Single-phase ground fault traveling wave positioning method based on intelligent terminal

Abstract

The application discloses a single-phase ground fault traveling wave positioning method based on an intelligent terminal, and is applied to the technical field of intelligent electric appliances, and the method comprises the following steps: acquiring line data and line mode wave velocity; when a fault traveling wave is detected, performing phase-mode transformation on the fault traveling wave to acquire a line-mode traveling wave; identifying a first traveling wave head and recording selected traveling wave data; taking the first traveling wave head as a reference, identifying a second traveling wave head and a third traveling wave head through a convolutional neural network model, and further calculating a fault point position. The application fully considers the correlation between different wave heads, effectively captures the complex relationship between different wave heads, so that the second traveling wave head and the third traveling wave head can be accurately detected through the initial wave head, and then the position of the fault occurrence can be more accurately determined, and the precision of fault distance measurement is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent electrical technology, in particular to a single-phase grounding fault traveling wave positioning method based on an intelligent terminal. BACKGROUND

[0002] Single-phase grounding fault traveling wave positioning is a fault positioning technology based on the principle of traveling wave, mainly used in distribution networks in power systems. The basic idea is that when a single-phase grounding fault occurs in a distribution network, a transient traveling wave signal will be generated at the fault point, which will propagate along the line to both ends. By installing traveling wave sensors at the corresponding positions, the time difference of the arrival of the traveling wave signal at each sensor is detected, combined with the wave speed and other parameters of the line, and the position of the fault point is calculated by using double-end positioning or single-end positioning algorithms.

[0003] For double-end positioning, the traveling wave detection devices at both ends need to be strictly synchronized, which is an obstacle in large-scale promotion. For single-end positioning, when a single-phase grounding fault occurs, the traveling wave generated at the fault point will propagate in all directions along the line, and when it encounters nodes with different impedances such as line branches, bifurcation points, transformers, switching devices, and line terminals, complex reflection and refraction phenomena will occur; therefore, it is difficult to accurately identify the subsequent reflected wave front from the traveling wave, which poses a great challenge to the accuracy and reliability of fault positioning.

[0004] In the prior art, a power distribution network fault positioning method based on wavelet transform and CNN is disclosed in Chinese Patent No. 201810541591.1, which proposes a fault positioning method through a CNN convolutional neural network. However, due to the uncertainty of the fault point, it is difficult to identify the traveling wave head, which is manifested in that even two wavelet-transformed traveling wave waveform graphs that look very similar may correspond to completely different fault points. In summary, the existing technology still has many problems for single-end fault positioning. SUMMARY

[0005] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the present application is to provide a single-phase grounding fault traveling wave positioning method based on an intelligent terminal to solve the above-mentioned problems.

[0006] The single-phase grounding fault traveling wave positioning method based on an intelligent terminal of the present application comprises:

[0007] Obtaining line data of a line to be detected, and calculating a line mode wave speed at a preset frequency according to the line data;

[0008] The traveling wave detection device is arranged at one end of the to-be-detected line, and a sliding window length is calculated according to the wave velocity of the line mode and the line length of the to-be-detected line; the sliding window length is the longest time length required for the wave crest of a second traveling wave and the wave crest of a third traveling wave to reach the traveling wave detection device after a first traveling wave reaches the traveling wave detection device; the first traveling wave is a wave head of a fault traveling wave reaching the traveling wave detection device; the second traveling wave is a wave head of the fault traveling wave reaching the other end of the to-be-detected line and then reflecting to reach the traveling wave detection device; and the third traveling wave is a wave head of the fault traveling wave reaching the traveling wave detection device and then reflecting at the fault point to reach the traveling wave detection device again;

[0009] When the traveling wave detection device detects the fault traveling wave, the fault traveling wave is subjected to phase-mode transformation to obtain a line-mode traveling wave;

[0010] The first traveling wave in the line-mode traveling wave is taken as the first traveling wave, and line-mode traveling waves in the line-mode traveling wave within the sliding window length range from the first traveling wave are recorded as selected traveling wave data;

[0011] The first traveling wave is taken as a reference to identify the second traveling wave and the third traveling wave from the selected traveling wave data through a convolutional neural network model, to calculate the wave crest time of the first traveling wave, the wave crest time of the second traveling wave, and the wave crest time of the third traveling wave at the preset frequency, and to further calculate the fault point position.

