Aerospace manufacturing process knowledge retrieval enhancement generation method based on dynamic graph neural network and large language model collaboration

Abstract

The invention relates to an aerospace manufacturing process knowledge retrieval enhancement generation method based on cooperation of a dynamic graph neural network and a large language model. The method comprises the following steps: constructing a process document set containing aerospace manufacturing process information, and constructing a knowledge graph and a correlation index matrix based on the large language model; constructing a dynamic graph neural network model based on query dependence, and predicting a matching score of each entity based on a user query problem; further calculating a correlation score of the process document and the user query problem based on the matching score and the correlation index matrix, and returning the process document highly related to the user query problem; and inputting the process document highly related to the user query question and the user query question into the large language model, and generating an aerospace manufacturing process knowledge answer matched with the user query question. According to the method, cross-document and cross-knowledge-graph complex association retrieval can be realized, and the knowledge utilization rate and reasoning precision of a large language model in the field of aerospace manufacturing are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and aerospace manufacturing technology, and in particular to an aerospace manufacturing process knowledge retrieval enhancement generation method based on the collaboration of a dynamic graph neural network and a large language model. Background Art

[0002] Aerospace products are characterized by high unit value, a high degree of customization, and a long manufacturing process chain, placing extremely high demands on the accuracy, reliability, and traceability of process solutions. A large amount of process design and optimization work relies on the accumulated experience of senior engineers, but this implicit knowledge is often difficult to systematically inherit and reuse. Large language models (LLMs) have powerful natural language understanding and generation capabilities. They can make implicit knowledge explicit and construct a shareable and trainable knowledge system, thereby alleviating the technological gap caused by talent loss to a certain extent. Therefore, research on large language models for aerospace manufacturing processes has important engineering value and strategic significance.

[0003] While current large language models demonstrate impressive reasoning capabilities, they remain limited when processing domain-specific expertise (particularly aerospace manufacturing process knowledge), hindering their ability to directly and efficiently leverage domain knowledge beyond pre-trained corpora. To address this shortcoming, retrieval-augmented generation (RAG) has gained widespread attention. Its core concept is to incorporate task-related external documents into the generation process, injecting new knowledge into the static large language model. However, existing RAG methods typically retrieve documents independently, making it difficult to capture the complex relationships between cross-document knowledge fragments. This hinders the performance of large language models in multi-hop reasoning and cross-document knowledge integration tasks. Recent research has demonstrated that incorporating graph structure information into text indexing and retrieval can enable efficient retrieval of entities and their relationships, effectively enhancing the accuracy and relevance of generated results. However, these methods often rely on training on specific graph structures and lack the ability to transfer to new manufacturing process knowledge graphs. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides an aerospace manufacturing process knowledge retrieval enhancement generation method based on the collaboration of dynamic graph neural network and large language model, which solves the problem that the traditional retrieval enhancement generation RAG method is difficult to capture the complex correlation relationship between cross-document knowledge fragments and lacks the ability to migrate to the new manufacturing process knowledge graph.

[0005] To solve the above technical problems, the present invention provides the following technical solution: a method for enhancing the generation of aerospace manufacturing process knowledge retrieval based on the collaboration of a dynamic graph neural network and a large language model, comprising the following steps: S1. Collect and construct a process document collection containing aerospace manufacturing process information, build a knowledge graph based on a large language model, and establish an association index matrix corresponding to the process document collection; S2. Build a query-dependent dynamic graph neural network model and optimize its parameters to predict the matching score of each entity in the knowledge graph based on the user's query question, so as to assist in locating process documents that are highly relevant to the user's query question; S3, based on the matching score and the associated index matrix, further calculate the relevance score between the process document and the user query question, and based on the relevance score, return the top documents in descending order. A process document that is highly relevant to the user's query; S4. Input the process documents and user query questions that are highly relevant to the user's query questions into the large language model to generate aerospace manufacturing process knowledge answers that match the user's query questions.

