Flexible interconnection device site selection method and system based on distributed resource clustering

Abstract

The invention provides a flexible interconnection device site selection method and system based on distributed resource clustering, and the method comprises the steps: constructing a photovoltaic output feature map Laplacian matrix and a load feature map Laplacian matrix according to the photovoltaic output and load data of a planning region, decomposing the constructed matrixes through DGRNMF, and obtaining a distributed resource cluster; solving a photovoltaic clustering center curve and a load clustering center curve according to a decomposition result through a k-means clustering algorithm, and dividing the photovoltaic clustering center curve and the load clustering center curve into an urban region and a rural region; constructing constraint conditions of the power supply grid segmentation model to obtain the power supply grid segmentation model; and solving the power supply grid segmentation model to obtain an optimal power grid partitioning scheme, and determining a candidate installation position of the flexible interconnection device according to the optimal power grid partitioning scheme. According to the invention, the precision and reliability of the flexible interconnection device site selection planning decision are improved.

Description

Technical Field

[0001] This invention belongs to the field of power system technology, specifically relating to the field of distribution network planning under the background of flexible interconnection, and more specifically, relating to a flexible interconnection network planning method and system that takes into account the clustering characteristics of distributed resources. Background Technology

[0002] In recent years, with the large-scale integration of distributed generators (DGs) and the increasing uncertainty of loads, the operation mode and planning methods of distribution networks have faced severe challenges.

[0003] Soft Open Points (SOPs), as flexible interconnection devices for efficient power flow regulation, provide crucial technical support for improving the flexibility and economy of distribution networks. How to achieve effective load balancing and voltage fluctuation control through the rational planning and deployment of SOPs has become an important research topic in the field of distribution networks.

[0004] With the advancement of new power system construction, the distribution network currently faces several challenges. For example, some 10kV urban distribution networks do not meet the "N-1" requirement, resulting in high load rates or even overloads on certain lines. Uneven load distribution in some urban areas leads to increased line losses, potentially causing equipment damage or line faults. Furthermore, the continuous integration of flexible resources places new demands on the distribution network for high reliability and operational flexibility. Most 10kV rural distribution networks are single-radial lines, characterized by low interconnectivity, poor load transfer capacity, and excessively long power supply radii. Moreover, with the development of new loads, line load rates continue to increase, making it difficult to meet future load growth needs.

[0005] With the increasing penetration of distributed photovoltaic, energy storage, and flexible loads, the complexity of distribution network operation is increasing: resources exhibit significant clustering characteristics, with urban resources having low and fluctuating output, while rural resources have high output but poor smoothness. This difference renders traditional "one-size-fits-all" planning methods ineffective, requiring targeted regional processing; the demand for operational flexibility is urgent, and the intermittency and uncertainty of distributed resources pose new requirements for high power supply reliability. Existing distribution networks rely on mechanical interconnect switches, which have slow control speeds, cannot achieve millisecond-level power balance, and are ill-suited to handling multi-scenario coupling problems. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for selecting flexible interconnection devices based on distributed resource clustering. In this invention, distributed resources refer to the sum of all distributed photovoltaic (PV) power generation units and loads within a planning area. First, the output of PV power generation and load changes over different time periods are collected within the planning area. The PV output and load characteristics of each area are used as clustering features. The PV output scenario and load clustering center curves for that area are clustered using the DGRNMF (Dual-Graph Regularized Nonnegative Matrix Factorization) clustering algorithm. Next, a power grid segmentation model based on source-load partitioning constraints is established, dividing the PV clustering center curves and load clustering center curves into two categories: urban and rural. The candidate locations for SOPs (Site-Operated Utility Units) are determined based on the results of solving the grid segmentation model. Then, a new flexible distribution network planning model is established, incorporating power balance into the constraints to ensure the system meets power balance conditions. Finally, the specific scheme for adding flexible interconnection devices to the distribution network is obtained by solving the flexible planning model. This invention achieves precise zoning and scientific planning of flexible interconnection devices in the power distribution network through a four-in-one technical approach of data-driven cluster analysis, urban-rural characteristic discrimination, constrained modeling, and intelligent optimization solution. It solves the core problem of large differences in distributed resource characteristics and the inapplicability of traditional planning methods, improves the safety, reliability, and economy of power grid operation, and reduces network losses, equipment overload risks, and the probability of power outages.

[0007] The present invention adopts the following technical solution.

[0008] A first aspect of the present invention provides a flexible interconnection device location method based on distributed resource clustering, comprising: Based on the photovoltaic output and load data of the planning area, the photovoltaic output feature map Laplace matrix and the load feature map Laplace matrix are constructed. The constructed matrices are decomposed by bi-graph regularized non-negative matrix decomposition clustering. The decomposition results are then used to determine the photovoltaic cluster center curve and the load cluster center curve by k-means clustering algorithm. The photovoltaic cluster center curve and the load cluster center curve are divided into urban and rural areas; The objective function of the power grid segmentation model is to minimize active power loss and load balance. The constraints of the power grid segmentation model are constructed based on the photovoltaic clustering center curves and load clustering center curves of urban and rural areas, thus obtaining the power grid segmentation model. Solve the power grid segmentation model to obtain the optimal power grid partitioning scheme, and determine the candidate installation locations of the flexible interconnection device based on the optimal power grid partitioning scheme.

[0009] Preferably, photovoltaic output data and load data for each time period in the planning area are read and the data is preprocessed; Based on the preprocessed data, construct the Laplace matrix of the photovoltaic power output sample graph and the Laplace matrix of the load sample graph; Construct the photovoltaic output feature map Laplace matrix and the load feature map Laplace matrix based on the photovoltaic output sample map Laplace matrix and the load sample map Laplace matrix; Based on the Laplace matrix of the photovoltaic power output characteristic map and the Laplace matrix of the load characteristic map, the DGRNMF algorithm is used to decompose them into photovoltaic coefficient matrix and load coefficient matrix, photovoltaic base matrix and load base matrix; Based on the photovoltaic coefficient matrix and the load coefficient matrix, the optimal number of clusters for photovoltaic output and load is determined by the elbow method. Based on the photovoltaic base matrix, the load base matrix, the optimal number of clusters for photovoltaic output and load, multiple photovoltaic cluster center curves and load cluster center curves are obtained by the k-means clustering algorithm.

