Power distribution network cluster division method considering distributed resource regulation potential

Abstract

The invention discloses a power distribution network cluster division method considering distributed resource regulation and control potential, is suitable for a high-permeability distributed photovoltaic, large-scale distributed power supply and energy storage equipment access complex power distribution network, and belongs to the technical field of power system partition optimization and distributed control. The method comprises the following steps: acquiring the sensitivity of node voltage to reactive power injection change through a power flow model, and extracting node reactive power regulation capability and a space density factor; normalizing the extracted electrical distance, reactive power and spatial density data, and splicing the data into node feature vectors; determining the cluster number K, and iteratively optimizing the clustering center and node attribution by taking the minimum Euclidean distance quadratic sum between the cluster internal node feature vector and the cluster center as a target; and outputting the cluster label of each node to complete power distribution network partitioning. The method provided by the invention has the advantages of simple steps and low calculation amount, can quickly identify node groups which are strong in cluster internal coupling and relatively independent among clusters, provides high-quality partition support for subsequent regionalization voltage and reactive power optimization, and remarkably improves distributed control efficiency and system operation stability.

Description

Technical Field

[0001] The present invention relates to a distribution network operation and control technology, and in particular to a distribution network cluster division method considering the potential of distributed resource regulation. Background Art

[0002] With the global energy transition and the goal of carbon neutrality, distributed photovoltaic (PV), wind power, user-side distributed power sources, and energy storage devices are being rapidly deployed in distribution networks. While large-scale renewable energy integration has improved the cleanliness of distribution networks, it has also significantly exacerbated voltage fluctuations, uneven reactive power distribution, and reverse power flow. On the one hand, the randomness and volatility of PV output can cause voltage limits at end nodes to exceed their limits. On the other hand, centralized reactive power compensation and voltage control methods, due to high communication traffic, slow response times, and difficulty in accurately locating problem areas, can no longer meet the current distribution network's demand for rapid dynamic regulation.

[0003] To achieve more efficient localized control, academia and engineering have proposed regionalized voltage / reactive power coordinated control based on network partitioning. Node clustering is a key prerequisite technology: by dividing distribution network nodes into clusters with strong internal coupling and weak external coupling, conditions are created for independent or hierarchical control of each cluster. Currently, common clustering methods are primarily based on electrical distance, network topology, or a single sensitivity metric. These methods fail to adequately consider inter-node electrical coupling, local resource distribution, and spatial clustering characteristics, resulting in poor reactive power compensation or voltage regulation after partitioning.

[0004] Therefore, a method is urgently needed that can simultaneously integrate multi-dimensional indicators such as electrical distance (reflecting the strength of voltage coupling), node reactive power regulation capability (reflecting the potential of local resource support), and distributed resource spatial density (reflecting the degree of spatial aggregation of resources such as photovoltaics and DG), and complete node partitioning through an efficient clustering algorithm to significantly improve the quality of partitioning and subsequent regional control performance. Summary of the Invention

[0005] To address the existing issues of partitioning, which often rely on single indicators and produce poor results, the present invention provides a distribution network cluster partitioning method that considers the potential for distributed resource control. Partitioning calculations can be completed in a small number of steps, ensuring strong intra-cluster coupling while reducing cross-cluster interference, providing a solid partitioning foundation for distributed control. This method is characterized by the following steps:

[0006] Step S1: Based on the distribution network power flow model, solve the sensitivity matrix of voltage to changes in active and reactive power injection, and construct the electrical distance E between nodes based on it. ij .

[0007]

[0008] ΔU=(N-HJ-1 L) -1 ΔP+(L-JH -1 N) -1 ΔQ (2)

[0009]

[0010] Where H, N, M, and L are the elements of the Jacobian matrix, ΔP represents the change in node active power, and ΔQ represents the change in node reactive power. By simplifying the power flow equation, we obtain the reactive voltage sensitivity matrix K ik And the active voltage sensitivity matrix L ik , these two matrices respectively characterize the response characteristics of node voltage to reactive power injection and active power injection.

[0011] Step S2: Extract the reactive power regulation capability F of each node i and spatial density factors of distributed photovoltaics and distributed generation

[0012] Step S3: For indicator E ij 、F i , ρ i Perform linear normalization on each node and combine the normalized values to form the node feature vector X i .

[0013] Step S4: Set the number of clusters K, and use the K-means algorithm to iteratively update the cluster center μ based on the principle of minimizing the objective function. k and node set C k , until f(C) converges.

[0014]

[0015] Step S5: Output the cluster label of each node to realize regional partitioning of the distribution network.

[0016] (1) In step 1, accurately define the reactive power regulation capability F i And electrical distance E i,j , which is crucial for the node partitioning of distribution networks:

[0017]

[0018] Where Q i,max , Q i,min The adjustable maximum and minimum reactive power are used to quantify the node's responsiveness in reactive power optimization. The adjustable reactive power range reflects the node's potential to cope with reactive power fluctuations. By calculating its value relative to the node's actual reactive power Q iThe square ratio can clearly reflect the node regulation capability. For example, large industrial nodes have a large reactive power regulation range, F i A high value can serve as an important adjustment node during partitioning.

