A multi-point grounding wire location method based on multi-layer segment separation graph neural network

Abstract

The present invention proposes a multi-point grounding wire positioning method of a multi-layer interval segment separation graph neural network. The main steps of the method include: building a simulation detection line, simulating various grounding conditions and collecting data as samples for training the graph neural network; using the graph attention network to adaptively assign weights to nodes and aggregate them; using the gradient descent method with momentum to calculate the deviation of each layer of the trained graph neural network; converting the deviation into weights and assigning them to each trained graph neural network layer; inputting the actual detection signal; linearly superimposing the output of each layer; finding the nodes that may be grounded and separating them and verifying them; and obtaining grounding segments of multiple different branches. The proposed distribution network grounding detection method based on a multi-layer graph neural network can solve the grounding detection problem of multiple grounding points, realize the function of obtaining multiple grounding segments under the condition of multiple grounding points, and improve the detection accuracy.

Description

Technical Field

[0001] The invention belongs to the field of power system detection, relates to an artificial intelligence method and a distribution network grounding detection method, and is suitable for detecting a grounding line of a power system distribution network. Background Art

[0002] The distribution network refers to a power grid that receives electrical energy from the transmission network or regional power plants and distributes it locally or according to voltage to various types of users through distribution facilities. The distribution network has a complex structure and uses overhead wires to achieve long-distance transmission. The actual working environment is harsh, and there are often uncertain factors such as trees and houses that cause grounding faults. The current common troubleshooting method is large-scale searches, cutting down trees that may contact overhead wires, which consumes huge manpower and environmental resources. Therefore, there is a special need for detection devices to predict the grounding point of the distribution network. The existing distribution network fault detection method can only determine whether there is grounding, and the detection accuracy is poor, and it is impossible to locate the specific location of the grounding point.

[0003] In addition, there are methods for applying graph neural networks to distribution network fault detection, but the single-layer graph neural network fault detection method can only give the grounding fault point in the topology map and is uncertain. It cannot deal with the problems caused by multiple grounding points, and there is no practical interval where the grounding point is located.

[0004] Therefore, a multi-point grounding wire positioning method based on a multi-layer interval separation graph neural network is proposed to solve the problem of multi-point grounding in the distribution network and the measurement accuracy problem caused by the uncertainty brought by the multi-point positions, and it can give a probabilistic grounding interval segment. Summary of the invention

[0005] The present invention proposes a multi-point grounding wire positioning method using a multi-layer interval segment separation graph neural network, which combines the multi-layer graph neural network with the gradient descent method with momentum, introduces segmentation operation and separation operation, and can obtain multiple grounding segments and grounding probabilities; the steps in the use process are:

[0006] Step 1: Build a simulation detection circuit, simulate various grounding conditions and collect data as samples for training the graph neural network;

[0007] Step 2: According to the analog detection circuit structure diagram, assume that the analog detection circuit has K branches, and divide each branch into N duan segments, each segment is considered a node, where the kth branch and the nth node x k,n The segment represented by (x k,n ,x k,n +Δx k ]; k=1,2,…,K; n=1,2,…,N duan ; Δx k is the kth branch N duan The length after equal division, x in the segmentk,n represents the distance (n-1)Δx from the starting point of the k-th branch to the k-th branch k The total number of nodes is N′=K×N duan ;

[0008] Step 3: Draw the topology diagram of the simulated detection line according to the node connection relationship after segmentation, and obtain the adjacency matrix A; the adjacency matrix A describes the connection relationship of the nodes in the topology diagram, and the dimension of A is N′×N′; among them, if the node x k,i and node x t,j If they are connected, they are neighbors, t=1,2,…,K; i,j=1,2,…,N; the (k-1)Nth nodes in A duan +i row (t-1)×N duan +j column elements and (t-1)×N duan +j row (k-1)N duan +i column elements are all 1; if two nodes are not connected, they are all 0;

[0009] Step 4: Take a sample of the training network and train the graph neural network; the feature matrix of one of the samples is:

[0010]

