Ground fault detection method, device and equipment based on convolutional neural network

Abstract

The application discloses a ground fault detection method and device based on a convolutional neural network and equipment, relates to the technical field of power distribution network fault detection, and can obtain sample zero sequence current data and corresponding sample labels by constructing and running a heterogeneous line and heterogeneous phase ground fault simulation model, and train a convolutional neural network according to the sample zero sequence current data and the corresponding sample labels. Finally, the real-time data features of a real-time zero sequence current waveform are identified through the trained convolutional neural network, and a ground fault detection result of a small-resistance grounding system is obtained, which can effectively and definitely identify a cross-line two-point grounding fault, and has important significance for cross-line two-point grounding fault detection of a small-resistance grounding system.

Description

Technical Field

[0001] This application relates to the field of power distribution network fault detection technology, specifically to a ground fault detection method, device, and equipment based on convolutional neural networks. Background Technology

[0002] With the accelerating pace of urbanization, the cable coverage rate of urban medium-voltage distribution networks is increasing. Simultaneously, the significant increase in capacitive current makes it difficult for arc suppression coils to compensate for large capacitive currents, rendering traditional neutral grounding via arc suppression coils ineffective in urban cable distribution networks. Low-resistance grounding systems, with their advantages of rapid fault clearing and low overvoltage levels, are widely used in urban distribution networks. Current low-resistance grounding systems primarily rely on time-delay zero-sequence overcurrent protection, set based on the maximum capacitive current, typically with high settings of 40A to 60A. Based on practical experience, this setting can only detect faults with grounding resistances of 85Ω to 140Ω. Furthermore, with the increasing use of multi-circuit overhead lines on the same tower and multi-circuit cables in the same trench, the probability of simultaneous or successive grounding faults on multiple circuits on the same busbar in distribution networks has increased significantly. In such cases, the zero-sequence current exhibits different characteristics compared to single-circuit grounding faults. If these grounding faults are not detected promptly, the severity of the fault may worsen, disrupting the operational stability of the power system and threatening the safety of people's lives and property. Based on existing technology, when a high-resistance grounding fault occurs at two points across a low-resistance grounding system, the system cannot detect the fault. Summary of the Invention

[0003] The purpose of this application is to provide a ground fault detection method, device, and equipment based on convolutional neural networks, which solves the problem that existing technologies are unable to detect high-resistance ground faults at two points across a line.

[0004] This application is achieved through the following technical solution:

[0005] The first aspect of this application provides a ground fault detection method based on a convolutional neural network, comprising:

[0006] A simulation model of a low-resistance grounding system is constructed, and a grounding fault with different line and phase is set in the simulation model to obtain a simulation model of grounding fault with different line and phase.

[0007] Run the simulation model of the non-circuit and non-phase grounding fault and collect the sample zero-sequence current data corresponding to each outgoing line of the simulation model.

[0008] The zero-sequence current data of the samples is preprocessed to obtain the sample data features corresponding to the zero-sequence current data of the samples, and a corresponding sample label is set for each sample data feature; wherein, the sample label includes fault or non-fault;

[0009] The convolutional neural network is trained using the features of the sample data and their corresponding sample labels to obtain the trained convolutional neural network.

[0010] The real-time zero-sequence current waveform corresponding to each outgoing line in the low-resistance grounding system is collected, and the trained convolutional neural network is scheduled to identify the real-time data features of the real-time zero-sequence current waveform to determine the grounding fault detection result corresponding to the low-resistance grounding system.

[0011] In some possible design approaches, before acquiring the real-time zero-sequence current waveform corresponding to each outgoing line in the low-resistance grounding system, the following steps are also included:

[0012] Collect the zero-sequence current amplitude corresponding to the neutral line of the low-resistance grounding system;

[0013] If the amplitude of the zero-sequence current corresponding to the neutral line exceeds the preset setting value, then proceed to the step of collecting the real-time zero-sequence current waveform corresponding to each outgoing line in the small resistance grounding system; otherwise, continue to monitor the amplitude of the zero-sequence current corresponding to the neutral line until the amplitude of the zero-sequence current corresponding to the neutral line exceeds the setting value, and then proceed to the step of collecting the real-time zero-sequence current waveform corresponding to each outgoing line in the small resistance grounding system.

