Intelligent Recommendation System for Engineering Design Parameters Based on Knowledge Graph

By constructing a parameter-level knowledge graph in the marine engineering design collaboration platform, the problems of incomplete parameter recommendation and invisible dependency links in the existing technology are solved, and efficient and accurate parameter recommendation is achieved.

CN123115528APending Publication Date: 2026-11-17BEIJING YUNKE KAICHUANG INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611483655.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-22
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing marine engineering design collaboration platforms, the design parameter recommendation methods cannot automatically extract parameter-level calculation dependencies, resulting in incomplete results, invisible calculation dependency links, missing cross-disciplinary connection information, and time-consuming manual searches.

Method used

A parameter-level recommendation system is built based on knowledge graphs. It generates a graph with design files as nodes and computational dependencies between parameters as edges. The path search module searches for reachable paths from the starting parameters to the target parameters and returns intermediate parameter nodes and computational dependencies.

Benefits of technology

It enables rapid location of parameter transmission links, reduces cross-disciplinary search workload, improves parameter recommendation efficiency and accuracy, and ensures the completeness and visibility of recommendation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115528A_ABST
    Figure CN123115528A_ABST
Patent Text Reader

Abstract

This invention discloses an intelligent recommendation system for engineering design parameters based on knowledge graphs, belonging to the field of intelligent recommendation technology for engineering design parameters. The system includes a topology construction module, a graph generation module, an intent parsing module, a path search module, and a recommendation output module. It generates a data flow topology graph by parsing the data dependencies between multiple disciplines in a marine engineering design collaboration platform, establishes a knowledge graph with parameter entities as nodes and computational dependencies between parameters as edges, and stores it in a graph database. It receives design query statements input by users, extracts the target equipment type and the name of the parameter to be recommended, locates the starting parameter node in the graph database, searches for reachable paths along the computational dependencies between parameters, selects the path with the fewest edges as the recommended path, and extracts the parameter values ​​of intermediate parameter nodes and the target node parameter values, encapsulating them into a recommendation result dataset for output. This system can return cross-disciplinary parameter calculation chains, reducing manual searching and missing parameter links.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent recommendation technology for engineering design parameters, specifically an intelligent recommendation system for engineering design parameters based on knowledge graphs. Background Technology

[0002] In marine engineering design collaboration platforms, data transfer and computational dependencies exist between multidisciplinary design documents. Design parameters related to the target equipment type need to be derived step-by-step from upstream input parameters and intermediate parameters. Current engineering design parameters are often recommended to be found using keyword matching, reuse of parameters from historical projects, or manual searching based on experience. Keyword matching performs word-based similarity searches on design query statements and document content, making it difficult to identify the data transfer direction between design documents, and the returned results are often document fragments or isolated parameter values. Reusing parameters from historical projects stores parameter snapshots by equipment type or professional category, but does not record the order of calculation and derivation between parameters, making it difficult to adapt to changes in professional interfaces across different projects. Manual searching relies on designers' familiarity with upstream and downstream professional interfaces and document dependencies, and is time-consuming in scenarios with long cross-disciplinary parameter chains or parallel design.

[0003] The header information of design documents in the marine engineering design collaboration platform records upstream input file identifiers and downstream output file identifiers. The document content contains parameter fields of different categories. Existing parameter recommendation methods do not associate file-level data transfer relationships with the set of parameter fields within the file, and cannot automatically obtain the directed computational relationships between common transfer parameters and independent parameters. Existing technologies lack a mechanism to automatically extract parameter-level computational dependencies from the data flow of design documents. The granularity of knowledge representation remains at the file or document level, and the directed computational relationships between common transfer parameters and independent parameters within the same design document are not expressed as a queryable graph structure. After the user inputs the target equipment type and the name of the parameter to be recommended, existing retrieval methods cannot obtain the reachable path step by step along the parameter computational dependencies from the starting parameter, nor can they provide the intermediate parameter nodes and their parameter values ​​required to reach the target parameter in the recommendation results. Parameter recommendation suffers from incomplete results, invisible computational dependency links, and missing cross-disciplinary connection information. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent recommendation system for engineering design parameters based on knowledge graphs. By generating a parameter-level knowledge graph from the data dependencies of design documents in a marine engineering design collaboration platform, and searching for reachable paths from the starting parameters to the parameters to be recommended in the graph database, the system returns recommendation results containing intermediate parameter nodes and parameter calculation dependencies. This overcomes the problems of insufficient parameter recommendation granularity and invisible calculation dependency links in the prior art.

[0005] To achieve the above objectives, this invention provides the following technical solution: This invention provides an intelligent recommendation system for engineering design parameters based on a knowledge graph, including a topology construction module, a graph generation module, an intent parsing module, a path search module, and a recommendation output module. The topology construction module is used to parse the data dependencies between multiple disciplines in a marine engineering design collaboration platform, generating a data flow topology graph with design files as nodes and directed edges based on the data transmission direction. The graph generation module is used to establish a knowledge graph with parameter entities as nodes and computational dependencies between parameters as edges, based on the design file type and parameter categories of each node in the data flow topology graph, and stores it in a graph database. The intent parsing module is used to receive user-input design query statements, extract the target equipment type and the name of the parameter to be recommended through word segmentation and dependency parsing analysis, and group the target equipment type and the name of the parameter to be recommended. The system uses a combination of query criteria and path search modules. The path search module locates the starting parameter node matching the target device type in the graph database. Starting from this node, it searches for all reachable paths connecting to the target node corresponding to the parameter name to be recommended, following the edges indicating computational dependencies between parameters. Each reachable path contains at least one intermediate parameter node and edges connecting all nodes. The recommendation output module selects the path with the fewest edges from all reachable paths as the recommended path, extracts the parameter values ​​corresponding to each intermediate parameter node and the parameter values ​​of the target node from the recommended path, encapsulates these into a recommendation result dataset, and outputs it to the front-end interface of the design collaboration platform. This system utilizes a knowledge graph to express computational dependencies between parameters, enabling rapid location of parameter transmission links from massive amounts of engineering design data. This reduces the workload of designers searching for parameters across disciplines and performing trial calculations, improving the efficiency and accuracy of parameter recommendations.

