CN116298699A - Power distribution network fault identification method based on wavelet optimization and graph convolution neural network - Google Patents
Power distribution network fault identification method based on wavelet optimization and graph convolution neural network Download PDFInfo
- Publication number
- CN116298699A CN116298699A CN202310510415.2A CN202310510415A CN116298699A CN 116298699 A CN116298699 A CN 116298699A CN 202310510415 A CN202310510415 A CN 202310510415A CN 116298699 A CN116298699 A CN 116298699A
- Authority
- CN
- China
- Prior art keywords
- node
- fault
- power distribution
- distribution network
- phase
- 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.)
- Pending
Links
- 238000009826 distribution Methods 0.000 title claims abstract description 53
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 26
- 238000005457 optimization Methods 0.000 title claims abstract description 10
- 239000013598 vector Substances 0.000 claims abstract description 46
- 230000009466 transformation Effects 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 24
- 239000011159 matrix material Substances 0.000 claims description 23
- 238000000354 decomposition reaction Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 5
- 238000003860 storage Methods 0.000 claims description 4
- 230000001105 regulatory effect Effects 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 239000004973 liquid crystal related substance Substances 0.000 claims 2
- 238000013507 mapping Methods 0.000 abstract description 5
- 238000005070 sampling Methods 0.000 abstract 1
- 125000003275 alpha amino acid group Chemical group 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000000547 structure data Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
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
-
- 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Locating Faults (AREA)
Abstract
The invention discloses a power distribution network fault identification method based on wavelet optimization and graph convolution neural network, which comprises the following steps: sampling three-phase voltage and three-phase current when the power distribution network fails, and taking node codes of the power distribution network at the moment as original fault data; constructing a power distribution network fault identification model, comprising: performing wavelet transformation on the original fault data to obtain an initial feature vector; and adding an initial residual error technology and a unit mapping technology to the traditional GCN model to obtain the deep-drawing convolution neural network. The invention not only can solve the problem of unobvious characteristics when the power distribution network breaks down, but also can improve the fault identification precision through the constructed improved graph convolution neural network for fault positioning and fault classification.
Description
Technical Field
The invention belongs to the technical field of power distribution network fault positioning, and particularly relates to a method for identifying the type and the position of a power distribution network fault based on a graph convolution neural network.
Background
The graph rolling neural network is an emerging graph data learning technology, has great advantages when processing massive graph structure data and complex relevance problems, can perform feature extraction on irregular non-European data, and has strong modeling capability on dependency relations among nodes in a graph.
In the aspect of existing power distribution network fault positioning based on graph convolution neural network, most students adopt fault voltage and fault current as inputs of the neural network, and the tasks of power distribution network fault positioning and fault category identification are separately modeled. However, the middle-low voltage distribution network in China mostly adopts a small-current grounding mode, and the problem that the fault current is small and the fault characteristics are not obvious exists after the distribution network has a small-current grounding fault. The distribution network has a plurality of branches and a complex line structure, and the traditional measuring device provides incomplete information, has larger measuring error and is easy to generate data loss and distortion. And the main stream power distribution network fault positioning and fault category identifying tasks have certain relevance, and the two separate modeling can cause the phenomena of slow fault processing and resource waste.
Disclosure of Invention
The invention aims to solve the defects of the prior art, and provides a power distribution network fault identification method based on wavelet optimization and graph convolution neural network, so as to solve the problems that when a power distribution network breaks down, fault characteristics are not obvious, and relativity between fault location and fault category is ignored.
