Intelligent ship design reference data recommendation method and system based on knowledge graph

Abstract

The invention relates to the technical field of knowledge maps, in particular to a ship design reference data intelligent recommendation method and system based on a knowledge map, and the method comprises the steps: constructing a multi-dimensional ship knowledge map; analyzing the received design task to obtain a design basic parameter and a target performance index, constructing a task related sub-graph, extracting a task constraint condition, and generating a target index semantic vector; carrying out graph structure representation learning on the task related sub-graphs by utilizing a graph neural network coding model, and carrying out feature fusion on the task related sub-graphs and target index semantic vectors to generate comprehensive design reference vectors; and based on the comprehensive design reference vector and the task constraint condition, matching historical cases in the knowledge graph to obtain a design reference recommendation result. According to the method, by constructing the multi-dimensional ship knowledge graph, design basic parameters, performance indexes and constraint conditions are systematically managed; historical design cases can be intelligently matched, high-correlation design reference results are provided, and design efficiency and accuracy are improved.

Description

A Knowledge Graph-Based Intelligent Recommendation Method and System for Ship Design Reference Data Technical Field

[0001] This invention relates to the field of knowledge graph technology, specifically to an intelligent recommendation method and system for ship design reference data based on knowledge graphs. Background Technology

[0002] Ship design is a complex systems engineering project, involving multiple aspects such as overall ship layout, structural design, power system configuration, performance optimization, and code compliance. Traditional ship design methods mainly rely on designers' experience, historical design cases, and scattered design codes for reference and decision-making. This approach has the following problems: Information is scattered and difficult to manage systematically. Parameters, performance indicators, structural relationships, and code requirements involved in ship design are usually stored in different databases or documents, lacking a unified association and organization method, making it difficult for designers to quickly obtain comprehensive reference information. Historical design experience is difficult to fully utilize. Although historical design cases can provide reference for new designs, the lack of a systematic representation method often leads designers to spend a lot of time searching and comparing cases, and it is difficult to quantify their relevance to the current design task. Design constraints are complex and difficult to automate. Ship design involves multiple constraints, and traditional methods often rely on manual judgment, which is not only inefficient but also prone to omissions or inaccurate judgments. The design optimization process relies on experience and lacks intelligent support. When design objectives are diverse and conflicting, such as simultaneously needing to consider stability, speed, economy, and safety, traditional design methods are unable to provide comprehensive optimization guidance, which can easily lead to low design efficiency or suboptimal solutions. Summary of the Invention

[0003] (I) Purpose of the Invention The purpose of this invention is to provide an intelligent recommendation method and system for ship design reference data based on knowledge graphs. By constructing a multi-dimensional ship knowledge graph, the basic design parameters, performance indicators and constraints are systematically managed. By combining conditional retrieval strategies and graph neural networks, the structure representation of task-related subgraphs is learned and fused with the semantic vector of the target indicator. This enables intelligent matching of historical design cases, provides highly relevant design reference results, improves design efficiency and accuracy, supports multi-objective optimization design, adapts to different types of ship design tasks, and has good scalability and adaptability.

[0004] (II) Technical Solution To address the above problems, this invention provides an intelligent recommendation method for ship design reference data based on a knowledge graph, comprising: constructing a multi-dimensional ship knowledge graph; parsing the received design task to obtain basic design parameters and target performance indicators, wherein the basic design parameters are used to define the design space and the target performance indicators are used to define the design optimization direction; using a conditional retrieval strategy to locate the set of nodes matching the basic design parameters in the knowledge graph and constructing a task-related subgraph; extracting task constraints based on the task-related subgraph; generating a target indicator semantic vector based on the target performance indicators; using a graph neural network encoding model to perform graph structure representation learning on the task-related subgraph and fusing features with the target indicator semantic vector to generate a comprehensive design reference vector; and matching historical cases in the knowledge graph based on the comprehensive design reference vector and the task constraints to obtain design reference recommendation results.

