Power system section control method

Abstract

The invention relates to a power system section control method and device, computer equipment, a readable storage medium and a computer program product. The method comprises the following steps: constructing a power grid point-line model diagram according to a bus object set and a branch object set of a power system; acquiring a section object set of the power system, and constructing a section tripartite graph according to the bus object set, the branch object set and the section object set; performing pooling operation on the power grid point-line model graph to obtain a core network frame graph of the power system; performing pooling operation on the section tripartite graph to obtain a section clustering graph of the power system; fusing the core network frame graph and the section clustering graph to obtain fused graph features of the power system; and inputting the fusion image features into a pre-trained power system section control model to obtain a power system section control action for controlling the power system section. By adopting the method, the accuracy of section control of the power system can be improved.

Description

Technical Field

[0001] This application relates to the field of power system control technology, and in particular to a power system cross-section control method, apparatus, computer equipment, readable storage medium, and computer program product. Background Technology

[0002] With the increasing demands for real-time dispatch and security control in power systems, power grid operation modes are becoming increasingly complex and variable. Optimal control of the power grid requires consideration of unit output optimization, especially the satisfaction of physical constraints on the power grid, including constraints on numerous power system cross-sections.

[0003] Traditional optimization methods for control strategies based on effective cross sections in power systems mostly employ algorithms with fixed input dimensions, such as linear neural network structures, to process cross section information. These methods cannot effectively adapt to the dynamically changing number of cross sections. Furthermore, the processing of neural networks in the grid point-line model graph uses simple pooling operations to aggregate information, which cannot effectively handle the complex correspondence between nodes and edges in the model graph. This makes it difficult to achieve accurate fusion of grid topology and constraint information, resulting in insufficient accuracy of the generated power system cross section control strategies and failing to meet the urgent need for real-time and precise control of modern power grids.

[0004] Therefore, traditional technologies suffer from low accuracy in power system cross-sectional control strategies. Summary of the Invention

[0005] Therefore, it is necessary to provide a power system control method, device, computer equipment, readable storage medium, and computer program product based on power system cross-sectional model and point-line model that can improve the accuracy of power system control, addressing the aforementioned technical problems.

[0006] In a first aspect, this application provides a power system control method based on a power system cross-sectional model and a point-line model, the method comprising:

[0007] Based on the set of bus objects and the set of branch objects in the power system, a power grid point-line model diagram is constructed; the power grid point-line model diagram is used to represent the relationship between bus objects and branch objects in the power system.

[0008] Obtain the set of cross-sectional objects of the power system, and construct a cross-sectional three-part heterogeneous diagram based on the set of bus objects, the set of branch objects, and the set of cross-sectional objects; the cross-sectional three-part heterogeneous diagram is used to characterize the relationship between the bus objects and cross-sectional objects, and the relationship between the branch objects and cross-sectional objects in the power system.

[0009] A pooling operation is performed on the power grid point-line model diagram to obtain the core network diagram of the power system; a pooling operation is performed on the cross-sectional three-part heterogeneous diagram to obtain the cross-sectional cluster diagram of the power system; the core network diagram and the cross-sectional cluster diagram are fused to obtain the fused diagram features of the power system.

[0010] The fused graph features are input into a pre-trained power system cross-section control model to obtain power system cross-section control actions; the power system cross-section control actions are used to control the power system cross-section, and the pre-trained power system cross-section control model is a model obtained through reinforcement learning training.

[0011] In one embodiment, obtaining the set of cross-sectional objects of the power system and constructing a cross-sectional three-part heterogeneous diagram based on the set of bus objects, the set of branch objects, and the set of cross-sectional objects includes:

[0012] The cross-section object set is obtained based on the busbar object set and the branch object set;

[0013] Based on the busbar object set, the branch object set, and the cross-section object set, first association information and second association information are determined; the first association information is used to record the association relationship between the busbar object and the cross-section object; the second association information is used to record the association relationship between the branch object and the cross-section object.

[0014] Based on the bus object set, the branch object set, the cross-section object set, the first association information, and the second association information, the cross-section three-part heterogeneous diagram is constructed;

[0015] The cross-sectional triadic graph includes bus nodes corresponding to each bus object in the bus object set, branch nodes corresponding to each branch object in the branch object set, cross-sectional nodes corresponding to each cross-sectional object in the cross-sectional object set, as well as first-type edges and second-type edges. The first-type edges are used to characterize the association between a bus object and a cross-sectional object; the second-type edges are used to characterize the association between a branch object and a cross-sectional object.

[0016] In one embodiment, after the step of constructing the cross-sectional tri-heterogeneous diagram based on the bus object set, the branch object set, the cross-section object set, the first association information, and the second association information, the method further includes:

[0017] Obtain the current relation type of the current node in the cross-sectional heterogeneous graph, and the convolution weights corresponding to the current relation type in the network layer above the current network layer where the current node is located; the current node is any one of the bus nodes, branch nodes, and cross-sectional nodes; the current relation type is any one of the relation types corresponding to the first type edge and the relation types corresponding to the second type edge;

[0018] For the current node, obtain the neighbor node embeddings of the current node's neighbor nodes in the previous network layer under the current relationship type, and the normalization coefficient of the current node for the current relationship type;

[0019] Based on the convolution weights, the neighbor node embeddings, and the normalization coefficients, the current node embedding corresponding to the current node in the previous network layer under the current relation type is updated to obtain the current node embedding corresponding to the current node in the current network layer under the current relation type.

[0020] Based on the current node embedding corresponding to each current node in the current network layer under each current relation type, the node feature matrix of the cross-sectional ternary heterogeneous graph and the adjacency matrix of the cross-sectional ternary heterogeneous graph are obtained; and based on the node feature matrix and the adjacency matrix of the cross-sectional ternary heterogeneous graph, the node embedding matrix of the cross-sectional ternary heterogeneous graph is obtained.

[0021] In one embodiment, the step of performing a pooling operation on the cross-sectional triadic graph to obtain the cross-sectional clustering graph of the power system includes:

[0022] Obtain the adjacency matrix and node feature matrix of the cross-sectional ternary heterogeneous graph, and obtain the node embedding matrix based on the adjacency matrix and the node feature matrix;

[0023] The node embedding matrix is ​​updated according to the pre-set cross-section coefficient matrix to generate a pooled node feature matrix; the adjacency matrix is ​​updated according to the cross-section coefficient matrix to generate a pooled adjacency matrix; the elements in the cross-section coefficient matrix are the allocation ratios used to cluster bus nodes and branch nodes into cross-section nodes;

[0024] The cross-sectional clustering graph is generated based on the pooled node feature matrix and the pooled adjacency matrix.

[0025] In one embodiment, the pooling operation on the power grid point-line model diagram to obtain the core network diagram of the power system includes:

[0026] Based on the adjacency matrix and node feature matrix of the power grid point-line model diagram, the shrinkage priority score corresponding to each edge in the power grid point-line model diagram is obtained through linear transformation;

[0027] According to the shrinkage priority score corresponding to each edge in the power grid point-line model diagram, each edge in the power grid point-line model diagram is shrunk to obtain the shrunk power grid point-line model diagram, which serves as the core network diagram.

