A distribution network topology identification method based on voltage amplitude

Abstract

The present invention belongs to the technical field of distribution network, and discloses a distribution network topology identification method based on voltage amplitude, which collects the voltage amplitude of multiple time sections of each node, uses a similarity algorithm for primary identification, and then uses a Lasso regression algorithm for secondary identification to generate a topology; the generated topology is verified, and if the voltage drop requirement is met, the topology is output; if the voltage drop requirement is not met, the topology is corrected until the voltage drop requirement is met. Compared with a single similarity algorithm, the present invention has a high recognition accuracy and well solves the problem that the voltage amplitude data of the distribution network has strong multiple correlations.

Description

Technical Field

[0001] The invention belongs to the technical field of distribution networks, and in particular relates to a distribution network topology identification method based on voltage amplitude. Background Art

[0002] As the core area and important link of building a new power system, the safety, reliability and intelligence level of the distribution network are directly related to the stable operation and efficient development of the entire power system. Accurate distribution network topology is an important prerequisite for a series of complex operations such as state estimation, fault location and relay protection of the distribution network, and it is also the basis for realizing flexible regulation, optimized operation and intelligent interaction of the distribution network. However, on the one hand, the structure of the distribution network is complex, and it is necessary to install precise measurement electronic devices or invest a lot of manpower to obtain the accurate topology of the distribution network, which undoubtedly increases the operating cost of the distribution network; on the other hand, the distribution network is often transformed and reconstructed, and some users make illegal connections, which changes the original structure of the topology. The design files are difficult to update in real time, and often cannot accurately reflect the current state of the distribution network. Therefore, how to accurately and effectively identify the topology of the distribution network and provide strong support for building a clean, low-carbon, safe and efficient new power system has become the focus of existing research.

[0003] At present, the topology identification methods of distribution networks are mainly divided into tree search method, signal injection method and data analysis method. The tree search method regards the topology of the entire distribution network system as a tree, takes a certain node as the root node, and traverses and searches from the root node. According to the search results, the topological structure of the network is identified, including the location and connection relationship of each node. The tree search method mainly includes depth-first search (DFS) and breadth-first search (BFS) according to different search strategies. In large-scale or complex problems, the search process will become very time-consuming. The signal injection method observes the flow trajectory of various signals in the line through communication methods such as power carrier communication or wireless communication to determine the connection relationship between components. Although the signal injection method is simple in principle and has a small amount of calculation, it requires the installation of additional identification equipment, which is costly, and the quality requirements of communication are relatively high, which is not suitable for widespread application. The data analysis method uses the voltage, power, current and other electrical quantity data collected by the Advanced Metering Infrastructure (AMI) and smart meters to identify the relationship between households and transformers, the phase relationship between users, and the connection relationship between upstream and downstream nodes. However, the data analysis method cannot handle some common problems well. For example, in the actual operation process, due to the physical characteristics of the distribution network itself, the voltage values ​​of multiple nodes show a "same rise and fall" trend, that is, the voltage amplitude data has a strong multi-correlation. Using a single similarity algorithm and a single judgment dimension will lead to a decrease in the accuracy of topology recognition. Summary of the invention

[0004] To solve the above problems, the present invention provides a distribution network topology identification method based on voltage amplitude, which adopts multiple similarity algorithms to avoid single judgment dimension and eliminate the error caused by collinearity. At the same time, the Lasso regression algorithm is used for secondary identification to further improve the accuracy of distribution network topology structure identification, improve the identification speed and reduce the equipment installation cost.

[0005] The present invention provides a method for identifying a distribution network topology based on voltage amplitude, comprising the following steps:

[0006] S1. Assuming that the distributed power supply is not considered, the advanced metering infrastructure (AMI) and smart meters are used to collect node voltage data of multiple time sections of each node, and the user's power consumption information is recorded and transmitted at pre-set time intervals to ensure data integrity and synchronization;

[0007] S2. Analyze the voltage data of the distribution network using a similarity algorithm to obtain a heat map between nodes, set corresponding thresholds to filter nodes, perform an identification, and obtain a possible neighbor node set of all nodes;

[0008] S3, taking the node connected to the main power supply as the parent node, the parent node as the response variable, and the neighbor node set of the parent node as the variable, the regression coefficient is calculated using the Lasso regression algorithm; the node with a high regression coefficient is selected as the child node, and then the child node is used as the new parent node to find a new child node, and the second identification is performed cyclically to obtain the topological structure of the distribution network;

[0009] S4. Check the distribution network topology. If the voltage amplitude of all parent nodes is higher than that of child nodes, the topology is directly output; if not, it is corrected until the above conditions are met and the topology is output.

