CN-115905574-B - Knowledge graph construction method and device for ship electric power system design task

Abstract

The invention discloses a knowledge graph construction method and device for a ship electric power system design task. The method comprises the steps of carrying out pattern layer design, defining concepts and relations among the concepts of a ship electric power system, carrying out data preprocessing on unstructured data, marking the entities and relations among the entities, training the marked entity data set and entity relation data set to obtain corresponding models, carrying out triplet extraction based on the obtained models, importing the three-tuple extraction into a graph database to form a ship electric power system concept knowledge graph, carrying out modeling to form a ship electric power system design task knowledge graph, and carrying out graph fusion to form a final knowledge graph. The method can realize the construction of the knowledge graph of the ship power system design task, so that the knowledge graph construction in the field has higher convenience, high efficiency and accuracy, a knowledge base is provided for the subsequent construction of a query platform by utilizing the graph, and the retrieval efficiency of ship designers is improved.

Inventors

• XU XIAOBIN
• LIAO QING
• ZHANG ZHENJIE
• SHEN XUFENG
• LI JIANNING
• YUAN ZHAOFENG
• FENG JING
• LUO YANG

Assignees

• 杭州电子科技大学
• 杭州钱航船舶修造有限公司
• 杭州壹诺工业科技有限公司

Dates

Publication Date

20260508

Application Date

20221222

Claims (3)

1. 1. The knowledge graph construction method for the ship electric power system design task is characterized by comprising the following steps of: s1, designing and defining a ship power system concept and a relationship among concepts by a mode layer; S2, carrying out data preprocessing on unstructured data in a ship design instruction manual, design specifications and design standards; S3, marking the entities and the relationships among the entities according to the pattern layer concepts and the relationships among the concepts to obtain an entity data set and an entity relationship data set; S4, training the marked entity data set and the entity relation data set by adopting ALBERT +bidirectional LSTM+CRF algorithm and BERT+bidirectional GRU+attribute+FC algorithm respectively to obtain an entity recognition model and an entity relation recognition model; the ALBERT layer in ALBERT +bidirectional LSTM+CRF algorithm is used for extracting text features, embedding vectors are converted into word vectors output by ALBERT pre-training models, a large amount of semantic information is migrated to perform fine adjustment on downstream tasks, ALBERT output is used as bidirectional LSTM input, the bidirectional LSTM layer can acquire sequence information of a context, and finally the CRF layer decodes and outputs a predicted label sequence which accords with label transfer constraint conditions and is most possible; The BERT layer in the BERT+bidirectional GRU+attention+FC algorithm is used for extracting text features, the bidirectional GRU layer acquires feature vectors of original sentences through a bidirectional GRU neural network, the Attention layer is an Attention mechanism layer, weight vectors are generated, feature vectors based on word or sentence levels are obtained after the feature vectors obtained by the GRU layer are multiplied by corresponding weight vectors, FC is a fully connected layer, and feature integration is carried out on the feature vectors obtained by the front layer; S5, performing triplet extraction based on the obtained entity recognition model and entity relation recognition model, and importing the triplet extraction into a Neo4j graph database to form a conceptual knowledge graph of the ship power system; s6, combining field expert knowledge and a ship power system product structure, analyzing a design business flow, extracting a design task, and performing hierarchical modeling by utilizing protege to form a ship power system design task knowledge graph; S7, fusing the obtained ship power system concept knowledge graph with the design task knowledge graph through entity alignment and disambiguation to form a final ship power system design task-oriented knowledge graph; the relation defined in the step S1 comprises 8 kinds of entity and 10 kinds of entity relations altogether; the 8 kinds of entities are functional components, component attributes, personnel, characteristics, situations, measures, products and positions respectively; Wherein the 10 entity relationships are Belong to, trait of, part of, kind of, condition of, position of, apply to, infer to, PARTICIPATE IN, and Equipment to, respectively; In the step S2, data cleaning is carried out on unstructured data in a ship design instruction manual, design specifications and design standards, and txt texts are obtained through conversion; The entity labeling in the step S3 is based on the concept defined in the step S1, a label labeling tool is adopted for labeling, data is derived into a csv format and is converted into a BIO format through codes, and the entity relationship labeling is based on the concept relationship defined in the step S1 and is converted into a txt format through codes; The step S6 design business flow comprises radio communication and navigation system design, lighting system design, electric traction system design, electric propulsion system design, ship electric power system design, ship automation design and communication system design, and the design task knowledge graph is formed by constructing a body in protege and then importing the body into a Neo4j graph database.
2. 2. The knowledge graph construction method for the ship electric power system design task according to claim 1, wherein the triplet extraction in the step S5 is combined entity extraction and relation extraction, the obtained triplet form is < entity, relation and entity >, the triplet data is imported into a Neo4j graph database, the triplet is firstly processed into the formats of entity.csv and relation.csv, and the triplet is imported into the Neo4j graph database in a Neo4 j-report mode to form the ship electric power system concept knowledge graph.
3. 3. A knowledge graph construction apparatus for a marine vessel power system design task, the apparatus comprising a processor, and a memory arranged to store computer executable instructions which, when executed, cause the processor to perform the knowledge graph construction method of any one of claims 1 or 2.

