Method and system for partitioning AC / DC power grid and identifying weak nodes based on spectral clustering

Abstract

The invention discloses an AC / DC power grid partitioning and weak node identification method and system based on spectral clustering, and the method comprises the steps: calculating a similarity index according to the change of a set state variable between any two nodes in a power grid, and building a similarity index matrix; the short-circuit ratio between any two nodes in the power grid is calculated, an AC-DC coupling strength matrix is established, the mutual influence of the AC-DC power grid is considered, and the short-circuit ratio between two DC drop points in the nodes is a comprehensive short-circuit ratio index for introducing AC-DC multi-infeed; fusing the similarity index matrix and the AC / DC coupling strength matrix to obtain an improved similarity matrix; based on the improved similarity matrix, carrying out dynamic partitioning by adopting spectral clustering; and calculating singular value entropies of all nodes of each partition, and taking the node with the maximum singular value entropy in each partition as a weak node. According to the method, the correlation degree of state parameters and the alternating current and direct current mutual coupling relation in the alternating current and direct current hybrid system are considered, and partitioning and weak node identification can be accurately carried out.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system operation, and more specifically, relates to a method and system for AC-DC power grid partitioning and weak node identification based on spectral clustering. Background Art

[0002] DC power transmission has become one of the important ways of power transmission in China due to its advantages in long-distance, large-capacity, and cross-regional power transmission. The multi-infeed DC power transmission system can expand the system transmission capacity and increase the flexibility of operation modes, but it will also cause negative feedback interactions between the AC-DC systems and between the DC-DC systems. The increase in DC power transmission capacity will lead to the weakening of the strength of the receiving-end AC system, and the contradiction between "strong AC and weak DC" is prominent. Once the voltage of the receiving-end power grid AC system is abnormal or a fault occurs, it may cause commutation failure of the inverter station, and in severe cases, it may even lead to the interruption of multi-circuit DC power transmission, ultimately threatening the safe and stable operation of the entire system. Therefore, accurately and quickly partitioning the AC-DC hybrid power grid dynamically and identifying its weak nodes is of great significance for ensuring the safe and stable operation of large-scale AC-DC power grids.

[0003] CN111652469A discloses a method and system for identifying weak links in an AC-DC hybrid power grid, which constructs a vulnerability index calculation model for identifying weak links in an AC-DC hybrid power grid; establishes an equivalent topological structure of the AC-DC hybrid power grid to be identified; according to the equivalent topological structure, calculates identification parameters, that is: statistically calculates the probability of a transmission line failure in the power grid to be identified, calculates the ratio of the power flow transfer correlation degree of each transmission line, calculates the voltage ratio of each bus node, and determines the correlation degree between each transmission line and each bus node; inputs the identification parameters into the vulnerability index calculation model, calculates the vulnerability index of each transmission link, and the larger the value of the vulnerability index, the weaker the corresponding transmission link, thereby realizing the identification of weak links. However, this invention only considers the correlation degree and voltage, does not consider the mutual influence of the AC-DC power grid and the mutual coupling relationship between AC and DC in the AC-DC hybrid system, and does not consider partitioning to identify weak nodes in different regions. Summary of the Invention

[0004] To solve the deficiencies in the prior art, the present invention provides a method and system for AC-DC power grid partitioning and weak node identification based on spectral clustering.

[0005] The present invention adopts the following technical solutions.

[0006] The first aspect of the present invention proposes a method for AC-DC power grid partitioning and weak node identification based on spectral clustering, which is characterized by including: Calculate the similarity index between any two nodes according to the change of the set state variables between any two nodes in the power grid, and establish a similarity index matrix; Calculate the short-circuit ratio between any two nodes in the power grid, and establish an AC-DC coupling strength matrix, where the mutual influence between the AC and DC power grids is taken into account. For the short-circuit ratio between two DC landing points in a node, the comprehensive short-circuit ratio index considering multi-infeed of AC and DC is introduced; Fuse the similarity index matrix and the AC-DC coupling strength matrix to obtain an improved similarity matrix; Based on the improved similarity matrix, perform dynamic partitioning using spectral clustering that adaptively determines the number of clusters based on eigenvalues; According to the dynamic partitioning result, calculate the singular value entropy of all nodes in each partition, and take the node with the largest singular value entropy in each partition as the weak node.

