CN112180210B - Power distribution network single-phase earth fault line selection method and system - Google Patents

Power distribution network single-phase earth fault line selection method and system Download PDF

Info

Publication number
CN112180210B
CN112180210B CN202011018649.8A CN202011018649A CN112180210B CN 112180210 B CN112180210 B CN 112180210B CN 202011018649 A CN202011018649 A CN 202011018649A CN 112180210 B CN112180210 B CN 112180210B
Authority
CN
China
Prior art keywords
fault
line
sequence current
zero
distribution network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
CN202011018649.8A
Other languages
Chinese (zh)
Other versions
CN112180210A (en
Inventor
殷浩然
苗世洪
牛荣泽
孙芊
王磊
彭磊
徐恒博
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huazhong University of Science and Technology
State Grid Corp of China SGCC
State Grid Henan Electric Power Co Ltd
Electric Power Research Institute of State Grid Henan Electric Power Co Ltd
Original Assignee
Huazhong University of Science and Technology
State Grid Corp of China SGCC
State Grid Henan Electric Power Co Ltd
Electric Power Research Institute of State Grid Henan Electric Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huazhong University of Science and Technology, State Grid Corp of China SGCC, State Grid Henan Electric Power Co Ltd, Electric Power Research Institute of State Grid Henan Electric Power Co Ltd filed Critical Huazhong University of Science and Technology
Priority to CN202011018649.8A priority Critical patent/CN112180210B/en
Publication of CN112180210A publication Critical patent/CN112180210A/en
Application granted granted Critical
Publication of CN112180210B publication Critical patent/CN112180210B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/08—Locating faults in cables, transmission lines, or networks
    • G01R31/081—Locating faults in cables, transmission lines, or networks according to type of conductors
    • G01R31/086—Locating faults in cables, transmission lines, or networks according to type of conductors in power transmission or distribution networks, i.e. with interconnected conductors
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/08—Locating faults in cables, transmission lines, or networks
    • G01R31/088—Aspects of digital computing
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/50—Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
    • G01R31/52—Testing for short-circuits, leakage current or ground faults
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/50—Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
    • G01R31/58—Testing of lines, cables or conductors
    • G01R31/60—Identification of wires in a multicore cable
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
    • Y04S10/52—Outage or fault management, e.g. fault detection or location

Abstract

The invention discloses a method and a system for single-phase earth fault line selection of a power distribution network, and belongs to the field of power system protection. The method comprises the steps of firstly obtaining zero-sequence current time-frequency information by utilizing S transformation, calculating line fault information correlation degree based on full-frequency-band information of zero-sequence current of each line, then building an S transformation correlation degree image SCF for improving identifiability and anti-interference of fault characteristics, building a convolutional neural network deep learning model S-CNN containing an SCF layer on the basis, training structural parameters and hyper-parameters of the convolutional neural network deep learning model S-CNN step by utilizing fault data generated by a Simulink simulation model, and finally extracting deep characteristics of fault zero-sequence current of a power distribution network through the S-CNN to realize fault line selection. The line selection effect is not influenced by the system running condition and the fault condition, and has stronger robustness under the conditions of strong noise interference and asynchronous sampling.

