New equipment starting scheduling method and system based on graph neural network and knowledge graph

Abstract

The invention discloses a new equipment starting scheduling method and system based on a graph neural network and a knowledge graph, and relates to the technical field of power scheduling, and the method comprises the steps: constructing a power grid equipment knowledge graph according to power grid physical equipment and related knowledge elements; generating starting scheme power grid model data based on the power grid equipment knowledge graph; converting the generated starting scheme power grid model data into graph vector data by adopting a graph embedding method, and determining a historical typical starting scheme which is most matched with the new equipment starting scheme by adopting a graph similarity method; and on the basis of the historical typical starting scheme, a node alignment method is adopted to search for a device node pair, corresponding to the historical matching device and the new device, of the historical matching device, the new device node is replaced with the historical matching device node based on the device node pair, and a new device starting scheme is obtained. The method can effectively simulate the complex topological structure of the power grid and the mutual dependence relation between elements, accurately measures the operation characteristics of the power grid, and achieves the intelligent compilation of a new equipment starting scheme to reduce the burden of power grid operation personnel.

Description

Technical Field

[0001] The present invention relates to the field of power dispatching technology, and in particular to a new equipment startup scheduling method and system based on graph neural network and knowledge graph. Background Art

[0002] With the rapid development of new power systems, the pace of grid expansion and construction has accelerated in recent years, resulting in a large number of startup and dispatch implementation plans for new equipment being compiled by provincial and local grid operators. However, compilation of startup plans is complex and requires a high level of personal knowledge and experience, placing significant pressure on grid operators. The traditional method of manually compiling startup and dispatch implementation plans for new equipment is characterized by low efficiency, high pressure, and the risk of errors.

[0003] Manually developing a plan requires multiple steps, including equipment parameter collection, grid topology analysis, startup procedure design, and safety verification. Each step relies on manual operation, resulting in a lengthy process. Furthermore, startup plans for similar equipment often share similar frameworks, yet require repeated manual parameter adjustments, creating a significant amount of repetitive work. Furthermore, repeated communication and confirmation with multiple departments, including dispatching, operations, and maintenance, reduces communication efficiency. Inefficient information transfer can lead to repeated plan revisions, extending the development cycle.

[0004] Because startup plans involve the safe and stable operation of the power grid, plan developers bear significant responsibility and pressure. With the rapid advancement of power grid technology, manual planning methods are no longer adaptable to the startup requirements of new equipment. Human negligence, resulting in parameter entry errors, missed steps, or oversight of safety checks, can lead to equipment damage, power grid accidents, and even casualties. Furthermore, inconsistent data versions during multi-departmental collaboration can easily lead to plan errors. Manual planning relies on individual experience, resulting in inconsistent format and content, making it difficult to ensure quality. In response, the introduction of intelligent tools, the establishment of standardized processes, and the promotion of technological innovation can significantly improve planning efficiency and quality, reduce the pressure on operators, and ensure the safe and stable operation of the power grid. In the future, as the level of power grid intelligence increases, startup plan planning will also need to gradually evolve towards automation and intelligence. Leveraging next-generation artificial intelligence technologies to support the intelligent dispatching of new power systems will enhance the ability to ensure safe power grid operation. Summary of the Invention

[0005] The purpose of the present invention is to address the problems in the above-mentioned prior art and provide a new equipment startup scheduling method and system based on graph neural networks and knowledge graphs, which can effectively simulate the complex topology structure of the power grid and the interdependence between components, accurately measure the operating characteristics of the power grid, and realize the intelligent compilation of new equipment startup plans to reduce the burden on power grid operators.

[0006] In order to achieve the above object, the present invention has the following technical solutions:

[0007] First, a new device startup scheduling method based on graph neural network and knowledge graph is provided, including:

[0008] Construct a knowledge graph of power grid equipment based on physical power grid equipment and related knowledge elements;

[0009] Generate startup plan grid model data based on the grid equipment knowledge graph;

[0010] The generated startup plan grid model data is converted into graph vector data using a graph embedding method, and the historical typical startup plan that best matches the new equipment startup plan is determined using a graph similarity method.

[0011] A node alignment method is used on historical typical startup plans to find device node pairs corresponding to historical matching devices and new devices. Based on the device node pairs, the new device nodes are replaced with historical matching device nodes to obtain a new device startup plan.