[0019] Electrical distance E i,j Calculated by formula (7), where K ik , L ik The voltage sensitivity matrix elements are the voltage sensitivity matrix elements and the active sensitivity matrix elements when node i injects reactive power into node k. This technical feature takes into account the actual electrical coupling relationship between nodes. The voltage sensitivity matrix elements reflect the impact of active and reactive power injection between nodes on the voltage. ik After normalization, we can get the sum of E i,j The smaller the value, the closer the electrical connection between the nodes is, and the more suitable it is to be divided into the same area during partitioning.

[0020] (2) Spatial density factor The kernel density estimation (KDE) method is used to describe the spatial aggregation degree of distributed resources:

[0021]

[0022] Where, d ij is the electrical distance between nodes i and j, h is the bandwidth, K is the kernel function, and is applicable to all distribution network nodes i∈N bus The electrical distance reflects the physical location relationship of the nodes, the bandwidth h controls the influence range of the kernel function, and the kernel function K weights the nodes at different distances. The spatial density factor ρ is obtained by summing i The larger the value, the denser the distributed resources around the node. When partitioning, resource-intensive areas can be divided into the same cluster, which is convenient for centralized management and coordinated control.

[0023] (3) The execution of the K-means clustering algorithm includes:

[0024] (a) K-means is used to initialize K cluster centers in the feature vector space. Traditional K-means random initialization is prone to falling into local optimality. K-means first randomly selects a point as the first cluster center, and then selects new centers based on the principle of minimizing the distance between the point and the selected cluster center. This ensures that the initial cluster centers are evenly distributed in the feature vector space, improving the algorithm's convergence speed and clustering accuracy.

[0025] (b) Assign each node to the closest cluster based on Euclidean distance. Calculate the Euclidean distance between each node and all cluster centers. Based on the principle that closely spaced nodes have similar features and should belong to the same region, assign the nodes to the corresponding cluster.

[0026] (c) Update the cluster center. Recalculate the cluster center based on the average of the feature vectors of the member nodes within the cluster. Add the corresponding elements of the feature vectors of all nodes within the cluster and divide by the number of nodes. This makes the cluster center better represent the overall characteristics of the cluster and provides a reasonable reference for subsequent node allocation.

[0027] (d) Repeat the node assignment and cluster center update steps until the node cluster labels are stable or the reduction in the sum of squared distances (SSE) within the cluster falls below a preset threshold. During this iteration, nodes are reallocated based on the new cluster centers, and the cluster centers are updated. When the SSE reduction falls below a preset threshold, the clustering is stable and iteration can be stopped. The preset threshold needs to be adjusted based on the scenario and data.

[0028] (e) Output the final cluster center and node cluster labels to complete the distribution network regionalization. This result provides a reference for distribution network operation management, allowing the development of reactive power optimization control strategies for different regions to improve operational efficiency and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:

[0030] Figure 1 Flowchart of the distribution network node resource cluster division method

[0031] Figure 2 Comprehensive workflow diagram for the proposed cluster partitioning strategy. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0033] See also Figure 1 ,A distribution network cluster partitioning method considering the ,potential of distributed resource regulation, includes the following steps,

[0034] Step 1: Extraction of electrical distance and resource characteristics. In this step, the power flow model of the distribution network is first established through power flow calculation, ensuring that the complete network topology, line parameters, node loads, and distributed resource data are input. After solving using the Newton-Raphson method, the sensitivity of voltage to active power injection and the sensitivity of voltage to reactive power injection L can be extracted from the Jacobian matrix. ik After generating the sensitivity matrix, the electrical coupling strength K between any two nodes is quantified using the following formula: ik :

[0035]

[0036] ΔU=(N-HJ -1 L) -1 ΔP+(L-JH -1 N) -1 ΔQ (2)

[0037]

[0038] Among them, H, N, M, and L are the elements of the Jacobian matrix, ΔP represents the change in node active power, and ΔQ represents the change in node reactive power. By simplifying the power flow equation, we get the reactive voltage sensitivity matrix K ik And the active voltage sensitivity matrix L ik , these two matrices respectively characterize the response characteristics of node voltage to reactive power injection and active power injection.

[0039]

[0040] This electrical distance comprehensively reflects the difference in the impact of the node on the voltage distribution of the entire network.

[0041] While obtaining the electrical distance, the system will read the parameters of the reactive compensation device equipped at each node, including the maximum reactive output Q i,max , minimum reactive output Q i,min And the reference reactive load Q i , and calculate the reactive power regulation capability of the node by the following formula:

[0042]

[0043] This indicator can accurately reflect the potential contribution of the node in reactive power optimization.

