Multi-level power distribution network cluster division method and device

Abstract

The invention belongs to the technical field of power distribution network cluster division, and particularly relates to a multi-level power distribution network cluster division method and device, and the method comprises the steps: building a high-voltage class cluster division model based on a modularity index and a power balance index, and outputting a cluster division result through a K-means algorithm; on the basis of the inter-cluster power exchange index and the cluster dynamic fitness index, constructing a low-voltage class cluster division model, and outputting a sub-cluster division result of each cluster by using a genetic algorithm on the basis of a cluster division result by adopting a dynamic sub-cluster division optimization method; and establishing a multi-level optimization model and constraint conditions thereof, and based on the multi-level optimization model, coordinating objective functions of the high-voltage-level cluster division model and the low-voltage-level cluster division model by using a multi-level optimization algorithm to realize global optimization. According to the invention, collaborative optimization of high and low voltages is realized, power loss and voltage deviation are reduced, and dynamic adaptability to distributed energy is improved.

Description

Technical Field

[0001] This invention belongs to the field of power distribution network cluster division technology, specifically relating to a multi-level power distribution network cluster division method and apparatus. Background Technology

[0002] With the transformation of the energy structure and the rapid development of renewable energy, the penetration rate of distributed energy in distribution networks has increased significantly. Modern distribution networks have evolved from the traditional unidirectional power supply model to complex networks encompassing various distributed power sources, energy storage devices, and dynamic loads. Against this backdrop, distribution network optimization methods are also constantly evolving.

[0003] Photovoltaic power generation is widely used in power distribution networks due to its clean and low-carbon characteristics. However, photovoltaic output is affected by factors such as weather and time, exhibiting significant intermittency and uncertainty, which poses challenges to the power balance and voltage stability of the power distribution network. Smart grids emphasize efficient, reliable, and adaptive power management, requiring the power distribution network to have real-time optimization capabilities to cope with the dynamic changes in distributed energy resources and loads. Traditional power distribution network optimization methods are mostly based on static models, which are difficult to meet the dynamic needs of smart grids.

[0004] Cluster partitioning, as an important means of distribution network optimization, achieves power allocation and voltage control by grouping grid nodes. However, existing research mostly focuses on partitioning at single voltage levels, lacking consideration for the coordinated operation of high and low voltage levels, thus limiting optimization effectiveness. Furthermore, optimization algorithms for large-scale distribution networks have high computational complexity, making real-time applications difficult.

[0005] Traditional distribution network clustering methods often focus on a single voltage level, making it difficult to balance global power balance at high voltage levels with local voltage stability at low voltage levels. This results in low energy utilization efficiency and significant voltage fluctuations. Distributed energy sources such as photovoltaic power generation exhibit strong randomness and temporal volatility, making traditional static clustering methods ineffective in addressing dynamically changing load and generation demands, which can easily lead to power imbalances or voltage deviations.

[0006] Existing optimization algorithms suffer from high computational complexity when dealing with large-scale distribution networks, making it difficult to meet real-time optimization requirements. Traditional methods often neglect the comprehensive optimization of power loss and voltage stability during cluster partitioning, leading to energy waste and grid operation risks.

[0007] Therefore, how to develop a cluster partitioning method that can adapt to the dynamic characteristics of distributed energy, take into account the coordination of multiple voltage levels, and meet the real-time optimization needs of large-scale distribution networks has become a key issue that urgently needs to be addressed in the field of distribution network optimization. Summary of the Invention

[0008] In view of the shortcomings of the prior art, the purpose of this invention is to provide a multi-level distribution network cluster partitioning method and device. By partitioning the multi-level distribution network cluster, combined with K-means and genetic algorithms, high and low voltage collaborative optimization is achieved, power loss and voltage deviation are reduced, the dynamic adaptability to distributed energy is improved, and the real-time optimization needs of large-scale distribution networks can be met.

[0009] To achieve the above objectives, this invention provides a method for dividing a multi-level distribution network cluster, comprising the following steps: S1. Based on the modularity index and power balance index, a high-voltage level cluster partitioning model is constructed, its constraints are set, and the cluster partitioning result is output using the K-means algorithm. S2. Based on the power exchange index between clusters and the dynamic fitness index of clusters, a low-voltage level cluster partitioning model is constructed, its constraints are set, and a dynamic sub-cluster partitioning optimization method is adopted. Based on the cluster partitioning results, the sub-cluster partitioning results of each cluster are output using a genetic algorithm. S3. Establish a multi-level optimization model and its constraints. Based on the multi-level optimization model, use the multi-level optimization algorithm to coordinate the objective function of the high and low voltage level cluster partitioning model to achieve global optimization.

