A single-phase ground fault detection method, system and storage medium for a small current grounding system

Abstract

The application discloses a single-phase ground fault detection method and system for a small-current grounding system and a storage medium, and the method comprises the following steps: acquiring zero sequence current and zero sequence voltage of a to-be-detected line in a to-be-detected area; if the zero sequence voltage of the to-be-detected line is greater than a starting value or a zero sequence voltage mutation value is greater than a set value, entering a single-phase ground fault detection step; otherwise, repeating the above steps by taking another line in the to-be-detected area as the to-be-detected line until entering the single-phase ground fault detection step; the single-phase ground fault detection step comprises the following steps: pre-processing the zero sequence current and the zero sequence voltage of the to-be-detected line to obtain pre-processed image data; and detecting the pre-processed image data by using a trained single-phase ground fault detection model to obtain a single-phase ground fault detection result; and the application can improve the accuracy and anti-interference capability of single-phase ground fault detection in the small-current grounding system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of the cross discipline of artificial intelligence and power distribution network, and in particular to a single-phase grounding fault detection method and system for a small current grounding system and a storage medium. BACKGROUND

[0002] The low-voltage distribution network in China mainly adopts a small current grounding mode for operation, and common small current grounding systems include three types of neutral point non-grounding, grounding through an arc suppression coil and grounding through a resistor. After long-term practice and exploration, the first two grounding modes are mainly used in the power distribution network in China. In these systems, the single-phase grounding fault has the highest probability of occurrence, accounting for more than 80% of the total number of faults. However, when a fault occurs, it is disturbed by various external factors, the fault current is small and the situation is complex, which brings great challenges to fault line selection (detecting the fault line).

[0003] In the line selection of a small current grounding system, traditional methods include zero sequence current amplitude and phase comparison method, zero sequence admittance method, etc., which are based on the relationship between zero sequence voltage and current as a criterion. However, these methods more or less cannot identify the fault line. Using artificial intelligence methods to identify images can better collect zero sequence current and zero sequence voltage information. With the rapid development of artificial intelligence technology, image classification, as one of the core tasks in the field of computer vision, has attracted widespread attention and research. Traditional image classification methods mainly rely on manual feature design, such as color, texture and shape, etc. These methods are limited in handling complex and diverse images. With the rise of deep learning, especially the introduction of convolutional neural networks (CNN), image classification has brought revolutionary breakthroughs. Through automatic extraction of multi-level features, CNN can achieve high-precision classification on large-scale image datasets. In recent years, more and more people use artificial intelligence methods for fault detection, but the activation function used in the network structure is not perfect, and the accuracy of small current grounding system grounding fault detection is not good, and the anti-noise ability is not good.

[0004] The information disclosed in this BACKGROUND section is only for the purpose of increasing the understanding of the background of the present application and should not be taken as an acknowledgment or any form of suggestion that this information forms prior art with respect to the present application. SUMMARY

[0005] The purpose of the present application is to provide a single-phase grounding fault detection method and system for a small current grounding system, which can improve the accuracy and anti-interference ability of single-phase grounding fault detection in a small current grounding system.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0007] In a first aspect, the present application provides a single-phase ground fault detection method for a small current grounding system, comprising:

[0008] obtaining zero sequence current and zero sequence voltage of a to-be-tested line in a to-be-tested area;

[0009] if the zero sequence voltage of the to-be-tested line is greater than a starting value, or the zero sequence voltage mutation value is greater than a set value, entering a single-phase ground fault detection step, otherwise, repeating the above steps with another line in the to-be-tested area as the to-be-tested line until entering the single-phase ground fault detection step;

[0010] the single-phase ground fault detection step comprises:

[0011] preprocessing the zero sequence current and the zero sequence voltage of the to-be-tested line to obtain preprocessed image data;

[0012] detecting the preprocessed image data using a trained single-phase ground fault detection model to obtain a single-phase ground fault detection result.