Description

Knowledge graph construction method and device for ship electric power system design task Technical Field The invention belongs to the field of intelligent design of ships, and particularly relates to a knowledge graph construction method and device for a ship power system design task. Background In the field of marine power system design, designers need to formulate personalized design tasks according to different design business requirements. However, since the ship power system has many functional modules, such as a power supply device, a distribution device meter, a power grid, power system protection, etc., the design of each functional module involves diversified design knowledge. Meanwhile, complex association relations exist among different design knowledge. Therefore, a designer often needs to manually query a large number of standards and specifications during design, and even seeks remote technical support of field experts, so that the whole design process is long in time consumption, low in efficiency and easy to make mistakes, the design process is affected, and the intelligence level needs to be improved. At present, artificial intelligence technology is rapidly developed, and has more mature application in a plurality of fields. The knowledge graph is used as a big branch of artificial intelligence, and has been widely applied in the related fields of search engines, intelligent recommendation, intelligent question-answering and the like. If the knowledge graph can be applied to the ship design stage, various design knowledge can be stored and expressed in a graph form, and the knowledge graph oriented to the ship design is constructed, the intelligent level of the ship design is hopefully improved. Disclosure of Invention In order to solve the defects existing in the prior art, the invention provides a knowledge graph construction method and device for a ship electric power system design task. According to the invention, the ship design knowledge and the association relation between the knowledge can be effectively modeled, so that support is provided for designers to quickly acquire design tasks and related information in the field of ship electric power, and the field knowledge graph is constructed by analyzing the original unstructured data, so that the information retrieval efficiency and the user experience are improved. The specific technical scheme of the invention is as follows: The invention provides a knowledge graph construction method for a ship electric power system design task, which comprises the following steps: s1, designing and defining a ship power system concept and a relationship among concepts by a mode layer; S2, carrying out data preprocessing on unstructured data in a ship design instruction manual, design specifications and design standards; S3, marking the entities and the relationships among the entities according to the pattern layer concepts and the relationships among the concepts to obtain an entity data set and an entity relationship data set; S4, training the marked entity data set and the entity relation data set by adopting ALBERT +bidirectional LSTM+CRF algorithm and BERT+bidirectional GRU+attribute+FC algorithm respectively to obtain an entity recognition model and an entity relation recognition model; S5, performing triplet extraction based on the obtained entity recognition model and entity relation recognition model, and importing the triplet extraction into a Neo4j graph database to form a conceptual knowledge graph of the ship power system; and S6, analyzing the design business flow by combining the field expert knowledge and the product structure of the ship power system, extracting the design task, and performing hierarchical modeling by utilizing protege to form a ship power system design task knowledge graph. And S7, fusing the obtained ship power system concept knowledge graph with the design task knowledge graph through entity alignment and disambiguation to form a final ship power system design task-oriented knowledge graph. Further, the pattern layer design in the step S1 is designed according to expert experience, and defines the concept type of the ship power system and the relationship existing between the concepts. Together, 8 types of entities, namely, functional parts, part attributes, personnel, features, situations, measures, products and locations, and 10 types of entity relationships, namely Belong to, trait of, part of, kind of, condition of, position of, apply to, infer to, PARTICIPATE IN and value to. In step S2, unstructured data such as a manual for ship design, design specifications, design standards, etc. are subjected to data cleaning, and then converted into txt files. Further, the entity labeling in the step S3 is based on the concept defined in the step S1, a label labeling tool is adopted for labeling, data is exported into a csv format and is converted into a BIO format through codes, and entity relation labeling is ba