A Method and System for Object Optimization Based on Large Model Fusion Knowledge Graph

By constructing a knowledge graph for optimal equipment selection and using a model based on the Transformer architecture to parse equipment selection task information, and by filtering and evaluating path evidence, the problem of difficulty in distinguishing between basic parameters and actual performance path constraints in existing equipment selection systems is solved, thereby improving the accuracy and stability of equipment ranking.

CN123117632APending Publication Date: 2026-11-17WUXI JINDADI HUINONG AGRICULTURAL SCIENCE & TECHNOLOGY DEVELOPMENT CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202611274758.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-21
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing knowledge graph-based industrial equipment selection systems, it is difficult to effectively distinguish the constraint support of basic parameter paths and actual performance paths, resulting in the problem of candidate equipment ranking deviation and the mismatch between the optimal selection results and industrial application conditions.

Method used

By collecting equipment parameters and operational performance data, a knowledge graph for equipment selection is constructed. A pre-trained language understanding and structured output model based on the Transformer architecture is used to parse equipment selection task information, filter attribute path evidence and performance path evidence, evaluate the degree of support for single path constraints, select effective contributing paths, and finally select and rank candidate equipment models.

Benefits of technology

It enables the differentiation between basic technical parameters of equipment and performance in actual process scenarios, improves the matching degree between recommended equipment and actual process application conditions, reduces the impact of differences in data standards between different business systems on the optimization results, and enhances the pertinence and stability of candidate equipment ranking.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117632A_ABST
    Figure CN123117632A_ABST
Patent Text Reader

Abstract

This invention discloses an object selection method and system based on a large-scale model fusion knowledge graph, relating to the field of intelligent selection technology for industrial equipment. The method includes the following steps: S1, collecting candidate equipment parameter data and operational performance data, performing data preprocessing, and constructing an equipment selection knowledge graph; S2, using a pre-trained language understanding and structured output model based on a Transformer architecture to parse the equipment selection task information, and based on the parsing results, filtering attribute path evidence and performance path evidence in the equipment selection knowledge graph, and determining the satisfaction of task constraints; S3, evaluating the support degree of single-path constraints and the marginal contribution of paths, filtering effective contributing paths, and ranking the candidate equipment models to obtain recommended candidate equipment models. This solves the problem of difficulty in distinguishing the supporting roles of attribute paths and performance paths in industrial equipment selection, leading to ranking bias and selection mismatch.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent optimization technology for industrial equipment, specifically to an object optimization method and system based on a large model fusion knowledge graph. Background Technology

[0002] With the continuous advancement of intelligent manufacturing, the Industrial Internet, and the construction of digital workshops, industrial robots, CNC equipment, sensors, controllers, and other equipment are widely used in various process scenarios. Different equipment models vary in structural specifications, technical parameters, applicable process ranges, and actual operational performance. Enterprises typically need to select equipment based on specific production tasks, process conditions, and equipment capabilities to support business activities such as production line construction, equipment upgrades, capacity expansion, and process optimization. Simultaneously, industrial information systems such as equipment asset management systems and manufacturing execution systems continuously accumulate data on equipment parameters, process tasks, and operational performance, providing a data foundation for industrial equipment selection and management.

[0003] For example, Chinese patent CN120822636A discloses an optimization method, knowledge graph data structure, and apparatus for a knowledge graph-enhanced large-scale model for the governance of black and gray industries. The method includes: constructing a knowledge graph of the target industry chain using empirical data; obtaining case information of historical cases within the target industry chain; training a preliminary large-scale model using the case information of the historical cases to obtain a large-scale model of the black and gray industry chain; and performing knowledge graph enhancement and fusion training on the large-scale model of the black and gray industry chain using the knowledge graph to obtain a large-scale governance model for black and gray industries. By adopting this technical solution, the governance model can be optimized using the knowledge graph of the target industry chain, enabling accurate and efficient processing of cases within the target industry chain based on the governance model, improving the efficiency and accuracy of case processing, and reducing the operational difficulty for staff.

[0004] However, existing knowledge graph-based industrial equipment selection systems typically use multi-hop relationship paths between equipment objects, technical attributes, application scenarios, and performance to associate, score, and rank candidate objects. In the actual knowledge graph structure, the technical attribute path mainly reflects the matching between the equipment's basic parameters and task requirements, while the performance result path mainly reflects the equipment's actual operating performance in the corresponding process scenario. The two paths have different constraints and support effects on specific selection tasks. If a uniform scoring method or a fixed weighting method is used for different paths, the basic parameter matching path can easily have an excessive impact on the ranking results, weakening the ranking effect of the performance result path, which directly reflects actual operating performance. This can lead to problems such as candidate object ranking shifts, key technical constraints not being effectively reflected, and mismatches between the selected results and industrial application conditions.

[0005] Therefore, in order to address the above problems, there is an urgent need for an object optimization method and system based on large-scale model fusion knowledge graph. Summary of the Invention

[0006] Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides an object optimization method and system based on a large model fusion knowledge graph. This solves the problems in the existing knowledge graph-based industrial equipment selection process, where different types of paths are scored uniformly or have fixed weights superimposed, making it difficult to effectively distinguish the constraint support of basic parameter paths and actual performance paths, causing candidate equipment ranking to be offset, and the optimization results to not match the industrial application conditions.

