Micro-grid group dynamic division method and system based on graph theory and non-dominated sorting

Abstract

The invention discloses a micro-grid group dynamic division method and system based on a graph theory and non-dominated sorting, and the method comprises the steps: generating a graph theory model of a micro-grid group through a weighted undirected graph according to the corresponding data of the micro-grid group; based on a graph theory model, respectively defining an improved modularity index, a power matching degree index and a partition supply and demand coordination index; dividing a micro-grid group by using a second-generation non-dominated sorting genetic algorithm; calculating difference values of division indexes at different moments and dynamically updating the weighted adjacency matrix; by calculating the target gradient change rate of the solution, selecting the solution with the most balanced tradeoff between the targets; according to the updated weighted adjacency matrix, outputting a visual division result of the micro-grid group; and iteratively updating the division result, and outputting a final division result of the micro-grid group. According to the method, the problem that electrical network parameter characteristics are difficult to characterize, the problem of the number of divided partitions of the microgrid group and the problem of partitions to which the microgrids belong, and the problem that nonlinear conflicts exist among targets are solved.

Description

Technical Field

[0001] The present invention belongs to the field of power grids, relates to the technology of microgrid optimization and control, and particularly relates to a method and system for dynamically partitioning a microgrid group based on graph theory and non-dominated sorting. Background Art

[0002] Affected by the development of new power systems and power electronics technologies, the number of adjustable resources in a microgrid group (such as flexible adjustable loads represented by air conditioners and electric vehicles, distributed controllable power generation devices represented by diesel generators and fuel cells, and various energy storage devices, etc.) has increased exponentially, and the adjustable parameters have transformed from a single dimension to multiple dimensions. Existing centralized and distributed optimization methods are difficult to obtain an optimal scheduling strategy within a limited time. Based on this, current research mostly adopts the method of "dimensionality reduction control" to solve the above problems, that is, dividing the microgrid group into multiple self-sufficient partitions, and each partition realizes the energy management and control of the microgrid group through the optimization within the microgrid and the coordinated optimization between the microgrids. This method only requires the microgrids within the partition to exchange information, greatly reducing the communication requirements and the amount of information transmission, and simplifying the complexity of controlling a large number of devices.

[0003] The reasonable allocation, scheduling, and energy management of various internal resources are based on the reasonable division of regions. Reasonably designing division indicators and adopting efficient division algorithms are the basis and key of regional division. Division indicators can be roughly divided into structural indicators and functional indicators. Structural indicators aim at close connections between nodes within a sub-region and loose connections between regions, and mainly include indicators such as modularity, node membership, and stability; functional indicators aim at regional optimization and control, and different optimization and control requirements will result in certain differences in functional indicators in different studies, mainly including indicators such as power balance degree, cluster accommodation capacity, and source-load-storage matching degree. Division algorithms mainly include clustering algorithms, intelligent optimization algorithms, and community detection algorithms based on complex networks. It should be noted that existing research mainly focuses on typical operating scenarios and fixed network topologies, and mostly uses fixed division methods to divide the microgrid group, and its division results do not change with time. However, the fixed division method does not consider the temporal variation and spatial migration of renewable energy output and load, which may lead to the division results of the fixed microgrid group being unable to adapt to the dynamic spatio-temporal power changes of renewable energy and load or the changes in network topology, making its optimization strategy unable to adapt to energy management in different operating scenarios. On this basis, some research has proposed dynamic division methods. The dynamic division method can update the division results of the microgrid group according to the real-time operating state of the system, which helps to improve the applicability of the optimization strategy under different operating scenarios. Therefore, constructing reasonable division indicators and using efficient division algorithms to dynamically divide the microgrid group helps to improve the applicability of the optimization strategy of the microgrid group in different operating scenarios.

[0004] The existing technical solution mainly combines the reactive power balance degree, active power balance degree index and modularity index as the division index. Then, the adjacency matrix in the undirected graph is used to encode the chromosomes in the genetic algorithm, and new individuals are generated by performing operations such as crossover and mutation on the adjacency matrix. The changed adjacency matrix consists of multiple block matrices, and each block matrix represents a partition in the division result of the distribution network. The row or column number of the non-zero square matrix in the block matrix represents the distribution network node number in that partition. However, this solution has the following disadvantages:

[0005] 1. Since the adjacency matrix in the undirected graph only contains two elements, 0 and 1, it can only represent whether there is a line connection between microgrids and cannot represent electrical parameter characteristics such as the resistance and transmission capacity of the line.

