Arc fault identification method and system based on multi-feature fusion

Abstract

The invention provides an arc fault identification method and system based on multi-feature fusion, and relates to the technical field of arc fault detection. The method comprises the following steps: acquiring a current signal generated in an operation process of to-be-detected electrical equipment; denoising is carried out on the current signal; the denoised current signals are sent to a combined channel neural network; wherein the first channel neural network is a convolutional long-short-term memory network and is used for extracting spatial-temporal characteristics of the current signals; the second channel neural network is an FFT-ELM network and is used for extracting depth features of the current signals; and performing feature fusion on the spatial-temporal features and the depth features through a feature fusion layer in a combined channel neural network, calculating the fused features by using a trained full connection layer, and outputting an arc fault recognition result. According to the invention, accurate identification of the arc fault can be realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of arc fault detection, and in particular relates to an arc fault identification method and system based on multi-feature fusion. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Arc is a gas discharge phenomenon. When current passes through certain insulating media (such as air), it will produce instantaneous sparks. The generation of arc is due to the escape of electrons from the metal surface of the contact and the ionization of gas atoms or molecules in the gap. The arc has high temperature and high heat characteristics, and its temperature can reach 3000℃~4000℃, accompanied by the splashing of molten metal, which can easily ignite the line insulation layer and surrounding combustibles, thereby causing a fire; therefore, in the power system, arc fault detection and identification is a key issue to ensure the safe operation of the system.

[0004] For the detection and identification of arc faults, the traditional detection method is to build a monitoring system based on the physical characteristics of the fault arc, such as electromagnetic radiation, arc light and noise, and use capture equipment to capture specific physical characteristics, such as using a Hilbert fractal antenna to capture the electromagnetic radiation of a DC arc, and use this to determine whether an arc fault has occurred. This identification method has strong interpretability and high accuracy; however, it can only detect arc faults at fixed locations and is easily affected by environmental factors; therefore, its universality is poor. Another commonly used method for fault arc identification is to detect the fault arc by analyzing the time domain and frequency domain characteristics of the fault arc, such as directly using Fourier transform to extract the frequency domain characteristics of the arc, and using wavelet transform to realize the identification of the fault arc. This type of method can completely extract the intrinsic characteristics that can characterize the fault arc, but the final detection result still requires manual judgment, which is inefficient. With the continuous development of machine learning, more and more technicians in this field have begun to use machine learning related models for fault arc detection and identification. However, the inventors found that there are still the following technical problems, such as: after obtaining the current or voltage signal of the electrical equipment, directly input it into the machine learning model for detection and identification, and learn and extract features through a single model to finally complete the identification and diagnosis of the fault arc. This method is relatively simple and easy to implement, but lacks secondary processing of the signal, resulting in the current or voltage signal used for learning cannot fully characterize its inherent characteristics; in addition, the model is too single, which makes the model's feature extraction ability limited, and ultimately leads to a low accuracy rate for the identification of the fault arc. Summary of the invention

[0005] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides an arc fault identification method and system based on multi-feature fusion, which can realize accurate identification of arc faults.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: A first aspect of the present invention provides an arc fault identification method based on multi-feature fusion.

[0007] An arc fault identification method based on multi-feature fusion, comprising: Acquire the current signal generated during the operation of the electrical equipment to be tested; De-noising the current signal; The denoised current signal is sent to the combined channel neural network; the first channel neural network is a convolutional long short-term memory network, which is used to extract the spatiotemporal characteristics of the current signal; the second channel neural network is an FFT-ELM network, which is used to extract the depth characteristics of the current signal; The spatiotemporal features are fused with the depth features through the feature fusion layer in the combined channel neural network, the fused features are calculated using the trained fully connected layer, and the arc fault recognition result is output.

[0008] Furthermore, the convolutional long short-term memory network includes a convolution layer, a pooling layer and an LSTM layer; wherein the convolution layer is used to extract the spatial characteristics of the current signal, the pooling layer is used to further reduce the feature dimension, and the LSTM layer is used to extract the temporal characteristics of the current signal.

[0009] Furthermore, the convolutional long short-term memory network is used to extract the spatiotemporal characteristics of the current signal, including: The convolution layer performs a convolution operation on the input current signal window. The convolution kernel slides on the signal and extracts the spatial features of the current signal through weighted summation and activation function processing. After the dimension of the feature map is further reduced by the pooling layer, the temporal features of the current signal are extracted by the internal gate structure of the LSTM layer.