[0006] Furthermore, when resolving data dependencies, the topology construction module reads the header information of each design file in the design collaboration platform. This header information includes upstream input file identifiers and downstream output file identifiers. The design file corresponding to the upstream input file identifier is used as the starting node, and the design file corresponding to the downstream output file identifier is used as the ending node. A first directed edge is established between the starting and ending nodes. The parameter fields commonly contained in the two design files connected by the first directed edge are used as the edge attributes of the first directed edge and stored in the edge attribute set of the data flow topology graph. This configuration allows the data flow topology graph to carry the actual parameter information transferred between files, providing accurate input for subsequent parameter-level knowledge graph construction.

[0007] As a technical solution of this invention, when establishing computational dependencies between parameters, the graph generation module extracts the upstream and downstream design files corresponding to each directed edge in the data flow topology graph; reads the first set of parameter fields contained in the upstream design file and the second set of parameter fields contained in the downstream design file; compares the first and second set of parameter fields, and marks the common parameter fields that exist in both sets as transit parameters; for independent parameter fields in the second set that do not belong to the first set of parameter fields, a second directed edge is established between the downstream parameter node corresponding to the independent parameter field and the transit parameter node corresponding to the transit parameter, pointing from the transit parameter node to the downstream parameter node, as a computational dependency between parameters. This construction method treats common parameters as transit links, making the computational dependency between newly added downstream parameters and source parameters explicit, enabling the knowledge graph to express parameter derivation relationships and improving the targeting of subsequent path searches.

[0008] Furthermore, when the path search module searches for all reachable paths, it uses the starting parameter node as the current traversal node and queries the graph database for all downstream nodes adjacent to the current traversal node. Each downstream node is connected to the current traversal node by a second directed edge. The downstream node is added to the candidate queue, and the path length from the starting parameter node to that downstream node is recorded. The node with the shortest path length is selected from the candidate queue and used as the new current traversal node. This query and addition process is repeated until the parameter name of the current traversal node matches the parameter name to be recommended. The traversed node sequence and edge sequence are recorded as a reachable path. When the candidate queue is empty and the target node has not been reached, it is determined that no reachable path exists. This search process prioritizes expanding along the shortest path, which can reduce invalid traversals and improve the efficiency of reachable path search when the number of parameter nodes is large.

[0009] Preferably, when multiple nodes with the same path length exist in the candidate queue, the path search module prioritizes selecting the node with the highest semantic similarity between its name and the name of the parameter to be recommended as the new current traversal node. By introducing semantic similarity to distinguish candidate nodes of the same length, the rationality of path selection can be further improved, avoiding the recommended path deviating from the user's intent due to relying solely on graph structure.

[0010] Preferably, when the candidate queue is empty and the target node has not been reached, the path search module expands the starting parameter node along the direction of the second directed edge to include all subgraph nodes within two layers as a backup recommendation set, and marks the nodes in the backup recommendation set that are directly connected to the parameter name to be recommended as approximate target nodes. In this way, even if there is no perfectly matching reachable path, an approximate recommendation result can be given based on the local subgraph, enhancing the applicability of the system under conditions where the parameter names do not perfectly match or the graph is incomplete.

[0011] Furthermore, when the intent parsing module extracts the target device type and the name of the parameter to be recommended, it performs Chinese word segmentation on the designed query statement to obtain a word sequence, and labels each word in the word sequence with a part-of-speech tag. From the words labeled with noun parts of speech, words that match the preset device name dictionary are selected as the target device type. Based on the position of the target device type in the word sequence, consecutive word combinations that follow the target device type and match the preset parameter name dictionary are selected as the name of the parameter to be recommended. When there are multiple words in the word sequence that match the preset device name dictionary, the word that appears last in the order of appearance is selected as the target device type. This can accurately separate device information and parameter information from natural language queries, reducing the probability of recommendation link deviation caused by incorrect query condition construction.

[0012] As a technical solution of this invention, when the recommendation output module encapsulates the recommendation result dataset, it sequentially reads the parameter values ​​corresponding to each intermediate parameter node and the calculation logic identifier in the edge attributes of the edges connected to each intermediate parameter node, according to the order from the starting parameter node to the target node in the recommendation path. Using the calculation logic identifier as the basis for parameter conversion, it performs unit unification processing on the parameter values ​​of the intermediate parameter nodes to obtain standardized parameter values. The standardized parameter values ​​are associated with the node names of the intermediate parameter nodes and stored in the first data area, the parameter values ​​of the target nodes are stored in the second data area, and the edge sequence of the recommendation path is stored in the third data area. The first, second, and third data areas are then merged to form the recommendation result dataset. By encapsulating the parameter values, target values, and path information in separate areas, the front-end interface can completely display the parameter recommendation chain and its basis, facilitating verification and reuse by designers.

[0013] Furthermore, when the recommended output module performs unit unification processing on the parameter values ​​of intermediate parameter nodes, it reads the source unit identifier and target unit identifier corresponding to the calculation logic identifier from the edge attributes; it calls the preset unit conversion mapping table to query the conversion coefficient that matches the source unit identifier and target unit identifier; it multiplies the parameter value of the intermediate parameter node by the conversion coefficient to obtain the standardized parameter value, and uses the target unit identifier as the unit label of the standardized parameter value. This eliminates unit differences between different professional design documents, ensures that the recommended parameter values ​​are output under a unified unit system, and improves the usability of the recommendation results.