In order to achieve the aim of the invention, the invention adopts the following technical scheme:
the invention discloses a power distribution network fault identification method based on wavelet optimization and graph convolution neural network, which is characterized by comprising the following steps:
step 1: constructing a data set of a graph convolution neural network;
step 1.1: abstracting a tree-shaped radial power distribution network into a topological graph consisting of vertexes and edges, and marking the topological graph as G= (V, E); v represents a node set abstracted by the load of the power distribution network, E represents an edge set abstracted by an overhead line or a cable line;
step 1.2: collecting three-phase voltage and three-phase current of each node when the power distribution network fails, and taking the three-phase voltage and the three-phase current as original fault data, wherein the original fault data is recorded as B= (B) 1 ，B 2 ，…，B i ，…，B K ) T K is the total number of nodes in a topological graph G of the power distribution network, wherein B i Is the set of three-phase voltages and three-phase currents of the ith node at the time of failure, an
Wherein (1)>
Phase A voltage representing the ith node, < +.>
B-phase voltage of the i-th node, < ->
C-phase voltage of the i-th node, < ->
Phase a current representing the i-th node, +.>
B-phase current representing the ith node, < +.>
C phase current of the ith node is represented, T represents transposition, i is more than or equal to 1 and less than or equal to K;
for original fault data b= (B 1 ，B 2 ，…，B i ，…，B K ) T After wavelet transformation and normalization processing, a set of K node initial feature vectors is obtained and is marked as Q= (Q) 1 ，Q 2 ，…，Q i ，…，Q K ) T Wherein Q is i An initial feature vector representing an i-th node at the time of failure;
step 1.3: define initial feature vector q= (Q) 1 ，Q 2 ，…，Q i ，…，Q K ) T The tag information set is c= (C 1，j ，C 2，j ，…，C i，j ，…，C K，j ) T Wherein C i，j True label information representing a j-th failure of an i-th node; 1.ltoreq.i.ltoreq.K, 1.ltoreq.j.ltoreq.J, J being the number of fault categories;
step 1.4: according to the topological graph G of the power distribution network, if the ith node is connected with the jth node, the jth row data a in the adjacent matrix A is caused to be displayed ij =1, otherwise, let a ij =0; thereby constructing an adjacency matrix A, and A epsilon R K×K ；
Step 2: constructing an improved graph roll-up neural network, comprising: l graph convolution layers and a full connection layer;
step 2.1: defining the serial number of the current graph convolution layer as l, and initializing l=0; when l=0, let H (l) ＝Q；
Step 2.2: calculating the feature vector H of the output of the first +1th picture volume layer by using the method (1) (l+1) ：
In the formula (1), the components are as follows,
a symmetric normalized Laplace matrix representing the topological graph G and obtained by the formula (2); alpha and beta are super parameters, and sigma (·) is an activation function; i is an identity matrix; w (W) (l) A weight matrix to be trained for the first graph convolution layer;
in the formula (2), A is an adjacent matrix,
a degree matrix of A+I;
step 2.3: after l+1 is assigned to L, the step 2.2 is returned to be executed until L > L, thereby obtaining a high-order feature vector
A high-order feature vector representing an i-th node;
step 2.4: the high-order feature vector H (L ) Inputting the fault probability P into a full-connection layer, and outputting the probability P of the j-th fault of the i-th node in the topological graph G of the power distribution network after the treatment of the softmax activation function i，j And according to the probability P i，j Obtaining the ith node senderPredictive label information of failure;
step 2.5: constructing a cross entropy loss function according to the predicted label information and the real label information of the faults of the ith node, training the improved graph convolution neural network by using a back propagation algorithm, and updating network parameters by minimizing the cross entropy loss function until the training times or the convergence position of the cross entropy loss function are reached, so that a power distribution network fault identification model is obtained and is used for classifying whether the overhead line of the power distribution network has faults and fault types.