[0012] As a preferred solution, in the step of constructing a knowledge graph of power grid equipment based on power grid physical equipment and related knowledge elements, the related knowledge elements include the association relationship of power grid physical equipment, and the power grid physical equipment and the association relationship of power grid physical equipment are stored in a Neo4j graph database. By generating corresponding Cphyer statements and executing Cphyer statements, the related knowledge elements are written into the Neo4j graph database to form a knowledge graph of power grid equipment.

[0013] As a preferred solution, the steps of generating startup plan grid model data based on the grid equipment knowledge graph include:

[0014] Retrieve the name of the starting substation from the starting plan, and retrieve the grid physical equipment and equipment attributes of the grid physical equipment related to the starting substation from the grid equipment knowledge graph;

[0015] The retrieved grid physical device collection is arranged in a list form and then encapsulated in JSON data format to form the startup plan grid model data.

[0016] As a preferred solution, a graph embedding method is used to convert the generated startup plan grid model data into graph vector data, and a graph similarity method is used to determine the historical typical startup plan that best matches the new equipment startup plan. The following steps are included:

[0017] Parsing the structures of nodes, edges, attributes, and graphs in each startup scenario power grid model data, and loading the structures of nodes, edges, attributes, and graphs in each startup scenario power grid model data into a graph object;

[0018] Initialize the vector representations of nodes and edges for the point and edge features in the graph object of the start-up scenario power grid model data for each case. The calculation expression is as follows:

[0019]

[0020] e ij = MLP edge (x ij )

[0021] In the formula, represents the node initialization vector; x i represents the node feature; e ij represents the edge vector; x ij represents the edge feature; MLP node represents the multi-layer perceptron for generating the node vector; MLP edge represents the multi-layer perceptron for generating the edge vector;

[0022] Update the features based on the surrounding node information, edge information, and the interaction information between graphs according to the following formula:

[0023]

[0024] In the formula, m j-＞i represents the message information transmitted from the neighbor node j to the current node i; represents the vector of node i in the t-th iteration; u j-＞i represents the cross information of the two graphs; f message represents the multi-layer perceptron based on the concatenated input; a j-＞i represents the attention weight; f node represents the multi-layer perceptron for updating the node representation;

[0025] Generate the corresponding graph vector data based on all the node vectors of each case's power grid model;

[0026]

[0027] In the formula, h G represents the whole graph vector; f G represents the aggregation function for generating the whole graph vector;

[0028] Measure the similarity between two graph vectors by the cosine similarity method;

[0029]

[0030] In the formula, h G1 ·h G2 is the dot product of the vectors h G1 and h G2 ; |h G1 | and |hG2 | are vectors h G1 and h G2 The model;

[0031] By calculating the similarity between the grid model graph vectors of the new equipment startup plan and multiple historical typical startup plans, the historical typical startup plan with the highest similarity is selected as the matching template.

[0032] As a preferred solution, the node alignment method is used to find the device node pair corresponding to the historical matching device and the new device based on the historical typical startup plan, and the new device node is replaced with the historical matching device node based on the device node pair. In the step of obtaining the new device startup plan, the node alignment method is used to find the node pair h' with similar structure or characteristics in the two power grid model diagrams. n With h n , where h' n represents the power grid model equipment node with the startup substation as the core, h n Represents the power grid model device node with the historical matching substation as the core; based on the node vector generated by iterative calculation, node alignment is achieved by calculating the cosine similarity, and the aligned node h' n The device name replaces the node h in the historical typical startup scenario n Generate a new device startup plan by typing the device name.

[0033] Secondly, a new device startup scheduling system based on graph neural networks and knowledge graphs is provided, including:

[0034] A knowledge graph construction module is used to construct a knowledge graph of power grid equipment based on physical power grid equipment and related knowledge elements;

[0035] A power grid model data generation module is used to generate startup plan power grid model data based on the power grid equipment knowledge graph;

[0036] The historical typical startup plan matching module is used to convert the generated startup plan grid model data into graph vector data using a graph embedding method, and determine the historical typical startup plan that best matches the new equipment startup plan using a graph similarity method;

[0037] The node alignment module is used to use the node alignment method to find the device node pair corresponding to the historical matching device and the new device based on the historical typical startup plan, replace the new device node with the historical matching device node based on the device node pair, and obtain the new device startup plan.

[0038] As a preferred solution, when the knowledge graph construction module constructs the power grid equipment knowledge graph based on the power grid physical equipment and related knowledge elements, the related knowledge elements include the association relationship of the power grid physical equipment, and the power grid physical equipment and the association relationship of the power grid physical equipment are stored in the Neo4j graph database. By generating corresponding Cphyer statements and executing Cphyer statements, the related knowledge elements are written into the Neo4j graph database to form the power grid equipment knowledge graph.