[0006] Furthermore, in step S1, the specific process includes the following steps: S11. Collect a set of process documents related to aerospace manufacturing process information, convert the collected process documents into a processable plain text format, and perform segmentation and sentence processing to obtain a process document collection. ; S12. Use a large language model as an information extraction engine to parse the process document collection For each process document in the aerospace manufacturing domain, identify the entity set and relationship set , organize and construct triple sets ;in 、 Respectively represent , No. entities, Indicates the Entity and the relationships between entities; S13. Use the pre-trained dense vector encoding model to encode each entity and each relationship into an entity vector and a relationship vector respectively; S14. Calculate the cosine similarity between entity vectors; if the cosine similarity of two entity vectors exceeds the set threshold, the two entities corresponding to them are considered to be similar entities; for each pair of similar entities, add additional relations to obtain additional triple sets , thereby enriching the connectivity of the knowledge graph; S15. Final knowledge graph Defined as ,in 、 Represents two entities, Indicate the relationship between them; S16. Define a binary matrix, whose row index represents the entity index and column index represents the process document index; traverse the process document collection For each process document in the document, find all entities it contains and set the elements at the corresponding positions in the binary matrix to 1 to form an associated index matrix , used to quickly locate process documents.

[0007] Furthermore, in step S2, the specific process includes the following steps: S21. Initialize the user query question, entities and relationships in the knowledge graph using the pre-trained dense vector encoding model to obtain the query vector , entity node feature set , relationship feature set ; S22. The query-dependent dynamic graph neural network model includes: The first layer is a Bellman-Ford message passing layer and the last layer is an output layer for fusing query information. S23, query vector , entity node feature set , relationship feature set Input into the query dependency-based dynamic graph neural network model, after After the Bellman-Ford message passing layer, the entity vector set is obtained ; S24, the entity vector set Each entity node feature in Post-concatenation query vector , sent to the output layer to obtain the matching score between the entity and the current user query question.

[0008] Furthermore, the Bellman-Ford messaging layer updates each entity node feature and relationship feature based on the following formula: ; ; ; ; in, Indicates the Layer entity node features, and , Indicates the Layer Entity and The relationship characteristics of entities, and , ⊙ represents element-by-element multiplication, Indicates the Entity to the The information transmitted by the entity is the attention weight, expressed based on the query vector , No. Entity relative to the The importance of other entities in the neighborhood of an entity, represents the concatenation of feature dimensions, and Represents the weights of two linear layers processing query vector and entity node features respectively, Represented in the knowledge graph Middle and First The index set of all entities that the entity has a relationship with, Represents a perceptron consisting of two fully connected layers; the entity node features output by the last layer of the network Constituent entity vector set .

[0009] Furthermore, the output layer includes a perceptron consisting of a fully connected layer and a Sigmoid activation layer, which integrates the query information from the query vector into the entity node features.

[0010] Furthermore, the parameters of the query-dependent dynamic graph neural network model are optimized by maximizing the likelihood probability of query-related entities, and a loss function is constructed. Defined as: ; In the formula Indicates a set of related entities to query, Represents a collection of query-irrelevant entities, Indicates that the query is related to the entity The matching score predicted by the query-dependent dynamic graph neural network model, Indicates that entities are irrelevant to the query Matching scores predicted by the query dependency-based dynamic graph neural network model.

[0011] Furthermore, all entities in the knowledge graph that overlap with entities in the user's query question and all entities related to this Entities connected by relationships constitute the query-related entity set ; The remaining entities in the knowledge graph constitute the query-irrelevant entity set .

[0012] Furthermore, in step S3, the specific process includes the following steps: S31. Obtain the matching score of each entity in the knowledge graph based on the user query question Then, based on the association index matrix , calculate the relevance score between each process document in the process document collection and the user query question as follows: ; S32. Correlation score Sort the data in descending order by value, and take the first The index number corresponding to the data is obtained, and the corresponding process document is searched in the process document collection based on the index number, that is, the process document that is highly relevant to the user's query question.

[0013] Furthermore, step S4 specifically includes: incorporating process documents that are highly relevant to the user's query questions into the context input of the large language model, and combining them with the user's query questions to generate accurate and detailed aerospace manufacturing process knowledge answers to support intelligent question answering and technical guidance in aerospace manufacturing scenarios.

[0014] Through the above technical solution, the present invention provides an enhanced generation method for aerospace manufacturing process knowledge retrieval based on the collaboration of dynamic graph neural network and large language model, which has at least the following beneficial effects: (1) This invention accurately calculates relevance scores by predicting matching scores based on a query-dependent dynamic graph neural network model. This process dynamically adjusts the message delivery process based on the specific characteristics and semantic information of the user's query question, locates the most relevant information on different knowledge graphs, and implements complex association retrieval and multi-hop reasoning across documents and knowledge graphs.