[0007] Preferably, the calculating the similarity index between any two nodes according to the change of the set state variables between any two nodes in the power grid and establishing a similarity index matrix is specifically as follows: Select one of the operating states of the AC-DC power grid as the state variable, and collect the states of nodes i and node j at a set time interval. The state variables of the two nodes within the set time period form two trajectories. Using two continuous non-decreasing functions defined on , stretch or compress the two trajectories locally on the time axis respectively, solve the maximum distance between each corresponding position of the two stretched or compressed trajectories, find two optimal continuous non-decreasing functions to minimize the maximum distance, and the minimized maximum distance is the similarity index between these two nodes. Calculate the similarity index between any two nodes and establish a similarity index matrix.

[0008] Preferably, the comprehensive short-circuit ratio index for the nodes between two DC landing points is:

[0009] Where: is the comprehensive short-circuit ratio index between DC landing point p and DC landing point q ; is the commutation bus voltage of DC landing point p ; , are the DC powers of DC landing points p , q respectively; is the element in the th row and p th column of the equivalent nodal impedance matrix p viewed from each DC commutation bus; The equivalent nodal impedance matrix seen from each DC commutation busbar of the p th q row and the

[0010] Preferably, the fusion of the similarity index matrix and the AC-DC coupling strength matrix to obtain an improved similarity matrix is specifically as follows: Regarding node similarity and electrical distance as two factors, the importance weights of node similarity and electrical distance are obtained through the analytic hierarchy process. The importance weight of node similarity is used as the weight of the similarity index matrix, and the importance weight of electrical distance is used as the weight of the AC-DC coupling strength matrix. The similarity index matrix and the AC-DC coupling strength matrix are weighted to obtain an improved similarity matrix.

[0011] Preferably, based on the improved similarity matrix, spectral clustering based on adaptively determining the number of clusters by eigenvalues is used for dynamic partitioning, specifically as follows: According to the improved similarity matrix, the adjacency matrix is calculated. By performing a row-sum operation on the adjacency matrix, the degree of each node can be calculated. All degree values are used as diagonal elements to form a diagonal matrix, which is the degree matrix; The degree matrix is subtracted from the adjacency matrix to obtain the Laplacian matrix, and the Laplacian matrix is normalized with the help of the degree matrix; The normalized Laplacian matrix is subjected to eigenvalue decomposition. The decomposed eigenvalues are sorted, and the eigenvectors corresponding to the largest set number of eigenvalues are selected. These eigenvectors are arranged by column to form an eigenvector matrix. Each row in the eigenvector matrix is used as a clustering sample for K-means clustering.

[0012] Preferably, the calculation of the adjacency matrix according to the improved similarity matrix is specifically as follows: Using the element in the i th j row and i th j column of the improved similarity matrix to represent the square of the distance between the

[0013] th node and the i th j node, setting a scale parameter, which determines the attenuation speed of the similarity between nodes, and using the Gaussian kernel function to calculate the adjacency matrix according to the improved similarity matrix. The formula is: For the element in the i th row and j th column of the improved similarity matrix; Set scale parameter.

[0014] Preferably, the number of clusters is adaptively determined based on eigenvalues, specifically: Arrange the eigenvalues of the Laplacian matrix from largest to smallest, calculate the difference between each eigenvalue and its next eigenvalue as the eigenvalue gap, form an eigenvalue gap sequence, and sequentially find the first maximum value in the eigenvalue gap sequence. The label of this value in the eigenvalue gap sequence is the number of clusters.

[0015] Preferably, the calculation of the singular value entropy of all nodes in each partition is specifically: For each node in each partition, perform AC-DC power flow calculations to obtain all power flow calculation equations for each node. Take the partial derivatives of each variable in each power flow calculation equation. These partial derivatives form a Jacobian matrix. After performing singular value decomposition on this matrix, normalize the singular values into a probability distribution and calculate the singular value entropy of the node, specifically:

[0016] In the formula, is the singular value entropy, is the result of normalizing the l th singular value of the Jacobian matrix, is the order of the Jacobian matrix.