[0014] In some possible design approaches, a cross-line, cross-phase grounding fault is set up in the simulation model to obtain a cross-line, cross-phase grounding fault simulation model, including:

[0015] In the simulation model, ground faults are set on different outgoing lines, and the phase lines that cause ground faults are different on different outgoing lines, thus obtaining a simulation model of ground faults with different outgoing lines and different phases.

[0016] In some possible design approaches, the off-line, off-phase grounding fault simulation model is run, and sample zero-sequence current data corresponding to each outgoing line of the off-line, off-phase grounding fault simulation model is collected, including:

[0017] Run the simulation model of the different phases and different lines grounding fault, and sample the analog signal output by each outgoing line at a preset sampling time to obtain the discrete sample points corresponding to each outgoing line.

[0018] For any discrete sample point corresponding to an outgoing line, the discrete sample point is quantized to obtain a quantized discrete sample point, and the quantized discrete sample point is converted into a digital signal to obtain sample zero-sequence current data.

[0019] In some possible design approaches, the sample zero-sequence current data is preprocessed to obtain sample data features corresponding to the sample zero-sequence current data, and a corresponding sample label is set for each sample data feature, including:

[0020] The sample zero-sequence current data is converted into a two-dimensional image by Gram angle field to obtain the sample data features corresponding to the sample zero-sequence current data.

[0021] If a grounding fault is set on the outgoing line corresponding to the sample data feature, then the sample label corresponding to the sample data feature is set as fault.

[0022] If the outgoing line corresponding to the sample data feature is not set to ground fault, then the sample tag corresponding to the sample data feature is set to non-fault.

[0023] In some possible design approaches, the sample zero-sequence current data is converted into a two-dimensional image using a Gram angle field to obtain the sample data features corresponding to the sample zero-sequence current data, including:

[0024] The sample zero-sequence current data is normalized to obtain normalized sample zero-sequence current data.

[0025] The normalized sample zero-sequence current data is encoded as a polar coordinate angle, and the sampling time corresponding to the normalized sample zero-sequence current data is encoded as a polar coordinate radius to obtain the target encoded data.

[0026] Extract the Gram angle and field or Gram angle difference field corresponding to the target encoded data to obtain the sample data features corresponding to the sample zero-sequence current data.

[0027] In some possible design approaches, the normalized sample zero-sequence current data is encoded as polar coordinates with the following angles:

[0028] ;

[0029] in, Let i represent the i-th data point in the normalized sample zero-sequence current data, where i = 1, 2, ..., I, and I represents the total number of data points in the normalized sample zero-sequence current data. This represents the set of data points in the normalized sample zero-sequence current data. This represents the polar coordinate angle obtained from the encoding.

[0030] In some possible design approaches, the sampling time corresponding to the normalized sample zero-sequence current data is encoded as a polar coordinate radius of:

[0031] ;

[0032] in, This represents the sampling time corresponding to the i-th data point in the normalized zero-sequence current data. Represents the normalization factor. This represents the polar coordinate radius obtained from the encoding.

[0033] Based on the same inventive concept, a second aspect of this application provides a ground fault detection device based on a convolutional neural network, comprising:

[0034] The fault simulation module is used to construct a simulation model corresponding to a low-resistance grounding system, and to set up a cross-line and cross-phase grounding fault in the simulation model to obtain a cross-line and cross-phase grounding fault simulation model.

[0035] The sample data acquisition module is used to run the simulation model of the non-linear and non-phase grounding fault and to acquire the sample zero-sequence current data corresponding to each outgoing line of the simulation model.

[0036] The sample data processing module is used to preprocess the sample zero-sequence current data to obtain the sample data features corresponding to the sample zero-sequence current data, and to set a corresponding sample label for each sample data feature; wherein, the sample label includes fault or non-fault.

[0037] The deep learning module is used to train the convolutional neural network using the features of the sample data and their corresponding sample labels, and to obtain the trained convolutional neural network.

[0038] The ground fault detection module is used to collect the real-time zero-sequence current waveform corresponding to each outgoing line in the low-resistance grounding system, and to schedule the trained convolutional neural network to identify the real-time data features of the real-time zero-sequence current waveform to determine the ground fault detection result corresponding to the low-resistance grounding system.