[0028] In one embodiment, the fusion of the core network diagram and the cross-sectional clustering diagram to obtain the fused diagram features of the power system includes:

[0029] Based on the bus identifier of the bus object set and the branch identifier of the branch object set, record the first association relationship between the bus node in the core network diagram and the bus node in the cross-sectional three-part heterogeneous diagram, and record the second association relationship between the branch edge in the core network diagram and the branch node in the cross-sectional three-part heterogeneous diagram.

[0030] Based on the first and second association relationships, the core network diagram and the cross-sectional cluster diagram are weighted and fused to obtain the fused diagram features of the power system.

[0031] Secondly, this application provides a power system cross-section control device, the device comprising:

[0032] The power grid point-line model diagram construction module is used to construct a power grid point-line model diagram based on the set of bus objects and the set of branch objects in the power system; the power grid point-line model diagram is used to represent the relationship between bus objects and branch objects in the power system;

[0033] The cross-sectional three-part heterogeneous graph construction module is used to obtain the set of cross-sectional objects of the power system, and construct a cross-sectional three-part heterogeneous graph based on the set of bus objects, the set of branch objects, and the set of cross-sectional objects; the cross-sectional three-part heterogeneous graph is used to characterize the relationship between the bus objects and cross-sectional objects, as well as the relationship between the branch objects and cross-sectional objects in the power system.

[0034] The dual-graph pooling and fusion module is used to perform pooling operations on the power grid point-line model graph to obtain the core network diagram of the power system; to perform pooling operations on the cross-sectional three-part heterogeneous graph to obtain the cross-sectional clustering graph of the power system; and to fuse the core network diagram and the cross-sectional clustering graph to obtain the fused graph features of the power system.

[0035] The control action output module is used to input the fused graph features into a pre-trained power system cross-section control model to obtain power system cross-section control actions; the power system cross-section control actions are used to control the power system cross-section, and the pre-trained power system cross-section control model is a model obtained through reinforcement learning training.

[0036] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described power system section control method.

[0037] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0038] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0039] The aforementioned power system cross-section control method, device, computer equipment, readable storage medium, and computer program product construct a power grid point-line model diagram that accurately represents the physical connection relationships in the power system based on the set of bus objects and branch objects. Furthermore, based on the set of bus objects, branch objects, and cross-section objects, cross-sections are treated as independent nodes to construct a "bus-branch-cross-section" three-part heterogeneous cross-section diagram. This maps dynamically changing cross-section information to the same graph structure, achieving adaptive processing of the dynamically changing number of power grid cross-sections, thereby improving the model's generalization ability and practicality. By performing pooling operations on the power grid point-line model diagram and the three-part heterogeneous cross-section diagram, power grid cross-section control methods are obtained. The core network diagram and cross-sectional cluster diagram of the power system are fused together to obtain the fused diagram features of the power system. This process considers the node and edge features of the two diagrams, simplifies the topology of the power grid model, and achieves deep focusing and accurate fusion of the power grid point-line model diagram and the cross-sectional three-part heterogeneous diagram, enhancing the ability to extract key operating state features of the power grid. Then, the fused diagram features are input into a pre-trained power system cross-sectional control model to obtain power system cross-sectional control actions for controlling the power system cross-sections. This achieves high-precision optimized control of the power system cross-sections and solves the problem of low accuracy of power system cross-sectional control strategies in traditional power system control technologies. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating a power system cross-section control method in one embodiment;

[0042] Figure 2 This is a schematic diagram of hierarchical pooling of a cross-sectional trigonometric graph in one embodiment;

[0043] Figure 3 This is a schematic diagram of hierarchical pooling of a power grid point-line model diagram in one embodiment;

[0044] Figure 4 This is a flowchart illustrating a power system cross-section control method in another embodiment;

[0045] Figure 5 This is a structural block diagram of a power system cross-section control device in one embodiment;

[0046] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0048] In the actual operation of power systems, the number of active cross-sections requiring monitoring and control is dynamically changing due to the commissioning and decommissioning of equipment such as generating units, busbars, lines, and transformers. Traditional methods typically employ algorithmic models with fixed structures, such as linear neural networks, whose number of neurons in the input layer cannot be changed once set during training. This means that the trained algorithmic model can only adapt to the fixed number of cross-sections used in the training data. In practical applications, if the number of cross-sections decreases, it can be barely handled by padding; however, if the number of cross-sections increases, it exceeds the input dimension of the algorithmic model, which cannot handle it. This necessitates re-collecting data, adjusting the algorithmic model structure, and retraining, which is time-consuming, labor-intensive, and difficult to apply in real time. Therefore, the rigid structure of linear neural network methods, unable to adaptively handle dynamic changes in the number of cross-sections, severely restricts their practicality and generalization ability in real-world, variable power scenarios.

[0049] Furthermore, traditional technologies face structural mismatch challenges when integrating power grid point-line model diagrams and cross-sectional diagrams. In the power grid point-line model diagram, equipment such as generating units, buses, and loads are nodes, while branches such as lines and windings are edges; however, in the power grid three-part diagram, generating units, branches, and cross-sections are all treated as nodes. This means that elements representing the same physical entity (such as a single line) in both diagrams are "edges" in the power grid point-line model diagram but "nodes" in the three-part diagram. Traditional average pooling or fixed-weighted pooling methods cannot effectively handle this complex correspondence between nodes and edges. They mechanically fuse information without establishing focused, dynamic connections based on the actual physical meaning of nodes / edges in the diagram (such as equipment type, ID association) and real-time operating status (such as power flow, coefficient weights). Therefore, this simple pooling method cannot accurately integrate information, affecting the accuracy of subsequent optimization decisions.

[0050] To address the aforementioned issues, this application provides a power system cross-section control method. The method aims to construct a three-part heterogeneous graph of power system cross-section information, mapping multiple cross-section information onto a single graph and processing it using a graph neural network. This addresses the problem that linear neural networks, with their fixed input layer neurons, cannot adaptively handle changes in the number of effective cross-sections. Furthermore, the three-part heterogeneous graph representing cross-section information and the power grid point-line model graph representing the power grid point-line model are combined using graph-level pooling after considering node and edge feature attributes. This allows for the fusion of the two graphs and better capture of cross-section and power grid point-line model information.

[0051] Before describing the specific implementation methods of this application, the relevant terms in this application will be explained first:

[0052] Power System Section: Also known as Power Flow Section, or simply Section, it refers to the set of electrical components (such as generators, transmission lines, transformers, reactors, etc.) with clear safety relationships (such as the same transmission channel, the same zone boundary, the same critical power supply path) that are monitored, calculated, and controlled in real time at a specific moment to ensure the safety and stability of the power grid and to constrain the power transmission boundaries between regions. The total power flow (active / reactive power) or individual / overall power limits of these components are monitored.

[0053] Grid point-line model: That is, the grid bus-branch model, which is a basic mathematical model used to represent the network topology and electrical connection relationship. This model abstracts the physical power grid into a topological network composed of "points" and "lines". Among them, "points" represent bus nodes, which are electrical nodes for power generation, convergence, and distribution. Their states are described by output power, load, voltage amplitude, and phase angle; "lines" represent branches, which are transmission components connecting buses (such as lines, transformers), and their characteristics are described by impedance parameters such as resistance and reactance. By simplifying details such as switches inside substations, this model focuses on the overall electrical connection of the network and provides an essential underlying structure that reflects the essence of the physical connection of the power grid for subsequent integration with effective section information.

