A power grid dynamic double-layer partition optimization method considering regional voltage control capability

Abstract

The application discloses a power grid dynamic double-layer partition optimization method considering regional voltage control capability, and relates to the field of power grid reactive power control. The steps are as follows: (1) data preparation. The current power grid operation condition is confirmed based on historical operation power data or real-time measured power data of the power grid. (2) calculation of an extended reactive voltage sensitivity matrix. The extended reactive voltage sensitivity between nodes of the power grid is calculated according to power data of the power grid. (3) first-layer partition based on modularity. The connection weight between nodes is calculated through the extended reactive voltage sensitivity matrix, a regional modularity function is constructed, and an improved K-means algorithm is adopted for clustering. (4) second-layer partition based on regional reactive power margin. A regional reactive power margin index is constructed according to the reactive power compensation capability of reactive power equipment, an improved K-means algorithm is adopted for clustering, and a plurality of partition results are obtained by repeating the process several times. (5) combination of double-layer partition. A partition similarity index is constructed, the partition result with the highest similarity to the result of step (3) in step (4) is taken, and the two partition results are integrated by adopting agglomerative clustering. (6) partition optimization. The region where the voltage out-of-limit node is located is judged, and the reactive power source in the region is dispatched for optimization. The application has the advantages that the reactive power margin and the reactive voltage sensitivity are integrated into the field of power grid partition, the reactive power margin of each region after power grid partition is reasonably distributed, the regional reactive voltage control capability is improved, and the power grid partition optimization efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of dynamic double-layer partition optimization method of power grid considering regional voltage control capability, belong to power grid reactive power control field. BACKGROUND

[0002] Power grid reactive power optimization is an important means to improve the security and stability of power system and economy, and reasonable allocation of reactive power can reduce the active power loss of power grid, reduce the cost of reactive power dispatching, improve system stability and improve voltage quality. With the large-scale access of renewable energy and distributed energy and the increasing demand for electricity, the management of voltage and reactive power becomes particularly important. Power grid partition optimization is the basis for fine scheduling of reactive power equipment, and is crucial for ensuring stable operation of power grid and improving power quality.

[0003] The present application is partitioned from the reactive voltage sensitivity between nodes and the partition reactive margin under the existing power grid framework, and the two partition results are integrated, and the power grid is optimized based on this, to realize fine scheduling of reactive power equipment. It has important significance for ensuring stable operation of power grid and improving power quality. SUMMARY

[0004] The technical problem to be solved by the present application is to overcome the technical defects of the prior art, and a kind of dynamic double-layer partition optimization method of power grid considering regional voltage control capability is proposed. The steps include:

[0005] (1) Data preparation. Obtain the reactive power compensation capability data of power grid reactive power equipment, and monitor whether the node voltage of power grid is out of limit, if yes, confirm the current power grid operating power based on historical operating power data or real-time measured power data, and jump to step (2), otherwise continue to monitor the node voltage.

[0006] (2) Calculate the extended reactive voltage sensitivity matrix. According to the power data of power grid, the reactive voltage sensitivity between each PQ node of power grid is calculated, then the corresponding reactive voltage sensitivity is calculated by taking each PV node as a PQ node, and then the extended reactive voltage sensitivity matrix containing all nodes except the balance node is obtained.

[0007] (3) First layer partition based on modularity. The connection weight between nodes is calculated by the extended reactive voltage sensitivity matrix, the regional modularity function is constructed, and the improved K-means algorithm is used for power grid partition based on modularity, and the steps are as follows:

[0008] ① Select initial partition point. The number of partition targets α is determined by experts. α nodes are randomly selected, and the sum of the connection weight between nodes is calculated as the basis for judgment. Random selection is carried out for several times, and the set of random nodes with the lowest connection weight is taken as the initial partition point.

[0009] ②Node partition iteration. Calculate the corresponding modularity when the remaining nodes are incorporated into different partitions, respectively, and record the partition with the maximum modularity.

[0010] ③Update the partition result. Update the node partition according to the partition result with the maximum modularity recorded in step ②. If the number of partition iterations t is less than half of the maximum iteration number T, jump to step ②, otherwise jump to step ④.

[0011] ④Randomly select β nodes, calculate the corresponding modularity when they are incorporated into different partitions, and record the partition with the maximum modularity. Repeat step ④ until the iteration number reaches the maximum value. The value of β is shown in formula (1):

[0012] β = ceil(1-t / T) (1)

[0013] Where ceil() is the ceiling function; t and T are the current iteration number and the maximum iteration number, respectively.

[0014] ⑤Output the partition result. Balance the nodes into the node partition with the largest connection weight.

[0015] (4)Sub-layer partition based on regional reactive power margin. Construct a regional reactive power margin index according to the reactive power compensation capacity of the reactive power equipment, and use the improved Kmeans algorithm for clustering combined with the topology structure of the power grid. Repeat several times to obtain multiple partition results.