[0005] In another aspect of the present invention, preferably, the knowledge graph uses design fundamental parameters as nodes and constraints as edges, the constraints including design logical relationships, performance constraint relationships, and specification association relationships; the parsing of the received design task to obtain design fundamental parameters and target performance indicators includes: dividing the design task to obtain a design fundamental information part and a performance requirement description part; performing feature extraction on the design fundamental information part to obtain design fundamental parameters, the design fundamental parameters including: ship type, dimensional parameters, displacement, structural layout form, and sea state level; performing semantic parsing and keyword extraction on the performance requirement description part to identify target performance indicators, the target performance indicators including: stability, speed, resistance, structural strength, economy, and safety.

[0006] In another aspect of the present invention, preferably, the step of locating the set of nodes matching the design basis parameters in the knowledge graph using a conditional retrieval strategy and constructing a task-related subgraph includes: matching the design basis parameters with each node in the knowledge graph based on at least one of ship type, dimensional parameters, displacement, structural layout form, and sea state level to obtain a first candidate node set; performing multi-hop expansion with the first candidate node set as the starting node to obtain a second candidate node set associated with the first candidate node set; and performing path consistency detection on the first candidate node set and the second candidate node set to obtain a target node set and construct a task-related subgraph.

[0007] In another aspect of the present invention, preferably, the step of performing path consistency detection on the first candidate node set and the second candidate node set to obtain a target node set and construct a task-related subgraph includes: traversing paths along the edges between nodes in the first and second candidate node sets to generate a set of candidate paths between nodes; calculating a path consistency score for each candidate path, the consistency score being based on the similarity of node attributes in the path, edge type matching, and correlation with design parameters; setting a path consistency threshold, filtering paths with scores greater than the threshold, the nodes involved constituting the target node set; and combining the target node set and its corresponding edge information to form a task-related subgraph.

[0008] In another aspect of the present invention, preferably, the step of extracting task constraints based on the task-related subgraph includes: extracting an edge set from the task-related subgraph, the edge set including logical relationships, performance constraint relationships, and specification association relationships; and performing path traversal on the nodes in the task-related subgraph along the design logical relationships, performance constraint relationships, and specification association relationships to form task constraints describing the design task constraint boundaries.

[0009] In another aspect of the present invention, preferably, generating a target indicator semantic vector based on the target performance indicator includes: converting the category, value, or level information of each target performance indicator into a structured vector representation, wherein the structured vector is used to describe the computable attributes of each indicator; semantically encoding the text description of the target performance indicator and mapping it to a high-dimensional semantic vector space to form a text semantic vector; fusing the structured vector and the text semantic vector to generate a preliminary performance indicator vector; and adjusting the preliminary performance indicator vector based on the correlation between the target performance indicators to generate a comprehensive target indicator vector.

[0010] In another aspect of the present invention, preferably, the step of using a graph neural network encoding model to learn the graph structure representation of the task-related subgraph and fusing it with the target indicator semantic vector to generate a comprehensive design reference vector includes: initializing the node and edge features of the task-related subgraph to form a graph structure feature matrix; inputting the graph structure feature matrix into the graph neural network encoding model, learning the structure representation of the task-related subgraph through a graph attention mechanism to generate a task subgraph embedding vector; performing a linear transformation on the target indicator semantic vector to match the dimension of the target indicator semantic vector with the dimension of the task subgraph embedding vector to obtain a transformed target indicator semantic vector; and fusing the task subgraph embedding vector with the transformed target indicator semantic vector to generate a comprehensive feature vector.

[0011] In another aspect of the present invention, preferably, the task subgraph embedding vector and the transformed target indicator semantic vector are fused to generate a comprehensive feature vector, including: concatenating the task subgraph embedding vector and the transformed target indicator semantic vector to form a preliminary fusion vector; weighting each dimension of the preliminary fusion vector based on an attention mechanism to obtain a weighted fusion vector; and normalizing the weighted fusion vector to generate a comprehensive feature vector.

[0012] In another aspect of the present invention, preferably, based on the comprehensive design reference vector and the task constraints, matching historical cases in the knowledge graph to obtain design reference recommendation results includes: constructing case feature vectors for each historical case in the knowledge graph; filtering out historical cases that do not meet the task constraints based on a preset constraint consistency judgment rule; calculating the similarity between the case feature vectors of the remaining historical cases and the comprehensive design reference vector; sorting the historical cases according to the calculated similarity values, filtering out historical cases with similarity higher than a preset threshold to form a candidate design reference set; and matching the candidate design reference set with the received design task to obtain design reference recommendation results.

