Distributed photovoltaic cluster elastic grid division method based on space-time dynamic association

Abstract

The invention discloses a distributed photovoltaic cluster elastic grid division method based on space-time dynamic association. The method comprises the following steps: constructing a space-time high-dimensional feature tensor for describing the running state of a power distribution network; a comprehensive space-time incidence matrix is obtained; a Laplacian matrix reflecting source-load interaction characteristics is constructed; performing characteristic decomposition and dimension reduction mapping on the Laplacian matrix to obtain initial sub-grids of a plurality of distributed photovoltaic clusters; verifying whether the frequency change rate caused by the maximum power shortage in the isolated island operation mode is out of limit or not; calculating the critical clearing time of the system through time domain simulation, and taking the critical clearing time as a quantitative index of transient stability; and calculating a migration priority index of the boundary node, carrying out iterative updating until the whole network meets the multi-dimensional constraint, and outputting a final division scheme. The method can adapt to the spatial-temporal fluctuation characteristics of high-proportion distributed photovoltaic, effectively guarantees the frequency safety and transient stability of the power distribution network, and achieves the dynamic balance of physical topology, spatial-temporal association and safe operation.

Description

Technical Field

[0001] This invention belongs to the field of power grids and relates to distribution network operation and control technology, specifically to a method for elastic grid partitioning of distributed photovoltaic clusters based on spatiotemporal dynamic correlation. Background Technology

[0002] Currently, the penetration rate of new energy sources, represented by distributed photovoltaic (PV) power, in power distribution networks is experiencing explosive growth, driving the transformation of these networks from traditional unidirectional radial networks to multi-source interactive active distribution networks. However, distributed PV exhibits significant randomness and intermittency, and the grid connection of numerous power electronic inverters results in a "dual high" characteristic of both "high proportion of renewable energy" and "high proportion of power electronic equipment" in the power distribution network. This transformation leads to a significant reduction in the system's equivalent rotating inertia and weakened disturbance immunity, posing a severe challenge to the safe and stable operation of the power grid.

[0003] Existing methods for distribution network cluster partitioning or reconfiguration mostly focus on optimizing static electrical distances, network modularity, or power balance. These methods have three main limitations: First, they ignore the spatiotemporal dynamic correlation between sources and loads. Traditional methods are often based on steady-state data from a single section, making it difficult to characterize the power fluctuation correlation of distributed photovoltaic clusters under different meteorological conditions and time scales, resulting in partitioning results that are difficult to adapt to rapid changes in photovoltaic output. Second, they lack consideration for the dynamic stability margin of the system. Existing partitioning strategies often only focus on steady-state indicators, neglecting the frequency security and transient stability of subgrids under islanded or disturbed operating modes. In low-inertia scenarios, once a fault occurs, the partitioned "vulnerable subgrids" are highly susceptible to system instability due to exceeding the frequency change rate limit or insufficient critical cut-off time. Third, they lack flexible boundary adjustment mechanisms. Faced with the uncertainty of sources and loads, fixed network boundaries cannot achieve dynamic resource mutual assistance.

[0004] Therefore, there is an urgent need for a distributed photovoltaic cluster elastic grid partitioning method that can integrate spatiotemporal dynamic correlation characteristics and take into account the system's transient stability margin, so as to enable the distribution network to actively adapt to and defend against photovoltaic fluctuations. Summary of the Invention

[0005] Purpose of the invention: To address the shortcomings of existing distribution network partitioning methods, which are unable to adapt to the spatiotemporal fluctuations of high-proportion distributed photovoltaic (PV) systems and neglect system dynamic stability constraints, this invention provides an elastic grid partitioning method for distributed PV clusters based on spatiotemporal dynamic correlation. This method can adapt to the spatiotemporal fluctuation characteristics of high-proportion distributed PV systems, effectively ensure the frequency security and transient stability of the distribution network, and achieve a dynamic balance between physical topology, spatiotemporal correlation, and safe operation.