[0017] The second aspect of the present invention proposes a spectrum clustering-based AC-DC power grid partitioning and weak node identification system using the method described in the first aspect of the present invention, including a similarity index matrix construction model, an AC-DC coupling strength matrix construction module, a fusion module, a dynamic partitioning module, and a weak node identification module, characterized in that: Similarity index matrix construction model: Calculate the similarity index between any two nodes in the power grid according to the change of the set state variables between any two nodes, and establish a similarity index matrix; AC-DC coupling strength matrix construction module: Calculate the short-circuit ratio between any two nodes in the power grid and establish an AC-DC coupling strength matrix. Considering the mutual influence of the AC-DC power grid, the short-circuit ratio between two DC landing points in a node is the comprehensive short-circuit ratio index introducing AC-DC multi-infeed; Fusion module: Fuse the similarity index matrix and the AC-DC coupling strength matrix to obtain an improved similarity matrix; Dynamic partitioning module: Based on the improved similarity matrix, perform dynamic partitioning using spectrum clustering with the number of clusters adaptively determined based on eigenvalues; Weak node identification module: According to the dynamic partitioning result, calculate the singular value entropy of all nodes in each partition, and take the node with the largest singular value entropy in each partition as the weak node.

[0018] The third aspect of the present invention proposes a computer device, including a memory and a processor, the memory stores a computer program, and is characterized in that when the processor executes the computer program, it implements the steps of the method for AC-DC power grid partitioning and weak node identification based on spectral clustering according to the first aspect of the present invention.

[0019] The fourth aspect of the present invention proposes a computer-readable storage medium, on which a computer program is stored, and is characterized in that when the computer program is executed by a processor, it implements the steps of the method for AC-DC power grid partitioning and weak node identification based on spectral clustering according to the first aspect of the present invention.

[0020] The beneficial effects of the present invention are as follows. Compared with the prior art, the present invention first calculates the similarity index between any two nodes in the system for the trajectory of the operation state change of the AC-DC system, describes the correlation between each node, forms a similarity matrix, and reflects the similarity of nodes within a certain time scale. Then, considering the mutual influence of the AC-DC power grid, based on the short-circuit ratio to reflect the coupling degree between nodes in the AC-DC hybrid system, a multi-infeed short-circuit ratio index of AC-DC is introduced at the DC landing point to form an AC-DC coupling strength matrix, which is more accurate than the short-circuit ratio obtained by the theoretical model. It not only considers the correlation of state parameters but also the mutual coupling relationship between AC and DC in the AC-DC hybrid system. The weights are selected based on the analytic hierarchy process, and the similarity matrix and the AC-DC coupling strength matrix are weighted to form an improved similarity matrix, and the spectral clustering method is used for dynamic partitioning to adapt to the dynamic changes of complex environments. Finally, the singular value entropy of the nodes in the region is calculated, and the node with the largest singular value entropy is used as the weak node in the region to obtain the identification result of the weak nodes in the AC-DC hybrid system, which can capture complex non-linear relationships and improve the accuracy and sensitivity of regional identification by combining partitioning, avoiding the influence of node data with low correlation. Description of the Drawings

[0021] Figure 1 It is a flowchart of the method for AC-DC power grid partitioning and weak node identification based on spectral clustering. Detailed Embodiments

[0022] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only a part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present invention.

[0023] As Figure 1 shown, Embodiment 1 of the present invention provides a method for identifying AC / DC power grid partitions and weak nodes based on spectral clustering, which is characterized by including: Calculate the similarity index between any two nodes in the power grid according to the change of the set state variable between any two nodes, and establish a similarity index matrix; Calculate the short-circuit ratio between any two nodes in the power grid, and establish an AC / DC coupling strength matrix. Considering the mutual influence between the AC and DC power grids, the short-circuit ratio between two DC landing points in a node is the comprehensive short-circuit ratio index introducing multi-infeed of AC / DC; Fuse the similarity index matrix and the AC / DC coupling strength matrix to obtain an improved similarity matrix; Based on the improved similarity matrix, perform dynamic partitioning using spectral clustering that adaptively determines the number of clusters based on eigenvalues; According to the dynamic partitioning result, calculate the singular value entropy of all nodes in each partition, and the node with the largest singular value entropy in each partition is the weak node.