Description

Power distribution network single-phase earth fault line selection method and system
Technical Field
The invention belongs to the field of power system protection, and particularly relates to a single-phase earth fault line selection method and system for a power distribution network.
Background
The distribution network in China generally adopts a small current grounding mode, namely a neutral point is not grounded or is grounded through an arc suppression coil, and single-phase grounding faults account for more than 80% of the total number of faults. When a single-phase earth fault occurs in a power distribution network, the line voltage still keeps symmetrical, and the fault current is small, so that the regulation specifies that the system can continue to operate for 1-2 hours, but the fault is not removed for a long time, which may cause the expansion of a fault range and the deterioration of fault properties, and bring about great potential safety hazards. Therefore, the faulty line must be determined and cut out in time.
In the existing research, the fault line selection principle of the power distribution network is mainly divided into two types: fault line selection based on steady state signals and fault line selection based on transient state signals. The existing fault line selection method mostly takes zero-sequence current or specific frequency band information thereof as fault characteristics, has poor noise resistance, is difficult to cope with complex scenes of operating environment and fault conditions, and limits the application of the fault line selection method in practical engineering. Therefore, in order to realize a fault line selection method with strong noise immunity and generalization ability, a fault feature extraction method and a line selection criterion need to be further improved.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a power distribution network single-phase earth fault line selection method and a power distribution network single-phase earth fault line selection system, and aims to solve the technical problems that the existing fault line selection method is poor in noise resistance and difficult to cope with complex scenes of operating environments and fault conditions.
In order to achieve the purpose, the invention provides a new method for selecting the single-phase earth fault line of the power distribution network on the one hand, which sequentially comprises the following steps:
step 1: establishing a simulation working condition of the single-phase earth fault of the power distribution network, setting a voltage of the power distribution network, a node load, a line fault position, a transition resistance and a fault phase angle value range by considering the actual running condition of the power distribution network, generating a fault scene set covering the parameter range, and collecting fault zero-sequence current of each line under all fault scenes to form a deep learning training data set;
step 2: obtaining full-band time-frequency information of zero-sequence current of each line by utilizing S transformation, calculating Euclidean distance between the full-band time-frequency information of each line to serve as the correlation degree of fault zero-sequence current of each line, and filling the correlation degrees of the fault zero-sequence current of all lines into a two-dimensional matrix to form a fault zero-sequence current correlation degree graph SCF;
and step 3: and (2) constructing a convolutional neural network model S-CNN containing the SCF, training by using the deep learning training data set generated in the step (1), adjusting the structural parameters and the hyper-parameters of the S-CNN according to the training result, and acquiring the trained S-CNN model with fault line selection capability.
Further, in step 1, the fault scene set is a parameter history table formed by values of the grid voltage, the node load, the line fault position, the transition resistance and the fault phase angle within a certain range according to the actual running condition of the power distribution network. And arranging and combining the parameter values in the parameter history table to form a large number of simulation scenes, namely a fault scene set.
The deep learning training data set is used for operating the single-phase fault simulation working condition of the power distribution network under all fault scenes contained in the fault scene set, collecting zero sequence current of each line and recording corresponding fault lines; and taking the zero sequence current as a data main body and the corresponding fault line as a data label to jointly form a deep learning training data set.
Further, in step 2, S is transformed into a time-series signal processing method, which is discrete form:
Figure BDA0002699947280000021
in the formula，x[i]Is a discrete time sequence signal, i-0, 1, N-1;
Figure BDA0002699947280000031
fsfor discrete signal sampling frequency, N is the number of sampling points,
Figure BDA0002699947280000032
representing a time point, k is an integer and has a value range of 1,2, ·, N;
Figure BDA0002699947280000033
represents the frequency point, N is an integer and has a value range of 0,1, N/2 in consideration of the sampling theorem. According to the formula, the full-band time-frequency information of the transient zero-sequence current of each line in the fault period of the power distribution network can be obtained by utilizing S transformation.
Further, the fault zero-sequence current correlation graph SCF is a two-dimensional matrix formed by arranging the zero-sequence current correlations of each line according to the above formula. For example, line 1 and line 2 fault zero sequence current is transformed by S to obtain corresponding full-band time frequency information, and then the Euclidean distance S between two lines of time frequency information is calculated12As SrelThe elements in the matrix.
Figure BDA0002699947280000034
Wherein s isijAnd the Euclidean distance between the full-frequency-band time-frequency information of the fault zero-sequence current of the line i and the line j is represented, i is more than or equal to 1 and less than or equal to y, j is more than or equal to 1 and less than or equal to y, and y is the total number of the lines.