[0015] (2) The query-dependency-based dynamic graph neural network model constructed by this invention can be directly applied to different query scenarios and various aerospace manufacturing process knowledge graphs. The model can simultaneously consider the semantic features and structural information of the aerospace manufacturing process knowledge graph, thereby learning the complex association relationship between queries and entities.

[0016] (3) This method captures the complex relationships between queries, documents, and aerospace manufacturing knowledge graphs in a unified and portable representation space, identifies relevant entities in the knowledge graph, and quickly locates corresponding process documents. This key information is efficiently incorporated into the context of a large language model, generating highly accurate answers to aerospace manufacturing process knowledge. This effectively improves the knowledge utilization and reasoning accuracy of the large language model in the aerospace manufacturing field. This method is applicable to multiple scenarios, including process design, optimization, and decision support. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 This is a flow chart of the aerospace manufacturing process knowledge retrieval enhancement generation method based on the collaboration of a dynamic graph neural network and a large language model of the present invention; Figure 2 This is a simplified diagram of the aerospace manufacturing process knowledge graph structure in an embodiment of the present invention; Figure 3 This is the first part of the query-dependent dynamic graph neural network model constructed in the embodiment of the present invention. A simple diagram of the Bellman-Ford message passing layer; Figure 4 Schematic diagram of the enhanced generation paradigm for aerospace manufacturing process knowledge retrieval based on the collaboration of dynamic graph neural network and large language model in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the above-mentioned objectives, features, and advantages of the present invention more clearly understood, the present invention is further described below in detail with reference to the accompanying drawings and specific embodiments. This will enable a full understanding of how this application uses technical means to solve technical problems and achieve technical effects, and to implement the invention accordingly.

[0019] Those skilled in the art will appreciate that all or part of the steps in the above-mentioned embodiment methods can be accomplished by instructing the relevant hardware through a program. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0020] Please refer to Figures 1-4 , showing a specific implementation of this embodiment. This embodiment constructs a knowledge graph and association index matrix in conjunction with a large language model. This query-dependent dynamic graph neural network model is then used to locate process documents highly relevant to the user's query. The large language model is then used again to generate aerospace manufacturing process knowledge answers that match the user's query. This embodiment enables complex association retrieval and multi-hop reasoning across documents and knowledge graphs, effectively improving the knowledge utilization and reasoning accuracy of large language models in the aerospace manufacturing field. It is suitable for scenarios such as process design, optimization, and decision support.

[0021] Please refer to Figure 1 This embodiment proposes a method for enhancing the generation and retrieval of aerospace manufacturing process knowledge based on the collaboration of a dynamic graph neural network and a large language model. The method includes the following steps: S1. Collect and construct a collection of process documents containing aerospace manufacturing process information. Based on a large language model, accurately extract knowledge elements such as entities and relationships from the process documents, construct a knowledge graph, and establish an association index matrix corresponding to the process document collection. As a preferred implementation of step S1, the specific process includes the following steps: S11. Collect a set of process documents related to aerospace manufacturing process information, convert the collected process documents into a processable plain text format (such as txt format), and perform segmentation and sentence processing to obtain a process document collection ; S12. Use a large language model (e.g., OpenBMB / MiniCPM-2.6) as an information extraction engine to parse a collection of process documents. For each process document in the aerospace manufacturing domain, identify the entity set (such as: part name, processing technology, material, equipment model, etc.) and relationship set (e.g., "adopt," "connect," "process by," etc.), organizing and constructing a set of triples ;in 、 Respectively represent , No. entities, Indicates the Entity and the relationships between entities; Figure 2 A simplified diagram of the aerospace manufacturing process knowledge graph is shown. For process documents 1-5 in the figure, the processes include welding, milling, heat treatment, surface engineering, and additive manufacturing, and the materials include titanium alloys, aluminum alloys, magnesium alloys, high-temperature alloys, carbon fiber composites, glass fiber composites, ceramic matrix composites, metal matrix composites, and high-strength steel. S13. Use a pre-trained dense vector encoding model (e.g., Sentence-BERT) to encode each entity and each relationship into an entity vector and a relationship vector, respectively, to form an entity vector set and a relationship vector set, respectively. S14. Calculate the cosine similarity between entity vectors. If the cosine similarity of two entity vectors exceeds the set threshold, the two entities corresponding to them are considered to be the same entity or strongly semantically related entities, and are defined as similar entities. For each pair of similar entities, add additional relations, such as "synonymous with" and "alias", to obtain an additional set of triples. , thereby enriching the connectivity of the knowledge graph; in this embodiment, the threshold involved is 0.85; S15. Final knowledge graph Defined as ,in 、 Represents two entities, Represent the relationship between them; Stored in a graph database to support subsequent retrieval and reasoning of dynamic graph neural networks; S16. Define a binary matrix, whose row index represents the entity index and column index represents the process document index; traverse the process document collection For each process document in , find all entities it contains and set the elements at the corresponding positions in the binary matrix to 1, that is: Indicates "entity Appears in the document ", forming the associated index matrix , used to quickly locate process documents; associate the index matrix Stored as efficient and queryable data structures (such as sparse matrices and hash tables).