[0024] Preferably, the calculating the similarity index between any two nodes in the power grid according to the change of the set state variable between any two nodes and establishing a similarity index matrix is specifically: Select one of the operating states of the AC / DC power grid as the state variable, and collect the state variables of nodes i and node j at a set time interval. The state variables of the two nodes within the set time period form two trajectories. Using two continuous non-decreasing functions defined on , stretch or compress the two trajectories locally on the time axis respectively, solve the maximum distance between each corresponding position of the stretched or compressed two trajectories, find two optimal continuous non-decreasing functions to minimize the maximum distance, and the minimized maximum distance is the similarity index between these two nodes. Calculate the similarity index between any two nodes and establish a similarity index matrix.

[0025] It should be noted that the operating state includes but is not limited to the voltage, frequency, power flow, etc. of each node in the power grid.

[0026] Calculate the similarity index between any two nodes , and the formula is:

[0027] In the formula: and are the trajectories composed of the state variables of node i and node j ; and are all continuous non-decreasing functions; represents the distance between two trajectories after stretching or compression at the th moment, represents at all moments, represents for all possible and find the minimum value of the maximum distance.

[0028] Similarity index matrix is:

[0029] In the formula, n is the total number of nodes.

[0030] Preferably, for the comprehensive short-circuit ratio index between two nodes with DC landing points, it is:

[0031] In the formula: is the comprehensive short-circuit ratio index between the DC landing point p and the DC landing point q ; is the commutation bus voltage of the DC landing point p ; , are respectively the DC powers of the DC landing points p , q ; is the element of the th row and p th column of the equivalent nodal impedance matrix p viewed from each DC commutation bus; is the element of the th row and p th column of the equivalent nodal impedance matrix q viewed from each DC commutation bus; For the short-circuit ratio between two nodes in other cases, it is the reciprocal of the absolute value of the per-unit value of the impedance between the two nodes, and the formula is:

[0032] In the formula: is the short-circuit ratio between node and node i when j ; is the short-circuit capacity of node i ; is the rated voltage of node i ; is the rated power; is the equivalent impedance between nodes i and nodes j ; is the per-unit value of the impedance between nodes i and nodes j .

[0033] The AC-DC coupling strength matrix composed of

[0034] Preferably, the similarity index matrix and the AC-DC coupling strength matrix are fused to obtain an improved similarity matrix, specifically: Regarding the node similarity and the electrical distance as two factors, the importance weights of the node similarity and the electrical distance are obtained through the analytic hierarchy process. The importance weight of the node similarity is used as the weight of the similarity index matrix, and the importance weight of the electrical distance is used as the weight of the AC-DC coupling strength matrix. The similarity index matrix and the AC-DC coupling strength matrix are weighted to obtain an improved similarity matrix. The formula is:

[0035] In the formula, and are the importance weights of the node similarity and the electrical distance respectively, is the improved similarity matrix.

[0036] Preferably, based on the improved similarity matrix, spectral clustering with the number of clusters adaptively determined based on eigenvalues is used for dynamic partitioning, specifically: According to the improved similarity matrix, the adjacency matrix is calculated. By performing the row-sum operation on the adjacency matrix, the degree of each node can be calculated. All the degree values are used as the diagonal elements to form a diagonal matrix, and this diagonal matrix is the degree matrix; the degree matrix D The formula is:

[0037] In the formula, is the element in the i th row and j th column of the adjacency matrix; is the degree of the i th node.

[0038] Subtracting the adjacency matrix from the degree matrix gives the Laplacian matrix. The formula is:

[0039] In the formula: is the Laplacian matrix, A is the adjacency matrix, is the element in thei The element in the j row and

[0040] column. And normalize the Laplacian matrix with the help of the degree matrix. The normalization formula is:

[0041] In the formula: is the normalized Laplacian matrix.