Furthermore, the S-CNN model is composed of three links of an SCF construction layer, a deep feature extraction layer and a line selection output layer. S-CNN uses sequence data I of zero sequence current of each line after fault0(x is y, x is the number of data points of zero sequence current of each line, and y is the total number of the lines) as input, and a two-dimensional matrix S is generated through the SCF construction layerrel(y × y). Then, the layer pair S is extracted in the deep layerrelThe fault line selection features are extracted, and the deep feature extraction layer is composed of several convolution layers and pooling layers alternatelyComprises the following steps:
by a convolution kernel (k)1×k1，k1The order of the first layer convolution kernel matrix) and the sample matrix, and obtaining a characteristic surface (m multiplied by m, m is the order of the special surface matrix) through an activation function, wherein the formula is as follows:
Figure BDA0002699947280000035
wherein l represents the sequence of the network layers,
Figure BDA0002699947280000036
the characteristic surface representing the output of the l-th layer,
Figure BDA0002699947280000037
the feature plane representing the output of layer l-1, as the input of layer l,
Figure BDA0002699947280000041
represents the l-th layer of the convolution kernel matrix,
Figure BDA0002699947280000042
represents the sliding step convolution operation of the convolution kernel to the characteristic surface, and f (-) represents the activation function.
Then pool the kernel (k)2×k2，k2Order of the second layer convolution kernel matrix) to obtain a quadratic feature plane (s × s), which is expressed by a formula:
Figure BDA0002699947280000043
the above equation represents a mean pooling process, which is next to the convolutional layer, and the pooled layer performs a deresolution sampling of the feature plane output by the convolutional layer.
Finally, the final characteristic surface is integrated by a full-connection network in the line selection output layer to form a line selection vector OL(yx 1) where each element represents the possibility of y-line failure, respectively, i.e.The larger the value is, the higher the possibility of the line fault is, and the line with the highest possibility is selected by the Softmax layer as the line selection result.
The S-CNN structural parameters refer to the structure of a deep feature extraction layer and comprise the number of convolution and pooling layers, the number of feature planes of each layer and the order of a convolution kernel matrix and a pooling kernel matrix.
And the S-CNN super-parameter is the learning efficiency and batch training number of the model, and is adjusted according to the training test result.
In another aspect, the present invention provides a single-phase earth fault line selection system for a power distribution network, including:
the training data set acquisition module is used for establishing a single-phase earth fault simulation working condition of the power distribution network, setting a power grid voltage, a node load, a line fault position, a transition resistance and a fault phase angle value range, generating a fault scene set covering the parameter range, and acquiring fault zero-sequence current of each line under all scenes to form a deep learning training data set;
the SCF acquisition module is used for acquiring full-band time-frequency information of the fault zero-sequence current of each line by utilizing S conversion, calculating the Euclidean distance between the full-band time-frequency information of each line to serve as the correlation degree of the fault zero-sequence current of each line, and filling the correlation degrees of the fault zero-sequence current of all lines into a two-dimensional matrix to form a fault zero-sequence current correlation degree graph SCF;
the model acquisition module is used for constructing a convolutional neural network model S-CNN containing the SCF, training the S-CNN by using a deep learning training data set, adjusting S-CNN structural parameters and hyper-parameters according to a training result, acquiring a trained S-CNN model with fault line selection capability, and realizing single-phase earth fault line selection of a power grid.
Further, the fault scene set is a parameter history table formed by values of the grid voltage, the node load, the line fault position, the transition resistance and the fault phase angle within a preset range according to the actual operation condition of the power distribution network, and the fault scene set is formed by arranging and combining the parameter values in the parameter history table.
Further, the discrete form of the S transform is:
Figure BDA0002699947280000051
in the formula, x [ i ]]Is a discrete time sequence signal, i-0, 1, N-1;
Figure BDA0002699947280000052
fsfor discrete signal sampling frequency, N is the number of sampling points,
Figure BDA0002699947280000053
representing a time point, k is an integer and has a value range of 1,2, ·, N;
Figure BDA0002699947280000054
and the frequency points are represented, N is an integer and has the value range of 0,1, N/2.
Further, the fault zero-sequence current correlation graph SCF is a two-dimensional matrix formed by arranging fault zero-sequence current correlations of each line, as follows:
Figure BDA0002699947280000055
wherein S isijAnd representing that the fault zero-sequence current of the line i and the fault zero-sequence current of the line j acquire corresponding full-frequency-band time frequency information through S transformation, and then calculating the Euclidean distance between the two lines of line time frequency information, wherein i is more than or equal to 1 and less than or equal to y, j is more than or equal to 1 and less than or equal to y, and y is the total number of lines.
Compared with the prior art, the invention provides a new method for single-phase earth fault line selection of the power distribution network by the technical scheme. Firstly, acquiring zero-sequence current time-frequency information by utilizing S transformation, and calculating the line fault information correlation degree based on the full-frequency-band information of the zero-sequence current of each line; secondly, in order to improve the identifiability and anti-interference of fault characteristics, an S transformation correlation graph SCF is constructed; on the basis, a convolutional neural network deep learning model S-CNN containing an SCF layer is established, and structural parameters and hyper-parameters of the convolutional neural network deep learning model are trained by utilizing fault data; and finally, extracting deep characteristics of the fault zero-sequence current of the power distribution network through S-CNN to realize fault line selection. Compared with the existing line selection method, the line selection method has the advantages that the line selection effect is not influenced by the system operation condition and the fault condition, and the robustness is stronger under the conditions of strong noise interference and asynchronous sampling. The method is not influenced by the system running condition and the fault condition, and has stronger robustness under the conditions of strong noise interference and asynchronous sampling.