[0006] 2. The nodes in the same block matrix in the adjacency matrix are classified into the same partition, and the division result of the microgrid group can be directly obtained from the block matrix. However, when the node numbers in the same partition in the adjacency matrix are not continuous, there will be multiple block matrices in the same partition in the adjacency matrix, and the division result of the microgrid group cannot be directly obtained from the block matrix in the adjacency matrix. It is necessary to convert the adjacency matrix into the form of an undirected graph to obtain the number of division partitions of the microgrid group and the partitions to which each microgrid belongs, and then calculate the modularity index, power matching degree index and partition supply and demand coordination index, resulting in low algorithm operation efficiency.

[0007] 3. In the existing technology, multiple division indexes are usually normalized by the linear weighting method and then weighted and summed after multiplying by the weight coefficient. First, the weight setting in the linear weighting method usually depends on subjective judgment or repeated trial and error and lacks a theoretical basis; second, the linear weighting method assumes that the division indexes are independent of each other, but in fact, there are non-linear conflicts between different division indexes. Summary of the Invention

[0008] Objective of the Invention: To overcome the deficiencies in the prior art, a dynamic partitioning method and system for microgrid clusters based on graph theory and non-dominated sorting are provided. By using a weighted undirected graph to construct a graph theory model of the microgrid cluster, a weighted adjacency matrix containing electrical parameters such as line impedance and transmission capacity is obtained to solve the problem that it is difficult for the adjacency matrix encoding based on the undirected graph to characterize the characteristics of electrical network parameters; the depth first search (DFS) algorithm is used to identify the connected components in the weighted adjacency matrix, and the number of partitions of the microgrid cluster and the partition numbers to which each microgrid belongs are obtained to solve the problem that when the numbering order of the microgrids in the microgrid cluster is chaotic, the number of partitions of the microgrid cluster and the partitions to which each microgrid belongs cannot be directly obtained from the block matrix in the adjacency matrix; the second-generation non-dominated sorting genetic algorithm (NSGA-II) is used to optimize different partitioning indicators simultaneously to solve the problems of the existing linear weighted method for partitioning indicators relying on subjective weight setting and the non-linear conflict between multiple optimization objectives.

[0009] Technical Solution: To achieve the above objective, the present invention provides a dynamic partitioning method for microgrid clusters based on graph theory and non-dominated sorting, including the following steps:

[0010] S1: According to the data corresponding to the microgrid cluster, use a weighted undirected graph to generate a graph theory model of the microgrid cluster;

[0011] S2: Based on the graph theory model, define the improved modularity index, power matching degree index, and partition supply-demand coordination index respectively;

[0012] S3: According to the indicators defined in step S2, use the second-generation non-dominated sorting genetic algorithm to partition the microgrid cluster;

[0013] S4: Calculate the difference of the partitioning indicators at different times and dynamically update the weighted adjacency matrix;

[0014] S5: By calculating the objective gradient change rate of the solution, select the solution with the most balanced trade-off between objectives;

[0015] S6: According to the updated weighted adjacency matrix, output the visualized partitioning result of the microgrid cluster;

[0016] S7: Iteratively update the partitioning result and output the final partitioning result of the microgrid cluster.

[0017] Further, the data corresponding to the microgrid cluster in step S1 includes the microgrid cluster topology structure, node numbers, line resistances and transmission capacities, photovoltaic power outputs, wind turbine power outputs, and load forecast power data of each microgrid.

[0018] Furthermore, in step S1, a weighted undirected graph is used. Generate a graph model of microgrid clusters. The microgrid cluster is constructed as a vertex set , Representative Microgrids, edge collection represents the transmission lines connecting adjacent microgrids, and each edge in the edge set , weighted adjacency matrix Used to store the network topology and weight information of the microgrid group.