The power distribution network fault identification method based on wavelet optimization and graph convolution neural network is characterized in that the wavelet transformation processing and normalization processing in the step 1.2 comprise the following steps:
step 1.2.1: original fault data b= (B) with db4 wavelet 1 ，B 2 ，…，B i ，…，B K ) T Vector composed of three-phase voltage and three-phase current of ith node in (a)
Four layers of discrete wavelet decomposition are carried out to obtain 4 layers of detail coefficient sequences which are respectively marked as m i，1 、m i，2 、m i，3 、m i，4 And a layer of approximation coefficient sequences, denoted as a i，4 ；
Step 1.2.2: the threshold function pair m constructed by using (3) i，1 、m i，2 、m i，3 、m i，4 Processing to obtain a processed 4-layer detail coefficient sequence
In the formula (3), m i，n (k) For the kth detail coefficient in the nth layer detail coefficient sequence in the ith node, |m i，n (k) I represents m i，n (k) Is used for the amplitude of (a) and (b),
the method comprises the steps that the k estimated detail coefficients in an n-th estimated detail coefficient sequence in an i-th node are estimated, n is the number of wavelet decomposition layers, and n is more than or equal to 1 and less than or equal to 4; lambda is a threshold value and is obtained by the formula (4); z is a regulating factor; e is the base of the exponential function;
in the formula (4), S is the length of the original signal data B;
step 1.2.3: for the processed 4-layer detail coefficient sequence
And approximation coefficient sequence a i，4 Reconstructing to obtain 4-layer wavelet detail component +.>
And concatenates the 4-layer wavelet detail components into a vector +.>
Thereby obtaining a vector m= (M) 1 ，M 2 ，…，M i ，…，M K ) T ；
Step 1.2.4: and carrying out standard normalization processing on the vector M so as to obtain a set Q of initial feature vectors.
The label information of the j-th fault of the i-th node is encoded by adopting a single-heat vector encoding mode, so that a label information set C= (C) is obtained 1，j ，C 2，j ，…，C i，j ，…，C K，j ) T The method comprises the steps of carrying out a first treatment on the surface of the If C i，j If =000, the label information indicating the j-th failure of the i-th node is failure-free, if C i，j If =001, the label information indicating the j-th failure of the i-th node is a single-phase ground fault, and if C i，j If 010, the label information indicating the j-th failure of the i-th node is a two-phase ground failure, if C i，j 011, the label information indicating the j-th failure of the i-th node is a two-phase short-circuit failure, if C i，j And=100, the label information indicating the j-th fault of the i-th node is a three-phase ground fault.
The electronic equipment comprises a memory and a processor, wherein the memory is used for storing a program for supporting the processor to execute the power distribution network fault identification method, and the processor is configured to execute the program stored in the memory.
The invention relates to a computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and the computer program is executed by a processor to execute the steps of the power distribution network fault identification method.
Compared with the prior art, the invention has the beneficial effects that:
1. the method for extracting the fault characteristics based on wavelet transformation fully considers the connection and influence among the nodes, improves the precision and the robustness of a fault prediction model on the basis of fusing a network topological structure, can be effectively applied to comprehensive fault situations with frequent actual topological changes, and solves the problem of unobvious fault characteristics.
2. The invention provides a deep graph convolution neural network model, which adopts an initial residual error technology and a unit mapping technology, and the two methods can enable the graph convolution network to stack more convolution layers so as to aggregate the characteristic information of a deeper neighbor. The method provided by the invention can effectively relieve the problem of over-smoothness and improve the accuracy of the model on fault identification.
3. The invention considers the connection between fault location and fault type, adopts a single-heat vector coding mode to code single-phase earth fault, two-phase short circuit fault and three-phase earth fault as the output layer of the graph convolution neural network, and provides a combined model of fault location and fault classification, which can identify specific fault positions and fault types.
Drawings
FIG. 1 is a flow chart of a method for identifying faults of a power distribution network;
FIG. 2 is a topology of an IEEE 33 node power distribution network;
fig. 3 is a four-layer multi-resolution decomposition step diagram of a wavelet transform.
Detailed Description
The following describes the embodiments of the present invention further with reference to the drawings.
