A simulation error-dominated region search method based on multi-level block decoupling algorithm

Abstract

A method for searching for simulation error-dominated regions based on a multi-level block decoupling algorithm, S1, using a primary block decoupling algorithm based on network topology, dividing the nodes into blocks according to the node's outgoing line degree, and decomposing the original power grid into multiple subnets according to the aggregated adjacency matrix; S2, performing hybrid dynamic simulation by injecting measured PMU data at the boundary points of each subnet to reduce the error range, and then using a secondary block decoupling algorithm to further reduce the error region into blocks, performing hybrid dynamic simulation on smaller sub-regions to find error elements. Compared with traversing all nodes for hybrid dynamic simulation, the method proposed in the present invention can achieve efficient search for error-causing regions, quickly locate error elements, and reduce computational costs.

Description

Technical Field

[0001] The invention belongs to the technical field of power system simulation modeling, and in particular relates to a simulation error dominant region search method based on a multi-level block decoupling algorithm. Background Art

[0002] In order to improve the efficiency of error area search, existing research mainly has the following methods: (1) Considering the dynamic distribution characteristics of system frequency, the frequency disturbance trajectory of each area is analyzed based on measured data. (2) Considering the dynamic distribution characteristics of system voltage, the spatiotemporal distribution characteristics after disturbance are quantified. The path and spatiotemporal characteristics of the disturbance propagation process are analyzed from time and space multi-dimensional dimensions. The disturbance characteristics of each node are analyzed by the spatiotemporal distribution characteristics after the system is disturbed, and the area with less disturbance is preliminarily divided. (3) The power grid is abstracted as an undirected graph, and the credibility is used as the line weight to obtain the minimum spanning tree and the optimal cut set to achieve system decoupling, and the error source is traced through multiple hybrid dynamic simulations. (4) The wide area first search method is used to identify the simulation correctable domain, and the error source is quickly located according to the intersection of the disturbance depth and the simulation error. The shortcomings of the existing technology are: the power grid area is large, the nodes are numerous, and the node numbers are chaotic. How to reasonably divide the power grid into blocks and then determine the location of the error. Current research has not yet provided a specific solution on how to decouple the system, that is, how to determine the decoupling point when using the hybrid dynamic simulation method and how to efficiently search for error components. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a simulation error dominant area search method based on a multi-level block decoupling algorithm. Compared with traversing all nodes for hybrid dynamic simulation, the method proposed by the present invention can realize efficient search of the error-causing area, quickly locate the error components, and reduce the computational cost.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0005] A simulation error dominant region search method based on a multi-level block decoupling algorithm, the steps are:

[0006] S1. Using the primary block decoupling algorithm based on network topology, the blocks are divided according to the node outgoing line degree, and the original power grid is decomposed into multiple subnets according to the aggregated adjacency matrix;

[0007] S2. Hybrid dynamic simulation is performed by injecting measured PMU data at each subnet boundary point to reduce the error range. Then, the secondary block decoupling algorithm is used to further reduce the error area and perform hybrid dynamic simulation on smaller sub-areas to find the error components.

[0008] Preferably, sub-step S1 is as follows:

[0009] S1.1. Search for nodes with a degree of 1 in the adjacency matrix, mark them as terminal nodes, start from the terminal node V with the largest natural number, search for nodes connected to the terminal node V with a degree less than m based on the adjacency matrix, where m represents any degree greater than 1, and add them to the sub-region set RV, then continue to search for nodes connected to the nodes added to the set with a degree less than m, and continue to add them to RV until the search set RV contains all nodes connected to the terminal node and with a degree less than m;

[0010] S1.2, remove the nodes in the set RV, do not participate in the next search, and regenerate the adjacency matrix;

[0011] S1.3, repeat steps S1.1-S1.2 until there is no node with degree 1 in the adjacency matrix; that is, the distribution network can be divided into multiple subnets based on the network topology.