[0019] Furthermore, the weighted adjacency matrix of step S1 is middle:

[0020]

[0021] In the formula, the edge weight set For storage of transmission lines Resistance And transport capacity .

[0022] Furthermore, the definition of the improved modularity index in step S2 includes:

[0023] Introducing modularity index The indicators are quantified and expressed as follows:

[0024]

[0025] In the formula, for The weighted adjacency matrix at the moment Line Elements of a column; express 1 / 2 of the sum of the weights of all edges in the partition at that moment; express All nodes at the moment The sum of the weights of the connected edges; express All nodes at the moment The sum of the weights of the connected edges; Used to indicate Time Node and Are they in the same partition and connected? ,express Time Node and In the same partition and connected, express Time Node and Not connected if not in the same partition.

[0026] Furthermore, the definition of the power matching degree index in step S2 includes:

[0027] Introduce the power matching degree index to quantify the index, and the expression is as follows:

[0028]

[0029]

[0030]

[0031]

[0032] In the formula, and respectively represent the power matching degree index values of all partitions and the average value of the power matching degree index of the th partition at time and respectively represent the number of partitions and the number of microgrids in the th partition; and respectively represent the adjustable active power and the net load power in the th microgrid at time , , , and respectively represent the adjustable active powers of PV, wind turbines, adjustable loads, energy storage, and diesel generators in the th microgrid at time , and respectively represent the real-time active powers of PV, wind turbines, and loads in the th microgrid at time

[0033] Furthermore, the definition of the partition supply-demand coordination index in step S2 includes:

[0034] Introduce the partition supply-demand coordination index for quantification, and the expression is as follows:

[0035]

[0036]

[0037]

[0038] In the formula, and respectively represent the average value of the supply - demand coordination index of the time - period partition and the supply - demand coordination index value of the th partition; represents the difference between the adjustable capacity of the controllable electrical equipment at time and the net load power within the

[0039] Further, the process of using the second - generation non - dominated sorting genetic algorithm to divide the micro - grid cluster in step S3 includes:

[0040] A1: Initialize parameters, including population size, maximum number of iterations, crossover probability range, mutation probability range, dynamic update threshold;

[0041] A2: Encode the chromosomes according to the weighted adjacency matrix ;

[0042] A3: Initialize the population, including: randomly changing the upper - triangular part of the initial weighted matrix to 0 or keeping it unchanged, and then synchronously copying the upper - triangular elements to the lower - triangular part to ensure the symmetry of the weighted matrix, and randomly generating the population;

[0043] A4: Weight threshold filtering, including: performing partition detection on the generated weighted adjacency matrix, and only retaining the edges that satisfy to construct the binary adjacency matrix , as follows:

[0044]

[0045] A5: Traverse and identify connected components, including: using the depth - first search algorithm to traverse and identify connected components; mark all nodes as unvisited, starting from any unvisited node , recursively visit its adjacent nodes along the valid edges ( ); all visited nodes are marked as the same partition and assigned a unique partition label number, repeat this process until all nodes are visited; after the traversal, the number of different labels is the number of partitions, and the label number of each node represents its partition number;

[0046] A6: Based on the initial partition results of each individual in the population obtained from the partition detection, use multi - thread parallel computing to calculate the fitness function values of the modularity index, power matching index, and partition supply - demand coordination index for each individual in the population;

[0047] A7: Non - dominated sorting, including: performing fast non - dominated sorting on the population to divide the Pareto front levels;

[0048] A8: Target crowding degree calculation, including: calculating the crowding degree of each solution in the target space:

[0049]

[0050] In the formula, represents the crowding degree of the th individual at time is the th objective function value; and respectively represent the maximum and minimum values of all individuals under the objective functions ;

[0051] A9: Elite selection, including: selecting 2 individuals with the highest non-dominated rank and the largest crowding degree from the population;

[0052] A10: Crossover operation, including: performing crossover on the elements in the th row and th column of the upper triangular part of the binary adjacency matrix of the selected two parents, and then synchronously copying the crossed elements in the upper triangular part of the binary adjacency matrix of the parents to the lower triangular part to maintain the symmetry of the binary adjacency matrix of the parents;