[0010] As a preferred embodiment of the present invention, in S1, the objective function of the high voltage level cluster partitioning model is... for: (1); In the formula, The electrical distance between nodes i and j is the modularity index. The active power balance of the entire distribution network, i.e., the power balance index; , They are respectively , Weighting coefficients; The constraints of the high-voltage level cluster partitioning model include: Node exclusivity constraint: Each node can only be assigned to one cluster, represented as: (2); In the formula, As an indicator variable, This indicates that node i belongs to cluster c, otherwise it is 0; N is the total number of nodes in the distribution network. Indicates the number of clusters; Cluster size constraints mean that the number of nodes in each cluster must be within a reasonable range, expressed as: (3); In the formula, The number of nodes in cluster c; , These are the floor functions for rounding up and rounding down, respectively. Power capacity constraint: The total power capacity of each cluster must meet the operational requirements of the distribution network, expressed as: (4); In the formula, Provide photovoltaic power to node i; Let be the energy storage power of node i; , These are the minimum and maximum power generation capacities of cluster c, respectively. Electrical connectivity constraint: Nodes within the same cluster are electrically connected.

[0011] As a preferred embodiment of the present invention, the process of outputting the cluster partitioning result in S1 is as follows: S1.1 Initialize the input node feature vector and the number of clusters. For the distribution network topology data, the K-means++ method is used to select initial centers, maximizing the distance between initial centers. One node serves as the initial cluster center. Ensure that the initial center meets the electrical connectivity constraints; S1.2 For each node i, calculate its relationship with each cluster center. The Euclidean distance is used to assign node i to the cluster with the closest distance; After each node allocation, the cluster size constraint, power capacity constraint, and electrical connectivity constraint are checked. If any constraint is not met, the node allocation is adjusted, nodes exceeding the cluster size constraint are reassigned, and a penalty function is used to adjust the objective function. (5); In the formula, This represents the adjusted objective function, which is used as the new objective function. use; , , This is the penalty coefficient; This represents the net power of cluster c; S1.3 Update the cluster center. For each cluster c, recalculate the center: (6); In the formula, This represents all nodes in cluster c; This represents the voltage magnitude at node i; S1.4, Repeat S1.2 and S1.3 until the maximum number of iterations is reached; In each iteration, calculate the objective function value and verify that all constraints are satisfied; S1.5 Output cluster partitioning results Each cluster center Objective function value .

[0012] As a preferred embodiment of the present invention, in S2, the inter-cluster power exchange index Represented as: (7); In the formula, This refers to the active power exchange index; This refers to the reactive power exchange index. , They are respectively , Weighting coefficients; in, Represented as: (8); In the formula, SC represents the set of all sub-clusters of cluster c; Sub-cluster and The amount of active power exchanged between them; Represented as: (9); In the formula, Sub-cluster and The amount of reactive power exchanged between them; Cluster dynamic fitness index Represented as: (10); (11); (12); (13); In the formula, is an intermediate variable; T represents the evaluation period; This represents the net change in active power of sub-cluster sc at time t; This represents the net change in reactive power of sub-cluster sc at time t; , These represent the maximum active and reactive power capacities of the sub-cluster sc, respectively. , Let represent the active power of sub-cluster sc at times t and t-1, respectively; , Let represent the reactive power of sub-cluster sc at times t and t-1, respectively.

[0013] As a preferred embodiment of the present invention, in S2, the objective function of the low-voltage level cluster partitioning model is... for: (14); In the formula, , They represent , Weighting coefficients; The constraints of the low-voltage level cluster partitioning model include: Node exclusivity constraint: Each node can only be assigned to one sub-cluster, represented as: (15); In the formula, This represents the number of sub-clusters. As an indicator variable, This indicates that node i belongs to sub-cluster sc; otherwise, it is 0. This represents all nodes in cluster c; Sub-cluster size constraints: (16); In the formula, The number of nodes in the sub-cluster sc; , These represent the minimum and maximum number of nodes in the cluster, respectively.

[0014] As a preferred embodiment of the present invention, in S2, the prediction-based dynamic sub-cluster partitioning optimization method involves using a genetic algorithm to obtain the optimal solution for sub-cluster partitioning, while simultaneously monitoring the deviation between the actual power and the predicted value to trigger a re-optimization of the sub-cluster partitioning: real-time acquisition of the actual power of each node, and calculation of the deviation at time t. And set a threshold, when At that time, starting from the current optimal solution, the genetic algorithm is run to re-optimize the sub-cluster partitioning, and a cooldown time limit is set to avoid frequent adjustments; Represented as: (17); In the formula, This represents the actual net power of node i at time t; This represents the predicted net power of node i at time t; N is the total number of nodes in the distribution network.