In this embodiment, a power distribution network fault identification method based on wavelet optimization and graph convolution neural network, referring to fig. 1, includes the following steps:
step 1: constructing a data set of a graph convolution neural network;
step 1.1: as shown in fig. 2, the tree-shaped radial power distribution network is abstracted into a topological graph consisting of vertexes and edges, and is denoted as g= (V, E); v represents a node set abstracted by the load of the power distribution network, E represents an edge set abstracted by an overhead line or a cable line;
step 1.2: the feeder terminal device FTU is utilized to collect three-phase voltage and three-phase current of each node when the power distribution network fails, and the three-phase voltage and the three-phase current are used as original fault data and recorded as B= (B) 1 ，B 2 ，…，B i， …，B K ) T K is the total number of nodes in a topological graph G of the power distribution network, wherein B i Is the set of three-phase voltages and three-phase currents of the ith node at the time of failure, an
Wherein (1)>
Phase A voltage representing the ith node, < +.>
B-phase voltage of the i-th node, < ->
C-phase voltage of the i-th node, < ->
Phase a current representing the i-th node, +.>
B-phase current representing the ith node, < +.>
Phase C current representing the i node; t represents transposition, i is more than or equal to 1 and less than or equal to K;
because the db4 wavelet has better local information analysis capability in the time domain and the frequency domain and is more sensitive to non-periodic signals, the invention selects the parent wavelet as the db4 wavelet, and utilizes the db4 wavelet to respectively perform the analysis on the original fault data B= (B) 1 ，B 2 ，…，B i ，…，B K ) T Vector composed of three-phase voltage and three-phase current of each node
Performing four-layer discrete wavelet transform, wherein the four-layer discrete wavelet decomposition steps are shown in fig. 3; for original fault data b= (B 1 ，B 2 ，…，B i， …，B K ) T After wavelet transformation and normalization processing, a set of K node initial feature vectors is obtained and is marked as Q= (Q) 1 ，Q 2 ，…，Q i ，…，Q K ) T Wherein Q is i An initial feature vector representing an i-th node at the time of failure;
step 1.2.1: original fault data b= (B) with db4 wavelet 1 ，B 2 ，…，B i ，…，B K ) T Vector composed of three-phase voltage and three-phase current of ith node in (a)
Four layers of discrete wavelet decomposition are carried out to obtain 4 layers of detail coefficient sequences which are recorded as m i，1 、m i，2 、m i，3 、m i，4 And a layer of approximation coefficient sequences, denoted as a i，4 ；
Step 1.2.2: the amplitude of the wavelet coefficient of the original signal is larger than the amplitude of the coefficient of noise, so the invention constructs a new threshold function, reserves the coefficient with larger amplitude in the wavelet coefficient, sets the coefficient with lower amplitude in the wavelet coefficient to 0, and achieves the purpose of removing the original fault data B= (B) 1 ，B 2 ，…，B i ，…，B K ) T The noise in (2) is calculated by using the threshold function pair m constructed by the formula (1) i，1 、m i，2 、m i，3 、m i，4 Processing to obtain a processed 4-layer detail coefficient sequence
In the formula (1), m i，n (k) For the kth detail coefficient in the nth layer detail coefficient sequence in the ith node, |m i，n (k) I represents m i，n (k) Is used for the amplitude of (a) and (b),
the method comprises the steps that the k estimated detail coefficients in an n-th estimated detail coefficient sequence in an i-th node are estimated, n is the number of wavelet decomposition layers, and n is more than or equal to 1 and less than or equal to 4; lambda is a threshold value and is obtained by the formula (2); z is a regulating factor; e is the base of the exponential function;
in the formula (2), S is the length of the original signal data B;
step 1.2.3: for the processed 4-layer detail coefficient sequence
And approximation coefficient sequence a i，4 Reconstructing to obtain 4-layer wavelet detail component +.>
And concatenates the 4-layer wavelet detail components into a vector +.>
Thereby obtaining a vector m= (M) 1 ，M 2 ，…，M i， …，M K ) T ；
Step 1.2.4: and carrying out standard normalization processing on the vector M so as to obtain a set Q of initial feature vectors.