[0012] In a possible implementation, the training process of the convolutional neural network model includes:

[0013] Line-mode traveling wave sample data is obtained, the line-mode traveling wave sample data being line-mode traveling wave data containing a first traveling wave, a second traveling wave, and a third traveling wave;

[0014] The line-mode traveling wave sample data is decomposed into first frequency component data at multiple frequencies through wavelet transform;

[0015] First traveling wave sampling data is extracted from the multiple first frequency component data to form multiple first vectors corresponding to different frequencies; second traveling wave sampling data is extracted from the multiple first frequency component data to form multiple second vectors corresponding to different frequencies; and third traveling wave sampling data is extracted from the multiple first frequency component data to form multiple third vectors corresponding to different frequencies;

[0016] The first vectors and the second vectors in the same group of line-mode traveling wave sample data are arranged in a first feature matrix in a longitudinal direction in an alternating manner according to frequencies, and the first vectors and the third vectors in the same group of line-mode traveling wave sample data are arranged in a second feature matrix in a longitudinal direction in an alternating manner according to frequencies.

[0017] Assign a second wavefront label to the first feature matrix, and assign a third wavefront label to the second feature matrix;

[0018] The convolutional neural network model is formed by training a convolutional neural network using the first feature matrix or the second feature matrix as input data and the corresponding labels as output data.

[0019] In one possible implementation, the convolutional neural network includes a first convolutional layer, a second convolutional layer, and a fully connected layer connected in sequence.

[0020] Training a convolutional neural network includes:

[0021] The input data is input into the first convolutional layer, and the input data is convolved by a 2*2 convolutional kernel in the first convolutional layer to generate the first convolutional data; the horizontal stride in the first convolutional layer is 1, and the vertical stride is 2.

[0022] The first convolutional data is input into the second convolutional layer, and the first convolutional data is processed by a 2*2 convolutional kernel in the second convolutional layer to generate the second convolutional data; the horizontal stride in the second convolutional layer is 1, and the vertical stride is 1.

[0023] The second convolutional data is input into the fully connected layer, and the parameters of the fully connected layer are adjusted according to the difference between the output of the fully connected layer and the corresponding label.

[0024] In one possible implementation, the sampling of the first row vector, the second row vector, and the third row vector includes:

[0025] The first traveling wave front in the first frequency component data is divided into a preset number of intervals from the initial abrupt change point to the wave crest, and each interval node is sampled as a sampling point and then arranged horizontally to form the first row vector of the corresponding frequency.

[0026] The second row wavefront in the first frequency component data is divided into an average number of intervals from the initial abrupt change point to the wave crest, and each interval node is sampled as a sampling point and then arranged horizontally to form the second row vector of the corresponding frequency.

[0027] The third row wavefront in the first frequency component data is divided into a predetermined number of intervals from the initial abrupt change point to the wave crest, and each interval node is sampled as a sampling point and then arranged horizontally to form the third row vector corresponding to the frequency.

[0028] In one possible implementation, identifying the second and third traveling wavefronts from the selected traveling wave data using a convolutional neural network model, based on the first traveling wavefront, includes:

[0029] The selected traveling wave data is decomposed into second frequency component data at multiple frequencies using wavelet transform;

[0030] Sample data of the first traveling wavefront are extracted from multiple second frequency component data to form multiple reference row vectors corresponding to different frequencies; sample data of wavefronts other than the first traveling wavefront are extracted from multiple second frequency component data to form multiple detection row vectors corresponding to different frequencies.

[0031] The reference row vector and the row vector to be detected are arranged vertically alternately according to frequency to form the feature matrix to be detected;

[0032] The feature matrix to be detected is input into the convolutional neural network model, and the result output by the convolutional neural network model is received to determine whether the feature matrix to be detected corresponds to the second row wavefront, the third row wavefront, or other wavefronts.

[0033] In one possible implementation, sampling of the baseline row vector and the row vector to be detected includes:

[0034] The range from the initial abrupt change point to the peak of the first traveling wave in the second frequency component data is divided into a preset number of intervals, and each interval node is sampled as a sampling point and then arranged horizontally to form the reference row vector of the corresponding frequency.

[0035] The wavefronts of the second frequency component data that are not the first traveling wavefront are divided into a predetermined number of intervals from the initial abrupt change point to the wave crest. Each interval node is then sampled and arranged horizontally to form the corresponding frequency of the row vector to be detected.

[0036] In one possible implementation, determining whether the feature matrix to be detected corresponds to a second wavefront, a third wavefront, or another wavefront includes:

[0037] When the output of the convolutional neural network model corresponds to the second row wavefront, it is determined that the feature matrix to be detected corresponds to the second row wavefront.

[0038] When the output of the convolutional neural network model corresponds to the third row wave head, it is determined that the feature matrix to be detected corresponds to the third row wave head.

[0039] When the output of the convolutional neural network model does not correspond to the second row wavefront or the third row wavefront, it is determined that the feature matrix to be detected corresponds to other wavefronts.