[0013] In another aspect, preferably, a knowledge graph-based intelligent recommendation system for ship design reference data includes: a first construction module for constructing a multi-dimensional ship knowledge graph; a parsing module for parsing received design tasks to obtain basic design parameters and target performance indicators, wherein the basic design parameters are used to define the design space and the target performance indicators are used to define the design optimization direction; a second construction module for locating a set of nodes matching the basic design parameters in the knowledge graph using a conditional retrieval strategy to construct a task-related subgraph; an extraction module for extracting task constraints based on the task-related subgraph; a first generation module for generating a target indicator semantic vector based on the target performance indicators; a second generation module for learning the graph structure representation of the task-related subgraph using a graph neural network encoding model and fusing features with the target indicator semantic vector to generate a comprehensive design reference vector; and an acquisition module for matching historical cases in the knowledge graph based on the comprehensive design reference vector and the task constraints to obtain design reference recommendation results.

[0014] (III) Beneficial Effects The above-mentioned technical solution of this invention has the following beneficial technical effects: By constructing a multi-dimensional ship knowledge graph, this invention systematically manages the basic parameters, performance indicators, constraints, and historical design cases in ship design, realizing the structured, queryable, and associative nature of design information, thus solving the problems of scattered information and low retrieval efficiency in traditional ship design. Through a conditional retrieval strategy, the set of nodes matching the basic design parameters is located in the knowledge graph, and a task-related subgraph is constructed. Path traversal and extraction are performed on design logical relationships, performance constraint relationships, and specification association relationships, achieving efficient acquisition of design task constraints. Based on this, the target performance indicators are transformed into structured vectors and semantic vectors, and a comprehensive target indicator vector is generated by combining the correlation between indicators, providing a quantitative basis for design optimization. Through graph neural network learning of the structural representation of the task-related subgraph and feature fusion with the target indicator semantic vector, a comprehensive design reference vector is generated. This enables intelligent matching of historical design cases in the knowledge graph, thereby providing highly relevant design reference results and improving the efficiency and accuracy of ship design. Meanwhile, this method can be adapted to ship design tasks of different types, sizes and operating sea state levels, supports multi-objective optimization, has good scalability and long-term adaptability, and provides intelligent and precise decision support for the ship design process. Attached Figure Description

[0015] Figure 1 is a schematic diagram of the overall process of an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0017] Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0018] In the description of this invention, it should be noted that the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0019] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0020] Example 1: A knowledge graph-based intelligent recommendation method for ship design reference data. Figure 1 shows a schematic diagram of the overall process of an embodiment of the present invention. As shown in Figure 1, the method includes: constructing a multi-dimensional ship knowledge graph; constructing a multi-dimensional ship knowledge graph based on multi-source heterogeneous ship data, including historical ship type overall parameters, structural design parameters, propulsion system configuration, performance test results, standard clause data, and expert experience knowledge. The knowledge graph uses design basic parameters as nodes and constraints as edges. The constraints include design logical relationships, performance constraint relationships, and standard association relationships. The data is processed through steps such as entity extraction, relation extraction, data cleaning, and semantic alignment to finally form a unified and queryable multi-dimensional knowledge graph.

[0021] The system parses the received design task to obtain basic design parameters and target performance indicators. The basic design parameters define the design space, and the target performance indicators define the design optimization direction. In this embodiment, parsing the received design task to obtain these parameters includes: dividing the design task to obtain a basic design information section and a performance requirement description section; receiving user-inputted design task information, which can originate from text descriptions, templated task forms, design requirement documents, etc. To extract structured parameters, the system semantically divides the design task and automatically identifies two types of information: the basic design information section, which mainly describes the ship's basic characteristics and configuration requirements, typically possessing explicit parameter properties or inferable fixed attributes; and the performance requirement description section, which mainly includes the designer's performance requirements, such as target speed, stability rating, and economic expectations, given in natural language description or indicator form. This division can be achieved through rule templates, natural language processing, semantic segmentation models, etc., making subsequent information extraction more targeted and efficient.