[0039] As a preferred solution, the power grid model data generation module retrieves the name of the starting substation from the starting plan, and retrieves the power grid physical equipment and equipment attributes of the power grid physical equipment related to the starting substation from the power grid equipment knowledge graph;

[0040] The retrieved grid physical device collection is arranged in a list form and then encapsulated in JSON data format to form the startup plan grid model data.

[0041] As a preferred solution, the historical typical startup plan matching module parses the structure of nodes, edges, attributes and graphs in the power grid model data of each startup plan, and loads the structure of nodes, edges, attributes and graphs in the power grid model data of each startup plan as a graph object;

[0042] Initialize the vector representation of nodes and edges for the point and edge features in the graph object of each startup scenario power grid model data. The calculation expression is as follows:

[0043]

[0044] e ij =MLP edge (x ij )

[0045] Where, represents the node initialization vector; x i Represents node features; e ij represents the edge vector; x ij Represents edge features; MLP node Represents a multilayer perceptron that generates node vectors; MLP edge represents a multilayer perceptron that generates edge vectors;

[0046] Update the features based on the surrounding node information, edge information, and interaction information between graphs as follows:

[0047]

[0048] Where m j-＞i Represents the message information transmitted from neighbor node j to current node i; Denote the vector of node i at the t-th round of iteration; u j-＞i Denote the cross-information of two graphs; f message Represent a multi-layer perceptron based on concatenated input; a j-＞i Represent the attention weights; f node Represent a multi-layer perceptron for updating node representations;

[0049] Generate corresponding graph vector data based on the vector of all nodes in each power grid model;

[0050]

[0051] In the formula, h[[ID=u18]] G Denote the whole graph vector; f G Denote the aggregation function for generating the whole graph vector;

[0052] Measure the similarity between two graph vectors by the cosine similarity method;

[0053]

[0054] In the formula, h G1 ·h G2 Is the dot product of vectors h G1 And h G2 ; |h G1 | And |h G2 | Are the norms of vectors h G1 And h G2 respectively;

[0055] By calculating the similarity of the graph vectors of the power grid model between the new equipment startup plan and multiple historical typical startup plans, select the historical typical startup plan with the highest similarity as the matching template.

[0056] As a preferred solution, the node alignment module finds node pairs h' n And h n With similar structures or features in the power grid model graph through the node alignment method, h' n Denote the equipment nodes of the power grid model with the startup substation as the core, h n Denote the equipment nodes of the power grid model with the historical matching substation as the core; Based on the node vectors generated by iterative calculation, node alignment is achieved by calculating the cosine similarity, and replace the equipment name of node h n In the historical typical startup plan with the equipment name of the aligned node h' n To generate a new equipment startup plan.

[0057] In a third aspect, an electronic device is provided, comprising a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement the new device startup scheduling method based on graph neural network and knowledge graph as described in the first aspect.

[0058] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the new device startup scheduling method based on graph neural network and knowledge graph as described in the first aspect is implemented.

[0059] Compared with the prior art, the first aspect of the present invention has at least the following beneficial effects:

[0060] Graph neural networks are a deep learning technology for processing graph data. They can effectively simulate the complex topology of power grids and the interdependencies between components, thereby more accurately measuring power grid operational characteristics. To enable intelligent compilation of new equipment startup plans and reduce the burden on power grid operators, this paper proposes a new equipment startup scheduling method based on graph neural networks and knowledge graphs. This method ensures efficient generation and reliable steps for new equipment startup plans, addressing the low efficiency, high pressure, and error-prone nature of manual compilation of new equipment startup schedules by power grid operators. Graph similarity techniques typically calculate the similarity between two graphs based on graph vectors, lacking interactive information between the graphs. To this end, this paper employs an attention mechanism to match similar nodes in the graph. A node alignment method is then used to identify device node pairs corresponding to historically matched devices and new devices based on historically matched device startup plans. Based on these device node pairs, the new device nodes are replaced with historically matched device nodes to obtain new equipment startup plans. This method improves the expressive power of graph embedding technology and the effectiveness of graph similarity matching, providing near-optimal results for matching historically matched startup plans. This method can be widely applied to power grid operation and scheduling at the provincial and local levels, addressing the increasing types and number of grid scheduling objects and the diversification of power grid operations, providing a technical foundation for intelligent compilation of startup plans.