[0010] After the spatial features and temporal features are extracted, the LSTM outputs of different time steps are fused through a fully connected layer to form an abstract representation of the entire signal window; the abstract representation includes the spatial features and temporal features of the current signal.

[0011] Furthermore, the FFT-ELM network extracts the deep features of the current signal, including: The current signals of different frequencies are decomposed by discrete Fourier transform FFT, key samples are selected from the decomposed current signals, and the key samples are sent to the ELM model; the ELM model includes an input layer, a hidden layer and an output layer; wherein the key samples are received by the input layer, the key samples are learned and deep features are extracted by the hidden layer, and then sent out through the output layer.

[0012] Furthermore, when selecting key samples from the decomposed current signal, the current amplitudes of the DC component and the harmonic component are selected as the characteristic levels, that is, the current peak value and phase angle of each frequency band are selected as the key samples.

[0013] Furthermore, a nonlinear transformation is required before calculating the fused features using the fully connected layer, namely: ; in, Represents the output after the activation function Sigmoid is applied. represents the output of the fully connected layer, Represents the larger value between the output S of the fully connected layer and the value 0.

[0014] A second aspect of the present invention provides an arc fault identification system based on multi-feature fusion.

[0015] An arc fault identification system based on multi-feature fusion, comprising: The data acquisition module is configured to: obtain the current signal generated during the operation of the electrical equipment to be detected; A denoising module is configured to: denoise the current signal; The feature extraction module is configured to: send the denoised current signal to a combined channel neural network; wherein the first channel neural network is a convolutional long short-term memory network, which is used to extract the spatiotemporal features of the current signal; and the second channel neural network is an FFT-ELM network, which is used to extract the depth features of the current signal; The recognition module is configured to: fuse the spatiotemporal features with the depth features through a feature fusion layer in a combined channel neural network, calculate the fused features using a trained fully connected layer, and output an arc fault recognition result.

[0016] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of an arc fault identification method based on multi-feature fusion as described in the first aspect of the present invention.

[0017] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of an arc fault identification method based on multi-feature fusion as described in the first aspect of the present invention are implemented.

[0018] One or more of the above technical solutions have the following beneficial effects: The present invention provides an arc fault identification method based on multi-feature fusion, by sending the denoised current signal into a combined channel neural network; wherein the first channel neural network is a convolutional long short-term memory network, which can learn both the temporal dependency of the signal and the spatial characteristics of the signal; the second channel neural network is an FFT-ELM network, which is used to extract the deep features of the current signal. Therefore, the model constructed in this application can simultaneously learn the temporal features, spatial features and deep features of the signal, solving the problem of limited feature extraction capability caused by the common use of a single model in the prior art, and greatly improving the recognition accuracy of arc faults.

[0019] The present invention adopts a combined channel neural network for parallel processing. In the second channel neural network, the current signals of different frequencies are decomposed by discrete Fourier transform FFT, key samples are selected from the decomposed current signals, and the current amplitudes of the DC component and the harmonic component are selected as feature levels and then input into the ELM network. This solves the problem in the prior art that the signal to be detected is directly input into the detection model, resulting in poor recognition effect due to the lack of secondary processing of the signal, and further improves the recognition accuracy of arc faults.

[0020] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0022] Figure 1 This is a flow chart of an arc fault identification method based on multi-feature fusion in Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0023] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0024] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0025] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0026] Embodiment 1 This embodiment discloses an arc fault identification method based on multi-feature fusion.

[0027] like Figure 1 As shown, an arc fault identification method based on multi-feature fusion includes: Step S1, obtaining a current signal generated during the operation of the electrical equipment to be detected; Step S2, denoising the current signal; Step S3, sending the denoised current signal to a combined channel neural network; wherein the first channel neural network is a convolutional long short-term memory network, which is used to extract the spatiotemporal characteristics of the current signal; and the second channel neural network is an FFT-ELM network, which is used to extract the depth characteristics of the current signal; Step S4: Fusing the spatiotemporal features with the depth features through the feature fusion layer in the combined channel neural network, calculating the fused features using the trained fully connected layer, and outputting arc fault recognition results.

[0028] In step S1, the current signal generated during the operation of the electrical equipment to be detected is obtained, which can be achieved in the following way.