[0012] Preferably, the method of the secondary block decoupling algorithm is as follows:

[0013] S2.1 Count the number of nodes in the error sub-region;

[0014] S2.2 If the number of nodes is an even number, i.e., 2n, then the sub-region is divided into two sub-regions, each of which contains n nodes in order;

[0015] S2.3 If the number of nodes is an odd number, i.e. 2n-1, the sub-region is divided into two ordered sub-regions of n and n-1 respectively;

[0016] S2.4 sets the maximum node number to be placed in n-1 sub-areas; hybrid dynamic simulation is performed on smaller sub-areas to find the error element.

[0017] Preferably, the adjacency matrix is ​​established as follows:

[0018] 1) Generate a weighted adjacency matrix;

[0019] 2) Define the matrix distance index D and the adjacent numbered node number index N;

[0020] 3) Use the simulated annealing algorithm to solve the new adjacency matrix.

[0021] Preferably, 1) the process of generating a weighted adjacency matrix is ​​as follows:

[0022] The weighted adjacency matrix introduces edge weight information based on the original adjacency matrix. These weights represent the influence of the line distance between two adjacent nodes. The longer the line, the lower the weight. The symbol W represents the weighted adjacency matrix, and its expression is as follows:

[0023] ;

[0024] ;

[0025] in, is the line distance between node i and node j. If the two nodes are not directly connected, the value is 0. and are the shortest distance and maximum distance between two adjacent nodes respectively; represents the weighted adjacency matrix; w ij Represents an element in the weighted adjacency matrix. When the value is 0, it means that nodes i and j are not directly connected. When the value is not 0, it reflects the influence of the line distance between two adjacent nodes.

[0026] Preferably, 2) the method of defining the matrix distance index D and the adjacent numbered node number index N is as follows;

[0027] The expression is as follows:

[0028] ;

[0029] ;

[0030] Among them, among them, is a sign function. If i and j differ by only 1, then , otherwise the value is equal to 0; similarly, if If not zero, , otherwise the value is equal to 0; the smaller the D value, the more the elements in the adjacency matrix are concentrated on the main diagonal, and the larger the N value, the higher the degree of connection between the elements in the adjacency matrix.

[0031] Preferably, 3) the process of using the simulated annealing algorithm to solve the new adjacency matrix is ​​as follows:

[0032] The following objective function is established:

[0033] ;

[0034] in is the weight, is the minimum matrix distance, is the maximum number of adjacent numbered nodes.

[0035] Preferably, in order to centralize the numbers of nodes that are closely connected in the network topology to facilitate block decoupling, topology aggregation should comply with the following conditions:

[0036] Condition 1: The number of adjacent numbered nodes, N, should be as large as possible, while the matrix distance, D, should be as small as possible;

[0037] Condition 2: The main operation of topological aggregation is to exchange the rows and columns of the adjacency matrix. Since the adjacency matrix is ​​a symmetric square matrix, the exchange of rows and columns should be consistent.

[0038] An error tracing system based on network topology aggregation and multi-level block decoupling includes a processing module. The program instructions executed by the processing module are a simulation error dominant area search method based on a multi-level block decoupling algorithm.

[0039] A processing module executes a program instruction for the simulation error dominant region search method based on a multi-level block decoupling algorithm.

[0040] The present invention can achieve the following beneficial effects:

[0041] Compared with the prior art, the present invention considers the problem of network number confusion and difficulty in block partitioning before block decoupling, introduces the idea of ​​topological aggregation, including the adjacent numbered node quantity index and the matrix distance index, and constructs an objective function based on the weighted adjacency matrix to rearrange the matrix. Consider the line distance problem, and perform topological aggregation on the matrix on this basis. Consider the multi-level block decoupling problem, rather than performing block partitioning through a single decoupling. The primary multi-level block decoupling considers the node outgoing degree, and searches step by step from the node with the smallest outgoing degree. The secondary block decoupling considers the node outgoing degree of the sub-region that is less than or equal to 2, and then further blocks. Compared with traversing all nodes for hybrid dynamic simulation, the method proposed in the present invention can achieve efficient search for difference-causing areas, quickly locate error components, and reduce computational costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0043] Figure 1 It is a typical radial distribution network topology structure of the present invention;

[0044] Figure 2 is the adjacency matrix before and after the network topology aggregation of the present invention;