[0016] (5)Combination of double-layer partition. Construct a partition similarity index, calculate the similarity of the partition results in steps (3) and (4), take the partition result in step (4) with the highest similarity to the result in step (3), construct a node partition similarity matrix, and integrate the two partition results using agglomerative clustering.

[0017] (6)Partition optimization. Determine the area where the voltage out-of-limit node is located, and dispatch the regional reactive power source for optimization.

[0018] Further, the confirmation of the current power grid operating power in step (1). Since the partition of the power grid in the present application is based on the current operating state of the system, the partition results obtained under different operating states will be different, so it is necessary to obtain the current operating power parameters of the power grid. The power parameter determination method is as follows:

[0019] If historical operating data is used, the daily power data is divided into one-hour time periods, the past 30-day historical power data of the current time period is taken, and is processed according to the following formulas (2) and (3).

[0020]

[0021] Wherein: P(t) and Q(t) are the power data obtained by processing, and are the power of the power grid at t period; And are the average active power of the power grid at t period in the past week and the past month, respectively; And are the average reactive power of the power grid at t period in the past week and the past month, respectively; and α1 and α2 are weight coefficients, α1=0.7 and α2=0.3.

[0022] The average value is taken as the power of the power grid at the current period; if real-time measured power data is used, the average value of the power data in the past 30 minutes is taken as the power data of the electric appliance.

[0023] Further, the calculation of the extended reactive voltage sensitivity matrix in step (2). The sensitivity matrix can intuitively reflect the coupling relationship between nodes, and is an important basis for the power grid partitioning of the application. The calculation method of the extended reactive voltage sensitivity matrix is as follows:

[0024] First, the reactive voltage sensitivity matrix of the PQ node is calculated. Newton method is used to calculate the power flow distribution of the power grid, and the power correction equation in the last iteration process is taken, and its expression is shown in formula (4):

[0025]

[0026] Wherein: J is the Jacobian matrix, which can indicate the partial derivative relationship between the power and the node voltage; ΔP and ΔQ represent the deviation matrix of the node injected active power and reactive power, respectively; Δθ and ΔV represent the change matrix of the node voltage phase angle and voltage amplitude, respectively.

[0027] The sensitivity matrix is obtained by inverting the above formula:

[0028]

[0029] Wherein: S is the sensitivity matrix; S PV and S QV are the active voltage and reactive voltage sensitivity factors, respectively, which are used to indicate that the change of the power of the system node causes the change of the node voltage; S Pθ and S Qθ reflect the change of the voltage phase angle caused by the change of the power of the system node.

[0030] The relationship between the change amount of the node voltage ΔV and the change amounts of the active power ΔP and the reactive power ΔQ of the power grid can be expressed as:

[0031] ΔV=S PV ·ΔP+S QV ·ΔQ (6)

[0032] Among them: ΔQ=[ΔQ1, ΔQ2,...ΔQ n ] T ,ΔP=[ΔP1,ΔP2,...ΔP n ] T

[0033] Ignoring the effect of active power ΔP on voltage change in formula (6), the reactive power voltage sensitivity matrix S of node PQ is obtained:

[0034] ΔV·ΔQ -1 =S QV =S (7)

[0035] Calculate the reactive voltage sensitivity matrix of the PV node. PV nodes have the ability to maintain their own voltage; therefore, a generalized reactive voltage sensitivity is defined to reflect the coupling characteristics between the PV node and other nodes. The generalized reactive voltage sensitivity is defined as follows: For a given PV node, if it is replaced by a PQ node with the same electrical characteristics, the voltage change at that node caused by the change in reactive power of the other nodes in the power grid is calculated. The calculation process is as follows:

[0036] Treating the PV nodes for which reactive voltage sensitivity needs to be calculated as PQ nodes, the reactive voltage sensitivity of the entire network is calculated using the same method described above for calculating the reactive voltage sensitivity of PQ nodes. The remaining PV nodes are then treated as PQ nodes, and their corresponding reactive voltage sensitivities are calculated accordingly.

[0037] By integrating the reactive voltage sensitivity of the PQ node with that of the PV node, an extended reactive voltage sensitivity matrix is ​​obtained:

[0038]

[0039] Where: m and n are the number of PQ nodes and PV nodes, respectively; matrix S m×m medium element S ij Meaning the reactive voltage sensitivity of node i to node j of PQ; matrix M m×(n-m-1) Element M ij Meaning the reactive voltage sensitivity of PQ node i to PV node j; matrix N (n-m-1)×m element N ij This refers to the reactive voltage sensitivity of PV node i to PQ node j; matrix Y (n-m-1)×(n-m-1) element Y ij This refers to the reactive voltage sensitivity of PV node i to PV node j.