[0029] Use a multimeter or high-precision ammeter or other measuring equipment to connect the positive and negative poles or corresponding measuring terminals of the measuring equipment to the current loop of the electrical equipment to be tested, while ensuring safety. It should be noted that when connecting the measuring equipment, it is necessary to select an appropriate measuring range to avoid equipment damage or inaccurate measurement. In the selection process, it is necessary to consider both the current range of the electrical equipment to be tested and the measuring range of the measuring equipment to be selected. For example: assuming that the rated current of the electrical equipment is 10A, the maximum operating current may reach 12A, and considering certain current fluctuations and overload conditions, a multimeter or high-precision ammeter with a measuring range of 20A or more can be selected for measurement. This can ensure the accuracy of the measurement and provide a certain safety margin.

[0030] After the current signal is acquired by using the selected measuring device, step S2 is executed to denoise the current signal.

[0031] When denoising the current signal, median filtering, low-pass filtering, high-pass filtering, band-stop filtering, wavelet transform denoising or sliding average method can be used for denoising. In view of the good suppression effect of median filtering on pulse noise (such as salt and pepper noise) and the ability to effectively protect the edge information of the signal, the median filtering method is selected in this embodiment to denoise the current signal in step S1, specifically: The size of the selected filter window is 5*5. For each point in the current signal, take it as the center point, extract the values ​​of all points in the selected filter window and sort these values ​​from small to large. Then take the median, that is: if there are an odd number of elements in the filter window, take the value of the element in the middle position after sorting as the new value of the center point; if there are an even number of elements in the filter window, usually take the average of the two middle element values ​​after sorting as the new value of the center point. Finally, update the signal, that is: replace the original value of the center point with the calculated median, repeat the above process for each point in the signal until all points have been processed; at this point, denoising is completed.

[0032] When denoising is completed, step S3 is executed to send the denoised current signal to the combined channel neural network; wherein the first channel neural network is a convolutional long short-term memory network, which is used to extract the spatiotemporal characteristics of the current signal; and the second channel neural network is an FFT-ELM network, which is used to extract the depth characteristics of the current signal.

[0033] Specifically, the convolutional long short-term memory network C-LSTM includes 2 convolutional layers, 2 pooling layers and 1 LSTM layer; wherein, the convolution kernel size of the two convolutional layers is set to 2*2, and the number of the two convolution kernels is set to 16 and 24 respectively, which are used to extract the spatial features of the current signal; the pooling window size of the pooling layer is also set to 2*2, and the step size is also 2, which is used to further reduce the feature dimension; the LSTM layer is used to extract the temporal features of the current signal.

[0034] Furthermore, the convolution layer performs a convolution operation on the input current signal window, and the convolution kernel slides on the signal. The spatial features of the current signal are extracted through the weighted summation and activation function processing; wherein the convolution operation is performed using the ReLU activation function, that is: ; ; in, represents the activation function, Represents the convolutional layer i Middle k The output of the filter, represents convolution; Represents the convolutional layer The data set, Indicates The convolution kernel in the layer Inputs, Indicates The input and The first The convolution kernel in the layer, It means the The offset of the input.

[0035] Subsequently, the output of the convolutional layer enters the pooling layer, which is used to further reduce the feature dimension, namely: ;in, Represents maximum pooling.

[0036] After the dimension of the feature map is further reduced by the above pooling layer, the time series features of the current signal are extracted by the internal gate structure of the LSTM layer.

[0037] In the internal gate structure of the LSTM layer, the forget gate determines which information is discarded from the cell state of the previous time step. It receives the hidden state of the previous time step and the input of the current time step, and outputs a value between 0 and 1 through the sigmoid function. This value determines which information in the cell state of the previous time step is retained. The internal gate structure of the LSTM layer consists of an input gate, a forget gate, and an output gate; among them, the input gate determines which new information is added to the cell state. It also receives the hidden state of the previous time step and the input of the current time step, determines which values ​​will be updated through the sigmoid function, and generates a new candidate value vector through the tanh function. Then, the output of the forget gate and the input gate is combined to update the cell state. The output gate of the LSTM layer determines what information is output from the hidden state of the current time step. It receives the hidden state of the previous time step and the input of the current time step, determines which parts of the cell state will be output through the sigmoid function, and after processing the cell state through the tanh function, it is multiplied with the output of the sigmoid function to obtain the final hidden state output.

[0038] After the spatial features and temporal features are extracted, the LSTM outputs of different time steps are fused through the fully connected layer to form an abstract representation of the entire signal window; the abstract representation includes the spatial features and temporal features of the current signal; thus, the extraction of the spatiotemporal features of the current signal is completed.