[0042] Perform eigenvalue decomposition on the normalized Laplacian matrix, sort the decomposed eigenvalues, select the eigenvectors corresponding to the largest set number of eigenvalues, and arrange these eigenvectors in columns to form an eigenvector matrix. The eigenvector matrix is expressed as:

[0043] In the formula: is the eigenvector matrix; is the eigenvector corresponding to the largest set number of eigenvalues; c is the set number; Take each row in the eigenvector matrix as a clustering sample for K - means clustering.

[0044] Preferably, calculating the adjacency matrix according to the improved similarity matrix is specifically as follows: Use the element in the i row and j column of the improved similarity matrix to represent the square of the distance between the i th node and the j th node. Set a scale parameter, which determines the attenuation rate of the similarity between nodes. Use the Gaussian kernel function to calculate the adjacency matrix according to the improved similarity matrix. The formula is:

[0045] In the formula, is the element in the i row and j column of the adjacency matrix; is the element in the i row and j column of the improved similarity matrix; is the set scale parameter.

[0046] Preferably, adaptively determining the number of clusters based on eigenvalues is specifically as follows: Arrange the eigenvalues of the Laplacian matrix from largest to smallest, calculate the difference between each eigenvalue and its next eigenvalue and use it as the eigenvalue gap to form an eigenvalue gap sequence. The sequence formula is expressed as:

[0047] In the formula, is the o th eigenvalue gap after sorting the eigenvalues corresponding to all nodes, , are respectively the o th and the o +1th eigenvalues after sorting the eigenvalues corresponding to all nodes.

[0048] Search for the first maximum value in the eigenvalue gap sequence in turn. The label of this value in the eigenvalue gap sequence is the number of clusters. The formula for the number of clusters is:

[0049] In the formula, is all the eigenvalue gaps before the o th eigenvalue gap after sorting the eigenvalues corresponding to all nodes; is the o+ 1th eigenvalue gap after sorting the eigenvalues corresponding to all nodes; is the smallest value when reaching o ; Preferably, calculating the singular value entropy of all nodes in each partition specifically includes: For each node in each partition, perform AC-DC power flow calculations to obtain all power flow calculation equations for each node. Take the partial derivatives of each variable in each power flow calculation equation. These partial derivatives form a Jacobian matrix. After performing singular value decomposition on this matrix, normalize the singular values into a probability distribution, and calculate the singular value entropy of the node, specifically:

[0050] In the formula, is the singular value entropy, is the result of normalizing the l th singular value of the Jacobian matrix, is the order of the Jacobian matrix.

[0051] It should be noted that define the sensitivity i of the active power of node as:

[0052] In the formula, is the active power of node i .

[0053] The larger the value of The greater the change, the change in the node load at this time will cause significant fluctuations in the node voltage amplitude and phase angle. From this, it can be reflected the change of the singular value entropy when the node active power changes, thus indicating that it is effective to identify weak nodes by calculating the singular value entropy.

[0054] Embodiment 2 of the present invention proposes a AC-DC power grid partitioning and weak node identification system based on spectral clustering using the method described in Embodiment 1 of the present invention, including a similarity index matrix construction model, an AC-DC coupling strength matrix construction module, a fusion module, a dynamic partitioning module, and a weak node identification module, characterized in that: Similarity index matrix construction model: Calculate the similarity index between any two nodes according to the change of the set state variables between any two nodes in the power grid, and establish a similarity index matrix; AC-DC coupling strength matrix construction module: Calculate the short-circuit ratio between any two nodes in the power grid, and establish an AC-DC coupling strength matrix, where the mutual influence of the AC-DC power grid is taken into account, and for the short-circuit ratio between two DC landing points in the node, the comprehensive short-circuit ratio index introducing multi-infeed of AC-DC is used; Fusion module: Fuse the similarity index matrix and the AC-DC coupling strength matrix to obtain an improved similarity matrix; Dynamic partitioning module: Based on the improved similarity matrix, perform dynamic partitioning using spectral clustering with the number of clusters adaptively determined based on eigenvalues; Weak node identification module: According to the dynamic partitioning result, calculate the singular value entropy of all nodes in each partition, and take the node with the largest singular value entropy in each partition as the weak node.