[0054] In an exemplary embodiment, as Figure 1 shown, a power system section control method is provided. In this embodiment, taking the application of this method to a terminal as an example, it can be understood that this method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, this method includes the following steps S102 to step S108. Among them:

[0055] Step S102, construct a grid point-line model diagram according to the set of bus objects and the set of branch objects of the power system.

[0056] Among them, the grid point-line model diagram can be used to represent the relationship between bus objects and branch objects in the power system.

[0057] Exemplarily, a bus can represent a combination of a power generation unit that can independently complete energy conversion (for example, convert energy such as fuel, water power, wind energy, and light energy into electric energy) and its directly supporting equipment, and is connected to the power grid for power operation and dispatching. Its specific composition is not limited in this application and may include, but is not limited to: various equipment (units) required for power generation such as generators, steam turbines, boilers, water turbines, photovoltaic panels, and wind turbines, as well as a combination of equipment composed of grid buses, load equipment, and other electrical topology nodes. It can be understood that the set of bus objects can refer to the bus entities (also known as bus objects) included when abstracting the power system into a graph structure and one or more of their attributes, such as bus type, voltage, active power output, reactive power output, active load, reactive load, etc.

[0058] A branch can represent an electrical energy transmission path or channel connecting to the power grid in a power system. Its specific components are not limited in this application and may include, but are not limited to, transmission / distribution lines, transformers, reactors, resistors, etc. Understandably, the set of branch objects can refer to the branch entities (also called branch objects) and one or more attributes they possess when abstracting the power system as a graph structure, such as branch type, impedance (resistance, reactance), admittance, active power at the beginning and end, reactive power at the beginning and end, etc.

[0059] In practical applications, the terminal can define the bus objects in the power grid as bus nodes in the diagram. Each bus node It can include bus node characteristics, such as voltage, active and reactive power, etc., and multiple bus objects constitute a bus object collection. , representing graph structure The set of nodes; define the branch objects in the power grid as connection nodes in the graph. and another node side road Each branch edge can contain edge features, such as the first and last nodes, impedance, admittance, and active and reactive power at the first and last nodes. Multiple branch objects constitute a branch object set. , representing graph structure The set of edges allows us to abstract the power system into a graph structure representing a point-line model of the power grid. From this, the node feature matrix of the power grid point-line model diagram can be obtained. and adjacency matrix This leads to the construction of a point-line model of the power grid system. This reflects the physical connection relationships between busbar and branch objects in the power system, thus providing a basic network structure representation for subsequent fusion with the cross-sectional triangulation diagram. It is understandable that this is achieved by constructing the graph structure of the power grid point-line model. Furthermore, graph neural network algorithms can be used to analyze the graph structure of the power grid point-line model. Feature extraction is performed to facilitate subsequent pooling and fusion processing.

[0060] Step S104: Obtain the set of cross-sectional objects of the power system, and construct a three-part heterogeneous cross-sectional diagram based on the set of bus objects, the set of branch objects, and the set of cross-sectional objects.

[0061] Among them, the cross-sectional ternary heterogeneous diagram, also known as the ternary heterogeneous diagram, is used to characterize the relationship between bus objects and cross-sectional objects in a power system, as well as the relationship between branch objects and cross-sectional objects.

[0062] Understandably, based on the aforementioned explanation of "power system cross-section", the set of cross-section objects can refer to the cross-section entities (or cross-section objects, i.e., the set of electrical components in the power system that embody safety relationships) and one or more attributes included when the power system is abstracted into a graph structure.

[0063] In practical implementation, the terminal can be based on the set of bus objects in the power system. and branch object collection This yields a collection of cross-section objects. This allows cross-section objects to be introduced as independent nodes into the graph structure, and then buses, branches, and cross-sections are treated as nodes, representing the relationship between buses and cross-sections in the power system (bus-cross-section, U2S). And the relationship between branches and cross sections (Branch-Cross Section, B2S). As edges, an abstract graph structure is derived, which can intuitively model cross-sectional information of the power grid. Therefore, graph neural network algorithms suitable for heterogeneous graphs can be used to analyze the graph structure of the power grid point-line model graph. Feature extraction is performed to construct a cross-sectional trigonometric heterogeneous diagram of the power system.

[0064] Understandably, the characteristics of bus nodes and branch nodes in the cross-sectional heterogeneous diagram are consistent with the characteristics of the corresponding bus nodes and branch edges in the power grid point-line model diagram. The characteristics of cross-sectional nodes may include constraint information such as the upper and lower limits of the attributes possessed by the cross-sectional object. The edge characteristics between the bus and the cross-section or between the branch and the cross-section can be derived from the cross-sectional definition file, specifically manifested as the dynamic coefficients of the bus or branch in the cross-section, thus fully representing the composition structure of a cross-section.

[0065] Step S106: Perform pooling operation on the power grid point-line model diagram to obtain the core network diagram of the power system; perform pooling operation on the cross-sectional three-part heterogeneous diagram to obtain the cross-sectional cluster diagram of the power system; merge the core network diagram and the cross-sectional cluster diagram to obtain the fused diagram features of the power system.

[0066] In practical implementation, the terminal can perform pooling operations on the power grid point-line model graph to retain key nodes and edges in the graph structure, simplify the topology of the power grid point-line model, and obtain the core network diagram of the power system; perform pooling operations on the cross-sectional three-part heterogeneous graph to cluster bus nodes and branch nodes into cross-sectional nodes, and obtain the cross-sectional clustering graph of the power system; and fuse the core network diagram and the cross-sectional clustering graph to correlate the nodes and edges in the graph structures of the two graphs, and obtain the fused graph features of the power system.

[0067] Step S108: Input the fused graph features into the pre-trained power system section control model to obtain the power system section control actions.

[0068] The power system section control actions can be continuous or discrete action values ​​obtained from a pre-trained power system section control model, which can be used to control the power system section. These actions can include, but are not limited to, active / reactive power output from the bus. It is understood that if the power system section control model outputs continuous action values, these values ​​can be discretized to obtain discrete action values ​​that the power system can actually execute and control.

[0069] Among them, the pre-trained power system section control model can be a model trained through reinforcement learning, including a value network and a policy network.

[0070] In one possible example, the terminal can input the fused graph features into the value network and policy network of a pre-trained power system cross-section control model to obtain power system cross-section control actions.

[0071] The aforementioned power system cross-section control method constructs a grid point-line model diagram that accurately represents the physical connections within the power system based on the bus object set and branch object set. Then, based on the bus object set, branch object set, and cross-section object set, it constructs a "bus-branch-cross-section" three-part heterogeneous cross-section diagram, treating cross-sections as independent nodes. This maps dynamically changing cross-section information to the same graph structure, achieving adaptive handling of the dynamically changing number of grid cross-sections, thereby improving the model's generalization ability and practicality. By performing pooling operations on the grid point-line model diagram and the three-part heterogeneous cross-section diagram, the core network structure diagram and cross-section diagram of the power system are obtained respectively. Clustering graphs are generated, and the core network diagram and cross-sectional clustering graphs are fused to obtain the fused graph features of the power system. This takes into account the node and edge features of the two graphs, simplifies the topology of the power grid model, and achieves deep focusing and accurate fusion of the power grid point-line model graph and the cross-sectional three-part heterogeneous graph, enhancing the ability to extract key operating state features of the power grid. Then, the fused graph features are input into a pre-trained power system cross-sectional control model to obtain power system cross-sectional control actions for controlling the power system cross-sections. This achieves high-precision optimized control of the power system cross-sections and solves the problem of low accuracy of power system cross-sectional control strategies in traditional power system control technologies.