[0013] Optionally, the preprocessing the zero sequence current and the zero sequence voltage of the to-be-tested line to obtain preprocessed image data comprises:

[0014] setting a time window T, and generating a time-domain image of the zero sequence current and the zero sequence voltage based on the time window T and the zero sequence current and the zero sequence voltage of the to-be-tested line;

[0015] normalizing the time-domain image of the zero sequence current and the zero sequence voltage to obtain a normalized time-domain image of the zero sequence current and the zero sequence voltage;

[0016] superimposing the normalized time-domain image of the zero sequence current and the zero sequence voltage to obtain the preprocessed image data.

[0017] Optionally, the normalizing the time-domain image of the zero sequence current and the zero sequence voltage is performed according to the following formula:

[0018]

[0019] wherein, is a modulus value of an electrical quantity at a current time in the normalized time-domain image, the electrical quantity being represented by the zero sequence voltage or the zero sequence current; is a modulus value of the electrical quantity at the current time in the time-domain image; is a modulus value of a maximum electrical quantity within the time window T.

[0020] Optionally, the training step of the single-phase ground fault detection model comprises:

[0021] Obtain the image dataset, which includes time-domain images of zero-sequence current and voltage preprocessed for normal and faulty lines;

[0022] The image dataset was used as the training set for the single-phase ground fault detection model, and "line normal" and "line fault" were used as labels.

[0023] Based on the training set, the single-phase ground fault detection model is trained to determine the optimal parameters of the single-phase ground fault detection model;

[0024] Based on the optimal parameters of the single-phase ground fault detection model, the trained single-phase ground fault detection model is obtained.

[0025] Optionally, the single-phase grounding fault detection model is obtained by improving the ResNet neural network using the Kolmogorov-Arnold theorem, and includes: a pooling layer, a first convolutional layer, a first residual module, a second residual module, a third residual module, a fourth residual module, a fully connected layer, and a KAN classifier connected in sequence; the first residual module includes three first residual blocks connected in sequence; the first residual block includes a second convolutional layer, a KAN activation function layer, a third convolutional layer, a KAN activation function layer, and a fourth convolutional layer connected in sequence. The second residual module includes four sequentially connected second residual blocks; each second residual block includes a sequentially connected fifth convolutional layer, a KAN activation function layer, a sixth convolutional layer, a KAN activation function layer, and a seventh convolutional layer. The third residual module includes six sequentially connected third residual blocks; each third residual block includes a sequentially connected fourth convolutional layer, a KAN activation function layer, an eighth convolutional layer, a KAN activation function layer, and a ninth convolutional layer. The fourth residual module includes three sequentially connected fourth residual blocks; each fourth residual block includes a sequentially connected seventh convolutional layer, a KAN activation function layer, a tenth convolutional layer, a KAN activation function layer, and an eleventh convolutional layer.

[0026] Optionally, the first convolutional layer includes 64 7x7 convolutional kernels; the second convolutional layer includes 64 1x1 convolutional kernels; the third convolutional layer includes 64 3x3 convolutional kernels; the fourth convolutional layer includes 256 1x1 convolutional kernels; the fifth convolutional layer includes 128 1x1 convolutional kernels; the sixth convolutional layer includes 128 3x3 convolutional kernels; the seventh convolutional layer includes 512 1x1 convolutional kernels; the eighth convolutional layer includes 256 3x3 convolutional kernels; the ninth convolutional layer includes 1024 1x1 convolutional kernels; the tenth convolutional layer includes 512 3x3 convolutional kernels; and the eleventh convolutional layer includes 2048 1x1 convolutional kernels.

[0027] Optionally, the KAN activation function layer uses the KAN activation function, as shown in the following formula:

[0028]

[0029] in, and All of these are trainable parameters; Let x be the SiLu activation function; This is an intermediate function; It is a spline activation function.

[0030] Optionally, the starting value is set to 0.1 times the rated voltage.

[0031] Secondly, the present invention provides a single-phase grounding fault detection system for a low-current grounding system, comprising:

[0032] The data acquisition module is used to acquire the zero-sequence current and zero-sequence voltage of the circuit under test in the test area;

[0033] The judgment module is used to enter the single-phase ground fault detection step if the zero-sequence voltage of the line under test is greater than the starting value or the sudden change of the zero-sequence voltage is greater than the set value. Otherwise, the above steps are repeated in the test area with another line as the line under test until the single-phase ground fault detection step is entered.