[0053] A11: Mutation operation, including: when the element in the th row and th column of the upper triangular part of the binary adjacency matrix of the offspring is , randomly mutate it to or ; when the element in the th row and th column of the upper triangular part of the binary adjacency matrix of the offspring is , only when the element in the th row and th column of the upper triangular part of the original binary adjacency matrix is , randomly mutate it to or , and when the element in the upper triangular part of the original binary adjacency matrix is , the element in the upper triangular part of the binary adjacency matrix of the offspring remains unchanged; subsequently, synchronously copy the mutated elements in the upper triangular part of the binary adjacency matrix of the offspring to the lower triangular part to maintain the symmetry of the binary adjacency matrix of the offspring;

[0054] A12: Elite retention, including: retaining high-quality solutions in the non-dominated front, avoiding loss of potential optimal solutions, and at the same time merging the parent and offspring populations.

[0055] Further, the specific steps of step S4 include:

[0056] Judge whether the modularity index function value obtained from the binarized adjacency matrix at time two is the same as the modularity index function value obtained from the binarized adjacency matrix at time one; if the same, do not update the binarized adjacency matrix; if different, then judge whether the modularity index function value obtained from the binarized adjacency matrix at time one satisfies , if and at the power matching degree index and the partition supply - demand coordination index function value obtained from the binarized adjacency matrix at time two and the difference between the power matching degree index obtained from the binarized adjacency matrix at time one and the difference between the partition supply - demand coordination index function values exceed the threshold, update the binarized adjacency matrix; the expression is as follows:

[0057]

[0058] In the formula: and are respectively the binarized adjacency matrices at time one and time two; is the dynamic update threshold;

[0059] When there is an update in the binarized adjacency matrix at time one , obtain the corresponding weighted adjacency matrix through the binarized adjacency matrix at time one .

[0060] The present invention also provides a dynamic partitioning system for a micro - grid group based on graph theory and non - dominated sorting, including:

[0061] A graph - theory model generation module, which is used to generate a graph - theory model of the micro - grid group using a weighted undirected graph;

[0062] An index definition module, which is used to define the improved modularity index, power matching degree index, and partition supply - demand coordination index respectively;

[0063] A partitioning module, which is used to partition the micro - grid group;

[0064] An update module, which is used to dynamically update the weighted adjacency matrix;

[0065] An iterative output module, which is used to iteratively update the partitioning result and output the final partitioning result of the micro - grid group.​

[0066] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0067] 1. By using a weighted undirected graph to construct a graph theory model of the microgrid group, on this basis, a weighted adjacency matrix containing electrical parameters such as line impedance and transmission capacity is used to encode the chromosome to achieve double coding of network topology and electrical characteristics, making the partitioning result more in line with actual physical constraints.

[0068] 2. Use the depth-first search algorithm to identify the connected components in the weighted adjacency matrix to directly obtain the partitioning result of the microgrid group, avoiding the steps of converting the adjacency matrix into an undirected graph form and then reading the partitioning result from the undirected graph, effectively improving the operation efficiency of the algorithm.

[0069] 3. Use the second-generation non-dominated sorting genetic algorithm to optimize multiple objectives respectively. Through non-dominated sorting and crowding degree calculation, a uniformly distributed Pareto optimal solution set is generated, avoiding defects such as subjectivity in weight selection, single solution set, and poor adaptability caused by dynamically modifying weights through conditional judgment. Brief description of the drawings

[0070] Figure 1 It is a schematic flow diagram of the method of the present invention;

[0071] Figure 2 It is a schematic diagram of the graph theory model of the microgrid group;

[0072] Figure 3 It is a schematic diagram of the chromosome coding method;

[0073] Figure 4 It is a diagram of the dynamic partitioning result of the microgrid group. Specific implementation manners

[0074] The following further clarifies the present invention in conjunction with the drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. After reading the present invention, various equivalent forms of modification of the present invention by those skilled in the art all fall within the scope defined by the appended claims of this application.

[0075] Embodiment 1:

[0076] As Figure 1 shown, this embodiment provides a dynamic partitioning method for a microgrid group based on graph theory and non-dominated sorting, including the following steps:

[0077] Step 1: Input the topology structure of the microgrid group, node numbers, line resistances and transmission capacities, photovoltaic power generation, wind turbine power generation, and load prediction power data of each microgrid.