[0022] Feature extraction is performed on the basic design information to obtain basic design parameters, including: ship type, dimensional parameters, displacement, structural layout, and sea state level. For the basic design information, the system employs entity recognition algorithms, parameter pattern matching, and terminology database comparison to extract features from text containing content with clear numerical, categorical, or structural meaning. This ultimately generates a set of basic parameters for the design task. Ship type: for example, multipurpose cargo ship, container ship, tanker, tugboat, research vessel, engineering vessel, etc. Dimensional parameters: including main dimensional indicators such as length (L), beam (B), depth (D), and design draft (T); displacement: including design displacement or displacement range; structural layout: such as single bottom / double bottom, single hull / double hull, longitudinal / transverse frame structure, and compartment distribution; sea state level: including navigation area level, sea state level (e.g., sea state 6, 7), and whether ice reinforcement is required.

[0023] The performance requirement description is semantically parsed and keywords are extracted to identify target performance indicators, including stability, speed, drag, structural strength, economy, and safety. The performance requirement description employs semantic parsing models, keyword extraction algorithms, and indicator vocabulary matching to identify the specific performance requirements of the task. The system extracts target performance indicators from the description and constructs a set of performance indicators representing the design optimization direction. Stability performance includes lateral stability indicators, restoring moment requirements, and stability requirements under full load and partial load. Speed ​​targets include design speed, maximum speed, or speed range. Drag performance includes drag curve prediction, line optimization direction, and fluid performance targets reflecting motion drag. Structural strength requirements include strength assessment under specific sea states, structural fatigue life requirements, and longitudinal strength requirements. Economic indicators include fuel consumption, operating costs, and cost constraints related to life-cycle economics. Safety indicators include watertight compartment layout requirements, damage stability requirements, and emergency system configuration requirements.

[0024] A conditional retrieval strategy is employed to locate the set of nodes in the knowledge graph that match the basic design parameters, constructing a task-related subgraph. For example, if the task is specified as a "3000-ton multipurpose cargo ship," historical case nodes with similar displacement ranges are selected; if the specified navigation area is "coastal navigation area," parameter nodes that meet the corresponding navigation area design specifications are selected; if a specific structural type is specified, such as a double-bottom structure, design nodes containing that structure are selected. Through these conditional selections, a task-related subgraph containing nodes and edges directly related to the task is constructed. This subgraph limits the model input range, avoids interference from irrelevant information, and improves computational efficiency and result accuracy. In this embodiment, the step of locating the set of nodes matching the design basic parameters in the knowledge graph using a conditional retrieval strategy and constructing a task-related subgraph includes: matching the design basic parameters with each node in the knowledge graph based on at least one of ship type, dimensional parameters, displacement, structural layout, and sea state level to obtain a first candidate node set; and comparing the design basic parameters with corresponding nodes in the knowledge graph using category matching, numerical range matching, or semantic similarity matching, respectively, according to the attribute type and numerical characteristics of each parameter, to construct a first candidate node set for describing the basic constraints of the design task.

[0025] Starting from the first candidate node set, a multi-hop expansion is performed to obtain a second candidate node set associated with the first candidate node set. Based on the edge relationship types in the knowledge graph (including design logic relationships, performance coupling relationships, normative constraint relationships, and historical case association relationships), the first candidate node set is expanded in multiple hops. During the expansion process, according to the preset maximum number of hops (e.g., 2 to 4 hops) and relationship weights, the adjacent nodes reachable from the starting node are searched layer by layer to filter nodes that are directly or indirectly related to the ship's overall parameters, structural parameters, performance indicators, or normative clauses, forming a second candidate node set. The second candidate node set is used to supplement the potential constraint information related to the task and expand the knowledge coverage of the task.

[0026] The first and second candidate node sets are subjected to path consistency detection to obtain the target node set. A task-related subgraph is constructed, including: traversing paths along the edges between nodes in the first and second candidate node sets to generate a set of candidate paths between nodes; based on the multiple edge relationships defined in the knowledge graph, starting from any node in the first candidate node set, performing a depth-first or breadth-first traversal within a preset maximum hop count to search for nodes in the second candidate node set that can be reached, and recording all possible paths to reach them. During the path traversal, to avoid path explosion and cycle accumulation, historical visit markers are set for visited nodes; when a marked node is visited again, the system terminates the current path extension to ensure that the paths are finite and acyclic. The above traversal process ultimately generates multiple sets of candidate paths reflecting the constraints between nodes, with each path containing a sequence of nodes and their corresponding edge sequences.