[0015] As a preferred embodiment of the present invention, in S2, the process of using a genetic algorithm to output the sub-cluster partitioning results of each cluster is as follows: S2.1 Use K-means clustering to generate an initial partitioning scheme as high-quality individuals; S2.2 Initialize parameters and population, setting the population size n, the number of algorithm iterations K, and the crossover probability. and mutation probability The objective function f is used to evaluate the fitness of individuals, and the initial population and subpopulation are randomly generated. S2.3. Select chromosomes that meet the cluster size constraints. Select each individual in the current population and subpopulation to meet the power capacity constraints and electrical connectivity constraints. Remove unqualified individuals and regenerate individuals. S2.4. Assess the fitness of the current population and subpopulation, and calculate the fitness value of the current generation for each individual in the population and each individual in the subpopulation. S2.5. Use the roulette wheel selection algorithm to select parent individuals, perform crossover and mutation operations on the parent individuals selected in the subpopulation, and then use a disjoint-set data structure to verify electrical connectivity. S2.6 Merge individuals from the two generations of the secondary population, retaining the globally optimal individual. Meanwhile, the subpopulation duplication rate is controlled; if it is greater than 0.9, individuals are regenerated. S2.7 Dynamic Monitoring ,like ,from Start a new round of optimization, output the adjusted partitioning scheme, and output the optimal sub-cluster partitioning result when the maximum number of iterations K is reached or the fitness converges. .

[0016] As a preferred embodiment of the present invention, in S3, the objective function J of the multi-level optimization model is: (18); In the formula, This represents the objective function of the high-voltage level cluster partitioning model; Let represent the objective function of the low-voltage level cluster partitioning model within the c-th cluster; For hierarchical coordination weights; The constraints are: Power flow constraints are expressed as: (19); In the formula, This represents all nodes in cluster c; Provide photovoltaic power to node i; Let be the energy storage power of node i; Let be the load power of node i; This refers to the power loss within cluster c; Node voltage constraints: (20); In the formula, , , Let N represent the voltage amplitude of node i and its minimum and maximum values, respectively; N is the total number of nodes in the distribution network.

[0017] As a preferred embodiment of the present invention, the multi-level optimization algorithm in S3 is specifically as follows: S3.1 High voltage level optimization: K-means algorithm is used to obtain cluster partitioning results. Objective function value ; S3.2 Low-voltage level optimization: For cluster c, optimize its sub-clusters by using a genetic algorithm to generate the optimal sub-cluster partitioning result. and the objective function value ; S3.3, through dynamic adjustment Given power flow constraints, the optimization objectives of balancing high-voltage and low-voltage levels are optimized, and the global objective function J is optimized as follows: S3.3.1 Setting Initial Weights , obtain and Check whether each cluster c satisfies the cluster size constraint, power capacity constraint, and electrical connectivity constraint. If not, run the genetic algorithm to readjust the sub-cluster partitioning. S3.3.2, Calculate J pair gradient: (twenty one); In the formula, Indicates the number of clusters; Update weights: (twenty two); In the formula, The learning rate; , These are the current level coordination weight and the updated level coordination weight, respectively. Recalculate and ,renew and ; S3.3.3, Repeat S3.3.1 and S3.3.2 until the termination condition is met, i.e., the weight changes. If the change is less than the set threshold, the power flow constraint is verified in each iteration; S3.3.4, Output the final result and Objective function value J and optimized .

[0018] A multi-level distribution network cluster partitioning device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The processor executes the computer program to implement the method described in the claims.

[0019] The beneficial effects of this invention are: This invention divides the distribution network into large-scale clusters at high voltage levels and fine-grained sub-clusters at low voltage levels. Optimization at high voltage levels focuses on global power distribution, while optimization at low voltage levels focuses on local voltage stability. Through collaborative optimization, a balance between global and local considerations is achieved, effectively addressing the problem that traditional single-voltage-level optimization struggles to simultaneously address both global and local needs. Furthermore, a multi-objective optimization model is constructed, comprehensively considering indicators such as electrical distance, power balance, power loss, voltage stability, and node importance. An initial solution is generated using the K-means algorithm, combined with optimization using a genetic algorithm, reducing computational complexity and meeting the real-time optimization requirements of large-scale distribution networks.