[0014] Furthermore, when the intent parsing module combines the target device type and the name of the parameter to be recommended into query conditions, it reads the project stage tags carried by each design file node in the data flow topology diagram from the graph database. These project stage tags include basic design stage tags and detailed design stage tags. When the design query statement contains stage keywords corresponding to the basic design stage tags, the name of the parameter to be recommended is associated with the basic design stage tags to generate the first query sub-condition. When the design query statement contains stage keywords corresponding to the detailed design stage tags, the name of the parameter to be recommended is associated with the detailed design stage tags to generate the second query sub-condition. The target device type is then logically ANDed with the first and second query sub-conditions to obtain the query conditions. By introducing the project stage dimension, the values ​​and dependencies of the same device parameter in different design stages can be distinguished, avoiding confusion between parameters in the basic design stage and the detailed design stage, and improving the consistency of the recommendation results with the current design stage.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0016] Starting with the data dependencies in the design documents of the marine engineering design collaboration platform, the input-output relationships between design documents from various disciplines are first analyzed to generate a data flow topology graph with design documents as nodes and directed edges representing data transfer directions. Then, based on the upstream and downstream design documents corresponding to each directed edge in the data flow topology graph, the parameter field sets contained in both are read. Common parameter fields are marked as transferred parameters, and directed edges are established from the transferred parameter nodes to the downstream parameter nodes for independent parameter fields in the downstream design documents that do not belong to the upstream parameter field set, forming parameter entity nodes and computational dependency edges between parameters. This parameter-level graph can express the computational derivation relationship between common transferred parameters and downstream independent parameters in the same design document, refining cross-disciplinary data transfer into queryable parameter dependency paths. When recommending the starting parameter node that matches the target equipment type, the computational dependency order can be obtained along the directed edges between parameter entities, instead of returning isolated single parameter values ​​or lists of similar files.

[0017] In the parameter-level knowledge graph, starting from the initial parameter node matching the target device type, all reachable paths connecting to the target node corresponding to the parameter name to be recommended are searched along the edges of the computational dependencies between parameters. During traversal, all downstream nodes adjacent to the currently traversed node are queried, and each downstream node is added to a candidate queue with its path length from the initial parameter node to that downstream node recorded. The node with the shortest path length is then selected from the candidate queue as the new current traversed node, until the parameter name of the current traversed node matches the parameter name to be recommended. This search strategy can select the recommended path with the fewest edges from multiple reachable paths, avoiding the inclusion of long links unrelated to the calculation of the target parameter in the recommendation results. The parameter values ​​corresponding to each intermediate parameter node in the recommended path and the parameter values ​​of the target node are extracted and encapsulated and output together in the order from the initial parameter node to the target node. The recommendation results include the parameter calculation dependency sequence and the source of parameters at each level, rather than just providing the final parameter values, reducing the information loss caused by cross-disciplinary manual searches and file-level matching. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a schematic diagram of the module structure of an intelligent recommendation system for engineering design parameters based on knowledge graphs;

[0020] Figure 2 It is a flowchart for constructing a data flow topology diagram;

[0021] Figure 3 This is a flowchart of the parameter dependency graph generation process;

[0022] Figure 4 It involves designing a probability density distribution diagram of the word sequence length in a query statement;

[0023] Figure 5 It is a probability density distribution diagram corresponding to the intermediate parameter node parameter values ​​and the calculation logic identifier. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] See Figure 1 This invention provides an intelligent recommendation system for engineering design parameters based on a knowledge graph, comprising a topology construction module, a graph generation module, an intent parsing module, a path search module, and a recommendation output module. The topology construction module parses the data dependencies between multiple disciplines in a marine engineering design collaboration platform, generating a data flow topology graph with design files as nodes and directed edges along the data transfer direction. The graph generation module, based on the design file type and parameter categories of each node in the data flow topology graph, establishes a knowledge graph with parameter entities as nodes and computational dependencies between parameters as edges, and stores it in a graph database. The intent parsing module receives user-input design query statements, extracts the target equipment type and the name of the parameter to be recommended through word segmentation and dependency parsing, and combines the target equipment type and the name of the parameter to be recommended as query conditions. The path search module locates the starting parameter node matching the target equipment type in the graph database, and searches for all reachable paths connecting to the target node corresponding to the name of the parameter to be recommended along the direction of the edges of computational dependencies between parameters, with each reachable path containing at least one intermediate parameter node and edges connecting all nodes. The recommendation output module selects the path with the fewest edges from all reachable paths as the recommended path, extracts the parameter values ​​of each intermediate parameter node and the parameter values ​​of the target node in the recommended path, encapsulates them into a recommendation result dataset, and outputs it to the front-end interface of the design collaboration platform.

[0026] In specific implementation, please refer to Figure 2 The topology building module retrieves a list of registered design files from the design collaboration platform. Each design file in the platform records an upstream input file identifier and a downstream output file identifier in its header information. This header information is stored in the design file's metadata area, with the upstream input file identifier and downstream output file identifier corresponding to structured fields. The topology building module calls the design collaboration platform's file parsing interface, processes each design file according to the order in the design file list, reads the header information of each design file, and extracts the values ​​of the upstream input file identifier field and the downstream output file identifier field.

[0027] The topology construction module uses the design file corresponding to the value of the upstream input file identifier field as the starting node and the design file corresponding to the value of the downstream output file identifier field as the ending node. The topology construction module queries the starting node identifier and ending node identifier in the data flow topology graph. If the starting node identifier does not exist in the data flow topology graph, the topology construction module creates the starting node; if the ending node identifier does not exist in the data flow topology graph, the topology construction module creates the ending node. The data flow topology graph uses an adjacency list to store the design file nodes and the first directed edge. The topology construction module establishes a first directed edge between the starting node and the ending node, with the starting end of the first directed edge connected to the starting node and the ending end connected to the ending node.

[0028] The topology construction module reads the parameter field set contained in the design file corresponding to the starting node and the parameter field set contained in the design file corresponding to the ending node. The module performs an intersection operation on these two sets of parameter fields to obtain a common parameter field set. Each parameter field in the common parameter field set exists in both the design files corresponding to the starting and ending nodes. The module stores this common parameter field set as an edge attribute of the first directed edge in the edge attribute set of the data flow topology graph. Each record in the edge attribute set of the data flow topology graph contains the first directed edge identifier, the starting node identifier, the ending node identifier, and a list of common parameter fields.

[0029] Optionally, the upstream input file identifier and downstream output file identifier in the file header information use the unique number of the design file. When the field value of the upstream input file identifier field is empty, the topology building module does not create the corresponding first directed edge; when the field value of the downstream output file identifier field is empty, the topology building module does not create the corresponding first directed edge. When the set of parameter fields contained in the design file corresponding to the starting node and the set of parameter fields contained in the design file corresponding to the ending node do not intersect, the first directed edge is still created, and the list of common parameter fields in the edge attributes of the first directed edge is empty.