[0022] S2. Build a query-dependent dynamic graph neural network model that dynamically adjusts the weights of nodes and edges in the graph structure based on the user's query input to adapt to the semantic characteristics of different queries. Optimize the model parameters to predict the matching score of each entity in the knowledge graph based on the user's query, thereby assisting in locating process documents that are highly relevant to the user's query. As a preferred implementation of step S2, the specific process includes the following steps: S21. User query question (Natural language problems / questions in retrieval tasks), knowledge graphs Entities in and relationships Use pre-trained dense vector encoding models (such as Sentence-BERT) to initialize and obtain query vectors , entity node feature set , relationship feature set ; S22. The query-dependent dynamic graph neural network model includes The first layer is a Bellman-Ford message passing layer and the last layer is an output layer for fusing query information. S23, query vector , entity node feature set , relationship feature set Input into the query dependency-based dynamic graph neural network model, after After the Bellman-Ford message passing layer, the entity vector set is obtained ; S24, the entity vector set Each entity node feature in Post-concatenation query vector , sent to the output layer to obtain the matching score between the entity and the current user query question.

[0023] More specifically, the Bellman-Ford messaging layer updates each entity node feature and relationship feature based on the following formula: ; ; ; ; in, Indicates the Layer entity node features, and , Indicates the Layer Entity and The relationship characteristics of entities, and , ⊙ represents element-by-element multiplication, Indicates the Entity to the The information transmitted by the entity is the attention weight, expressed based on the query vector , No. Entity relative to the The importance of other entities in the neighborhood of an entity, represents the concatenation of feature dimensions, and Represents the weights of two linear layers processing query vector and entity node features respectively, Represented in the knowledge graph Middle and First The index set of all entities that the entity has a relationship with, Represents a perceptron consisting of two fully connected layers; the entity node features output by the last layer of the network Constituent entity vector set . Figure 3 The first part of the dynamic graph neural network model based on query dependency is presented. A simplified diagram of the Bellman-Ford messaging layer.

[0024] More specifically, the output layer includes a perceptron consisting of a fully connected layer and a Sigmoid activation layer, which integrates query information from the query vector into entity node features.

[0025] More specifically, the parameters of the query-dependent dynamic graph neural network model are optimized by maximizing the likelihood probability of query-related entities and constructing a loss function Defined as: ; In the formula Indicates a set of related entities to query, Represents a collection of query-irrelevant entities, Indicates that the query is related to the entity The matching score predicted by the query-dependent dynamic graph neural network model, Indicates that entities are irrelevant to the query The matching score predicted by the query-dependent dynamic graph neural network model. In this embodiment, the network training process uses the AdamW optimizer to update the network parameters, and the learning rate is set to 5.0e-4.

[0026] More specifically, traversing the user query question Entity Set in , retrieve entity sets from knowledge graphs , All entities in Overlapping entities constitute the overlapping entity set , Entities in and all Entities connected by relationships (refer to the relationship set of the knowledge graph) constitute the query-related entity set ; The remaining entities, namely The query-irrelevant entity set .

[0027] In this embodiment, the query dependency-based dynamic graph neural network model constructed can be directly applied to different queries and aerospace manufacturing process knowledge graphs. The model can simultaneously consider the semantic features and structural information of the aerospace manufacturing process knowledge graph, thereby learning the complex association relationship between queries and entities. S3, based on the matching score and the associated index matrix, further calculate the relevance score between the process document and the user query question, and based on the relevance score, return the top documents in descending order. A process document that is highly relevant to the user's query; As a preferred embodiment of step S3, the specific process includes the following steps: S31, after obtaining the knowledge graph , calculate the matching score of each entity in the knowledge graph based on the user query question ; S32, calculate the relevance score of each process document in the process document set and the user query question based on the association index matrix ; ; S33, sort the data in the relevance score from large to small, take the index number corresponding to the first data, and find the corresponding process document in the process document set based on the index number, which is the process document highly relevant to the user query question .