[0011] Among them, the dimension of U is N′×7; V 1,1,1 、V 1,1,2 and V 1,1,3 For node x 1，1 The three-phase voltage amplitude information; I 1,1,1 ,I 1,1,2 and I 1,1,3 For node x 1，1 The three-phase current amplitude information; P 1,1 The value is 1, indicating that the node x in this sample 1，1 Ground; P 1,1 The value is 0, indicating that the node x in this sample 1，1 Non-grounded; V 1,2,1 、V 1,2,2 and V 1,2,3 For node x 1,2 The three-phase voltage amplitude information; I 1,2,1 ,I 1,2,2 and I 1,2,3 For node x 1,2 The three-phase current amplitude information; P 1,2 The value is 1, indicating that the node x in this sample 1,2 Ground; P 1,2 The value is 0, indicating that the node x in this sample 1,2 Non-grounded; V k,n,1 、V k,n,2 and Vk,n,3 For node x k,n The three-phase voltage amplitude information; I k,n,1 ,I k,n,2 and I k,n,3 For node x k,n The three-phase current amplitude information; P k,n The value is 1, indicating that the node x in this sample k,n Ground; P k,n A value of 0 indicates that node x in this sample k,n Non-grounded; and For Node The three-phase voltage amplitude information; and For Node The three-phase current amplitude information; A value of 1 indicates that the node in this sample Grounding; A value of 0 indicates that the node in this sample Non-grounded; take the 7th column in U to form a probability vector Take out the (k-1)Nth duan +n rows as node x k,n The state feature vector

[0012] Step 5: Input the adjacency matrix A and feature matrix U into the graph neural network; the input of each node is the state feature vector The graph neural network iteratively aggregates the information of the target node’s neighborhood through embedding propagation, stacks B layers of propagation layers to obtain high-order neighborhood information, and updates the node status; the node x k,n After calculation and aggregation, the node status is:

[0013]

[0014] in, For node x k,n The node state in the propagation layer of the lth layer of the graph neural network, l = 1, 2, …, B; f(·,·,·,·) is the state transfer function, For node x k,n The eigenvectors of the connected edges, For node x k,n The neighbor node state feature vector of For node x k,n At the node state of the l-1th layer propagation layer; node x is obtained by aggregation through the B layer propagation layer k,n Final State and the state feature vector V′k,n,1 , V′ k,n,2 and V′ k,n,3 For node x k,n After calculating and aggregating the three-phase voltage amplitude information, I′ k,n,1 , I′ k,n,2 and I′ k,n,3 For node x k,n After calculating and aggregating the three-phase current amplitude information, P′ k,n For node x k,n The grounding probability generated after calculation and aggregation;

[0015] Step 6: Introduce the graph attention network to adaptively assign weights to nodes and aggregate them; aggregate the state feature vectors of N′ nodes As input, the state feature vector set of the new N′ nodes is obtained after calculation through the graph attention network V″ k,n,1 , V″ k,n,2 and V″ k,n,3 For node x k,n The three-phase voltage amplitude information calculated by the graph attention network, I″ k,n,1 , I″ k,n,2 and I″ k,n,3 For node x k,n The three-phase current amplitude information calculated by the graph attention network, P″ k,n For node x k,n The ground probability calculated by the graph attention network; the specific method is to k,n and node x t,j Apply the weight matrix W, W with a dimension of 7×7 respectively to calculate the node attention coefficient:

[0016]

[0017] Where L(·) represents the activation function exp(·) represents an exponential function with e as the base, a k(n),t(j) represents the attention coefficient, N(x k，n ) represents node x k，n The set of adjacent points of represents the weight matrix, The dimension of node x is 1×14; W represents the linear transformation weight matrix, and the dimension of W is 7×7; then the attention coefficient and the node state feature vector are linearly combined to obtain the node x k,n The output state feature vector for:

[0018]

[0019] in,

[0020] Step 7: Reintegrate the state feature vectors of all N′ nodes into the state matrix U″, and take out the 7th column to form the probability vector U″ is:

[0021]

[0022] in, for:

[0023]

[0024] Step 8: Based on steps 5, 6, and 7, take Q training samples to train the graph neural network and obtain the probability vector of each training sample, where the probability vector of the qth training sample is By using the gradient descent method with momentum, we can obtain the deviation δ after the graph neural network training; k,n The ground probability estimate y of the node C(k,n) for:

[0025]

[0026] Where β is the exponential decay rate, f(W) is a random objective function differentiable with respect to W; f 1 (W),f 2 (W),…,f C (W) represents the random objective function corresponding to the step size 1, 2, ..., C, Gradient of the random objective function over c steps Node x k,n Expected grounding probability E(y C(k,n) ) and the actual grounding situation is:

[0027]

[0028] in,

[0029] The average deviation of all nodes is:

[0030]

[0031] Step 9: From step 4 to step 8, get L Graph The average deviation of the trained graph neural network for all nodes in the lth layer is δ′ l , and assign weight coefficient α to the trained graph neural network in layer l l ;

[0032] The weights of the trained graph neural network in the lth layer are:

[0033]

[0034] Step 10, give L Graph The trained graph neural network inputs the feature matrix of the actual detection line and outputs the probability vector respectively; is the output probability vector of the trained graph neural network at the lth layer;

[0035] Step 11: Graph The trained graph neural network is linearly superimposed; the probability vector Y is:

[0036]

[0037] in is the grounding probability after linear superposition of each node;

[0038] Step 12: Specify the probability threshold ε, ε∈(0,1), and discuss the nodes whose grounding probability is greater than the probability threshold. And consider that other nodes are not grounded;

[0039] Step 13, separate and verify the nodes under discussion; due to the influence of neighboring nodes, the grounding probability of a single node contains uncertainty, and the grounding probability will increase under this influence, so the separation operation is introduced; one of the nodes under discussion x k,n The state feature vector and is the actual detection node x k,n The three-phase voltage amplitude information, and is the actual detection node x k,n The three-phase current amplitude information of the ground detection device is replaced by the state feature vector of the node when it is operating normally after the signal is generated by the ground detection device, and the state feature vector of the node when it is operating normally is obtained. k,n Whether the grounding characteristic matrix for:

[0040]

[0041] Where V 1,1,1(0) 、V 1,1,2(0) and V 1,1,3(0) After the ground detection device generates a signal, the node x 1,1 Three-phase voltage amplitude information of normal operation; I 1,1,1(0) ,I 1,1,2(0) and I 1,1,3(0) After the ground detection device generates a signal, the node x 1,1 Normal operation three-phase current amplitude information; V 1,2,1(0) 、V1,2,2(0) and V 1,2,3(0) After the ground detection device generates a signal, the node x 1,2 Three-phase voltage amplitude information of normal operation; I 1,2,1(0) ,I 1,2,2(0) and I 1,2,3(0) After the ground detection device generates a signal, the node x 1,2 Normal operation of the three-phase current amplitude information; V k,n-1,1(0) 、V k,n-1,2(0) and V k,n-1,3(0) After the ground detection device generates a signal, the node x k,n-1 Three-phase voltage amplitude information of normal operation; I k,n-1,1(0) ,I k,n-1,2(0) and I k,n-1,3(0) After the ground detection device generates a signal, the node x k,n-1 Normal operation of the three-phase current amplitude information; V k,n+1,1(0) 、V k,n+1,2(0) and V k,n+1,3(0) After the ground detection device generates a signal, the node x k,n+1 Three-phase voltage amplitude information of normal operation; I k,n+1,1(0) ,I k,n+1,2(0) and I k,n+1,3(0) After the ground detection device generates a signal, the node x k,n+1 Three-phase current amplitude information of normal operation; and The node after the ground detection device generates a signal Three-phase voltage amplitude information during normal operation; and The node after the ground detection device generates a signal Normal three-phase current amplitude information; Input L Graph The trained graph neural network of the layer is obtained by step 11 about the node x k,n The probability vector in The nodes x 1,1 ,x 1,2 ,…,x k,n-1 ,x k,n+1 ,…,x K,N At node x k,n The probability of grounding under the influence, To discuss node x separately k,n Is node x grounded? k,n The grounding probability; verify again If true, then it is considered that at node x k,n The segment (x k,n ,x k,n +Δx k] exists in the ground; if not, node x is excluded k,n ;

[0042] Step 14, repeat step 13 to add the node x that needs to be discussed k1,n1 ,x k2,n2 ,…,x kt,nt Separate and verify to obtain multiple grounding segments (x k1,n1 ,x k1,n1 +Δx k1 ],(x k2,n2 ,x k2,n2 +Δx k2 ],…,(x kt,nt ,x kt,nt +Δx kt ].