[0010] Preferably, the photovoltaic coefficient matrix and load factor matrix, photovoltaic base matrix and load base matrix are decomposed using the DGRNMF algorithm, including: The DGRNMF objective function of photovoltaics is constructed based on the Laplace matrix of the photovoltaic power output sample map and the Laplace matrix of the photovoltaic power output feature map; the photovoltaic diagonal matrix is ​​constructed; based on the photovoltaic diagonal matrix, the photovoltaic basis matrix, the photovoltaic coefficient matrix and the photovoltaic sparse error matrix are updated by alternating multiplication; the iteration continues until the DGRNMF objective function of photovoltaics converges. The DGRNMF objective function of the load is constructed based on the Laplacian matrix of the load sample map and the Laplacian matrix of the load feature map; the load diagonal matrix is ​​constructed; based on the load diagonal matrix, the load basis matrix, the load coefficient matrix and the load sparsity error matrix are updated by alternating multiplication; the iteration continues until the DGRNMF objective function of the load converges.

[0011] Preferably, the objective function of DGRNMF for photovoltaics is expressed by the following formula:

[0012] In the formula, Represents a photovoltaic matrix. Represents the photovoltaic coefficient matrix. This indicates photovoltaic power output data. express Norm, Represents the photovoltaic sparse error matrix; The weights that control time smoothness are represented by tr, and tr represents the trace of the matrix. The weights represent the control space similarity. Represents the photovoltaic power output characteristic matrix. Represents the Laplace matrix of the photovoltaic sample graph. The superscript indicates the weights of the photovoltaic sparse error matrix. This indicates the matrix transpose.

[0013] Preferably, the photovoltaic base matrix, photovoltaic coefficient matrix, and photovoltaic sparse error matrix are updated as follows:

[0014] In the formula, This represents element-wise multiplication. Denotes the first photovoltaic diagonal matrix. This represents the second photovoltaic diagonal matrix. Represents the photovoltaic power output sample degree matrix. This represents the similarity matrix of photovoltaic power output samples. express The main part, express The absolute value of the negative part.

[0015] Preferably, the photovoltaic clustering center curve and the load clustering center curve are divided into urban and rural areas, including: The photovoltaic peak value characteristics, photovoltaic smoothness characteristics, and photovoltaic output stability characteristics are determined based on the photovoltaic cluster center curves. The scores of all photovoltaic cluster center curves are constructed based on the photovoltaic peak value characteristics, photovoltaic smoothness characteristics, and photovoltaic output stability characteristics. Among the scores of all photovoltaic cluster center curves, if the score of a certain type of photovoltaic cluster center curve is greater than the set photovoltaic discrimination threshold, that type of photovoltaic cluster center curve is judged as urban; otherwise, it is judged as rural. The load peak-to-valley difference characteristics, load rate characteristics, and load fluctuation characteristics are determined based on the load cluster center curves. The scores of all load cluster center curves are constructed based on the load peak-to-valley difference characteristics, load rate characteristics, and load fluctuation characteristics. Among the scores of all load cluster center curves, if the score of a certain type of load cluster center curve is greater than the set photovoltaic discrimination threshold, the photovoltaic cluster center curve of that type is judged as urban; otherwise, it is judged as rural.

[0016] Preferably, the constraints include topology constraints, system security constraints, switching operation constraints, power flow constraints, distributed generation output constraints, and source-load partitioning constraints. The source-load zoning constraints are constructed based on the photovoltaic clustering center curves and load clustering center curves, which divide the region into urban and rural areas, and are expressed by the following formula:

[0017] In the formula, Let be the Euclidean distance between curves a and b; Let p be the photovoltaic power output curve for the p-th city area. Let be the photovoltaic power output curve for the q-th rural region. The photovoltaic cluster center curve for urban areas, The photovoltaic cluster center curve for rural areas, This indicates the maximum permissible Euclidean distance within the photovoltaic category in urban areas. This indicates the maximum permissible Euclidean distance within the photovoltaic category in rural areas. and These are categorized as urban and rural photovoltaic systems, respectively. Let m be the load electricity consumption curve for the m-th city area. This is the load electricity consumption curve for the nth rural area. The load cluster center curve for the urban area. The load cluster center curve for rural areas, This represents the maximum permissible Euclidean distance within the urban area load class. This indicates the maximum permissible Euclidean distance within the rural photovoltaic load category. and The categories are urban and rural areas, respectively, for load.

[0018] Preferably, solving the power supply grid segmentation model includes: Generate a random adjacency matrix as the initial population, calculate the fitness of each adjacency matrix under the objective function, and determine the individual with the highest fitness in the current population; The adjacency matrix is ​​cross-crossed and mutated. The fitness of each individual in the new population after cross-crossing and mutation is recalculated and compared with the fitness of the individual with the highest fitness in the current population. The individual with the highest fitness is updated. The current iteration count is checked to see if it meets the maximum iteration count. If not, the next iteration continues. If it does, the loop is exited, and the optimized final population, the optimal adjacency matrix and its fitness are obtained. Set the optimal adjacency matrix as the optimal power grid partitioning scheme.

[0019] Preferably, determining the candidate installation locations for the flexible interconnection device based on the optimal power grid zoning scheme includes: The optimal power grid partitioning scheme is found by depth-first search, which identifies all sets of nodes connected by closed branches. Each set of connected nodes is a power supply area. Identify the set of boundary branches of the power supply area, and select the top 100 locations with the highest comprehensive priority scores from the set of boundary branches of the power supply area to form a list of candidate SOP locations.

[0020] A second aspect of the present invention provides a flexible interconnection device location system based on distributed resource clustering, which, when running the flexible interconnection device location method based on distributed resource clustering described in the first aspect of the present invention, includes: The cluster center curve solving module is used to construct the photovoltaic output feature map Laplace matrix and the load feature map Laplace matrix based on the photovoltaic output and load data of the planning area. The constructed matrices are decomposed by dual-graph regularized non-negative matrix decomposition clustering. The decomposition results are then used by the k-means clustering algorithm to determine the photovoltaic cluster center curve and the load cluster center curve. The classification module is used to divide the photovoltaic cluster center curve and the load cluster center curve into urban and rural areas; The model building module is used to construct the constraints of the power grid segmentation model based on the photovoltaic clustering center curves and load clustering center curves of urban and rural areas, with the goal of minimizing active power loss and load balance. The location determination module is used to solve the power grid segmentation model, obtain the optimal power grid partitioning scheme, and determine the candidate installation locations of the flexible interconnection device based on the optimal power grid partitioning scheme.

