High-resistance small-current grounding line selection method and system based on a full convolutional network

Abstract

This invention provides a method and system for selecting high-resistivity, low-current grounding faults based on a fully convolutional network, relating to the field of digital power technology. This invention transforms the high-resistivity, low-current grounding fault selection problem into an image recognition problem. By converting the spectral components of the transient current traveling wave of a fault into a spatiotemporal image, the grounding fault selection problem is transformed into an image recognition and classification problem with a mature learning model, reducing the difficulty of model training and improving the accuracy of fault selection. The fully convolutional network achieves pixel-level analysis in the three-dimensional spatiotemporal image through transposed convolutional layers. The fully convolutional network only needs to compare the polarity and amplitude characteristics of the transient current traveling wave through image comparison to complete the high-resistivity, low-current grounding fault selection; it features accurate and fast image analysis, combining high efficiency, practicality, convenience, and accuracy.

Description

Technical Field

[0001] This invention relates to the field of digital power technology, and in particular to a high-resistivity, low-current grounding line selection method and system based on a fully convolutional network. Background Technology

[0002] With the gradual advancement of China's energy internet construction, the requirements for power supply reliability are increasing. China's 10kV distribution network widely adopts low-current grounding methods, with single-phase grounding faults being the primary type of fault. When a single-phase grounding fault occurs in the distribution network, the higher the grounding resistance, the smaller the transient current in some non-faulty lines, making accurate fault location difficult. Currently, some power supply companies still use manual stringing to locate high-resistance, low-current grounding faults. This method needs improvement in terms of efficiency, accuracy, and safety.

[0003] To address this, numerous scholars both domestically and internationally have conducted research on low-current grounding fault location. This research primarily categorizes low-current grounding fault location into two methods: passive and active. Passive methods analyze transient current characteristics using correlation coefficients and wavelet transforms to select the fault. Active methods employ frequency-converted signals for low-current grounding fault location. However, based on field analysis, these methods are only suitable for low-current grounding fault location within 2 kiloohms, and their accuracy is low for high-resistance low-current grounding faults ranging from 2 kiloohms to 5 kiloohms.

[0004] Therefore, it is necessary to provide a high-resistance, low-current grounding selection method and system based on fully convolutional networks to solve one of the above-mentioned technical problems. Summary of the Invention

[0005] To address one of the aforementioned technical problems, this invention proposes a high-resistivity, low-current grounding fault location method based on a fully convolutional network. This method collects transient current data from a distribution network experiencing a single-phase grounding fault and performs high-resistivity, low-current grounding fault location through transient current traveling wave extraction, traveling wave image generation, and grounding fault location discrimination steps.

[0006] Specifically, the transient current traveling wave extraction steps are as follows: deep learning is performed on the transient current traveling wave feature data through a convolutional neural network to obtain a transient current traveling wave extraction model; the transient current data is then subjected to convolutional wavelet transform through the transient current traveling wave extraction model to obtain transient current traveling wave data.

[0007] Specifically, the traveling wave image generation steps are as follows: the time-domain change of transient current traveling wave data is used as a spatial coordinate system for analysis and three-dimensional projection to obtain the three-dimensional coordinates of the transient current traveling wave in the spatial coordinate system; the image is created in the three-dimensional spatiotemporal model using the three-dimensional coordinates to obtain the three-dimensional spatiotemporal image of the transient current traveling wave in the spatial coordinate system.

[0008] Specifically, the grounding selection process involves: using a fully convolutional network to perform deep learning on the features of the three-dimensional spatiotemporal image to obtain a grounding selection analysis model; using the grounding selection analysis model to perform pixel-level analysis on the three-dimensional spatiotemporal image to obtain the high-resistance, low-current grounding selection result; the analysis includes transient current traveling wave amplitude polarity analysis and transient current traveling wave graphic feature analysis.

[0009] As a further solution, the transient current traveling wave extraction model performs frequency decomposition of the transient current traveling wave, reconstructs the transient current traveling wave approximation coefficients, and extracts transient current fault traveling wave data to obtain transient current traveling wave data W. s for:

[0010]

[0011] Where Ф(t) is the basic wavelet of the transient current traveling wave; β is the scale of the transient current traveling wave wavelet transform, used to compress or stretch the basic wavelet; k is the transient current traveling wave shift; and g(t) is the input transient current data.