[0043] Compared with the prior art, the present invention has the following advantages and effects:

[0044] (1) The existing technology cannot accurately locate the grounding point, but can only determine whether there is grounding, and cannot deal with the problem of grounding at multiple grounding points. The present invention introduces a multi-layer graph neural network to solve the problem of multi-point grounding detection in the distribution network.

[0045] (2) The present invention proposes segmented operation, which can obtain multiple grounding segments and the grounding probability of each grounding segment.

[0046] (3) The present invention proposes a separation operation, which can solve the uncertainty problem caused by multi-point grounding and thus improve the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The invention discloses a flow chart of grounding detection of a distribution network. DETAILED DESCRIPTION

[0048] A multi-point grounding wire positioning method proposed by the present invention is described in detail as follows in conjunction with the accompanying drawings:

[0049] Figure 1 The invention discloses a flow chart of grounding detection of a distribution network.

[0050] Step S1, build a simulation detection circuit, simulate various grounding conditions and collect data as samples for training the graph neural network.

[0051] Step S2: according to the simulation detection circuit structure diagram, each branch is divided into N duan segments, and regard each segment as a node.

[0052] Step S3, draw a topological diagram of the simulated detection line according to the node connection relationship after segmentation, and obtain the adjacency matrix A.

[0053] Step S4, take a sample of the training network and train the graph neural network.

[0054] Step S5, input the adjacency matrix A and the feature matrix U of the sample into the graph neural network.

[0055] Step S6: Introduce the graph attention network to adaptively assign weights to nodes and aggregate them.

[0056] Step S7: Reintegrate the state feature vectors of all nodes into the state matrix U″, and take out the 7th column to form the probability vector

[0057] Step S8, from steps S5, S6, and S7, take Q training samples to train the graph neural network and obtain the probability vector of each training sample; then use the gradient descent method with momentum to obtain the deviation δ′ after the graph neural network training.

[0058] Step S9, obtain L from step S4 to step S8 Graph The trained graph neural network of the first layer is assigned a weight coefficient α to the trained graph neural network of the first layer. l .

[0059] Step S10, respectively give L Graph The trained graph neural network inputs the feature matrix of the actual detection line and outputs the probability vectors respectively.

[0060] Step S11: L Graph The trained graph neural network layers are linearly superimposed to obtain the superimposed probability vector Y.

[0061] Step S12, specify a probability threshold ε, discuss nodes whose grounding probability is greater than the probability threshold, and consider other nodes not to be grounded.

[0062] Step S13, separate and verify the node under discussion. Replace the state feature vector of the node under discussion with the state feature vector of the node when the grounding detection device generates a signal when the node is operating normally, and obtain the state feature vector for verifying the node x k,n Whether the grounding characteristic matrix Will Input L Graph The trained graph neural network of the layer obtains the information about node x from step S11. k,n The probability vector Verify again If true, then it is considered that at node x k,n There is a ground connection in the segment represented; if not, node x is excluded k,n .

[0063] Step S14, repeating step S13 to obtain multiple grounding segments of different branches.

