A new energy distribution area topology correlation analysis method and a storage medium

Abstract

The application discloses a new energy distribution area topology correlation analysis method. Based on voltage characteristic images, the new energy distribution area characteristics are considered to classify node correlation relationships based on knowledge matching, the node pair correlation relationships are modeled in the form of triplets, link prediction is adopted to mine potential node correlation relationships, and finally visual area topology is obtained and output. The application also discloses a corresponding storage medium. By implementing the application, the node correlation relationship can be accurately inferred, a complete area topology graph can be quickly obtained, and the low-voltage distribution network situation awareness capability can be improved.

Description

Technical Field

[0001] This invention relates to the field of distribution network topology identification, and in particular to a method and storage medium for topology association analysis of new energy distribution transformer areas. Specifically, it involves using power measurement data of transformer nodes to perform topology association analysis of new energy low-voltage distribution transformer areas through a knowledge graph-based relational graph convolutional network algorithm, thereby improving the situational awareness capability of low-voltage distribution networks. Background Technology

[0002] As a crucial link connecting the power grid and electricity users, the low-voltage distribution network's state awareness capability is essential for the lean operation of the power grid, with accurate transformer substation topology identification being its core. In recent years, due to the widespread integration of high-proportion distributed photovoltaic power and large-scale electric vehicles into low-voltage distribution substations, bidirectional time-varying dynamic power flows exist between substation topology nodes, fundamentally changing the electrical coupling characteristics in the substation topology network and bringing more challenges to substation topology correlation analysis.

[0003] Current research on distribution network topology correlation analysis based on data features at home and abroad mainly relies on electrical characteristic quantities such as voltage and power for judgment. Various algorithms and other technologies are combined to extract feature analysis of node connection relationships in target data. For example, the correlation of nodes is analyzed based on distribution network measurement data, and the topology is inferred using greedy algorithms, maximum spanning tree algorithms, etc.

[0004] However, the widespread integration of numerous distributed resources into low-voltage distribution networks has exacerbated the challenges of active distribution networks and bidirectional power flow, making the analysis of voltage correlation and electrical coupling characteristics in transformer substation topologies more difficult and affecting the accuracy of traditional methods. Therefore, there is an urgent need for a method for analyzing low-voltage transformer substation topology correlations that considers the integration of new energy sources. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a method and storage medium for topology correlation analysis of new energy distribution substations. This method can accurately infer node relationships, quickly obtain a complete substation topology map, and improve the situational awareness capability of low-voltage distribution networks.

[0006] The technical solution adopted in this invention is as follows: One aspect of this invention provides a method for topological correlation analysis of new energy distribution substations, comprising the following steps:

[0007] Step S1: Define the relationships between node ontologies based on the characteristics of new energy distribution areas, and perform knowledge graph ontology modeling for low-voltage distribution areas.

[0008] Step S2: Update the feature profile of the transformer area nodes based on the voltage time series data, and integrate the prior topology to construct the entity and relationship knowledge graph of the low-voltage transformer area;

[0009] Step S3: Store the node relationships in the form of triples, generate a set of positive and negative samples, and complete the construction of the knowledge graph of the transformer area;

[0010] Step S4: Initialize the parameters of the relational graph convolutional network model and divide it into training and test sets;

[0011] Step S5: Input the node association triples and adjacency matrix into the relation graph convolutional network to update the node and association features;

[0012] Step S6: Aggregate normalized neighbor node information and handle overfitting, update hidden layer node information, and output node adjacency matrix.

[0013] Step S7: Use the DisMult decoder to evaluate the reference weights of the node association vector, and optimize the association of the node with the highest output ranking through cross-entropy loss;

[0014] Step S8: Determine whether the loss function has converged or reached the maximum number of iterations. If it has converged or the condition has been met, save the model parameters; otherwise, proceed to step S5.

[0015] Step S9: Use the trained network model to perform online topology association analysis on the test set data; output the prediction result with the highest reference weight, and output the visualized transformer area topology based on the adjacency matrix.

[0016] Preferably, step S1 further includes:

[0017] Step S10: In the low-voltage distribution network area where a high proportion of distributed photovoltaic power generation and large-scale electric vehicles are widely connected, the topology is simplified.