[0034] A single-phase ground fault detection module is connected to a data acquisition module and a judgment module, respectively. It is used to preprocess the zero-sequence current and zero-sequence voltage of the line under test to obtain preprocessed image data; and to use a trained single-phase ground fault detection model to detect the preprocessed image data to obtain the single-phase ground fault detection result.

[0035] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed, implements the single-phase grounding fault detection method for a low-current grounding system as described in the first aspect of the claim.

[0036] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0037] The application provides a single-phase grounding fault detection method, system and storage medium for a small current grounding system, which is based on the obtained zero sequence current and zero sequence voltage of a to-be-detected line, and first judges whether the to-be-detected line needs to be detected, and then performs single-phase grounding fault detection or reselects a line as the to-be-detected line based on the judgment result; wherein the single-phase grounding fault detection is to use a trained single-phase grounding fault detection model to accurately identify whether there is a single-phase grounding fault, obtain a detection result, improve the accuracy of fault detection, and has anti-interference ability under the proof of experimental results.

[0038] The application provides a single-phase grounding fault detection method for a small current grounding system, and an image data set is used for a training set of a single-phase grounding fault detection model, wherein the image data set includes a time domain image after zero sequence current voltage preprocessing of a normal line and a fault line, the time domain image is obtained by normalizing and superimposing a zero sequence current time domain image and a zero sequence voltage time domain image, so that the trained single-phase grounding fault detection model has good anti-interference ability.

[0039] The application provides a single-phase grounding fault detection method for a small current grounding system, and a single-phase grounding fault detection model thereof is obtained based on an improved ResNet neural network according to the Kolmogorov-Arnold theorem, and compared with original convolutional neural networks (VGG16 network, AlexNet network, ResNet network and DenseNet network), the accuracy of fault detection is improved; and the single-phase grounding fault detection model still has a high detection accuracy in a noisy environment, which indicates that the method has good anti-interference ability.

[0040] The application provides a single-phase grounding fault detection method for a small current grounding system, and an activation function in a single-phase grounding fault detection model is improved, so that the activation function becomes more complete, and the accuracy of the original network is higher while the activation function becomes complex, thereby increasing the accuracy of single-phase grounding fault detection. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 Fig. 1 shows a single-phase grounding fault detection method flowchart in an embodiment of the application;

[0042] Figure 2 Fig. 2 shows a preprocessed image data example diagram of a fault line in an embodiment of the application;

[0043] Figure 3 Fig. 3 shows a preprocessed image data example diagram of a normal line in an embodiment of the application;

[0044] Figure 4The diagram shown is an R-KAN network structure diagram in one embodiment of the present invention;

[0045] Figure 5 The figure shown is a simulation model diagram of the power distribution network after adding DG in one embodiment of the present invention;

[0046] Figure 6 The diagram shown illustrates the change in loss value during the training process of the R-KAN network in one embodiment of the present invention. Detailed Implementation

[0047] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0048] Example 1

[0049] like Figure 1 As shown in the figure, this invention provides a method for detecting single-phase grounding faults in low-current grounding systems, comprising the following steps:

[0050] S1: Obtain the zero-sequence current and zero-sequence voltage of the circuit under test in the area under test;

[0051] S2: If the zero-sequence voltage of the line under test is greater than the starting value, or the sudden change in zero-sequence voltage is greater than the set value, then proceed to the single-phase ground fault detection step; otherwise, replace the line under test in the test area with another line and repeat the above steps until the single-phase ground fault detection step is entered.

[0052] The single-phase ground fault detection steps include:

[0053] S3: Preprocess the zero-sequence current and zero-sequence voltage of the circuit under test to obtain preprocessed image data;

[0054] S4: Use the trained single-phase ground fault detection model to detect the preprocessed image data to obtain the single-phase ground fault detection result.