[0040] In one possible implementation, the calculation of the fault location includes:

[0041] The difference between the peak time of the third wavefront and the peak time of the first wavefront is calculated as the first duration, and the difference between the peak time of the second wavefront and the peak time of the first wavefront is calculated as the second duration.

[0042] The first ranging distance is generated by multiplying the first duration by the line mode wave velocity and then dividing by 2; the second ranging distance is generated by multiplying the second duration by the line mode wave velocity and then dividing by 2.

[0043] When the difference between the sum of the first ranging distance and the second ranging distance and the length of the line to be detected is less than or equal to a preset value, the first ranging distance is taken as the distance from the fault point to the traveling wave detection device.

[0044] When the difference between the sum of the first ranging distance and the second ranging distance and the length of the line to be tested is greater than a preset value, the third ranging distance is taken as the distance from the fault point to the traveling wave detection device; the third ranging distance is the line length minus the second ranging distance plus the first ranging distance and then divided by 2.

[0045] In one possible implementation, the number of frequencies after wavelet transform decomposition is 6 to 10.

[0046] In one possible implementation, the preset frequency is 100kHz.

[0047] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0048] This invention presents a traveling wave location method for single-phase grounding faults based on intelligent terminals. It utilizes the powerful capabilities of convolutional neural networks to extract and analyze features from traveling wave signals, fully considering the correlation between different wavefronts and effectively capturing the complex relationships between them. This allows for accurate detection of the second and third traveling wavefronts based on the initial wavefront, leading to a more precise determination of the fault location and significantly improving fault location accuracy. This not only helps in quickly locating fault points, reducing power outage time and maintenance costs, but also improves the overall reliability and operational efficiency of the power system. Attached Figure Description

[0049] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0050] Figure 1 This is a schematic diagram of the method steps in an embodiment of this application;

[0051] Figure 2 This is a schematic diagram of single-end fault location in an embodiment of this application;

[0052] Figure 3 This is a schematic diagram of convolution processing in an embodiment of this application;

[0053] Figure 4 This is a waveform diagram of a 100kHz line mode traveling wave in an embodiment of this application. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0055] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0056] Please refer to the following: Figure 1 This is a flowchart illustrating the traveling wave location method for a single-phase ground fault provided in an embodiment of the present invention. Further, the traveling wave location method for a single-phase ground fault may specifically include the contents described in steps S1-S5.

[0057] S1: Obtain the line data of the line to be tested, and calculate the line mode wave velocity at a preset frequency based on the line data;

[0058] S2: A traveling wave detection device is installed at one end of the line to be tested, and the sliding window length is calculated based on the line mode wave velocity and the line length of the line to be tested; the sliding window length is the longest time required for the peaks of the second and third traveling wave waves to reach the traveling wave detection device after the first traveling wave wavehead reaches the traveling wave detection device; the first traveling wave wavehead is the wavehead of the fault traveling wave reaching the traveling wave detection device; the second traveling wave wavehead is the wavehead of the fault traveling wave reaching the other end of the line to be tested and then reflected before reaching the traveling wave detection device; the third traveling wave wavehead is the wavehead of the fault traveling wave reaching the traveling wave detection device, reflected again at the fault point, and then reaching the traveling wave detection device.

[0059] S3: When the traveling wave detection device detects a fault traveling wave, it performs a phase mode transformation on the fault traveling wave to obtain a line mode traveling wave;

[0060] S4: Take the first traveling wave head in the linear model traveling wave as the first traveling wave head, and record the linear model traveling waves within the sliding window length range from the first traveling wave head as the selected traveling wave data;

[0061] S5: Using the first traveling wave wavehead as a reference, the second and third traveling wave waveheads are identified from the selected traveling wave data through a convolutional neural network model. The peak times of the first, second, and third traveling wave waveheads at the preset frequency are calculated, and the fault location is further calculated.

[0062] In implementing this application embodiment, it is necessary to first calculate the line-mode wave velocity at a preset frequency based on the line data. This is prior art and will not be elaborated upon in this application embodiment. It should be understood that although the line-mode wave velocity tends to stabilize at a certain frequency, it still undergoes slight changes with frequency. Therefore, in this application embodiment, it is necessary to accurately locate a preset frequency first, and perform subsequent location calculations at that frequency. For example, since a large number of waveform features are lost at excessively high frequencies, and even some wavefronts are lost, 100kHz is preferred as the preset frequency in this application embodiment. It should be understood that the preset frequency mentioned in this application refers to the frequency after performing wavelet transform on the traveling wave. At this preset frequency, the line-mode wave velocity is generally between 1 and 2 × 10⁻⁶. 8 Approximately m / s.