[0018] Step S11: Using distribution transformer nodes and user nodes as entities in the knowledge graph, construct the knowledge graph from top to bottom; based on the characteristics of new energy distribution areas and topology-related knowledge, divide the relationship between distribution area topology nodes into separation relationship, series relationship, parallel relationship, photovoltaic acceptance relationship and electric vehicle acceptance relationship;

[0019] Step S12: Based on the topological node association relationship of the distribution area, the topology of the new energy distribution area is described in the form of a directed adjacency matrix Y. The association relationship of all nodes in the distribution area is characterized and the node type is represented in the directed adjacency matrix.

[0020] Preferably, step S2 further includes:

[0021] Step S20: Extract the voltage of each node, set a time window, extract the time series of the initial data of equal length, and construct a feature profile of the distribution substation node based on the voltage time series curve.

[0022] Step S21: Update the feature profile of the transformer area nodes based on the voltage time series data, and construct the entity and relationship knowledge graph of the low-voltage transformer area by integrating prior topological knowledge.

[0023] Preferably, step S3 further includes:

[0024] The obtained node associations are stored in the form of triples of "node1-association-node2". The node entities are connected through relationships, forming a knowledge graph of the low-voltage distribution substation topology.

[0025] Since all possible relationships between nodes have been defined, there are only positive samples and no negative samples in the sample. A negative sample set is constructed by uniformly sampling and replacing any node in the node pair with equal probability.

[0026] Preferably, step S4 further includes:

[0027] Step S40: Initialize the parameters of the graph convolutional network; set the learning rate a0, the number of iterations Epoch, the hidden layer dimension n, and the hidden layer neuron h_dim parameter of the graph convolutional network.

[0028] Step S41: Input the voltage data of the transformer substation nodes and divide the dataset into a training set, a validation set, and a test set. The number of samples in the relational graph convolutional network is taken as the total number of nodes and edges. The training set and the test set are divided in a 4:1 ratio.

[0029] Preferably, step S5 further includes:

[0030] Step S50: Input the triplet of the node association relationship and the adjacency matrix of the transformer area into the relation graph convolutional network to extract features of the voltage profile of all nodes in the transformer area.

[0031] Step S51: Based on the node time series features formed by the node voltage time series data, a relational graph convolutional network is used as an encoder to extract features from them, and the high-dimensional features of the node time series voltage are mapped to the low-dimensional vectors of the topological nodes.

[0032] Preferably, step S6 further includes:

[0033] Basis function decomposition and block diagonal decomposition are introduced to define the model's weight updates, aggregate and normalize neighbor node information, and handle overfitting to update the hidden layer node information H. i l+1 Output the node vectors and the association matrix.

[0034] Preferably, step S7 further includes:

[0035] Step S70: Based on the node association relationship adjacency matrix output by the relation graph convolutional network, and combined with the DisMult decoder, link prediction is performed on the association relationship between the nodes in the transformer area using the link prediction model. The link prediction model includes at least a relation graph convolutional network encoder and a DisMult decoder. The DisMult decoder evaluates the reference weight for each possible edge relationship in the topology graph, and the edge loss is calculated for each edge.

[0036] Step S71: The cross-entropy loss is used to optimize the reference weight evaluation results of the DistMult decoder, so that the scores of the observable positive samples are higher than those of the negative samples. The reference weight evaluation results are sorted, and the triplet of the node association relationship with the largest reference weight value is output as the prediction result.

[0037] Preferably, the simplification of the topology in low-voltage distribution substations with a high proportion of distributed photovoltaic power and widespread access to large-scale electric vehicles further includes:

[0038] Only the relationship identification of the end nodes is considered, namely regular user nodes, distributed photovoltaic nodes, and electric vehicle nodes. Branch nodes without measurement devices are not considered.

[0039] Only normally closed switches are retained, and the circuit is simplified to a regular power line;

[0040] Redundant power lines are merged, and only one connection line is retained between adjacent different nodes.

[0041] Accordingly, another aspect of the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the method as described above.