[0027] A path consistency score is calculated for each candidate path. This score is based on the similarity of node attributes, edge type matching, and relevance to design fundamental parameters. Node attribute similarity is calculated based on the similarity between the attributes of each node in the path (e.g., structural parameter nodes, performance index nodes, specification nodes, etc.) and the design fundamental parameters. Similarity can be measured using Euclidean distance, cosine similarity, or semantic embedding vector distance. Edge type matching is based on whether the edge relationships in the path conform to a preset design logical order (e.g., overall parameters → structural parameters → performance indexes → specification clauses). Edges that conform to the logical order are assigned higher scores, while edges that are semantically inconsistent or logically conflicting are assigned lower scores. Relevance to design fundamental parameters is calculated based on the degree of matching between the overall design semantics reflected by the path and the design fundamental parameters. For example, whether the performance parameters involved in the path are related to the target performance index, and whether the structural parameters belong to a common arrangement form of the target ship type. The above scores are weighted and summed according to preset weights to form the path consistency score for each candidate path.

[0028] Set a path consistency threshold and filter paths with scores greater than the threshold. The nodes involved constitute the target node set. The path consistency threshold can be set uniformly during the model training phase or dynamically adjusted according to the task type.

[0029] The target node set and its corresponding edge information are combined to form a task-related subgraph. The edge relationships between all nodes in the target node set are extracted from the original knowledge graph, including direct constraints, performance coupling relationships and canonical reference relationships between nodes, and a task-related subgraph with complete structure, performance and canonical semantic chains is constructed.

[0030] Based on the task-related subgraph, task constraints are extracted, including: extracting the edge set from the task-related subgraph, whereby the edge set includes logical relationships, performance constraint relationships, and specification relationships; traversing paths along the design logical relationships, performance constraint relationships, and specification relationships to each node in the task-related subgraph to form task constraints describing the design task constraint boundaries. The task constraints are output in a structured form for subsequent fusion and reasoning processes with the target performance index vector.

[0031] Based on the target performance indicators, a target indicator semantic vector is generated, including: converting the category, value, or level information of each target performance indicator into a structured vector representation, wherein the structured vector is used to describe the computable attributes of each indicator; and decomposing the target performance indicators into attributes, including indicator name, indicator category (such as stability, propulsion efficiency, seakeeping, structural strength, speed, etc.), indicator unit (such as m, kN, °, s, m / s), numerical range (such as "≥20", "5~8", "not greater than 0.35"), and level information (such as "Class B", "Category II of the Navigation Area", etc.).

[0032] The textual descriptions of the target performance indicators are semantically encoded and mapped to a high-dimensional semantic vector space to form textual semantic vectors. Each target performance indicator typically has a natural language description, such as "requires strong seakeeping," "speed not less than 20 knots," and "longitudinal strength meets specifications." A pre-trained language model can be used for this process.

[0033] The structured vector and text semantic vector are fused to generate a preliminary performance index vector. Vector concatenation directly connects the structured vector and text semantic vector along the feature dimension. The preliminary performance index vector is adjusted based on the correlation between target performance indicators to generate a comprehensive target index vector. The correlation between different performance indicators is calculated, including positive coupling (e.g., seakeeping and stability) and negative coupling (e.g., speed and energy consumption). A performance correlation matrix is ​​constructed to represent the coupling strength between indicators. The preliminary performance index vector is input into the performance correlation matrix for correlation propagation to strengthen the semantic consistency of related indicators. Vector constraints are applied to indicators with conflicting relationships (e.g., "drag reduction" and "increased displacement") to balance conflicting semantics. Graph convolutional networks, attention networks, or correlation propagation algorithms are used to further enhance the global consistency between different indicators. The correlation-adjusted vector accurately reflects the overall optimization direction of the target performance indicators and their internal relationships, forming the final comprehensive target index vector.