[0061] It can be understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0063] Figure 1Flowchart of a new device startup scheduling method based on graph neural networks and knowledge graphs in an embodiment of the present invention;

[0064] Figure 2 Schematic diagram of the relationship between the graph representation model and the graph matching model architecture according to an embodiment of the present invention;

[0065] Figure 3 In the embodiment of the present invention, a schematic diagram of a new equipment startup plan is obtained by taking the Fuwan substation in Anhui Province as an example. DETAILED DESCRIPTION

[0066] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0067] To achieve intelligent compilation of new equipment startup plans and reduce the burden on power grid operators, an embodiment of the present invention designs a new equipment startup scheduling method based on graph neural networks and knowledge graphs. This method includes the construction of a provincial power grid equipment knowledge graph, the generation of startup plan grid model data based on the knowledge graph, and grid model topology matching based on graph similarity calculation. The present invention uses a knowledge graph to replace the prior knowledge of planners to store historical typical startup plans; a graph embedding method is used to convert the startup plans containing knowledge into graph vector data, and a graph similarity method is used to determine the matching typical startup plans; and a joint graph neural network technology generates new equipment startup plans through node alignment, thereby ensuring efficient plan generation and reliable steps.

[0068] See also Figure 1 The embodiment of the present invention provides a new device startup scheduling method based on a graph neural network and a knowledge graph, including:

[0069] S1. Construct a knowledge graph of power grid equipment based on physical power grid equipment and related knowledge elements;

[0070] S2. Generate startup plan grid model data based on the grid equipment knowledge graph;

[0071] S3. Convert the generated startup plan grid model data into graph vector data using a graph embedding method, and determine the historical typical startup plan that best matches the new equipment startup plan using a graph similarity method;

[0072] S4. Using a node alignment method on a historical typical startup plan, find a device node pair corresponding to a historical matching device and a new device. Based on the device node pair, replace the new device node with the historical matching device node to obtain a new device startup plan.

[0073] A knowledge graph is a structured semantic knowledge base used to describe concepts and their relationships in the physical world in symbolic form. The basic building block of a knowledge graph is the "entity-relationship-entity" triple. Entities are interconnected through relationships, forming a network-like knowledge structure. Neo4j, as a high-performance graph database, is specifically designed for storing and querying graph-structured data. Designed based on graph theory, it intuitively represents data and its relationships through nodes, relationships, and properties, allowing structured data to be stored on the network. Accordingly, the present invention uses the Cypher tool to construct and store provincial-level power grid equipment information in the Neo4j graph database. Cypher is Neo4j's dedicated query language for efficiently querying and manipulating graph data. It describes data schemas using a declarative syntax (similar to SQL, which allows users to query, insert, update, delete data, and manage database structures) and is particularly well suited for processing complex relationships between nodes, relationships, and properties.

[0074] In one possible implementation, the relevant knowledge elements of step S1 include the association relationships of power grid physical equipment. The power grid physical equipment and the association relationships of the power grid physical equipment are stored in a Neo4j graph database. Based on the knowledge elements such as provincial power grid physical equipment and their association relationships stored in a relational database in the scheduling platform, corresponding Cphyer statements are generated. By executing the Cphyer statements, the relevant knowledge elements are written into the Neo4j graph database to form a knowledge graph of power grid equipment.

[0075] In one possible implementation, step S2 specifies that the scope of site activation typically includes: activating the substation's newly built lines and the line switches on both sides; activating the substation's busbars, bus tie switches, sectionalizers, main transformers, and their associated equipment. Because the equipment involved in the activation plan covers a local area of ​​the provincial power grid, the provincial power grid knowledge graph can be used to extract all equipment involved in the activation plan and generate power grid model data to provide data support for step S3.

[0076] Step S2 generates startup plan grid model data based on the grid equipment knowledge graph, including:

[0077] 1) Retrieve the name of the starting substation from the startup plan, and retrieve the grid physical equipment and equipment attributes related to the starting substation from the grid equipment knowledge graph, as shown in Table 1;

[0078] 2) The retrieved grid physical device sets are arranged in a list and then encapsulated in JSON data format to form the startup plan grid model data, thus realizing the standardization of the startup plan grid model representation.