[0045] Figure 3 is the adjacency matrix before and after the network topology aggregation of the present invention;

[0046] Figure 4 This is a diagram of the initial network topology block decoupling process of the present invention;

[0047] Figure 5 This is a multi-level block decoupling error tracing flow chart of the present invention;

[0048] Figure 6 This is a system diagram of a park in Conghua according to the present invention;

[0049] Figure 7 This is a system diagram after decoupling of the present invention. DETAILED DESCRIPTION

[0050] The distribution network can be regarded as a main feeder with multiple branches, and multiple sub-branches are branched out from the branch line. Therefore, the topological structure of the distribution network can be decoupled into several single branches with disconnected edges, and the topological structure of the distribution network is stored hierarchically according to each single branch. The busbar nodes in the power system are abstracted as points in the graph, all nodes are naturally numbered, and the transmission lines are abstracted as edges. The system adjacency matrix A is established, and its expression is as follows:

[0051] ;

[0052] In the formula, a ij When the element is 0, it means that nodes i and j are not directly connected; ij When it is 1, it means that nodes i and j are directly connected.

[0053] Since the initial numbering of nodes does not take into account the mutual connection between nodes, the elements in the corresponding adjacency matrix are often distributed in a disorderly manner, and the power grid cannot be divided into blocks and partitions. In order to more clearly show the relationship between nodes, the present invention proposes the topological aggregation idea of ​​the adjacency matrix: by exchanging the rows and columns of the adjacency matrix, the non-zero elements in the adjacency matrix are concentrated as much as possible near the diagonal.

[0054] Specifically, based on the research of the prior art, the present invention proposes a simulation error dominant region search method based on a multi-level block decoupling algorithm, the method is:

[0055] 1. First, we generate a weighted adjacency matrix. This matrix introduces edge weight information based on the original adjacency matrix. These weights represent the influence of the line distance between two adjacent nodes. The longer the line, the lower the weight. The symbol W represents the weighted adjacency matrix, and its expression is as follows.

[0056] ;

[0057] ;

[0058] in, is the line distance between nodes i and j. If two nodes are not directly connected, the value is 0. and are the shortest and maximum distances between two adjacent nodes, respectively.

[0059] 2. Define the matrix distance index D and the adjacent numbered node number index N. The expression is as follows:

[0060] ;

[0061] ;

[0062] Among them, among them, is a sign function. If i and j differ by only 1, then , otherwise the value is equal to 0; similarly, if If not zero, , otherwise the value is equal to 0. The smaller the D value, the more the elements in the adjacency matrix are clustered on the main diagonal, and the larger the N value, the higher the degree of connection between the elements in the adjacency matrix.

[0063] In order to centralize the numbers of nodes that are closely connected in the network topology for easy block decoupling, the topological aggregation should follow the following principles: the number of adjacent numbered nodes index N should be as large as possible, and the matrix distance index D should be as small as possible. The main operation of topological aggregation is to exchange the rows and columns of the adjacency matrix. Since the adjacency matrix is ​​a symmetric square matrix, the exchange of rows and columns should be consistent.

[0064] 3. Use the simulated annealing algorithm to solve the new adjacency matrix and establish the following objective function:

[0065] ;

[0066] in is the weight, is the minimum matrix distance, is the maximum number of adjacent numbered nodes.

[0067] by Figure 1 Take the distribution network shown in the figure as an example. There are ten nodes in the network. The color blocks represent the adjacency matrix. The blue blocks are As shown in the figure, it can be seen that the initial node numbers are scattered and difficult to decouple. The adjacency matrix is ​​reconstructed using the simulated annealing algorithm to obtain Figure 2 The new adjacency matrix shown is Figure 3 This is the topology diagram rearranged after simulated annealing. It can be seen that after topological aggregation, the nodes are aggregated to the main diagonal, creating conditions for block decoupling of hybrid dynamic simulation.

[0068] Decoupling simulation requires the power grid to be segmented according to certain principles. The reasonable segmentation of the power grid will directly affect the efficiency of error tracing. Therefore, the present invention proposes a multi-level block decoupling algorithm based on network topology, which gradually decomposes the original power grid into multiple subnets for hybrid dynamic simulation and quickly locks the error area.