[0078] Step 2: Use a weighted undirected graph Generate the graph theory model of the microgrid group, as Figure 2 shown. The microgrid group is constructed as a vertex set , representing the th microgrid. The edge set represents the transmission lines connecting adjacent microgrids. Each edge in the edge set . The weighted adjacency matrix is used to store the network topology structure and weight information of the microgrid group, as follows:

[0079]

[0080] In the formula, the weight set of the edges is used to store the resistance of the transmission line and the transmission capacity .

[0081] Step 3: Define the improved modularity index.

[0082] To ensure that the partitions obtained after the division of the microgrid group maintain sufficient structural strength, the modularity index is introduced to quantify it, as follows:

[0083]

[0084] In the formula, is the element in the th row and th column of the weighted adjacency matrix at time; represents 1 / 2 of the sum of the weights of all edges within the partition at time; represents the sum of the weights of all edges connected to node at time; represents the sum of the weights of all edges connected to node at time; is used to represent whether nodes and are in the same partition and connected at time, , representing that nodes and are in the same partition and connected at time, represents that nodes and are not in the same partition and not connected at time.

[0085] Step 4: Define the power matching degree index.

[0086] The randomness and volatility of renewable energy and load power increase the net load fluctuation. To utilize the adjustable resources in the microgrid cluster to suppress the net load fluctuation, a power matching degree index is introduced to quantify it, which is expressed as follows:

[0087]

[0088]

[0089]

[0090]

[0091] In the formula, and respectively represent the values of the power matching degree index for all partitions and the average value of the power matching degree index for the th partition at time and respectively represent the number of partitions and the number of microgrids in the th partition; and respectively represent the adjustable active power and the net load power in the th microgrid at time , , , and respectively represent the adjustable active power of PV, wind turbine, adjustable load, energy storage and diesel generator in the th microgrid at time , and respectively represent the real-time active power of PV, wind turbine and load in the th microgrid at time

[0092] Step 5: Define the partition supply-demand coordination index.

[0093] To reflect the supply-demand coordination relationship between distributed energy storage and net load in the microgrid cluster division, and at the same time show the coordination degree between adjustable resources and system operation requirements in each partition, a partition supply-demand coordination index is introduced for quantification, which is defined as follows:

[0094]

[0095]

[0096]

[0097] In the formula, and respectively represent the average value of the supply - demand coordination index in the time - period partition and the supply - demand coordination index value of the th partition; represents the difference between the adjustable capacity of the controllable electrical equipment at time and the net load power in the

[0098] Step 6: Use the second - generation non - dominated sorting genetic algorithm to partition the micro - grid group. The specific process includes the following steps:

[0099] Step 6.1: Initialize parameters. Set the population size to 200, the maximum number of iterations to 50, the crossover probability range to [0.2, 0.5], the mutation probability range to [0.1, 0.3], and the dynamic update threshold to .

[0100] Step 6.2: Encode the chromosomes according to the weighted adjacency matrix generated in Step 2 as shown in Figure 3 .

[0101] Step 6.3: Initialize the population. Randomly change the elements in the upper - triangular part of the initial weighted matrix to 0 or keep them unchanged, and then synchronously copy the upper - triangular elements to the lower - triangular part to ensure the symmetry of the weighted matrix, and randomly generate 200 populations.

[0102] Step 6.4: Weight - threshold filtering. Perform partition detection on the generated weighted adjacency matrix, and only retain the edges that satisfy to construct a binary adjacency matrix as follows:

[0103]

[0104] Step 6.5: Traverse and identify connected components.

[0105] Use the depth - first search algorithm to traverse and identify connected components; mark all nodes as unvisited, start from any unvisited node , and recursively visit its adjacent nodes along the valid edges ( ); all visited nodes are marked as the same partition and assigned a unique partition label number. Repeat this process until all nodes are visited; after the traversal, the number of different labels is the number of partitions, and the label number of each node represents its partition number.