[0006] Technical Solution: To achieve the above objectives, this invention provides a method for elastic grid partitioning of distributed photovoltaic clusters based on spatiotemporal dynamic correlation, comprising the following steps:

[0007] S1: Obtain the corresponding data of the target distribution network and construct a spatiotemporal high-dimensional feature tensor describing the operating status of the distribution network;

[0008] S2: Based on the spatiotemporal high-dimensional feature tensor, the electrical distance sensitivity matrix and power time-series fluctuation correlation matrix between nodes are calculated respectively, and a comprehensive spatiotemporal correlation matrix is ​​obtained by dynamic weighting factor fusion, which quantifies the dual coupling strength between nodes at the physical and information levels.

[0009] S3: Abstract the distribution network as a weighted undirected graph, with nodes as vertices and the element values ​​of the comprehensive spatiotemporal correlation matrix as edge weights, and construct a Laplace matrix that reflects the source-load interaction characteristics.

[0010] S4: Perform eigenvalue decomposition and dimensionality reduction mapping on the Laplace matrix, and use the K-means++ clustering algorithm to divide the nodes in the feature space to obtain several initial sub-grids of distributed photovoltaic clusters.

[0011] S5: For each initial subgrid, establish a virtual inertia evaluation model that considers the penetration rate of power electronic equipment, and verify whether the rate of change of frequency (RoCoF) caused by the maximum power deficit exceeds the limit in islanded operation mode;

[0012] S6: Preset an N-1 short-circuit fault set in each subgrid, and calculate the critical clearing time (CCT) of the system through time-domain simulation, which is used as a quantitative indicator of transient stability.

[0013] S7: For “fragile subgrids” that fail the verification in step S5 or step S6, calculate the migration priority index of their boundary nodes, strip the nodes with the highest priority index and merge them into the adjacent “strong subgrids”, iterate and update until the entire network satisfies the multidimensional constraints, and output the final partitioning scheme.

[0014] Furthermore, in step S2, the elements in the spatiotemporal correlation matrix W are integrated. The coupling weight between node i and node j is represented by the following formula:

[0015]

[0016] in: Normalized voltage sensitivity characterizes spatial electrical distance; The power fluctuation correlation coefficient characterizes the time-dependent power output characteristics. The meteorological driving weighting factor varies with the time window t; when the meteorological data is determined to be "high-fluctuation weather", Take the smaller value to emphasize power correlation; when judged as "stable weather", Take the larger value to emphasize electrical distance.

[0017] Furthermore, in step S2, the normalized voltage sensitivity... Inverse matrix calculation based on the Jacobian matrix of the distribution network:

[0018]

[0019] in, The elements in the Jacobian inverse matrix represent the changes in active power injection at node j. The amount of change in voltage amplitude at node i The sensitivity of the index is calculated; the denominator is the maximum value of the sensitivity of the entire network, which is used to normalize the index to the [0, 1] interval.

[0020] Furthermore, in step S2, the power fluctuation correlation coefficient The calculation incorporates a time decay factor, the formula of which is:

[0021]

[0022] in: , Let be the net injection power at nodes i and j at time τ, respectively; , λ represents the average power within the time window T; λ∈(0,1] is the forgetting factor, which is used to give higher weight to recent data, thereby capturing the latest trend of photovoltaic power output.

[0023] Furthermore, in step S3, a normalized Laplace matrix is ​​constructed, expressed as follows:

[0024]

[0025] Where: I is the identity matrix, D is the degree matrix, and its diagonal elements .

[0026] Further, step S4 includes:

[0027] A1: Calculate the Laplacian matrix The eigenvalues ​​are calculated and arranged in ascending order. The optimal number of subgrids K is determined based on the eigenvalue difference maximization principle (Eigengap Heuristic).

[0028] A2: Select the eigenvectors corresponding to the first K eigenvalues. Construct the feature matrix And standardize each row vector of U;

[0029] A3: Use the standardized feature matrix row vectors as sample points, input them into the K-means++ algorithm for clustering, and output the subgrid label to which the node belongs.

[0030] Furthermore, in step S5, the virtual inertia and frequency constraint verification satisfies the rate of change of frequency (RoCoF) constraint:

[0031]

[0032] in: The rate of change of the system's frequency at the instant the system encounters a disturbance; For subgrid The maximum power deficit that may occur within the system; The inherent inertia constant of a traditional rotating electric motor within the subgrid; The virtual inertia constant provided by the control strategy for inverter-type power sources (photovoltaics, energy storage) within the subgrid; The maximum permissible rate of frequency change threshold specified in the standards for safe operation of power distribution networks.