[0030] In specific implementation, please refer to Figure 3 The graph generation module extracts the upstream and downstream design file identifiers corresponding to each first directed edge from the data flow topology graph. Each record in the edge attribute set of the data flow topology graph contains the first directed edge identifier, the starting node identifier, the ending node identifier, and a list of common parameter fields. The graph generation module reads each record in the edge attribute set, designating the design file corresponding to the starting node identifier as the upstream design file and the design file corresponding to the ending node identifier as the downstream design file.

[0031] The graph generation module reads the first set of parameter fields from the upstream design file. This first set consists of all parameter field names appearing in the upstream design file. The graph generation module then reads the second set of parameter fields from the downstream design file. This second set also consists of all parameter field names appearing in the downstream design file. When reading the first and second parameter field sets, the graph generation module calls the file content parsing interface of the design collaboration platform to extract the parameter fields according to their names and stores the extracted parameter field names in a list format.

[0032] The knowledge graph generation module compares the first set of parameter fields with the second set of parameter fields. The comparison operation uses set intersection, marking parameter fields that exist in both sets as transmitted parameters. The knowledge graph generation module records the corresponding parameter entity nodes as transmitted parameter nodes. Transmitted parameter nodes correspond to parameter entity nodes in the knowledge graph, and these entity nodes are identified by the parameter field names.

[0033] The graph generation module marks parameter fields in the second parameter field set that do not belong to the first parameter field set. The difference between the second and first parameter field sets is taken as the independent parameter field set. Each parameter field in the independent parameter field set appears in the downstream design file but not in the upstream design file. The graph generation module records the parameter entity node corresponding to each independent parameter field in the independent parameter field set as a downstream parameter node.

[0034] For each independent parameter field in the set of independent parameter fields, the graph generation module establishes a second directed edge between the transmitting parameter node and the downstream parameter node. This second directed edge, pointing from the transmitting parameter node to the downstream parameter node, indicates a computational dependency between the transmitted parameter corresponding to the transmitting parameter node and the independent parameter field corresponding to the downstream parameter node. The edge attributes of the second directed edge record the type of computational dependency between the parameters, using a fixed identifier.

[0035] In some embodiments, when the first parameter field set and the second parameter field set have multiple common parameter fields, the graph generation module establishes a transmission parameter node for each transmission parameter. When the second parameter field set has multiple independent parameter fields, the graph generation module establishes a downstream parameter node for each independent parameter field. For each downstream parameter node, the graph generation module establishes a second directed edge between the downstream parameter node and all transmission parameter nodes.

[0036] In some embodiments, when the first parameter field set and the second parameter field set do not have a common parameter field, the graph generation module does not mark the transmitted parameters, nor does it establish a second directed edge. When the second parameter field set does not have an independent parameter field, the graph generation module does not establish a second directed edge.

[0037] In some embodiments, the graph generation module writes the established parameter entity nodes and second directed edges into the graph database. The graph database uses a graph data structure to store the parameter entity nodes and second directed edges. Parameter entity nodes are distinguished by node type identifiers, and second directed edges are distinguished by edge type identifiers. The graph generation module writes the node identifier, parameter field name, and design file identifier of each parameter entity node into the graph database. The graph generation module writes the edge identifier, start node identifier, end node identifier, and parameter computational dependency type of each second directed edge into the graph database.

[0038] In practice, the path search module reads the target device type from the query conditions and the set of parameter entity nodes in the graph database. The parameter entity nodes store node identifiers, parameter names, device type identifiers, and design document identifiers in the graph database. The path search module queries the graph database for all parameter entity nodes whose "device type identifier" attribute matches the target device type, and uses these found parameter entity nodes as the starting parameter node.

[0039] The path search module uses the starting parameter node as the current traversal node. It queries the graph database for all downstream nodes adjacent to the current traversal node. An adjacent downstream node is a parameter entity node pointed to by the current traversal node via a second directed edge. Each second directed edge points from the current traversal node to an adjacent downstream node, and its edge attributes record the computational dependency type between parameters. The path search module adds all found adjacent downstream nodes to a candidate queue and records the path length from the starting parameter node to each adjacent downstream node. The path length is represented by the number of second directed edges traversed from the starting parameter node to the adjacent downstream node. When there is only one second directed edge from the starting parameter node to the adjacent downstream node, the path length is recorded as 1; when there are multiple second directed edges traversed from the starting parameter node to the adjacent downstream node, the path length is recorded as the total number of second directed edges traversed.

[0040] The path search module retrieves the node with the shortest path length from the candidate queue as the new current traversal node. When there is only one node with the shortest path length in the candidate queue, the module directly retrieves it as the new current traversal node. When there are multiple nodes with the same path length in the candidate queue, and that path length is the shortest path length in the queue, the module calculates the semantic similarity between the node name and the name of the parameter to be recommended for each of these nodes, and prioritizes the node with the highest semantic similarity as the new current traversal node. Semantic similarity is calculated using the formula:

[0041]

[0042] in, This represents the semantic similarity value between the node name and the name of the parameter to be recommended. subscript It is an abbreviation for semantic similarity. The value of is greater than 0 and less than or equal to 1; This represents the edit distance between the node name string and the parameter name string to be recommended. subscript It is an abbreviation for edit distance. The value of is a non-negative integer. Edit distance is defined as the minimum number of single-character edit operations required to transform one string into another. Single-character edit operations include inserting a character, deleting a character, and replacing a character. Edit distance is calculated using a dynamic programming algorithm. The dynamic programming algorithm constructs a two-dimensional matrix with the number of rows equal to the length of the node name string plus one and the number of columns equal to the length of the recommended parameter name string plus one. The value of is the number of rows in the two-dimensional matrix. Line 1 The element value of the column represents the first part of the node name string. The first character of the string is the name of the parameter to be recommended. Edit distance between characters, where This represents the row index of a two-dimensional matrix, with values ​​ranging from 0 to the length of the node name string. This represents the column index of the two-dimensional matrix, ranging from 0 to the length of the string containing the recommended parameter name. The dynamic programming algorithm fills the two-dimensional matrix from left to right and top to bottom. For each matrix element, the minimum of the following three candidate values ​​is used: the left element value plus 1, the top element value plus 1, and the top-left element value plus the replacement cost. The replacement cost is the value at the 1st position in the node name string. The character and the first character of the string of the parameter name to be recommended If all characters are the same, take 0; in the node name string, the first... The character and the first character of the string of the parameter name to be recommended If all characters are different, the value is 1. When multiple nodes with the same path length have the same semantic similarity value, the path search module selects the node with the first name in lexicographical order as the new current traversal node.