[0020] This invention combines real-time monitoring data with a dynamic adjustment mechanism to adapt to the dynamic changes of distributed energy resources and enhance real-time optimization capabilities. Verified through actual distribution network simulation, compared with single non-hierarchical cluster partitioning and the traditional static K-means method, this invention has significant advantages in reducing power imbalance, voltage deviation, and power loss, while also greatly reducing computation time. It effectively improves the operating efficiency and voltage stability of the distribution network, enhances its adaptability to distributed energy resources, and provides strong technical support for the efficient management of modern smart grids. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the principle of the method of the present invention; Figure 2 This refers to the geographical layout of the power distribution network during the verification process in this invention. Figure 3 This is the result of the distribution network cluster division during the verification process in this invention. Detailed Implementation

[0022] The embodiments of the present invention will be further described below with reference to the accompanying drawings: Example 1: As Figure 1 As shown, a method for partitioning a multi-level distribution network cluster includes the following steps: S1. Based on the modularity index and power balance index, a high-voltage level cluster partitioning model is constructed, its constraints are set, and the cluster partitioning result is output using the K-means algorithm. S2. Based on the power exchange index between clusters and the dynamic fitness index of clusters, a low-voltage level cluster partitioning model is constructed, its constraints are set, and a dynamic sub-cluster partitioning optimization method is adopted. Based on the cluster partitioning results, the sub-cluster partitioning results of each cluster are output using a genetic algorithm. S3. Establish a multi-level optimization model and its constraints. Based on the multi-level optimization model, use the multi-level optimization algorithm to coordinate the objective function of the high and low voltage level cluster partitioning model to achieve global optimization.

[0023] In S1, the objective function of the high-voltage level cluster partitioning model is... for: (1); In the formula, The electrical distance between nodes i and j is the modularity index. The active power balance of the entire distribution network, i.e., the power balance index; , They are respectively , Weighting coefficients; The modularity index and power balance index can be obtained from existing technologies, such as the relevant indexes disclosed in CN202411404504.X.

[0024] The constraints of the high-voltage level cluster partitioning model include: Node exclusivity constraint: Each node can only be assigned to one cluster, represented as: (2); In the formula, As an indicator variable, This indicates that node i belongs to cluster c, otherwise it is 0; N is the total number of nodes in the distribution network. Indicates the number of clusters (i.e., high-voltage level clusters); Cluster size constraints mean that the number of nodes in each cluster must be within a reasonable range, avoiding being too small or too large, as expressed as: (3); In the formula, The number of nodes in cluster c; , These are the floor functions for rounding up and rounding down, respectively. Constraints ensure that each cluster contains at least one Number of nodes, at most no more than [number] nodes Each node.

[0025] Power capacity constraint: The total power capacity of each cluster must meet the operational requirements of the distribution network, expressed as: (4); In the formula, Provide photovoltaic power to node i; Let be the energy storage power of node i; , These are the minimum and maximum power generation capacities of cluster c, respectively. , , This represents the total power generation capacity of the entire power distribution network.

[0026] Electrical connectivity constraint: Nodes within the same cluster are electrically connected.

[0027] The process of outputting the cluster partitioning results is as follows: S1.1 Initialize the input node feature vector and the number of clusters. For distribution network topology data, the K-means++ method is used to select initial centers, maximizing the distance between initial centers and reducing the risk of convergence to local optima. One node serves as the initial cluster center. Ensure that the initial center meets the electrical connectivity constraints; S1.2 For each node i, calculate its relationship with each cluster center. The Euclidean distance is used to assign node i to the cluster with the closest distance; After each node allocation, check the cluster size constraint, power capacity constraint, and electrical connectivity constraint. If any constraint is not met (if any one of them is not met, the node allocation is reassigned), adjust the node allocation, reassign nodes that exceed the cluster size constraint, and adjust the objective function using a penalty function. (5); In the formula, This represents the adjusted objective function, which is used as the new objective function. Use (i.e., after adjustment) Updated to ); , , This is the penalty coefficient; This represents the net power of cluster c; S1.3 Update the cluster center. For each cluster c, recalculate the center: (6); In the formula, This represents all nodes in cluster c; This represents the voltage magnitude at node i; S1.4, Repeat S1.2 and S1.3 until the maximum number of iterations is reached; In each iteration, calculate the objective function value and verify that all constraints are satisfied; S1.5 Output cluster partitioning results Each cluster center Objective function value .

[0028] By introducing a minimum inter-cluster power exchange metric, cross-cluster power flow can be reduced, power transmission distances shortened, and network losses lowered. This also improves system stability and enhances cluster autonomy. In S2, the inter-cluster power exchange metric... Represented as: (7); In the formula, This refers to the active power exchange index; This refers to the reactive power exchange index. , They are respectively , Weighting coefficients; in, Represented as: (8); In the formula, SC represents the set of all sub-clusters of cluster c; Sub-cluster and The amount of active power exchanged between them; Represented as: (9); In the formula, Sub-cluster and The amount of reactive power exchanged between them; Cluster dynamic fitness index Represented as: (10); (11); (12); (13); In the formula, is an intermediate variable; T represents the evaluation period; This represents the net change in active power of sub-cluster sc at time t; This represents the net change in reactive power of sub-cluster sc at time t; , The maximum active and reactive power capacities of sub-clusters sc are used for normalization, and the index values ​​are... The value is between 0 and 1, with smaller values ​​indicating a stronger ability of the cluster to adapt to dynamic changes; , Let represent the active power of sub-cluster sc at times t and t-1, respectively; , Let represent the reactive power of sub-cluster sc at times t and t-1, respectively.