[0042] Implementing the embodiments of the present invention has the following beneficial effects:

[0043] This invention provides a method and storage medium for topology correlation analysis of new energy distribution substations. Addressing the diverse node types and complex electrical quantity correlation characteristics of new energy distribution substations, this invention employs five correlation relationships—separation, series, parallel, photovoltaic integration, and electric vehicle integration—to characterize the heterogeneous topology of the substations. The obtained node correlation relationships are stored as triples of "node 1 - correlation relationship - node 2," with node entities connected through relationships, forming a knowledge graph of the low-voltage distribution substation topology. This invention performs attribute analysis and feature extraction on voltage measurement data, combining it with prior topological knowledge to form a feature profile of substation nodes. A knowledge graph-based relational graph convolutional network algorithm is used to predict node correlations, determining node types and identifying missing node pair correlations, thus generating a complete distribution substation topology. This improves the efficiency and accuracy of obtaining the complete distribution substation topology, thereby enhancing the situational awareness capability of the low-voltage distribution network. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, obtaining other drawings based on these drawings without creative effort still falls within the scope of the present invention.

[0045] Figure 1 This is a schematic diagram of the main flow of an embodiment of a new energy distribution area topology correlation analysis method provided by the present invention;

[0046] Figure 2 for Figure 1 A diagram illustrating the types of node-to-node relationships involved;

[0047] Figure 3 for Figure 1 A schematic diagram illustrating the topological adjacency matrix of the new energy distribution station area involved in the document;

[0048] Figure 4 for Figure 1 A schematic diagram of the update of a single node in the relational graph convolutional network involved in the process;

[0049] Figure 5 for Figure 1 A schematic diagram of the link prediction model involved;

[0050] Figure 6 This is a schematic diagram of the node area in one example of the present invention;

[0051] Figure 7 This is a schematic diagram comparing the results of different algorithms in one example of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.

[0053] like Figure 1 The diagram shown illustrates the main flowchart of an embodiment of a new energy distribution area topology correlation analysis method provided by the present invention; in conjunction with... Figures 2 to 5 As shown, in this embodiment, the method includes at least the following steps:

[0054] Step S1: Define the relationships between node ontologies based on the characteristics of new energy distribution areas, and perform knowledge graph ontology modeling for low-voltage distribution areas.

[0055] In a specific example, step S1 further includes:

[0056] Step S11: In the low-voltage distribution network area where a high proportion of distributed photovoltaic power generation and large-scale electric vehicles are widely connected, the topology is simplified.

[0057] Understandably, in low-voltage distribution substations with a high proportion of distributed photovoltaic power and widespread large-scale electric vehicle integration, the topology information is more complex than in traditional substations. The relationships between different nodes can no longer be simply divided into connection and separation relationships, but have evolved into a multi-relationship heterogeneous graph. To facilitate substation topology correlation analysis and avoid the redundancy of the vast and complex low-voltage distribution network topology information, the following simplification is performed:

[0058] (1) Only the relationship identification of the end nodes is considered, namely, conventional user nodes, distributed photovoltaic nodes, and electric vehicle nodes. Branch nodes without measurement devices are not considered.

[0059] (2) Only normally closed switches are retained, and the circuit is simplified to a regular power line;

[0060] (3) The redundant power lines are merged, and only one connection line is retained between adjacent different nodes.

[0061] Step S11: Knowledge graph technology can clearly analyze the relationships between entities, meeting the requirements of distribution transformer area topology association analysis. Using distribution transformer nodes and user nodes as entities in the knowledge graph, the knowledge graph is constructed from top to bottom. First, the relationships between node entities are defined. Based on the characteristics of new energy distribution areas and topology-related knowledge, the relationships between distribution area topology nodes are divided into separation relationships, series relationships, parallel relationships, photovoltaic acceptance relationships, and electric vehicle acceptance relationships. The node relationship representation is shown in Table 1, and the schematic diagram of the node-to-node relationship types is shown below. Figure 1 As shown.

[0062] Table 1 Node Association Types

[0063]

[0064] Among them, the separation relationship: If there is no direct electrical connection between nodes, that is, there are no other nodes on the line connecting the two ends of the node pair, the node relationship is called a "separation relationship". The separation relationship is the most common relationship in transformer substations, and it is also the only "non-adjacent" relationship. The expression for the separation relationship is as follows:

[0065] x ij =x ji =0 (1)

[0066] Series Relationship: If there is a direct electrical connection between nodes and a stable power transmission direction, meaning both nodes are conventional user nodes, this node relationship is called a "series relationship." Taking the power supply path from transformer to line to node as the main power supply line, the node that is electrically closer to the transformer is the parent node in this node pair. The expression for the series relationship is as follows:

[0067] x ij =-x ji =1 (2)

[0068] Parallel connection: If there is no direct electrical connection between nodes, but they share a common power source direction, i.e., they share the same parent node, the node relationship is called a "parallel connection." Parallel connections differ from separation connections; they are closer in electrical distance and may involve branching branches in practice. The expression for a parallel connection is shown below:

[0069] x ij =x ji =-2 (3)

[0070] Photovoltaic Adoption Relationship: If there is a direct electrical connection between nodes, and one of these nodes is a distributed photovoltaic (PV) node, this node association is called a "PV adoption relationship." The PV adoption relationship is used specifically to describe the connection relationships between PV nodes. The existence of distributed PV nodes makes the low-voltage distribution substation a multi-source network, but it lacks the continuous power supply capability of a distribution transformer, exhibiting distinct characteristics different from conventional user nodes and distribution transformer nodes. Therefore, the relationship between PV nodes and adjacent nodes should not be discussed in conjunction with the conventional "series relationship." The expression for the PV adoption relationship is as follows:

[0071] x ij =-x ji =3 (4)

[0072] Electric Vehicle Admission Relationship: If there is a direct electrical connection between nodes, and one of these nodes is an electric vehicle node, this node association is called an "electric vehicle admission relationship." Similar to photovoltaic admission relationships, electric vehicle admission relationships are specifically used to describe the connection relationships of electric vehicle nodes. The spatiotemporal distribution characteristics of electric vehicle nodes differ significantly from those of conventional user nodes, therefore it is necessary to define their association relationships separately. The expression for an electric vehicle admission relationship is as follows:

[0073] x ij =-x ji =4 (5)

[0074] Step S12: Based on the topological node association relationship of the distribution area, the topology of the new energy distribution area is described in the form of a directed adjacency matrix Y. The association relationship of all nodes in the distribution area is characterized and the node type is represented in the directed adjacency matrix.

[0075] It is understandable that each node has a unique association type. Based on the above relationships, the topology of the new energy distribution area is described in the form of a directed adjacency matrix Y, as follows: Figure 3 As shown, Figure 3 In the diagram, nodes 1, 2, and 5 represent regular user nodes, node 4 represents a distributed photovoltaic node, and node 3 represents an electric vehicle node. Unlike traditional directed adjacency matrices, the directed adjacency matrix for new energy distribution areas not only depicts the relationships between all nodes within the area but also simultaneously represents the node types.

[0076] Step S2: Update the feature profile of the transformer area nodes based on the voltage time series data, and integrate the prior topology to construct the entity and relationship knowledge graph of the low-voltage transformer area.

[0077] Specifically, the process includes: Step S20, extracting the voltage of each node, setting a time window, extracting equal-length time series data from the initial data, and constructing a node feature profile of the distribution substation based on the voltage time series curve. The topology correlation analysis results are related to the node feature dimension. Increasing the node feature dimension, i.e., shortening the voltage time series data sampling interval, will increase the accuracy of the topology correlation analysis. However, when the node feature dimension reaches a certain level, the improvement effect on the accuracy of the topology correlation analysis is no longer significant, and it increases the difficulty of data analysis. Generally, the sampling interval is set to 15 minutes.

[0078] Step S21: Update the feature profile of the transformer area nodes based on the voltage time series data, and construct the entity and relationship knowledge graph of the low-voltage transformer area by integrating prior topological knowledge.

[0079] Step S3: Store the node relationships in the form of triples, generate a set of positive and negative samples, and complete the construction of the knowledge graph of the transformer area.

[0080] Specifically, the obtained node associations are stored in the form of triples of "node1-association-node2". The node entities are connected through relationships, forming a knowledge graph of the low-voltage distribution substation topology.

[0081] Since all possible relationships between nodes have been defined, there are only positive samples and no negative samples in the sample. By uniformly sampling and replacing any node in the node pair with equal probability, a negative sample set is constructed to improve the generalization ability of the model.