[0055] Embodiment 3 of the present invention proposes a computer device, including a memory and a processor, the memory stores a computer program, characterized in that when the processor executes the computer program, it implements the steps of the AC-DC power grid partitioning and weak node identification method described in Embodiment 1 of the present invention.

[0056] Embodiment 4 of the present invention proposes a computer-readable storage medium, on which a computer program is stored, characterized in that when the computer program is executed by a processor, it implements the steps of the AC-DC power grid partitioning and weak node identification method described in Embodiment 1 of the present invention.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: still can modify the specific implementation manners of the present invention or make equivalent replacements, and any modification or equivalent replacement without departing from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A method for AC / DC power grid partitioning and weak node identification based on spectral clustering, characterized in that: include: According to the changes of the set state variables between any two nodes in the power grid, the similarity index between any two nodes is calculated, and a similarity index matrix is ​​established; Calculate the short-circuit ratio between any two nodes in the power grid and establish an AC / DC coupling strength matrix, which takes into account the mutual influence of AC and DC power grids. The short-circuit ratio between two DC points in a node is a comprehensive short-circuit ratio index that introduces AC / DC multi-feed. The similarity index matrix and the AC / DC coupling strength matrix are fused to obtain an improved similarity matrix; Based on the improved similarity matrix, spectral clustering is used to adaptively determine the number of clusters based on eigenvalues ​​for dynamic partitioning. According to the dynamic partitioning results, the singular value entropy of all nodes in each partition is calculated, and the node with the largest singular value entropy in each partition is regarded as a weak node.

2. The method for AC / DC power grid partitioning and weak node identification based on spectral clustering according to claim 1, characterized in that: According to the change of the set state variables between any two nodes in the power grid, the similarity index between any two nodes is calculated, and the similarity index matrix is ​​established, which is specifically: Select one of the operating states of the AC and DC power grids as the state variable and collect node data at a set time interval. i and nodes j The state variables of the two nodes in the set time period are combined into two trajectories, using two definitions in The two trajectories are locally stretched or compressed on the time axis based on a continuous non-decreasing function. The maximum distance between each corresponding position of the two trajectories after stretching or compression is solved. The two optimal continuous non-decreasing functions are found to minimize the maximum distance. The minimized maximum distance is the similarity index of the two nodes. The similarity index between any two nodes is calculated, and a similarity index matrix is ​​established.

3. The method for AC / DC power grid partitioning and weak node identification based on spectral clustering according to claim 1, characterized in that: The comprehensive short-circuit ratio index between the nodes of two DC drop points is: Where: DC drop point p With DC drop point q The comprehensive short-circuit ratio indicator between them; DC drop point p The commutation bus voltage; , DC drop point p , q DC power; is the equivalent node impedance matrix seen from each DC conversion bus No. p Row, No. p Column elements; Equivalent node impedance matrix viewed from each DC conversion bus No. p Row, No. q Column elements; For other cases, the short-circuit ratio between two nodes is the reciprocal of the absolute value of the per-unit impedance between the two nodes.

4. The method for AC / DC power grid partitioning and weak node identification based on spectral clustering according to claim 1, characterized in that: The similarity index matrix and the AC / DC coupling strength matrix are fused to obtain an improved similarity matrix, which is specifically: Taking node similarity and electrical distance as two factors, the importance weights of node similarity and electrical distance are obtained through hierarchical analysis method. The importance weight of node similarity is used as the weight of the similarity index matrix, and the importance weight of electrical distance is used as the weight of the AC / DC coupling strength matrix. The similarity index matrix and the AC / DC coupling strength matrix are weighted to obtain an improved similarity matrix.