[0106] ​Step 6.6: Based on the initial partition results of each individual in the population obtained from partition detection, use multi-threaded parallel computing to calculate the fitness function values of the modularity index, power matching degree index, and partition supply-demand coordination index for each individual in the population.

[0107] Step 6.7: Non-dominated sorting.

[0108] Perform fast non-dominated sorting on the population to divide the Pareto front levels.

[0109] Step 6.8: Target crowding degree calculation.

[0110] Calculate the crowding degree of each solution in the target space to ensure the diversity of the front solutions. The calculation formula is as follows:

[0111]

[0112] In the formula, represents the crowding degree of the th individual at time ; is the and respectively represent the maximum and minimum values of all individuals in objective functions under.

[0113] Step 6.9: Elite selection. Select the 2 individuals with the highest non-dominated rank and the largest crowding degree from the population.

[0114] Step 6.10: Crossover operation. For the elements in the th row and th column of the upper triangle of the binary adjacency matrix of the two selected parents, perform crossover. Then synchronously copy the crossed elements of the upper triangle of the binary adjacency matrix of the parents to the lower triangle to maintain the symmetry of the binary adjacency matrix of the parents.

[0115] Step 6.11: Mutation operation. When the element in the th row and th column of the upper triangle of the binary adjacency matrix of the offspring, randomly mutate it to or ; when the element in the th row and th column of the upper triangle of the binary adjacency matrix of the offspring, only when the element in the th row and th column of the upper triangle of the original binary adjacency matrix is or When the element in the upper triangle of the original binary adjacency matrix is, the element in the upper triangle of the offspring binary adjacency matrix remains unchanged; subsequently, the mutated elements in the upper triangle of the offspring binary adjacency matrix are synchronously copied to the lower triangle to maintain the symmetry of the offspring binary adjacency matrix.

[0116] Step 6.12: Elite retention. Retain the high-quality solutions in the non-dominated front to avoid losing potential optimal solutions, and at the same time merge the parent and offspring populations.

[0117] Step 7: Calculate the difference in the partitioning index at different times and dynamically update the binary adjacency matrix. Judge whether the value of the modularity index function obtained from the binary adjacency matrix at time is the same as the value of the modularity index function obtained from the binary adjacency matrix at time . If they are the same, the binary adjacency matrix is not updated; if they are not the same, judge whether the value of the modularity index function obtained from the binary adjacency matrix at time satisfies . And if , and when the difference between the power matching degree index and the partition supply-demand coordination index function value obtained from the binary adjacency matrix at time and the difference between the power matching degree index and the partition supply-demand coordination index function value obtained from the binary adjacency matrix at time exceed the threshold, update the binary adjacency matrix. Specifically, it is as follows: where:

[0118]

[0119] In the formula: and are the binary adjacency matrices at time and time respectively; is the dynamic update threshold.

[0120] Step 8: Solution set screening, select the solution with the most balanced trade-off between objectives by calculating the target gradient change rate of the solution.

[0121] Step 9: When there is an update in the binary adjacency matrix at time , obtain the corresponding weighted adjacency matrix through the binary adjacency matrix at time , and update the visualization partition result of the microgrid group at time based on the weighted adjacency matrix . Update the output the visualization partition result of the microgrid group at time

[0122] Step 10: Determine whether the current moment is the termination moment. If it is not the termination moment, perform time-series rolling optimization. Use the current Pareto solution set as the initial population, divide the microgrid according to the operating state at the moment, and repeat steps 6 to 9; if it is the termination moment, stop the operation. The final dynamic division result of the microgrid group obtained in this embodiment is as Figure 4 shown.

[0123] Embodiment 2:

[0124] Based on the method of Embodiment 1, this embodiment provides a dynamic division system for a microgrid group based on graph theory and non-dominated sorting, including:

[0125] A graph theory model generation module, configured to generate a graph theory model of the microgrid group using a weighted undirected graph;

[0126] An index definition module, configured to respectively define an improved modularity index, a power matching degree index, and a partition supply-demand coordination index;

[0127] A division module, configured to divide the microgrid group;

[0128] An update module, configured to dynamically update the weighted adjacency matrix;

[0129] An iterative output module, configured to iteratively update the division result and output the final division result of the microgrid group.