[0021] Compared with the prior art, the beneficial effects of the present invention include at least the following: By using spatiotemporal characteristic clustering analysis based on DGRNMF, accurate pattern recognition and feature extraction of massive distributed photovoltaic and load curves were achieved. This solved the problem that traditional methods could not effectively characterize complex operating characteristics such as "low and large fluctuations in urban resource output and high and smooth rural resource output". It improved the depth of resource characteristic cognition, reduced the blindness of site selection planning decisions for flexible interconnection devices, and improved the accuracy of site selection planning decisions for flexible interconnection devices. By automatically identifying urban and rural labels with multi-feature weighting, the system achieves quantitative classification of urban and rural clustering patterns, solving the subjectivity and inconsistency problems of existing technologies that rely on human experience to classify resource types. This improves the scientific basis of zoning, reduces human bias in planning results, and enhances the accuracy of site selection planning decisions for flexible interconnection devices. By embedding source-load partitioning constraints into the power grid segmentation model, the resource clustering characteristics are directly transformed into mathematical constraints for power grid physical segmentation. This solves the problem of disconnect between power grid topology optimization and resource characteristics in traditional planning, improves the matching degree between the partitioning scheme and resource distribution, reduces the risk of line overload and voltage exceeding limits caused by "source-load mismatch", and improves the reliability and safety of the site selection planning decision for flexible interconnection devices. Solving the partitioning model and decoding the candidate SOP locations enables efficient searching of the optimal power grid partitioning scheme in a large solution space and automatic identification of key flexible interconnection points. This solves the problems of slow mechanical switch control and inability to cope with random fluctuations in distributed resources, improves the distribution network's adaptability to uncertainty and the flexibility of power flow control, reduces line losses, and enhances the load transfer capability and power supply reliability of the flexible interconnection device's site selection and planning decisions. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the flexible interconnection device location method based on distributed resource clustering provided in accordance with an embodiment of the present invention; Figure 2 This is a schematic diagram of the sum of squared distances to photovoltaic cluster centers provided in accordance with an embodiment of the present invention; Figure 3 This is a schematic diagram of the photovoltaic cluster center curve based on the DGRNMF method clustering provided in accordance with an embodiment of the present invention; Figure 4 This is a schematic diagram of the sum of squared distances to the cluster centers of the load provided in accordance with an embodiment of the present invention; Figure 5 This is a schematic diagram of the load cluster center curve based on DGRNMF clustering provided in accordance with an embodiment of the present invention; Figure 6 This is a schematic diagram of the population algorithm flow provided according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the power supply network segmentation in the planning area according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the alternative locations of the planning area SOP provided in accordance with the embodiments of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0024] like Figure 1 As shown, Embodiment 1 of the present invention provides a site selection and planning method for flexible interconnected devices based on distributed resource clustering, including the following steps: Step 1: Construct the Laplace matrix of the photovoltaic output feature map and the Laplace matrix of the load feature map based on the photovoltaic output and load data of the planning area. The constructed matrices are decomposed using DGRNMF, and the decomposition results are used to solve for the photovoltaic cluster center curve and the load cluster center curve using the k-means clustering algorithm.

[0025] In a preferred but non-limiting embodiment of the present invention, step 1 includes: Step 1.1: Read the photovoltaic output data and load data of the planning area for each time period, and preprocess the data, including filling in missing values ​​in the data by interpolation or mean substitution, removing negative values ​​and pruning outliers to ensure that the data is non-negative.

[0026] Step 1.2: Construct the Laplace matrix of the photovoltaic power output sample map and the Laplace matrix of the load sample map based on the preprocessed data in Step 1.1.

[0027] More preferably, step 1.2 includes: Step 1.2.1: Calculate the Euclidean distance between every two photovoltaic power output data points; calculate the Euclidean distance between every two load data points. The smaller the Euclidean distance, the more similar the power output patterns of the two data points throughout the day.

[0028] Step 1.2.2: For each photovoltaic power output data, find the k photovoltaic power output data that are most similar to it from all photovoltaic power output data and construct a k nearest neighbor graph; for each load data, find the k load data that are most similar to it from all load data and construct a k nearest neighbor graph.

[0029] Step 1.2.3: For each pair of photovoltaic power output data within the k nearest neighbor range, calculate the connection weight between each pair of photovoltaic power output data using a Gaussian kernel function, and symmetricize the weight matrix to obtain the photovoltaic power output sample similarity matrix; For each pair of load data within the k nearest neighbor range, calculate the connection weight between each pair of load data using a Gaussian kernel function, and symmetricize the weight matrix to obtain the load sample similarity matrix.

[0030] Weights reflect the local similarity between data points; the higher the weight, the higher the similarity. Data pairs not within the k-nearest neighbor range have a weight of 0.

[0031] Step 1.2.4: Determine the photovoltaic output sample degree matrix by summing the rows of the photovoltaic output sample similarity matrix; obtain the photovoltaic output sample graph Laplace matrix by subtracting the photovoltaic output sample degree matrix from the photovoltaic output sample similarity matrix; calculate the load sample degree matrix based on the load sample similarity matrix; obtain the load sample graph Laplace matrix by subtracting the load sample degree matrix from the load sample similarity matrix.

[0032] Step 1.3: Construct the photovoltaic output feature map Laplace matrix and the load feature map Laplace matrix based on the photovoltaic output sample map Laplace matrix and the load sample map Laplace matrix.

[0033] More preferably, step 1.3 includes: Step 1.3.1: Considering that photovoltaic power output and load have a natural temporal continuity, the power output / load values ​​at adjacent times are usually gradual rather than abrupt. The most common method is the chain adjacency method, which connects the data at adjacent times in the photovoltaic power output data with weights, and does not connect the data at non-adjacent times, to obtain a photovoltaic power output tridiagonal matrix; and connects the data at adjacent times in the load data with weights, and does not connect the data at non-adjacent times, to obtain a load tridiagonal matrix.

[0034] Step 1.3.2: Symmetrically process the tridiagonal matrix of photovoltaic output to obtain the photovoltaic output feature similarity matrix. Calculate the photovoltaic output feature degree matrix based on the photovoltaic output feature similarity matrix, and calculate the Laplace matrix of the photovoltaic output feature map based on the photovoltaic output feature degree matrix. Symmetrically process the tridiagonal matrix of load to obtain the load feature similarity matrix. Calculate the load feature degree matrix based on the load feature similarity matrix, and calculate the Laplace matrix of the load feature map based on the load feature degree matrix.