Drawings
Fig. 1 is a simulation model diagram of a single-phase earth fault of a power distribution network in an embodiment of the invention.
FIG. 2 is a diagram of the S-CNN fault line selection model structure of the present invention.
FIG. 3 is a diagram of the S-transform correlation matrix construction process according to the present invention.
Fig. 4 is a waveform diagram of a set of zero sequence currents of each line superimposed with white gaussian noise in the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The invention provides a new method for selecting a single-phase earth fault line of a power distribution network, which sequentially comprises the following steps:
step 1: establishing a simulation working condition of the single-phase earth fault of the power distribution network, setting a voltage of the power distribution network, a node load, a line fault position, a transition resistance and a fault phase angle value range by considering the actual running condition of the power distribution network, generating a fault scene set covering the parameter range, and collecting fault zero-sequence current of each line under all fault scenes to form a deep learning training data set;
step 2: obtaining full-band time-frequency information of zero-sequence current of each line by utilizing S transformation, calculating Euclidean distance between the full-band time-frequency information of each line to serve as the correlation degree of fault zero-sequence current of each line, and filling the correlation degrees of the fault zero-sequence current of all lines into a two-dimensional matrix to form a fault zero-sequence current correlation degree graph SCF;
and step 3: and (2) constructing a convolutional neural network model S-CNN containing the SCF, training by using the deep learning training data set generated in the step (1), adjusting the structural parameters and the hyper-parameters of the S-CNN according to the training result, and acquiring the trained S-CNN model with fault line selection capability.
Specifically, in step 1, the fault scene set is a parameter history table formed by taking values of the grid voltage, the node load, the line fault position, the transition resistance and the fault phase angle within a certain range according to the actual running condition of the power distribution network. And arranging and combining the parameter values in the parameter history table to form a large number of simulation scenes, namely a fault scene set.
The deep learning training data set is used for operating the single-phase fault simulation working condition of the power distribution network under all fault scenes contained in the fault scene set, collecting zero sequence current of each line and recording corresponding fault lines; and taking the zero sequence current as a data main body and the corresponding fault line as a data label to jointly form a deep learning training data set.
Specifically, in step 2, S is transformed into a time-series signal processing method whose discrete form is:
Figure BDA0002699947280000071
in the formula, x [ i ]]Is a discrete time sequence signal, i-0, 1, N-1;
Figure BDA0002699947280000072
fsfor discrete signal sampling frequency, N is the number of sampling points,
Figure BDA0002699947280000073
representing a time point, k is an integer and has a value range of 1,2, ·, N;
Figure BDA0002699947280000081
represents the frequency point, N is an integer and has a value range of 0,1, N/2 in consideration of the sampling theorem. According to the formula, the full-band time-frequency information of the transient zero-sequence current of each line in the fault period of the power distribution network can be obtained by utilizing S transformation.
Specifically, the fault zero-sequence current correlation graph SCF is a two-dimensional matrix formed by arranging the zero-sequence current correlations of each line according to the above formula. For example, line 1 and line 2 fault zero sequence current is transformed by S to obtain corresponding full-band time frequency information, and then the Euclidean distance S between two lines of time frequency information is calculated12As SrelThe elements in the matrix.
Figure BDA0002699947280000082
Wherein S isijAnd representing that the fault zero-sequence current of the line i and the fault zero-sequence current of the line j acquire corresponding full-frequency-band time frequency information through S transformation, and then calculating the Euclidean distance between the two lines of line time frequency information, wherein i is more than or equal to 1 and less than or equal to y, j is more than or equal to 1 and less than or equal to y, and y is the total number of lines.
Examples
Step 1: a single-phase earth fault simulation model of the power distribution network shown in FIG. 1 is constructed in a simulink simulation platform, and the types and parameters of lines are shown in Table 1.
TABLE 1
Line parameters Overhead line Cable with a protective layer
Resistance (positive/zero sequence) 0.1250/0.2750 0.2700/2.7000
Inductor (positive/zero sequence) 1.3000/4.6000 0.2550/1.0190
Capacitor (positive sequence/zero sequence) 0.0096/0.0054 0.2800/0.3390
A parameter history table as shown in table 2 is formed. And arranging and combining the parameter values in the parameter history table to form a large number of simulation scenes, namely a fault scene set.
TABLE 2
Figure BDA0002699947280000083
Figure BDA0002699947280000091
And arranging and combining the operation parameters according to the table 2, collecting zero sequence current time sequence data at each line port under all scenes through circulation simulation, setting the sampling rate to 3200Hz, and generating 10800 groups of fault data and corresponding fault labels.
Step 2: the full-band time-frequency information of the zero-sequence current of each line is obtained by using S transformation, the Euclidean distance between the full-band time-frequency information of each line is calculated to be used as the correlation degree of the fault zero-sequence current of each line, and the correlation degrees of the fault zero-sequence currents of all lines are filled into a two-dimensional matrix to form a fault zero-sequence current correlation degree graph SCF, as shown in figure 2.