[0012] As a further solution, the three-term time-domain waveform expression of the transient current traveling wave after convolutional wavelet transform is as follows:

[0013]

[0014] Among them: I a I b I c These are the three-phase currents of the distribution network; I amax I bmax I cmax These are the peak values ​​of the three-phase currents in the distribution network; δ a The signal is a 50Hz power frequency signal; t a λ represents the fault time of a single-phase ground fault; λ represents the initial phase angle of the transient current fault traveling wave.

[0015] As a further solution, the three-dimensional coordinates (X, X) of the transient current traveling wave in the spatial coordinate system can be obtained by the following formula. a Y b Z c ):

[0016]

[0017] Where, x z y z These are the original x and y coordinates of the transient current traveling wave; r 11 r 12 r 21 r 22These are the linear transformation data of the original coordinate system of the transient current traveling wave; r 13 r 23 These are the translation data of the original coordinate system of the transient current traveling wave; r 31 r 32 These are perspective transformation data of the original coordinate system of the transient current traveling wave.

[0018] As a further solution, after the transient current traveling wave is projected in three dimensions, wave pattern normalization is also performed:

[0019] The transient current waveforms of the two cycles before and after a single-phase ground fault are collected to obtain the initial waveform data of the transient current traveling wave. Based on the initial waveform data, a three-dimensional spatiotemporal model is created.

[0020] As a further solution, the fully convolutional network achieves pixel-level analysis of three-dimensional spatiotemporal images through transposed convolutional layers. The sampling window F of the fully convolutional network for high-resistance, low-current grounding selection... a for:

[0021]

[0022] Among them, e x e y e z These represent the length, width, and height of the three-dimensional spatiotemporal image, respectively; s t q represents the search stride of the fully convolutional network; a The window size for sampling the fully convolutional network.

[0023] As a further solution, the fully convolutional network calculates the image residual output value g of the three-dimensional spatiotemporal image. a for:

[0024] g a =f(d,w a ,w b )+w j d

[0025] Where f is the convolution function of the fully convolutional network; d is the input 3D spatiotemporal image variable of the fully convolutional network; w a w b These are the weights of layers a and b of the fully convolutional network, respectively; w j Transform the input d to the output dimension.

[0026] A high-resistance, low-current ground fault location system based on a fully convolutional network is provided. It operates on hardware equipment and uses a high-resistance, low-current ground fault location method based on a fully convolutional network as described above to locate single-phase ground faults.

[0027] Compared with related technologies, the high-resistivity, low-current grounding line selection method and system based on fully convolutional networks provided by this invention have the following advantages:

[0028] This invention transforms the high-resistance, low-current grounding fault location problem into an image recognition problem. By converting the spectral components of the transient current traveling wave into a spatiotemporal image, the grounding fault location problem is transformed into an image recognition and classification problem with a mature learning model, thus reducing the difficulty of model training and improving the accuracy of fault location. The fully convolutional network achieves pixel-level analysis in the three-dimensional spatiotemporal image through transposed convolutional layers. The fully convolutional network only needs to compare the polarity and amplitude characteristics of the transient current traveling wave through image comparison to complete the high-resistance, low-current grounding fault location. It features accurate and fast image analysis, combining high efficiency, practicality, convenience, and accuracy. Attached Figure Description

[0029] Figure 1 A flowchart illustrating a preferred embodiment of the high-resistivity, low-current grounding line selection method based on a fully convolutional network provided by the present invention;

[0030] Figure 2 A step diagram illustrating the transient current traveling wave extraction process of a high-resistivity, low-current grounding line selection method based on a fully convolutional network provided by this invention.

[0031] Figure 3 The transient current traveling wave time-domain waveform diagram of a high-resistivity low-current grounding line selection method based on a fully convolutional network provided by the present invention;

[0032] Figure 4 The transient current traveling wave three-dimensional transformation diagram is provided for a high-resistivity low-current grounding line selection method based on a fully convolutional network provided by the present invention. Detailed Implementation

[0033] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0034] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0035] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0036] like Figures 1 to 2 As shown, the present invention provides a high-resistivity, low-current grounding fault location method based on a fully convolutional network, which collects transient current data of a distribution network experiencing a single-phase grounding fault; and performs high-resistivity, low-current grounding fault location through transient current traveling wave extraction, traveling wave image generation, and grounding fault location discrimination steps.