[0063] In this embodiment, a single-end positioning scheme is used for traveling wave detection. For its basic principle, please refer to [link to relevant documentation]. Figure 2Section AB is the line to be tested. A traveling wave detection device is installed at end A. When a ground fault occurs at point C, a fault traveling wave will be generated. One path, L1, will propagate along the direction from C to A, and when it reaches end A at time t1, it will be represented as the first traveling wave front, which is the clearest and easiest to capture. Another path, L2, will propagate along the direction from C to B and will be emitted when it reaches the bus at end B. When it reaches end A at time t2, it will be represented as the second traveling wave front. At the same time, L1 will be reflected by L3 when it reaches end A, and will be reflected again when it reaches the fault point C. When it reaches end A at time t3, it will be represented as the third traveling wave front. Since the location of point C is unknown, even for two seemingly very similar wavelet transform waveforms, the positions of the second and third traveling wave fronts may be interchanged. In addition, the fault traveling wave will also be reflected at other locations in the line, generating many wave fronts at end A. Therefore, accurate detection of the second and third traveling wave fronts is necessary to accurately locate the fault point.

[0064] In this embodiment, all data processing is implemented using a smart terminal. The smart terminal can be a feeder terminal, station terminal, converged terminal, or distribution transformer terminal, or other types of smart terminals; this embodiment does not impose any limitations. Traveling wave detection requires a traveling wave detection device, which can be installed at locations such as T-joints in the transformer substation, line ends, user-side meters, and poles. Those skilled in the art can arrange the traveling wave detection device according to the specific wiring conditions of the transformer substation. In the specific implementation process, the traveling wave-related data detected by the traveling wave detector is sent to the smart terminal, where it is calculated and analyzed.

[0065] After a fault traveling wave is generated, it may oscillate within the line, so it is necessary to calibrate the wavefronts over a certain period of time for identification. In this embodiment, a sliding window length is set, which is the longest time required for the peaks of the second and third traveling wave wavefronts to reach the traveling wave detection device. Specifically, it can be calculated according to the following formula:

[0066]

[0067] In the formula, v is the linear mode wave velocity, and L is the line length. The sliding window is denoted by α, which is a correction parameter to ensure full coverage of the complete wavefront shape. Specifically, it represents the time required for the third traveling wave wavefront to reach the traveling wave detection device when the fault point occurs near B, and also the time required for the second traveling wave wavefront to reach the traveling wave detection device when the fault point occurs near A. This sliding window allows for the accurate determination of the data range requiring traveling wave wavefront identification.

[0068] In this embodiment, when the traveling wave detection device detects a fault traveling wave, it converts the fault traveling wave into three modes (0, 1, and 2) through phase mode transformation. The zero-mode wave velocity is greatly affected by frequency and is difficult to calculate accurately, so it is discarded. The remaining two modes, namely the line mode traveling wave, are used for corresponding identification and calculation. The specific phase mode transformation method can be implemented using a transformation matrix, which is a mature existing technology, and this embodiment does not impose further limitations. For example, fault traveling wave detection can be performed using voltage detection or current detection; in this embodiment, voltage detection is preferred, and the corresponding traveling wave data is also voltage data along a time sequence.

[0069] In this embodiment, the initial wavefront, i.e., the first traveling wavefront, needs to be detected first for subsequent wavefront identification. This can be the first traveling wavefront in the linear model with an amplitude greater than a preset value, or the wavefront corresponding to the maximum value; generally, the detection results are consistent. It should be understood that wavefront identification can be performed by selecting from waveforms corresponding to each frequency after wavelet transform. After selecting the first traveling wavefront, the selected traveling wave data can be determined based on its position. It should be understood that the selected traveling wave data shown in this embodiment is generally divided into sliding window length ranges starting from the initial abrupt change point of the first traveling wavefront. Since the second and first traveling wavefronts are generated simultaneously at the fault point, they have a strong correlation. Similarly, the third traveling wavefront is generated by two reflections of the first traveling wavefront, so they also have a strong correlation. Furthermore, due to differences in the number of reflections and propagation distance, there will be certain differences between the second and third traveling wavefronts. Based on the above principles, the second and third traveling wave waveforms can be identified by capturing their corresponding features using the first traveling wave waveform as a reference. To capture latent features, this application employs a convolutional neural network model for identification. After identifying the second and third traveling wave waveforms, the fault location can be calculated based on the peak times of the first, second, and third traveling wave waveforms. This application, through the above technical means, utilizes the powerful capabilities of convolutional neural networks to extract and analyze features from traveling wave signals, fully considering the correlation between different waveforms and effectively capturing the complex relationships between them. This allows for accurate detection of the second and third traveling wave waveforms using the initial waveform.