[0069] (1) Phase 1: Initial network topology block decoupling

[0070] According to the adjacency matrix after topological aggregation, block decoupling is performed, and nodes with a degree of 1 in the adjacency matrix are searched and marked as terminal nodes. Starting from the terminal node V with the largest natural number, based on the adjacency matrix, nodes connected to it with a degree less than m are searched from the node, and added to the sub-region set RV. Then, nodes connected to it with a degree less than m are continued to be searched from the nodes added to the set, and continue to be added to RV until the set RV is searched to include all nodes connected to the terminal node and with a degree less than m. Remove the nodes in this set (not participating in the next search) and regenerate the adjacency matrix. Repeat the above steps until there are no nodes with a degree of 1 in the adjacency matrix. In this way, the distribution network can be divided into multiple subnets based on the network topology.

[0071] In actual power systems, the average node outgoing degree is generally 3-5. Different block degrees m will directly affect the effect of system block division. Too small will lead to too many blocks and too much calculation of hybrid dynamic simulation; too large will lead to too few blocks and it will be difficult to quickly locate the source of error. Therefore, the selection of block degree m should satisfy: 2<m<M, where M is the maximum outgoing degree.

[0072] (2) Second stage: multi-level network topology block decoupling

[0073] After the initial network topology block decoupling algorithm is used to decompose the power grid into multiple sub-areas, the measured data is injected at the boundary of each sub-area to perform hybrid dynamic simulation on each sub-area. When a sub-area with a large number of nodes is identified as an error area, it is necessary to further decouple the sub-area in order to find the difference-causing element as soon as possible. At this time, the maximum node degree of each sub-area does not exceed m: if m>2, the network topology block decoupling algorithm of the first stage is repeated, and then the hybrid dynamic simulation is performed on each decoupled sub-area; if m=2, the following dichotomy is used to gradually decouple the error sub-areas to narrow the search area.

[0074] Binary block decoupling algorithm: Count the number of nodes in the error sub-area. If the number of nodes is an even number (2n), divide the sub-area into two small areas, each with n ordered nodes. If the number of nodes is an odd number (2n-1), divide the sub-area into two ordered sub-areas of n and n-1. Here, the maximum node number is placed in the n-1 sub-area.

[0075] Taking the rearranged adjacency matrix above as an example, take m=3 and perform network topology block decoupling:

[0076] Start the search from node 10, whose node matrix L10=[0 0 0 0 0 0 0 0 1 0],

[0077] Nodes 10 and 9 are connected, the degree of node 9 is 3, and R1 = {10}; remove node 10 and regenerate the node adjacency matrix as follows Figure 3 A2.

[0078] Start the search from node 8 (node ​​8 is the terminal node with the largest number). Its node matrix L9=[0 0 0 0 0 00 0 1]. Nodes 8 and 9 are connected. The degree of node 9 is 2, and R2={8, 9}. Remove nodes 8 and 9 and regenerate the node adjacency matrix as shown in A3 in the figure below.

[0079] Repeat the search until all nodes are partitioned.

[0080] The block decoupling algorithm based on power grid topology information will Figure 1 The power grid is initially decomposed into four sub-areas, namely R1={10}, R2={8,9}, R3={5,6,7} and R4={1,2,3,4}.

[0081] This can greatly reduce the number of hybrid dynamic simulations and improve the speed of searching the error-dominated region. The initial decoupling diagram is shown in Figure 4 When the scale of the power grid is large, m is set according to the actual power grid demand to divide the power grid into a suitable number of sub-areas as soon as possible.

[0082] for Figure 4 After the initial decoupling, if the error element is identified in region R4, where there is an even number of nodes in R4, multi-level network topology block decoupling can be performed to divide R4 into two sub-regions R5_1 with 2 nodes and R5_2 with 2 nodes, where R5_1={1,2}, R5_2={3,4}. Then, the hybrid dynamic simulation is continued until the error element is found.