[0037] Specifically, the transient current traveling wave extraction steps are as follows: deep learning is performed on the transient current traveling wave feature data through a convolutional neural network to obtain a transient current traveling wave extraction model; the transient current data is then subjected to convolutional wavelet transform through the transient current traveling wave extraction model to obtain transient current traveling wave data.

[0038] Specifically, the traveling wave image generation steps are as follows: the time-domain change of transient current traveling wave data is used as a spatial coordinate system for analysis and three-dimensional projection to obtain the three-dimensional coordinates of the transient current traveling wave in the spatial coordinate system; the image is created in the three-dimensional spatiotemporal model using the three-dimensional coordinates to obtain the three-dimensional spatiotemporal image of the transient current traveling wave in the spatial coordinate system.

[0039] Specifically, the grounding selection process involves: using a fully convolutional network to perform deep learning on the features of the three-dimensional spatiotemporal image to obtain a grounding selection analysis model; using the grounding selection analysis model to perform pixel-level analysis on the three-dimensional spatiotemporal image to obtain the high-resistance, low-current grounding selection result; the analysis includes transient current traveling wave amplitude polarity analysis and transient current traveling wave graphic feature analysis.

[0040] It should be noted that traditional low-current grounding fault location methods are only applicable to grounding faults with currents below 2,000 ohms. Their accuracy is low for high-resistance, low-current grounding faults ranging from 2,000 to 5,000 ohms. To address this low accuracy issue, this embodiment transforms the high-resistance, low-current grounding fault location problem into an image recognition problem. By converting the spectral components of the transient fault current traveling wave into a spatiotemporal image, the grounding fault location problem is transformed into an image recognition and classification problem with a mature learning model. A fully convolutional neural network (FCN) image recognition method is then used for high-resistance, low-current grounding fault location.

[0041] Three-dimensional spatiotemporal images can visualize the characteristics of transient current traveling waves, greatly improving the difficulty of model training and the accuracy of line selection. Fully convolutional networks (FCNs) achieve pixel-level analysis in three-dimensional spatiotemporal images through transposed convolutional layers, offering advantages such as accurate and fast image analysis. Therefore, this paper uses FCNs for polarity and image feature comparison of transient current traveling waves.

[0042] As a further solution, the transient current traveling wave extraction model performs frequency decomposition of the transient current traveling wave, reconstructs the transient current traveling wave approximation coefficients, and extracts transient current fault traveling wave data to obtain transient current traveling wave data W. s for:

[0043]

[0044] Where Ф(t) is the basic wavelet of the transient current traveling wave; β is the scale of the transient current traveling wave wavelet transform, used to compress or stretch the basic wavelet; k is the transient current traveling wave shift; and g(t) is the input transient current data.

[0045] As a further solution, the three-term time-domain waveform expression of the transient current traveling wave after convolutional wavelet transform is as follows:

[0046]

[0047] Among them: I a I b I c These are the three-phase currents of the distribution network; I amax I bmax I cmax These are the peak values ​​of the three-phase currents in the distribution network; δ a The signal is a 50Hz power frequency signal; t a λ represents the fault time of a single-phase ground fault; λ represents the initial phase angle of the transient current fault traveling wave.

[0048] As a further solution, the three-dimensional coordinates (X, X) of the transient current traveling wave in the spatial coordinate system can be obtained by the following formula. a Y b Z c ):

[0049]

[0050] Where, x z y z These are the original x and y coordinates of the transient current traveling wave; r 11 r 12 r 21 r 22 These are the linear transformation data of the original coordinate system of the transient current traveling wave; r 13 r 23 These are the translation data of the original coordinate system of the transient current traveling wave; r 31 r 32 These are perspective transformation data of the original coordinate system of the transient current traveling wave.

[0051] As a further solution, after the transient current traveling wave is projected in three dimensions, wave pattern normalization is also performed:

[0052] The transient current waveforms of the two cycles before and after a single-phase ground fault are collected to obtain the initial waveform data of the transient current traveling wave. Based on the initial waveform data, a three-dimensional spatiotemporal model is created.