[0070] In one possible implementation, the training process of the convolutional neural network model includes:

[0071] Acquire line mode traveling wave sample data, wherein the line mode traveling wave sample data is line mode traveling wave data containing a first traveling wave front, a second traveling wave front, and a third traveling wave front;

[0072] The line-mode traveling wave sample data is decomposed into first frequency component data at multiple frequencies using wavelet transform;

[0073] Sample data of the first row wavefronts are extracted from multiple first frequency component data to form multiple first row vectors corresponding to different frequencies; sample data of the second row wavefronts are extracted from multiple first frequency component data to form multiple second row vectors corresponding to different frequencies; sample data of the third row wavefronts are extracted from multiple first frequency component data to form multiple third row vectors corresponding to different frequencies.

[0074] The first row vector and the second row vector in the same set of line mode traveling wave sample data are arranged vertically alternately according to frequency to form a first feature matrix, and the first row vector and the third row vector in the same set of line mode traveling wave sample data are arranged vertically alternately according to frequency to form a second feature matrix;

[0075] Assign a second wavefront label to the first feature matrix, and assign a third wavefront label to the second feature matrix;

[0076] The convolutional neural network model is formed by training a convolutional neural network using the first feature matrix or the second feature matrix as input data and the corresponding labels as output data.

[0077] In implementing this application embodiment, it is necessary to first obtain line model traveling wave sample data under different state scenarios. This data can be obtained through simulation calculations using simulation software such as PSCAD, or through laboratory and experimental environments. This application embodiment does not impose any limitations. In this application embodiment, wavelet transform is used to decompose multiple frequency component data to capture wavefront features at different frequencies. For example, seven frequency component data corresponding to seven frequencies are decomposed: 10kHz, 25kHz, 50kHz, 100kHz, 200kHz, 300kHz, and 500kHz. At this time, different first frequency component data are sampled to form a first row vector. The second row vector and the third row vector Where i is the i-th frequency and j is the j-th sampling point; at this time, the first feature matrix is ​​generated. In order for the convolutional neural network to better capture the relevant features of different wavefronts at the same frequency and the relevant features of different wavefronts at adjacent frequencies, in this embodiment of the application, a method of alternating vertical arrangement according to the frequency order is adopted to form the feature matrix. For example, the first feature matrix is ​​generated by the first row vector and the second row vector, and the generation result is as follows:

[0078]

[0079] Similarly, the second feature matrix is ​​generated using the first row vector and the third row vector, and the generation result is as follows:

[0080]

[0081] After data processing is complete, labels need to be assigned to the feature matrices of different sample data. The second wavehead label can be assigned a value of (1,0), and the third wavehead label can be assigned a value of (0,1). At this point, a convolutional neural network model can be trained using the feature matrices.

[0082] In one possible implementation, the convolutional neural network includes a first convolutional layer, a second convolutional layer, and a fully connected layer connected in sequence.

[0083] Training a convolutional neural network includes:

[0084] The input data is input into the first convolutional layer, and the input data is convolved by a 2*2 convolutional kernel in the first convolutional layer to generate the first convolutional data; the horizontal stride in the first convolutional layer is 1, and the vertical stride is 2.

[0085] The first convolutional data is input into the second convolutional layer, and the first convolutional data is processed by a 2*2 convolutional kernel in the second convolutional layer to generate the second convolutional data; the horizontal stride in the second convolutional layer is 1, and the vertical stride is 1.

[0086] The second convolutional data is input into the fully connected layer, and the parameters of the fully connected layer are adjusted according to the difference between the output of the fully connected layer and the corresponding label.

[0087] When implementing the embodiments of this application, please refer to Figure 3 This document illustrates the specific feature extraction process of an embodiment of this application. In the first convolutional layer, a 2x2 convolution kernel is used for convolution processing, with a horizontal stride of 1 and a vertical stride of 2. Due to the distribution characteristics of the first and second feature matrices, a vertical stride of 2 allows for the analysis of the correlation features between the first, second, and third wavefronts at the same frequency. For the second convolutional layer, since each row in the data matrix after the first convolutional layer represents a frequency, a convolution process with both a vertical and horizontal stride of 1 can be used to analyze the correlation between wavefronts at adjacent frequencies, effectively improving feature extraction efficiency. In this embodiment, the parameter adjustment of the fully connected layer can be achieved through a loss function, which is a very mature technology in the prior art, and this embodiment does not impose further limitations.

[0088] In one possible implementation, the sampling of the first row vector, the second row vector, and the third row vector includes:

[0089] The first traveling wave front in the first frequency component data is divided into a preset number of intervals from the initial abrupt change point to the wave crest, and each interval node is sampled as a sampling point and then arranged horizontally to form the first row vector of the corresponding frequency.

[0090] The second row wavefront in the first frequency component data is divided into an average number of intervals from the initial abrupt change point to the wave crest, and each interval node is sampled as a sampling point and then arranged horizontally to form the second row vector of the corresponding frequency.