[0083] The above-mentioned multi-level block decoupling algorithm based on network topology can gradually decompose the power system into multiple sub-areas. On this basis, hybrid dynamic simulation of the sub-areas can greatly reduce the number of hybrid dynamic simulations, thereby greatly improving the search efficiency. The overall process is as follows: Figure 5 shown.

[0084] Embodiment 1:

[0085] Let's take a distribution network application example to illustrate:

[0086] Figure 6 This is a system in a certain park in Conghua. The system is a medium-voltage distribution network system with a rated voltage of 10kV. There are 7 feeders and 46 nodes. Each node is equipped with a PMU device. First, the adjacency matrix is ​​reconstructed using the simulated annealing algorithm to generate a new node number; here m=4 is taken, and the multi-level decoupling algorithm based on network topology proposed in this invention is used to decouple the power grid into small systems, as follows Figure 7 As shown, the system is initially divided into 10 subsystems.

[0087] After decomposing the power grid into the 10 subsystems, the measured data are injected into the boundaries of each subsystem using the voltage source method to perform the first round of hybrid dynamic simulation, and the credibility of each sub-area is shown in Table 1.

[0088] Table 1

[0089]

[0090] As shown in Table 1, the minimum and average credibility in sub-region 10 are both less than 0.7, which is lower than the threshold set by the present invention. Region 10 contains only one node, so it can be determined that node 46 belongs to the error element. The dynamic simulation results in sub-region 10 are as follows: Figure 2 In addition, the average reliability of sub-region 5 is lower than 0.8, which is also lower than the threshold value set by the present invention, so it is determined that there are error elements in sub-region 5 as well.

[0091] There are 3 nodes in sub-area 5, so the second stage of block decoupling is required. According to the binary decoupling algorithm proposed in the present invention, sub-area 5 will be divided into two sub-areas, and the measured data will be injected from the boundary nodes of each sub-area to perform the second round of hybrid dynamic simulation. The credibility calculation results of the second round of hybrid dynamic simulation are shown in Table 2.

[0092] Table 2

[0093]

[0094] As shown in Table 2, the credibility of node 23 in sub-region 5_2 is higher than the threshold, and node 23 can be considered as an error-free node. The average credibility in sub-region 5_1 is lower than the threshold, and it is considered that there are error elements in this sub-region. Therefore, this sub-region will continue to decouple in blocks and perform a third round of dynamic simulation. The credibility calculation results of the third round of hybrid dynamic simulation are shown in Table 3 below.

[0095] Table 3

[0096]

[0097] As shown in Table 3, the average credibility in sub-region 5_1_2 is lower than 0.8, which means that there is an error element. Therefore, node 22 is identified as an error element and node 21 is identified as a non-error element. The dynamic simulation result of node 22 in sub-region 5_1_2 is shown in the figure. At this point, all nodes have been traversed and the search process is over.

[0098] It can be seen that the difference of the method proposed in the present invention is that it further explores the effect of the system itself (transmission line length, node density) on the system topology, and uses the simulated annealing algorithm to eliminate the impact of the unreasonable numbering of the original nodes, and uses a multi-level rather than a single decoupling system to partition and block the system, and further divide the subsystem into blocks before each hybrid dynamic simulation, which greatly reduces the error component tracing time. Compared with traversing all nodes for hybrid dynamic simulation, the method proposed in the present invention has higher timeliness.

[0099] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A simulation error dominant region search method based on a multi-level block decoupling algorithm, characterized by: S1. Using the primary block decoupling algorithm based on network topology, the blocks are divided according to the node outgoing line degree, and the original power grid is decomposed into multiple subnets according to the aggregated adjacency matrix; S2. Hybrid dynamic simulation is performed by injecting measured PMU data at each subnet boundary point to reduce the error range. Then, the secondary block decoupling algorithm is used to further reduce the error area, and hybrid dynamic simulation is performed on smaller sub-areas to find the error element. The S1 sub-steps are as follows: S1.