[0082] More specifically, the construction process of the positive and negative sample sets is as follows: For a given node pair, determine the relationship, and check if the relationship exists between the two nodes. If it exists, the node association relationship is taken as a positive sample. An incorrect node association relationship is generated by changing any node in the positive sample node pair, and this is taken as a negative sample. Assuming E is the set of triplet relationships of all transformer area nodes, and F is the generated negative sample set, the generation method is as follows:

[0083] F r ={(n i ,r F ,n j )|r F ∈E∧r F ≠r∧(n i ,r,n j )∈E} (6)

[0084] In the formula, (n i ,r,n j ) represents the true triplet of the association relationship between the nodes in the transformer area, (n i ,r F ,n j ) represents a negative triplet representing the association relationship between nodes in the distribution area.

[0085] Step S4: Initialize the parameters of the relational graph convolutional network model and divide it into training and test sets.

[0086] Specifically, the steps include: Step S40, initializing the parameters of the graph convolutional network. This involves setting parameters such as the learning rate a0, the number of iterations Epoch, the hidden layer dimension n, and the hidden layer neurons h_dim.

[0087] Step S41: Input the voltage data of the transformer substation nodes and divide the dataset into a training set, a validation set, and a test set. The number of samples in the graph convolutional network is equal to the total number of nodes and edges, and the training set and test set are divided in a 4:1 ratio.

[0088] Understandably, as the topology of the new energy access backend area evolves from a homogeneous graph to a heterogeneous graph, traditional topology correlation analysis methods are no longer applicable. To fully explore the complex electrical quantity correlation characteristics between nodes in the new energy distribution area, a Relational-graph Convolutional Network (R-GCN) is used to analyze node correlation characteristics through link prediction, determining node types while performing topology correlation analysis. Graph Convolutional Network (GCN) is a natural extension of convolutional neural networks in the graph domain. It can extract features from graph data to complete tasks such as node classification and link prediction. Compared to previous graph neural network algorithms, GCN can achieve shared weight parameters, reducing computational complexity and exhibiting local connectivity. However, it considers homogeneous graph modeling, where node pairs in the graph have consistent correlations, making it unsuitable for new energy distribution area topology identification. Relational graph convolutional networks are designed for heterogeneous graph modeling. Based on the node update method of graph convolutional networks, they take into account different relationship types between nodes. When processing neighbor nodes, different weight parameters are introduced for each type of relationship. After aggregating neighbors belonging to the same relationship type, a total aggregation is performed. This method can undertake the task of topological correlation analysis under new energy transformer substations.

[0089] Step S5: Input the node association triples and adjacency matrix into the relation graph convolutional network to update the node and association features.

[0090] Specifically, the steps include: Step S50, inputting the triplet of the node association relationship and the adjacency matrix of the transformer area into the relation graph convolutional network to extract features of the voltage profile of all nodes in the transformer area.

[0091] Step S51: Based on the node time series features formed by the node voltage time series data, a relational graph convolutional network is used as an encoder to extract features from them, and the high-dimensional features of the node time series voltage are mapped to the low-dimensional vectors of the topological nodes.

[0092] The coding principle will be introduced below.

[0093] The node information is updated based on the input node profile, and the calculation formula is as follows:

[0094] H 0 =V=[V1,V2,...,V N (7)

[0095]

[0096] In the formula, H 0 To input initial network information, this means inputting the node voltage profile information V into the R-GCN network; Represents the node n in the transformer area.i Hiding feature information; σ is the ReLU activation function; This is the regularized weight matrix. Let R be its own weight matrix; R = {separation relationship, series relationship, parallel relationship, photovoltaic acceptance relationship, electric vehicle acceptance relationship}, where r represents the r-th relationship under set R; c represents the set of neighboring nodes connected to node i when the relation is r; i,r Set a pre-defined standardization constant.

[0097] A schematic diagram of a single node update in a graph convolutional network is shown below. Figure 4 As shown in the diagram, the central node of each part is the node to be updated, and the surrounding nodes are its neighbors. There are N types of association relationships, denoted as rel_1 to rel_N. Each association relationship can be further divided into input and output categories. Each round of node updates requires collecting information from neighboring nodes (the square labeled 1), transforming it according to different association relationship types, standardizing it, and summing the results to obtain the result (the square labeled 2). Finally, this result is passed through an activation function (such as ReLU) to complete the update of the node information (the square labeled 3) for that round. This single-node update method can be computed in parallel with shared parameters across the entire graph.

[0098] In a specific example, a topology graph can be generated by reasoning about the relationships between nodes in the platform area. First, the specific relationships of each node are improved by aggregating the neighbor node information under each relationship. Then, the neighbor node information under all relationships is vector-normalized and superimposed to obtain the total neighbor node information of each node. Finally, this information is integrated with the node's own historical information to obtain the current hidden feature information of that node.