[0040] Furthermore, the modularity function is constructed in step (3). The modularity function is defined as follows:

[0041]

[0042] Where: S′ ij With S′ ji For the two elements in the i-th row and j-th column and the j-th row and i-th column of the extended reactive voltage sensitivity matrix; A ij δ(i,j) represents the connection weight between node i and node j; σ represents the sum of the weights of all edges in the network; if node i and node j are in the same partition, then δ(i,j) = 1, otherwise δ(i,j) = 0.

[0043] Furthermore, the definition of the regional reactive power margin index η in step (4) is as follows. The conventional definition of reactive power margin as the ratio of reactive load to maximum reactive power compensation has limitations, ignoring the different numbers of nodes in different zones, resulting in different reactive power margin weights. This invention proposes a regional reactive power margin index η, defined as follows:

[0044]

[0045] Where: η i ∑Q represents the reactive power margin of region i. Li With ∑Q i These represent the reactive load and maximum reactive power compensation for region i, respectively; n i ρ represents the number of nodes in region i; ρ is the penalty coefficient, which takes a large positive value.

[0046] Furthermore, step (4) employs an improved K-means-based grid partitioning method based on regional reactive power margin. The reactive power margin data distribution of nodes differs from reactive power voltage sensitivity; to prevent regional connectivity issues in the partitioning results, the partition selection part of the algorithm iteration process needs improvement. The steps for grid partitioning based on regional reactive power margin using the improved K-means method are as follows:

[0047] ① Construct the connection matrix. Based on the power grid topology diagram, construct the connection matrix C.

[0048] C ij =φ(i,j) (14)

[0049] Where: C ij Let φ(i,j) be the element in the i-th row and j-th column of the connection matrix C; φ(i,j) indicates whether node i and node j are directly connected. If they are connected, then φ(i,j) = 1, otherwise φ(i,j) = 0.

[0050] ② Select the initial partition point. Use the initial partition from "Modularity-based Partitioning". If the number of nodes with no active power source ε is greater than the number of partitions α, proceed to step ③; otherwise, terminate the partitioning operation.

[0051] ③ Node partitioning iteration. Based on the connection matrix C, record the reactive power margin index η of the region when each node is merged into an adjacent partition, and record the partition with the largest modularity.

[0052] ④ Update the partitioning results. Update the node partitioning information according to the maximum regional reactive power margin index η recorded in step ③. If the number of partitioning iterations t is less than half of the maximum number of iterations T, jump to step ③; otherwise, jump to step ⑤.

[0053] ⑤ Randomly select β nodes and calculate their modularity when they are incorporated into different partitions. Record the partition with the largest regional reactive power margin index η. Repeat step ⑤ until the number of iterations reaches the maximum value. The value of β is shown in formula (15):

[0054] β=ceil(1-t / T) (15) where: ceil() is the floor function; t and T are the current iteration number and the maximum iteration number, respectively.

[0055] ⑥ Repeat steps ② to ⑤ several times to obtain multiple partition results.

[0056] Furthermore, the integration of the two-level partitioning results in step (5) is as follows: The similarity of partitioning results is defined by the partitions of nodes in different partitioning results, and the similarity between nodes is defined by the partitioning results of multiple nodes within the same partition. The integration steps of the two-level partitioning results are as follows:

[0057] ① Select the integration base. Define a partitioning result similarity index. From the secondary partitions "grid partitions based on regional reactive power margin," select the partitioning result with the highest similarity to the primary partition "grid partitions based on modularity," and use this as the integration base. The partitioning similarity index d(p) i ,p j The definition is as follows:

[0058]

[0059] Among them: [p] ia ] indicates the partitioning result p i The set of nodes in the partition to which node a belongs; [p ja ] indicates the partitioning result p j The set of nodes in the partition to which node a belongs; |[p ia ]∩[p ja ]| represents partition p i The partition to which node a belongs and partition p j The number of intersections of the partitions to which node a belongs; |[p ia ]∪[p ja ]| represents partition p i The partition to which node a belongs and partition p j The number of the unions of the partitions to which node a belongs.

[0060] ② Construct the node similarity matrix. To determine whether node i and node j belong to the same region in different partitioning results, define the node similarity matrix Sim:

[0061]

[0062] Among them: [p] ai ] and [p aj ] represents the set of partitions to which nodes i and j belong in partition a; χ η (i,j) indicates whether node i and node j belong to the same region in the partition; k is the number of integration bases used; s ij The similarity between node i and node j in the partitioning of the ensemble is expressed, with a value ranging from [0, 1]. Here, node similarity is defined by whether different nodes belong to the same partition in multiple partitioning results, which differs from the similarity in step ① above.