5. The method for AC / DC power grid partitioning and weak node identification based on spectral clustering according to claim 4, characterized in that: Based on the improved similarity matrix, spectral clustering based on eigenvalue adaptive determination of cluster number is used for dynamic partitioning, specifically: According to the improved similarity matrix, the adjacency matrix is ​​calculated. The degree of each node can be calculated by performing row-wise summation operation on the adjacency matrix. All degree values ​​are used as diagonal elements to form a diagonal matrix, which is the degree matrix. Subtract the adjacency matrix from the degree matrix to get the Laplacian matrix, and normalize the Laplacian matrix with the help of the degree matrix; Perform eigenvalue decomposition on the normalized Laplace matrix, sort the decomposed eigenvalues, select the eigenvectors corresponding to the largest set number of eigenvalues, arrange these eigenvectors in columns to form an eigenvector matrix, and use each row in the eigenvector matrix as a clustering sample for K-means clustering.

6. The method for AC / DC power grid partitioning and weak node identification based on spectral clustering according to claim 5, characterized in that: The adjacency matrix is ​​calculated according to the improved similarity matrix, specifically: The improved similarity matrix i Line j The elements of the column represent i Nodes and j The square of the distance between nodes is used to set the scale parameter, which determines the decay rate of the similarity between nodes. The Gaussian kernel function is used to calculate the adjacency matrix based on the improved similarity matrix. The formula is: In the formula, The adjacency matrix i Line j Elements of a column; The improved similarity matrix i Line j Elements of a column; Set the scale parameter.

7. The method for AC / DC power grid partitioning and weak node identification based on spectral clustering according to claim 6, characterized in that: The adaptive determination of the number of clusters based on the eigenvalue is specifically as follows: Arrange the eigenvalues ​​of the Laplace matrix from large to small, calculate the difference between each eigenvalue and its next eigenvalue and use it as the eigenvalue gap to form an eigenvalue gap sequence, and find the first maximum value in the eigenvalue gap sequence in turn. The label of this value in the eigenvalue gap sequence is the cluster number.

8. The method for AC / DC power grid partitioning and weak node identification based on spectral clustering according to claim 1, characterized in that: The calculation of the singular value entropy of all nodes in each partition is specifically as follows: For each node in each partition, AC and DC power flow calculations are performed to obtain all power flow calculation equations for each node. Partial derivatives are calculated for each variable in each power flow calculation equation. These partial derivatives form a Jacobian matrix. After performing singular value decomposition on the matrix, the singular values ​​are normalized to probability distribution, and the singular value entropy of the node is calculated, which is specifically: In the formula, is the singular value entropy, is the Jacobian matrix l The result of singular value normalization is is the order of the Jacobian matrix.

9. An AC / DC power grid partitioning and weak node identification system based on spectral clustering using the method according to any one of claims 1 to 8, comprising a similarity index matrix construction model, an AC / DC coupling strength matrix construction module, a fusion module, a dynamic partitioning module and a weak node identification module, characterized in that: Similarity index matrix construction model: According to the changes of the set state variables between any two nodes in the power grid, the similarity index between any two nodes is calculated and the similarity index matrix is ​​established; AC / DC coupling strength matrix construction module: Calculate the short-circuit ratio between any two nodes in the power grid and establish an AC / DC coupling strength matrix, which takes into account the mutual influence of AC and DC power grids. The short-circuit ratio between two DC points in a node is a comprehensive short-circuit ratio indicator that introduces AC / DC multi-feed. Fusion module: fuses the similarity index matrix and the AC / DC coupling strength matrix to obtain an improved similarity matrix; Dynamic partitioning module: Based on the improved similarity matrix, dynamic partitioning is performed using spectral clustering based on adaptive determination of the number of clusters based on eigenvalues; Weak node identification module: According to the dynamic partitioning results, the singular value entropy of all nodes in each partition is calculated, and the node with the largest singular value entropy in each partition is regarded as a weak node.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method for AC / DC power grid partitioning and weak node identification based on spectral clustering as claimed in any one of claims 1 to 8 are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for AC / DC power grid partitioning and weak node identification based on spectral clustering as claimed in any one of claims 1 to 8 are implemented.

Images