[0055] This embodiment provides a single-phase grounding fault detection method for low-current grounding systems. Based on the acquired zero-sequence current and zero-sequence voltage of the line under test, it first determines whether the line under test needs to be fault detected, and then performs single-phase grounding fault detection or reselects a line as the line under test based on the judgment result. The single-phase grounding fault detection uses a trained single-phase grounding fault detection model to accurately identify whether a single-phase grounding fault exists, obtains the fault detection result, and improves the accuracy and anti-interference capability of fault line selection.

[0056] In this embodiment, in step S1, the zero-sequence current and zero-sequence voltage of the line under test in the area under test are acquired in real time through the sensors (zero-sequence current transformer and zero-sequence voltage transformer) built into the microcomputer protection device. The microcomputer protection device is a key device in modern power systems. It combines computer technology, signal processing technology and automatic control technology, and has the characteristics of high precision, high reliability and intelligence. Since this device is existing technology and is not the innovation of this invention, it will not be described in detail here.

[0057] In this embodiment, if the zero-sequence voltage of the line under test is greater than the starting value, or the sudden change in zero-sequence voltage is greater than the set value, then the single-phase ground fault detection step is entered; otherwise, the above steps are repeated with a different line in the test area until the single-phase ground fault detection step is entered. Specifically, this includes:

[0058] Based on the zero-sequence voltage U0 obtained in real time in step S1, it is determined whether the line under test needs to enter the single-phase ground fault detection. If it is needed, the single-phase ground fault detection steps S3-S4 are entered. If it is not needed, another line is selected as the line under test and the process returns to step S1.

[0059] The judgment criterion is: zero-sequence voltage U0 > starting value U set Or, the zero-sequence voltage mutation ΔU0 > the preset value ε;

[0060] Specifically, U set The value is 0.1 times the rated voltage, ΔU0 is the current zero-sequence voltage minus the previous zero-sequence voltage, and ε is 0.5-1V.

[0061] In this embodiment, step S3 involves preprocessing the zero-sequence current and zero-sequence voltage of the circuit under test to obtain preprocessed image data. The specific steps are as follows:

[0062] S31: Set the time window T, and generate time-domain images of zero-sequence current and zero-sequence voltage based on the time window T and the zero-sequence current and zero-sequence voltage of the circuit under test;

[0063] Specifically, such as Figures 2-3 As shown, the time window T is set to 0.06 seconds. Based on the zero-sequence current and zero-sequence voltage collected by the protection device, time-domain images of the zero-sequence current and zero-sequence voltage are generated using data processing software or algorithms, according to the setting of the time window T and the requirements of the sampling frequency. The time-domain image usually uses time as the horizontal axis and zero-sequence current or zero-sequence voltage as the vertical axis, and displays the trend of data change in the form of a curve.

[0064] S32: Normalize the time-domain images of zero-sequence current and zero-sequence voltage to obtain normalized time-domain images of zero-sequence current and zero-sequence voltage.

[0065] Specifically, the normalization formula is as follows:

[0066]

[0067] wherein, is the modulus value of the electrical quantity at the current time in the normalized time-domain image, and the electrical quantity is represented as zero sequence voltage or zero sequence current; is the modulus value of the electrical quantity at the current time in the time-domain image; is the modulus value of the maximum electrical quantity in the time window T.

[0068] S33: superimpose the normalized time-domain images of the zero sequence current and the zero sequence voltage to obtain the preprocessed image data, as shown in Figure 2 and Figure 3 The preprocessed image data of the fault line and the normal line are shown in

[0069] In the present embodiment, the training step of the single-phase ground fault detection model in step S4 is as follows:

[0070] Obtain an image data set, wherein the image data set includes the preprocessed time-domain images of the zero sequence current and voltage of the normal line and the fault line;

[0071] Take the image data set as the training set of the single-phase ground fault detection model, and take “line normal” and “line fault” as labels; wherein “line fault” represents that the line has a single-phase ground fault;

[0072] Based on the training set, train the single-phase ground fault detection model to determine the optimal parameters of the single-phase ground fault detection model;

[0073] According to the optimal parameters of the single-phase ground fault detection model, obtain the trained single-phase ground fault detection model.