[0044] In addition, to reflect the clustering effect of distributed photovoltaic (PV) and distributed generation (DG) devices in the electrical space, the kernel density estimation (KDE) method is used to calculate the spatial density factor based on the node coordinates and resource location data as follows:

[0045]

[0046] Among them, d ij is the electrical distance between node i and node j, h is the smoothing bandwidth, and K is the selected kernel function. Through the above processing, the “three-dimensional” characteristics of each node are completed: electrical coupling (E i,j ), reactive power response (F i ) and resource aggregation (ρ i ) comprehensive extraction.

[0047] Step 2: Normalization and eigenvector construction. In order to eliminate the dimensional differences of different indicators, step S2 normalizes the E obtained in step S1. i,j (the average distance from all other nodes can be taken), F i and ρ i The minimum-maximum linear normalization is performed separately and mapped to the interval [0,1]. After normalization, the three are concatenated in sequence to form the characteristic vector of each node. The electrical coupling degree, reactive power regulation capability and resource spatial distribution of the comprehensive node are used to provide structured and dimensional data input for subsequent clustering.

[0048] Step 3: K-means clustering iteration

[0049] In this step, the desired number of clusters K is determined based on engineering experience or the “elbow method”; then, based on the K-means method and the above-mentioned feature vector set, K cluster centers {θ K The clustering iteration process includes: for each feature vector X j , calculate its relationship with all centers θ K , and assign it to the cluster with the nearest center; then recalculate the center of each cluster and take the arithmetic average of the eigenvectors of all members of the cluster:

[0050]

[0051] Repeat the assignment and update until the cluster labels are stable or the reduction in the sum of squares within the cluster is lower than the set threshold.

[0052]

[0053] The objective function and update formula used in this process are consistent with those in the claims, ensuring the global rationality of cluster division and local optimal convergence.

[0054] Step 4: Output the partitioning results. After K-means converges, the final cluster label for each node is written to the distribution network management system or GIS platform, and a visual map of the node distribution of different clusters is displayed (for example, using different colors), allowing operations personnel to quickly identify the boundaries of each region. Subsequently, each region can independently perform voltage regulation and reactive power optimization control based on the clustering results, reducing cross-region coordination overhead while improving control response speed and overall system stability.

Claims

1. A distribution network cluster division method considering the potential of distributed resource regulation, characterized in that: The steps include: Step S1: Based on the distribution network power flow model, solve the sensitivity matrix of voltage to changes in active and reactive power injection, and construct the electrical distance E between nodes based on it. ij ; ΔU=(N-HJ -1 L) -1 ΔP+(L-JH -1 N) -1 ΔQ (2) Where H, N, M, and L are the elements of the Jacobian matrix, ΔP represents the change in node active power, and ΔQ represents the change in node reactive power. By simplifying the power flow equation, the reactive voltage sensitivity matrix K is obtained: ik And the active voltage sensitivity matrix L ik , these two matrices represent the response characteristics of node voltage to reactive power injection and active power injection respectively; Step S2: Extract the reactive power regulation capability F of each node i and spatial density factors of distributed photovoltaics and distributed generation Step S3: For indicator E ij 、F i , ρ i Perform linear normalization on each node and combine the normalized values to form the node feature vector X i ; Step S4: Set the number of clusters K, and use the K-means algorithm to iteratively update the cluster center μ based on the principle of minimizing the objective function. k and node set C k , until f(C) converges; Step S5: Output the cluster label of each node to realize regional partitioning of the distribution network.

2. The method for dividing distribution network clusters considering the potential of distributed resource regulation according to claim 1, wherein in step 1, the reactive regulation capability F i And electrical distance E ij It is defined by the following formula: Where Q i,max , Q i,min is the adjustable maximum / minimum reactive power, which is used to quantify the response capability of the node in reactive power optimization. ik , L ik The reactive voltage sensitivity and active sensitivity matrix elements are injected for node i to node k respectively to ensure that the actual electrical coupling relationship of the nodes is considered when partitioning.

3. The method for dividing distribution network clusters considering the potential for distributed resource regulation according to claim 1, wherein in step 2, the spatial density factor Calculated by kernel density estimation KDE method: Where, d ij is the distance between nodes i and j, h is the bandwidth, and K is the kernel function; it reflects the degree of spatial aggregation of distributed resources.

4. The method for clustering a distribution network considering the potential for distributed resource regulation according to claim 1, wherein in step 4, executing the K-means clustering algorithm comprises: (a) Using the K-means++ strategy to initialize K cluster centers in the feature vector space; (b) Assign each node to the cluster with the smallest distance based on the Euclidean distance to the center of each cluster; (c) Recalculate the cluster center of each cluster based on the average value of the feature vectors of its member nodes; (d) Repeat steps (b) and (c) until the node cluster label is stable or the reduction in the sum of squared distances within the cluster is lower than a preset threshold; (e) Output the final cluster center and node affiliation labels to complete the regional division of the distribution network.

Images