[0043] The path search module repeatedly performs the query and addition steps for the new current traversal node. It queries the graph database for all downstream nodes adjacent to the new current traversal node, adds these downstream nodes to the candidate queue, and records the path length from the starting parameter node to the downstream node. Each time, the path search module selects the node with the shortest path length from the candidate queue as the new current traversal node and checks if the parameter name of the current traversal node matches the parameter name to be recommended. When the parameter name of the current traversal node matches the parameter name to be recommended, the path search module stops searching and records the traversed node sequence and edge sequence as a reachable path. The node sequence consists of all parameter entity nodes traversed sequentially from the starting parameter node to the target node, and the edge sequence consists of the second directed edge between adjacent parameter entity nodes in the node sequence.

[0044] In some embodiments, the path search module maintains a set of visited nodes during the search process. When a node retrieved from the candidate queue already exists in the set of visited nodes, the path search module does not use that node as the new current traversal node, but instead retrieves the next node with the shortest path length from the candidate queue. The path search module adds each node that becomes the new current traversal node to the set of visited nodes. When the path search module reaches the target node, it backtracks to obtain the reachable path based on the node markers and edge connections in the set of visited nodes.

[0045] When the candidate queue is empty and the target node has not been reached, the path search module determines that no reachable path exists. The path search module obtains the starting parameter node and expands it along the direction of the second directed edge to include all subgraph nodes within two layers. Expanding within two layers means reaching the first-layer nodes from the starting parameter node by traversing the second directed edge in the positive direction, and reaching the second-layer nodes by continuing along the second directed edge in the positive direction from the first-layer nodes. The path search module adds all the expanded first-layer and second-layer nodes to the backup recommendation set. The path search module queries each node in the backup recommendation set. When a node in the backup recommendation set has a direct edge connection to the parameter entity node corresponding to the parameter name to be recommended, the node in the backup recommendation set with the direct edge connection is marked as an approximate target node. A direct edge connection means that a node in the backup recommendation set is directly connected to the parameter entity node corresponding to the parameter name to be recommended by a single second directed edge, without passing through other parameter entity nodes.

[0046] In practice, the intent parsing module receives the design query statement input by the user. The design query statement is in string form and is obtained from the front-end interface of the design collaboration platform. The intent parsing module stores the design query statement as a raw query string.

[0047] The intent parsing module performs Chinese word segmentation on the design query statement. The Chinese word segmentation operation uses the forward maximum matching algorithm, which employs a pre-defined word segmentation dictionary. This dictionary stores standard terms and commonly used parameter names in the field of marine engineering design. The execution process of the forward maximum matching algorithm is as follows: Starting from the first character position of the design query statement, the longest word in the pre-defined word segmentation dictionary is taken as the pre-defined maximum word length. A string fragment with a length equal to the pre-defined maximum word length is extracted from the current character position. This string fragment is then precisely matched against the pre-defined word segmentation dictionary. If the string fragment exists in the dictionary, it is segmented into a word, and the current character position is shifted forward by the length of that word. If the string fragment does not exist in the dictionary, one character is removed from the end of the fragment, and a new precise match is performed against the dictionary until the remaining string fragment has a length of 1. This process continues until the remaining string fragment has a length of 1, at which point it is segmented into a word. This extraction and matching process is repeated until all characters of the design query statement are segmented, resulting in a word sequence. Each word in the word sequence is arranged according to its order of appearance in the design query statement.

[0048] The intent parsing module labels each word in the word sequence with a part-of-speech tag. The part-of-speech tags use a pre-defined part-of-speech tagging set, which includes noun, verb, adjective, and other parts of speech. The intent parsing module matches each word in the word sequence against a pre-defined part-of-speech dictionary. If a word exists in the pre-defined part-of-speech dictionary, its corresponding part-of-speech tag from the dictionary is used as the word's part-of-speech tag; if a word does not exist in the dictionary, it is tagged as a noun. The pre-defined part-of-speech dictionary stores terms related to marine engineering design and their corresponding part-of-speech tags.

[0049] The intent parsing module selects words from those tagged with nouns as target equipment types that match a pre-defined equipment name dictionary. This dictionary stores all equipment type names in the marine engineering design field. The matching method is exact equality matching. The intent parsing module iterates through each word in the word sequence. If a word is tagged with a noun and is exactly equal to any equipment type name in the pre-defined dictionary, it is added to the candidate equipment type set. If there is only one word in the candidate equipment type set, that word is selected as the target equipment type. If there are multiple words in the candidate equipment type set, the word that appears last in the candidate set is selected as the target equipment type. "Last in the candidate set" refers to the word with the highest position index in the word sequence.

[0050] The intent parsing module uses the position of the target device type in the word sequence as a benchmark, selecting consecutive word combinations that follow the target device type and match a preset parameter name dictionary as the recommended parameter name. The intent parsing module locates the position index of the target device type in the word sequence and, starting from the first word after the target device type's position index, sequentially checks whether each word is completely equal to any parameter name in the preset parameter name dictionary. The preset parameter name dictionary stores all known parameter names in the field of marine engineering design. When the first word matching the preset parameter name dictionary is encountered, the module begins recording the recommended parameter name fragment, adding subsequent consecutive matching words to the fragment until the first word not matching the preset parameter name dictionary is encountered. All words in the recommended parameter name fragment are then directly concatenated in their original order to obtain the recommended parameter name. If no word matching the preset parameter name dictionary exists after the target device type, the recommended parameter name is set to null.