[0035] Step 1.4: The photovoltaic output characteristic map Laplace matrix and the load characteristic map Laplace matrix are determined by the DGRNMF algorithm to determine the photovoltaic coefficient matrix, the load coefficient matrix, the photovoltaic base matrix, and the load base matrix, decomposing the original load data into physically meaningful typical load patterns and corresponding pattern contributions.

[0036] More preferably, step 1.4 includes: Step 1.4.1: Construct the DGRNMF objective function for photovoltaic power generation based on the Laplace matrix of the photovoltaic power output sample map and the Laplace matrix of the photovoltaic power output feature map, expressed by the following formula:

[0037] In the formula, This represents a photovoltaic matrix, where each column represents a sunrise power generation mode, such as, but not limited to, sunny day mode and cloudy day mode. This represents the photovoltaic coefficient matrix, where each column describes the time-varying weights of the photovoltaic data on the photovoltaic base matrix. This represents the photovoltaic power output data, which is the 24-hour power output data of n photovoltaic power stations. express Norm, This represents the photovoltaic sparse error matrix, which captures outliers in photovoltaic data. The weights represent the control over time smoothness, used to ensure that the sunrise power curve conforms to the natural variation of solar irradiance; tr represents the trace of the matrix. Represents the Laplace matrix of the photovoltaic feature map; The weights represent the control spatial similarity, ensuring that photovoltaic sites located close to each other have similar power output patterns. Represents the photovoltaic power output characteristic matrix. Represents the Laplace matrix of the photovoltaic sample graph. The weights represent the sparse error matrix of photovoltaics.

[0038] Construct a photovoltaic diagonal matrix, expressed by the following formula:

[0039] In the formula, Denotes the first photovoltaic diagonal matrix. Represents the second photovoltaic diagonal matrix; Based on the photovoltaic diagonal matrix, the photovoltaic basis matrix, photovoltaic coefficient matrix, and photovoltaic sparse error matrix are updated by alternating multiplication, as expressed by the following formula:

[0040] In the formula, This represents element-wise multiplication. Represents the photovoltaic power output sample degree matrix. This represents the similarity matrix of photovoltaic power output samples. Represents the positive part of H. It represents the absolute value of the negative part of H.

[0041] Step 1.4.2: Construct the DGRNMF objective function for the load based on the Laplace matrix of the load sample map and the Laplace matrix of the load feature map, expressed by the following formula:

[0042] In the formula, This represents the load base matrix, where each column represents a daily load pattern, such as, but not limited to, residential electricity consumption patterns and industrial electricity consumption patterns. This represents the load factor matrix, where each column represents the weight of the electricity consumption characteristics of the load data on the load basis matrix. This represents load data, specifically the 24-hour electricity consumption data for m load points. This represents the load sparse error matrix; Weights representing the temporal continuity of controlled electricity consumption behavior, maintaining a smooth change in the load cluster center curve. Represents the Laplace matrix of the load sample graph. , Represents the load sample degree matrix, Represents the similarity matrix of the load samples; The weights representing the similarity of user types are represented by similar patterns using load points with similar electrical characteristics. The Laplace matrix represents the load characteristic map. , Represents the load characteristic matrix, Represents the load feature similarity matrix. The superscript represents the weighting parameters of the sparse error matrix in the control load data. This indicates the matrix transpose.

[0043] Construct the load diagonal matrix, expressed by the following formula:

[0044] In the formula, Represents the first load diagonal matrix. This represents the second load diagonal matrix; The load basis matrix, load factor matrix, and load sparsity error matrix are updated by alternating multiplication, as expressed by the following formula:

[0045] Iterate separately until the DGRNMF objective function of photovoltaic and the DGRNMF objective function of load converge.

[0046] Step 1.5: Based on the photovoltaic coefficient matrix and load coefficient matrix obtained in Step 1.4, determine the optimal number of photovoltaic power output and load clusters using the elbow method; based on the optimal number of photovoltaic power output and load clusters, as well as the photovoltaic basis matrix and load basis matrix obtained in Step 1.4, determine the photovoltaic cluster center curve and the load cluster center curve using the k-means clustering algorithm.

[0047] More preferably, step 1.5 includes: Step 1.5.1: Based on the photovoltaic coefficient matrix and load coefficient matrix obtained in Step 1.4, determine the optimal number of clusters for photovoltaic output and load using the elbow method.

[0048] More preferably, step 1.5.1 includes: For the photovoltaic power output data, a range of values ​​for K is defined; for each K value, a k-means clustering algorithm is run on the photovoltaic coefficient matrix. The column vectors represent the low-dimensional representation of a sample in the latent space. The sum of squares (SSE) within each cluster corresponding to each K value is recorded; this is the sum of squared distances from all samples to the center of their respective clusters. A curve is plotted showing the SSE as a function of K. The K value corresponding to the inflection point of the SSE curve is selected as the optimal number of clusters (Kpv) for photovoltaic (PV) output, i.e., the optimal number of PV output modes. The K value at the inflection point indicates that further increasing the number of clusters no longer significantly improves the model fit. PV output data is taken from 24-hour data from multiple PV sites, and load data is taken from 24-hour data from multiple load points. Each column vector of the PV output and load data represents a 24-hour curve for the corresponding PV site or load point; that is, each sample of the PV output and load data represents a 24-hour curve for the corresponding PV site or load point.

[0049] For the load data, define the range of values ​​for K; for each K value, run the k-means clustering algorithm on the load coefficient matrix. The column vectors represent a low-dimensional representation of a sample in the latent space; the sum of squares within a cluster (SSE) corresponding to each K value is recorded, which is the sum of squared distances from all samples to the center of their respective clusters; the curve of SSE as a function of K value is plotted; the K value corresponding to the inflection point of the SSE curve is selected as the optimal number of clusters Kload, i.e., the optimal number of load patterns.

[0050] Step 1.5.2: Use the photovoltaic base matrix and load base matrix obtained in step 1.4 as the initial cluster center curves for the k-means clustering algorithm.