[0099] Step S6: Aggregate normalized neighbor node information and handle overfitting, update hidden layer node information, and output node adjacency matrix.

[0100] Specifically, when applying equations (7) and (8) to the calculation of multi-relationship station node data, the parameters of the relation graph convolutional network increase rapidly, leading to overfitting. To address this, basis function decomposition and block diagonal decomposition are introduced to define the model's weight updates, aggregate and normalize neighbor node information, and handle overfitting, updating the hidden layer node information H. i l+1 The output node vectors and association matrix are then explained. The principles of basis function decomposition and block diagonal decomposition are introduced below.

[0101] The basis functions allow weights from different relation types to be shared, and the decomposition formula is shown below:

[0102]

[0103] In the formula, As base With coefficient linear combination, coefficients It depends only on relation r.

[0104] The block diagonal decomposition representation applies sparse constraints to the weight matrix for each relation type, which can be obtained by directly summing low-dimensional matrices, as shown in the following formula:

[0105]

[0106] In the formula, Defined as a low-dimensional sparse matrix The diagonal matrix obtained by direct summation can be regarded as sparse constraints of different relation types.

[0107] Basis function decomposition and block diagonal decomposition both reduce the number of parameters required to learn multi-relation data. At the same time, since parameter updates for rare relations and regular relations are shared, overfitting to rare relations can be avoided.

[0108] Step S7: Use the DisMult decoder to evaluate the reference weights of the node association vector, and optimize the association of the node with the highest output ranking through cross-entropy loss.

[0109] Specifically, the process includes the following steps: Step S70, based on the node association adjacency matrix output by the relational graph convolutional network, and combined with the DistMult decoder, link prediction is performed on the association relationships between nodes in the transformer area. The link prediction model is as follows: Figure 5 As shown, it consists of a relational graph convolutional network encoder and a DisMult decoder. The DisMult decoder evaluates the reference weights for each possible edge relationship in the topology graph, and the edge loss is calculated for each edge.

[0110] By evaluating the similarity of positive and negative sample node association triples using reference weights, the correlation between node pairs and their associations is obtained, thereby assessing the node pairs and their association triples (n) that appear during training. i ,r,n j ), (n i ,r F ,n j The higher the reference weight value, the greater the probability that the fact is true. Taking the evaluation of the triplet of positive sample node association as an example, the calculation formula is as follows:

[0111]

[0112] In the formula, These represent the node pairs, with the first and last nodes n respectively. i nj Vector, R r Let r be the matrix-vector representation of the relation.

[0113] Step S71: The cross-entropy loss is used to optimize the reference weight evaluation results of the DistMult decoder, so that the scores of the observable positive samples are higher than those of the negative samples. The calculation formula is shown in Equation (12). The reference weight evaluation results are sorted, and the triplet of the node association relationship with the largest reference weight value is output as the prediction result.

[0114]

[0115] In the formula, T represents the set of real node associations; l is the sigmoid function; y represents the positive and negative sample states, where y = 1 indicates that the sample is a positive sample, and y = -1 indicates that the sample is a negative sample.

[0116] Step S8: Determine whether the loss function has converged or whether the maximum number of iterations is met. If it has converged or the maximum number of iterations is met, save the model parameters; otherwise, return to step S5.

[0117] Step S9: Use the trained network model to perform online topology association analysis on the test set data; output the prediction result with the highest reference weight, and output the visualized transformer area topology based on the adjacency matrix.

[0118] Specifically, the model parameters are saved, and the trained relational graph convolutional network model is used to perform online topological association analysis on the node relationships of the test set. The inference result with the highest reference weight is output and the result is saved to the node association dataset. Based on the adjacency matrix of node association relationships, a visualized low-voltage distribution transformer area topology is output.

[0119] It is understood that the method provided by this invention is based on voltage feature profiling, considers the characteristics of new energy distribution substations, classifies node association relationships based on knowledge matching, models node pair association relationships in the form of triples, and uses link prediction to mine potential node association relationships. Compared with traditional methods, it can more accurately infer node association relationships, and the substation topology map will also be more complete.

[0120] The following will use a specific simulation example to further understand the essence and advantages of the method of the present invention.