[0033] Furthermore, the method for calculating the critical resection time in step S6 includes:

[0034] For each tie line l within the subgrid, a three-phase short-circuit fault is set, and the rotor motion equations are solved by numerical integration, gradually increasing the fault duration. Until the system loses synchronization stability;

[0035]

[0036] in: Critical cut-off time for the k-th subgrid; Let k be the set of critical paths within subgrid k; if the calculated... Less than the minimum operating time of the relay protection device If the transient stability of the subgrid is not up to standard, then the subgrid is deemed to be unqualified.

[0037] Furthermore, in step S7, the migration priority index of the boundary nodes... Defined as:

[0038]

[0039] in: The sensitivity of boundary node b to the voltage support within its current subgrid; The sensitivity of boundary node b to the voltage support of adjacent candidate subgrids; , These represent the power imbalance between the current subgrid and its adjacent subgrids, respectively. and These are the first weighting coefficient and the second weighting coefficient, respectively.

[0040] Refactoring logic: Selection The largest node is migrated, moving it from a weak subgrid with "less demand" to a strong subgrid with "stronger coupling".

[0041] Beneficial effects: Compared with existing technologies, this invention integrates spatiotemporal dynamic correlation characteristics and takes into account the dynamic stability margin of the system. Its advantages are as follows:

[0042] 1. Adapting to photovoltaic fluctuations: This invention overcomes the shortcomings of traditional static division that ignores time-varying characteristics. By introducing a comprehensive spatiotemporal correlation matrix with meteorological driving weights, it accurately quantifies the dual coupling between nodes at the physical and power levels, thus adapting to the randomness of photovoltaic output.

[0043] 2. Ensuring operational safety: This invention establishes a multi-dimensional evaluation system that considers virtual inertia and transient stability, and uses the rate of change of frequency and critical cut-off time as rigid constraints, thus solving the problem of instability in low-inertia systems.

[0044] 3. Flexible dynamic optimization: This invention proposes a boundary reconstruction mechanism based on migration priority index, which dynamically migrates nodes of vulnerable subgrids to robust subgrids, realizing a dynamic balance between the physical topology, spatiotemporal correlation and safe operation of the distribution network. Attached Figure Description

[0045] Figure 1 This is a schematic flowchart of the method of the present invention;

[0046] Figure 2 A schematic diagram illustrating the evolution of the transient stability margin CCT;

[0047] Figure 3 This is a schematic diagram illustrating the evolution of the frequency security index RoCoF. Detailed Implementation

[0048] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0049] Example 1:

[0050] like Figure 1 As shown, this embodiment provides a method for elastic grid partitioning of distributed photovoltaic clusters based on spatiotemporal dynamic correlation, including the following steps:

[0051] S1: Panoramic Data Mapping: Obtain the physical topology, line impedance parameters, historical output sequence of photovoltaic nodes, historical power sequence of load nodes, and micro-meteorological data of the target distribution network, and construct a spatiotemporal high-dimensional feature tensor describing the operating status of the distribution network;

[0052] S2: Spatiotemporal Coupling Metric: Based on the high-dimensional spatiotemporal feature tensor, the electrical distance sensitivity matrix and power time-series fluctuation correlation matrix between nodes are calculated respectively, and a comprehensive spatiotemporal correlation matrix is ​​obtained by fusing through dynamic weighting factors, which quantifies the dual coupling strength between nodes at the physical and informational levels.

[0053] Elements in the spatiotemporal correlation matrix W The coupling weight between node i and node j is represented by the following formula:

[0054]

[0055] in: Normalized voltage sensitivity characterizes spatial electrical distance; The power fluctuation correlation coefficient characterizes the time-dependent power output characteristics. The meteorological driving weighting factor varies with the time window t; when the meteorological data is determined to be "high-fluctuation weather", Take the smaller value to emphasize power correlation; when judged as "stable weather", Take the larger value to emphasize electrical distance.