[0074] Specifically, the single-phase ground fault detection model (R-KAN network, ResNet-Kolmogorov-Arnold Networks network) is obtained by improving the ResNet neural network based on the Kolmogorov-Arnold theorem;

[0075] wherein, the Kolmogorov-Arnold theorem means that if is a multi-variable continuous function on a bounded domain, then can be written as a combination of a finite number of single-variable continuous functions and binary function addition operations.

[0076] Further, for a smooth function ;

[0077]

[0078] in, Let it be a function, and ; For a mapping function, , Let n represent the set of real numbers, and p, q represent the sequence numbers.

[0079] Specifically, such as Figure 4 As shown, the structure of the single-phase ground fault detection model, i.e., the R-KAN network, includes:

[0080] The pooling layer, the first convolutional layer (conv1), the first residual module, the second residual module, the third residual module, the fourth residual module, the fully connected layer, and the KAN classifier are connected in sequence.

[0081] The first residual module includes three sequentially connected first residual blocks; the first residual block includes a sequentially connected second convolutional layer (conv2), a KAN activation function layer (KAN), a third convolutional layer (conv3), a KAN activation function layer (KAN), and a fourth convolutional layer (conv4).

[0082] The second residual module includes four sequentially connected second residual blocks; each second residual block includes a sequentially connected fifth convolutional layer (conv5), a KAN activation function layer, a sixth convolutional layer (conv6), a KAN activation function layer, and a seventh convolutional layer (conv7).

[0083] The third residual module consists of six sequentially connected third residual blocks; each third residual block consists of a sequentially connected fourth convolutional layer (conv4), a KAN activation function layer, an eighth convolutional layer (conv8), a KAN activation function layer, and a ninth convolutional layer (conv9).

[0084] The fourth residual module consists of three sequentially connected fourth residual blocks; the fourth residual block consists of a sequentially connected seventh convolutional layer (conv7), a KAN activation function layer, a tenth convolutional layer (conv10), a KAN activation function layer, and an eleventh convolutional layer (conv11).

[0085] Specifically, the first convolutional layer includes 64 convolutional kernels of 7x7 size; the second convolutional layer includes 64 convolutional kernels of 1x1 size; the third convolutional layer includes 64 convolutional kernels of 3x3 size; the fourth convolutional layer includes 256 convolutional kernels of 1x1 size; the fifth convolutional layer includes 128 convolutional kernels of 1x1 size; the sixth convolutional layer includes 128 convolutional kernels of 3x3 size; the seventh convolutional layer includes 512 convolutional kernels of 1x1 size; the eighth convolutional layer includes 256 convolutional kernels of 3x3 size; the ninth convolutional layer includes 1024 convolutional kernels of 1x1 size; the tenth convolutional layer includes 512 convolutional kernels of 3x3 size; and the eleventh convolutional layer includes 2048 convolutional kernels of 1x1 size.

[0086] In the embodiment, the formula of the neural network forward transmission is changed by using the above-mentioned Kolmogorov-Arnold theorem (KAN theorem) as follows:

[0087]

[0088] wherein, is gradually nested into the check-in entry result; is the mapping function of the kth formula nesting, is the input of the network, is the output of the network, the role of is to combine the weight bias operation in with the activation function, is called KAN activation function.

[0089]

[0090]

[0091] wherein, and are trainable parameters; is a SiLu activation function, and x is an object; is an intermediate function; is a certain spline activation function.

[0092] In the embodiment, the model training and verification hardware and software environment of the patent are as follows: 13th Gen intelcore (TM) i5-13600KF processor, NVIDIA 4070 ti super (16G), memory size of 32G, python3.11, CUDN121, pytorch2.2.1.