[0051] Step 1.5.3: The target number of clusters for this k-means clustering algorithm is the optimal number of clusters Kpv for photovoltaic power output and the optimal number of clusters Kload for load. Calculate the Euclidean distance from each sample in the photovoltaic coefficient matrix and the load coefficient matrix to the initial cluster center curves of the current optimal number of clusters Kpv for photovoltaic power output and Kload for load. Assign the corresponding sample to the cluster closest to it, resulting in Kpv photovoltaic power output clusters and Kload load clusters. Step 1.5.4: Calculate the mean of all samples within each of the Kpv photovoltaic output clusters and Kload load clusters, respectively. Use these mean values ​​as the new Kpv and Kload cluster center curves for the next iteration. Repeat steps 1.5.3 and 1.4. The algorithm stops when the position changes of the Kpv and Kload cluster center curves reach the preset maximum number of iterations in two consecutive iterations, and the final K cluster center curves are obtained, thus yielding the photovoltaic cluster center curves. Cluster center curve of load .

[0052] Assuming there are 9 photovoltaic power stations in areas A and B of the planning region, cluster analysis is performed using 24-hour output data from each power station for one day each in spring, summer, autumn, and winter. The elbow method is used to determine the number of clusters, k. As shown in the Squashed Squared Distance (SSE)(k) curve of the cluster centers, the error decreases most significantly when k=2, therefore this is the optimal number of clusters. Figure 2 As shown.

[0053] The DGRNMF clustering algorithm is used to cluster the photovoltaic (PV) output scenarios within the planning area, providing a clear picture of the overall PV output in the distribution network of the planned area. The clustering results show that the PV cluster center curves are divided into two categories: urban PV cluster center curves and rural PV cluster center curves. The center curves for these two categories are shown below. Figure 3 As shown, the green curve has a higher peak value and a smoother curve with smaller fluctuations, indicating a more open space and less shading, thus representing the cluster center curve for rural photovoltaic systems with better ambient light conditions. In contrast, the red curve has a lower output peak value, representing the cluster center curve for urban photovoltaic systems with dense high-rise buildings and poor air quality.

[0054] Based on 19 loads within the assumed planning area, 24-hour data for each load were selected to obtain 19 load cluster center curves, which were then subjected to cluster analysis. The elbow method was used to determine the cluster categories, and the optimal number of clusters k was found to be 2. Figure 4 As shown. Therefore, the load cluster center curve is divided into urban load cluster center curve and rural load cluster center curve, as shown. Figure 5 As shown.

[0055] Depend on Figure 1-5 As can be seen, the loads in areas A and B are divided into two categories, representing urban load and rural load, respectively. The graph shows that the red curve has a higher peak value and a larger peak-to-valley difference, exhibiting significant fluctuations, characteristic of urban load; while the green curve is relatively flat, with a lower peak value and smaller fluctuations, exhibiting significant characteristics of rural load.

[0056] Based on the collection and cluster analysis of power sources and loads, regional resource endowment is determined as an important basis for grid partitioning and distribution network planning. The results can divide the entire system into two regions: first, rural areas with abundant power sources and weak loads, which can be divided into independent grids for self-sufficiency; second, urban areas with scarce power sources and strong loads, which can be coupled with adjacent areas to reduce dependence on the main grid.

[0057] Step 2: Divide the clustered photovoltaic cluster center curve and load cluster center curve into urban and rural areas.

[0058] Step 2.1, based on the photovoltaic cluster center curve Solving for photovoltaic peak characteristics Photovoltaic smoothness characteristics and photovoltaic power output stability characteristics It can be expressed by the following formula:

[0059] In the formula, Indicates the first Photovoltaic cluster center curve , Indicates the first The standardized output value of the photovoltaic cluster center curve at time t. Indicates the first Peak power generation capacity of photovoltaic cluster center curves Indicates the first Nighttime power output levels of photovoltaic cluster center curves This represents the total number of time points; in this invention, the value is 24, representing 24 hours in a day. This indicates the proportion of the change from an adjacent time t-1 to t to the total range of change throughout the day; express The standard deviation of is expressed by the following formula:

[0060] In the formula, express The average value is expressed by the following formula:

[0061] Based on photovoltaic peak characteristics Photovoltaic smoothness characteristics and photovoltaic power output stability Construct the scores for all photovoltaic cluster center curves. The score of the photovoltaic cluster center curve is expressed by the following formula:

[0062] In the formula, Indicates the first The score of the photovoltaic cluster center curve, Indicates the weighting coefficient; Among all the scores of photovoltaic cluster center curves, if the score of a certain type of photovoltaic cluster center curve is greater than the set photovoltaic discrimination threshold, that type of photovoltaic cluster center curve is judged as an urban area; otherwise, it is judged as a rural area. If the score of the photovoltaic cluster center curve is greater than the set photovoltaic discrimination threshold, the photovoltaic cluster center curve is determined to be an urban area; otherwise, it is determined to be a rural area.

[0063] Step 2.2, based on the load cluster center curve Solving the characteristics of load peak-valley difference Load factor characteristics and load fluctuation characteristics It can be expressed by the following formula:

[0064] In the formula, This represents the cluster center curve for the k2th load. , This represents the standardized output value of the k2th load cluster center curve at time t. This represents the peak generating capacity of the cluster center curve for the k2th load class. This represents the nighttime power output level of the center curve for the k2 type of photovoltaic power generation. This represents the total number of time points; in this invention, the value is 24, representing 24 hours in a day. express The standard deviation of is expressed by the following formula:

[0065] In the formula, express The average value is expressed by the following formula:

[0066] Based on the characteristics of load peak-valley difference Load factor characteristics Characteristics of load fluctuation Construct the scores of all load cluster center curves. The score of the k2th load cluster center curve is expressed by the following formula:

[0067] In the formula, This represents the score of the center curve of the k-th type of photovoltaic system. Indicates the weighting coefficient; Among the scores of all load cluster center curves, if the score of a certain load cluster center curve is greater than the set photovoltaic discrimination threshold, the photovoltaic cluster center curve of that type is determined to be an urban area; otherwise, it is determined to be a rural area. That is, if the score of the k2th load cluster center curve is greater than the set load discrimination threshold, the load cluster center curve of that type is determined to be an urban area; otherwise, it is determined to be a rural area.

[0068] Step 3: With the goal of minimizing active power loss and load balance, construct the objective function of the power grid segmentation model. Based on the photovoltaic cluster center curves and load cluster center curves that divide the city and rural areas, construct the constraint conditions of the power grid segmentation model to obtain the power grid segmentation model.