[0063] ③ Agglomerative Clustering. Using the elements of each row of the similarity matrix Sim as the attributes of the corresponding nodes, and the number of partition targets α as the termination condition, agglomerative clustering is used to cluster the similarity matrix between nodes.

[0064] Furthermore, the partitioning optimization in step (6) is as follows:

[0065] ① Data Reading. Obtain power grid operating parameters and check for any node voltage exceeding limits. If found, proceed to step ②. ② Determine the primary partition to which the voltage-exceeding node belongs. Schedule reactive power equipment within the primary partition for optimization simulation. If the voltage exceeding limit problem is successfully resolved, end the optimization and output the optimization solution; otherwise, proceed to step ③.

[0066] ③ Traverse the node voltages in the neighboring partitions of the main partition, select the partition to which the node with the highest voltage belongs and merge it into the main partition. Schedule the reactive power sources in the main partition for reactive power optimization simulation. If the voltage over-limit problem is successfully resolved, end the optimization and output the optimization scheme; otherwise, repeat step ③. If all reactive power sources have participated in scheduling and there are still nodes with voltage over-limit, output the current control scheme and issue a warning. Attached Figure Description

[0067] Figure 1 Overall Flowchart

[0068] Figure 2 Improved K-means partitioning process based on modularity

[0069] Figure 3 Improved K-means partitioning process based on regional reactive power margin index

[0070] Figure 4 Partitioning Results Integration Process

[0071] Figure 5 Partition optimization process Detailed Implementation

[0072] The present invention will be further described in conjunction with the given embodiments, but is not limited thereto. A dynamic two-level zoning optimization method for a power grid considering regional voltage control capability, the process of which is as follows: Figure 1 As shown, it includes the following steps:

[0073] (1) Data preparation. Obtain reactive power compensation capacity data of the power grid reactive equipment, and at the same time monitor whether the voltage of the power grid nodes exceeds the limit. If so, confirm the current power grid operating status based on the historical operating power data or real-time measured power data, and jump to step (2). Otherwise, continue to monitor the node voltage.

[0074] If historical operating data is used, the daily power data is divided into one hour and one time period. The historical power data of the past 30 days for the current time period is taken and processed according to the following formulas (1) and (2).

[0075]

[0076] Where: P(t) and Q(t) are the processed power data, which are considered as the power of the power grid during time period t; and These are the average active power of the power grid during time period t over the past week and the past month, respectively. and α1 represents the average reactive power of the power grid during time period t over the past week and month; α2 are weighting coefficients, α1 = 0.7 and α2 = 0.3.

[0077] The average value is taken as the power grid power for the current period; if real-time measured power data is used, the average value of the power data over the past 30 minutes is taken as the power grid power data for electrical appliances.

[0078] (2) Calculate the extended reactive voltage sensitivity matrix. Based on the power grid data, calculate the reactive voltage sensitivity between each PQ node of the power grid.

[0079] First, the reactive voltage sensitivity matrix of node PQ is calculated. The power flow distribution of the power grid is calculated using Newton's method, and the power correction equation from the last iteration is expressed as shown in equation (3):

[0080]

[0081] Where: J is the Jacobian matrix, which can show the partial derivative relationship between power and node voltage; ΔP and ΔQ represent the deviation matrices of injected active power and reactive power, respectively; Δθ and ΔV represent the variation matrices of node voltage phase angle and voltage amplitude, respectively.

[0082] Inverting the above equation yields the sensitivity matrix:

[0083]

[0084] Where: S is the sensitivity matrix; S PV With S QV These represent the active voltage and reactive voltage sensitivity factors, respectively, used to illustrate how changes in power at system nodes lead to changes in node voltage; S Pθ With S Qθ This reflects the change in voltage phase angle caused by changes in the power of system nodes.

[0085] The relationship between the voltage change ΔV at grid nodes and the changes in active and reactive power ΔP and ΔQ can be expressed as:

[0086] ΔV=S PV ·ΔP+S QV ·ΔQ (5)

[0087] Among them: ΔQ=[ΔQ1, ΔQ2,...ΔQ n ] T ,ΔP=[ΔP1,ΔP2,...ΔP n ] T

[0088] Ignoring the effect of active power ΔP on voltage change in formula (5), the reactive power voltage sensitivity matrix S of node PQ is obtained:

[0089] ΔV·ΔQ -1 =S QV =S (6)

[0090] Then, each PV node is treated as a PQ node, and the corresponding reactive voltage sensitivity is calculated, thus obtaining an extended reactive voltage sensitivity matrix that includes all nodes except the slack node. Each PV node is still treated as a PV node in the sensitivity calculation of the other PV nodes.