[0093] Embodiment 2

[0094] Based on the single-phase ground fault detection method for small current grounding system provided in Embodiment 1, this embodiment provides an example, which is introduced as follows:

[0095] This embodiment simulates the system by using PSCAD / EMTDC to build a simulation model. The simulation structure diagram is as shown in Figure 5 The simulation model diagram of the distribution network after adding DG (Distributed Generation) is shown in the figure, the simulation model has four feeder lines, the lengths are 4km, 5km, 4km and 5km respectively, and the remaining parameters are set: the system power voltage is 10kV, the frequency is 50Hz, the number of DGs is 2, one of which is a wind turbine model and the other is a photovoltaic model, the inductance value Lg of the arc suppression coil is 0.3H. The simulation sampling frequency is 40MHz, and the time of starting the fault is 0.8s. The time window of the neural network training and verification set is set to 0.005s. The first feeder line is divided into 1.5km, 1.5km and 1km, and the other feeder lines are set to half. The line segment close to the bus is an overhead line, and the line segment far away is a cable line. The line parameters are shown in Table 1:

[0096] Table 1 Line parameters

[0097]

[0098] The R-KAN neural network is first trained, and the loss value change of the training process is as shown in Figure 6 With the increase of the number of iterations, the loss value gradually decreases and finally approaches to 0.

[0099] To verify the R-KAN neural network, a test set needs to be constructed, and the method of constructing the test set is as follows:

[0100] A plurality of sets of zero sequence voltage and zero sequence current data are obtained by simulation, and the superimposed waveforms of the zero sequence voltage and the zero sequence current are made. The composition of the training set is shown in Table 2. In Table 2 and Table 3, (a, b km) means that the access point is in the feeder line a, and the distance from the first section is b kilometers.

[0101] The accuracy of the data set is defined as the ratio of the number of correct samples to the total number of samples.

[0102] Table 2 Ground fault parameters of test data set

[0103]

[0104] Verification test 1 (accuracy comparison of different convolutional neural networks):

[0105] The ground resistance values and positions of the fault points are changed, and a test data set is made as in Table 2. The test data set is respectively put into a VGG16 network, an AlexNet network, a ResNet network, a DenseNet network and an R-KAN network, and the detection accuracy results are as shown in Table 3:

[0106] Table 3 Detection accuracy of different types of convolutional neural networks

[0107]

[0108] From the above table, it can be seen that the detection accuracy of R-KAN reaches 100%, which is higher than the accuracy of VGG16 network, AlexNet network, ResNet network and DenseNet network. The R-KAN neural network, i.e. the single-phase ground fault detection model proposed in the application, effectively improves the accuracy of single-phase ground fault detection in a small-current grounding system.

[0109] Verification test 2 (noise data set accuracy comparison):

[0110] In an active power distribution network, power electronic devices such as inverters can introduce noise, which can interfere with the voltage and current of the system, thereby possibly affecting the accuracy of fault identification. Therefore, it is necessary to verify the fault identification accuracy of the line affected by noise. For this purpose, 20Hz Gaussian white noise is added to the original zero sequence voltage and zero sequence current waveform to generate a noise data set. The test data set is respectively put into a VGG16 network, an AlexNet network, a ResNet network, a DenseNet network and an R-KAN network, and the detection accuracy results are as shown in Table 4.

[0111] Table 4 Detection accuracy of different types of convolutional neural networks on noise set

[0112]

[0113] From the above table, it can be seen that R-KAN has the best anti-interference ability. Under the influence of Gaussian white noise, the detection accuracy is still 99.11%.

[0114] Verification test 3 (field recording verification test):

[0115] The data after the fault occurs in the field is extracted by recording software, and a zero sequence voltage and zero sequence current superimposed graph is made, which is put into the R-KAN network for field recording verification test. Through field verification, the accuracy of R-KAN for small-current single-phase ground fault detection is 100%.

[0116] In summary, the patent improves the ResNet neural network to R-KAN neural network by using the Kolmogorov-Arnold theorem, so that the accuracy of single-phase ground fault line detection in the small current grounding system is improved compared with the original convolutional neural network (VGG16 network, AlexNet network, ResNet network, DenseNet network). The R-KAN neural network has better anti-interference ability than the original convolutional neural network. In addition, the R-KAN network is verified by field recording data, and the accuracy of fault detection is 100%. The new artificial intelligence network proposed in the patent improves the accuracy of single-phase ground fault line detection in the small current grounding system.