[0079] Table 1

[0080] Classification Specific content Entity Type Substations, feeders, switches, circuit breakers, busbars Entity attributes Name, ID, voltage level, rated capacity, equipment type, connected equipment ID, and station

[0081] In one possible implementation, see Figure 2 In order to realize the intelligent compilation of new equipment startup plans and reduce the burden on power grid operators, a graph embedding method can be used to convert the startup plans containing knowledge into graph vector data. The graph similarity method is used to determine the historical typical startup plan that best matches the new equipment startup plan. Step S3 includes the following steps:

[0082] 1) Parse the structure of nodes, edges, attributes, and graphs in the json file of each startup scenario power grid model data, and load the structure of nodes, edges, attributes, and graphs in each startup scenario power grid model data into a NetworkX graph object;

[0083] The NetworkX graph object is the core data structure for representing and processing graph structures in Python. NetworkX provides four main graph object types to meet different graph structure requirements:

[0084] Graph: An undirected graph, where only one edge is allowed between two nodes, and a node is allowed to have an edge pointing to itself (self-loop).

[0085] DiGraph: A directed graph, a subclass of Graph. The edges between nodes are directional.

[0086] MultiGraph: allows multiple undirected edges between two nodes.

[0087] MultiDiGraph: A directed graph version of MultiGraph that allows multiple directed edges between two nodes.

[0088] NetworkX graph objects can be created in a variety of ways, including by adding nodes and edges directly, by creating from an adjacency matrix, and by using graph generators.

[0089] 2) Initialize the vector representation of nodes and edges for the point and edge features in the graph object of each startup scenario power grid model data. The calculation expression is as follows:

[0090]

[0091] e ij =MLP edge (xij )

[0092] In the formula, represents the node initialization vector; x i represents the node feature; e ij represents the edge vector; x ij represents the edge feature; MLP[[ID=1十三条]] node represents the multi-layer perceptron for generating the node vector; MLP edge represents the multi-layer perceptron for generating the edge vector;

[0093] 3) Update the features based on the surrounding node information, edge information, and the interaction information between graphs according to the following formula:

[0094]

[0095] In the formula, m j-＞i represents the message information transmitted from the neighbor node j to the current node i; represents the vector of node i in the t-th iteration; u j->i represents the cross information of two graphs; f message represents the multi-layer perceptron based on the concatenated input; a j->i represents the attention weight; f node represents the multi-layer perceptron for updating the node representation;

[0096] 4) Generate the corresponding graph vector data based on all node vectors of each power grid model; <oo00341>

[0098] In the formula, h G represents the whole graph vector; f[[ID=4九条]] G represents the aggregation function for generating the whole graph vector;

[0099] 5) Measure the similarity between two graph vectors by the cosine similarity method;

[0100]

[0101] In the formula, h G1 ·h G2 is the dot product of the vectors h G1 and h G2 ; |h G1 | and |h G2 | are the norms of the vectors h G1 and h G2 respectively;

[0102] By calculating the similarity of the graph vectors of the power grid models between the new device startup plan and multiple historical typical startup plans, select the historical typical startup plan with the highest similarity as the matching template.

[0103] In a possible implementation, step S4, after the historical typical startup plan is determined, node pairs h' with similar structures or characteristics in the two power grid model graphs are searched one by one by using the node alignment method. n With h n , where h' n represents the power grid model equipment node with the startup substation as the core, h n Represents the power grid model device node with the historical matching substation as the core; Based on the node vector generated by the iterative calculation in step S3, node alignment is achieved by calculating the cosine similarity. Finally, the aligned node h' n The device name replaces the node h in the historical typical startup scenario n Generate a new device startup plan by typing the device name.

[0104] See also Figure 3 Taking the Fuwan Substation in Anhui Province as an example of a substation to be started, according to the new equipment startup scheduling method based on graph neural network and knowledge graph in the embodiment of the present invention, there are 12 historical substations such as Daihe Substation, Rulin Substation, Gaixia Substation, and Caofang Substation with historical typical startup scenarios. Among them, the Fuwan Substation grid model has a total of 38 nodes and 42 edges. The local grid model represented by the historical typical startup scenario is matched with the local grid model with Fuwan Substation as the core. The closer the evaluation result is to 0, the more similar the two graphs are. From the results in Table 2, it can be seen that the "Rulin Substation" has the highest similarity, which is -2.6e-12. Its number of nodes and edges are 37 and 42 respectively. The topology structure is very similar to that of Fuwan Substation, so Rulin Substation is selected as the matching historical substation. Table 3 shows the node alignment results of Fuwan Substation and Rulin Substation. The results show that the aligned nodes are very similar in type and content.