[0072] In an exemplary embodiment, a set of cross-sectional objects of the power system is obtained, and a three-part heterogeneous cross-section diagram is constructed based on the set of bus objects, the set of branch objects, and the set of cross-sectional objects. This includes: obtaining the set of cross-sectional objects based on the set of bus objects and the set of branch objects; determining first association information and second association information based on the set of bus objects, the set of branch objects, and the set of cross-sectional objects; and constructing a three-part heterogeneous cross-section diagram based on the set of bus objects, the set of branch objects, the set of cross-sectional objects, the first association information, and the second association information.

[0073] The first association information is used to record the association between the busbar object and the section object (busbar-section, U2S); the second association information is used to record the association between the branch object and the section object (branch-section, B2S).

[0074] It can be understood that the cross-sectional three-part heterogeneous graph includes the bus nodes corresponding to each bus object in the bus object set, the branch nodes corresponding to each branch object in the branch object set, the cross-sectional nodes corresponding to each cross-sectional object in the cross-sectional object set, as well as the first type of edge and the second type of edge; wherein, the first type of edge (bus-cross-section, U2S) is used to represent the association relationship between a bus object and a cross-sectional object, and can also represent the connection relationship between a bus object and a cross-sectional object; the second type of edge (branch-cross-section, B2S) is used to represent the association relationship between a branch object and a cross-sectional object, and can also represent the connection relationship between a branch object and a cross-sectional object.

[0075] In practical implementation, the terminal can be based on the set of bus objects in the power system. and branch object collection As those skilled in the art will understand, power grid operation mode data (including the connection relationships of the power grid topology model, equipment parameters, etc.) are also needed to obtain a set of cross-sectional objects. Alternatively, it can be understood that the set of cross-section objects can be obtained from a file predefined by the scheduling personnel; and based on the set of bus objects. Branch object set and cross-section object collection This yields the first association information used to record the association between the busbar object and the section object. And second association information used to record the association between branch objects and cross-section objects. ; and then the collection of bus objects The set of bus nodes and branch objects corresponding to each bus object in the middle The set of branch nodes and cross-section objects corresponding to each branch object in the middle. The cross-section nodes corresponding to each cross-section object are used as nodes in the graph structure, and the first type of edges used to represent the association between a busbar object and a cross-section object are used. And a second type of edge used to characterize the association between a branch object and a section object. As edges in the graph structure, we obtain the graph structure of the cross-sectional ternary heterogeneous graph of the power system. Thus, a three-part heterogeneous diagram of the cross section is constructed.

[0076] The technical solution of this embodiment introduces a set of cross-section objects and determines, based on the set of bus objects, the set of branch objects, and the set of cross-section objects, the first association relationship information representing the connection relationship between bus objects and cross-section objects and the second association relationship information representing the connection relationship between branch objects and cross-section objects. Then, based on the set of bus objects, the set of branch objects, the set of cross-section objects, the set of cross-section objects, the first association relationship information, and the second association relationship information, the cross-section, bus, and branch are used as node types in the graph structure, and the edge types of branch-cross-section and bus-cross-section are used to represent the association relationship between nodes, thereby constructing a three-part heterogeneous cross-section graph of the power system.

[0077] Therefore, in the cross-sectional heterogeneous graph, cross sections can be introduced as independent nodes into the power grid topology model (graph structure). By mapping multiple potentially dynamically changing cross-sectional information to the same graph structure, power grid cross-sectional information can be intuitively modeled within the graph structure, thereby achieving structured modeling of multiple cross-sectional constraint information. Furthermore, when the number of cross sections changes (i.e., when the power system adds or removes cross sections), it can be reflected in the cross-sectional object set. By adding or removing corresponding cross-section nodes and adjusting their edges with busbar and branch nodes, the complexity of traditional methods requiring overall model reconstruction can be effectively avoided.

[0078] After constructing the graph structure of the "bus-branch-section" heterogeneous cross-section graph, a suitable graph neural network (GNN) algorithm is needed to process it to extract node representation embeddings for subsequent graph pooling. For example, Relational Graph Convolutional Network (R-GCN) or Heterogeneous Graph Attention Network (HAN) can be used to process the heterogeneous cross-section graph; this application does not specifically limit the application to this method. The following uses R-GCN as an example to illustrate the processing of the heterogeneous cross-section graph using a graph neural network algorithm.

[0079] Therefore, in one possible embodiment, after constructing the cross-sectional ternary heterogeneous graph based on the set of bus objects, the set of branch objects, the set of cross-sectional objects, the first association information, and the second association information, the method further includes: obtaining the current relationship type of the current node in the cross-sectional ternary heterogeneous graph, and the convolution weights corresponding to the current relationship type in the network layer above the current network layer where the current node is located; for the current node, obtaining the neighbor node embeddings corresponding to the current node's neighbor nodes in the network layer above the current relationship type, and the normalization coefficients of the current node for the current relationship type; updating the current node embedding corresponding to the current node in the network layer above the current relationship type based on the convolution weights, neighbor node embeddings, and normalization coefficients, to obtain the current node embedding corresponding to the current node in the current network layer under the current relationship type; obtaining the node feature matrix and the adjacency matrix of the cross-sectional ternary heterogeneous graph based on the current node embeddings corresponding to each current node in the current network layer under each current relationship type; and obtaining the node embedding matrix of the cross-sectional ternary heterogeneous graph based on the node feature matrix and the adjacency matrix.

[0080] The current node can be any one of the bus nodes, branch nodes, and cross-section nodes; the current relationship type can be any one of the relationship types corresponding to the first type of edge (bus-cross-section, U2S) and the relationship types corresponding to the second type of edge (branch-cross-section, B2S).

[0081] In its implementation, R-GCN introduces independent transformation matrices for different types of edge relationships. Therefore, in the cross-sectional heterogeneous graph of this application embodiment, different convolution weights are used for message passing between network layers on the node representation embedding (also known as node embedding) for the two types of edges: bus-section (U2S, i.e., first type edge) and branch-section (B2S, i.e., second type edge). Based on this, the update rule for node embedding can be formalized as follows:

[0082]

[0083] In equation (1), Indicates the current relation type (e.g., U2S for type 1 edge or B2S for type 2 edge). Indicates the current relation type Next current node Each neighboring node A set; It is the normalization coefficient of the current node for the current relation type, used to avoid the aggregation result "exploding" due to too many neighboring nodes of higher-order nodes. This is a relation-specific parameter matrix, i.e., the current relation type. The corresponding convolution weights; The number of layers in the graph neural network (in this embodiment, a relational graph convolutional network); This represents the activation function, which is not specifically defined here. Then, For the current relation type Next current node In the current network layer The current node embedding corresponding to the layer, For the current relation type Next current node The next network layer The current node embedding corresponding to the layer, For the current relation type Next current node neighboring nodes The next network layer The corresponding neighbor nodes in the layer are embedded.