[0064] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A multi-point grounding wire location method based on multi-layer segment separation graph neural network, It is characterized in that Combining the multi-layer graph neural network with the gradient descent method with momentum, introducing segmentation and separation operations, multiple grounding segments and grounding probabilities can be obtained, which can solve the uncertainty problem caused by multi-point grounding and improve detection accuracy. The steps in use are: Step 1: Build a simulation detection circuit, simulate various grounding conditions and collect data as samples for training the graph neural network; Step 2: According to the analog detection circuit structure diagram, assume that the analog detection circuit has K branches, and divide each branch into N duan segments, each segment is considered a node, where the kth branch and the nth node x k，n The segment represented by (x k，n , x k，n +Δx k ];k=1,2,…,K;n=1,2,…,N duan ; Δx k is the kth branch N duan The length after equal division, x in the segment k，n Indicates the distance (n-1) Δx from the starting point of the k-th branch to the k-th branch k The total number of nodes is N′=K×N duan ; Step 3: Draw the topology diagram of the simulated detection line according to the node connection relationship after segmentation, and obtain the adjacency matrix A; the adjacency matrix A describes the connection relationship of the nodes in the topology diagram, and the dimension of A is N′×N′; among them, if the node x k，i and node x t，j If they are connected, they are neighbors, t = 1, 2, ..., K; i, j = 1, 2, ..., N; the (k-1)Nth nodes in A duan +i row (t-1)×N duan +j column elements and (t-1)×N duan +j row (k-1)N duan +i column elements are all 1; if two nodes are not connected, they are all 0; Step 4: Take a sample of the training network and train the graph neural network. The feature matrix of one of the samples is: Among them, the dimension of U is N′×7; V 1，1，1 、V 1，1，2 and V 1，1，3 For node x 1，1 The three-phase voltage amplitude information; I 1，1，1 ,I 1，1，2 and I 1，1，3 For node x 1，1 The three-phase current amplitude information; P 1，1 The value is 1, indicating that the node x in this sample 1，1 Ground; P 1，1 The value is 0, indicating that the node x in this sample 1，1 Non-grounded; V 1，2，1 、V 1，2，2 and V 1，2，3 For node x 1，2 The three-phase voltage amplitude information; I 1，2，1 ,I 1，2，2 and I 1，2，3 For node x 1，2 The three-phase current amplitude information; P 1，2 The value is 1, indicating that the node x in this sample 1，2 Ground; P 1，2 A value of 0 indicates that node x in this sample 1，2 Non-grounded; V k，n，1 、V k，n，2 and V k，n，3 For node x k，n The three-phase voltage amplitude information; I k，n，1 ,I k，n，2 and I k，n，3 For node x k，n The three-phase current amplitude information; P k，n The value is 1, indicating that the node x in this sample k，n Ground; P k，n The value is 0, indicating that the node x in this sample k，n Non-grounded; and For Node The three-phase voltage amplitude information; and For Node The three-phase current amplitude information; A value of 1 indicates that the node in this sample Grounding; A value of 0 indicates that the node in this sample Non-grounded; take the 7th column in U to form a probability vector Take out the (k-1)Nth duan +n rows as node x k，n The state feature vector Step 5: Input the adjacency matrix A and feature matrix U into the graph neural network; the input of each node is the state feature vector The graph neural network iteratively aggregates the information of the target node’s neighborhood through embedding propagation, stacks B layers of propagation layers to obtain high-order neighborhood information, and updates the node status; the node x k，n After calculation and aggregation, the node status is: in, For node x k，n The node state in the propagation layer of the graph neural network, l = 1, 2, …, B; f(·, ·, ·, ·) is the state transfer function, For node x k，n The eigenvectors of the connected edges, For node x k，n The neighbor node state feature vector of For node x k，n At the node state of the l-1th layer propagation layer; node x is obtained by aggregation through the B layer propagation layer k，n Final State and the state feature vector V′ k，n，1 , V′ k，n，2 and V′ k，n，3 For node x k，n After the three-phase voltage amplitude information is calculated and aggregated, I′ k，n，1 , I′ k，n，2 and I′ k，n，3 For node x k，n After calculating and aggregating the three-phase current amplitude information, P′ k，n For node x k，n The grounding probability generated after calculation and aggregation; Step 6, introduce a graph attention network to adaptively assign weights between nodes and aggregate; the set of state feature vectors of N' nodes is used as the input, and after calculation by the graph attention network, a new set of state feature vectors of N' nodes V″ k，n，1 、V″ k，n，2 and V″ k，n，3 are the three-phase voltage amplitude information of node x k，n after calculation by the graph attention network, and I″ k，n，1 、I″ k，n，2 and I″ k，n，3 are the three-phase current amplitude information of node x k，n after calculation by the graph attention network, and P″ k，n is the grounding probability of node x k，n after calculation by the graph attention network; the specific method is to apply the weight matrix W to node x k，n and node x t，j respectively. The dimension of W is 7×7, and calculate the node attention coefficient: Where L(·) represents the activation function exp(·) represents