[0056] Normalized voltage sensitivity Inverse matrix calculation based on the Jacobian matrix of the distribution network:

[0057]

[0058] in, The elements in the Jacobian inverse matrix represent the changes in active power injection at node j. The amount of change in voltage amplitude at node i The sensitivity of the index is calculated; the denominator is the maximum value of the sensitivity of the entire network, which is used to normalize the index to the [0, 1] interval.

[0059] Power fluctuation correlation coefficient The calculation incorporates a time decay factor, the formula of which is:

[0060]

[0061] in: , Let be the net injection power at nodes i and j at time τ, respectively; , λ represents the average power within the time window T; λ∈(0,1] is the forgetting factor, which is used to give higher weight to recent data, thereby capturing the latest trend of photovoltaic power output.

[0062] S3: The main construction of the spectral clustering graph is as follows: the distribution network is abstracted into a weighted undirected graph, with nodes as vertices and the element values ​​of the comprehensive spatiotemporal correlation matrix as edge weights, to construct a Laplace matrix that reflects the source-load interaction characteristics.

[0063] In this embodiment, a normalized Laplace matrix is ​​constructed, expressed as follows:

[0064]

[0065] Where: I is the identity matrix, D is the degree matrix, and its diagonal elements .

[0066] S4: Initial subnetting: Perform eigenvalue decomposition and dimensionality reduction mapping on the Laplacian matrix, and use the K-means++ clustering algorithm to divide the nodes in the feature space to obtain several initial subnets for distributed photovoltaic clusters.

[0067] Step S4 includes:

[0068] A1: Calculate the Laplacian matrix The eigenvalues ​​are calculated and arranged in ascending order. The optimal number of subgrids K is determined based on the eigenvalue difference maximization principle (Eigengap Heuristic).

[0069] A2: Select the eigenvectors corresponding to the first K eigenvalues. Construct the feature matrix And standardize each row vector of U;

[0070] A3: Use the standardized feature matrix row vectors as sample points, input them into the K-means++ algorithm for clustering, and output the subgrid label to which the node belongs.

[0071] S5: Dynamic Inertia and Frequency Constraint Verification: For each initial subgrid, establish a virtual inertia evaluation model that considers the penetration rate of power electronic equipment, and verify whether the rate of change of frequency (RoCoF) caused by the maximum power deficit exceeds the limit in islanded operation mode.

[0072] Virtual inertia and frequency constraint verification satisfy the rate of change of frequency (RoCoF) constraint:

[0073]

[0074] in: The rate of change of the system's frequency at the instant the system encounters a disturbance; For subgrid The maximum power deficit that may occur within the system; The inherent inertia constant of a traditional rotating electric motor within the subgrid; The virtual inertia constant provided by the control strategy for inverter-type power sources (photovoltaics, energy storage) within the subgrid; The maximum permissible rate of frequency change threshold specified in the standards for safe operation of power distribution networks.

[0075] S6: Transient stability margin assessment: Preset N-1 short-circuit fault sets in each subgrid, calculate the critical clearing time (CCT) of the system through time-domain simulation, and use it as a quantitative indicator of transient stability.

[0076] Methods for calculating the critical resection time include:

[0077] For each tie line l within the subgrid, a three-phase short-circuit fault is set, and the rotor motion equations are solved by numerical integration, gradually increasing the fault duration. Until the system loses synchronization stability;

[0078]

[0079] in: Critical cut-off time for the k-th subgrid; Let k be the set of critical paths within subgrid k; if the calculated... Less than the minimum operating time of the relay protection device If the transient stability of the subgrid is not up to standard, then the subgrid is deemed to be unqualified.

[0080] S7: Flexible Boundary Reconstruction: For “fragile subgrids” that fail the verification in step S5 or step S6, calculate the migration priority index of their boundary nodes, strip and merge the nodes with the highest priority index into the adjacent “strong subgrids”, iterate and update until the entire network satisfies the multidimensional constraints, and output the final partitioning scheme.

[0081] migration priority index of boundary nodes Defined as:

[0082]

[0083] in: The sensitivity of boundary node b to the voltage support within its current subgrid; The sensitivity of boundary node b to the voltage support of adjacent candidate subgrids; , These represent the power imbalance between the current subgrid and its adjacent subgrids, respectively. and These are the first weighting coefficient and the second weighting coefficient, respectively.