[0105] Table 2

[0106]

[0107]

[0108] Table 3

[0109]

[0110] To intelligently compile new equipment startup plans and reduce the burden on power grid operators, this paper designs a method based on graph neural networks and knowledge graphs. Graph similarity techniques typically calculate the similarity between two graphs based on graph vectors, lacking interactive information between the graphs. Therefore, this paper uses an attention mechanism to match similar nodes in two graphs, improving the expressive power of graph embedding techniques and the effectiveness of graph similarity matching, providing near-optimal results for historical plan matching.

[0111] Another embodiment of the present invention further proposes a new device startup scheduling system based on a graph neural network and a knowledge graph, including:

[0112] A knowledge graph construction module is used to construct a knowledge graph of power grid equipment based on physical power grid equipment and related knowledge elements;

[0113] A power grid model data generation module is used to generate power grid model data for startup plans based on the power grid equipment knowledge graph;

[0114] The historical typical startup plan matching module is used to convert the generated startup plan grid model data into graph vector data using a graph embedding method, and determine the historical typical startup plan that best matches the new equipment startup plan using a graph similarity method;

[0115] The node alignment module is used to use the node alignment method to find the device node pair corresponding to the historical matching device and the new device based on the historical typical startup plan, replace the new device node with the historical matching device node based on the device node pair, and obtain the new device startup plan.

[0116] In one possible implementation, when the knowledge graph construction module constructs a knowledge graph of power grid equipment based on power grid physical equipment and related knowledge elements, the related knowledge elements include the association relationship of power grid physical equipment, and the power grid physical equipment and the association relationship of power grid physical equipment are stored in a Neo4j graph database. By generating corresponding Cphyer statements and executing Cphyer statements, the related knowledge elements are written into the Neo4j graph database to form a knowledge graph of power grid equipment.

[0117] Furthermore, the power grid model data generation module retrieves the name of the starting substation from the starting plan, and retrieves the power grid physical equipment and equipment attributes of the power grid physical equipment related to the starting substation from the power grid equipment knowledge graph;

[0118] The retrieved grid physical device collection is arranged in a list form and then encapsulated in JSON data format to form the startup plan grid model data.

[0119] In one possible implementation, the historical typical startup plan matching module parses the structures of nodes, edges, attributes, and graphs in each startup plan power grid model data, and loads the structures of nodes, edges, attributes, and graphs in each startup plan power grid model data as a graph object;

[0120] Initialize the vector representation of nodes and edges for the point and edge features in the graph object of each startup scenario power grid model data. The calculation expression is as follows:

[0121]

[0122] e ij = MLP edge (x ij )

[0123] In the formula, represents the node initialization vector; x i represents the node feature; e ij represents the edge vector; x ij represents the edge feature; MLP node represents the multi-layer perceptron for generating the node vector; MLP edge represents the multi-layer perceptron for generating the edge vector;

[0124] Update the features based on the surrounding node information, edge information, and the interaction information between graphs according to the following formula:

[0125]

[0126] In the formula, m j-＞i represents the message information transmitted from the neighbor node j to the current node i; represents the vector of node i at the t-th iteration; u j-＞i represents the cross information of the two graphs; f message represents the multi-layer perceptron based on the concatenated input; a j-＞i represents the attention weight; f node represents the multi-layer perceptron for updating the node representation;

[0127] Generate the corresponding graph vector data based on all node vectors of each power grid model;

[0128]

[0129] In the formula, h G represents the whole graph vector; f G represents the aggregation function for generating the whole graph vector;

[0130] Measure the similarity between two graph vectors by the cosine similarity method;

[0131]

[0132] In the formula, h G1 · h G2 is the dot product of the vectors h G1 and h G2 ; |h G1 | and |h G2 | are the norms of the vectors h G1 and h G2 respectively;

[0133] By calculating the similarity between the grid model graph vectors of the new equipment startup plan and multiple historical typical startup plans, the historical typical startup plan with the highest similarity is selected as the matching template.

[0134] In a possible implementation, the node alignment module searches for node pairs h' with similar structures or features in the power grid model graph by using a node alignment method. n With h n , h' n represents the power grid model equipment node with the startup substation as the core, h n Represents the power grid model device node with the historical matching substation as the core; based on the node vector generated by iterative calculation, node alignment is achieved by calculating the cosine similarity, and the aligned node h' n The device name replaces the node h in the historical typical startup scenario n Generate a new device startup plan by typing the device name.