an exponential function with e as the base, a k(n)，t(j) represents the attention coefficient, N(x k，n ) represents node x k，n The set of adjacent points of represents the weight matrix, The dimension of node x is 1×14; W represents the linear transformation weight matrix, and the dimension of W is 7×7; then the attention coefficient and the node state feature vector are linearly combined to obtain the node x k，n The output state feature vector for: in, Step 7: Reintegrate the state feature vectors of all N′ nodes into the state matrix U″, and take out the 7th column to form the probability vector U″ is: in, for: Step 8: Based on steps 5, 6, and 7, take Q training samples to train the graph neural network and obtain the probability vector of each training sample, where the probability vector of the qth training sample is q = 1, 2, ..., Q; the deviation δ after graph neural network training is obtained by the gradient descent method with momentum; x k，n The ground probability estimate y of the node C(k，n) for: Where β is the exponential decay rate, f(W) is a random objective function differentiable with respect to W; f 1 (W), f 2 (W),…,f C (W) represents the random objective function corresponding to the step size 1, 2, ..., C, Gradient of the random objective function over c steps Node x k，n Expected grounding probability E(y C(k，n) ) and the actual grounding situation is: in, The average deviation of all nodes is: Step 9: From step 4 to step 8, get L Graph The average deviation of the trained graph neural network for all nodes in the lth layer is δ′ l , and assign weight coefficient α to the trained graph neural network in layer l l ; The weights of the trained graph neural network in the lth layer are: Step 10, give L Graph The trained graph neural network inputs the feature matrix of the actual detection line and outputs the probability vector respectively; is the output probability vector of the trained graph neural network at the lth layer; Step 11: Graph The trained graph neural network is linearly superimposed; the probability vector Y is: in is the grounding probability after linear superposition of each node; Step 12: specify a probability threshold ε, ε∈(0,1), and discuss nodes whose grounding probability is greater than the probability threshold And consider that other nodes are not grounded; Step 13, separate and verify the nodes under discussion; due to the influence of neighboring nodes, the grounding probability of a single node contains uncertainty, and the grounding probability will increase under this influence, so the separation operation is introduced; one of the nodes under discussion x k，n The state feature vector and is the actual detection node x k，n The three-phase voltage amplitude information, and is the actual detection node x k，n The three-phase current amplitude information of the ground detection device is replaced by the state feature vector of the node when it is operating normally after the signal is generated by the ground detection device, and the state feature vector of the node when it is operating normally is obtained. k，n Whether the grounding characteristic matrix for: Where V 1，1，1(0) 、V 1，1，2(0) and V 1，1，3(0) After the ground detection device generates a signal, the node x 1，1 Three-phase voltage amplitude information of normal operation; I 1，1，1(0) ,I 1，1，2(0) and I 1，1，3(0) After the ground detection device generates a signal, the node x 1，1 Normal operation three-phase current amplitude information; V 1，2，1(0) 、V 1，2，2(0) and V 1，2，3(0) After the ground detection device generates a signal, the node x 1，2 Three-phase voltage amplitude information of normal operation; I 1，2，1(0) ,I 1，2，2(0) and I 1，2，3(0) After the ground detection device generates a signal, the node x 1，2 Normal operation three-phase current amplitude information; V k，n-1，1(0) 、V k，n-1，2(0) and V k，n-1，3(0) After the ground detection device generates a signal, the node x k，n-1 Three-phase voltage amplitude information of normal operation; I k，n-1，1(0) ,I k，n-1，2(0) and I k，n-1，3(0) After the ground detection device generates a signal, the node x k，n-1 Normal operation three-phase current amplitude information; V k，n+1，1(0) 、V k，n+1，2(0) and V k，n+1，3(0) After the ground detection device generates a signal, the node x k，n+1 Three-phase voltage amplitude information of normal operation; I k，n+1，1(0) ,I k，n+1，2(0) and I k，n+1，3(0) After the ground detection device generates a signal, the node x k，n+1 Three-phase current amplitude information of normal operation; and The node after the ground detection device generates a signal Three-phase voltage amplitude information during normal operation; and The node after the ground detection device generates a signal Normal operation of the three-phase current amplitude information; Input L Graph The trained graph neural network of the layer is obtained by step 11 about the node x k，n The probability vector in The nodes x 1，1 , x 1，2 , …, x k，n-1 , x k，n+1 , …, x K，N At node x k，n The probability of grounding under the influence, To discuss node x separately k，n Is node x grounded? k，n The grounding probability; verify again If true, then it is considered that at node x k，n The segment (x k，n , x k，n +Δx k ] exists in the ground; if not, node x is excluded k，n ; Step 14, repeat step 13 to add the node x that needs to be discussed k1，n1 , x k2，n2 , …, x kt，nt Separate and verify to obtain multiple grounding segments (x k1，n1 , x k1，n1 +Δx k1 ],(x k2，n2 , x k2，n2 +Δx k2 ],…,(x kt，nt , x kt，nt +Δx kt ].

Images