[0028] In this embodiment, the relevance score is calculated based on the matching score predicted by the dynamic graph neural network model based on the query dependency, which can dynamically adjust the message passing process according to the user query question, locate the most relevant information on different knowledge graphs, and realize complex association retrieval and multi-hop reasoning across documents and knowledge graphs.

[0029] S4, input the filtered process document highly relevant to the user query question and the user query question into the large language model to generate an aerospace manufacturing process knowledge answer highly matched, accurate and professional to the user query question; As a preferred embodiment of step S4, the specific process includes the following steps: S41, input the process document highly relevant to the user query question into the context input of the large language model LMM, and combine the user query question to generate an aerospace manufacturing process knowledge answer highly matched, accurate and professional to the user query question to support intelligent question answering and technical guidance in aerospace manufacturing scenarios, that is: ; Figure 4 This embodiment shows the aerospace manufacturing process knowledge retrieval enhancement generation paradigm based on the dynamic graph neural network. In this embodiment, the dynamic graph neural network model based on query dependency receives the aerospace manufacturing process question proposed by the user, constructs the aerospace manufacturing process knowledge graph, sorts the aerospace manufacturing process documents and locates the high-relevance process documents, then sends the aerospace manufacturing process question proposed by the user and the located high-relevance process documents into the large language model LMM to obtain the aerospace manufacturing process reply.

[0030] In this embodiment, by capturing the complex relationship between queries, documents and aerospace manufacturing knowledge graphs in a unified and portable representation space, identifying relevant entities in the knowledge graph, quickly locating corresponding process documents, and incorporating them into the context of the large language model, highly accurate aerospace manufacturing process knowledge answers are generated, effectively improving the knowledge utilization and reasoning accuracy of the large language model in the field of aerospace manufacturing, and being suitable for scenarios such as process design, optimization and decision support.

[0031] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless they are mutually inconsistent.

[0032] The logic and / or steps represented in the flowchart or otherwise described herein may be considered, for example, as an ordered list of executable instructions for implementing logical functions, and may be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device).

[0033] The above embodiments provide a detailed introduction to the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A method for enhancing the generation and retrieval of aerospace manufacturing process knowledge based on the collaboration of dynamic graph neural network and large language model, characterized by: The following steps are involved: S1. Collect and construct a process document collection containing aerospace manufacturing process information, build a knowledge graph based on a large language model, and establish an association index matrix corresponding to the process document collection; S2. Build a query-dependent dynamic graph neural network model and optimize its parameters to predict the matching score of each entity in the knowledge graph based on the user's query question, thereby assisting in locating process documents that are highly relevant to the user's query question. S3, based on the matching score and the associated index matrix, further calculate the relevance score between the process document and the user query question, and based on the relevance score, return the top documents in descending order. A process document that is highly relevant to the user's query; S4. Input the process documents and user query questions that are highly relevant to the user's query questions into the large language model to generate aerospace manufacturing process knowledge answers that match the user's query questions.

2. The method for enhancing the generation and retrieval of aerospace manufacturing process knowledge based on the collaboration of a dynamic graph neural network and a large language model according to claim 1 is characterized by: The specific process of step S1 includes the following steps: S11. Collect a set of process documents related to aerospace manufacturing process information, convert the collected process documents into a processable plain text format, and perform segmentation and sentence processing to obtain a process document collection. ; S12. Use a large language model as an information extraction engine to parse the process document collection For each process document in the aerospace manufacturing domain, identify the entity set and relationship set , organize and construct triple sets ;in 、 Respectively represent , No. entities, Indicates the Entity and the relationships between entities; S13. Use the pre-trained dense vector encoding model to encode each entity and each relationship into an entity vector and a relationship vector respectively; S14. Calculate the cosine similarity between entity vectors; if the cosine similarity of two entity vectors exceeds the set threshold, the two entities corresponding to them are considered to be similar entities; for each pair of similar entities, add additional relations to obtain additional triple sets , thereby enriching the connectivity of the knowledge graph; S15. Final knowledge graph Defined as ,in 、 Represents two entities, Indicates the relationship between them; S16. Define a binary matrix, whose row index represents the entity index and column index represents the process document index; traverse the process document collection For each process document in the document, find all entities it contains and set the elements at the corresponding positions in the binary matrix to 1 to form an associated index matrix , used to quickly locate process documents.