[0069] With the goals of minimizing active power loss and load balancing, the objective function of the power grid segmentation model is constructed and expressed by the following formula:

[0070] In the formula, and Indicates the coefficient weight. , The total active power loss is expressed by the following formula:

[0071] In the formula, Indicates the total number of time periods. It represents the set of all lines in the power grid. This represents the current in branch ij at time t. This represents the resistance of branch ij. The load balancing metric is expressed by the following formula:

[0072] In the formula, This represents the transmission complex power of branch ij at time t. This indicates the maximum transmission capacity of line ij. This represents the active power of branch ij at time t. This represents the reactive power of branch ij at time t.

[0073] The constraints include topology constraints, system security constraints, switching operation constraints, power flow constraints, distributed power output constraints, and source-load partitioning constraints.

[0074] Topological constraints are expressed by the following formula:

[0075] In the formula, This represents the set of all remote control switch branches. This represents the remote control switch state of branch ij at time t, where 0 is open and 1 is closed. Branch ij represents the line from node i to node j. This represents the total number of normally closed branches. This represents the total number of nodes in the network. Indicates the number of root nodes. Indicates the first The set of remote control switch branches in each loop. This represents the remote control switch state of branch ij in the l-th power supply loop at time t, where 0 is open and 1 is closed. This represents the number of normally closed branches in the l-th loop. This represents the total number of branches in the l-th power supply loop.

[0076] System safety constraints include: voltage constraints, current constraints, and tie-line capacity constraints, expressed by the following formula:

[0077] In the formula, This represents the minimum voltage at node i. Let represent the voltage at node i at time t. This represents the maximum voltage at node i. Let represent the current in branch ij at time t. This represents the maximum current in branch ij. This indicates the transmission power of the tie line ij. This indicates the maximum transmission power of the tie line ij.

[0078] The switching operation constraints are expressed by the following formula:

[0079] In the formula, This indicates the number of times the remote control switch of branch ij operates during time period t. This indicates the maximum number of times the branch switch ij is allowed to operate throughout the day.

[0080] Power flow constraints are expressed by the following formula:

[0081] In the formula, This indicates that the distributed power source at node i has active power output at time t. This represents the active load of node i at time t. Indicates the conductance of branch ij. Indicates the susceptance of branch ij. This represents the voltage phase angle difference between nodes i and j. This indicates that the reactive power output of the distributed power source at node i at time t is zero. This represents the reactive load of node i at time t.

[0082] The output constraint of distributed power sources is expressed by the following formula:

[0083] In the formula, This represents the maximum active power output of the distributed power source. The power factor angle represents the power factor of a distributed power source. This indicates the maximum reactive power output of the distributed power source.

[0084] Source-load zoning constraints are constructed based on the photovoltaic clustering center curves and load clustering center curves that divide urban and rural areas, as expressed by the following formula:

[0085] In the formula, Let be the Euclidean distance between curves a and b; Let p be the photovoltaic power output curve for the p-th city area. Let be the photovoltaic power output curve for the q-th rural region. The photovoltaic cluster center curve for urban areas, The photovoltaic cluster center curve for rural areas, This indicates the maximum permissible Euclidean distance within the photovoltaic category in urban areas. This indicates the maximum permissible Euclidean distance within the photovoltaic category in rural areas. and These are categorized as urban and rural photovoltaic systems, respectively. Let m be the load electricity consumption curve for the m-th city area. This is the load electricity consumption curve for the nth rural area. The load cluster center curve for the urban area. The load cluster center curve for rural areas, This represents the maximum permissible Euclidean distance within the urban area load class. This indicates the maximum permissible Euclidean distance within the rural photovoltaic load category. and The categories are urban and rural areas, respectively, for load.

[0086] Step 4: Use GA-PSO (Hybrid Genetic Algorithm-Particle Swarm Optimization) to solve the power grid partitioning model and select the optimal partitioning scheme, specifically: Step 4.1: Generate a random adjacency matrix as the initial population, calculate the fitness of each adjacency matrix under the objective function, and determine the individual with the highest fitness in the current population and its fitness.

[0087] A binary adjacency matrix A is used for encoding. The matrix dimension is NxN, where N is the number of nodes. Element A[i][j]=1 indicates that the branch is closed and nodes i and j are connected. A[i][j]=0 indicates that the branch is open. Randomly generate P feasible adjacency matrices Population={A} that satisfy the constraints in step 3. 1， A2, ..., Ap}, forming the initial population; For each individual in the population, i.e., the adjacency matrix A kPerform power flow calculations to obtain network status, including voltage, current, and power; calculate the adjacency matrix A based on the objective function of the power grid segmentation model in step 3. k Performance: f(A) k The problem is transformed into a minimization problem: Fitness(A) = w1 × f1 + w2 × f2; This converts the minimization problem into a fitness maximization problem: Fitness(A) = w1 × f1 + w2 × f2; k )=1 / (1+f(A k Record the individual A with the highest fitness in the current population. best And its adaptability.

[0088] Step 4.2: Perform crossover and mutation on the adjacency matrix, recalculate the fitness of each individual in the new population after crossover and mutation, compare and update the individual with the highest fitness, and check if the current iteration count meets the maximum iteration count. If not, continue to the next iteration; if so, exit the loop and obtain the optimized final population and the optimal adjacency matrix A. best,global The fitness of the population and the optimized final population are the partitioning schemes, and the optimal adjacency matrix is ​​the optimal power grid partitioning scheme.

[0089] Step 4.3, for the optimal adjacency matrix A best,global A depth-first search (DFS) is used to find all sets of nodes connected by closed branches (elements with a value of 1). Each connected set of nodes is an independent power supply area.

[0090] Step 4.4: Generate a power supply area partitioning vector, identifying the power supply area number to which each node belongs; identify the set of boundary branches of the power supply area, i.e., the branches connecting different power supply areas within A. best,global Disconnected branches; select the top N positions with the highest overall priority scores from the set of branches at the region boundary to form a list L of candidate SOP positions. candidate .

[0091] The overall priority score is expressed by the following formula:

[0092] In the formula, Indicates the power transfer strength weight. The power transfer strength score is expressed by the following formula:

[0093] In the formula, Indicates voltage stability weight. The voltage stability score is expressed by the following formula:

[0094] In the formula, Let the voltage amplitudes at node i and node j be at time t, respectively. Indicates the weight of the region coupling strength. The regional coupling strength score is expressed by the following formula:

[0095] In the formula, This represents the total active power load of the r-th region at time t.