[0091] The third row wavefront in the first frequency component data is divided into a predetermined number of intervals from the initial abrupt change point to the wave crest, and each interval node is sampled as a sampling point and then arranged horizontally to form the third row vector corresponding to the frequency.

[0092] In the implementation of this application embodiment, the sampling process for the first row vector, the second row vector, and the third row vector is similar. This application embodiment uses the first row vector as an example for explanation. When dividing the sampling points, in order to represent the similarity of these three different wavefronts at similar positions, this application embodiment divides the range from the initial abrupt change point of the wave to the wave crest into the same number of intervals, and uses the interval nodes as sampling points. It should be understood that the interval nodes are the endpoints of all intervals, that is, they include the wave crest and the initial abrupt change point. Generally, the preset number can be set to 10, that is, generating 11 sampling points, making the first row vector, the second row vector, and the third row vector 11-dimensional vectors.

[0093] In one possible implementation, identifying the second and third traveling wavefronts from the selected traveling wave data using a convolutional neural network model, based on the first traveling wavefront, includes:

[0094] The selected traveling wave data is decomposed into second frequency component data at multiple frequencies using wavelet transform;

[0095] Sample data of the first traveling wavefront are extracted from multiple second frequency component data to form multiple reference row vectors corresponding to different frequencies; sample data of wavefronts other than the first traveling wavefront are extracted from multiple second frequency component data to form multiple detection row vectors corresponding to different frequencies.

[0096] The reference row vector and the row vector to be detected are arranged vertically alternately according to frequency to form the feature matrix to be detected;

[0097] The feature matrix to be detected is input into the convolutional neural network model, and the result output by the convolutional neural network model is received to determine whether the feature matrix to be detected corresponds to the second row wavefront, the third row wavefront, or other wavefronts.

[0098] In the implementation of this application embodiment, the recognition process and the model training process are quite similar. The only difference is that each time the feature matrix is ​​constructed, it is based on the reference row vector corresponding to the first wavefront and the row vector of the wavefront to be detected. When the feature matrix is ​​input into the trained convolutional neural network model, it can effectively classify the second wavefront, the third wavefront, and other wavefronts. If the output of the convolutional neural network model matches the label of the second wavefront, then the wavefront is a second wavefront; if the output of the convolutional neural network model matches the label of the third wavefront, then the wavefront is a third wavefront; if the output of the convolutional neural network model matches the labels of both the second and third wavefronts, then the wavefront is another type of wavefront.

[0099] In one possible implementation, sampling of the baseline row vector and the row vector to be detected includes:

[0100] The range from the initial abrupt change point to the peak of the first traveling wave in the second frequency component data is divided into a preset number of intervals, and each interval node is sampled as a sampling point and then arranged horizontally to form the reference row vector of the corresponding frequency.

[0101] The wavefronts of the second frequency component data that are not the first traveling wavefront are divided into a predetermined number of intervals from the initial abrupt change point to the wave crest. Each interval node is then sampled and arranged horizontally to form the corresponding frequency of the row vector to be detected.

[0102] When implementing the embodiments of this application, the specific sampling process is the same as the sampling process in the training process described above. It should be understood that the preset number of intervals corresponding to the sampling process in the embodiments of this application should remain consistent.

[0103] In one possible implementation, determining whether the feature matrix to be detected corresponds to a second wavefront, a third wavefront, or another wavefront includes:

[0104] When the output of the convolutional neural network model corresponds to the second row wavefront, it is determined that the feature matrix to be detected corresponds to the second row wavefront.

[0105] When the output of the convolutional neural network model corresponds to the third row wave head, it is determined that the feature matrix to be detected corresponds to the third row wave head.

[0106] When the output of the convolutional neural network model does not correspond to the second row wavefront or the third row wavefront, it is determined that the feature matrix to be detected corresponds to other wavefronts.

[0107] In one possible implementation, the calculation of the fault location includes:

[0108] The difference between the peak time of the third wavefront and the peak time of the first wavefront is calculated as the first duration, and the difference between the peak time of the second wavefront and the peak time of the first wavefront is calculated as the second duration.

[0109] Calculate the first duration by multiplying it by the line mode wave velocity and then dividing by 2 to generate the first ranging distance; calculate the second duration by multiplying it by the line mode wave velocity and then dividing by 2 to generate the second ranging distance;

[0110] When the difference between the sum of the first ranging distance and the second ranging distance and the length of the line to be detected is less than or equal to a preset value, the first ranging distance is taken as the distance from the fault point to the traveling wave detection device.

[0111] When the difference between the sum of the first ranging distance and the second ranging distance and the length of the line to be tested is greater than a preset value, the third ranging distance is taken as the distance from the fault point to the traveling wave detection device; the third ranging distance is the line length minus the second ranging distance plus the first ranging distance and then divided by 2.