Step 1.3: define initial feature vector q= (Q) 1 ，Q 2 ，…，Q i ，…，Q K ) T The tag information set is c= (C 1，j ，C 2，j ，…，C i，j ，…，C K，j ) T Wherein C i，j Tag information indicating a j-th failure of an i-th node; 1.ltoreq.i.ltoreq.K, 1.ltoreq.j.ltoreq.J, J being the number of fault categories;
in this embodiment, the label information of the j-th fault of the i-th node is encoded by adopting a single-hot vector encoding mode to obtain a label information set c= (C) 1，j ，C 2，j ，…，C i，j ，…，C K，j ) T The method comprises the steps of carrying out a first treatment on the surface of the If C i，j If =000, the label information indicating the j-th failure of the i-th node is failure-free, if C i，j If =001, the label information indicating the j-th failure of the i-th node is a single-phase ground fault, and if C i，j If 010, the label information indicating the j-th failure of the i-th node is a two-phase ground failure, if C i，j 011, the label information indicating the j-th failure of the i-th node is a two-phase short-circuit failure, if C i，j And=100, the label information indicating the j-th fault of the i-th node is a three-phase ground fault.
Step 1.4: root of Chinese characterAccording to the topological graph G of the power distribution network, if the ith node is connected with the jth node, the jth row data a in the adjacent matrix A is caused to be displayed ij =1, otherwise, let a ij =0; thereby constructing an adjacency matrix A.epsilon.R K×K ；
Step 2: at present, the traditional GCN is a shallow layer model, the optimal performance is realized when the depth is 2, and the performance of the model is reduced when the traditional GCN stacks more layers and the nonlinearity is increased; therefore, the invention provides an improved GCN model, which can enable the improved GCN model to extract characteristic information from a higher-order neighbor through an initial residual error technology and a unit mapping technology without an overcomplete phenomenon, thereby improving the accuracy of model identification; constructing an improved graph roll-up neural network, comprising: l graph convolution layers and a full connection layer;
step 2.1: defining the serial number of the current graph convolution layer as l, and initializing l=0; when l=0, H (l) ＝Q；
Step 2.2: calculating the feature vector H of the output of the first +1th picture volume layer by using the method (3) (l) ：
In the formula (3), the amino acid sequence of the compound,
a symmetric normalized Laplace matrix representing the topological graph G and obtained by the formula (4); alpha and beta are super parameters, and sigma (·) is an activation function; i is an identity matrix; w (W) (l) A weight matrix to be trained for the first graph convolution layer;
in the formula (4), A is an adjacent matrix,
a degree matrix of A+I;
step 2.2.1:according to the initial residual technique of step 2, an initial feature vector q= (Q) is added during the process of stacking the aggregated features per picture volume 1 ，Q 2 ，…，Q i ，…，Q K ) T Thus, the node feature vector of the layer l+1
It is necessary to comprehensively consider the node feature vector +.>
And an initial feature vector q= (Q 1 ，Q 2 ，…，Q i ，…，Q K ) T The characteristics of the node cannot be diluted along with the increase of the layer number, so that excessive smoothness is avoided, wherein L is more than or equal to 0 and less than or equal to L;
step 2.1.2: according to the unit mapping technique of step 2, in the weight matrix W (l) Adding identity matrix I n Can be used for
The direct mapping to the output allows us to alleviate the overfitting using regularization etc. techniques while still retaining higher order information;
step 2.3: after l+1 is assigned to L, the step 2.2 is returned to be executed until L > L, thereby obtaining a high-order feature vector
Higher order feature vector representing the ith node；
Step 2.4: high order feature vector H (L) Inputting the fault probability P into a full-connection layer, and outputting the probability P of the j-th fault of the i-th node in the topological graph G of the power distribution network after the treatment of the softmax activation function i，j And according to probability P i，j Obtaining predictive label information of the fault of the ith node;
step 2.5: constructing a cross entropy loss function according to the predicted label information and the real label information of the ith node failure, and calculating the total loss function value L of all nodes through the cross entropy loss function in the formula (5) c Training the improved graph convolution neural network by using a back propagation algorithm, and updating network parameters by minimizing a cross entropy loss function until the training times or the convergence position of the cross entropy loss function are reached, so as to obtain a power distribution network fault identification model, and classifying whether the overhead line of the power distribution network has faults and fault types;
in the formula (5), when the ith node has the jth fault, y i，，j When no j-th fault occurs in the ith node, y is =1 i，j ＝0。
In this embodiment, an electronic device includes a memory for storing a program supporting the processor to execute the above method, and a processor configured to execute the program stored in the memory.