[0091] By integrating the reactive voltage sensitivity of the PQ node with that of the PV node, an extended reactive voltage sensitivity matrix is ​​obtained:

[0092]

[0093] Where: m and n are the number of PQ nodes and PV nodes, respectively; matrix S m×m medium element S ij Meaning the reactive voltage sensitivity of node i to node j of PQ; matrix M m×(n-m-1) Element M ij Meaning the reactive voltage sensitivity of PQ node i to PV node j; matrix N (n-m-1)×m element Nij This refers to the reactive voltage sensitivity of PV node i to PQ node j; matrix Y (n-m-1)×(n-m-1) element Y ij This refers to the reactive voltage sensitivity of PV node i to PV node j.

[0094] (3) First-level partitioning based on modularity. The connection weight A between nodes is calculated using an extended reactive voltage sensitivity matrix. ij :

[0095]

[0096] Where: S′ ij With S′ ji For the two elements in the i-th row and j-th column and the j-th row and i-th column of the extended reactive voltage sensitivity matrix; A ij The connection weight between node i and node j.

[0097] The regional modularity function θ is constructed and defined as follows:

[0098]

[0099] Where: σ represents the sum of the weights of all edges in the network; if node i and node j are in the same partition, then δ(i,j) = 1, otherwise δ(i,j) = 0.

[0100] Clustering is performed using an improved K-means algorithm, the process is as follows: Figure 2 As shown, the steps are as follows:

[0101] ① Select initial partitioning points. The number of partitioning targets, α, is determined under expert guidance. α nodes are randomly selected, and the sum of their connection weights is used as the criterion. After several rounds of random selection, the group of random nodes with the lowest sum of connection weights is selected as the initial partitioning points.

[0102] ② Node partitioning iteration. Calculate the modularity of the remaining nodes when they are merged into different partitions, and record the partition with the highest modularity.

[0103] ③ Update the partitioning results. Based on the partitioning results with the highest modularity recorded in step ②, update the node partitioning information. If the number of partitioning iterations t is less than half of the maximum number of iterations T, jump to step ②; otherwise, jump to step ④.

[0104] ④ Randomly select β nodes and calculate their modularity when merged into different partitions. Record the partition with the highest modularity. Repeat step ④ until the number of iterations reaches the maximum value. The value of β is shown in formula (12):

[0105] β=ceil(1-t / T) (12)

[0106] Where: ceil() is the floor function; t and T are the current iteration number and the maximum iteration number, respectively.

[0107] ⑤ Output the partitioning results. Balanced nodes are assigned to the partition of the node with the highest connection weight.

[0108] (4) Sub-level partitioning based on regional reactive power margin. Based on the reactive power compensation capability of reactive power equipment, a regional reactive power margin index η is constructed:

[0109]

[0110] Where: η i ∑Q represents the reactive power margin of region i. Li With ∑Q i These represent the reactive load and maximum reactive power compensation for region i, respectively; n i ρ represents the number of nodes in region i; ρ is the penalty coefficient, which takes a large positive value.

[0111] Based on the power grid topology, an improved K-means algorithm is used for clustering, as follows: Figure 3 As shown, the steps are as follows:

[0112] ① Construct the connection matrix. Based on the power grid topology diagram, construct the connection matrix C.

[0113] C ij =φ(i,j) (14)

[0114] Where: C ij Let φ(i,j) be the element in the i-th row and j-th column of the connection matrix C; φ(i,j) indicates whether node i and node j are directly connected. If they are connected, then φ(i,j) = 1, otherwise φ(i,j) = 0.

[0115] ② Select the initial partition point. Use the initial partition from "Modularity-based Partitioning". If the number of nodes with no active power source ε is greater than the number of partitions α, proceed to step ③; otherwise, terminate the partitioning operation.

[0116] ③ Node partitioning iteration. Based on the connection matrix C, record the reactive power margin index η of the region when each node is merged into an adjacent partition, and record the partition with the largest modularity.

[0117] ④ Update the partitioning results. Update the node partitioning information according to the maximum regional reactive power margin index η recorded in step ③. If the number of partitioning iterations t is less than half of the maximum number of iterations T, jump to step ③; otherwise, jump to step ⑤.

[0118] ⑤ Randomly select β nodes and calculate their modularity when they are incorporated into different partitions. Record the partition with the largest regional reactive power margin index η. Repeat step ⑤ until the number of iterations reaches the maximum value. The value of β is shown in formula (15):

[0119] β=ceil(1-t / T) (15)

[0120] Where: ceil() is the floor function; t and T are the current iteration number and the maximum iteration number, respectively.

[0121] ⑥ Repeat steps ② to ⑤ several times to obtain multiple partition results.