[0010] Furthermore, in S2, the distribution network voltage data is analyzed using a similarity algorithm, specifically:

[0011] Assume there are two node n-dimensional voltage data X and Y, the Pearson correlation coefficient of the two nodes is:

[0012] ,

[0013] in, , are the means of variables X and Y, respectively. is the Pearson correlation coefficient between variables X and Y;

[0014] The Euclidean distance between two nodes is:

[0015] ,

[0016] Construct an n×n initial distance matrix M, the distance element M in the matrix M ij Represents the i-th data x of n-dimensional voltage data X i To the jth data y of the n-dimensional voltage data Y j The Euclidean distance from M 11 To M nn The path W with the minimum cumulative distance is the DTW distance of variables X and Y:

[0017] ,

[0018] W is the best path between variables X and Y, which consists of a set of steps of length K ( )composition.

[0019] Furthermore, in S2, an identification is performed to find the neighboring nodes of all nodes, specifically:

[0020] The Pearson correlation coefficient, Euclidean distance and Dynamic Time Warping (DTW) algorithm are used to generate their respective correlation coefficient matrices, and the thresholds eta1, eta2, and eta3 are set respectively;

[0021] In the Pearson correlation coefficient matrix, set the matrix element value greater than eta1 to 1, indicating that the Pearson correlation coefficient of the two nodes is high, otherwise it is 0, and generate a new matrix A;

[0022] In the Euclidean distance matrix, set the matrix element value less than eta2 to 1, indicating that the Euclidean distance between the two nodes is small, otherwise it is 0, and generate a new matrix B;

[0023] In the DTW matrix, set the matrix element value that is less than eta3 to 1, indicating that the DTW distance between the two nodes is small, otherwise it is 0, and generate a new matrix C;

[0024] The values ​​of the correlation coefficient obtained by using the Pearson correlation coefficient, Euclidean distance and DTW algorithm are used as evaluation indicators. If any two of the three evaluation indicators meet the threshold indicators, it means that the two nodes have a high degree of similarity, and one of the nodes is in the possible neighbor node set of the other node; the values ​​of matrices A, B, and C are added to generate matrix D, and the diagonal elements are set to 0, the element values ​​greater than or equal to 2 are 1, and the element values ​​less than 2 are 0, to generate the node adjacency matrix E.

[0025] Furthermore, S3 is specifically:

[0026] Select the child nodes with high regression coefficients as the parent nodes, and record the parent nodes and child nodes into the set Q;

[0027] Take the obtained child node as the new parent node, find the neighbor node set of the corresponding node as the variable, and the neighbor node set does not contain the nodes already recorded in the set Q;

[0028] Lasso regression is used to determine neighbor nodes. If there is only one neighbor node, there is no need to calculate the regression coefficient. The regression coefficient is directly set to 1, and the node is set as a child node. This cycle is repeated until all nodes are recorded in the set Q.

[0029] Furthermore, in S4, it is checked whether the generated topology satisfies that the voltage of all parent nodes is higher than that of child nodes. If so, the topology is output. If not, the positions of the parent nodes and child nodes of the unsatisfactory branches are swapped until all branches meet the conditions.

[0030] The beneficial effects of the present invention are as follows: the method of the present invention uses a variety of similarity algorithms to find a possible set of neighbor nodes of a node in one identification, and the node set includes real adjacent nodes; adjacent nodes are conditionally correlated, and the correlation is quantified using lasso regression, and irrelevant non-adjacent nodes are excluded using secondary identification. The method of the present invention only requires node data to complete the topological identification of the distribution network; the identification steps are simple, avoiding the problem of a single similarity algorithm causing a decrease in the recognition rate, and will not fall into a loop or a long time, and the cost is low, and does not require high-quality communication. The method of the present invention has a good recognition effect and has a very broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a flow chart of the method of the present invention;

[0032] Figure 2 It is the Pearson correlation coefficient heat map between nodes in the IEEE33-node distribution network;

[0033] Figure 3 It is the Euclidean distance heat map between nodes in the IEEE33-node distribution network;

[0034] Figure 4 It is the DTW distance heat map between nodes in the IEEE33-node distribution network;

[0035] Figure 5 It is a schematic diagram of the proportion of matrix elements in the Euclidean distance matrix;

[0036] Figure 6 It is a schematic diagram of the proportion of matrix elements in the DTW distance matrix;

[0037] Figure 7 It is a visualization diagram of matrix E;

[0038] Figure 8 It is a schematic diagram of the original topological structure of the distribution network;

[0039] Fig. 9 Schematic diagram of the topological structure obtained by secondary identification. DETAILED DESCRIPTION

[0040] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments in conjunction with the accompanying drawings.