[0029] The objective function for low-voltage cluster partitioning is to minimize the voltage deviation and power imbalance of the sub-clusters. for: (14); In the formula, , They represent , Weighting coefficients; The constraints of the low-voltage level cluster partitioning model include: Node exclusivity constraint: Each node can only be assigned to one sub-cluster, represented as: (15); In the formula, This refers to the number of sub-clusters (i.e., low-voltage level sub-clusters). As an indicator variable, This indicates that node i belongs to sub-cluster sc; otherwise, it is 0. This represents all nodes in cluster c; Sub-cluster size constraints: (16); In the formula, The number of nodes in the sub-cluster sc; , These represent the minimum and maximum number of nodes in the cluster, respectively.

[0030] The prediction-based dynamic sub-cluster partitioning optimization method uses a genetic algorithm to obtain the optimal solution for sub-cluster partitioning. Simultaneously, it monitors the deviation between actual power and predicted values, triggering a re-optimization of the sub-cluster partitioning: the actual power of each node is collected in real time, and the deviation at time t is calculated. And set a threshold, when At that time, starting from the current optimal solution, the genetic algorithm is run to re-optimize the sub-cluster partitioning, and a cooldown time limit (1 hour) is set to avoid frequent adjustments; Represented as: (17); In the formula, This represents the actual net power of node i at time t; The predicted net power of node i at time t is represented (the predicted value can be obtained using known methods such as the LSTM model); N is the total number of nodes in the distribution network.

[0031] An improved genetic algorithm is used to solve for the optimal cluster partitioning of the distribution network. While performing the normal genetic iteration process, a secondary population is constructed. This secondary population explores new solution spaces through independent evolutionary processes, while the main population A utilizes these exploration results through fusion and elite retention mechanisms. Including prediction model training and data preparation, and considering dynamic cluster partitioning, the process of using the genetic algorithm to output the sub-cluster partitioning results for each cluster is as follows: S2.1 Use K-means clustering to generate an initial partitioning scheme as high-quality individuals; S2.2 Initialize parameters and population, setting the population size n, the number of algorithm iterations K, and the crossover probability. and mutation probability The objective function f is used to evaluate the fitness of individuals, and the initial population and subpopulation are randomly generated. S2.3. Select chromosomes that meet the cluster size constraints. Select each individual in the current population and subpopulation to meet the power capacity constraints and electrical connectivity constraints. Remove unqualified individuals and regenerate individuals. S2.4. Assess the fitness of the current population and subpopulation, and calculate the fitness value of the current generation for each individual in the population and each individual in the subpopulation. S2.5. Use the roulette wheel selection algorithm to select parent individuals, perform crossover and mutation operations on the parent individuals selected in the subpopulation, and then use a disjoint-set data structure to verify electrical connectivity. S2.6 Merge individuals from the two generations of the secondary population, retaining the globally optimal individual. Meanwhile, the subpopulation duplication rate is controlled; if it is greater than 0.9, individuals are regenerated. S2.7 Dynamic Monitoring ,like ,from Start a new round of optimization, output the adjusted partitioning scheme, and output the optimal sub-cluster partitioning result when the maximum number of iterations K is reached or the fitness converges. .

[0032] The multi-level optimization algorithm aims to coordinate the optimization objectives of high-voltage levels (large clusters) and low-voltage levels (sub-clusters). Through independent optimization at each level and cross-level coordination, it achieves global and local power allocation and voltage stability optimization. The algorithm combines K-means, genetic algorithms, and gradient descent to ensure computational efficiency and optimization accuracy.

[0033] The multi-level optimization algorithm consists of two stages: independent optimization at each level and coordinated optimization across levels. The former optimizes the objective functions of the high-voltage and low-voltage levels separately, while the latter achieves global coordination through weight adjustment and iterative optimization.

[0034] In S3, the objective function J of the multi-level optimization model is: (18); In the formula, This represents the objective function of the high-voltage level cluster partitioning model; The objective function of the low-voltage level cluster partitioning model within the c-th cluster (i.e., the function corresponding to cluster c) is... ); For hierarchical coordination weights; The constraints are: Power flow constraints are expressed as: (19); In the formula, This represents all nodes in cluster c; Provide photovoltaic power to node i; Let be the energy storage power of node i; Let be the load power of node i; This refers to the power loss within cluster c; Node voltage constraints: (20); In the formula, , , Let N represent the voltage amplitude of node i and its minimum and maximum values, respectively; N is the total number of nodes in the distribution network.