[0084] Therefore, the update rule shown in equation (1) can be used to update the convolution weights. Neighbor node embedding and normalization coefficient Embed the current node in the previous network layer for the current node under the current relation type. Perform an update to obtain the current node embedding in the current network layer for the current relation type. Based on equation (1), the current relation type can be obtained. Next current node The embedding of the current node in the last layer of the graph neural network is then used. Furthermore, based on the embedding of the current node in the current network layer for each current relation type, the node feature matrix of the cross-sectional trigonometric graph can be obtained. and adjacency matrix Therefore, based on the nodal feature matrix of the cross-sectional trigonometric heterogeneous diagram... and adjacency matrix The node embedding matrix of the cross-sectional ternary heterogeneous graph is obtained as follows: ,in In the graph structure of the ternary heterogeneous diagram in the cross section Based on this, a cross-sectional trigonometric heterogeneous graph for subsequent pooling fusion was constructed. .

[0085] The technical solution of this embodiment, taking a relational graph convolutional network as an example, demonstrates the processing method of a three-dimensional heterogeneous cross-section graph using a graph neural network algorithm suitable for heterogeneous graphs. Multiple cross-section information is mapped to the same graph structure and processed by a graph neural network. The node embeddings of each node in the three-dimensional heterogeneous cross-section graph are extracted, thereby fully characterizing the different action modes of buses and branches on the cross-section. This allows the graph neural network to directly process different numbers of cross-section inputs, achieving adaptive processing for dynamically changing cross-section numbers. The constructed three-dimensional heterogeneous cross-section graph thus possesses good adaptive capabilities and engineering practicality, thereby improving the generalization ability and practicality of the power grid model in real, ever-changing, and complex power scenarios.

[0086] In an exemplary embodiment, a pooling operation is performed on a cross-sectional ternary heterogeneous graph to obtain a cross-sectional clustering graph of the power system. This includes: obtaining the adjacency matrix and node feature matrix of the cross-sectional ternary heterogeneous graph; obtaining a node embedding matrix based on the adjacency matrix and node feature matrix; updating the node embedding matrix according to a pre-constructed cross-sectional coefficient matrix to generate a pooled node feature matrix; updating the adjacency matrix according to the cross-sectional coefficient matrix to generate a pooled adjacency matrix; and generating a cross-sectional clustering graph based on the pooled node feature matrix and the pooled adjacency matrix.

[0087] The elements in the cross-section coefficient matrix can be the allocation ratio used to cluster bus nodes and branch nodes into cross-section nodes. This ratio comes from the cross-section information definition file and can represent information such as which equipment is included in a cross-section and the coefficient of each equipment.

[0088] In practical applications, the terminal can obtain the node feature matrix of the cross-sectional trigonometric heterogeneous graph. and adjacency matrix Alternatively, the node feature matrix of the cross-sectional trigonometric heterogeneous graph can be obtained based on the aforementioned graph neural network algorithms (such as R-GCN, HAN, etc.). and adjacency matrix Then it can be based on the adjacency matrix and node feature matrix The node embedding matrix is ​​learned through graph neural network algorithms (such as R-GCN, HAN, etc.): .

[0089] Pre-set cross-sectional coefficient matrix middle, is the number of nodes in the cross-sectional ternary heterogeneity graph. The number of clusters in the cross-section. elements in Represents device node To the cross-section node The coefficient can be used as the allocation ratio for clustering each device node (including bus nodes and branch nodes) into section nodes. When performing pooling operations on the section-based heterogeneous graph, the allocation process (i.e., the clustering process) considers the section composition relationship. If a device node (including bus nodes and branch nodes) belongs to only one section, the allocation ratio of the device node is fully allocated to that section cluster; if a device node belongs to multiple sections, the coefficient of the device node in its respective section is used as the allocation ratio to distribute it to the corresponding section clusters, and edges are added between the corresponding section clusters to represent the association relationship.

[0090] Furthermore, based on the cross-sectional coefficient matrix and node embedding matrix ,pass Generate the feature matrix of the pooled nodes. According to the cross-sectional coefficient matrix For adjacency matrix To update, i.e. Generate the adjacency matrix after pooling. Based on the feature matrix of the nodes after pooling Adjacency matrix after pooling Generate a cross-sectional clustering diagram. For example... Figure 2 The diagram shown is a hierarchical pooling diagram of a three-part heterogeneous cross-section. Based on this pooling operation, the generated pooled cross-section sub-graph is the cross-section clustering graph. Nodes represent cross-sections, and edges represent shared equipment between cross-sections, i.e., shared buses and branches.

[0091] The technical solution of this embodiment obtains a node embedding matrix based on the adjacency matrix and the node feature matrix, thereby considering node feature attributes in graph-level pooling to cluster equipment nodes (buses, branches) into cross-section nodes; furthermore, it generates a pooled node feature matrix based on a pre-constructed cross-section coefficient matrix and the node embedding matrix; the adjacency matrix is ​​updated according to the cross-section coefficient matrix to generate a pooled adjacency matrix, and the cross-section clustering graph generated based on the pooled node feature matrix and the pooled adjacency matrix can accurately abstract information at the cross-section level of the power system, thereby more effectively capturing power grid cross-section information.

[0092] In an exemplary embodiment, a pooling operation is performed on the power grid point-line model diagram to obtain the core network diagram of the power system. This includes: obtaining the shrinkage priority score corresponding to each edge in the power grid point-line model diagram through linear transformation based on the adjacency matrix and node feature matrix of the power grid point-line model diagram; and shrinking each edge in the power grid point-line model diagram according to the shrinkage priority score corresponding to each edge in the power grid point-line model diagram to obtain the shrunken power grid point-line model diagram, which serves as the core network diagram.

[0093] The shrinkage priority score (shrinkage priority score) can be used to represent the priority of shrinking each edge in the power grid point-line model diagram. The higher the shrinkage priority score of an edge, the higher the shrinkage priority of that edge, which usually indicates that the edge is a non-critical edge, such as: non-overloaded branches, non-section component equipment, etc.

[0094] In a specific implementation, the power grid point-line model diagram constructed above is as follows: The terminal can be based on the power grid point-line model diagram. adjacency matrix and node feature matrix By performing a linear transformation as shown in equation (2), the connecting nodes in the power grid point-line model diagram are obtained. and another node side road Corresponding contraction priority score :

[0095]

[0096] In equation (2), and For nodes and nodes The feature vectors of nodes include electrical quantities and bus attributes such as whether the node contains a generator. Representing an edge eigenvectors, This indicates a splicing operation. and These are learnable parameters.

[0097] For example, it can be done through the following partial Method, for shrinkage priority score Normalization and biasing are performed to obtain the normalized shrinkage priority score. :

[0098]

[0099] In equation (3), The normalized bias hyperparameters are used to obtain the normalized shrinkage priority score. It can enhance gradient stability.

[0100] Furthermore, based on the normalized contraction priority scores corresponding to each edge in the power grid point-line model diagram... The edges in the power grid point-line model diagram are shrunk. For example, the edge with the highest shrinkage priority score can be selected as the edge to be shrunk and iteratively shrunk. During the shrinking process, the edge to be shrunk will be... Corresponding connected nodes and nodes To merge new nodes .