[0051] The intent parsing module retrieves project phase tags from the graph database, based on the design file nodes in the data flow topology diagram. In the graph database, each design file node uses its design file identifier as the primary key and contains a project phase tag attribute. The project phase tag attribute can be either a basic design phase tag or a detailed design phase tag. The basic design phase tag corresponds to the basic design phase in marine engineering design, while the detailed design phase tag corresponds to the detailed design phase. The intent parsing module calls the graph database query interface to retrieve the project phase tag attribute values ​​for all design file nodes.

[0052] The intent parsing module pre-sets keyword sets for the basic design stage and the detailed design stage. Keywords in the basic design stage keyword set are associated with tags for the basic design stage, and this set includes "basic design," "basic stage," and "preliminary design." Keywords in the detailed design stage keyword set are associated with tags for the detailed design stage, and this set includes "detailed design," "detailed stage," and "construction drawing design." The intent parsing module matches the design query against the basic design stage keyword set. When the query contains any keyword from this set, it associates the name of the parameter to be recommended with the basic design stage tag, generating a first query sub-condition. The data structure of this first sub-condition includes a parameter name field and a project stage tag field. The parameter name field contains the name of the parameter to be recommended, and the project stage tag field contains the basic design stage tag. The intent parsing module then matches the design query against the detailed design stage keyword set. When the query contains any keyword from this set, it associates the name of the parameter to be recommended with the detailed design stage tag, generating a second query sub-condition. The data structure for the second query sub-condition also includes a parameter name field and a project stage label field. The parameter name field contains the name of the parameter to be recommended, and the project stage label field contains the detailed design stage label.

[0053] When the design query statement does not contain any keyword from either the basic design phase keyword set or the detailed design phase keyword set, the intent parsing module will not generate either the first or second query sub-condition, but will only use the name of the parameter to be recommended as a query sub-condition without conditional phase restrictions.

[0054] The intent parsing module performs a logical AND operation between the target device type and the first and second query sub-conditions to obtain the query conditions. The logical AND operation is as follows: the query condition data structure includes three fields: target device type, parameter name, and project stage label. The target device type field's value is the target device type; the parameter name field's value is inherited from the parameter name field value in the corresponding query sub-condition; and the project stage label field's value is inherited from the project stage label field value in the corresponding query sub-condition. When the first query sub-condition exists, the intent parsing module generates one query condition, where the project stage label field's value is the basic design stage label. When the second query sub-condition exists, the intent parsing module generates another query condition, where the project stage label field's value is the detailed design stage label. When both exist, the intent parsing module generates two query conditions. When neither exists, the intent parsing module generates one query condition, where the project stage label field's value is null.

[0055] See Figure 4 In the graph, the horizontal axis represents the length of the word sequence, in units of word count, ranging from 3 to 25 words; the vertical axis represents the probability density, ranging from 0.000 to 0.185. The graph includes five probability density curves for different equipment types, corresponding to platform equipment type (solid blue line), pump equipment type (dotted red line), pipeline equipment type (dashed orange line), valve equipment type (dotted green line), and container equipment type (dotted purple line).

[0056] The curve for platform equipment type shows a clear peak between 6 and 11 words in length, with a peak probability density of approximately 0.185. This indicates that most parameter name word sequences related to platform equipment type are concentrated between 7 and 10 words in length, and that the word sequences for this type are relatively short and concentrated. The peak of the curve for pump equipment type is located in the range of 8 to 11 words, with a peak probability density of approximately 0.18. This indicates that the parameter name word sequences for pump equipment are slightly longer than those for platform equipment, and the word sequence length distribution is more concentrated. The peak of the curve for pipeline equipment type appears in the range of 9 to 12 words, with a peak probability density of approximately 0.17. This shows that the word sequence length for pipeline equipment is longer than that for platform and pump equipment, and exhibits a higher frequency around 10 words.

[0057] The curve for valve equipment type is relatively flat, with a peak probability density of approximately 0.135. The peak position is between words 11 and 14, indicating that the parameter name word sequence for valve equipment is relatively long and the word sequence length is quite dispersed, with multiple local peaks. The curve for container equipment type is similar in shape to that of valve equipment, with a peak probability density of approximately 0.14. The peak position is between words 12 and 15, indicating that the parameter name word sequence for container equipment type is generally longer and the word sequence is more complex.

[0058] In its implementation, the recommendation output module obtains the recommended path output by the path search module. The recommended path contains an ordered sequence of nodes and an ordered sequence of edges from the starting parameter node to the target node. In the ordered node sequence, the first node is the starting parameter node, the last node is the target node, and the remaining nodes are intermediate parameter nodes. In the ordered edge sequence, each edge is a second directed edge connecting two adjacent nodes, with the direction of the second directed edge pointing from the previous node to the next node. The recommendation output module processes each intermediate parameter node in the ordered node sequence sequentially, according to the order from the starting parameter node to the target node.

[0059] For each intermediate parameter node in the ordered node sequence, the recommendation output module reads the corresponding parameter value from the graph database. The parameter entity nodes in the graph database contain a node name attribute, a parameter value attribute, and a unit identifier attribute. The recommendation output module uses the node identifier of the intermediate parameter node as the query key to read the attribute value of the intermediate parameter node's parameter value from the graph database, and uses this read attribute value as the parameter value of the intermediate parameter node.

[0060] The recommendation output module reads the computation logic identifier from the edge attributes of the edges connected to intermediate parameter nodes. An edge connected to an intermediate parameter node refers to the second directed edge in the ordered edge sequence of the recommendation path that points to that intermediate parameter node; that is, the incoming edge of that intermediate parameter node. Each second directed edge in the ordered edge sequence contains a computation logic identifier field in its edge attributes. The computation logic identifier is a fixed identifier used to identify the parameter computation logic corresponding to that second directed edge. The recommendation output module reads the value of the computation logic identifier field from the edge attributes of the incoming edges of that intermediate parameter node.

[0061] The recommended output module retrieves the source unit identifier and target unit identifier based on the calculation logical identifier. The calculation logical identifier is associated with a unit conversion configuration table, where each record contains a calculation logical identifier field, a source unit identifier field, and a target unit identifier field. The recommended output module uses the calculation logical identifier as the query key to search the unit conversion configuration table for a record that exactly matches the calculation logical identifier, and then reads the source unit identifier field value and target unit identifier field value from that record.