Step (ii) of3: and constructing an S-CNN model shown in the figure 3, wherein the S-CNN model consists of three links of an SCF construction layer, a deep feature extraction layer and a line selection output layer. S-CNN uses sequence data I of zero sequence current of each line after fault0(x is y, x is the number of data points of zero sequence current of each line, and y is the total number of the lines) as input, and a two-dimensional matrix S is generated through the SCF construction layerrel(y × y). Then, the layer pair S is extracted in the deep layerrelThe fault line selection characteristic that contains is drawed, and deep layer characteristic extraction layer comprises a plurality of convolution layer and pooling layer in turn, specifically is:
by a convolution kernel (k)1×k1，k1The order of the first layer convolution kernel matrix) and the sample matrix, and obtaining a characteristic surface (m multiplied by m, m is the order of the special surface matrix) through an activation function, wherein the formula is as follows:
Figure BDA0002699947280000092
wherein l represents the sequence of the network layers,
Figure BDA0002699947280000093
the characteristic surface representing the output of the l-th layer,
Figure BDA0002699947280000094
the feature plane representing the output of layer l-1, as the input of layer l,
Figure BDA0002699947280000095
represents the l-th layer of the convolution kernel matrix,
Figure BDA0002699947280000096
represents the sliding step convolution operation of the convolution kernel to the characteristic surface, and f (-) represents the activation function.
Then pool the kernel (k)2×k2，k2Order of the second layer convolution kernel matrix) to obtain a quadratic feature plane (s × s), which is expressed by a formula:
Figure BDA0002699947280000097
the above equation represents a mean pooling process, which is next to the convolutional layer, and the pooled layer performs a deresolution sampling of the feature plane output by the convolutional layer.
And (3) training the S-CNN model by using 10800 groups of fault data in the step (2), and adjusting the structural parameters and the hyper-parameters of the model according to the training result. Finally, the structural parameters of the model are shown in table 3, the learning efficiency is set to 1, and the number of batch trainings is 20.
TABLE 3
Serial number Structure of the product Parameter(s)
1 Inputting samples 64×10
2 SrelTwo-dimensional matrix 10×10
3 Convolutional layer 1 And (3) convolution kernel: 5 × 5, sliding step: 1
4 Characteristic surface 1 6 groups, 6X 6
5 Pooling layer 1 And (3) down-sampling kernel: 2 × 2, sampling step size: 2
6 Convolutional layer 2 And (3) convolution kernel: 2 × 2, sliding step: 1
7 Characteristic surface 2 12 groups, 2X 2
8 Pooling layer 2 And (3) down-sampling kernel: 1 × 1, sampling step size: 1
9 Full connection layer Line selection vector: 1X 10
10 Softmax layer /
And 4, step 4: changing the simulation model parameters shown in FIG. 1, generating 400 groups of fault data different from the training process under different fault positions, fault phase angles and transition resistances, and forming a model test set for checking the line selection effect of the trained S-CNN model.
The test process is illustrated by taking the test of 1 group of data as an example: inputting the group of fault zero-sequence currents into the trained S-CNN model, and obtaining line selection vectors output by the S-CNN, wherein elements of the line selection vectors represent the probability that the S-CNN judges fault data as faults of each line, namely the S-CNN considers that the line 2 corresponding to the group of fault data has single-phase earth faults and is consistent with the actual situation.
And (3) testing 400 groups of fault data according to the steps, wherein the line selection accuracy is 100%, namely the trained S-CNN model can realize hundred-percent correct line selection on simulation data under different fault positions, fault phase angles and transition resistances.
TABLE 4
Figure BDA0002699947280000101
Figure BDA0002699947280000111
And 5: in order to test the anti-interference capability of the S-CNN line selection model, Gaussian white noise with different signal-to-noise ratios is superimposed in the 400 groups of test data to form a noise interference data set. Taking a group of zero sequence current data when the single-phase ground fault occurs in the outgoing line L2 as an example, the waveform of the zero sequence current after 10dB gaussian white noise is superimposed is shown in fig. 4.
The trained S-CNN is tested by utilizing fault data under different signal-to-noise ratios, and the accuracy rate of line selection is shown in Table 5. The snr is calculated as snr ═ 10lg (Ps/Pn), where Ps is the source signal and Pn is the noise signal, so a smaller snr indicates a larger noise. Signal to noise ratio non in table 5 indicates that no noise is added. As can be seen from Table 5, the S-CNN line selection methods have strong anti-noise interference capability, and when the signal-to-noise ratio reaches 10dB, the line selection accuracy can still reach 98.5%.
TABLE 5
Signal to noise ratio/dB Line selection accuracy/%)
non 100
30 100
25 100
20 100
15 100
10 98.5
Step 6: in order to test the robustness of the S-CNN line selection model when the zero-sequence current sampling of each line fault is asynchronous, for the 400 groups of test data, on the basis of adding 20dB Gaussian white noise, a sampling time difference Td is set for the zero-sequence currents of the outgoing lines L1-5 and L6-10, a data set with noise sampling different steps is generated, and the sampling asynchronous scene existing in each outgoing line is simulated. As can be seen from Table 6, the S-CNN line selection model has strong robustness when the zero-sequence current sampling of each line fault is asynchronous.
TABLE 6
Sampling time difference Td/ms Line selection accuracy/%)
0.00 100
1.25 100
2.50 100
5.00 100
The new method for selecting the single-phase earth fault line of the power distribution network provided by the invention can be realized according to the steps, and the steps show that the line selection effect of the S-CNN model provided by the invention is not influenced by the system running condition and the fault condition, and has stronger robustness under the conditions of strong noise interference and asynchronous sampling.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (8)