[0041] Figure 1 The method of the present invention is a flowchart, comprising the following steps:

[0042] S1. Use the advanced measurement system AMI and smart meters to collect node voltage data of multiple time sections at each node, record and transmit the user's electricity consumption information at pre-set time intervals (such as 30 minutes, 60 minutes, etc.) to ensure data integrity and synchronization;

[0043] S2. Analyze the voltage data of the distribution network using a similarity algorithm to obtain a heat map between nodes, set corresponding thresholds to filter nodes, perform an identification, and obtain a possible neighbor node set of all nodes;

[0044] S3, taking the node connected to the main power supply as the parent node, the parent node as the response variable, and the neighbor node set of the parent node as the variable, the regression coefficient is calculated using the Lasso regression algorithm, and the node with the highest regression coefficient is selected as the child node, and then the child node is used as the new parent node to find a new child node, and the cycle is repeated for secondary identification to obtain the topological structure of the distribution network;

[0045] S4. Check the distribution network topology. If the voltage amplitude of all parent nodes is higher than that of child nodes, the topology is directly output. If not, it is corrected until the above conditions are met and the topology is output.

[0046] The present invention uses the 24-hour voltage data of the IEEE 33-node distribution network for verification, and collects data once every hour. Based on the collected data, the Pearson correlation coefficient, Euclidean distance and DTW algorithm are used to generate corresponding correlation coefficient matrices, which are represented by heat maps.

[0047] Assume there are two node n-dimensional voltage data X and Y, the Pearson correlation coefficient of the two nodes is:

[0048] ,

[0049] in, , are the means of variables X and Y, respectively. is the Pearson correlation coefficient between variables X and Y;

[0050] The Euclidean distance between two nodes is:

[0051] ,

[0052] Construct an n×n initial distance matrix M, the distance element M in the matrix M ij Represents the i-th data x of n-dimensional voltage data X i To the jth data y of the n-dimensional voltage data Y j The Euclidean distance from M 11 To M nn The path W with the minimum cumulative distance is the DTW distance of variables X and Y:

[0053] ,

[0054] W is the best path between variables X and Y, which consists of a set of steps of length K ( )composition.

[0055] Figure 2-Figure 4 It is a correlation coefficient heat map generated by the Pearson algorithm, Euclidean distance and DTW algorithm. The starting color of the gradient on the right side of the figure represents low density or low intensity, while the ending color represents high density or high intensity.

[0056] Using the above formula, generate the correlation coefficient matrix of Pearson correlation coefficient, Euclidean distance and DTW algorithm, and set the thresholds eta1, eta2, eta3 respectively. Figure 2 It can be seen that the Pearson correlation coefficient matrix is ​​a dense matrix, and the matrix elements are basically between 0.95 and 1. . Figure 5 and Figure 6 are the proportions of matrix elements in the Euclidean distance and DTW distance matrices, respectively. Figure 5 It can be seen that the matrix elements in the Euclidean distance matrix are concentrated between 0 and 0.06. Find all the elements between 0.04 and 0.06 and take the average value, which is 0.045. .exist Figure 6 In the matrix, the elements are mainly concentrated in the range of 0-0.1. Since each node is connected to at most three nodes, the present invention selects the three smallest elements in each row of the matrix that are not 0, adds all the elements, and obtains the average value of 0.0295. Therefore .

[0057] According to the above process, the matrix E is generated. Figure 7 This is a visualization of the matrix E. Matrix elements that are 1 are displayed in black, and those that are 0 are displayed in white. There are 33 nodes in the matrix, and the number of elements that are 1 in the matrix E is 212, which means there are 106 edges, so the graph corresponding to the adjacency matrix must have a cycle.

[0058] After obtaining the adjacency matrix of the distribution network nodes, Lasso regression is used to filter out branches without connection relationships. Starting from the nodes connected to the main power supply, child nodes are found. The search order and regression coefficients are shown in Table 1. First, find the edge (i, j) whose index contains the parent node among the 106 edges, traverse with the parent node as the response, and use the neighbor node set as the variable to calculate the respective regression coefficients. In the 33-node distribution network, the maximum number of neighbor nodes is 3 and the average is 2. If there are multiple (greater than or equal to 2) regression coefficients that are not 0, the two variables with the largest regression coefficient values ​​are selected, and the corresponding regression coefficient values ​​are a and b. If If it is greater than 0.5, it means there is only one child node, and if it is less than 0.5, it means there are two child nodes. This process is repeated layer by layer until all nodes are recorded in the set Q, as shown in Table 1. A regression coefficient value of 1 indicates that there is only one variable, and a value of 0 indicates that the variable corresponding to the edge has been filtered out.