[0053] It should be noted that when a high-resistance, low-current grounding event occurs (above 2000 ohms), the transient resonant frequency is low, resulting in a small transient current traveling wave after passing through the arc suppression coil. When displayed in a three-dimensional spatiotemporal model, the oscillation values ​​around the projection planes of the three phases represent the fault harmonic components. In low-current grounding systems, the transient components disappear after six cycles of the fault. Therefore, in this embodiment, the traveling wave pattern standardization process selects the transient current waveforms from the two cycles before and after the fault for standardization to obtain the initial waveform data of the transient current fault traveling wave, and then creates a three-dimensional spatiotemporal model.

[0054] As a further solution, the fully convolutional network achieves pixel-level analysis of three-dimensional spatiotemporal images through transposed convolutional layers. The sampling window F of the fully convolutional network for high-resistance, low-current grounding selection... a for:

[0055]

[0056] Among them, e x e y e z These represent the length, width, and height of the three-dimensional spatiotemporal image, respectively; s t q represents the search stride of the fully convolutional network; a The window size for sampling the fully convolutional network.

[0057] As a further solution, the fully convolutional network calculates the image residual output value g of the three-dimensional spatiotemporal image. a for:

[0058] g a =f(d,w a ,w b )+w j d

[0059] Where f is the convolution function of the fully convolutional network; d is the input 3D spatiotemporal image variable of the fully convolutional network; w a w b These are the weights of layers a and b of the fully convolutional network, respectively; w j Transform the input d to the output dimension.

[0060] A high-resistance, low-current ground fault location system based on a fully convolutional network is provided. It operates on hardware equipment and uses a high-resistance, low-current ground fault location method based on a fully convolutional network as described above to locate single-phase ground faults.

[0061] It should be noted that, in a specific embodiment, the sub-steps of the traveling wave image generation step include establishing the transient current traveling wave time-domain waveform, three-dimensional projection of the transient current traveling wave, traveling wave graphic normalization processing, and three-dimensional spatiotemporal model creation. In this embodiment, the step of establishing the transient current traveling wave time-domain waveform establishes the transient current traveling wave time-domain waveforms of three lines, such as... Figure 3 As shown; by Figure 3 It can be seen that the polarity of the transient current traveling wave waveform of line 2 is opposite to that of lines 1 and 3, and its maximum amplitude is larger than that of the other two lines.

[0062] like Figure 4 As shown in the figure, this embodiment performs a three-dimensional projection of the transient current traveling wave time-domain waveform to obtain a three-dimensional transformation diagram of the transient current traveling wave. As can be seen from the diagram, before a high-resistance single-phase ground fault occurs, the distribution line operates normally, and the current amplitude changes little, appearing elliptical in the three-dimensional model. When a high-resistance single-phase ground fault occurs in phase A of line 2, the traveling wave amplitude of line 2 is the largest, and its polarity is opposite to that of lines 1 and 2. The three-dimensional graphics enable rapid line selection for high-resistance, low-current ground faults.

[0063] In summary: In a power distribution network, when a single-phase ground fault occurs, the transient current traveling wave propagates along the network. During propagation, reflection and refraction occur at points of impedance change, resulting in the initial transient current traveling wave amplitude of the grounded line being greater than that of the ungrounded line, and the polarity of the transient current traveling wave being opposite. Therefore, by determining that the maximum value of the transient current traveling wave of the single-phase ground fault line is greater than that of other lines, and that the polarity of the fault traveling wave is opposite, the correct grounding line can be selected.

[0064] In this embodiment, the polarity and amplitude characteristics of transient current traveling waves are compared graphically within a three-dimensional spatiotemporal model to select grounding lines for high-resistance, low-current applications. The polarity and amplitude characteristics of the transient current traveling wave image are analyzed using a fully convolutional network to achieve the same result.