[0034] A graph neural network encoding model is used to learn the graph structure representation of the task-related subgraph and fuse it with the semantic vector of the target index to generate a comprehensive design reference vector. This includes: initializing the node and edge features of the task-related subgraph to form a graph structure feature matrix; combining the node feature vectors to form a node feature matrix and combining the edge feature vectors to form an edge feature matrix, thereby forming a complete graph structure feature matrix, which provides a data foundation for the graph neural network input.

[0035] The graph structure feature matrix is ​​input into the graph neural network encoding model. The graph attention mechanism is used to learn the structural representation of the task-related subgraphs and generate task subgraph embedding vectors. The node representations after being encoded by the graph neural network can be integrated into task subgraph embedding vectors through global pooling (such as mean pooling, max pooling or weighted pooling). These vectors can reflect the overall structural features, constraints and semantic information of the task-related subgraphs.

[0036] A linear transformation is performed on the target indicator semantic vector to match the dimension of the target indicator semantic vector with the dimension of the task subgraph embedding vector, thus obtaining the transformed target indicator semantic vector, which facilitates subsequent vector fusion.

[0037] The task subgraph embedding vector and the transformed target indicator semantic vector are fused to generate a comprehensive feature vector. This process includes: concatenating the task subgraph embedding vector and the transformed target indicator semantic vector to form a preliminary fusion vector; directly concatenating the task subgraph embedding vector and the transformed target indicator semantic vector along the feature dimensions to form a preliminary fusion vector, which simultaneously contains graph structure information and target indicator semantic information; weighting each dimension of the preliminary fusion vector based on an attention mechanism to obtain a weighted fusion vector; applying attention weights to each dimension of the preliminary fusion vector, assigning different weights according to the importance of each dimension to the design reference recommendation task, which can be set empirically to amplify feature dimensions with greater influence on design decisions and moderately suppress feature dimensions with lower weights; and normalizing the weighted fusion vector to generate a comprehensive feature vector, eliminating the influence of numerical scale differences on subsequent similarity calculations or optimization algorithms.

[0038] Based on the comprehensive design reference vector and the task constraints, historical cases in the knowledge graph are matched to obtain design reference recommendation results. This includes: constructing case feature vectors for each historical case in the knowledge graph; filtering out historical cases that do not meet the task constraints based on preset constraint consistency judgment rules; and setting constraint consistency judgment rules based on the task constraints extracted from the task-related subgraph to perform preliminary screening of all historical cases. This screening step is used to eliminate cases that conflict with or do not meet the design task in terms of size, material, performance constraints, or specification requirements, thereby ensuring that subsequent candidate cases all meet the basic design constraints.

[0039] The similarity between the case feature vectors of the remaining historical cases and the comprehensive design reference vector is calculated. The similarity calculation can use vector cosine similarity, Euclidean distance or other measurement methods applicable to multidimensional feature spaces to quantify the degree of matching between candidate historical cases and design tasks.

[0040] Historical cases are sorted according to the calculated similarity values, and those with similarity values ​​higher than a preset threshold are selected to form a candidate design reference set; only historical cases with similarity values ​​higher than the threshold are selected to enter the candidate design reference set.

[0041] The candidate design reference set is matched with the received design task to obtain design reference recommendation results. Further matching analysis is performed on the candidate design reference set and the received design task, reverting to the original received design task. Based on the target performance indicators of the received design task, the deviations of the candidate cases on each performance indicator are quantitatively evaluated. Through weighted calculation or multi-objective evaluation methods, the adaptability of each candidate case to the task objective is determined. The design reference recommendation result is the historical case with the best adaptability.

[0042] This invention constructs a multi-dimensional ship knowledge graph to systematically manage fundamental parameters, performance indicators, constraints, and historical design cases in ship design. This achieves structured, searchable, and associative design information, solving the problems of scattered information and low retrieval efficiency in traditional ship design. A conditional retrieval strategy locates the set of nodes matching the fundamental design parameters within the knowledge graph and constructs a task-related subgraph. Path traversal and extraction are performed on design logical relationships, performance constraint relationships, and specification relationships to efficiently obtain design task constraints. Based on this, target performance indicators are transformed into structured and semantic vectors, and a comprehensive target indicator vector is generated by combining the correlations between indicators, providing a quantitative basis for design optimization. A graph neural network is used to learn the structural representation of the task-related subgraph and fuse it with the target indicator semantic vector to generate a comprehensive design reference vector. This vector can intelligently match historical design cases in the knowledge graph, providing highly relevant design reference results and improving the efficiency and accuracy of ship design. Meanwhile, this method can be adapted to ship design tasks of different types, sizes and operating sea state levels, supports multi-objective optimization, has good scalability and long-term adaptability, and provides intelligent and precise decision support for the ship design process.