1. A single-phase earth fault line selection method for a power distribution network is characterized by comprising the following steps:
step 1: establishing a simulation working condition of a single-phase earth fault of a power distribution network, setting a value range of a power grid voltage, a node load, a line fault position, a transition resistance and a fault phase angle, generating a fault scene set covering the parameter range, and collecting fault zero-sequence current of each line under all scenes to form a deep learning training data set;
step 2: acquiring full-band time-frequency information of each line fault zero-sequence current by using S transformation, calculating Euclidean distance between the full-band time-frequency information of each line as the correlation of each line fault zero-sequence current, and filling the correlation of all line fault zero-sequence currents into a two-dimensional matrix to form a fault zero-sequence current correlation graph SCF;
and step 3: and (2) constructing a convolutional neural network model S-CNN containing the SCF, training the S-CNN by using the deep learning training data set generated in the step (1), adjusting the structural parameters and the hyper-parameters of the S-CNN according to the training result, acquiring the trained S-CNN model with fault line selection capability, and realizing single-phase ground fault line selection of the power grid.
2. The method according to claim 1, wherein the fault scenario set is a parameter history table formed by taking values of grid voltage, node load, line fault location, transition resistance and fault phase angle within a preset range according to the actual operation condition of the power distribution network, and the fault scenario set is formed by arranging and combining the parameter values in the parameter history table.
3. The method of claim 1, wherein the S-transform is a time sequence signal processing method, and the discrete form of the method is as follows:
Figure FDA0003087376040000011
Figure FDA0003087376040000012
in the formula, x [ i ]]Is a discrete time sequence signal, i-0, 1, N-1; f is fsDiscrete signal sampling frequency, N is the number of sampling points;
Figure FDA0003087376040000021
represents the time point, k ═ 1,2, ·, N;
Figure FDA0003087376040000022
represents a frequency point, N ═ 0,1, ·, N/2; t is an integer, m-0, 1, N-1.
4. The method according to claim 1, wherein the fault zero sequence current correlation graph SCF is a two-dimensional matrix formed by arranging fault zero sequence current correlations of each line, as follows:
Figure FDA0003087376040000023
wherein s isijAnd the Euclidean distance between the full-frequency-band time-frequency information of the fault zero-sequence current of the line i and the line j is represented, i is more than or equal to 1 and less than or equal to y, j is more than or equal to 1 and less than or equal to y, and y is the total number of the lines.
5. The utility model provides a distribution network single-phase earth fault route selection system which characterized in that includes:
the training data set acquisition module is used for establishing a single-phase earth fault simulation working condition of the power distribution network, setting a power grid voltage, a node load, a line fault position, a transition resistance and a fault phase angle value range, generating a fault scene set covering the parameter range, and acquiring fault zero-sequence current of each line under all scenes to form a deep learning training data set;
the SCF acquisition module is used for acquiring full-band time-frequency information of the fault zero-sequence current of each line by utilizing S conversion, calculating the Euclidean distance between the full-band time-frequency information of each line to serve as the correlation degree of the fault zero-sequence current of each line, and filling the correlation degrees of the fault zero-sequence current of all lines into a two-dimensional matrix to form a fault zero-sequence current correlation degree graph SCF;
the model acquisition module is used for constructing a convolutional neural network model S-CNN containing the SCF, training the S-CNN by using a deep learning training data set, adjusting S-CNN structural parameters and hyper-parameters according to a training result, acquiring a trained S-CNN model with fault line selection capability, and realizing single-phase earth fault line selection of a power grid.
6. The power distribution network single-phase earth fault line selection system of claim 5, wherein the fault scenario set is a parameter history table formed by taking values of a power grid voltage, a node load, a line fault position, a transition resistance and a fault phase angle within a preset range according to an actual operation condition of the power distribution network, and the fault scenario set is formed by arranging and combining parameter values in the parameter history table.
7. The single-phase ground fault line selection system of the power distribution network of claim 5, wherein the discrete form of the S-transform is:
Figure FDA0003087376040000031
Figure FDA0003087376040000032
in the formula, x [ i ]]Is a discrete time sequence signal, i-0, 1, N-1; f. ofsDiscrete signal sampling frequency, N is the number of sampling points;
Figure FDA0003087376040000033
represents the time point, k ═ 1,2, ·, N;
Figure FDA0003087376040000034
represents a frequency point, N ═ 0,1, ·, N/2; t is an integer, m-0, 1, N-1.
8. The single-phase earth fault line selection system of the power distribution network of claim 5, wherein the fault zero-sequence current correlation graph SCF is a two-dimensional matrix formed by arranging fault zero-sequence current correlations of each line as follows:
Figure FDA0003087376040000035
wherein s isijAnd the Euclidean distance between the full-frequency-band time-frequency information of the fault zero-sequence current of the line i and the line j is represented, i is more than or equal to 1 and less than or equal to y, j is more than or equal to 1 and less than or equal to y, and y is the total number of the lines.
CN202011018649.8A 2020-09-24 2020-09-24 Power distribution network single-phase earth fault line selection method and system Expired - Fee Related CN112180210B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011018649.8A CN112180210B (en) 2020-09-24 2020-09-24 Power distribution network single-phase earth fault line selection method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011018649.8A CN112180210B (en) 2020-09-24 2020-09-24 Power distribution network single-phase earth fault line selection method and system