[0065] The above are merely embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A high-resistivity, low-current grounding line selection method based on a fully convolutional network, characterized in that, Collect transient current data of the distribution network where a single-phase ground fault occurs; and select the high-resistance, low-current ground fault line through transient current traveling wave extraction, traveling wave image generation, and ground fault line selection discrimination steps; Transient current traveling wave extraction steps: Deep learning is performed on the transient current traveling wave feature data through a convolutional neural network to obtain a transient current traveling wave extraction model; Convolutional wavelet transform is performed on the transient current data through the transient current traveling wave extraction model to obtain transient current traveling wave data; The traveling wave image generation steps are as follows: the time-domain change of transient current traveling wave data is used as a spatial coordinate system for analysis and three-dimensional projection to obtain the three-dimensional coordinates of the transient current traveling wave in the spatial coordinate system; the three-dimensional coordinates are used to create an image in a three-dimensional spatiotemporal model to obtain a three-dimensional spatiotemporal image of the transient current traveling wave in the spatial coordinate system. The three-dimensional coordinates (X, X) of the transient current traveling wave in the spatial coordinate system can be obtained by the following formula. a Y b Z c ): Where, x z y z These are the original x and y coordinates of the transient current traveling wave; r 11 r 12 r 21 r 22 These are the linear transformation data of the original coordinate system of the transient current traveling wave; r 13 r 23 These are the translation data of the original coordinate system of the transient current traveling wave; r 31 r 32 These are perspective transformation data of the original coordinate system of the transient current traveling wave; Grounding selection process: Deep learning is performed on the features of the three-dimensional spatiotemporal image through a fully convolutional network to obtain a grounding selection analysis model; the grounding selection analysis model is used to perform pixel-level analysis on the three-dimensional spatiotemporal image to obtain the high-resistance, low-current grounding selection result; the analysis includes transient current traveling wave amplitude polarity analysis and transient current traveling wave graphic feature analysis. The fully convolutional network achieves pixel-level analysis of three-dimensional spatiotemporal images through transposed convolutional layers. The high-resistance, low-current grounding line selection sampling window F of the fully convolutional network... a for: Among them, e x e y e z These represent the length, width, and height of the three-dimensional spatiotemporal image, respectively; s t q represents the search stride of the fully convolutional network; a The window size for sampling the fully convolutional network.

2. The high-resistivity, low-current grounding line selection method based on a fully convolutional network according to claim 1, characterized in that, The transient current traveling wave extraction model performs frequency decomposition of the transient current traveling wave, reconstructs the transient current traveling wave approximation coefficients, and extracts transient current fault traveling wave data to obtain transient current traveling wave data W. s for: Where Ф(t) is the basic wavelet of the transient current traveling wave; β is the scale of the transient current traveling wave wavelet transform, used to compress or stretch the basic wavelet; k is the transient current traveling wave shift; and g(t) is the input transient current data.

3. The high-resistivity, low-current grounding line selection method based on a fully convolutional network according to claim 1, characterized in that, The three-term time-domain waveform expression of the transient current traveling wave after convolutional wavelet transform is as follows: Among them: I a I b I c These are the three-phase currents of the distribution network; I amax I bmax I cmax These are the peak values ​​of the three-phase currents in the distribution network; δ a The signal is a 50Hz power frequency signal; t a λ represents the fault time of a single-phase ground fault; λ represents the initial phase angle of the transient current fault traveling wave.

4. The high-resistivity, low-current grounding line selection method based on a fully convolutional network according to claim 1, characterized in that, After the transient current traveling wave is projected in three dimensions, wave pattern normalization is also performed: The transient current waveforms of the two cycles before and after a single-phase ground fault are collected to obtain the initial waveform data of the transient current traveling wave. Based on the initial waveform data, a three-dimensional spatiotemporal model is created.

5. The high-resistivity, low-current grounding line selection method based on a fully convolutional network according to claim 1, characterized in that, The fully convolutional network calculates the image residual output value g of the three-dimensional spatiotemporal image. a for: g a =f(d,w a ,w b )+w j d Where f is the convolution function of the fully convolutional network; d is the input 3D spatiotemporal image variable of the fully convolutional network; w a w b These are the weights of layers a and b of the fully convolutional network, respectively; w j Transform the input d to the output dimension.

6. A high-resistivity, low-current grounding line selection system based on a fully convolutional network, characterized in that, Running on hardware devices, and implementing high-resistance low-current ground fault location for single-phase ground faults through a high-resistance low-current ground fault location method based on a fully convolutional network as described in any one of claims 1 to 5.

Images