[0039] Based on the same inventive concept, a third aspect of this application provides an electronic device, including a processor and a memory;

[0040] The memory stores computer-executed instructions;

[0041] The processor executes computer execution instructions stored in the memory, causing the processor to perform the ground fault detection method based on a convolutional neural network as described in any of the first aspects.

[0042] Compared with the prior art, this application has the following advantages and beneficial effects:

[0043] This application provides a ground fault detection method, apparatus, and device based on convolutional neural networks. By constructing and running a simulation model of ground faults between different lines and phases, sample zero-sequence current data and corresponding sample labels can be obtained. A convolutional neural network is then trained based on the sample zero-sequence current data and corresponding sample labels. Finally, the real-time data features of the real-time zero-sequence current waveform are identified through the trained convolutional neural network to obtain the ground fault detection results of the low-resistance grounding system. This method can effectively and definitively identify two-point ground faults across lines, which is of great significance for the detection of two-point ground faults across lines in low-resistance grounding systems. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of the exemplary embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0045] Figure 1 A flowchart illustrating a ground fault detection method based on a convolutional neural network, provided as an embodiment of this application;

[0046] Figure 2 The zero-sequence current waveform diagram of the faulted line under the influence of grounding arc is provided in the embodiments of this application;

[0047] Figure 3 This is a zero-sequence current waveform diagram of a non-faulty line under the influence of a grounding arc, provided in an embodiment of this application.

[0048] Figure 4 A schematic diagram illustrating the process of obtaining sample data features provided in an embodiment of this application;

[0049] Figure 5 This is a schematic diagram of the structure of the simulation model for grounding faults between different lines and phases provided in the embodiments of this application;

[0050] Figure 6 The zero-sequence current waveform of the first-faulted line is provided in the embodiments of this application;

[0051] Figure 7 Zero-sequence current waveform diagram of a subsequently faulted line provided in an embodiment of this application;

[0052] Figure 8 Zero-sequence current waveform diagram of a non-faulty circuit provided in the embodiments of this application;

[0053] Figure 9 GASF diagram of the zero-sequence current waveform corresponding to the faulty line provided in the embodiments of this application;

[0054] Figure 10 GADF diagram of the zero-sequence current waveform corresponding to the faulty line provided in the embodiments of this application;

[0055] Figure 11 GASF diagram of zero-sequence current waveform corresponding to a non-faulty circuit provided in the embodiments of this application;

[0056] Figure 12 GADF diagram of the zero-sequence current waveform corresponding to the non-faulty line provided in the embodiments of this application;

[0057] Figure 13 A schematic diagram of a ground fault detection device based on a convolutional neural network provided in an embodiment of this application;

[0058] Figure 14 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0059] The attached diagram shows the markings and corresponding component names:

[0060] 1201-Fault simulation module, 1202-Sample data acquisition module, 1203-Sample data processing module, 1204-Deep learning module, 1205-Grounding fault detection module, 1401-Memory, 1402-Processor, 1403-Bus. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this application are only for explaining this application and are not intended to limit this application.

[0062] like Figure 1 As shown in the figure, this application provides a ground fault detection method based on a convolutional neural network, including:

[0063] S101. Construct a simulation model corresponding to a small resistance grounding system, and set up a different line and different phase grounding fault in the simulation model to obtain a simulation model of different line and different phase grounding fault.

[0064] S102. Run the simulation model of the non-circuit and non-phase grounding fault and collect the sample zero-sequence current data corresponding to each outgoing line of the simulation model.

[0065] S103. Preprocess the sample zero-sequence current data to obtain the sample data features corresponding to the sample zero-sequence current data, and set a corresponding sample label for each sample data feature; wherein, the sample label includes fault or non-fault.

[0066] S104. Using the sample data features and their corresponding sample labels, train the convolutional neural network to obtain the trained convolutional neural network.

[0067] S105. Collect the real-time zero-sequence current waveform corresponding to each outgoing line in the low-resistance grounding system, and schedule the trained convolutional neural network to identify the real-time data features of the real-time zero-sequence current waveform to determine the grounding fault detection result corresponding to the low-resistance grounding system.

[0068] Convolutional Neural Networks (CNNs) are deep learning models specifically designed for processing data with a grid structure, such as images and audio signals. This application uses a CNN to classify the two-dimensional graphs corresponding to the zero-sequence current data of each outgoing line, thus realizing the detection of two-point grounding faults across lines in a low-resistance grounding system.