[0101] Therefore, it is possible to base the contraction on the edge to be contracted. The new node after the merger Based on the normalized shrinkage priority score of the edge to be shrunk. For the node feature matrix in the power grid point-line model diagram Update the matrix to generate the pooled node feature matrix. .

[0102] For example, it can be done through The method is based on the normalized shrinkage priority score of the edge to be shrunk. The nodes connected to the edge to be shrunk eigenvectors and nodes eigenvectors Perform a weighted summation to obtain the new node. Node feature vectors Then, based on the node feature vectors of each new node... The feature matrix of the nodes after pooling is obtained. .

[0103] Simultaneously, based on the connection relationships corresponding to each new node, the adjacency matrix in the power grid point-line model graph can be updated by updating the edges between each new node. Update the adjacency matrix to generate the pooled adjacency matrix. Then, based on the feature matrix of the nodes after pooling... Adjacency matrix after pooling The shrunken grid point-line model diagram is obtained and used as the core grid diagram.

[0104] like Figure 3 The diagram shown illustrates hierarchical pooling of the power grid point-line model. Based on this pooling operation, the generated core network diagram can reduce the size of the power grid point-line model to a predetermined ratio (e.g., 220kV network, 500kV network, etc.), thus achieving hierarchical compression. This is similar to the network equivalence method in power systems, such as in power dispatching, where low-voltage networks (e.g., 110kV) are equated to high-voltage networks (e.g., 220kV and above) to highlight the main network structure. This edge contraction process provides learnable local pooling, ensuring that the pooled graph retains connectivity and key topological properties.

[0105] The technical solution of this embodiment analyzes the adjacency matrix and node feature matrix of the power grid point-line model graph to obtain the shrinkage priority score corresponding to each edge in the power grid point-line model graph, providing a reliable quantitative basis for the edge shrinkage strategy of hierarchical pooling of the power grid point-line model graph; then, according to the shrinkage priority score corresponding to each edge, each edge in the power grid point-line model graph is shrunk to remove redundant information in the graph. The resulting core network diagram can retain the key nodes and branches of the power grid point-line model graph, thereby reducing the number of nodes and simplifying the topology of the power grid model.

[0106] In an exemplary embodiment, the core network diagram and the cross-sectional cluster diagram are fused to obtain the fused diagram features of the power system. This includes: recording the first matching relationship between bus nodes in the core network diagram and bus nodes in the cross-sectional three-dimensional heterogeneous diagram based on the bus identifiers of the bus object set and the branch identifiers of the branch object set; and recording the second matching relationship between branch edges in the core network diagram and branch nodes in the cross-sectional three-dimensional heterogeneous diagram; and performing a weighted fusion of the core network diagram and the cross-sectional cluster diagram based on the first and second matching relationships to obtain the fused diagram features of the power system.

[0107] In practical applications, the terminal can record the first matching relationship between bus nodes in the core network diagram and bus nodes in the cross-sectional three-part heterogeneous diagram based on the bus identifiers of the bus object set and the branch identifiers of the branch object set, so as to match the corresponding bus based on the unique bus identifier. It can also record the second matching relationship between branch edges in the core network diagram and branch nodes in the cross-sectional three-part heterogeneous diagram, so as to match the corresponding branch based on the unique branch identifier. This allows for ID alignment association between the pooled grid point-line model diagram (i.e., the core network diagram) and the device nodes (bus nodes, branch nodes) in the cross-sectional three-part heterogeneous diagram, resulting in an association alignment diagram that matches corresponding nodes and edges based on unique identifiers. Furthermore, based on the first and second matching relationships, the association alignment diagram and the pooled cross-sectional sub-graph (i.e., the cross-sectional clustering diagram) are weighted and fused. The weights of each fused connection edge in the two graph fusion can be referenced from the elements in the cross-sectional coefficient matrix. Finally, the result can be obtained through... The operation is flattened into a fixed-dimensional feature vector, thereby obtaining the fused graph features of the power system used as input to the pre-trained power system cross-sectional control model.

[0108] The technical solution of this embodiment, by using the unique bus identifier of the bus object and the unique branch identifier of the branch object in the power system, can align and associate the bus nodes in the core network diagram with the bus nodes in the cross-sectional three-dimensional heterogeneous diagram, as well as the branch edges in the core network diagram with the branch nodes in the cross-sectional three-dimensional heterogeneous diagram, thereby effectively realizing the accurate association and alignment of the same physical entity in the two diagrams. Furthermore, based on the bus matching relationship and branch matching relationship of ID alignment, the core network diagram and the cross-sectional clustering diagram are weighted and fused, thereby simplifying the information loss problem in traditional pooling fusion. The resulting fused diagram features of the power system can focus on the key coupling relationships in the power system, thereby improving the representation quality of bus features, branch features, and cross-sectional features in the power system. This provides accurate fused information as input for subsequent output of power system control actions through the pre-trained power system control strategy network.

[0109] like Figure 4 As shown, this application provides a specific flowchart of a power system control method based on a power system cross-sectional model and a point-line model, including:

[0110] Input power grid model, power grid operation data and cross-sectional data: Power grid operation data including buses and branches can be obtained from the power system, including the entities and attributes of bus objects and the entities and attributes of branch objects, to obtain the set of bus objects and the set of branch objects, and then obtain the set of cross-sectional objects of the power system based on the set of bus objects and the set of branch objects, including the entities and attributes of cross-sectional objects.

[0111] Constructing a power grid point-line model diagram: Based on the set of bus objects and branch objects of the power system, define the bus objects in the power grid as bus nodes in the diagram. Define the branch objects in the power grid as connection nodes in the graph. and another node side road This allows the power system to be abstracted into a graph structure of a grid point-line model. Furthermore, feature extraction can be performed on the power grid point-line model diagram using graph neural networks for subsequent fusion processing, thereby obtaining the node feature matrix of the power grid point-line model diagram. and adjacency matrix This leads to the construction of a point-line model of the power grid system. .

[0112] Constructing a three-part heterogeneous diagram of cross-sections: Based on the set of bus objects, the set of branch objects, and the set of cross-section objects, the bus, branch, and cross-section are used as nodes, and the relationship between the bus and the cross-section in the power system (bus-cross-section, U2S) is represented. And the relationship between branches and cross sections (Branch-Cross Section, B2S). As edges, an abstract graph structure is derived, which can intuitively model cross-sectional information of the power grid. Simultaneously, deep learning methods (graph neural networks) suitable for heterogeneous graphs can be used to extract features from the cross-sectional ternary heterogeneous graph, obtaining the node feature matrix of the cross-sectional ternary heterogeneous graph. and adjacency matrix This allows for the construction of a cross-sectional tripartite heterogeneous diagram of the power system. .

[0113] Hierarchical pooling is performed on the power grid wire model diagram to extract the core network structure diagram: based on the power grid point and line model diagram adjacency matrix and node feature matrix The contraction priority score corresponding to each edge in the power grid point-line model diagram is obtained through linear transformation. According to the contraction priority score corresponding to each side in the power grid point-line model diagram. The edges of the power grid point-line model diagram are shrunk to obtain the shrunk power grid point-line model diagram, which serves as the core network diagram.