[0035] The multi-level optimization algorithm is as follows: S3.1 High voltage level optimization: K-means algorithm is used to obtain cluster partitioning results. Objective function value ; S3.2 Low-voltage level optimization: For cluster c, optimize its sub-clusters by using a genetic algorithm to generate the optimal sub-cluster partitioning result. and the objective function value ; S3.3, through dynamic adjustment Given power flow constraints, the optimization objectives of balancing high-voltage and low-voltage levels are optimized, and the global objective function J is optimized as follows: S3.3.1 Setting Initial Weights , obtain and Check whether each cluster c satisfies the cluster size constraint, power capacity constraint, and electrical connectivity constraint. If not, run the genetic algorithm to readjust the sub-cluster partitioning. S3.3.2, Calculate J pair gradient: (twenty one); In the formula, Indicates the number of clusters; Update weights: (twenty two); In the formula, The learning rate; , These are the current level coordination weight and the updated level coordination weight, respectively. Recalculate and ,renew and ; S3.3.3, Repeat S3.3.1 and S3.3.2 until the termination condition is met, i.e., the weight changes. If the change is less than the set threshold, the power flow constraint is verified in each iteration; S3.3.4, Output the final result and Objective function value J and optimized .

[0036] The verification process is as follows: To illustrate the superiority of the method proposed in this patent, a simulation verification was conducted using a real-world power distribution network as an example. The geographical layout of the power distribution network is as follows: Figure 2 As shown.

[0037] The cluster partitioning result obtained according to the method proposed in this embodiment is as follows: Figure 3 As shown, Figure 3 In the diagram, the yellow areas represent clusters, and the blue areas represent sub-clusters.

[0038] Meanwhile, the method proposed in this embodiment is compared with a single non-hierarchical cluster partitioning method and the traditional static K-means method, as shown in Table 1: Table 1 Comparison Results

[0039] As shown in Table 1, the unbalanced power of the method in this embodiment is reduced by 25% compared to a single-level method and by 44% compared to traditional methods, achieving better global power balance. Voltage deviation is reduced by 25% compared to a single-level method and by 45% compared to traditional methods, demonstrating more stable multi-level optimization. Power loss is reduced by 19% compared to a single-level method and by 35% compared to traditional methods, effectively reducing energy waste and significantly saving computation time.

[0040] Based on the simulation results above, the multi-level distribution network cluster partitioning proposed in this embodiment demonstrates significant advantages. Experiments show that this method, through coordinated optimization of high and low voltage levels, significantly reduces power imbalance, voltage deviation, and power loss, exhibiting clear advantages compared to single-level optimization and the traditional static K-means method. Furthermore, by combining prediction and dynamic adjustment mechanisms, it maintains high adaptability in dynamic scenarios with drastic fluctuations in photovoltaic output, achieving a prediction error (RMSE) of less than 0.1MW and reducing power imbalance and voltage deviation by 18%-30%, respectively.

[0041] Furthermore, through parallel computing optimization, the algorithm's computation time is reduced by 50%-75%, while still meeting real-time optimization requirements in large-scale distribution networks. These results fully validate the innovation and practicality of the method in this embodiment in improving distribution network operating efficiency, voltage stability, and distributed energy adaptability, providing reliable technical support for the efficient management of modern smart grids.

[0042] Example 2: Predicted net power of node i at time t Besides using LSTM model prediction, the following methods can also be used to obtain the result: Generated through a reinforcement learning model, which includes a state space, an action space, and a reward function, wherein: The state space is defined as the power feature vector of node i within the historical time window, including the actual net power. (node ​​i at time) Actual net power), photovoltaic output Energy storage capacity and load power ,in Indicates the time lag parameter; The action space is defined as the predicted power adjustment strategy, including the power change rate and the prediction bias compensation factor; Reward function at time t The design based on the reduction rate of prediction error is expressed as: (twenty three); In the formula, , These are the weighting coefficients; (ReoptimizationCost) represents the computational cost penalty triggered by reoptimization; The reinforcement learning model updates its policy online using the Q-learning algorithm to maximize cumulative reward, and... Time output optimization It is used for deviation calculation.

[0043] Reinforcement learning can handle uncertainties and dynamic changes in distribution networks (such as load fluctuations or new energy output fluctuations), and is more robust than static prediction methods, improving the stability and reliability of cluster partitioning. It can reduce computational overhead by minimizing unnecessary re-optimization, while ensuring that sub-cluster partitioning responds more promptly to actual deviations, thus optimizing distribution network operating efficiency.