[0084] Refactoring logic: Selection The largest node is migrated, moving it from a weak subgrid with "less demand" to a strong subgrid with "stronger coupling".

[0085] This embodiment also provides a dynamic update mechanism based on the rolling time domain, including:

[0086] Set the refactoring cycle At the beginning of each cycle, the latest ultra-short-term photovoltaic power prediction data is read, and the spatiotemporal coupling weights in step S2 are recalculated. If the predicted stability margin is... If the decrease exceeds the preset threshold δ, the boundary reconstruction process in step S7 is triggered to achieve a breathing-like adjustment of the grid shape as the photovoltaic output fluctuates.

[0087] Example 2:

[0088] To verify the effectiveness of the method of the present invention, the following simulation experiments and analyses were conducted in this embodiment:

[0089] A simulation example is built based on an improved IEEE 33-node distribution network system. The system contains 33 nodes, with 30% of the nodes connected to distributed photovoltaic (PV) power sources. The simulation time is... The window is set to 24 hours. The main parameter settings are as follows:

[0090] Spatiotemporal correlation parameter: forgetting factor Meteorological driving weighting factor (Electrical distance is emphasized under stable weather conditions). Stability constraint parameter: Maximum permissible rate of change of system frequency. Transient stable critical resection time threshold Reconstruction parameters: MPI calculation weights Clustering parameters: Preset target number of clusters. .

[0091] Initial mesh generation (Phase I)

[0092] Based on panoramic data mapping and spatiotemporal coupling measurement, a comprehensive spatiotemporal correlation matrix is ​​constructed, and a spectral clustering algorithm is used for initial network partitioning. The initial partitioning results show that although clustering based solely on spatiotemporal coupling strength ensures the tightness of electrical connections, it neglects the inertia support and transient stability within each subgrid.

[0093] Initial state assessment:

[0094] Subgrids 1, 2, and 3 were identified as “fragile subgrids.” Although the rate of change of frequency (RoCoF) met the requirements for some periods, the critical cut-off time (CCT) was generally below 0.15 s, indicating that they were highly susceptible to instability in the event of a short-circuit fault.

[0095] Subgrid 4: It is classified as a "strong subgrid" with all indicators within the safe range.

[0096] Elastic boundary reconstruction process (Phase II)

[0097] Since the initial partitioning fails to meet the multidimensional stability constraints, the system triggers an elastic boundary reconstruction mechanism. The algorithm iteratively migrates nodes from fragile subgrids to robust subgrids or balances them between subgrids by calculating the migration priority index (MPI) of the boundary nodes.

[0098] Analysis of typical iterative process:

[0099] Iterations 1-3 (Initial Adjustments):

[0100] Submesh 3 was reconstructed due to CCT exceeding the limit (0.134s), and boundary node 12 was migrated to submesh 4 based on the maximum MPI value (0.5565).

[0101] Subsequently, subgrid 4 experienced a decrease in stability due to carrying too many nodes (CCT dropped to 0.139s), and node 12 was re-identified and migrated to subgrid 2. This demonstrates the algorithm's dynamic search capability.

[0102] Iterations 4-8 (Deep Optimization):

[0103] The algorithm identified that subgrids 3 and 2 still have stability risks.

[0104] Key boundary nodes such as nodes 8, 16, 9, and 29 were successively stripped from the vulnerable region and merged into subgrid 1, which had a higher stability margin at this time.

[0105] In this phase, by sacrificing some electrical spacing tightness, the overall inertia level of each subnet was improved.

[0106] Iteration rounds 9-11 (fine-tuning convergence):

[0107] As a large number of nodes flooded into subgrid 1, its CCT index fluctuated (dropped to 0.091s), transforming it into a fragile subgrid.

[0108] The algorithm reacted quickly, backfilling nodes 16 and 14 into subgrid 4 and moving node 33 from subgrid 3 to subgrid 4.

[0109] This adjustment mechanism effectively mitigates the load pressure on individual subgrids.