[0114] Hierarchical pooling is performed on the cross-sectional ternary heterogeneous graph to obtain the cross-sectional clustering graph: based on the node feature matrix of the cross-sectional ternary heterogeneous graph. and adjacency matrix The node embedding matrix is ​​obtained through a graph neural network algorithm. And embed the nodes into the matrix Convert to section coefficient matrix According to the cross-sectional coefficient matrix and node embedding matrix Generate the feature matrix of the pooled nodes. According to the cross-sectional coefficient matrix For adjacency matrix Update the adjacency matrix to generate the pooled adjacency matrix. Based on the feature matrix of the nodes after pooling Adjacency matrix after pooling Generate a cross-sectional clustering diagram.

[0115] The core network diagram is associated with the cross-sectional clustering graph's buses and branches: Based on the bus identifier of the bus object set and the branch identifier of the branch object set, the corresponding nodes and edges in the core network diagram and the cross-sectional three-dimensional heterogeneous graph are matched based on the unique identifier. The buses in the core network diagram are associated and aligned with the buses in the cross-sectional three-dimensional heterogeneous graph (first matching relationship), and the branches in the core network diagram are associated and aligned with the branches in the cross-sectional three-dimensional heterogeneous graph (second matching relationship), resulting in an associated alignment graph based on the unique identifier matching the corresponding nodes and edges.

[0116] The core network diagram after association alignment is fused with the cross-sectional cluster diagram: Based on the first association relationship and the second association relationship, the association alignment diagram obtained after alignment is fused with the pooled cross-sectional sub-diagram (i.e., cross-sectional cluster diagram) through weighted fusion to obtain the fused diagram features of the power system.

[0117] The fused features are input into the value network and policy network: The fused graph features are input into the value network and policy network in the pre-trained power system cross-section control model to obtain the power system cross-section control actions used to control the power system cross-section.

[0118] It should be noted that the specific limitations of the above steps can be found in the above description of the specific limitations of a power system control method based on a power system cross-sectional model and a point-line model.

[0119] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0120] Based on the same inventive concept, this application also provides a power system cross-section control device for implementing the power system cross-section control method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more power system cross-section control device embodiments provided below can be found in the limitations of the power system cross-section control method described above, and will not be repeated here.

[0121] In one exemplary embodiment, such as Figure 5 As shown, a power system section control device is provided, including:

[0122] The power grid point-line model diagram construction module 510 is used to construct a power grid point-line model diagram based on the set of bus objects and the set of branch objects in the power system; the power grid point-line model diagram is used to represent the relationship between bus objects and branch objects in the power system.

[0123] The cross-sectional three-part heterogeneous graph construction module 520 is used to obtain the set of cross-sectional objects of the power system, and construct the cross-sectional three-part heterogeneous graph based on the set of bus objects, the set of branch objects, and the set of cross-sectional objects. The cross-sectional three-part heterogeneous graph is used to characterize the relationship between bus objects and cross-sectional objects, as well as the relationship between branch objects and cross-sectional objects in the power system.

[0124] The dual-graph pooling and fusion module 530 is used to perform pooling operations on the power grid point-line model graph to obtain the core network diagram of the power system; to perform pooling operations on the cross-sectional three-part heterogeneous graph to obtain the cross-sectional clustering graph of the power system; and to fuse the core network diagram and the cross-sectional clustering graph to obtain the fused graph features of the power system.

[0125] The control action output module 540 is used to input the fused graph features into the pre-trained power system section control model to obtain the power system section control action; the power system section control action is used to control the power system section, and the pre-trained power system section control model is a model obtained through reinforcement learning.

[0126] In one embodiment, the cross-section three-part heterogeneous graph construction module 520 is specifically used to obtain a cross-section object set based on a bus object set and a branch object set; determine first association information and second association information based on the bus object set, branch object set, and cross-section object set; the first association information is used to record the association relationship between bus objects and cross-section objects; the second association information is used to record the association relationship between branch objects and cross-section objects; and construct a cross-section three-part heterogeneous graph based on the bus object set, branch object set, cross-section object set, first association information, and second association information; wherein, the cross-section three-part heterogeneous graph includes bus nodes corresponding to each bus object in the bus object set, branch nodes corresponding to each branch object in the branch object set, cross-section nodes corresponding to each cross-section object in the cross-section object set, and first type edges and second type edges; the first type edge is used to characterize the association relationship between a bus object and a cross-section object; and the second type edge is used to characterize the association relationship between a branch object and a cross-section object.

[0127] In one embodiment, the apparatus further includes a cross-sectional graph processing module 550, which is further configured to obtain the current relation type of the current node in the cross-sectional triadic graph, and the convolution weights corresponding to the current relation type in the network layer above the current network layer where the current node is located; the current node is any one of the bus nodes, branch nodes, and cross-sectional nodes; the current relation type is any one of the relation types corresponding to the first type of edge and the second type of edge; for the current node, the embedding of the neighbor nodes corresponding to the current node in the network layer above the current relation type is obtained, and when The preceding node has a normalized coefficient for the current relation type. Based on the convolution weights, neighbor node embeddings, and normalized coefficients, the current node embedding corresponding to the current node in the previous network layer under the current relation type is updated to obtain the current node embedding corresponding to the current node in the current network layer under the current relation type. Based on the current node embedding corresponding to the current node in the current network layer under each current relation type, the node feature matrix of the cross-sectional ternary heterogeneous graph and the adjacency matrix of the cross-sectional ternary heterogeneous graph are obtained. And based on the node feature matrix and adjacency matrix of the cross-sectional ternary heterogeneous graph, the node embedding matrix of the cross-sectional ternary heterogeneous graph is obtained.

[0128] In one embodiment, the dual-graph pooling and fusion module 530 is further configured to obtain the adjacency matrix and node feature matrix of the cross-sectional ternary heterogeneous graph, obtain the node embedding matrix based on the adjacency matrix and node feature matrix; update the node embedding matrix according to a pre-set cross-sectional coefficient matrix to generate the pooled node feature matrix; update the adjacency matrix according to the cross-sectional coefficient matrix to generate the pooled adjacency matrix; the elements in the cross-sectional coefficient matrix are the allocation ratios used to cluster bus nodes and branch nodes into cross-sectional nodes; and generate a cross-sectional clustering graph based on the pooled node feature matrix and the pooled adjacency matrix.

[0129] In one embodiment, the dual-graph pooling and fusion module 530 is further configured to obtain the shrinkage priority score corresponding to each edge in the power grid point-line model graph through linear transformation based on the adjacency matrix and node feature matrix of the power grid point-line model graph; and to perform shrinkage processing on each edge in the power grid point-line model graph according to the shrinkage priority score corresponding to each edge in the power grid point-line model graph to obtain the shrunken power grid point-line model graph as the core network diagram.

[0130] In one embodiment, the dual-graph pooling and fusion module 530 is further configured to record, based on the bus identifier of the bus object set and the branch identifier of the branch object set, a first matching relationship between the bus node in the core grid diagram and the bus node in the cross-sectional three-dimensional heterogeneous diagram, and a second matching relationship between the branch edge in the core grid diagram and the branch node in the cross-sectional three-dimensional heterogeneous diagram; and to perform weighted fusion of the core grid diagram and the cross-sectional clustering diagram based on the first matching relationship and the second matching relationship to obtain the fused graph features of the power system.