[0062] The recommended output module calls a preset unit conversion mapping table to query the conversion factor that matches the source unit identifier and target unit identifier. The preset unit conversion mapping table contains a source unit identifier field, a target unit identifier field, and a conversion factor field. This table is stored in the system configuration database. The recommended output module uses the source unit identifier and target unit identifier as the join key to search the preset unit conversion mapping table for records where both the source unit identifier and target unit identifier values ​​are exactly the same, and then reads the conversion factor value from those records. If the source unit identifier and target unit identifier are exactly the same, the recommended output module directly sets the conversion factor to 1 and does not execute the query operation in the preset unit conversion mapping table.

[0063] The recommended output module multiplies the parameter values ​​of intermediate parameter nodes by a conversion factor to obtain standardized parameter values. The formula for calculating standardized parameter values ​​is:

[0064]

[0065] in, Represents the standardized parameter value. subscript It is a standardized English abbreviation symbol. The unit is the unit corresponding to the target unit identifier; This represents the parameter value of the intermediate parameter node. subscript It's the middle English abbreviation symbol. The unit is the unit corresponding to the source unit identifier. The numerical values ​​are derived from the parameter value attributes of intermediate parameter nodes in the graph database; This represents the conversion factor obtained from a preset unit conversion mapping table. subscript These are English abbreviations for conversion. The range of values ​​for is positive real numbers. The method of obtaining the target unit identifier is to precisely match the source unit identifier and the target unit identifier in a preset unit conversion mapping table. When the source unit identifier and the target unit identifier are completely consistent, The value of is 1.

[0066] The recommended output module uses the target unit identifier as the unit label for the standardized parameter value. The unit label of the standardized parameter value is stored together with the standardized parameter value. The recommended output module associates the standardized parameter value with the node name of the intermediate parameter node and stores it in the first data area. The first data area uses a key-value pair structure, where the key is the node name of the intermediate parameter node, and the value is the standardized parameter value and its unit label.

[0067] The recommendation output module reads the parameter values ​​of the target node from the graph database. The target node is the last node in the ordered node sequence. Using the target node's node identifier as the query key, the module reads the attribute values ​​of the target node's parameter values ​​from the graph database, uses these attribute values ​​as the target node's parameter values, and stores these parameter values ​​in the second data area. The second data area stores the target node's node name, parameter values, and unit identifier.

[0068] The recommendation output module stores the edge sequence of the recommended path in the third data area. The third data area stores the edge identifier, start node identifier, and end node identifier of each second directed edge in the ordered edge sequence. The third data area is stored in a list structure, and each element in the list is a triple consisting of the edge identifier, start node identifier, and end node identifier of a second directed edge.

[0069] The recommendation output module merges the first, second, and third data areas to form a recommendation result dataset. This dataset is stored using structured objects, each containing fields for the first, second, and third data areas. The first data area field contains the standardized parameter values ​​and unit labels of the intermediate parameter nodes stored in the first data area. The second data area field contains the target node parameter values ​​and unit identifiers stored in the second data area. The third data area field contains the list of recommended path edge sequences stored in the third data area. The recommendation output module then outputs the recommendation result dataset to the front-end interface of the design collaboration platform.

[0070] See Figure 5 In the graph, the horizontal axis represents the parameter values ​​of intermediate parameter nodes in the recommended path, ranging from 0 to 200; the vertical axis represents the probability density of the corresponding parameter values. The legend distinguishes the probability density curves corresponding to five different computational logic identifiers, from computational logic identifier 1 to computational logic identifier 5. Each curve reflects the distribution characteristics of the parameter values ​​of intermediate parameter nodes under the corresponding computational logic identifier.

[0071] Among them, the probability density curve of the calculation logic identifier 1 is a solid blue line with a peak value of about 0.028. The parameter value range corresponding to the peak value is about 55 to 65, indicating that the parameter values ​​of the intermediate parameter nodes associated with the calculation logic identifier are mainly concentrated in this range, and the probability density is high in the range of 50 to 80. The curve drops rapidly on both sides, showing that the parameter value distribution is relatively concentrated.

[0072] The probability density curve of computational logic identifier 2 is an orange dashed line with a peak value of approximately 0.019. The peak value is located between 80 and 95, and the parameter values ​​are mainly distributed in the range of 60 to 110, showing a relatively wide distribution range. Compared with computational logic identifier 1, the distribution is more dispersed.

[0073] The probability density curve of logic identifier 3 is a green dotted line with a peak value of approximately 0.022. The peak value is located between 95 and 115, indicating that the intermediate parameter node values ​​of this logic association are concentrated in the higher parameter value range, and the probability density is obvious in the range of 90 to 130, showing a right-skewed trend.

[0074] The probability density curve of the calculated logic identifier 4 is a red dotted line with a peak value of approximately 0.024. The peak value range covers the area between 70 and 85, and the parameter value distribution range is relatively wide, with obvious probability density from 40 to 100, indicating a relatively uniform distribution.

[0075] The probability density curve for calculation logic identifier 5 is a purple dotted line with a peak value of approximately 0.019. The peak value is between 110 and 140, and the parameter values ​​are mainly distributed in the range of 90 to 160, indicating that the parameter values ​​corresponding to this calculation logic identifier tend to be in the higher value range and have a wide distribution.

[0076] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A knowledge graph-based intelligent recommendation system for engineering design parameters, characterized in that, include: The topology construction module parses the data dependencies between multiple disciplines in the marine engineering design collaboration platform and generates a data flow topology graph with design files as nodes and data transmission directions as directed edges. The graph generation module, based on the design file type and parameter categories of each node in the data flow topology graph, establishes a knowledge graph with parameter entities as nodes and computational dependencies between parameters as edges, and stores it in the graph database. The intent parsing module receives the user's input design query statement, extracts the target device type and the name of the parameter to be recommended through word segmentation and dependency parsing, and combines the target device type and the name of the parameter to be recommended into query conditions. The path search module locates the starting parameter node that matches the target device type in the graph database, and searches for all reachable paths connecting the target node corresponding to the parameter name to be recommended, starting from the starting parameter node and following the direction of the edge that calculates the dependency relationship between parameters. Each reachable path contains at least one intermediate parameter node and an edge connecting each node. The recommendation output module selects the path with the fewest edges from all reachable paths as the recommended path, extracts the parameter values ​​corresponding to each intermediate parameter node in the recommended path and the parameter values ​​of the target node, encapsulates them into a recommendation result dataset, and outputs it to the front-end interface of the design collaboration platform.