[0069] To enable those skilled in the art to understand the technical solutions described in the embodiments of this application, the grounding fault detection principle of a low-resistance grounding system will be introduced first.

[0070] When a high-resistance grounding fault occurs in a low-resistance grounding system, it is often accompanied by an electric arc, i.e., an arc-induced high-resistance grounding fault. During an arc-induced high-resistance grounding fault, the thermal effect of the arc combustion causes a zero-slope phenomenon, which leads to different distortions in the faulty and non-faulty lines. The distortion in the faulty line is manifested by a decrease in the slope of its zero-sequence current waveform at the zero-crossing point; the distortion in the non-faulty line is manifested by significant abrupt changes in the zero-sequence current waveform at both the peaks and troughs. Therefore, this application uses the Mayr arc model for simulation experiments. The zero-sequence current waveforms of the faulty and non-faulty lines under the influence of the grounding arc are shown below. Figure 2 and Figure 3 As shown.

[0071] In some possible design approaches, before acquiring the real-time zero-sequence current waveform corresponding to each outgoing line in the low-resistance grounding system, the following steps are also included:

[0072] Collect the zero-sequence current amplitude corresponding to the neutral line of the low-resistance grounding system;

[0073] If the amplitude of the zero-sequence current corresponding to the neutral line exceeds the preset setting value, then proceed to the step of collecting the real-time zero-sequence current waveform corresponding to each outgoing line in the small resistance grounding system; otherwise, continue to monitor the amplitude of the zero-sequence current corresponding to the neutral line until the amplitude of the zero-sequence current corresponding to the neutral line exceeds the setting value, and then proceed to the step of collecting the real-time zero-sequence current waveform corresponding to each outgoing line in the small resistance grounding system.

[0074] When a ground fault occurs in a low-resistance grounding system, the amplitude of the zero-sequence current corresponding to the neutral line will increase. Therefore, this embodiment uses the amplitude of the zero-sequence current of the neutral line as the ground fault detection criterion for the low-resistance grounding system. When the amplitude of the zero-sequence current corresponding to the neutral line of the low-resistance grounding system is greater than the set value, it is determined that a two-point ground fault may have occurred in the low-resistance grounding system, and further fault detection is required. The setting of the zero-sequence current amplitude of the neutral line is as follows: According to general parameters based on engineering experience, the maximum unbalanced zero-sequence current corresponding to a single-phase ground fault in a pure overhead line and a pure cable line is approximately 0.37A and 0.26A, respectively. Since the protection device is connected to a zero-sequence current signal of 3 times, the set value corresponding to the zero-sequence current amplitude of the neutral line should be greater than 1.11A. This embodiment sets the set value to 1.5A, which can detect a ground fault of 3000 ohms.

[0075] In some possible design approaches, a cross-line, cross-phase grounding fault is set up in the simulation model to obtain a cross-line, cross-phase grounding fault simulation model, including:

[0076] In the simulation model, ground faults are set on different outgoing lines, and the phase lines that cause ground faults are different on different outgoing lines, thus obtaining a simulation model of ground faults with different outgoing lines and different phases.

[0077] This application primarily addresses the detection of high-resistance grounding faults at two points across a low-resistance grounding system. Therefore, a cross-line, cross-phase grounding fault can be set up, and corresponding zero-sequence current data can be obtained. This method can acquire a large number of different zero-sequence current data. Furthermore, since multiple outgoing lines exist in a low-resistance grounding system, the acquired zero-sequence current data includes both fault-related and non-fault-related zero-sequence current data. A cross-line, cross-phase grounding fault refers to a grounding fault being set up simultaneously on two outgoing lines, with the faults located on different outgoing lines. For example, assuming a grounding fault is set up on both the first and second outgoing lines, and the grounding fault on the first outgoing line occurs in phase A, then the grounding fault on the second outgoing line can be set up in phase B or phase C, thereby realizing the setting of a cross-line, cross-phase grounding fault and obtaining a simulation model of the cross-line, cross-phase grounding fault.

[0078] In some possible design approaches, the off-line, off-phase grounding fault simulation model is run, and sample zero-sequence current data corresponding to each outgoing line of the off-line, off-phase grounding fault simulation model is collected, including:

[0079] Run the simulation model of the different phases and different lines grounding fault, and sample the analog signal output by each outgoing line at a preset sampling time to obtain the discrete sample points corresponding to each outgoing line.