[0096] The corresponding algorithm steps are as follows, and the algorithm flowchart is shown below. Figure 6 As shown, the power supply network segmentation diagram of the planning area is obtained as follows: Figure 7 As shown, based on the four regions, where regions 1 and 2 are rural areas and regions 3 and 4 are urban areas, several potential locations for the SOP can be selected: 9-13, 10-41, 2-66, 30-54, 36-62, and 45-60. Figure 8 As shown.

[0097] Embodiment 2 of the present invention provides a flexible interconnection device location system based on distributed resource clustering, which runs the flexible interconnection device location method based on distributed resource clustering described in Embodiment 1, including: The cluster center curve solving module is used to construct the photovoltaic output feature map Laplace matrix and the load feature map Laplace matrix based on the photovoltaic output and load data of the planning area. The constructed matrices are decomposed by dual-graph regularized non-negative matrix decomposition clustering. The decomposition results are then used by the k-means clustering algorithm to determine the photovoltaic cluster center curve and the load cluster center curve. The classification module is used to divide the photovoltaic cluster center curve and the load cluster center curve into urban and rural areas; The model building module is used to construct the constraints of the power grid segmentation model based on the photovoltaic clustering center curves and load clustering center curves of urban and rural areas, with the goal of minimizing active power loss and load balance. The location determination module is used to solve the power grid segmentation model, obtain the optimal power grid partitioning scheme, and determine the candidate installation locations of the flexible interconnection device based on the optimal power grid partitioning scheme.

[0098] Compared with the prior art, the beneficial effects of the present invention include at least the following: By using spatiotemporal characteristic clustering analysis based on DGRNMF, accurate pattern recognition and feature extraction of massive distributed photovoltaic and load curves were achieved. This solved the problem that traditional methods could not effectively characterize complex operating characteristics such as "low and large fluctuations in urban resource output and high and smooth rural resource output". It improved the depth of resource characteristic cognition, reduced the blindness of site selection planning decisions for flexible interconnection devices, and improved the accuracy of site selection planning decisions for flexible interconnection devices. By automatically identifying urban and rural labels with multi-feature weighting, the system achieves quantitative classification of urban and rural clustering patterns, solving the subjectivity and inconsistency problems of existing technologies that rely on human experience to classify resource types. This improves the scientific basis of zoning, reduces human bias in planning results, and enhances the accuracy of site selection planning decisions for flexible interconnection devices. By embedding source-load partitioning constraints into the power grid segmentation model, the resource clustering characteristics are directly transformed into mathematical constraints for power grid physical segmentation. This solves the problem of disconnect between power grid topology optimization and resource characteristics in traditional planning, improves the matching degree between the partitioning scheme and resource distribution, reduces the risk of line overload and voltage exceeding limits caused by "source-load mismatch", and improves the reliability and safety of the site selection planning decision for flexible interconnection devices. Solving the partitioning model and decoding the candidate SOP locations enables efficient searching of the optimal power grid partitioning scheme in a large solution space and automatic identification of key flexible interconnection points. This solves the problems of slow mechanical switch control and inability to cope with random fluctuations in distributed resources, improves the distribution network's adaptability to uncertainty and the flexibility of power flow control, reduces line losses, and enhances the load transfer capability and power supply reliability of the flexible interconnection device's site selection and planning decisions.

[0099] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A site selection method for flexible interconnected devices based on distributed resource clustering, characterized in that: Based on the photovoltaic output and load data of the planning area, construct the photovoltaic output feature map Laplace matrix and the load feature map Laplace matrix. The constructed matrices are decomposed by the dual-graph regularized non-negative matrix decomposition clustering algorithm. The decomposition results are then used by the k-means clustering algorithm to determine the photovoltaic cluster center curve and the load cluster center curve. The photovoltaic cluster center curve and the load cluster center curve are divided into urban and rural areas; The objective function of the power grid segmentation model is to minimize active power loss and load balance. The constraints of the power grid segmentation model are constructed based on the photovoltaic clustering center curves and load clustering center curves of urban and rural areas, thus obtaining the power grid segmentation model. Solve the power grid segmentation model to obtain the optimal power grid partitioning scheme, and determine the candidate installation locations of the flexible interconnection device based on the optimal power grid partitioning scheme.

2. The flexible interconnection device location method based on distributed resource clustering according to claim 1, characterized in that: Determining the photovoltaic cluster center curve and the load cluster center curve includes: Read photovoltaic output and load data for each time period in the planning area and preprocess the data; Based on the preprocessed data, construct the Laplace matrix of the photovoltaic power output sample graph and the Laplace matrix of the load sample graph; Construct the photovoltaic output feature map Laplace matrix and the load feature map Laplace matrix based on the photovoltaic output sample map Laplace matrix and the load sample map Laplace matrix; Based on the Laplacian matrix of the photovoltaic output characteristic map and the Laplacian matrix of the load characteristic map, the photovoltaic coefficient matrix and the load coefficient matrix, as well as the photovoltaic basis matrix and the load basis matrix, are decomposed into photovoltaic coefficient matrix and load coefficient matrix, photovoltaic basis matrix and load basis matrix by the bi-graph regularized non-negative matrix decomposition clustering algorithm. Based on the photovoltaic coefficient matrix and the load coefficient matrix, the optimal number of clusters for photovoltaic output and load is determined by the elbow method. Based on the photovoltaic base matrix, the load base matrix, the optimal number of clusters for photovoltaic output and load, multiple photovoltaic cluster center curves and load cluster center curves are obtained by the k-means clustering algorithm.

3. The flexible interconnection device location method based on distributed resource clustering according to claim 2, characterized in that: The photovoltaic coefficient matrix and load factor matrix, as well as the photovoltaic basis matrix and load basis matrix, are decomposed using a dual-graph regularized nonnegative matrix factorization clustering algorithm, including: Based on the Laplacian matrix of the photovoltaic power output sample graph and the Laplacian matrix of the photovoltaic power output feature graph, a bigraph regularized nonnegative matrix decomposition clustering objective function for photovoltaics is constructed; a photovoltaic diagonal matrix is ​​constructed; based on the photovoltaic diagonal matrix, the photovoltaic basis matrix, the photovoltaic coefficient matrix, and the photovoltaic sparse error matrix are updated by alternating multiplication; the iteration continues until the bigraph regularized nonnegative matrix decomposition clustering objective function for photovoltaics converges; The bigraph regularized nonnegative matrix decomposition clustering objective function of the load is constructed based on the Laplacian matrix of the load sample map and the Laplacian matrix of the load feature map; the load diagonal matrix is ​​constructed; based on the load diagonal matrix, the load basis matrix, the load coefficient matrix and the load sparsity error matrix are updated by alternating multiplication; the iteration continues until the bigraph regularized nonnegative matrix decomposition clustering objective function of the load converges.