[0043] Example 2: A knowledge graph-based intelligent recommendation system for ship design reference data, comprising: a first construction module for constructing a multi-dimensional ship knowledge graph; a parsing module for parsing received design tasks to obtain basic design parameters and target performance indicators, wherein the basic design parameters define the design space and the target performance indicators define the design optimization direction; a second construction module for locating a set of nodes matching the basic design parameters in the knowledge graph using a conditional retrieval strategy to construct a task-related subgraph; an extraction module for extracting task constraints based on the task-related subgraph; a first generation module for generating a target indicator semantic vector based on the target performance indicators; a second generation module for learning the graph structure representation of the task-related subgraph using a graph neural network encoding model and fusing features with the target indicator semantic vector to generate a comprehensive design reference vector; and an acquisition module for matching historical cases in the knowledge graph based on the comprehensive design reference vector and the task constraints to obtain design reference recommendation results.

[0044] It should be understood that the specific embodiments described above are merely illustrative or explanatory of the principles of the invention and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of the invention should be included within the protection scope of the invention. Furthermore, the appended claims are intended to cover all variations and modifications falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.

[0045] The present invention has been described above with reference to embodiments thereof. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. The scope of the invention is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

[0046] Although embodiments of the present invention have been described in detail, it should be understood that various changes, substitutions, and modifications can be made to the embodiments of the present invention without departing from the spirit and scope of the invention.

[0047] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A knowledge graph-based intelligent recommendation method for ship design reference data, characterized in that, include: Constructing a multi-dimensional ship knowledge graph; The received design task is parsed to obtain basic design parameters and target performance indicators. The basic design parameters are used to define the design space, and the target performance indicators are used to define the design optimization direction. A conditional retrieval strategy is used to locate the set of nodes in the knowledge graph that match the design basic parameters, and a task-related subgraph is constructed; based on the task-related subgraph, task constraints are extracted. Based on the target performance index, a target index semantic vector is generated; a graph neural network encoding model is used to learn the graph structure representation of the task-related subgraph, and the feature is fused with the target index semantic vector to generate a comprehensive design reference vector; based on the comprehensive design reference vector and the task constraints, historical cases in the knowledge graph are matched to obtain design reference recommendation results.

2. The intelligent recommendation method for ship design reference data based on knowledge graphs according to claim 1, characterized in that, The knowledge graph uses basic design parameters as nodes and constraints as edges. The constraints include design logical relationships, performance constraints, and specification relationships. The process of parsing the received design task to obtain basic design parameters and target performance indicators includes: dividing the design task to obtain a basic design information part and a performance requirement description part; extracting features from the basic design information part to obtain basic design parameters, which include: ship type, dimensional parameters, displacement, structural layout form, and sea state level; and performing semantic parsing and keyword extraction on the performance requirement description part to identify target performance indicators, which include: stability, speed, resistance, structural strength, economy, and safety.

3. The intelligent recommendation method for ship design reference data based on knowledge graphs according to claim 2, characterized in that, The step of locating the set of nodes in the knowledge graph that match the design fundamental parameters using a conditional retrieval strategy and constructing a task-related subgraph includes: matching the design fundamental parameters with each node in the knowledge graph based on at least one of ship type, dimensional parameters, displacement, structural layout, and sea state level to obtain a first candidate node set; performing multi-hop expansion starting from the first candidate node set to obtain a second candidate node set associated with the first candidate node set; and performing path consistency detection on the first and second candidate node sets to obtain a target node set and constructing a task-related subgraph.

4. The intelligent recommendation method for ship design reference data based on knowledge graphs according to claim 3, characterized in that, The step of performing path consistency detection on the first candidate node set and the second candidate node set to obtain the target node set and construct a task-related subgraph includes: traversing paths along the edges between nodes in the first and second candidate node sets to generate a set of candidate paths between nodes; calculating the path consistency score for each candidate path, where the consistency score is based on the similarity of node attributes, edge type matching, and correlation with design parameters; setting a path consistency threshold, filtering paths with scores greater than the threshold, and the nodes involved constitute the target node set; and combining the target node set and its corresponding edge information to form a task-related subgraph.