[0121] Simulation example:

[0122] The parameters of the graph convolutional network model were set according to Table 2 for training. Regarding data encoding, since the size of the topological nodes in the transformer area to be identified is uncertain, the One-Hot encoding method (using an N-bit state register to encode N states) is no longer applicable. Instead, the LabelEncoder method can be used to encode the voltage characteristics of the transformer area nodes into integers between 0 and N-1 (where N is the number of categories) to reduce redundant information input and improve network operating efficiency.

[0123] Table 2 Training Parameters for Relationship Graph Convolutional Network

[0124]

[0125]

[0126] The verification was conducted in a 71-node distribution area. Node features (extracted from voltage time-series data) and adjacency matrices (obtained from initial records) were input into a graph convolutional network model. The model parameters were used to perform online inference of the topological node relationships. The resulting adjacency matrix was plotted as a topological graph for visualization. The results are as follows: Figure 6 As shown in the figure, node 0 is the distribution transformer node, nodes (58, 59, 61, 63, 64, 68, 71) are distributed photovoltaic nodes; nodes (60, 62, 65, 66, 67, 69, 70) are electric vehicle nodes, and the other nodes are conventional user nodes.

[0127] To verify the superiority of the proposed method, it was compared with non-machine learning algorithms such as the Greedy Algorithm (GA) and other commonly used machine learning algorithms such as Deep Neural Networks (DNN), Support Vector Machines (SVM), and Graph Convolutional Networks (GCN). The results are as follows: Figure 6 As shown.

[0128] The simulation results show that the new energy distribution area topology association analysis method (R-GCN) based on graph convolution relationship reasoning proposed in this invention can adapt to the new situation that arises after a high proportion of new energy is connected to the distribution area, determine the node type and accurately predict the association relationship of node pairs, and complete the task of new energy distribution area topology association analysis.

[0129] Accordingly, another aspect of the present invention also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the aforementioned... Figures 1 to 7 The method described. For more details, please refer to and combine with the foregoing discussion. Figures 1 to 7The description of that will not be repeated here.

[0130] Implementing the embodiments of the present invention has the following beneficial effects:

[0131] This invention provides a method and storage medium for topology correlation analysis of new energy distribution substations. Addressing the diverse node types and complex electrical quantity correlation characteristics of new energy distribution substations, this invention employs five correlation relationships—separation, series, parallel, photovoltaic integration, and electric vehicle integration—to characterize the heterogeneous topology of the substations. The obtained node correlation relationships are stored as triples of "node 1 - correlation relationship - node 2," with node entities connected through relationships, forming a knowledge graph of the low-voltage distribution substation topology. This invention performs attribute analysis and feature extraction on voltage measurement data, combining it with prior topological knowledge to form a feature profile of substation nodes. A knowledge graph-based relational graph convolutional network algorithm is used to predict node correlations, determining node types and identifying missing node pair correlations, thus generating a complete distribution substation topology. This improves the efficiency and accuracy of obtaining the complete distribution substation topology, thereby enhancing the situational awareness capability of the low-voltage distribution network.