[0135] Another embodiment of the present invention proposes an electronic device, including a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement the new device startup scheduling method based on graph neural network and knowledge graph.

[0136] Another embodiment of the present invention proposes a computer-readable storage medium, which stores at least one instruction. When the at least one instruction is executed by a processor, it implements the new device startup scheduling method based on graph neural network and knowledge graph.

[0137] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program code. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals. For ease of explanation, the above content only shows the part related to the embodiment of the present invention. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium is non-transitory and can be stored in a storage device formed by various electronic devices, and can implement the execution process recorded in the method of the embodiment of the present invention.

[0138] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts 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, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0140] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0141] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A new device startup scheduling method based on graph neural network and knowledge graph, characterized in that: include: Construct a knowledge graph of power grid equipment based on physical power grid equipment and related knowledge elements; Generate startup plan grid model data based on the grid equipment knowledge graph; The generated startup plan grid model data is converted into graph vector data using a graph embedding method, and the historical typical startup plan that best matches the new equipment startup plan is determined using a graph similarity method. A node alignment method is used on historical typical startup plans to find device node pairs corresponding to historical matching devices and new devices. Based on the device node pairs, the new device nodes are replaced with historical matching device nodes to obtain a new device startup plan.

2. The new device startup scheduling method based on graph neural network and knowledge graph according to claim 1 is characterized in that: In the step of constructing a knowledge graph of power grid equipment based on power grid physical equipment and related knowledge elements, the related knowledge elements include the association relationships of power grid physical equipment, the power grid physical equipment and the association relationships of power grid physical equipment are stored in a Neo4j graph database, and the related knowledge elements are written into the Neo4j graph database by generating corresponding Cphyer statements and executing the Cphyer statements to form a knowledge graph of power grid equipment.

3. The new device startup scheduling method based on graph neural network and knowledge graph according to claim 1 is characterized in that: The step of generating startup plan grid model data based on the grid equipment knowledge graph includes: Retrieve the name of the starting substation from the starting plan, and retrieve the grid physical equipment and equipment attributes of the grid physical equipment related to the starting substation from the grid equipment knowledge graph; The retrieved grid physical device collection is arranged in a list form and then encapsulated in JSON data format to form the startup plan grid model data.

4. The new device startup scheduling method based on graph neural network and knowledge graph according to claim 1 is characterized in that: The method of converting the generated startup plan grid model data into graph vector data by using a graph embedding method and determining the historical typical startup plan that best matches the new equipment startup plan by using a graph similarity method comprises the following steps: Parsing the structures of nodes, edges, attributes, and graphs in each startup scenario power grid model data, and loading the structures of nodes, edges, attributes, and graphs in each startup scenario power grid model data into a graph object; Initialize the vector representation of nodes and edges for the point and edge features in the graph object of each startup scenario power grid model data. The calculation expression is as follows: e ij =MLP edge (x ij ) Where, represents the node initialization vector; x i Represents node features; e ij represents the edge vector; x ij Represents edge features; MLP node Represents a multilayer perceptron that generates node vectors; MLP edge represents a multilayer perceptron that generates edge vectors; Update the features based on the surrounding node information, edge information, and interaction information between graphs as follows: Where m j-＞i Represents the message information transmitted from neighbor node j to current node i; represents the vector of node i at the tth iteration; u j->i Indicates the cross information of two images; f message represents a multilayer perceptron based on concatenated input; a j->i represents the attention weight; f node represents a multilayer perceptron used to update node representations; Generate corresponding graph vector data based on all node vectors of each power grid model; Where h G represents the entire graph vector; f G represents the aggregation function that generates the entire graph vector; The similarity between two graph vectors is measured by the cosine similarity method; where h G1 ·h G2 is the dot product of vectors h G1 and h G2 ; |h G1 | and |h G2 | are the magnitudes of vectors h G1 and h G2 respectively; By calculating the similarity between the grid model graph vectors of the new equipment startup plan and multiple historical typical startup plans, the historical typical startup plan with the highest similarity is selected as the matching template.

5. The new device startup scheduling method based on graph neural network and knowledge graph according to claim 4 is characterized in that: In the step of using a node alignment method to find a device node pair corresponding to a historical matching device and a new device based on a historical typical startup plan, replacing the new device node with a historical matching device node based on the device node pair, and obtaining a new device startup plan, the node alignment method is used to find a node pair h' having similar structures or characteristics in two power grid model graphs. n With h n , where h' n represents the power grid model equipment node with the startup substation as the core, h n Represents the power grid model device node with the historical matching substation as the core; based on the node vector generated by iterative calculation, node alignment is achieved by calculating the cosine similarity, and the aligned node h' n The device name replaces the node h in the historical typical startup scenario n Generate a new device startup plan by typing the device name.