5. The intelligent recommendation method for ship design reference data based on knowledge graphs according to claim 4, characterized in that, The step of extracting task constraints based on the task-related subgraph includes: extracting the edge set in the task-related subgraph, the edge set including logical relationships, performance constraint relationships and specification relationships; and performing path traversal on the nodes in the task-related subgraph along the design logical relationships, performance constraint relationships and specification relationships to form task constraints describing the design task constraint boundaries.

6. The intelligent recommendation method for ship design reference data based on knowledge graphs according to claim 5, characterized in that, The step of generating a target performance indicator semantic vector based on the target performance indicator includes: converting the category, value, or level information of each target performance indicator into a structured vector representation, wherein the structured vector is used to describe the computable attributes of each indicator; semantically encoding the text description of the target performance indicator and mapping it to a high-dimensional semantic vector space to form a text semantic vector; fusing the structured vector and the text semantic vector to generate a preliminary performance indicator vector; and adjusting the preliminary performance indicator vector based on the correlation between the target performance indicators to generate a comprehensive target performance indicator vector.

7. The intelligent recommendation method for ship design reference data based on knowledge graphs according to claim 6, characterized in that, The step of using a graph neural network encoding model to learn the graph structure representation of task-related subgraphs and fusing it with the target indicator semantic vector to generate a comprehensive design reference vector includes: initializing node and edge features of the task-related subgraphs to form a graph structure feature matrix; inputting the graph structure feature matrix into the graph neural network encoding model and learning the structure representation of the task-related subgraphs through a graph attention mechanism to generate task subgraph embedding vectors; performing a linear transformation on the target indicator semantic vectors to match the dimension of the target indicator semantic vectors with the dimension of the task subgraph embedding vectors to obtain the transformed target indicator semantic vectors; and fusing the task subgraph embedding vectors with the transformed target indicator semantic vectors to generate a comprehensive feature vector.

8. The intelligent recommendation method for ship design reference data based on knowledge graphs according to claim 7, characterized in that, The task subgraph embedding vector and the transformed target indicator semantic vector are fused to generate a comprehensive feature vector, including: concatenating the task subgraph embedding vector and the transformed target indicator semantic vector to form a preliminary fusion vector; weighting each dimension of the preliminary fusion vector based on an attention mechanism to obtain a weighted fusion vector; and normalizing the weighted fusion vector to generate a comprehensive feature vector.

9. The intelligent recommendation method for ship design reference data based on knowledge graphs according to claim 8, characterized in that, Based on the comprehensive design reference vector and the task constraints, historical cases in the knowledge graph are matched to obtain design reference recommendation results. This includes: constructing case feature vectors for each historical case in the knowledge graph; filtering out historical cases that do not meet the task constraints based on a preset constraint consistency judgment rule; calculating the similarity between the case feature vectors of the remaining historical cases and the comprehensive design reference vector; sorting the historical cases according to the calculated similarity values, and selecting historical cases with similarity higher than a preset threshold to form a candidate design reference set; and matching the candidate design reference set with the received design task to obtain design reference recommendation results.

10. A knowledge graph-based intelligent recommendation system for ship design reference data, characterized in that, include: First construction module: Constructing a multi-dimensional ship knowledge graph; Parsing module: Parses the received design task to obtain basic design parameters and target performance indicators. The basic design parameters are used to define the design space, and the target performance indicators are used to define the design optimization direction. The second construction module: uses a conditional retrieval strategy to locate the set of nodes in the knowledge graph that match the design basic parameters, and constructs a task-related subgraph; the extraction module: extracts task constraints based on the task-related subgraph; The first generation module generates a target indicator semantic vector based on the target performance indicator; the second generation module uses a graph neural network encoding model to learn the graph structure representation of the task-related subgraphs and fuses the features with the target indicator semantic vector to generate a comprehensive design reference vector; the obtaining module matches historical cases in the knowledge graph based on the comprehensive design reference vector and the task constraints to obtain design reference recommendation results.

Images