[0132] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0134] The above description is merely a preferred embodiment of the present invention and should not be construed as limiting the scope of the invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for analyzing the topological correlation of new energy distribution areas, characterized in that, Includes the following steps: Step S1: Define the relationships between node ontologies based on the characteristics of new energy distribution areas, and perform knowledge graph ontology modeling for low-voltage distribution areas. Step S2: Update the feature profile of the transformer area nodes based on the voltage time series data, and integrate the prior topology to construct the entity and relationship knowledge graph of the low-voltage transformer area; Step S3: Store the node relationships in the form of triples, generate a set of positive and negative samples, and complete the construction of the knowledge graph of the transformer area; Step S4: Initialize the parameters of the relational graph convolutional network model and divide it into training and test sets; Step S5: Input the node association triples and adjacency matrix into the relation graph convolutional network model to update the node association features; Step S6: Aggregate normalized neighbor node information and handle overfitting, update hidden layer node information, and output node adjacency matrix. Step S7: Use the DistMult decoder to evaluate the reference weights of the node association vector, and optimize the association of the node with the highest output ranking through cross-entropy loss; Step S8: Determine whether the loss function has converged or reached the maximum number of iterations. If it has converged or the condition has been met, save the parameters of the graph convolutional network model; otherwise, proceed to step S5. Step S9: Use the trained relational graph convolutional network model to perform online topological association analysis on the test set data; Output the prediction result with the highest reference weight, and output the visualized transformer area topology based on the node adjacency matrix; Step S1 further includes: Step S10: In the low-voltage distribution network area where a high proportion of distributed photovoltaic power generation and large-scale electric vehicles are widely connected, the topology is simplified. Step S11: Using distribution transformer nodes and user nodes as entities in the knowledge graph, construct the knowledge graph from top to bottom; based on the characteristics of new energy distribution areas and topology-related knowledge, divide the relationship between distribution area topology nodes into separation relationship, series relationship, parallel relationship, photovoltaic acceptance relationship and electric vehicle acceptance relationship; Step S12: Based on the topological node association relationship of the distribution area, the topology of the new energy distribution area is described in the form of a directed adjacency matrix Y. The association relationship of all nodes in the distribution area is depicted and the node type is characterized in the directed adjacency matrix Y. Step S3 further includes: The obtained node associations are stored in the form of triples of "node1-association-node2". The node entities are connected through relationships, forming a knowledge graph of the low-voltage distribution substation topology. Since all possible relationships between nodes have been defined, there are only positive samples and no negative samples in the sample. A negative sample set is constructed by uniformly sampling and replacing any node in the node pair with equal probability.

2. The method as described in claim 1, characterized in that, Step S2 further includes: Step S20: Extract the voltage of each node, set a time window, extract the time series of the initial data of equal length, and construct a feature profile of the transformer area nodes based on the voltage time series curve. Step S21: Update the feature profile of the transformer area nodes based on the voltage time series data, and construct the entity and relationship knowledge graph of the low-voltage transformer area by integrating prior topological knowledge.

3. The method as described in claim 2, characterized in that, Step S4 further includes: Step S40: Initialize the parameters of the graph convolutional network model; set the learning rate, number of iterations, hidden layer dimension, and hidden layer neuron parameters of the graph convolutional network model. Step S41: Input the voltage data of the transformer substation nodes and divide the dataset into a training set, a validation set, and a test set. The number of samples in the relational graph convolutional network model is taken as the total number of nodes and edges. The training set and the test set are divided according to a predetermined ratio.

4. The method as described in claim 3, characterized in that, Step S5 further includes: Step S50: Input the node association triplet and adjacency matrix into the relation graph convolutional network model to extract features of the voltage profile of all nodes in the transformer area; Step S51: Based on the node time-series data, the node time-series features are extracted using a relational graph convolutional network model as an encoder, and the high-dimensional features of the node time-series features are mapped to low-dimensional vectors of the topological nodes.

5. The method as described in claim 4, characterized in that, Step S6 further includes: Basis function decomposition and block diagonal decomposition are introduced to define the model's weight update, aggregate and normalize neighbor node information and handle overfitting, update hidden layer node information, and output node vectors and association matrix.

6. The method as described in claim 5, characterized in that, Step S7 further includes: Step S70: Based on the node association adjacency matrix output by the relational graph convolutional network model, the link prediction model is used to predict the association between the nodes in the transformer area in combination with the DistMult decoder. The link prediction model includes at least a relational graph convolutional network model encoder and a DistMult decoder. The DistMult decoder evaluates the reference weight for each possible edge relationship in the topology graph, and the edge loss is calculated for each edge. Step S71: The cross-entropy loss is used to optimize the reference weight evaluation results of the DistMult decoder, so that the scores of the model's observable positive samples are higher than those of the negative samples. The reference weight evaluation results are sorted, and the triplet of the node association relationship with the largest reference weight value is output as the prediction result.

7. The method according to any one of claims 2 to 6, characterized in that, In the low-voltage distribution substation area where a high proportion of distributed photovoltaic power generation and large-scale electric vehicles are widely connected, the topology is simplified, which further includes: Only the relationship identification of the end nodes is considered, namely regular user nodes, distributed photovoltaic nodes, and electric vehicle nodes. Branch nodes without measurement devices are not considered. Only normally closed switches are retained, and the circuit is simplified to a regular power line; Redundant power lines are merged, and only one connection line is retained between adjacent different nodes.

8. A computer-readable storage medium storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the method as described in any one of claims 1 to 7.

Images