[0080] For any discrete sample point corresponding to an outgoing line, the discrete sample point is quantized to obtain a quantized discrete sample point, and the quantized discrete sample point is converted into a digital signal to obtain sample zero-sequence current data.

[0081] For example, data acquisition can be performed using a zero-order holder, a quantizer, and a data type converter, with the sampling frequency set to 0.0001s. The sampler samples the analog signal output from the outgoing line at a preset sampling frequency to obtain discrete sample points corresponding to the outgoing line; the quantizer quantizes these discrete sample points, converting them into finite discrete values ​​to obtain quantized discrete sample points; and the encoder converts these quantized discrete sample points into digital signals to obtain the zero-sequence current data.

[0082] In some possible design approaches, the sample zero-sequence current data is preprocessed to obtain sample data features corresponding to the sample zero-sequence current data, and a corresponding sample label is set for each sample data feature, including:

[0083] The sample zero-sequence current data is converted into a two-dimensional image by Gram angle field to obtain the sample data features corresponding to the sample zero-sequence current data.

[0084] If a grounding fault is set on the outgoing line corresponding to the sample data feature, then the sample label corresponding to the sample data feature is set as fault.

[0085] If the outgoing line corresponding to the sample data feature is not set to ground fault, then the sample tag corresponding to the sample data feature is set to non-fault.

[0086] In some possible design approaches, the sample zero-sequence current data is converted into a two-dimensional image using a Gram angle field to obtain the sample data features corresponding to the sample zero-sequence current data, including:

[0087] The sample zero-sequence current data is normalized to obtain normalized sample zero-sequence current data.

[0088] For example, suppose the sample zero-sequence current data is as follows: ,in, Let represent the i-th data point in the sample zero-sequence current data. Then, the sample zero-sequence current data can be normalized as follows:

[0089] ;

[0090] in, Let i represent the i-th data point in the normalized sample zero-sequence current data, where i = 1, 2, ..., I, and I represents the total number of data points in the normalized sample zero-sequence current data. This represents the maximum value among all data points of the zero-sequence current data in the sample. This represents the minimum value among all data points of the zero-sequence current data in the sample.

[0091] like Figure 4 As shown, the normalized sample zero-sequence current data is encoded as a polar coordinate angle, and the sampling time corresponding to the normalized sample zero-sequence current data is encoded as a polar coordinate radius to obtain the target encoded data; the Gram angle and field or Gram angle difference field corresponding to the target encoded data are extracted to obtain the sample data features corresponding to the sample zero-sequence current data.

[0092] The interval [0,1] can be divided into several parts, and then 0 is discarded. The remaining points are associated with time series data, and each point falls within a unit circle in polar coordinates. As time increases, the time series of zero-sequence currents of each output line will also be continuously distorted within this unit circle at different angles and radii.

[0093] Here, [0,1] represents the normalized zero-sequence current value range (the original current data has been uniformly scaled to between 0 and 1 in previous steps). This range is divided to assign corresponding polar coordinate angles to the current values: for example, dividing it into 10 parts ([0,0.1), [0.1,0.2)...[0.9,1]), each part corresponding to an angle range (e.g., [0,0.1) corresponds to 0°~36°, [0.1,0.2) corresponds to 36°~72°), allowing different current values ​​to correspond to different polar coordinate angles.

[0094] After normalization, the current value is 0, which usually corresponds to a fault-free baseline state (no fault characteristics). Therefore, these 0-value data are removed, and only the current points with numerical changes are retained—these points may contain the characteristic information of ground faults.

[0095] The remaining non-zero current values ​​each correspond to a sampling time (e.g., the current collected in the 1st second, the 2nd second). The association here means that the current value (corresponding to the polar coordinate angle) and its sampling time (corresponding to the polar coordinate radius) are bound into a pair of data (current value → angle, sampling time → radius).

[0096] The unit circle in polar coordinates is a circle with a maximum radius of 1: the sampling time is normalized to between 0 and 1, and this radius is used as the polar coordinate radius; the angle corresponding to the split current value is used as the polar coordinate angle. In this way, each point corresponding to (current value, sampling time) will fall within a unit circle with a radius ≤ 1.