[0122] (5) Combination of two-level partitioning. The process is as follows: Figure 4 As shown, the partition similarity index d(p) is defined by the partitions of nodes in different partition results. i ,p j ):

[0123]

[0124] Among them: [p] ia ] indicates the partitioning result p i The set of nodes in the partition to which node a belongs; [p ja ] indicates the partitioning result p j The set of nodes in the partition to which node a belongs; |[p ia ]∩[p ja ]| represents partition p i The partition to which node a belongs and partition p j The number of intersections of the partitions to which node a belongs; |[p ia ]∪[p ja ]| represents partition p i The partition to which node a belongs and partition p j The number of the unions of the partitions to which node a belongs.

[0125] Calculate the similarity between the partitioning results in step (3) and step (4), and take the partitioning result in step (4) with the highest similarity to the result in step (3).

[0126] Construct a partition similarity matrix for each node to represent the similarity between different nodes in the same partition. The node similarity matrix Sim is defined as follows:

[0127]

[0128] Among them: [p] ai ] and [p aj ] represents the set of partitions to which nodes i and j belong in partition a; χ η(i,j) indicates whether node i and node j belong to the same region in the partition; k is the number of integration bases used; s ij The similarity between node i and node j in the partition of the integrated basis is reflected, with a value range of [0, 1].

[0129] Using the elements of each row of the similarity matrix Sim as the attributes of the corresponding nodes, and the number of partition targets α as the termination condition, agglomerative clustering is used to cluster the similarity matrix between nodes.

[0130] (6) Partition optimization. The process is as follows: Figure 5 As shown, to determine the region where the voltage over-limit node is located, optimize the reactive power sources within the dispatch area. The steps are as follows:

[0131] ① Data Reading. Obtain power grid operating parameters and check for any node voltage exceeding limits. If any are found, proceed to step ②.

[0132] ② Determine the main partition to which the voltage over-limit node belongs, schedule the reactive power equipment in the main partition for optimization simulation. If the voltage over-limit problem is successfully resolved, end the optimization and output the optimization solution; otherwise, jump to step ③.

[0133] ③ Traverse the node voltages in the neighboring partitions of the main partition, select the partition to which the node with the highest voltage belongs and merge it into the main partition. Schedule the reactive power sources in the main partition for reactive power optimization simulation. If the voltage over-limit problem is successfully resolved, end the optimization and output the optimization scheme; otherwise, repeat step ③. If all reactive power sources have participated in scheduling and there are still nodes with voltage over-limit, output the current control scheme and issue a warning.

Claims

1. A dynamic two-layer zoning optimization method for power grids that takes into account regional voltage control capabilities, characterized in that, Includes the following steps: (1) Data preparation: Obtain data on the reactive power compensation capability of the power grid reactive equipment, and at the same time monitor whether the voltage of the power grid node exceeds the limit. If so, confirm the current power grid operating status based on the historical operating power data or real-time measured power data, and jump to step (2); otherwise, continue to monitor the node voltage. (2) Calculate the extended reactive voltage sensitivity matrix. Based on the power grid data, calculate the reactive voltage sensitivity between each PQ node in the power grid. Then, treat each PV node as a PQ node and calculate the corresponding reactive voltage sensitivity. This will result in an extended reactive voltage sensitivity matrix that includes all nodes except the balancing node. (3) Based on modularity, the first-level partitioning is performed by calculating the connection weights between nodes through the extended reactive voltage sensitivity matrix, constructing the regional modularity function, and using the improved K-means algorithm for modularity-based power grid partitioning. The steps are as follows: ① Select initial partitioning points. Under the guidance of experts, determine the number of partitioning targets α. Randomly select α nodes and calculate the sum of connection weights between nodes as the basis for judgment. Perform several random selections and take the group of random nodes with the lowest sum of connection weights as the initial partitioning points. ② Iterate through node partitioning, calculate the modularity of the remaining nodes when they are merged into different partitions, and record the partition with the highest modularity. ③ Update the partitioning results. Based on the partitioning results with the largest modularity recorded in step ②, update the node partitioning status. If the number of partitioning iterations t is less than half of the maximum number of iterations T, jump to step ②; otherwise, jump to step ④. ④ Randomly select β nodes, calculate their modularity when they are merged into different partitions, record the partition with the highest modularity, and repeat step ④ until the number of iterations reaches the maximum value. The value of β is shown in formula (1): β=ceil(1-t / T) (1) Where: ceil() is the floor function; t and T are the current iteration number and the maximum iteration number, respectively; ⑤ Output the partitioning results, and assign balanced nodes to the partition of the node with the largest connection weight; (4) Based on the secondary partitioning of the regional reactive power margin, the regional reactive power margin index is constructed according to the reactive power compensation capability of the reactive power equipment. Combined with the power grid topology, the improved Kmeans algorithm is used for clustering, and the results of multiple partitions are obtained by repeating the process several times. (5) Combine the two-level partitioning, construct the partition similarity index, calculate the similarity between the partitioning results in step (3) and step (4), take the partitioning result with the highest similarity to the result in step (3) in step (4), construct the partition similarity matrix of each node, and use agglomerative clustering to integrate the two partitioning results. (6) Zone optimization: Determine the area where the voltage over-limit node is located, and optimize the reactive power source in the scheduling area.