Claims

1. A dynamic partitioning method for microgrid clusters based on graph theory and non-dominated sorting, characterized in that, It includes the following steps: S1: Generate a graph theory model of the microgrid cluster using a weighted undirected graph based on the corresponding data of the microgrid cluster; S2: Based on the graph theory model, define the improved modularity index, power matching degree index, and partition supply-demand coordination index respectively; S3: Divide the microgrid cluster using the second-generation non-dominated sorting genetic algorithm according to the indexes defined in step S2; S4: Calculate the difference of the partition indexes at different times and dynamically update the weighted adjacency matrix; S5: Select the solution with the most balanced trade-off between objectives by calculating the objective gradient change rate of the solution; S6: Output the visualized partition result of the microgrid cluster according to the updated weighted adjacency matrix; S7: Iteratively update the partition result and output the final partition result of the microgrid cluster.

2. The dynamic partitioning method for microgrid clusters based on graph theory and non-dominated sorting according to claim 1, characterized in that The corresponding data of the microgrid cluster in step S1 includes the microgrid cluster topology structure, node numbers, line resistances and transmission capacities, PV output power, wind turbine output power, and load prediction power data of each microgrid.

3. A dynamic partitioning method for a microgrid group based on graph theory and non-dominated sorting according to claim 1, characterized in that In the step S1, a weighted undirected graph is used to generate a graph theory model of the microgrid cluster, and the microgrid cluster is constructed as a vertex set , representing the th microgrid, and the edge set represents the transmission line connecting adjacent microgrids. Each edge in the edge set , and the weighted adjacency matrix is used to store the network topology structure and weight information of the microgrid cluster.

4. A dynamic partitioning method for a microgrid group based on graph theory and non-dominated sorting according to claim 3, characterized in that The weighted adjacency matrix of step S1 wherein: ； In the formula, the edge weight set used to store the transmission line resistance and transmission capacity .

5. A dynamic partitioning method for a microgrid group based on graph theory and non-dominated sorting according to claim 4, characterized in that The definition of the improved modularity index in step S2 includes: Introduce the modularity index Quantify the index, expressed as follows: ； In the formula, is the element in the th row and th column of the time-weighted adjacency matrix; represents 1 / 2 of the sum of the weights of all edges within the time partition; represents the sum of the weights of all edges connected to node at time; represents the sum of the weights of all edges connected to node at is used to represent whether nodes and are in the same partition and connected at time; , represents that nodes and are in the same partition and connected at time; represents that nodes and are not in the same partition and not connected at time.

6. A dynamic partitioning method for a microgrid group based on graph theory and non-dominated sorting according to claim 5, characterized in that, The definition of the power matching degree index in step S2 includes: Introduce the power matching degree index to quantify the index, and the expression is as follows: ； ； ； ； Wherein, and respectively represent the power matching degree index values of all partitions and at the -th partition, the average value of the power matching degree index; and respectively represent the number of partitions and the number of microgrids in the -th partition; and respectively represent at the -th microgrid, the adjustable active power and the net load power; , , , and respectively represent at the -th microgrid, the adjustable active powers of PV, wind turbine, adjustable load, energy storage and diesel generator; , and respectively represent at the -th microgrid, the real-time active powers of PV, wind turbine and load.

7. A dynamic partitioning method for a microgrid group based on graph theory and non-dominated sorting according to claim 6, characterized in that The definition of the partition supply-demand coordination index in step S2 includes: Introduce the partition supply-demand coordination index for quantification, and the expression is as follows: ； ； ； In the formula, and respectively represent the average value of the supply-demand coordination index in the time partition and the supply-demand coordination index value of the th partition; represents the difference between the adjustable capacity of the controllable electrical equipment at the th moment and the net load power in the th partition.