[0110] Final partitioning results

[0111] Network topology diagrams before and after optimization are as follows: Figure 2 and Figure 3 As shown. After 12 rounds of iterative calculations, the rate of change of frequency (RoCoF) and transient stability margin (CCT) of all subgrids in the entire network converged to within the safe threshold, and the elastic reconstruction was completed.

[0112] The final cluster partitioning scheme is as follows:

[0113] Cluster1 (Core Backbone Network): Contains nodes [1, 3, 4, 5, 7, 8, 9, 10, 19, 20, 24, 29]. This cluster is relatively large and typically includes a main power node, undertaking a significant portion of the system balancing tasks.

[0114] Cluster2 (distributed subnet): contains nodes [2, 6, 12, 23, 27, 30].

[0115] Cluster3 (End Micronet): Contains nodes [11, 15, 21, 22, 31].

[0116] Cluster4 (PV Consumption Group): Contains nodes [13, 14, 16, 17, 18, 25, 26, 28, 32, 33].

[0117] Figure 2 shows the evolution process of transient stability margin CCT: demonstrating the improvement effect of the elastic boundary reconstruction algorithm on the transient stability of each photovoltaic cluster in 12 iterations.

[0118] The red dashed line represents the minimum critical resection time threshold set by the system.

[0119] Cluster1 (orange line): This demonstrates a typical dynamic adjustment process. In the 9th iteration, due to the merging of too many nodes from Cluster2 and 3, its inertia support became insufficient, causing the CCT metric to drop significantly to 0.091s, marked as a "risk point." The algorithm responded quickly, triggering a reverse reconstruction mechanism to restore it to a safe level.

[0120] Cluster2 & 3 (blue / green lines): Initially in the unqualified area of ​​the red shaded zone, after multiple rounds of boundary node stripping and optimization, they gradually climb and converge to above the safe threshold.

[0121] Figure 3 The evolution of the frequency security index RoCoF demonstrates the enhanced immunity of the system to disturbances.

[0122] The initial RoCoF of Clusters 1, 2, and 3 all exceeded the safety limit of 2.0 Hz / s, indicating a risk of frequency collapse when running in an islanded environment.

[0123] As the iterations progressed, by introducing virtual inertia support from the robust subnet (Cluster4) and optimizing the network topology, the RoCoF of all subnets successfully converged to a convergence safe region below 2.0 Hz / s, verifying the effectiveness of the method of this invention in improving the frequency rigidity of the system.

[0124] Simulation results show that the partitioning method proposed in this invention can effectively identify and correct the risks of insufficient inertia and transient instability caused by photovoltaic access compared with traditional static spectral clustering. Through 12 iterations, all three sub-grids that were originally in a "vulnerable" state were optimized to a "robust" state, achieving the optimal balance of the distribution network in terms of physical topology, spatiotemporal correlation, and safety and stability.

Claims

1. A distributed photovoltaic cluster elastic grid partitioning method based on spatiotemporal dynamic association, characterized in that, The method comprises the following steps: S1: obtaining corresponding data of the target power distribution network, and constructing a high-dimensional space-time feature tensor describing the operation state of the power distribution network; S2: based on the high-dimensional space-time feature tensor, respectively calculating an electrical distance sensitivity matrix and a power time sequence fluctuation correlation matrix between nodes, and obtaining a comprehensive space-time correlation matrix through dynamic weight factor fusion, and quantifying the dual coupling strength between nodes in the physical and information layers; S3: abstracting the power distribution network as a weighted undirected graph, taking nodes as vertices, and taking element values of the comprehensive space-time correlation matrix as edge weights, and constructing a Laplace matrix reflecting source-load interaction characteristics; S4: performing characteristic decomposition and dimension reduction mapping on the Laplace matrix, and using a K-means++ clustering algorithm to split nodes in the feature space to obtain an initial sub-grid of a plurality of distributed photovoltaic clusters; S5: for each initial sub-grid, a virtual inertia evaluation model considering power electronic equipment penetration rate is established, and whether the frequency change rate is out of limit due to the maximum power shortage in the island operation mode is checked; S6: presetting an N-1 short-circuit fault set in each sub-grid, and taking the critical removal time of the system calculated through time domain simulation as a quantitative index of transient stability; S7: for the "fragile sub-grid" that does not pass the check of step S5 or step S6, the migration priority index of the boundary node is calculated, the node with the highest priority index is stripped and merged into the adjacent "strong sub-grid", and the iteration is updated until the multi-dimensional constraints of the whole network are met, and the final division scheme is output. 2.The method of claim 1, wherein, The elements in the spatio-temporal correlation matrix W in the step S2 represents the coupling weight between node i and node j, and the calculation formula is: ； where: is the normalized voltage sensitivity, representing the spatial electrical distance; is the power fluctuation correlation coefficient, representing the temporal output characteristics; is the meteorological driving weight factor varying with the time window t.