In this embodiment, a computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the method described above.
Claims (5)
1. A power distribution network fault identification method based on wavelet optimization and graph convolution neural network is characterized by comprising the following steps:
step 1: constructing a data set of a graph convolution neural network;
step 1.1: abstracting a tree-shaped radial power distribution network into a topological graph consisting of vertexes and edges, and marking the topological graph as G= (V, E); v represents a node set abstracted by the load of the power distribution network, E represents an edge set abstracted by an overhead line or a cable line;
step 1.2: collecting three-phase voltage and three-phase current of each node when the power distribution network fails, and taking the three-phase voltage and the three-phase current as original fault data, wherein the original fault data is recorded as B= (B) 1 ，B 2 ，…，B i ，…，B K ) T K is the total number of nodes in a topological graph G of the power distribution network, wherein B i Is the set of three-phase voltages and three-phase currents of the ith node at the time of failure, an
Wherein, the liquid crystal display device comprises a liquid crystal display device,
phase A voltage representing the ith node, < +.>
B-phase voltage of the i-th node, < ->
C-phase voltage of the i-th node, < ->
Phase a current representing the i-th node, +.>
B-phase current representing the ith node, < +.>
C phase current of the ith node is represented, T represents transposition, i is more than or equal to 1 and less than or equal to K;
for original fault data b= (B 1 ，B 2 ，…，B i ，…，B K ) T After wavelet transformation and normalization processing, a set of K node initial feature vectors is obtained and is marked as Q= (Q) 1 ，Q 2 ，…，Q i ，…，Q K ) T Wherein Q is i An initial feature vector representing an i-th node at the time of failure;
step 1.3: define initial feature vector q= (Q) 1 ，Q 2 ，…，Q i ，…，Q K ) T The tag information set is c= (C 1，j ，C 2，j ，…，C i，j ，…，C K，j ) T Wherein C i，j True label information representing a j-th failure of an i-th node; 1.ltoreq.i.ltoreq.K, 1.ltoreq.j.ltoreq.J, J being the number of fault categories;
step 1.4: according to the topological graph G of the power distribution network, if the ith node is connected with the jth node, the jth row data a in the adjacent matrix A is caused to be displayed ij =1, otherwise, let a ij =0; thereby constructing an adjacency matrix A, and A epsilon R K×K ；
Step 2: constructing an improved graph roll-up neural network, comprising: l graph convolution layers and a full connection layer;
step 2.1: defining the serial number of the current graph convolution layer as l, and initializing l=0; when l=0, let H (l) ＝Q；
Step 2.2: calculating the feature vector H of the output of the first +1th picture volume layer by using the method (1) (l+1) ：
In the formula (1), the components are as follows,
a symmetric normalized Laplace matrix representing the topological graph G and obtained by the formula (2); alpha and beta are super parameters, and sigma (·) is an activation function; i is an identity matrix; w (W) (l) A weight matrix to be trained for the first graph convolution layer;
in the formula (2), A is an adjacent matrix,
a degree matrix of A+I;
step 2.3: after l+1 is assigned to L, the step 2.2 is returned to be executed until L > L, thereby obtaining a high-order feature vector
A high-order feature vector representing an i-th node;
step 2.4: the high-order feature vector H (L) Inputting the fault probability P into a full-connection layer, and outputting the probability P of the j-th fault of the i-th node in the topological graph G of the power distribution network after the treatment of the softmax activation function i，j And according to the probability P i，j Obtaining predictive label information of the fault of the ith node;
step 2.5: constructing a cross entropy loss function according to the predicted label information and the real label information of the faults of the ith node, training the improved graph convolution neural network by using a back propagation algorithm, and updating network parameters by minimizing the cross entropy loss function until the training times or the convergence position of the cross entropy loss function are reached, so that a power distribution network fault identification model is obtained and is used for classifying whether the overhead line of the power distribution network has faults and fault types.
