CN-121984989-A - Plug-and-play method and system for Internet of things equipment of power grid based on multi-mode disambiguation and knowledge graph embedding

Abstract

The invention discloses a plug-and-play method and a plug-and-play system for power grid Internet of things equipment based on multi-mode disambiguation and knowledge map embedding, wherein the method is used for analyzing ambiguous user instructions through a multi-mode module, and generating clear configuration intention by fusing context and sensor data; and constructing a knowledge graph containing equipment configuration and topology resources, embedding the knowledge graph into vectors, matching an optimal access scheme with an object model through machine learning, and checking a power grid operation rule on a matching result at the same time to ensure that the scheme can be directly executed in a landing mode. The method solves the problems of inconsistent matching of the object models, dependence on manual configuration migration, weak analysis of ambiguous instructions and the like, realizes automatic access of novel switches, photovoltaics and other equipment, and improves the automation level of the Internet of things of the power grid.

Inventors

• LIANG ZHEHENG
• LU HONGZHI
• HUANG XIAOQIANG
• XU MINGJIE
• ZHANG ZIYANG
• ZENG JIJUN
• ZHANG XIAOLU
• ZHAO SHANLONG
• WANG YECHAO
• LU CHANGCAI
• YAO CHAOSHENG
• ZHANG JINBO

Assignees

• 广东电网有限责任公司
• 广东电网有限责任公司信息中心

Dates

Publication Date

20260505

Application Date

20251226

Claims (12)

1. 1. The plug and play method for the Internet of things equipment of the power grid based on multi-mode disambiguation and knowledge graph embedding is characterized by comprising the following steps of: acquiring a user multi-modal instruction, synchronously acquiring context data and environment perception data, inputting the multi-modal instruction, the context data and the environment perception data into a pre-trained large language model, and generating at least two candidate configuration intents; The method comprises the steps of extracting configuration information of new equipment, constructing a knowledge graph comprising equipment attributes, topological relations and determined equipment configuration intentions, modeling the knowledge graph by adopting RDF triples, generating triples by means of Schema alignment and rule mapping for structured data, generating triples by means of entity alignment and relation extraction for unstructured data, analyzing resource dependency relations between the new equipment and an upstream transformer substation and a downstream feeder line when the new equipment is connected to a power grid based on the knowledge graph, and generating the triples by means of entity alignment and relation extraction; Processing the determined equipment configuration intention, the extracted new equipment configuration information and the topological resource relation by adopting a knowledge graph embedding algorithm, converting the equipment configuration intention, the extracted new equipment configuration information and the topological resource relation into an intention vector, an equipment configuration vector and a topological resource vector, splicing the three vectors into a fixed-dimension total feature vector in sequence, and inputting the fixed-dimension total feature vector into a machine learning model; And when similar power grid Internet of things equipment is accessed, the matched historical data is called from the access knowledge base to directly generate the optimal configuration scheme.
2. 2. The method of claim 1, wherein the context data comprises GIS location information, line load data and topology state data, the environment awareness data comprises RFID identification data of a new device, the pre-trained large language model is a GPT-4o model, the pre-trained large language model is deployed on a cloud server, the cloud server interacts with a context storage module and a user preference storage module in real time through an API, the context storage module is used for storing user confirmation intention and corresponding multi-mode features, and the user preference storage module is used for storing historical operation and preference for intention generation.
3. 3. The method of claim 1, wherein when generating candidate collocation intents, the large language model is guided to generate the candidate collocation intents from three dimensions of physical location, load balance and operation and maintenance safety through the structured prompt, and the stored interaction data comprises user instruction text, equipment identity characteristics, power grid context characteristics and user confirmation results, and the interaction data is used for fine tuning the large language model regularly.
4. 4. The method of claim 1, wherein the Schema alignment comprises predefining entity types, core attributes and relationship types of the power distribution network knowledge graph, wherein the entity alignment comprises extracting entities in unstructured text by using a named entity recognition model, performing entity disambiguation by context semantic matching, and linking the disambiguated entities to an entity library of the knowledge graph, and the relationship extraction comprises extracting relationships among the entities from the unstructured text by using regular extraction or model extraction.
5. 5. The method of claim 1, wherein the knowledge-graph embedding algorithm is a RotatE algorithm, wherein the RotatE algorithm takes a triplet of knowledge-graphs as input, and converts entities and relationships into low-dimensional vectors through a complex vector rotation operation, wherein the complex vectors include real and imaginary parts, and the real and imaginary parts encode core semantics and uncertainty of the entities or relationships, respectively.
6. 6. The method of claim 5, wherein the training process of RotatE algorithm comprises initializing complex vectors of entities and relationships, sampling positive triples and negative triples in batches, wherein the positive triples correspond to real effective knowledge in a power distribution network and cover device attributes, resource dependencies and intention association core relationships, the negative triples are generated through rule constraint and random replacement and have semantic conflict with the positive triples, calculating L2 distance scores of the triples, calculating loss values by adopting marginal loss functions, back-propagating update vector parameters through an Adam optimizer, and iteratively training until the loss values are converged.
7. 7. The method of claim 1, wherein the machine learning model calculates the similarity between the intent vector and each feeder topology-intent association vector through cosine similarity to determine the target access point, then takes the total feature vector as input, generates probability distribution of the executable configuration parameters through the double hidden layer containing the ReLU activation function and the output layer of the Softmax activation function, calculates the protection setting value through mean square error, cross entropy and binary cross entropy, calculates the loss of the sampling frequency and protocol compatibility, and weights and sums the loss according to preset weights to obtain the total loss, and optimizes the model parameters based on the total loss.
8. 8. The method according to claim 1, wherein the object model consistency check result is obtained by calculating the euclidean distance between a new equipment protocol vector and a power distribution network object model vector, comparing the euclidean distance with a preset threshold value, and judging the compatibility of the new equipment and the power distribution network object model.
9. 9. The method of claim 1, wherein the grid internet of things device comprises a novel intelligent switch, a metering device, a distributed photovoltaic inverter, and a distributed energy access apparatus.
10. 10. The utility model provides a network internet of things equipment plug and play system based on multimode disambiguation and knowledge graph embedding which is characterized in that the system comprises: The multi-mode disambiguation and intention generation module is used for acquiring a multi-mode instruction of a user, synchronously acquiring context data and environment perception data, inputting the multi-mode instruction, the context data and the environment perception data into a pre-trained large language model, and generating at least two candidate collocation intentions; The system comprises a configuration constraint extraction and resource dependence analysis module, a resource dependence analysis module and a power grid analysis module, wherein the configuration constraint extraction and resource dependence analysis module is used for extracting configuration information of new equipment and constructing a knowledge graph comprising equipment attributes, topological relations and determined equipment configuration intentions; The device comprises an intention-configuration-object model mapping and embedding module, a machine learning model, a target access point, an executable configuration parameter and an object model consistency check result, wherein the intention-configuration-object model mapping and embedding module is used for processing the determined device configuration intention, the extracted new device configuration information and the topology resource relation by adopting a knowledge graph embedding algorithm, converting the device configuration intention, the extracted new device configuration information and the topology resource relation into an intention vector, a device configuration vector and a topology resource vector, splicing the three vectors into a fixed-dimension total feature vector in sequence, and inputting the fixed-dimension total feature vector into the machine learning model; The data storage module is used for storing the generated optimal configuration scheme, the corresponding multi-mode instruction, the knowledge graph data and the corresponding feature vector in an associated mode to construct an access knowledge base, and when the similar power grid Internet of things equipment is accessed, the matched historical data is called from the access knowledge base to directly generate the optimal configuration scheme.
11. 11. An electronic device comprising one or more processors, memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, which when executed by the processor implement the multi-modal disambiguation and knowledge graph embedding-based power grid internet of things device plug and play method of any of claims 1-9.
12. 12. A computer readable storage medium having stored thereon a computer program which when executed by a processor implements the multi-modal disambiguation and knowledge graph embedding based plug and play method of a grid internet of things device according to any of claims 1-9.