Publications (2)

Publication Number Publication Date
CN112180210A CN112180210A (en) 2021-01-05
CN112180210B true CN112180210B (en) 2021-08-10

Family

ID=73943425

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011018649.8A Expired - Fee Related CN112180210B (en) 2020-09-24 2020-09-24 Power distribution network single-phase earth fault line selection method and system

Country Status (1)

Country Link
CN (1) CN112180210B (en)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113625107A (en) * 2021-08-02 2021-11-09 四川轻化工大学 Power distribution network single-phase earth fault line selection method
CN114609468B (en) * 2021-12-16 2023-12-05 国网湖北省电力有限公司孝感供电公司 Cloud edge end cooperation-based single-phase earth fault interval distinguishing method for power distribution network
CN114252739B (en) * 2021-12-24 2023-11-03 国家电网有限公司 Power distribution network single-phase earth fault discrimination method, system, equipment and storage medium
CN114910742A (en) * 2022-05-05 2022-08-16 湖南腾河智慧能源科技有限公司 Single-phase fault grounding monitoring method and system, electronic equipment and storage medium
CN115097354B (en) * 2022-08-25 2023-01-20 北京智芯半导体科技有限公司 Method, system, electronic device and medium for determining fault line of power distribution network
CN116008731B (en) * 2023-02-15 2023-08-25 重庆大学 Power distribution network high-resistance fault identification method and device and electronic equipment
CN117031382B (en) * 2023-10-10 2024-02-20 云南电力试验研究院(集团)有限公司 Single-ended traveling wave device verification method based on signal-to-noise ratio and sine fitting method