[0117] Embodiment 3

[0118] The embodiment provides a single-phase ground fault detection system for a small current grounding system, comprising:

[0119] A data acquisition module is configured to acquire zero sequence current and zero sequence voltage of a to-be-detected line in a to-be-detected area.

[0120] A judgment module is configured to, if the zero sequence voltage of the to-be-detected line is greater than a starting value or a zero sequence voltage mutation value is greater than a set value, enter a single-phase ground fault detection step, otherwise, repeat the above steps with another line in the to-be-detected area as the to-be-detected line until the single-phase ground fault detection step is entered.

[0121] A single-phase ground fault detection module is connected with the data acquisition module and the judgment module, configured to pre-process the zero sequence current and the zero sequence voltage of the to-be-detected line to obtain pre-processed image data, and configured to detect the pre-processed image data by using a trained single-phase ground fault detection model to obtain a single-phase ground fault detection result.

[0122] Embodiment 4

[0123] The embodiment provides a computer readable storage medium storing a computer program, and the computer program is executed to realize the single-phase ground fault detection method for a small current grounding system in the embodiment 1.

[0124] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0125] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks. Figure 1 The flowchart and / or block diagram in the drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart and / or block diagrams can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical functions ("instruction"). It should also be noted that each block of the flowchart and / or block diagrams and a combination of flows and / or blocks in the flowchart and / or block diagrams can be implemented by computer program instructions. Although the computer program instructions can be written in any combination of one or more programming languages, those of ordinary skill in the art should understand that the computer program instructions can be written in any programming language of the present application. Figure 1 The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks.

[0126] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instruction means that implement the function specified in the flowchart block or blocks. Figure 1 The flowchart and / or block diagram in the drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart and / or block diagrams can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical functions ("instruction"). It should also be noted that each block of the flowchart and / or block diagrams and a combination of flows and / or blocks in the flowchart and / or block diagrams can be implemented by computer program instructions. Although the computer program instructions can be written in any combination of one or more programming languages, those of ordinary skill in the art should understand that the computer program instructions can be written in any programming language of the present application. Figure 1 The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks.

[0127] The embodiments of the present application are described above with reference to the accompanying drawings, and are not limited to the above-mentioned specific embodiments, and the above-mentioned specific embodiments are merely illustrative, but not restrictive, and those of ordinary skill in the art can make many forms under the guidance of the present application without departing from the purpose of the present application and the scope of protection of the claims, and these all belong to the protection of the present application.

Claims

1. A method for detecting single-phase grounding faults in low-current grounding systems, characterized in that, include: Obtain the zero-sequence current and zero-sequence voltage of the circuit under test in the area under test; If the zero-sequence voltage of the line under test is greater than the starting value, or the sudden change in zero-sequence voltage is greater than the set value, then the single-phase ground fault detection step is entered; otherwise, the above steps are repeated with another line in the test area until the single-phase ground fault detection step is entered. The single-phase ground fault detection steps include: The zero-sequence current and zero-sequence voltage of the circuit under test are preprocessed to obtain preprocessed image data; The pre-trained single-phase ground fault detection model is used to detect the pre-processed image data to obtain the single-phase ground fault detection result; The single-phase ground fault detection model is obtained by improving the ResNet neural network using the Kolmogorov-Arnold theorem, and includes: sequentially connected pooling layers, a first convolutional layer, a first residual module, a second residual module, a third residual module, a fourth residual module, a fully connected layer, and a KAN classifier; the first residual module includes three sequentially connected first residual blocks; the first residual block includes sequentially connected second convolutional layers, KAN activation function layers, third convolutional layers, KAN activation function layers, and fourth convolutional layers; the second residual module includes four sequentially connected second... The residual block; the second residual block includes a fifth convolutional layer, a KAN activation function layer, a sixth convolutional layer, a KAN activation function layer, and a seventh convolutional layer connected in sequence; the third residual module includes six third residual blocks connected in sequence; the third residual block includes a fourth convolutional layer, a KAN activation function layer, an eighth convolutional layer, a KAN activation function layer, and a ninth convolutional layer connected in sequence; the fourth residual module includes three fourth residual blocks connected in sequence; the fourth residual block includes a seventh convolutional layer, a KAN activation function layer, a tenth convolutional layer, a KAN activation function layer, and an eleventh convolutional layer connected in sequence.