[0039] It should be noted that when the denoised current signal is sent to the combined channel neural network, the current signal is sent to the first channel neural network and the second channel neural network at the same time. The above process has used the first channel neural network to complete the extraction of the spatiotemporal characteristics of the current signal. The following describes how to use the second channel neural network to extract the deep features of the current signal.

[0040] The FFT-ELM network is an ELM neural network model based on discrete Fourier transform. As an embodiment, the FFT-ELM network extracts the deep features of the current signal, including: The current signals of different frequencies are decomposed by discrete Fourier transform FFT, key samples are selected from the decomposed current signals, and the key samples are sent to the ELM model; the ELM model includes an input layer, a hidden layer and an output layer; wherein the key samples are received by the input layer, the key samples are learned and deep features are extracted by the hidden layer, and then sent out through the output layer.

[0041] Furthermore, the Fourier transform formula is expressed as: ; in, A discrete signal representing the original current signal, Represents the number of discrete signal sampling points, Represents the spectrum of the original current signal after discrete Fourier transform.

[0042] Furthermore, when selecting key samples from the decomposed current signal, the current amplitudes of the DC component and the harmonic component are selected as feature levels, that is, the current peak value and phase angle of each frequency band are used as key samples and these key samples are input into the ELM model.

[0043] The ELM model consists of an input layer, a hidden layer, and an output layer. When using the ELM model to solve the output value (i.e., obtain the deep features), the parameters from the input layer to the hidden layer are calculated randomly, while the parameters from the hidden layer to the output layer are calculated using the generalized inverse matrix. Assume that the sample input to the hidden layer is Total , that is ; Then, the ELM model can be expressed as: ; in, represents the output value of the ELM model, Indicates the number of nodes in the hidden layer of the ELM model, represents the connection weight between the hidden layer and the output layer, represents the activation function, represents the connection weight between the input layer and the output layer, that is ; Indicates the first The bias of hidden units, ” indicates the inner product operation.

[0044] After extracting the deep features of the current signal based on the above-mentioned FFT-ELM network, step S4 is executed to fuse the time series features with the deep features through the feature fusion layer in the combined channel neural network, and the fused features are calculated using the trained fully connected layer to output the arc fault recognition result.

[0045] Furthermore, when the spatiotemporal features and the deep features are fused by the feature fusion layer in the combined channel neural network, the data features extracted by the first channel neural network and the second channel neural network are spliced ​​through the cross attention mechanism to form the fused features, specifically: first, in the cross attention mechanism, the first channel features (spatial-temporal features) are usually used as queries, and the second channel features (depth features) are used as keys and values; of course, more complex strategies can also be adopted, such as alternately using different features as queries, keys and values, which are not limited here. Subsequently, the attention weights are calculated, the dot product is performed using the query and the key, and the results are normalized to the attention weights through the softmax function, which represent the correlation between different positions or features. Finally, the attention weights are multiplied by the values, and the results are weighted summed to obtain the fused feature representation; this process is actually a weighted fusion of different features.

[0046] Furthermore, taking PyTorch as an example, in the specific implementation, we can assume that query, key, and value represent query, key, and value, respectively, which are all three-dimensional tensors (batch size, sequence length, feature dimension); use torch.matmul or torch.einsum to perform matrix multiplication operations to calculate attention weights; apply torch.softmax to normalize the attention weights, and finally use the normalized attention weights to perform weighted summation on value.

[0047] After the first channel feature and the second channel feature are fused, the trained fully connected layer is used to calculate the fused features. It should be noted that a nonlinear transformation needs to be performed before the fully connected layer is used to calculate the fused features, namely: ; in, Represents the output after the activation function Sigmoid is applied. represents the output of the fully connected layer, Represents the larger value between the output S of the fully connected layer and the value 0.

[0048] The fused new features are input into the fully connected layer for nonlinear feature learning. Before feature learning, the number of neurons in the fully connected layer is set to 64. Finally, the probability of arc fault is calculated by the Sigmoid function. If the output arc fault probability is greater than 70%, it is considered that an arc fault exists; if the output arc fault probability does not exceed 70%, it is considered that no arc fault exists.

[0049] Embodiment 2 This embodiment discloses an arc fault identification system based on multi-feature fusion, including: The data acquisition module is configured to obtain the current signal generated during the operation of the electrical equipment to be detected.

[0050] The denoising module is configured to: denoise the current signal.

[0051] The feature extraction module is configured to: send the denoised current signal to a combined channel neural network; wherein the first channel neural network is a convolutional long short-term memory network, which is used to extract the spatiotemporal features of the current signal; and the second channel neural network is an FFT-ELM network, which is used to extract the deep features of the current signal.