[0044] Example 3: A multi-level distribution network cluster partitioning device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The processor executes the computer program to implement the method in Example 1 or Example 2.

Claims

1. A method for dividing a multi-level distribution network cluster, characterized in that... Includes the following steps: S1. Based on the modularity index and power balance index, a high-voltage level cluster partitioning model is constructed, its constraints are set, and the cluster partitioning result is output using the K-means algorithm. S2. Based on the power exchange index between clusters and the dynamic fitness index of clusters, a low-voltage level cluster partitioning model is constructed, its constraints are set, and a dynamic sub-cluster partitioning optimization method is adopted. Based on the cluster partitioning results, the sub-cluster partitioning results of each cluster are output using a genetic algorithm. S3. Establish a multi-level optimization model and its constraints. Based on the multi-level optimization model, use the multi-level optimization algorithm to coordinate the objective function of the high and low voltage level cluster partitioning model to achieve global optimization.

2. The method for dividing a multi-level distribution network cluster according to claim 1, characterized in that, In S1, the objective function of the high-voltage level cluster partitioning model is... for: （1）； In the formula, The electrical distance between nodes i and j is the modularity index. The active power balance of the entire distribution network, i.e., the power balance index; , They are respectively , Weighting coefficients; The constraints of the high-voltage level cluster partitioning model include: Node exclusivity constraint: Each node can only be assigned to one cluster, represented as: （2）； In the formula, As an indicator variable, This indicates that node i belongs to cluster c, otherwise it is 0; N is the total number of nodes in the distribution network. Indicates the number of clusters; Cluster size constraints mean that the number of nodes in each cluster must be within a reasonable range, expressed as: （3）； In the formula, The number of nodes in cluster c; , These are the floor functions for rounding up and rounding down, respectively. Power capacity constraint: The total power capacity of each cluster must meet the operational requirements of the distribution network, expressed as: （4）； In the formula, Provide photovoltaic power to node i; Let be the energy storage power of node i; , These are the minimum and maximum power generation capacities of cluster c, respectively. Electrical connectivity constraint: Nodes within the same cluster are electrically connected.

3. The method for dividing a multi-level distribution network cluster according to claim 2, characterized in that, In S1, the process of outputting the cluster partitioning results is as follows: S1.1 Initialize the input node feature vector and the number of clusters. For the distribution network topology data, the K-means++ method is used to select initial centers, maximizing the distance between initial centers. One node serves as the initial cluster center. Ensure that the initial center meets the electrical connectivity constraints; S1.2 For each node i, calculate its relationship with each cluster center. The Euclidean distance is used to assign node i to the cluster with the closest distance; After each node allocation, the cluster size constraint, power capacity constraint, and electrical connectivity constraint are checked. If any constraint is not met, the node allocation is adjusted, nodes exceeding the cluster size constraint are reassigned, and a penalty function is used to adjust the objective function. （5）； In the formula, This represents the adjusted objective function, which is used as the new objective function. use; , , This is the penalty coefficient; This represents the net power of cluster c; S1.3 Update the cluster center. For each cluster c, recalculate the center: （6）； In the formula, This represents all nodes in cluster c; This represents the voltage magnitude at node i; S1.4, Repeat S1.2 and S1.3 until the maximum number of iterations is reached; In each iteration, calculate the objective function value and verify that all constraints are satisfied; S1.5 Output cluster partitioning results Each cluster center Objective function value .

4. The method for dividing a multi-level distribution network cluster according to claim 1, characterized in that, In S2, the inter-cluster power exchange index Represented as: （7）； In the formula, This refers to the active power exchange index; This refers to the reactive power exchange index. , They are respectively , Weighting coefficients; in, Represented as: （8）； In the formula, SC represents the set of all sub-clusters of cluster c; Sub-cluster and The amount of active power exchanged between them; Represented as: （9）； In the formula, Sub-cluster and The amount of reactive power exchanged between them; Cluster dynamic fitness index Represented as: （10）； （11）； （12）； （13）； In the formula, is an intermediate variable; T represents the evaluation period; This represents the net change in active power of sub-cluster sc at time t; This represents the net change in reactive power of sub-cluster sc at time t; , These represent the maximum active and reactive power capacities of the sub-cluster sc, respectively. , Let represent the active power of sub-cluster sc at times t and t-1, respectively; , Let represent the reactive power of sub-cluster sc at times t and t-1, respectively.