[0112] When implementing the embodiments of this application, please refer to Figure 4 The figure shows the waveform of a line-mode traveling wave at 100kHz, where the horizontal axis represents time in μs and the vertical axis represents voltage in V. The figure shows the position of the first traveling wavefront, with its peak corresponding to time t1; the peak of the second traveling wavefront corresponds to time t2, and the peak of the third traveling wavefront corresponds to time t3. The accurate fault location can be calculated using the differences between these times and the line-mode wave velocity at 100kHz. It should be understood that this embodiment identifies different wavefronts at multiple frequencies and performs ranging calculations at a preset frequency, thus ensuring the accuracy of the line-mode wave velocity and effectively improving the ranging accuracy.

[0113] In this embodiment, the first ranging distance is the distance from the fault point to the traveling wave detection device calculated using a third traveling wave wavefront, and the second ranging distance is the distance from the fault point to the end of the line furthest from the traveling wave detection device calculated using a second traveling wave wavefront. If the calculation results are completely accurate, the sum of these two should equal the line length. If the sum differs significantly from the line length, it indicates that the detection process may be affected by some factors. In this case, a third ranging distance is generated by averaging the results for ranging prediction.

[0114] In one possible implementation, the number of frequencies after wavelet transform decomposition is 6 to 10.

[0115] In one possible implementation, the preset frequency is 100kHz.

[0116] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0117] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices or units, or may be electrical, mechanical or other forms of connection.

[0118] The units described as separate components may or may not be physically separate. As will be apparent to those skilled in the art, the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0119] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0120] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or grid device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0121] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for locating single-phase ground fault traveling wave based on a smart terminal, characterized in that, include: Obtain the line data of the line to be tested, and calculate the line mode wave velocity at a preset frequency based on the line data; A traveling wave detection device is installed at one end of the line to be tested, and the sliding window length is calculated based on the line mode wave velocity and the line length of the line to be tested. The sliding window length is the longest time required for the peaks of the second and third traveling wave waves to reach the traveling wave detection device after the first traveling wave wavehead reaches the traveling wave detection device. The first traveling wave wavehead is the wavehead of the fault traveling wave that reaches the traveling wave detection device. The second traveling wave wavehead is the wavehead of the fault traveling wave that reaches the other end of the line to be tested and is reflected before reaching the traveling wave detection device. The third traveling wave wavehead is the wavehead of the fault traveling wave that reaches the traveling wave detection device, is reflected again at the fault point, and then reaches the traveling wave detection device. When the traveling wave detection device detects a fault traveling wave, it performs a phase mode transformation on the fault traveling wave to obtain a line mode traveling wave; The first traveling wave head in the linear model traveling wave is taken as the first traveling wave head, and the linear model traveling waves within the sliding window length range from the first traveling wave head are recorded as selected traveling wave data. Using the first traveling wave wavehead as a reference, a convolutional neural network model is used to identify the second and third traveling wave waveheads from the selected traveling wave data. The peak times of the first, second, and third traveling wave waveheads at the preset frequency are calculated, and the location of the fault point is further calculated. The training process of the convolutional neural network model includes: Acquire line mode traveling wave sample data, wherein the line mode traveling wave sample data is line mode traveling wave data containing a first traveling wave front, a second traveling wave front, and a third traveling wave front; The line-mode traveling wave sample data is decomposed into first frequency component data at multiple frequencies using wavelet transform; Sample data of the first row wavefronts are extracted from multiple first frequency component data to form multiple first row vectors corresponding to different frequencies; sample data of the second row wavefronts are extracted from multiple first frequency component data to form multiple second row vectors corresponding to different frequencies; sample data of the third row wavefronts are extracted from multiple first frequency component data to form multiple third row vectors corresponding to different frequencies. The first row vector and the second row vector in the same set of line mode traveling wave sample data are arranged vertically alternately according to frequency to form a first feature matrix, and the first row vector and the third row vector in the same set of line mode traveling wave sample data are arranged vertically alternately according to frequency to form a second feature matrix; Assign a second wavefront label to the first feature matrix, and assign a third wavefront label to the second feature matrix; The convolutional neural network model is formed by training a convolutional neural network using the first feature matrix or the second feature matrix as input data and the corresponding labels as output data. Identifying the second and third traveling wavefronts from the selected traveling wave data using a convolutional neural network model, based on the first traveling wavefront, includes: The selected traveling wave data is decomposed into second frequency component data at multiple frequencies using wavelet transform; Sample data of the first traveling wavefront are extracted from multiple second frequency component data to form multiple reference row vectors corresponding to different frequencies; sample data of wavefronts other than the first traveling wavefront are extracted from multiple second frequency component data to form multiple detection row vectors corresponding to different frequencies. The reference row vector and the row vector to be detected are arranged vertically alternately according to frequency to form the feature matrix to be detected; The feature matrix to be detected is input into the convolutional neural network model, and the result output by the convolutional neural network model is received to determine whether the feature matrix to be detected corresponds to the second row wavefront, the third row wavefront, or other wavefronts.