2. The intelligent recommendation system for engineering design parameters based on knowledge graphs according to claim 1, characterized in that, The topology building module parses the data dependencies, specifically including: Read the header information of each design file in the design collaboration platform. The header information includes the upstream input file identifier and the downstream output file identifier. Take the design file corresponding to the upstream input file identifier as the starting node and the design file corresponding to the downstream output file identifier as the ending node, and establish a first directed edge between the starting node and the ending node. The parameter fields commonly contained in the two design files connected by the first directed edge are used as the edge attributes of the first directed edge and stored in the edge attribute set of the data flow topology graph.

3. The intelligent recommendation system for engineering design parameters based on knowledge graphs according to claim 1, characterized in that, The map generation module establishes the computational dependencies between the parameters, specifically including: Extract the upstream and downstream design files corresponding to each directed edge in the data flow topology graph; Read the first set of parameter fields contained in the upstream design file and the second set of parameter fields contained in the downstream design file; By comparing the first set of parameter fields and the second set of parameter fields, common parameter fields that exist in both the first set of parameter fields and the second set of parameter fields are marked as passed parameters; For independent parameter fields in the second parameter field set that do not belong to the first parameter field set, a second directed edge is established between the downstream parameter node corresponding to the independent parameter field and the transmission parameter node corresponding to the transmission parameter, pointing from the transmission parameter node to the downstream parameter node, as the computational dependency relationship between the parameters.

4. The intelligent recommendation system for engineering design parameters based on knowledge graphs according to claim 3, characterized in that, The path search module searches all reachable paths, specifically including: Using the starting parameter node as the current traversal node, query all downstream nodes adjacent to the current traversal node in the graph database. Each downstream node is connected to the current traversal node by a second directed edge. Add the downstream node to the candidate queue and record the path length from the starting parameter node to the downstream node; Take the node with the shortest path length from the candidate queue as the new current traversal node, and repeat the query and add steps until the parameter name of the current traversal node is consistent with the parameter name to be recommended. Record the traversed node sequence and edge sequence as a reachable path. When the candidate queue is empty and the target node has not been reached, it is determined that there is no reachable path.

5. The intelligent recommendation system for engineering design parameters based on knowledge graphs according to claim 1, characterized in that, The intent parsing module extracts the target device type and the name of the parameter to be recommended, specifically including: Perform Chinese word segmentation on the designed query statement to obtain a word sequence, and label each word in the word sequence with a part-of-speech tag; From words marked with noun parts of speech, select words that match the preset device name dictionary as the target device type; Based on the position of the target device type in the word sequence, a series of consecutive word combinations that are located after the target device type and match the preset parameter name dictionary are selected as the parameter name to be recommended. When there are multiple words in the word sequence that match the preset device name dictionary, the word that appears last in the sequence is selected as the target device type.

6. The intelligent recommendation system for engineering design parameters based on knowledge graphs according to claim 1, characterized in that, The recommendation output module encapsulates the recommendation result dataset, specifically including: According to the order from the starting parameter node to the target node in the recommended path, the parameter value corresponding to each intermediate parameter node and the calculation logic identifier in the edge attribute of the edge connected to each intermediate parameter node are read sequentially. Using the computational logic identifier as the basis for parameter conversion, the parameter values ​​of the intermediate parameter nodes are processed to unify the units, resulting in standardized parameter values. The standardized parameter values ​​are associated with the node names of the intermediate parameter nodes and stored in the first data area. The parameter values ​​of the target node are stored in the second data area. The edge sequence of the recommended path is stored in the third data area. The first data area, the second data area, and the third data area are merged to form the recommendation result dataset.

7. The intelligent recommendation system for engineering design parameters based on knowledge graphs according to claim 6, characterized in that, The recommendation output module performs unit unification processing on the parameter values ​​of the intermediate parameter nodes, specifically including: Read the source unit identifier and target unit identifier corresponding to the computation logic identifier from the edge attributes; Call the preset unit conversion mapping table to query the conversion coefficient that matches the source unit identifier and the target unit identifier; The parameter value of the intermediate parameter node is multiplied by the conversion factor to obtain the standardized parameter value, and the target unit identifier is used as the unit label of the standardized parameter value.

8. The intelligent recommendation system for engineering design parameters based on knowledge graphs according to claim 1, characterized in that, The intent parsing module combines the target device type and the name of the parameter to be recommended into query conditions, specifically including: Read the project stage labels carried by each design file node in the data flow topology diagram from the graph database. The project stage labels include basic design stage labels and detailed design stage labels. When the design query statement contains the stage keyword corresponding to the basic design stage tag, the name of the parameter to be recommended is associated with the basic design stage tag to generate a first query sub-condition; when the design query statement contains the stage keyword corresponding to the detailed design stage tag, the name of the parameter to be recommended is associated with the detailed design stage tag to generate a second query sub-condition. The target device type is combined with the first query sub-condition and the second query sub-condition by performing a logical AND operation to obtain the query conditions.

9. The intelligent recommendation system for engineering design parameters based on knowledge graphs according to claim 4, characterized in that, When searching all reachable paths, if there are multiple nodes with the same path length in the candidate queue, the path search module will preferentially select the node with the highest semantic similarity between its name and the name of the parameter to be recommended as the new current traversal node.

10. The intelligent recommendation system for engineering design parameters based on knowledge graphs according to claim 4, characterized in that, When the candidate queue is empty and the target node has not been reached, the path search module expands the starting parameter node along the direction of the second directed edge to include all subgraph nodes within two layers as a backup recommendation set, and marks the nodes in the backup recommendation set that have a direct edge connection with the parameter name to be recommended as approximate target nodes.