[0097] In some possible design approaches, the normalized sample zero-sequence current data is encoded as polar coordinates with the following angles:

[0098] ;

[0099] in, Let i represent the i-th data point in the normalized sample zero-sequence current data, where i = 1, 2, ..., I, and I represents the total number of data points in the normalized sample zero-sequence current data. This represents the set of data points in the normalized sample zero-sequence current data. This represents the polar coordinate angle obtained from the encoding.

[0100] In some possible design approaches, the sampling time corresponding to the normalized sample zero-sequence current data is encoded as a polar coordinate radius of:

[0101] ;

[0102] in, This represents the sampling time corresponding to the i-th data point in the normalized zero-sequence current data. This represents the normalization factor, set as the total sampling time of the zero-sequence current data for this sample. This represents the polar coordinate radius obtained from the encoding.

[0103] The normalized zero-sequence current data corresponding to each outgoing line, after polar coordinate transformation, contains time information and can therefore be reconstructed using the Gramian Angular Field (GAF). Two different images can be generated based on different formulas: the Gramian Angular Sum Field (GASF) and the Gramian Angular Difference Field (GADF), with the expressions as follows:

[0104] ;

[0105] ;

[0106] in, This represents the Gram matrix corresponding to the Gram angle and the field. This represents the Gram matrix corresponding to the Gram angle difference field; for the same exit line, Indicates and Different polar coordinate angles.

[0107] Based on the above technical solution, this application provides a detailed example to illustrate the technical solution, as follows.

[0108] Step 1: Build a system in the MATLAB / Simulink environment as follows Figure 5A simulation model of a non-linear, non-phase grounding fault was constructed. In this model, the neutral grounding resistance was set to 10 ohms, and it included six feeders with lengths of 2km, 6km, 8km, 10km, 12km, and 20km respectively. The positive sequence parameters of the line were: R1=0.28Ω / km, X1=0.26mh / km, C1=0.38μF / km; the zero sequence parameters were: R0=2.8Ω / km, X0=1.11mh / km, C0=0.28μF / km. Different fault conditions, different fault occurrence times, and different transition resistances (150 ohms-3000 ohms) could be set for the non-linear, non-phase grounding fault, allowing the acquisition of sample zero-sequence current data for each outgoing line under different grounding fault conditions. The zero-sequence current waveforms of the faulty line, the line that failed later, and the non-faulty line are shown below. Figure 6 , Figure 7 and Figure 8 As shown.

[0109] exist Figure 5 In the diagram, L1-L6 are outgoing lines, Ln is the neutral line, and Rn is the neutral line grounding resistance. M1-M6 are Mayr arc models used to simulate grounding arcs. F1-F6 are fault-inducing elements, which simulate various fault conditions by setting the fault occurrence time, fault line, fault phase, and fault transition resistance.

[0110] Step 2: Use the zero-sequence current amplitude corresponding to the neutral line as the trigger criterion for ground fault detection. Set the corresponding zero-sequence current amplitude to 1.5A. When the detected zero-sequence current amplitude exceeds the set value, a fault is determined to have occurred, and data can be collected. The collected data here can refer to sample zero-sequence current data used for training or real-time zero-sequence current data used for detection.

[0111] Step 3: Convert the sample zero-sequence current data or real-time zero-sequence current data into a two-dimensional image using Gram angle field, thereby obtaining the sample data characteristics or real-time data characteristics. For example, the GASF and GADF graphs of the zero-sequence current waveform corresponding to the faulty line are shown below. Figure 9 , Figure 10 As shown, the GASF and GADF graphs of the zero-sequence current waveform corresponding to the non-faulty line are as follows: Figure 11 , Figure 12 As shown.

[0112] Without training a convolutional neural network, sample labels can be set for the sample data features corresponding to each outgoing line. The sample label corresponding to the sample data features of grounding faults is set to 1, and the sample label corresponding to the sample data features of no grounding faults is set to 0, thereby obtaining training samples.

[0113] Step 4: For the sample data features, the training samples can be divided into a training set and a test set in a 4:1 ratio, and then fed into a convolutional neural network for training. This will result in a trained convolutional neural network. Based on the trained neural network, real-time data features can be identified, thereby enabling the detection of cross-line two-point grounding faults in low-resistance grounding systems.

[0114] like Figure 13 As shown, based on the same inventive concept, this application provides a ground fault detection device based on a convolutional neural network, comprising:

[0115] The fault simulation module 1301 is used to construct a simulation model corresponding to a low-resistance grounding system, and to set up a different-line and different-phase grounding fault in the simulation model to obtain a different-line and different-phase grounding fault simulation model.