[0059] Table 1 The order and correlation coefficient of secondary identification using Lasso regression

[0060]

[0061] Figure 8 and Fig. 9 They are the original topology of the 33-node distribution network and the topology generated by the secondary identification without verification. It can be seen that 31 of the 32 branches in the original topology were prepared to be identified, and branches 24-25 were identified incorrectly. The above-mentioned correction method is used to correct the topology based on whether the parent node voltage is higher than the child node voltage, and the obtained topology is the same as the original topology.

[0062] The above description is only a preferred embodiment of the present invention and is not intended to be a further limitation of the present invention. All equivalent changes made using the contents of the present specification and drawings are within the protection scope of the present invention.

Claims

1. A distribution network topology identification method based on voltage amplitude, characterized in that: The following steps are involved: S1. Assuming that the distributed power supply is not considered, the advanced measurement system AMI and smart meters are used to collect node voltage data of multiple time sections of each node, and the user's power consumption information is recorded and transmitted at pre-set time intervals to ensure data integrity and synchronization; S2. Analyze the voltage data of the distribution network using a similarity algorithm to obtain a heat map between nodes, set corresponding thresholds to filter nodes, perform an identification, and obtain a possible neighbor node set of all nodes; Among them, the similarity algorithm is used to analyze the distribution network voltage data, specifically: Assume there are two node n-dimensional voltage data X and Y, the Pearson correlation coefficient of the two nodes is: ， in, , are the means of variables X and Y, respectively. is the Pearson correlation coefficient between variables X and Y; The Euclidean distance between two nodes is: ， Construct an n×n initial distance matrix M, the distance element M in the matrix M ij Represents the i-th data x of n-dimensional voltage data X i To the jth data y of the n-dimensional voltage data Y j The Euclidean distance from M 11 To M nn The path W with the minimum cumulative distance is the DTW distance of variables X and Y: ， W is the best path between variables X and Y, consisting of a set of steps of length K composition; Perform an identification to find the neighbor nodes of all nodes, specifically: The Pearson correlation coefficient, Euclidean distance and DTW algorithm are used to generate their respective correlation coefficient matrices, and the thresholds eta1, eta2 and eta3 are set respectively; In the Pearson correlation coefficient matrix, set the matrix element value greater than eta1 to 1, indicating that the Pearson correlation coefficient of the two nodes is high, otherwise it is 0, and generate a new matrix A; In the Euclidean distance matrix, set the matrix element value less than eta2 to 1, indicating that the Euclidean distance between the two nodes is small, otherwise it is 0, and generate a new matrix B; In the DTW matrix, set the matrix element value that is less than eta3 to 1, indicating that the DTW distance between the two nodes is small, otherwise it is 0, and generate a new matrix C; The values ​​of the correlation coefficient obtained by using the Pearson correlation coefficient, Euclidean distance and DTW algorithm are used as evaluation indicators. If any two of the three evaluation indicators meet the threshold indicators, it means that the two nodes have a high degree of similarity, and one of the nodes is in the possible neighbor node set of the other node; add the values ​​of matrices A, B, and C to generate matrix D, set the diagonal elements to 0, the element values ​​greater than or equal to 2 to 1, and the element values ​​less than 2 to 0, and generate the node adjacency matrix E; S3, taking the node connected to the main power supply as the parent node, the parent node as the response variable, and the neighbor node set of the parent node as the variable, the regression coefficient is calculated using the Lasso regression algorithm; the node with a high regression coefficient is selected as the child node, and then the child node is used as the new parent node to find a new child node, and the second identification is performed cyclically to obtain the topological structure of the distribution network; S4. Check the distribution network topology. If the voltage amplitude of all parent nodes is higher than that of child nodes, the topology is directly output; if not, make corrections until the conditions are met and output the topology.

2. A distribution network topology identification method based on voltage amplitude according to claim 1, characterized in that: S3 is specifically: Select the child nodes with high regression coefficients as the parent nodes, and record the parent nodes and child nodes into the set Q; Take the obtained child node as the new parent node, find the neighbor node set of the corresponding node as the variable, and the neighbor node set does not contain the nodes already recorded in the set Q; Lasso regression is used to determine neighbor nodes. If there is only one neighbor node, there is no need to calculate the regression coefficient. The regression coefficient is directly set to 1, and the node is set as a child node. This cycle is repeated until all nodes are recorded in the set Q.

3. A distribution network topology identification method based on voltage amplitude according to claim 1, characterized in that: In S4, it is checked whether the generated topology satisfies that the voltage of all parent nodes is higher than that of child nodes. If so, the topology is output; if not, the positions of the parent nodes and child nodes of the unsatisfactory branches are swapped until all branches meet the conditions.

Images