5. The method for dividing a multi-level distribution network cluster according to claim 4, characterized in that, In S2, the objective function of the low-voltage level cluster partitioning model is... for: （14）； In the formula, , They represent , Weighting coefficients; The constraints of the low-voltage level cluster partitioning model include: Node exclusivity constraint: Each node can only be assigned to one sub-cluster, represented as: （15）； In the formula, This represents the number of sub-clusters. As an indicator variable, This indicates that node i belongs to sub-cluster sc; otherwise, it is 0. This represents all nodes in cluster c; Sub-cluster size constraints: （16）； In the formula, The number of nodes in the sub-cluster sc; , These represent the minimum and maximum number of nodes in the cluster, respectively.

6. The method for dividing a multi-level distribution network cluster according to claim 5, characterized in that, In S2, the prediction-based dynamic sub-cluster partitioning optimization method uses a genetic algorithm to obtain the optimal solution for sub-cluster partitioning. Simultaneously, it monitors the deviation between actual power and predicted values, triggering a re-optimization of the sub-cluster partitioning: the actual power of each node is collected in real time, and the deviation at time t is calculated. And set a threshold, when At that time, starting from the current optimal solution, the genetic algorithm is run to re-optimize the sub-cluster partitioning, and a cooldown time limit is set to avoid frequent adjustments; Represented as: （17）； In the formula, This represents the actual net power of node i at time t; This represents the predicted net power of node i at time t; N is the total number of nodes in the distribution network.

7. The method for dividing a multi-level distribution network cluster according to claim 6, characterized in that, In S2, the process of using a genetic algorithm to output the sub-cluster partitioning results of each cluster is as follows: S2.1 Use K-means clustering to generate an initial partitioning scheme as high-quality individuals; S2.2 Initialize parameters and population, setting the population size n, the number of algorithm iterations K, and the crossover probability. and mutation probability The objective function f is used to evaluate the fitness of individuals, and the initial population and subpopulation are randomly generated. S2.

3. Select chromosomes that meet the cluster size constraints. Select each individual in the current population and subpopulation to meet the power capacity constraints and electrical connectivity constraints. Remove unqualified individuals and regenerate individuals. S2.

4. Assess the fitness of the current population and subpopulation, and calculate the fitness value of the current generation for each individual in the population and each individual in the subpopulation. S2.

5. Use the roulette wheel selection algorithm to select parent individuals, perform crossover and mutation operations on the parent individuals selected in the subpopulation, and then use a disjoint-set data structure to verify electrical connectivity. S2.6 Merge individuals from the two generations of the secondary population, retaining the globally optimal individual. Meanwhile, the subpopulation duplication rate is controlled; if it is greater than 0.9, individuals are regenerated. S2.7 Dynamic Monitoring ,like ,from Start a new round of optimization, output the adjusted partitioning scheme, and output the optimal sub-cluster partitioning result when the maximum number of iterations K is reached or the fitness converges. .

8. The method for dividing a multi-level distribution network cluster according to claim 1, characterized in that, In S3, the objective function J of the multi-level optimization model is: （18）； In the formula, This represents the objective function of the high-voltage level cluster partitioning model; Let represent the objective function of the low-voltage level cluster partitioning model within the c-th cluster; For hierarchical coordination weights; The constraints are: Power flow constraints are expressed as: （19）； In the formula, This represents all nodes in cluster c; Provide photovoltaic power to node i; Let be the energy storage power of node i; Let be the load power of node i; This refers to the power loss within cluster c; Node voltage constraints: （20）； In the formula, , , Let N represent the voltage amplitude of node i and its minimum and maximum values, respectively; N is the total number of nodes in the distribution network.

9. A method for dividing a multi-level distribution network cluster according to claim 8, characterized in that, In S3, the multi-level optimization algorithm is specifically as follows: S3.1 High voltage level optimization: K-means algorithm is used to obtain cluster partitioning results. Objective function value ; S3.2 Low-voltage level optimization: For cluster c, optimize its sub-clusters by using a genetic algorithm to generate the optimal sub-cluster partitioning result. and the objective function value ; S3.3, through dynamic adjustment Given power flow constraints, the optimization objectives of balancing high-voltage and low-voltage levels are optimized, and the global objective function J is optimized as follows: S3.3.1 Setting Initial Weights , obtain and Check whether each cluster c satisfies the cluster size constraint, power capacity constraint, and electrical connectivity constraint. If not, run the genetic algorithm to readjust the sub-cluster partitioning. S3.3.2, Calculate J pair gradient: （21）； In the formula, Indicates the number of clusters; Update weights: （22）； In the formula, The learning rate; , These are the current level coordination weight and the updated level coordination weight, respectively. Recalculate and ,renew and ; S3.3.3, Repeat S3.3.1 and S3.3.2 until the termination condition is met, i.e., the weight changes. If the change is less than the set threshold, the power flow constraint is verified in each iteration; S3.3.4, Output the final result and Objective function value J and optimized .

10. A multi-level distribution network cluster partitioning device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the method described in any one of claims 1-9.

Images