2. The dynamic two-layer zoning optimization method for power grids considering regional voltage control capability according to claim 1, characterized in that, Confirmation of the current power grid operating status in step (1); If historical operating data is used, the daily power data is divided into hourly time periods. The historical power data of the past 30 days for the current time period is taken and processed according to the following formulas (2) and (3): Where: P(t) and Q(t) are the processed power data, which are considered as the power of the power grid during time period t; and These are the average active power of the power grid during time period t over the past week and the past month, respectively. and α1 represents the average reactive power of the power grid during time period t over the past week and month; α2 are weighting coefficients, α1 = 0.7, α2 = 0.

3. The average value is taken as the power grid power for the current period; if real-time measured power data is used, the average value of the power data over the past 30 minutes is taken as the power grid power data for electrical appliances.

3. The dynamic two-layer zoning optimization method for power grids considering regional voltage control capability according to claim 1, characterized in that, The calculation of the extended reactive voltage sensitivity matrix in step (2); First, the reactive voltage sensitivity matrix of node PQ is calculated. Then, the power flow distribution of the power grid is calculated using the Newton-Raphson method. The power correction equation in the last iteration is taken, and its expression is shown in formula (3): Where: J is the Jacobian matrix, which can show the partial derivative relationship between power and node voltage; ΔP and ΔQ represent the deviation matrices of injected active power and reactive power, respectively; Δθ and ΔV represent the variation matrices of node voltage phase angle and voltage amplitude, respectively. Inverting the above equation yields the sensitivity matrix: Where: S is the sensitivity matrix; S PV With S QV These represent the active voltage and reactive voltage sensitivity factors, respectively, used to illustrate how changes in power at system nodes lead to changes in node voltage; S Pθ With S Qθ This reflects the change in voltage phase angle caused by changes in system node power; The relationship between the voltage change ΔV at grid nodes and the changes in active and reactive power ΔP and ΔQ can be expressed as: ΔV=S P ·ΔP+S QV ·ΔQ (6) Where: ΔQ = [ΔQ1, ΔQ2, ...ΔQ n ] T ,ΔP=[ΔP1,ΔP2,...ΔP n ] T ; Ignoring the effect of active power ΔP on voltage change in formula (6), the reactive power voltage sensitivity matrix S of node PQ is obtained: ΔV·ΔQ -1 =S QV =S (7) The reactive voltage sensitivity matrix of a PV node is calculated. PV nodes have the ability to maintain their own voltage; therefore, a generalized reactive voltage sensitivity is defined to reflect the coupling characteristics between the PV node and other nodes. The generalized reactive voltage sensitivity is defined as follows: For a given PV node, if it is replaced by a PQ node with the same electrical characteristics, the voltage change at that node caused by the change in reactive power of other nodes in the power grid is calculated as follows: Treat the PV nodes whose reactive voltage sensitivity needs to be calculated as PQ nodes, calculate the power flow of the entire network, and use the above method to calculate the reactive voltage sensitivity of PQ nodes to obtain the reactive voltage sensitivity of the node. Treat the remaining PV nodes as PQ nodes one by one to calculate the corresponding reactive voltage sensitivity. By integrating the reactive voltage sensitivity of the PQ node with that of the PV node, an extended reactive voltage sensitivity matrix is ​​obtained: Where: m and n are the number of PQ nodes and PV nodes, respectively; matrix S m×m medium element S ij Meaning the reactive voltage sensitivity of node i to node j of PQ; matrix M m×(n-m-1) Element M ij Meaning the reactive voltage sensitivity of PQ node i to PV node j; matrix N (n-m-1)×m element N ij This refers to the reactive voltage sensitivity of PV node i to PQ node j; matrix Y (n-m-1)×(n-m-1) element Y ij This refers to the reactive voltage sensitivity of PV node i to PV node j.

4. The dynamic two-layer zoning optimization method for power grids considering regional voltage control capability according to claim 1, characterized in that, The construction of the modularity function in step (3); The modularity function is applied in weighted networks to measure the community structure characteristics of complex networks and determine the optimal number of partitions. The definition of the modularity function is as follows: Where: S′ ij With S′ ji For the two elements in the i-th row and j-th column and the j-th row and i-th column of the extended reactive voltage sensitivity matrix; A ij δ(i,j) represents the connection weight between node i and node j; σ represents the sum of the weights of all edges in the network; if node i and node j are in the same partition, then δ(i,j) = 1, otherwise δ(i,j) = 0.