[0116] The sample data acquisition module 1302 is used to run the simulation model of the non-linear and non-phase grounding fault and to acquire the sample zero-sequence current data corresponding to each outgoing line of the simulation model.

[0117] The sample data processing module 1303 is used to preprocess the sample zero-sequence current data to obtain the sample data features corresponding to the sample zero-sequence current data, and to set a corresponding sample label for each sample data feature; wherein, the sample label includes fault or non-fault.

[0118] The deep learning module 1304 is used to train the convolutional neural network using the features of the sample data and their corresponding sample labels, and to obtain the trained convolutional neural network.

[0119] The ground fault detection module 1305 is used to collect the real-time zero-sequence current waveform corresponding to each outgoing line in the low-resistance grounding system, and to schedule the trained convolutional neural network to identify the real-time data features of the real-time zero-sequence current waveform to determine the ground fault detection result corresponding to the low-resistance grounding system.

[0120] like Figure 14 As shown, based on the same inventive concept, this application also provides an electronic device, including a processor 302 and a memory 1401; the memory 1401 and the processor 1402 are interconnected via a bus 1403.

[0121] The memory 1401 stores computer-executed instructions;

[0122] The processor 1402 executes the computer execution instructions stored in the memory 1401, causing the processor 1402 to execute a ground fault detection method based on a convolutional neural network as described in any embodiment of this application.

[0123] For specific examples, memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). Furthermore, the processor may include a main processor and coprocessors. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0124] This application embodiment may also provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the ground fault detection method based on a convolutional neural network as described in any of the above embodiments.

[0125] This application embodiment can also provide a computer program product, including a computer program that, when executed by a processor, implements the ground fault detection method based on convolutional neural networks described in any of the above embodiments.

[0126] All or part of the steps in the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a readable memory. When the program is executed, it performs the steps of the above method embodiments; and the aforementioned memory (storage medium) includes: read-only memory (ROM), RAM, flash memory, hard disk, solid-state drive, magnetic tape, floppy disk, optical disk, and any combination thereof.

[0127] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processing unit of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processing unit of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0128] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0129] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0130] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0131] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A grounding fault detection method based on convolutional neural networks, applied to a low-resistance grounding system, characterized in that, include: A simulation model of a low-resistance grounding system is constructed, and a grounding fault with different line and phase is set in the simulation model to obtain a simulation model of grounding fault with different line and phase. Run the simulation model of the non-circuit and non-phase grounding fault and collect the sample zero-sequence current data corresponding to each outgoing line of the simulation model. The zero-sequence current data of the samples is preprocessed to obtain the sample data features corresponding to the zero-sequence current data of the samples, and a corresponding sample label is set for each sample data feature; wherein, the sample label includes fault or non-fault; The convolutional neural network is trained using the features of the sample data and their corresponding sample labels to obtain the trained convolutional neural network. The real-time zero-sequence current waveform corresponding to each outgoing line in the low-resistance grounding system is collected, and the trained convolutional neural network is scheduled to identify the real-time data features of the real-time zero-sequence current waveform to determine the grounding fault detection result corresponding to the low-resistance grounding system.

2. The ground fault detection method based on convolutional neural networks according to claim 1, characterized in that, Before acquiring the real-time zero-sequence current waveform corresponding to each outgoing line in the low-resistance grounding system, the following steps are also included: Collect the zero-sequence current amplitude corresponding to the neutral line of the low-resistance grounding system; If the amplitude of the zero-sequence current corresponding to the neutral line exceeds the preset setting value, then proceed to the step of collecting the real-time zero-sequence current waveform corresponding to each outgoing line in the small resistance grounding system; otherwise, continue to monitor the amplitude of the zero-sequence current corresponding to the neutral line until the amplitude of the zero-sequence current corresponding to the neutral line exceeds the setting value, and then proceed to the step of collecting the real-time zero-sequence current waveform corresponding to each outgoing line in the small resistance grounding system.

3. The grounding fault detection method based on convolutional neural networks according to claim 1, characterized in that, A non-linear, non-phase grounding fault is set up in the simulation model to obtain a simulation model of the non-linear, non-phase grounding fault, including: In the simulation model, ground faults are set on different outgoing lines, and the phase lines that cause ground faults are different on different outgoing lines, thus obtaining a simulation model of ground faults with different outgoing lines and different phases.