[0131] Each module in the aforementioned power system section control device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0132] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data required for power system section control methods. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a power system section control method.

[0133] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0134] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the various embodiments of the power system cross-section control method described above.

[0135] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps described in the embodiments of the power system cross-section control method.

[0136] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps described in various embodiments of a computational power system cross-section control method.

[0137] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic resistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence processors, etc., and are not limited to these.

[0138] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0139] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for cross-sectional control of a power system, characterized in that, The method includes: Based on the set of bus objects and the set of branch objects in the power system, a power grid point-line model diagram is constructed; the power grid point-line model diagram is used to represent the relationship between bus objects and branch objects in the power system. Obtain the set of cross-sectional objects of the power system, and construct a cross-sectional three-part heterogeneous diagram based on the set of bus objects, the set of branch objects, and the set of cross-sectional objects; the cross-sectional three-part heterogeneous diagram is used to characterize the relationship between the bus objects and cross-sectional objects, and the relationship between the branch objects and cross-sectional objects in the power system. A pooling operation is performed on the power grid point-line model diagram to obtain the core network diagram of the power system; a pooling operation is performed on the cross-sectional three-part heterogeneous diagram to obtain the cross-sectional cluster diagram of the power system; the core network diagram and the cross-sectional cluster diagram are fused to obtain the fused diagram features of the power system. The fused graph features are input into a pre-trained power system cross-section control model to obtain power system cross-section control actions; the power system cross-section control actions are used to control the power system cross-section, and the pre-trained power system cross-section control model is a model obtained through reinforcement learning training.

2. The method according to claim 1, characterized in that, The step of obtaining the set of cross-sectional objects of the power system, and constructing a cross-sectional three-part heterogeneous diagram based on the set of bus objects, the set of branch objects, and the set of cross-sectional objects includes: The cross-section object set is obtained based on the busbar object set and the branch object set; Based on the busbar object set, the branch object set, and the cross-section object set, first association information and second association information are determined; the first association information is used to record the association relationship between the busbar object and the cross-section object; the second association information is used to record the association relationship between the branch object and the cross-section object. Based on the bus object set, the branch object set, the cross-section object set, the first association information, and the second association information, the cross-section three-part heterogeneous diagram is constructed; The cross-sectional triadic graph includes bus nodes corresponding to each bus object in the bus object set, branch nodes corresponding to each branch object in the branch object set, cross-sectional nodes corresponding to each cross-sectional object in the cross-sectional object set, as well as first-type edges and second-type edges. The first-type edges are used to characterize the association between a bus object and a cross-sectional object; the second-type edges are used to characterize the association between a branch object and a cross-sectional object.

3. The method according to claim 2, characterized in that, After the step of constructing the cross-sectional tri-heterogeneous diagram based on the bus object set, the branch object set, the cross-section object set, the first association information, and the second association information, the method further includes: Obtain the current relation type of the current node in the cross-sectional heterogeneous graph, and the convolution weights corresponding to the current relation type in the network layer above the current network layer where the current node is located; the current node is any one of the bus nodes, branch nodes, and cross-sectional nodes; the current relation type is any one of the relation types corresponding to the first type edge and the relation types corresponding to the second type edge; For the current node, obtain the neighbor node embeddings of the current node's neighbor nodes in the previous network layer under the current relationship type, and the normalization coefficient of the current node for the current relationship type; Based on the convolution weights, the neighbor node embeddings, and the normalization coefficients, the current node embedding corresponding to the current node in the previous network layer under the current relation type is updated to obtain the current node embedding corresponding to the current node in the current network layer under the current relation type. Based on the current node embedding of each current node in each current network layer under each current relation type, the node feature matrix of the cross-sectional ternary heterogeneous graph and the adjacency matrix of the cross-sectional ternary heterogeneous graph are obtained; and based on the node feature matrix and the adjacency matrix of the cross-sectional ternary heterogeneous graph, the node embedding matrix of the cross-sectional ternary heterogeneous graph is obtained.

4. The method according to claim 1, characterized in that, The step of performing a pooling operation on the cross-sectional triadic graph to obtain the cross-sectional clustering graph of the power system includes: Obtain the adjacency matrix and node feature matrix of the cross-sectional ternary heterogeneous graph, and obtain the node embedding matrix based on the adjacency matrix and the node feature matrix; The node embedding matrix is ​​updated according to the pre-set cross-section coefficient matrix to generate a pooled node feature matrix; the adjacency matrix is ​​updated according to the cross-section coefficient matrix to generate a pooled adjacency matrix; the elements in the cross-section coefficient matrix are the allocation ratios used to cluster bus nodes and branch nodes into cross-section nodes; The cross-sectional clustering graph is generated based on the pooled node feature matrix and the pooled adjacency matrix.

5. The method according to claim 1, characterized in that, The process of performing a pooling operation on the power grid point-line model diagram to obtain the core network structure diagram of the power system includes: Based on the adjacency matrix and node feature matrix of the power grid point-line model diagram, the shrinkage priority score corresponding to each edge in the power grid point-line model diagram is obtained through linear transformation; According to the shrinkage priority score corresponding to each edge in the power grid point-line model diagram, each edge in the power grid point-line model diagram is shrunk to obtain the shrunk power grid point-line model diagram, which serves as the core network diagram.

6. The method according to claim 1, characterized in that, The process of fusing the core network diagram and the cross-sectional clustering diagram to obtain the fused diagram features of the power system includes: Based on the bus identifier of the bus object set and the branch identifier of the branch object set, record the first matching relationship between the bus node in the core network diagram and the bus node in the cross-sectional three-dimensional heterogeneous diagram, and record the second matching relationship between the branch edge in the core network diagram and the branch node in the cross-sectional three-dimensional heterogeneous diagram. Based on the first matching relationship and the second matching relationship, the core network diagram and the cross-sectional cluster diagram are weighted and fused to obtain the fused diagram features of the power system.

7. A power system cross-sectional control device, characterized in that, The device includes: The power grid point-line model diagram construction module is used to construct a power grid point-line model diagram based on the set of bus objects and the set of branch objects in the power system; the power grid point-line model diagram is used to represent the relationship between bus objects and branch objects in the power system; The cross-sectional three-part heterogeneous graph construction module is used to obtain the set of cross-sectional objects of the power system, and construct the cross-sectional three-part heterogeneous graph based on the set of bus objects, the set of branch objects, and the set of cross-sectional objects; the cross-sectional three-part heterogeneous graph is used to characterize the relationship between the bus objects and cross-sectional objects, and the relationship between the branch objects and cross-sectional objects in the power system. The dual-graph pooling and fusion module is used to perform pooling operations on the power grid point-line model graph to obtain the core network diagram of the power system; to perform pooling operations on the cross-sectional three-part heterogeneous graph to obtain the cross-sectional clustering graph of the power system; and to fuse the core network diagram and the cross-sectional clustering graph to obtain the fused graph features of the power system. The control action output module is used to input the fused graph features into a pre-trained power system cross-section control model to obtain power system cross-section control actions; the power system cross-section control actions are used to control the power system cross-section, and the pre-trained power system cross-section control model is a model obtained through reinforcement learning training.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images