2. The power distribution network fault identification method based on wavelet optimization and graph convolution neural network according to claim 1, wherein the wavelet transformation processing and normalization processing in step 1.2 comprises:
step 1.2.1: original fault data b= (B) with db4 wavelet 1 ，B 2 ，…，B i ，…，B K ) T Vector composed of three-phase voltage and three-phase current of ith node in (a)
Four layers of discrete wavelet decomposition are carried out to obtain 4 layers of detail coefficient sequences which are respectively marked as m i，1 、m i，2 、m i，3 、m i，4 And a layer of approximation coefficient sequences, denoted as a i，4 ；
Step 1.2.2: the threshold function pair m constructed by using (3) i，1 、m i，2 、m i，3 、m i，4 Processing to obtain a processed 4-layer detail coefficient sequence
In the formula (3), m i，n (k) For the kth detail coefficient in the nth layer detail coefficient sequence in the ith node, |m i，n (k) I represents m i The magnitude of n (k),
the method comprises the steps that the k estimated detail coefficients in an n-th estimated detail coefficient sequence in an i-th node are estimated, n is the number of wavelet decomposition layers, and n is more than or equal to 1 and less than or equal to 4; lambda is a threshold value and is obtained by the formula (4); z is a regulating factor; e is the base of the exponential function;
in the formula (4), S is the length of the original signal data B;
step 1.2.3: for the processed 4-layer detail coefficient sequence
And approximation coefficient sequence a i，4 Reconstructing to obtain 4-layer wavelet detail component +.>
And concatenates the 4-layer wavelet detail components into a vector +.>
Thereby obtaining a vector m= (M) 1 ，M 2 ，…，M i ，…，M K ) T ；
Step 1.2.4: and carrying out standard normalization processing on the vector M so as to obtain a set Q of initial feature vectors.
3. The power distribution network fault identification method based on wavelet optimization and graph convolution neural network according to claim 1, wherein the method is characterized in that a unique thermal vector coding mode is adopted to code label information of j-th fault of i-th node to obtain a label information set C= (C) 1，j ，C 2，j ，…，C i，j ，…，C K，j ) T The method comprises the steps of carrying out a first treatment on the surface of the If C i，j If =000, the label information indicating the j-th failure of the i-th node is failure-free, if C i，j If =001, the label information indicating the j-th failure of the i-th node is a single-phase ground fault, and if C i，j If 010, the label information indicating the j-th failure of the i-th node is a two-phase ground failure, if C i，j 011, the label information indicating the j-th failure of the i-th node is a two-phase short-circuit failure, if C i，j And=100, the label information indicating the j-th fault of the i-th node is a three-phase ground fault.
4. An electronic device comprising a memory and a processor, wherein the memory is configured to store a program for supporting the processor to perform the power distribution network fault identification method of any one of claims 1-3, the processor being configured to execute the program stored in the memory.
5. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor performs the steps of the power distribution network fault identification method according to any of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310510415.2A CN116298699A (en) | 2023-05-08 | 2023-05-08 | Power distribution network fault identification method based on wavelet optimization and graph convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310510415.2A CN116298699A (en) | 2023-05-08 | 2023-05-08 | Power distribution network fault identification method based on wavelet optimization and graph convolution neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116298699A true CN116298699A (en) | 2023-06-23 |
Family
ID=86829063
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310510415.2A Pending CN116298699A (en) | 2023-05-08 | 2023-05-08 | Power distribution network fault identification method based on wavelet optimization and graph convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116298699A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117406024A (en) * | 2023-10-19 | 2024-01-16 | 国网湖北省电力有限公司荆门供电公司 | Negative sequence reconstruction technology based on MK (modeling verification) and application method thereof in fault section positioning |
| CN117406024B (en) * | 2023-10-19 | 2024-05-24 | 国网湖北省电力有限公司荆门供电公司 | Negative sequence reconstruction technology based on MK (modeling verification) and application method thereof in fault section positioning |
-
2023
- 2023-05-08 CN CN202310510415.2A patent/CN116298699A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117406024A (en) * | 2023-10-19 | 2024-01-16 | 国网湖北省电力有限公司荆门供电公司 | Negative sequence reconstruction technology based on MK (modeling verification) and application method thereof in fault section positioning |
| CN117406024B (en) * | 2023-10-19 | 2024-05-24 | 国网湖北省电力有限公司荆门供电公司 | Negative sequence reconstruction technology based on MK (modeling verification) and application method thereof in fault section positioning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109635928B (en) | Voltage sag reason identification method based on deep learning model fusion | |
| CN107749757B (en) | Data compression method and device based on stack type self-coding and PSO algorithm | |
| CN111159638B (en) | Distribution network load missing data recovery method based on approximate low-rank matrix completion | |
| CN113628059B (en) | Associated user identification method and device based on multi-layer diagram attention network | |
| CN114118150A (en) | Power distribution network single-phase earth fault line selection method and system | |
| CN113987910A (en) | Method and device for identifying load of residents by coupling neural network and dynamic time planning | |
| CN114660375A (en) | Method for identifying power equipment fault | |
| Liu et al. | CNN-based fault phase identification method of double circuit transmission lines | |
| CN116298699A (en) | Power distribution network fault identification method based on wavelet optimization and graph convolution neural network | |
| CN116127343A (en) | Household transformer relation and phase identification method and system based on voltage sequence similarity | |
| CN115409122A (en) | Method, system, equipment and medium for analyzing concurrent faults of power transformation equipment | |
| CN113466630B (en) | Power distribution network fault cause classification method based on RSSPN model | |
| CN115661539A (en) | Less-sample image identification method embedded with uncertainty information | |
| CN115128402A (en) | Power distribution network fault type identification method and system based on data driving | |
| CN115409317A (en) | Transformer area line loss detection method and device based on feature selection and machine learning | |
| CN114596464A (en) | Multi-feature interactive unsupervised target detection method and system, electronic device and readable storage medium | |
| CN114021424A (en) | PCA-CNN-LVQ-based voltage sag source identification method | |
| Zhu et al. | Fault Phase Identification Method Based on Convolutional Neural Network for Double Circuit Transmission Lines | |
| CN117725454B (en) | Transmission line defect characteristic distribution learning method | |
| CN112163613A (en) | Rapid identification method for power quality disturbance | |
| CN111797732A (en) | Video motion identification anti-attack method insensitive to sampling | |
| CN112445939A (en) | Social network group discovery system, method and storage medium | |
| CN115407162B (en) | Low-current ground fault line selection method and system based on sparse representation | |
| CN117097624B (en) | Network topology structure enhancement method and device, electronic equipment and storage medium | |
| CN114638555B (en) | Power consumption behavior detection method and system based on multilayer regularization extreme learning machine |
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 |