4. The flexible interconnection device location method based on distributed resource clustering according to claim 3, characterized in that: The objective function for bi-graph regularized nonnegative matrix factorization clustering in photovoltaics is expressed by the following formula: In the formula, Represents a photovoltaic matrix. Represents the photovoltaic coefficient matrix. This indicates photovoltaic power output data. express Norm, Represents the photovoltaic sparse error matrix; The weights that control time smoothness are represented by tr, and tr represents the trace of the matrix. The weights represent the control space similarity. Represents the photovoltaic power output characteristic matrix. Represents the Laplace matrix of the photovoltaic sample graph. The superscript represents the weights of the photovoltaic sparse error matrix. This indicates the matrix transpose.

5. The flexible interconnection device location method based on distributed resource clustering according to claim 4, characterized in that: The photovoltaic basis matrix, photovoltaic coefficient matrix, and photovoltaic sparse error matrix are updated as follows: In the formula, This represents element-wise multiplication. Denotes the first photovoltaic diagonal matrix. This represents the second photovoltaic diagonal matrix. Represents the photovoltaic power output sample degree matrix. This represents the similarity matrix of photovoltaic power output samples. express The main part, express The absolute value of the negative part.

6. The flexible interconnection device location method based on distributed resource clustering according to claim 1, characterized in that: The photovoltaic cluster center curve and the load cluster center curve are divided into urban and rural areas, including: The photovoltaic peak value characteristics, photovoltaic smoothness characteristics, and photovoltaic output stability characteristics are determined based on the photovoltaic cluster center curves. The scores of all photovoltaic cluster center curves are constructed based on the photovoltaic peak value characteristics, photovoltaic smoothness characteristics, and photovoltaic output stability characteristics. Among the scores of all photovoltaic cluster center curves, if the score of a certain type of photovoltaic cluster center curve is greater than the set photovoltaic discrimination threshold, that type of photovoltaic cluster center curve is judged as urban; otherwise, it is judged as rural. The load peak-to-valley difference characteristics, load rate characteristics, and load fluctuation characteristics are determined based on the load cluster center curves. The scores of all load cluster center curves are constructed based on the load peak-to-valley difference characteristics, load rate characteristics, and load fluctuation characteristics. Among the scores of all load cluster center curves, if the score of a certain type of load cluster center curve is greater than the set photovoltaic discrimination threshold, the photovoltaic cluster center curve of that type is judged as urban; otherwise, it is judged as rural.

7. The flexible interconnection device location method based on distributed resource clustering according to claim 1, characterized in that: Constraints include topology constraints, system security constraints, switching operation constraints, power flow constraints, distributed generation output constraints, and source-load partitioning constraints. The source-load zoning constraints are constructed based on the photovoltaic clustering center curves and load clustering center curves, which divide the region into urban and rural areas, and are expressed by the following formula: In the formula, Let be the Euclidean distance between curves a and b; Let p be the photovoltaic power output curve for the p-th city area. Let be the photovoltaic power output curve for the q-th rural region. The photovoltaic cluster center curve for urban areas, The photovoltaic cluster center curve for rural areas, This indicates the maximum permissible Euclidean distance within the photovoltaic category in urban areas. This indicates the maximum permissible Euclidean distance within the photovoltaic category in rural areas. and These are categorized as urban and rural photovoltaic systems, respectively. Let m be the load electricity consumption curve for the m-th city area. This is the load electricity consumption curve for the nth rural area. The load cluster center curve for the urban area. The load cluster center curve for rural areas, This represents the maximum permissible Euclidean distance within the urban area load class. This indicates the maximum permissible Euclidean distance within the rural photovoltaic load category. and The categories are urban and rural areas, respectively, for load.

8. The flexible interconnection device location method based on distributed resource clustering according to claim 1, characterized in that: Solving the power supply grid partitioning model includes: Generate a random adjacency matrix as the initial population, calculate the fitness of each adjacency matrix under the objective function, and determine the individual with the highest fitness in the current population; The adjacency matrix is ​​cross-crossed and mutated. The fitness of each individual in the new population after cross-crossing and mutation is recalculated and compared with the fitness of the individual with the highest fitness in the current population. The individual with the highest fitness is updated. The current iteration count is checked to see if it meets the maximum iteration count. If not, the next iteration continues. If it does, the loop is exited, and the optimized final population, the optimal adjacency matrix and its fitness are obtained. Set the optimal adjacency matrix as the optimal power grid partitioning scheme.

9. The flexible interconnection device location method based on distributed resource clustering according to claim 8, characterized in that: Based on the optimal power grid zoning scheme, the candidate installation locations for flexible interconnection devices include: The optimal power grid partitioning scheme is found by depth-first search, which identifies all sets of nodes connected by closed branches. Each set of connected nodes is a power supply area. Identify the set of boundary branches of the power supply area, and select the top 100 locations with the highest comprehensive priority scores from the set of boundary branches of the power supply area to form a list of candidate SOP locations.

10. A flexible interconnection device location system based on distributed resource clustering, comprising the flexible interconnection device location method based on distributed resource clustering as described in any one of claims 1-9, characterized in that: The cluster center curve solving module is used to construct the photovoltaic output feature map Laplace matrix and the load feature map Laplace matrix based on the photovoltaic output and load data of the planning area. The constructed matrices are decomposed by the dual-graph regularized non-negative matrix decomposition clustering algorithm. The decomposition results are then used by the k-means clustering algorithm to determine the photovoltaic cluster center curve and the load cluster center curve. The classification module is used to divide the photovoltaic cluster center curve and the load cluster center curve into urban and rural areas; The model building module is used to construct the constraints of the power grid segmentation model based on the photovoltaic clustering center curves and load clustering center curves of urban and rural areas, with the goal of minimizing active power loss and load balance. The location determination module is used to solve the power grid segmentation model, obtain the optimal power grid partitioning scheme, and determine the candidate installation locations of the flexible interconnection device based on the optimal power grid partitioning scheme.

Images