Description

Plug-and-play method and system for Internet of things equipment of power grid based on multi-mode disambiguation and knowledge graph embedding Technical Field The invention relates to the technical field of network security, in particular to a plug-and-play method and a plug-and-play system of power grid Internet of things equipment based on multi-mode disambiguation and knowledge graph embedding. Background In existing grid automation or distributed energy access processes, device registration and parameter configuration mainly rely on manual setting or semi-automation processes, which causes a series of problems. The device has poor consistency in matching with the object model. The object model is a standardized description of the system on the properties and behaviors of the equipment, and takes an intelligent switch in IEC standard as an example, and the object model can define readable and writable parameters, communication protocols, parameter units, upper and lower limits and the like. However, in actual situations, manufacturer equipment specifications are various, protocol implementation is different, omission or errors are easy to occur in manual input, and object models stored in the system are inconsistent with actual capacities of field equipment. Configuration information migration and upstream and downstream resource dependence lack intelligent analysis. Configuration information migration refers to the fact that when a new device is accessed, the system needs to associate the new device with upstream and downstream devices, and automatically map device parameters, protection logic, communication links and the like into the existing configuration of the whole system. However, the current method is to manually consult the circuit diagram and the topology structure to determine the dependency relationship of the new equipment, and then manually update the topology database of the master station, the protection configuration file and the like, so that errors such as miss-matching, mismatch and the like are very easy to occur. The main reasons are that the system analysis is not performed by utilizing methods such as atlas, dependency relationship reasoning, machine learning and the like, the upstream and downstream resource dependence of equipment cannot be automatically analyzed, and the parameters cannot be intelligently transferred and aligned in the whole network. When a user or an operation and maintenance person gives out a natural language instruction, the system has insufficient resolving power on fuzzy or ambiguous intentions. For example, when a user issues a "grid-tie the piece of photovoltaic access to the distribution" instruction, existing systems often fail to translate this ambiguous intent into a specific executable configuration, either by default, error prone, or by relying on a manual dispatcher for translation. Therefore, a solution is needed to solve the problems that in the existing network Internet of things equipment access process, user instruction ambiguity is difficult to resolve, equipment configuration depends on manual decision, matching consistency of an object model is poor, analysis of resource dependence is inaccurate, and the like. Disclosure of Invention Aiming at the problems in the background technology, the invention provides the plug-and-play method of the power grid Internet of things equipment based on multi-mode disambiguation and knowledge map embedding, which can automatically disambiguate user instructions, accurately construct knowledge association and intelligently generate an optimal configuration scheme, reduce manual intervention, ensure the safety and compatibility of equipment access and improve the access efficiency of the power grid Internet of things equipment. In order to achieve the aim of the invention, the invention adopts the following technical scheme: a plug-and-play method of network equipment of an electric network based on multi-mode disambiguation and knowledge graph embedding comprises the following steps: acquiring a user multi-modal instruction, synchronously acquiring context data and environment perception data, inputting the multi-modal instruction, the context data and the environment perception data into a pre-trained large language model, and generating at least two candidate configuration intents; The method comprises the steps of extracting configuration information of new equipment, constructing a knowledge graph comprising equipment attributes, topological relations and determined equipment configuration intentions, modeling the knowledge graph by adopting RDF triples, generating triples by means of Schema alignment and rule mapping for structured data, generating triples by means of entity alignment and relation extraction for unstructured data, analyzing resource dependency relations between the new equipment and an upstream transformer substation and a downstream feeder line when the new equipment is connected