1. Search for nodes with a degree of 1 in the adjacency matrix, mark them as terminal nodes, start from the terminal node V with the largest natural number, search for nodes connected to the terminal node V with a degree less than m based on the adjacency matrix, where m represents any degree greater than 1, and add them to the sub-region set RV, then continue to search for nodes connected to the nodes added to the set with a degree less than m, and continue to add them to RV until the search set RV contains all nodes connected to the terminal node and with a degree less than m; S1.2, remove the nodes in the set RV, do not participate in the next search, and regenerate the adjacency matrix; S1.3, repeat steps S1.1-S1.2 until there is no node with degree 1 in the adjacency matrix; that is, the distribution network can be divided into multiple subnets based on the network topology structure; The method of the secondary block decoupling algorithm is as follows: S2.1 Count the number of nodes in the error sub-region; S2.2 If the number of nodes is an even number, i.e. 2n, then the sub-region is divided into two sub-regions, each of which contains n nodes in order; S2.3 If the number of nodes is an odd number, i.e. 2n-1, the sub-region is divided into two ordered sub-regions of n and n-1 respectively; S2.4 sets the maximum node number to be placed in n-1 sub-areas; hybrid dynamic simulation is performed on smaller sub-areas to find the error element.

2. The method for searching for a simulation error-dominated region based on a multi-level block decoupling algorithm according to claim 1, characterized in that: The adjacency matrix is ​​constructed as follows: 1) Generate a weighted adjacency matrix; 2) Define the matrix distance index D and the adjacent numbered node number index N; 3) Use the simulated annealing algorithm to solve the new adjacency matrix.

3. The method for searching for a simulation error-dominated region based on a multi-level block decoupling algorithm according to claim 2, characterized in that: 1) The process of generating a weighted adjacency matrix is ​​as follows: The weighted adjacency matrix introduces edge weight information based on the original adjacency matrix. These weights represent the influence of the line distance between two adjacent nodes. The longer the line, the lower the weight. The symbol W represents the weighted adjacency matrix, and its expression is as follows: ； ； in, is the line distance between node i and node j. If the two nodes are not directly connected, the value is 0. and are the shortest distance and maximum distance between two adjacent nodes respectively; represents the weighted adjacency matrix; w ij Represents an element in the weighted adjacency matrix. When the value is 0, it means that nodes i and j are not directly connected. When the value is not 0, it reflects the influence of the line distance between two adjacent nodes.

4. The method for searching for a simulation error-dominated region based on a multi-level block decoupling algorithm according to claim 2, characterized in that: 2) The method of defining the matrix distance index D and the adjacent numbered node number index N is as follows; The expression is as follows: ； ； Among them, among them, is a sign function. If i and j differ by only 1, then , otherwise the value is equal to 0; similarly, if If not zero, , otherwise the value is equal to 0; the smaller the D value, the more the elements in the adjacency matrix are concentrated on the main diagonal, and the larger the N value, the higher the degree of connection between the elements in the adjacency matrix.

5. The method for searching for a simulation error-dominated region based on a multi-level block decoupling algorithm according to claim 2, characterized in that: 3) The process of using the simulated annealing algorithm to solve the new adjacency matrix is ​​as follows: The following objective function is established: ； in is the weight, is the minimum matrix distance, is the maximum number of adjacent numbered nodes.

6. The method for searching for a simulation error-dominated region based on a multi-level block decoupling algorithm according to claim 4, characterized in that: In order to centralize the numbers of nodes that are closely connected in the network topology for easy block decoupling, the topology aggregation should comply with the following conditions: Condition 1: The number of adjacent numbered nodes, N, should be as large as possible, while the matrix distance, D, should be as small as possible; Condition 2: The main operation of topological aggregation is to exchange the rows and columns of the adjacency matrix. Since the adjacency matrix is ​​a symmetric square matrix, the exchange of rows and columns should be consistent.

7. An error tracing system based on network topology aggregation and multi-level block decoupling, comprising a processing module, characterized in that: The program instructions executed by the processing module are a simulation error dominant region search method based on a multi-level block decoupling algorithm according to claim 1.

8. A processing module, characterized in that: The executed program instructions are a simulation error dominant area search method based on a multi-level block decoupling algorithm according to claim 1.

Images