8. A dynamic partitioning method for microgrid clusters based on graph theory and non-dominated sorting according to claim 7, characterized in that, The process of dividing the microgrid cluster using the second-generation non-dominated sorting genetic algorithm in step S3 includes: A1: Initialize parameters, including population size, maximum number of iterations, crossover probability range, mutation probability range, and dynamic update threshold; A2: Encode the chromosome according to the weighted adjacency matrix Encode the chromosome; A3: Initialize the population, including: randomly changing to 0 or remaining unchanged the elements in the upper triangular part of the initial weighted matrix, and then synchronously copying the upper triangular elements to the lower triangular part to ensure the symmetry of the weighted matrix, and randomly generating the population; ​ A4: Weight threshold filtering, including: performing partition detection on the generated weighted adjacency matrix and only retaining the edges that satisfy to construct a binary adjacency matrix as follows: ； A5: Traverse and identify connected components, including: using the depth-first search algorithm to traverse and identify connected components; mark all nodes as unvisited, starting from any unvisited node begin, recursively visit its adjacent nodes along valid edges; all visited nodes are marked as the same partition and assigned a unique partition label number, repeat this process until all nodes are visited; after the traversal is completed, the number of different labels is the number of partitions, and the label number of each node represents the partition number it belongs to; A6: Based on the initial partition results of each individual in the population obtained by partition detection, use multi-threaded parallel computing to calculate the fitness function values of the modularity index, power matching degree index, and partition supply-demand coordination index of each individual in the population; A7: Non-dominated sorting, including: performing fast non-dominated sorting on the population and dividing the Pareto front ranks; A8: Objective crowding degree calculation, including: calculating the crowding degree of each solution in the objective space: ； In the formula, represents the crowding degree of the th individual at time ; is the th objective function value; and respectively represent the maximum and minimum values of all individuals under the objective functions. A9: Elite selection, including: selecting 2 individuals with the highest non-dominated rank and the largest crowding degree from the population; A10: Crossover operation, including: for the elements in the row and column of the upper triangle of the binary adjacency matrix of the two selected parents, perform crossover, and then synchronously copy the crossed elements of the upper triangle of the binary adjacency matrix of the parents to the lower triangle to maintain the symmetry of the binary adjacency matrix of the parents; A11: Mutation operation, including: when the element at the row and column in the upper triangle of the offspring binary adjacency matrix is, randomly mutate it to or ; when the element at the row and column in the upper triangle of the offspring binary adjacency matrix is, only when the element at the row and column in the upper triangle of the original binary adjacency matrix is, randomly mutate it to or , when the element in the upper triangle of the original binary adjacency matrix is, the element in the upper triangle of the offspring binary adjacency matrix remains unchanged; subsequently, synchronously copy the mutated elements in the upper triangle of the offspring binary adjacency matrix to the lower triangle to maintain the symmetry of the offspring binary adjacency matrix; A12: Elite retention, including: retaining the high-quality solutions in the non-dominated front, avoiding losing potential optimal solutions, and merging the parent and offspring populations at the same time.

9. A dynamic partitioning method for a microgrid group based on graph theory and non-dominated sorting according to claim 8, characterized in that, Step S4 specifically includes: Judge whether the modularity index function value obtained from the binarized adjacency matrix at time is the same as the modularity index function value obtained from the binarized adjacency matrix at time; if they are the same, the binarized adjacency matrix is not updated; if they are different, judge whether the modularity index function value obtained from the binarized adjacency matrix at time satisfies , if and at the power matching degree index and the partition supply-demand coordination index function value obtained from the binarized adjacency matrix at time and the difference in the power matching degree index obtained from the binarized adjacency matrix at time and the difference between the partition supply-demand coordination index function values exceed the threshold, update the binarized adjacency matrix; the expression is as follows: ； Wherein: and are respectively the time and the binary adjacency matrix at the time; is the dynamically updated threshold value; When the binary adjacency matrix at a moment has an update, through the binary adjacency matrix at a moment obtain the corresponding weighted adjacency matrix .

10. A microgrid group dynamic partitioning system based on graph theory and non-dominated sorting, characterized in that, It includes: A graph theory model generation module for generating a graph theory model of the microgrid cluster using a weighted undirected graph; An index definition module for defining the improved modularity index, power matching degree index, and partition supply-demand coordination index respectively; A division module for dividing the microgrid cluster; An update module for dynamically updating the weighted adjacency matrix; An iterative output module for iteratively updating the partition result and outputting the final partition result of the microgrid cluster.

Images