3. The method for enhancing the generation and retrieval of aerospace manufacturing process knowledge based on the collaboration of a dynamic graph neural network and a large language model according to claim 1 is characterized by: The specific process of step S2 includes the following steps: S21. Initialize the user query question, entities and relationships in the knowledge graph using the pre-trained dense vector encoding model to obtain the query vector , entity node feature set , relationship feature set ; S22. The query-dependent dynamic graph neural network model includes: The first layer is a Bellman-Ford message passing layer and the last layer is an output layer for fusing query information. S23, query vector , entity node feature set , relationship feature set Input into the query dependency-based dynamic graph neural network model, after After the Bellman-Ford message passing layer, the entity vector set is obtained ; S24, the entity vector set Each entity node feature in Post-concatenation query vector , sent to the output layer to obtain the matching score between the entity and the current user query question.

4. The method for enhancing the generation and retrieval of aerospace manufacturing process knowledge based on the collaboration of a dynamic graph neural network and a large language model according to claim 3 is characterized by: The Bellman-Ford messaging layer updates each entity node feature and relationship feature based on the following formula: ； ； ； ； in, Indicates the Layer entity node features, and , Indicates the Layer Entity and The relationship characteristics of entities, and , ⊙ represents element-by-element multiplication, Indicates the Entity to the The information transmitted by the entity is the attention weight, expressed based on the query vector , No. Entity relative to the The importance of other entities in the neighborhood of an entity, represents the concatenation of feature dimensions, and Represents the weights of two linear layers processing query vector and entity node features respectively, Represented in the knowledge graph Middle and First The index set of all entities that the entity has a relationship with, Represents a perceptron consisting of two fully connected layers; the entity node features output by the last layer of the network Constituent entity vector set .

5. The method for enhanced generation and retrieval of aerospace manufacturing process knowledge based on the collaboration of dynamic graph neural network and large language model according to claim 4 is characterized by: The output layer includes a perceptron composed of a fully connected layer and a Sigmoid activation layer, which integrates the query information from the query vector into the entity node features.

6. The method for enhancing the generation and retrieval of aerospace manufacturing process knowledge based on the collaboration of a dynamic graph neural network and a large language model according to claim 5 is characterized by: The parameters of the query-dependent dynamic graph neural network model are optimized by maximizing the likelihood probability of query-related entities and constructing a loss function Defined as: ； In the formula Indicates a collection of related entities to query, Represents a collection of query-irrelevant entities, Indicates that the query is related to the entity The matching score predicted by the query-dependent dynamic graph neural network model, Indicates that entities are irrelevant to the query Matching scores predicted by the query dependency-based dynamic graph neural network model.

7. The method for enhanced generation and retrieval of aerospace manufacturing process knowledge based on the collaboration of dynamic graph neural network and large language model according to claim 5 is characterized by: All entities in the knowledge graph that overlap with the entities in the user's query and all entities related to this Entities connected by relationships constitute the query-related entity set ; The remaining entities in the knowledge graph constitute the query-irrelevant entity set .

8. The method for enhancing the generation and retrieval of aerospace manufacturing process knowledge based on the collaboration of dynamic graph neural network and large language model according to claim 1 is characterized by: The specific process of step S3 includes the following steps: S31. Obtain the matching score of each entity in the knowledge graph based on the user query question Then, based on the association index matrix , calculate the relevance score between each process document in the process document collection and the user query question as follows: ； S32. Correlation score Sort the data in descending order by value, and take the first The index number corresponding to the data is obtained, and the corresponding process document is searched in the process document collection based on the index number, that is, the process document that is highly relevant to the user's query question.

9. The method for enhancing the generation and retrieval of aerospace manufacturing process knowledge based on the collaboration of dynamic graph neural network and large language model according to claim 1 is characterized by: Step S4 specifically includes: incorporating process documents that are highly relevant to the user's query questions into the context input of the large language model, and combining them with the user's query questions to generate accurate and detailed aerospace manufacturing process knowledge answers to support intelligent question answering and technical guidance in aerospace manufacturing scenarios.

Images