[0052] The recognition module is configured to: fuse the time series features with the depth features through the feature fusion layer in the combined channel neural network, calculate the fused features using the trained fully connected layer, and output the arc fault recognition result. Embodiment 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0053] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of an arc fault identification method based on multi-feature fusion as described in the first embodiment of the present disclosure.

[0054] Embodiment 4 The purpose of this embodiment is to provide an electronic device.

[0055] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of an arc fault identification method based on multi-feature fusion as described in the first embodiment of the present disclosure are implemented.

[0056] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0057] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0058] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for arc fault identification based on multi-feature fusion, characterized in that: include: Acquire the current signal generated during the operation of the electrical equipment to be tested; De-noising the current signal; The denoised current signal is sent to the combined channel neural network; the first channel neural network is a convolutional long short-term memory network, which is used to extract the spatiotemporal characteristics of the current signal; the second channel neural network is an FFT-ELM network, which is used to extract the depth characteristics of the current signal; The spatiotemporal features are fused with the depth features through the feature fusion layer in the combined channel neural network, the fused features are calculated using the trained fully connected layer, and the arc fault recognition result is output.

2. The arc fault identification method based on multi-feature fusion according to claim 1, characterized in that: A convolutional long short-term memory network is used to extract the spatiotemporal characteristics of the current signal, wherein the convolutional long short-term memory network includes a convolution layer, a pooling layer and an LSTM layer; Among them, the convolution layer is used to extract the spatial features of the current signal, the pooling layer is used to further reduce the feature dimension, and the LSTM layer is used to extract the temporal features of the current signal.

3. The arc fault identification method based on multi-feature fusion according to claim 2, characterized in that: The convolutional long short-term memory network is used to extract the spatiotemporal characteristics of the current signal, including: The convolution layer performs a convolution operation on the input current signal window. The convolution kernel slides on the signal and extracts the spatial features of the current signal through weighted summation and activation function processing. After the dimension of the feature map is further reduced by the pooling layer, the temporal features of the current signal are extracted by the internal gate structure of the LSTM layer.

4. The arc fault identification method based on multi-feature fusion according to claim 3, characterized in that: The convolutional long short-term memory network is used to extract the spatiotemporal characteristics of the current signal, including: After the spatial features and temporal features are extracted, the LSTM outputs of different time steps are fused through a fully connected layer to form an abstract representation of the entire signal window; the abstract representation includes the spatial features and temporal features of the current signal.

5. The arc fault identification method based on multi-feature fusion according to claim 1, characterized in that: The FFT-ELM network extracts deep features of the current signal, including: The current signals of different frequencies are decomposed by discrete Fourier transform FFT, key samples are selected from the decomposed current signals, and the key samples are sent to the ELM model; the ELM model includes an input layer, a hidden layer and an output layer; wherein the key samples are received by the input layer, the key samples are learned and deep features are extracted by the hidden layer, and then sent out through the output layer.

6. The arc fault identification method based on multi-feature fusion according to claim 5, characterized in that: When selecting key samples from the decomposed current signal, the current amplitudes of the DC component and the harmonic component are selected as the characteristic levels, that is, the current peak value and phase angle of each frequency band are selected as the key samples.

7. The arc fault identification method based on multi-feature fusion according to claim 1, characterized in that: Before using the fully connected layer to calculate the fused features, a nonlinear transformation is required, namely: ； in, Represents the output after the activation function Sigmoid is applied. represents the output of the fully connected layer, Represents the larger value between the output S of the fully connected layer and the value 0.

8. An arc fault identification system based on multi-feature fusion, characterized in that: include: The data acquisition module is configured to: obtain the current signal generated during the operation of the electrical equipment to be detected; A denoising module is configured to: denoise the current signal; The feature extraction module is configured to: send the denoised current signal to a combined channel neural network; wherein the first channel neural network is a convolutional long short-term memory network, which is used to extract the spatiotemporal features of the current signal; and the second channel neural network is an FFT-ELM network, which is used to extract the depth features of the current signal; The recognition module is configured to: fuse the spatiotemporal features with the depth features through a feature fusion layer in a combined channel neural network, calculate the fused features using a trained fully connected layer, and output an arc fault recognition result.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the arc fault identification method based on multi-feature fusion as described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the arc fault identification method based on multi-feature fusion as described in any one of claims 1 to 7 are implemented.

Images