5. The dynamic two-layer zoning optimization method for power grids considering regional voltage control capability according to claim 1, characterized in that, The definition of the reactive power margin index η in step (4) is as follows: Where: η i ∑Q represents the reactive power margin of region i. Li With ∑Q i These represent the reactive load and maximum reactive power compensation for region i, respectively; n i ρ represents the number of nodes in region i; ρ is the penalty coefficient, which takes a large positive value.

6. The dynamic two-layer zoning optimization method for power grids considering regional voltage control capability according to claim 1, characterized in that, Step (4) adopts an improved K-means-based power grid partitioning method based on regional reactive power margin; The steps for improving the K-means method for grid partitioning based on regional reactive power margin are as follows: ① Construct the connection matrix: Based on the power grid topology diagram, construct the connection matrix C: C ij =φ(i,j) (14) Where: C ij Let φ(i,j) be the element in the i-th row and j-th column of the connection matrix C; φ(i,j) indicates whether node i and node j are directly connected. If they are connected, then φ(i,j) = 1, otherwise φ(i,j) = 0. ② Select the initial partition point and use the initial partition in "Modularity-based partitioning". If the number of nodes with no power source ε is greater than the number of partitions α, jump to step ③; otherwise, terminate the partitioning operation. ③ Node partitioning iteration: According to the connection matrix C, the reactive power margin index η of each node when it is merged into the adjacent partition is recorded, and the partition with the largest modularity is recorded. ④ Update the partitioning results. According to the maximum regional reactive power margin index η recorded in step ③, update the node partitioning status. If the number of partitioning iterations t is less than half of the maximum number of iterations T, jump to step ③; otherwise, jump to step ⑤. ⑤ Randomly select β nodes, calculate their modularity when they are incorporated into different partitions, and record the partition with the largest regional reactive power margin index η. Repeat step ⑤ until the number of iterations reaches the maximum value. The value of β is shown in formula (15): β=ceil(1-t / T) (15) Where: ceil() is the floor function; t and T are the current iteration number and the maximum iteration number, respectively; ⑥ Repeat steps ② to ⑤ several times to obtain multiple partition results.

7. The dynamic two-layer zoning optimization method for power grids considering regional voltage control capability according to claim 1, characterized in that, The integration of the two-level partitioning results in step (5); The steps for integrating the results of the two-level partitioning are as follows: ① Select the integration basis and define the partitioning result similarity index. From the secondary partition "grid partitioning based on regional reactive power margin" and the primary partition "grid partitioning based on modularity", select the partitioning result with the highest similarity to the primary partition "grid partitioning based on modularity" as the integration basis. The partitioning similarity index d(p i ,p j The definition is as follows: Among them: [p] ia ] indicates the partitioning result p i The set of nodes in the partition to which node a belongs; [p ja ] indicates the partitioning result p j The set of nodes in the partition to which node a belongs; |[p ia ]∩[p ja ]| represents partition p i The partition to which node a belongs and partition p j The number of intersections of the partitions to which node a belongs; |[p ia ]∪[p ja ]| represents partition p i The partition to which node a belongs and partition p j The number of partitions to which node a belongs; ② Construct a similarity matrix between nodes. Define the similarity matrix Sim to determine whether node i and node j belong to the same region in different partitioning results: Among them: [p] ai ] and [p aj ] represents the set of partitions to which nodes i and j belong in partition a; χ η (i,j) indicates whether node i and node j belong to the same region in the partition; k is the number of integration bases used; s ij The similarity between node i and node j in the partition of the ensemble is reflected, with a value range of [0, 1]. Here, the node similarity is defined by whether different nodes belong to the same partition in multiple partition results, which is different from the similarity in step ① above. ③ Agglomerative clustering: The elements of each row of the similarity matrix Sim are used as the attributes of the corresponding nodes, and the number of partition targets α is used as the termination condition. Agglomerative clustering is used to cluster the similarity matrix between nodes.

8. The dynamic two-layer zoning optimization method for power grids considering regional voltage control capability according to claim 1, characterized in that, The partitioning optimization in step (6) is as follows: ① Data reading: Obtain power grid operating parameters and check if there are any node voltage over-limits. If so, proceed to step ②. ② Determine the main partition to which the voltage over-limit node belongs, schedule the reactive power equipment in the main partition for optimization simulation. If the voltage over-limit problem is successfully solved, end the optimization and output the optimization solution; otherwise, jump to step ③. ③ Traverse the node voltages in the neighboring partitions of the main partition, select the partition to which the node with the highest voltage belongs and merge it into the main partition, schedule the reactive power sources in the main partition for reactive power optimization simulation, and if the voltage limit problem is successfully solved, the optimization ends and the optimization scheme is output. Otherwise, repeat step ③. If all reactive power sources have participated in the scheduling and there are still voltage over-limit nodes, output the current control scheme and issue a warning.

Images