2. The method for detecting single-phase grounding faults in a low-current grounding system according to claim 1, characterized in that, The zero-sequence current and zero-sequence voltage of the circuit under test are preprocessed to obtain preprocessed image data, including: Set a time window T, and generate time-domain images of zero-sequence current and zero-sequence voltage based on the time window T and the zero-sequence current and zero-sequence voltage of the circuit under test. The time-domain images of zero-sequence current and zero-sequence voltage are normalized to obtain normalized time-domain images of zero-sequence current and zero-sequence voltage. The time-domain images of the zero-sequence current and zero-sequence voltage after normalization are superimposed to obtain the preprocessed image data.

3. The method for detecting single-phase grounding faults in a low-current grounding system according to claim 2, characterized in that, The time-domain images of zero-sequence current and zero-sequence voltage are normalized using the following formula: ； in, The magnitude of the electrical quantity at the current moment in the normalized time-domain image is represented as zero-sequence voltage or zero-sequence current. The magnitude of the electrical quantity at the current moment in the time-domain image; It represents the magnitude of the largest electrical quantity within the time window T.

4. The method for detecting single-phase grounding faults in a low-current grounding system according to claim 3, characterized in that, The training steps of the single-phase ground fault detection model include: Obtain the image dataset, which includes time-domain images of zero-sequence current and voltage preprocessed for normal and faulty lines; The image dataset was used as the training set for the single-phase ground fault detection model, and "line normal" and "line fault" were used as labels. Based on the training set, the single-phase ground fault detection model is trained to determine the optimal parameters of the single-phase ground fault detection model; Based on the optimal parameters of the single-phase ground fault detection model, the trained single-phase ground fault detection model is obtained.

5. The method for detecting single-phase grounding faults in a low-current grounding system according to claim 4, characterized in that, The first convolutional layer includes 64 7x7 convolutional kernels; the second convolutional layer includes 64 1x1 convolutional kernels; the third convolutional layer includes 64 3x3 convolutional kernels; the fourth convolutional layer includes 256 1x1 convolutional kernels; the fifth convolutional layer includes 128 1x1 convolutional kernels; the sixth convolutional layer includes 128 3x3 convolutional kernels; the seventh convolutional layer includes 512 1x1 convolutional kernels; the eighth convolutional layer includes 256 3x3 convolutional kernels; the ninth convolutional layer includes 1024 1x1 convolutional kernels; the tenth convolutional layer includes 512 3x3 convolutional kernels; and the eleventh convolutional layer includes 2048 1x1 convolutional kernels.

6. The method for detecting single-phase grounding faults in a low-current grounding system according to claim 5, characterized in that, The KAN activation function layer uses the KAN activation function, as shown in the following formula: ； in, and All of these are trainable parameters; Let x be the SiLu activation function; This is an intermediate function; It is a spline activation function.

7. The method for detecting single-phase grounding faults in a low-current grounding system according to claim 1, characterized in that, The starting value is set to 0.1 times the rated voltage.

8. A single-phase ground fault detection system for a low-current grounding system, implementing the single-phase ground fault detection method for a low-current grounding system as described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire the zero-sequence current and zero-sequence voltage of the circuit under test in the test area; The judgment module is used to enter the single-phase ground fault detection step if the zero-sequence voltage of the line under test is greater than the starting value or the sudden change of the zero-sequence voltage is greater than the set value. Otherwise, the above steps are repeated in the test area with another line as the line under test until the single-phase ground fault detection step is entered. A single-phase ground fault detection module is connected to a data acquisition module and a judgment module respectively. It is used to preprocess the zero-sequence current and zero-sequence voltage of the line under test to obtain preprocessed image data. This is used to detect single-phase grounding faults by using a trained single-phase grounding fault detection model on the preprocessed image data to obtain single-phase grounding fault detection results.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed, implements the single-phase grounding fault detection method for a low-current grounding system as described in any one of claims 1-8.

Images