), or similar storage medium, or a combination thereof.

[0137] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which may be in the form of a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email transceiver, a game console, a tablet computer, a wearable device or a combination of any of these devices.

[0138] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0139] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the embodiments of the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0140] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0141] Moreover, these computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which is implemented in the process Figure 1 A process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0142] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0143] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. An arc detection method, characterized in that: The method comprises: Collecting current signals, and performing frame division and windowing processing on the collected current signals to obtain a plurality of frame current signals; Calculate the Mel-frequency cepstral coefficients of each frame of the current signal, and construct a Mel-frequency cepstral coefficient matrix according to the Mel-frequency cepstral coefficients of each frame of the current signal; Constructing a convolutional neural network model, and training the convolutional neural network model to obtain a trained convolutional neural network model; Using a pre-trained convolutional neural network model, the Mel-frequency cepstral coefficient matrix is ​​identified and classified, and whether an arc occurs in the circuit is determined according to the identification and classification results; Before performing frame division and windowing processing on the collected current signal, the method further includes: A high-pass filter is used to perform pre-emphasis processing on the collected current signal to obtain a pre-emphasized current signal; The steps of performing frame division and windowing processing on the collected current signal to obtain a plurality of frame current signals include: Performing frame-by-frame windowing processing on the current signal after the pre-emphasis processing to obtain a plurality of frame current signals; The pre-emphasis processing of the collected current signal includes the following steps: The sampled vibration signal S(z) is passed through a high-pass filter; The high-pass filter is configured with a Gaussian filter function, and the Gaussian filter function is: H(z)=1-uz -1 ; Among them, H(z) is the Gaussian filter function, z is the input signal, and u is the weighting coefficient, whose value is between 0.9 and 1.

0.

2. The method according to claim 1, characterized in that The steps of calculating the Mel frequency cepstrum coefficients of each frame of current signal include: A fast Fourier transform is performed on each frame of current signal, and a preset triangular bandpass filter is used to extract features from the transform results. A discrete cosine transform is performed on the features extracted by each filter to obtain the corresponding Mel-frequency cepstrum coefficients.

3. The method according to claim 1, characterized in that The steps of constructing a two-dimensional convolutional neural network model and training the two-dimensional convolutional neural network model to obtain the trained two-dimensional convolutional neural network model include: Build a convolutional neural network model and initialize weight parameters; The sample data is input into the convolutional neural network model, and the sample data is forward propagated through the convolution layer, pooling layer, and fully connected layer to obtain the output value; Calculating the error between the output value and the target value; If the error is greater than the expected value, the error is passed back to the convolutional neural network model, and the errors of the fully connected layer, the pooling layer, and the convolutional layer are obtained in turn. The weights of each layer of the convolutional neural network model are adjusted according to the errors of each layer, and the sample data is input into the convolutional neural network model. The sample data is forward propagated through the convolutional layer, the pooling layer, and the fully connected layer to obtain the output value. If the error is less than or equal to the expected value, it is determined that the training of the convolutional neural network model is completed, and a trained convolutional neural network model is obtained.

4. The method according to claim 3, characterized in that The overall structure of the convolutional neural network includes: an input layer, five stacked convolutional layers and pooling layers, and a fully connected layer.

5. An arc detection device, characterized in that: The device comprises: The acquisition unit is used to acquire the current signal and perform frame division and windowing processing on the acquired current signal to obtain a plurality of frames of current signals; A calculation unit, used for calculating the Mel-frequency cepstral coefficients of each frame of the current signal, and constructing a Mel-frequency cepstral coefficient matrix according to the Mel-frequency cepstral coefficients of each frame of the current signal; A construction unit, used to construct a convolutional neural network model and train the convolutional neural network model to obtain a trained convolutional neural network model; A judgment unit, used to use a pre-trained convolutional neural network model to identify and classify the Mel-frequency cepstral coefficient matrix, and judge whether an arc occurs in the circuit according to the identification and classification results; Before performing frame division and windowing processing on the collected current signal, the device further includes: A pre-processing unit, used for performing pre-emphasis processing on the collected current signal by using a high-pass filter to obtain a pre-emphasized current signal; When the collected current signal is subjected to frame-by-frame windowing processing to obtain a plurality of frame current signals, the collection unit is specifically used to perform frame-by-frame windowing processing on the pre-emphasized current signal to obtain a plurality of frame current signals; When the acquisition unit performs pre-emphasis processing on the acquired current signal, it is used to pass the sampled vibration signal S(z) through a high-pass filter; The high-pass filter is configured with a Gaussian filter function, and the Gaussian filter function is: H(z)=1-uz -1 ; Among them, H(z) is the Gaussian filter function, z is the input signal, and u is the weighting coefficient, whose value is between 0.9 and 1.

0.

6. The device according to claim 5, characterized in that When calculating the Mel-frequency cepstrum coefficients of each frame of the current signal, the calculation unit is specifically used to: A fast Fourier transform is performed on each frame of current signal, and a preset triangular bandpass filter is used to extract features from the transform results. A discrete cosine transform is performed on the features extracted by each filter to obtain the corresponding Mel-frequency cepstrum coefficients.

7. The device according to claim 5, characterized in that When constructing a two-dimensional convolutional neural network model and training the two-dimensional convolutional neural network model to obtain a trained two-dimensional convolutional neural network model, the construction unit is specifically used to: Build a convolutional neural network model and initialize weight parameters; The sample data is input into the convolutional neural network model, and the sample data is forward propagated through the convolution layer, pooling layer, and fully connected layer to obtain the output value; Calculating the error between the output value and the target value; If the error is greater than the expected value, the error is passed back to the convolutional neural network model, and the errors of the fully connected layer, the pooling layer, and the convolutional layer are obtained in turn. The weights of each layer of the convolutional neural network model are adjusted according to the errors of each layer, and the sample data is input into the convolutional neural network model. The sample data is forward propagated through the convolutional layer, the pooling layer, and the fully connected layer to obtain the output value. If the error is less than or equal to the expected value, it is determined that the training of the convolutional neural network model is completed, and a trained convolutional neural network model is obtained.

8. The device according to claim 7, characterized in that The overall structure of the convolutional neural network includes: an input layer, five stacked convolutional layers and pooling layers, and a fully connected layer.

Citation Information

Patent Citations

  • Method and system for detecting arc faults in power circuits

    JP2014509396A

  • Classification method for quality of power signal using speech processing techniques, monitoring apparatus for quality of power signal using speech processing techniques and monitoring system for quality of power signal using speech processing techniques

    KR101299183B1

  • Electrical load determination apparatus and method using modified mfcc

    WO2019182166A1