6. A new equipment startup scheduling system based on graph neural network and knowledge graph, characterized by: include: A knowledge graph construction module is used to construct a knowledge graph of power grid equipment based on physical power grid equipment and related knowledge elements; A power grid model data generation module is used to generate power grid model data for startup plans based on the power grid equipment knowledge graph; The historical typical startup plan matching module is used to convert the generated startup plan grid model data into graph vector data using a graph embedding method, and determine the historical typical startup plan that best matches the new equipment startup plan using a graph similarity method; The node alignment module is used to use the node alignment method to find the device node pair corresponding to the historical matching device and the new device based on the historical typical startup plan, replace the new device node with the historical matching device node based on the device node pair, and obtain the new device startup plan.

7. The new device startup scheduling system based on graph neural network and knowledge graph according to claim 6 is characterized in that: When the knowledge graph construction module constructs the power grid equipment knowledge graph based on the power grid physical equipment and related knowledge elements, the related knowledge elements include the association relationship of the power grid physical equipment, and the power grid physical equipment and the association relationship of the power grid physical equipment are stored in the Neo4j graph database. By generating corresponding Cphyer statements and executing Cphyer statements, the related knowledge elements are written into the Neo4j graph database to form the power grid equipment knowledge graph.

8. The new device startup scheduling system based on graph neural network and knowledge graph according to claim 6 is characterized in that: The grid model data generation module retrieves the name of the starting substation from the startup plan, and retrieves the grid physical equipment and equipment attributes of the grid physical equipment related to the starting substation from the grid equipment knowledge graph; The retrieved grid physical device collection is arranged in a list form and then encapsulated in JSON data format to form the startup plan grid model data.

9. The new device startup scheduling system based on graph neural network and knowledge graph according to claim 6 is characterized in that: The historical typical startup plan matching module parses the structure of nodes, edges, attributes and graphs in each startup plan power grid model data, and loads the structure of nodes, edges, attributes and graphs in each startup plan power grid model data as a graph object; Initialize the vector representation of nodes and edges for the point and edge features in the graph object of each startup scenario power grid model data. The calculation expression is as follows: e ij =MLP edge (x ij ) Where, represents the node initialization vector; x i Represents node features; e ij represents the edge vector; x ij Represents edge features; MLP node Represents a multilayer perceptron that generates node vectors; MLP edge represents a multilayer perceptron that generates edge vectors; Update the features based on the surrounding node information, edge information, and interaction information between graphs as follows: Where m j->i Represents the message information transmitted from neighbor node j to current node i; represents the vector of node i at the tth iteration; u j->i Indicates the cross information of two images; f message represents a multilayer perceptron based on concatenated input; a j->i represents the attention weight; f node represents a multilayer perceptron used to update node representations; Generate corresponding graph vector data based on all node vectors of each power grid model; Where h G represents the entire graph vector; f G represents the aggregation function that generates the entire graph vector; The similarity between two graph vectors is measured by the cosine similarity method; where h G1 ·h G2 is the dot product of the vectors h G1 and h G2 ; |h G1 | and |h G2 | are the magnitudes of the vectors h G1 and h G2 respectively; By calculating the similarity between the grid model graph vectors of the new equipment startup plan and multiple historical typical startup plans, the historical typical startup plan with the highest similarity is selected as the matching template.

10. The new device startup scheduling system based on graph neural network and knowledge graph according to claim 9 is characterized in that: The node alignment module uses the node alignment method to find node pairs h' with similar structures or features in the power grid model graph. n With h n , h' n represents the power grid model equipment node with the startup substation as the core, h n Represents the power grid model device node with the historical matching substation as the core; based on the node vector generated by iterative calculation, node alignment is achieved by calculating the cosine similarity, and the aligned node h' n The device name replaces the node h in the historical typical startup scenario n Generate a new device startup plan by typing the device name.

11. An electronic device, characterized in that: It includes a processor and a memory, and the processor is used to execute a computer program stored in the memory to implement a new device startup scheduling method based on a graph neural network and a knowledge graph as described in any one of claims 1 to 5.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by the processor, it implements the new device startup scheduling method based on graph neural network and knowledge graph as described in any one of claims 1 to 5.

Images