3. The method of claim 2, wherein, The step S2 of normalizing the voltage sensitivity Inverse matrix calculation based on the Jacobian matrix of the power distribution network: ； where, is the element in the inverse Jacobian matrix, representing the change in active power injection at node j is the sensitivity of the change in voltage magnitude at node i; the denominator is the maximum value of the network sensitivity, used to normalize the index to the interval [0, 1]. is the sensitivity of the change in voltage magnitude at node i; the denominator is the maximum value of the network sensitivity, used to normalize the index to the interval [0, 1].

4. The distributed photovoltaic cluster elastic grid partitioning method based on space-time dynamic association according to claim 3, characterized in that, The power fluctuation correlation coefficient in the step S2 The calculation of the power fluctuation correlation coefficient in the step S2 introduces a time decay factor, which is given by ； in: , Let be the net injection power at nodes i and j at time τ, respectively; , λ represents the average power within the time window T; λ∈(0,1] is the forgetting factor, which is used to give higher weight to recent data, thereby capturing the latest trend of photovoltaic power output.

5. The distributed photovoltaic cluster elastic grid partitioning method based on space-time dynamic association according to claim 4, characterized in that, The step S3 constructs a normalized Laplace matrix, which is expressed as follows: ； where: I is the identity matrix, D is the degree matrix, whose diagonal elements .

6. The distributed photovoltaic cluster elastic grid partitioning method based on space-time dynamic association according to claim 5, characterized in that, The step S4 comprises: A1: compute eigenvalues of Laplacian matrix and arrange them in ascending order, and determine the optimal number of subgrids K according to the principle of maximum eigenvalue difference A2: select eigenvectors corresponding to the top K eigenvalues , construct a feature matrix and normalize each row vector of U; A3: inputting the standardized feature matrix row vector as a sample point into the K-means++ algorithm for clustering, and outputting the sub-grid label to which the node belongs.

7. The distributed photovoltaic cluster elastic grid partitioning method based on space-time dynamic association according to claim 6, characterized in that, In the step S5, the virtual inertia and frequency constraint check meet the frequency change rate constraint: ； in: The rate of change of the system's frequency at the instant the system encounters a disturbance; For subgrid The maximum power deficit that may occur within the system; The inherent inertia constant of a traditional rotating electric motor within the subgrid; The virtual inertia constant provided by the control strategy for inverter-type power supplies within the subgrid; The maximum permissible rate of frequency change threshold specified in the standards for safe operation of power distribution networks. 8.The method of claim 7, wherein, The calculation method of the critical removal time in the step S6 comprises: For each tie line l within the subnetwork, a three-phase short-circuit fault is set, and the rotor motion equation is solved by numerical integration, gradually increasing the fault duration , until the system loses synchronous stability; ； wherein: critical clearing time of the kth sub-grid; is the critical line set within the kth sub-grid; if the calculated is less than the minimum operating time of the relay protection device then the transient stability of the sub-grid is determined to be unqualified.

9. The distributed photovoltaic cluster elastic grid partitioning method based on space-time dynamic association according to claim 8, characterized in that, the migration priority index of the border node in the step S7 is defined as: ； wherein: is the sensitivity of the boundary node b to the voltage support within the current sub-grid; is the sensitivity of the boundary node b to the voltage support of the adjacent candidate sub-grid; , are the power imbalance degrees of the current sub-grid and the adjacent sub-grid, respectively; and are the first weight coefficient and the second weight coefficient, respectively.

Images