2. The method for locating a single-phase ground fault traveling wave based on a smart terminal according to claim 1, characterized in that, The convolutional neural network includes a first convolutional layer, a second convolutional layer, and a fully connected layer connected in sequence. Training a convolutional neural network includes: The input data is input into the first convolutional layer, and the input data is convolved by a 2*2 convolutional kernel in the first convolutional layer to generate the first convolutional data; the horizontal stride in the first convolutional layer is 1, and the vertical stride is 2. The first convolutional data is input into the second convolutional layer, and the first convolutional data is processed by a 2*2 convolutional kernel in the second convolutional layer to generate the second convolutional data; the horizontal stride in the second convolutional layer is 1, and the vertical stride is 1. The second convolutional data is input into the fully connected layer, and the parameters of the fully connected layer are adjusted according to the difference between the output of the fully connected layer and the corresponding label.

3. The method for locating a single-phase ground fault traveling wave based on a smart terminal according to claim 1, characterized in that, The sampling of the first, second, and third row vectors includes: The first traveling wave front in the first frequency component data is divided into a preset number of intervals from the initial abrupt change point to the wave crest, and each interval node is sampled as a sampling point and then arranged horizontally to form the first row vector of the corresponding frequency. The second row wavefront in the first frequency component data is divided into an average number of intervals from the initial abrupt change point to the wave crest, and each interval node is sampled as a sampling point and then arranged horizontally to form the second row vector of the corresponding frequency. The third row wavefront in the first frequency component data is divided into a predetermined number of intervals from the initial abrupt change point to the wave crest, and each interval node is sampled as a sampling point and then arranged horizontally to form the third row vector corresponding to the frequency.

4. The method for locating a single-phase ground fault traveling wave based on a smart terminal according to claim 1, characterized in that, The sampling of the baseline row vector and the row vector to be detected includes: The range from the initial abrupt change point to the peak of the first traveling wave in the second frequency component data is divided into a preset number of intervals, and each interval node is sampled as a sampling point and then arranged horizontally to form the reference row vector of the corresponding frequency. The wavefronts of the second frequency component data that are not the first traveling wavefront are divided into a predetermined number of intervals from the initial abrupt change point to the wave crest. Each interval node is then sampled and arranged horizontally to form the corresponding frequency of the row vector to be detected.

5. The method for locating a single-phase ground fault traveling wave based on a smart terminal according to claim 1, characterized in that, Determining whether the feature matrix to be detected corresponds to the second wavefront, the third wavefront, or other wavefronts includes: When the output of the convolutional neural network model corresponds to the second row wavefront, it is determined that the feature matrix to be detected corresponds to the second row wavefront. When the output of the convolutional neural network model corresponds to the third row wave head, it is determined that the feature matrix to be detected corresponds to the third row wave head. When the output of the convolutional neural network model does not correspond to the second row wavefront or the third row wavefront, it is determined that the feature matrix to be detected corresponds to other wavefronts.

6. The method for locating a single-phase ground fault traveling wave based on a smart terminal according to claim 1, characterized in that, The calculation of the fault location includes: The difference between the peak time of the third wavefront and the peak time of the first wavefront is calculated as the first duration, and the difference between the peak time of the second wavefront and the peak time of the first wavefront is calculated as the second duration. Calculate the first duration by multiplying it by the line mode wave velocity and then dividing by 2 to generate the first ranging distance; calculate the second duration by multiplying it by the line mode wave velocity and then dividing by 2 to generate the second ranging distance; When the difference between the sum of the first ranging distance and the second ranging distance and the length of the line to be detected is less than or equal to a preset value, the first ranging distance is taken as the distance from the fault point to the traveling wave detection device. When the difference between the sum of the first ranging distance and the second ranging distance and the length of the line to be tested is greater than a preset value, the third ranging distance is taken as the distance from the fault point to the traveling wave detection device; the third ranging distance is the line length minus the second ranging distance plus the first ranging distance and then divided by 2.

7. The method for locating a single-phase ground fault traveling wave based on a smart terminal according to claim 1, characterized in that, The number of frequencies after wavelet transform decomposition is 6 to 10.

8. The method for locating a single-phase ground fault traveling wave based on a smart terminal according to claim 1, characterized in that, The preset frequency is 100kHz.

Images