4. The grounding fault detection method based on convolutional neural networks according to claim 1, characterized in that, Run the simulation model of the non-circuit and non-phase grounding fault, and collect sample zero-sequence current data corresponding to each outgoing line of the simulation model, including: Run the simulation model of the different phases and different lines grounding fault, and sample the analog signal output by each outgoing line at a preset sampling time to obtain the discrete sample points corresponding to each outgoing line. For any discrete sample point corresponding to an outgoing line, the discrete sample point is quantized to obtain a quantized discrete sample point, and the quantized discrete sample point is converted into a digital signal to obtain sample zero-sequence current data.

5. The grounding fault detection method based on convolutional neural networks according to claim 1, characterized in that, The sample zero-sequence current data is preprocessed to obtain the sample data features corresponding to the sample zero-sequence current data, and a corresponding sample label is set for each sample data feature, including: The sample zero-sequence current data is converted into a two-dimensional image by Gram angle field to obtain the sample data features corresponding to the sample zero-sequence current data. If a grounding fault is set on the outgoing line corresponding to the sample data feature, then the sample label corresponding to the sample data feature is set as fault. If the outgoing line corresponding to the sample data feature is not set to ground fault, then the sample tag corresponding to the sample data feature is set to non-fault.

6. The grounding fault detection method based on convolutional neural networks according to claim 5, characterized in that, The sample zero-sequence current data is converted into a two-dimensional image using Gram angle field to obtain the sample data features corresponding to the sample zero-sequence current data, including: The sample zero-sequence current data is normalized to obtain normalized sample zero-sequence current data. The normalized sample zero-sequence current data is encoded as a polar coordinate angle, and the sampling time corresponding to the normalized sample zero-sequence current data is encoded as a polar coordinate radius to obtain the target encoded data. Extract the Gram angle and field or Gram angle difference field corresponding to the target encoded data to obtain the sample data features corresponding to the sample zero-sequence current data.

7. The grounding fault detection method based on convolutional neural networks according to claim 6, characterized in that, The normalized sample zero-sequence current data is encoded into polar coordinates with the following angles: ； in, Let i represent the i-th data point in the normalized sample zero-sequence current data, where i = 1, 2, ..., I, and I represents the total number of data points in the normalized sample zero-sequence current data. This represents the set of data points in the normalized sample zero-sequence current data. This represents the polar coordinate angle obtained from the encoding.

8. The grounding fault detection method based on convolutional neural networks according to claim 6, characterized in that, The sampling time corresponding to the normalized zero-sequence current data is encoded into polar coordinates with a radius of: ； in, This represents the sampling time corresponding to the i-th data point in the normalized zero-sequence current data. Represents the normalization factor. This represents the polar coordinate radius obtained from the encoding.

9. A ground fault detection device based on a convolutional neural network, wherein the ground fault detection device based on a convolutional neural network is used to perform the ground fault detection method based on a convolutional neural network as described in any one of claims 1 to 8, characterized in that, include: The fault simulation module is used to construct a simulation model corresponding to a low-resistance grounding system, and to set up a cross-line and cross-phase grounding fault in the simulation model to obtain a cross-line and cross-phase grounding fault simulation model. The sample data acquisition module is used to run the simulation model of the non-linear and non-phase grounding fault and to acquire the sample zero-sequence current data corresponding to each outgoing line of the simulation model. The sample data processing module is used to preprocess the sample zero-sequence current data to obtain the sample data features corresponding to the sample zero-sequence current data, and to set a corresponding sample label for each sample data feature; wherein, the sample label includes fault or non-fault. The deep learning module is used to train the convolutional neural network using the features of the sample data and their corresponding sample labels, and to obtain the trained convolutional neural network. The ground fault detection module is used to collect the real-time zero-sequence current waveform corresponding to each outgoing line in the low-resistance grounding system, and to schedule the trained convolutional neural network to identify the real-time data features of the real-time zero-sequence current waveform to determine the ground fault detection result corresponding to the low-resistance grounding system.

10. An electronic device, characterized in that, Including processor and memory; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the ground fault detection method based on a convolutional neural network as described in any one of claims 1 to 8.

Images