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20130034097A (en) * 2011-09-28 2013-04-05 인하대학교 산학협력단 System for distortion compensation of ultra fast impulse signal on coaxial cable
CN107765156A (en) * 2017-12-07 2018-03-06 江苏省电力试验研究院有限公司 A kind of system and method for the power distribution network main equipment live detection based on power line carrier, PLC
CN108107324A (en) * 2017-12-22 2018-06-01 北京映翰通网络技术股份有限公司 A kind of electrical power distribution network fault location method based on depth convolutional neural networks
CN108279364A (en) * 2018-01-30 2018-07-13 福州大学 Wire selection method for power distribution network single phase earthing failure based on convolutional neural networks
CN109142969A (en) * 2018-07-20 2019-01-04 西南交通大学 A kind of power transmission line fault phase selection based on Continuous Hidden Markov Model
CN109635928A (en) * 2018-12-06 2019-04-16 华北电力大学 A kind of voltage sag reason recognition methods based on deep learning Model Fusion
CN112698215A (en) * 2019-10-23 2021-04-23 诺乌姆工程有限公司 Estimating battery state from electrical impedance measurements using a convolutional neural network device

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20130034097A (en) * 2011-09-28 2013-04-05 인하대학교 산학협력단 System for distortion compensation of ultra fast impulse signal on coaxial cable
CN107765156A (en) * 2017-12-07 2018-03-06 江苏省电力试验研究院有限公司 A kind of system and method for the power distribution network main equipment live detection based on power line carrier, PLC
CN108107324A (en) * 2017-12-22 2018-06-01 北京映翰通网络技术股份有限公司 A kind of electrical power distribution network fault location method based on depth convolutional neural networks
CN108279364A (en) * 2018-01-30 2018-07-13 福州大学 Wire selection method for power distribution network single phase earthing failure based on convolutional neural networks
CN109142969A (en) * 2018-07-20 2019-01-04 西南交通大学 A kind of power transmission line fault phase selection based on Continuous Hidden Markov Model
CN109635928A (en) * 2018-12-06 2019-04-16 华北电力大学 A kind of voltage sag reason recognition methods based on deep learning Model Fusion
CN112698215A (en) * 2019-10-23 2021-04-23 诺乌姆工程有限公司 Estimating battery state from electrical impedance measurements using a convolutional neural network device

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于LOF和SVM的智能配电网故障辨识方法;胡伟 等;《电力自动化设备》;20160630;第7-12页 *

Also Published As

Publication number Publication date
CN112180210A (en) 2021-01-05

Similar Documents

Publication Publication Date Title
CN112180210B (en) Power distribution network single-phase earth fault line selection method and system
Lin et al. A fault classification method by RBF neural network with OLS learning procedure
CN103728535A (en) Extra-high-voltage direct-current transmission line fault location method based on wavelet transformation transient state energy spectrum
CN101762774B (en) Method for identifying high voltage direct current transmission line fault location based on genetic algorithm parameter
CN114441900A (en) Power distribution network single-phase earth fault line selection method and system
CN115587329A (en) Power distribution network fault classification method and system based on convolutional neural network extraction features
CN114137358B (en) Power transmission line fault diagnosis method based on graph convolution neural network
CN113610119B (en) Method for identifying power transmission line development faults based on convolutional neural network
CN107462810A (en) A kind of fault section location method suitable for active power distribution network
CN112557950B (en) Fault line selection method of power distribution network resonance grounding system based on matrix similarity
CN115494349B (en) Active power distribution network single-phase earth fault section positioning method
Sahel et al. Wavelet energy moment and neural networks based particle swarm optimisation for transmission line protection
CN114755530A (en) Robust fault positioning method for power transmission line
CN114062832A (en) Method and system for identifying short-circuit fault type of power distribution network
Silva et al. Sampling frequency influence at fault locations using algorithms based on artificial neural networks
CN112557808A (en) Single-phase earth fault positioning method for power distribution network
Kezunovic et al. Advanced approaches for detecting and diagnosing transients and faults
CN115407162B (en) Low-current ground fault line selection method and system based on sparse representation
CN113177717B (en) Quick evaluation method for toughness of power transmission system based on influence increment sensitivity
CN111665413A (en) Distribution line single-phase earth fault section positioning method
CN112345876A (en) Fault positioning method and system suitable for interval DTU
Eldin et al. High impedance fault detection in EHV series compensated lines using the wavelet transform
Jain et al. Classification and location of single line to ground faults in double circuit transmission lines using artificial neural networks
CN114895140B (en) Power distribution network fault positioning method based on virtual injection current ratio and two-stage reconstruction
Galvez et al. Fault location in hybrid AC/DC transmission grids containing DERs and HVDC lines

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20210810

CF01 Termination of patent right due to non-payment of annual fee