[0008] Technical solution

[0009] To achieve the above objectives, the present invention provides the following technical solution: an object selection method based on a large model fusion knowledge graph, comprising the following steps: S1, collecting candidate equipment parameter data and operational performance data, preprocessing the candidate equipment parameter data and operational performance data, and constructing an equipment selection knowledge graph; S2, receiving equipment selection task information, and parsing the equipment selection task information using a pre-trained language understanding and structured output model based on the Transformer architecture, filtering attribute path evidence and performance path evidence in the equipment selection knowledge graph based on the parsing results of the equipment selection task information, and determining the task constraint satisfaction of each path evidence based on the parsing results of the equipment selection task information; S3, evaluating the single path constraint support degree for path evidence that satisfies the task constraints, evaluating the path marginal contribution based on the single path constraint support degree, and filtering effective contribution paths; combining the effective contribution paths and the path marginal contribution, optimizing and ranking the candidate equipment models to obtain recommended candidate equipment models.

[0010] Furthermore, the specific process for collecting candidate equipment parameter data and operational performance data is as follows: Candidate equipment parameter data is collected through the equipment asset management system. This data includes: candidate equipment model, candidate equipment type, and technical attribute records. The technical attribute records include: technical attribute identifier, technical attribute name, technical attribute value, and technical attribute unit. Operational performance data is collected through the equipment status records and process task records in the manufacturing execution system. This data includes: equipment number, equipment model, process scenario number, and actual performance index records. The actual performance index records include: performance index identifier, performance index name, performance index value, and performance index unit.

[0011] Furthermore, the specific process for preprocessing candidate equipment parameter data and operational performance data to construct an equipment selection knowledge graph is as follows: Read the pre-defined standardized dictionary of technical attributes and standardized dictionary of performance indicators; standardize the technical attribute identifier, technical attribute name, technical attribute value, and technical attribute unit in each technical attribute record based on the standardized dictionary of technical attributes; standardize the performance indicator identifier, performance indicator name, performance indicator value, and performance indicator unit in each actual performance indicator record based on the standardized dictionary of performance indicators; delete candidate equipment parameter data and operational performance data with empty fields, and remove duplicate technical attribute records and actual performance indicator records. The process involves: using candidate equipment models as equipment object nodes, equipment numbers as equipment instance nodes, equipment numbers and process scenario numbers as equipment operating condition nodes, technical attribute identifiers, technical attribute values, and technical attribute units as technical attribute nodes, and performance index identifiers, performance index values, and performance index units as performance index nodes; establishing model attribution relationships between equipment object nodes and equipment instance nodes, technical attribute relationships between equipment object nodes and technical attribute nodes, scenario operation relationships between equipment instance nodes and equipment operating condition nodes, and performance result relationships between equipment operating condition nodes and performance index nodes; and constructing an equipment optimization knowledge graph.

[0012] Furthermore, the specific process of receiving equipment selection task information and parsing it using a pre-trained language understanding and structured output model based on the Transformer architecture is as follows: The equipment selection task information, including the selection task number and task requirement description, is received. The task requirement description is input into the pre-trained language understanding and structured output model based on the Transformer architecture to obtain the target equipment type, target process scenario number, and a set of task constraint records. Each task constraint record includes: constraint record number, constraint index category, constraint index name, constraint value, constraint unit, and constraint comparison relationship. The constraint index categories include technical attribute constraints and performance index constraints. Based on the constraint index category of the task constraint record, the corresponding technical attribute standardization dictionary or performance index standardization dictionary is queried to obtain the standard index identifier, standard index name, and standard index unit for the corresponding category. The constraint index name, constraint value, and constraint unit are standardized, and the corresponding standard index identifier is used as the constraint index identifier of the task constraint record.

[0013] Furthermore, based on the analysis results of equipment selection task information, the specific process for filtering attribute path evidence and performance path evidence in the equipment selection knowledge graph is as follows: Candidate equipment models with the same type as the target equipment are selected from the equipment selection knowledge graph to obtain a candidate equipment set; each candidate equipment model in the candidate equipment set is obtained; for task constraint records where the constraint index category is technical attribute constraint, the technical attribute record is read; when the technical attribute identifier in the technical attribute record is the same as the constraint index identifier in the task constraint record, an attribute path evidence record is established between the candidate equipment model, technical attribute identifier, technical attribute value, and constraint record number; for task constraint records where the constraint index category is performance index constraint, the operating performance data where the equipment model is the same as the candidate equipment model and the process scenario number is the same as the target process scenario number is read; when the performance index identifier in the actual performance index record is the same as the constraint index identifier in the task constraint record, a performance path evidence record is established between the candidate equipment model, equipment number, process scenario number, performance index identifier, performance index value, and constraint record number; the attribute path evidence record and performance path evidence record are associated with their respective task constraint records.

[0014] Furthermore, the specific process for determining the satisfaction of task constraints for each path's evidence based on the analysis results of equipment selection task information is as follows: For each attribute path evidence record and performance path evidence record, the technical attribute value or performance index value is matched with the constraint value according to the constraint comparison relationship in the corresponding task constraint record; when the matching judgment result satisfies the constraint comparison relationship, a constraint-satisfied flag is generated; when the matching judgment result does not satisfy the constraint comparison relationship, a constraint-unsatisfied flag is generated; wherein, the constraint comparison relationship includes greater than, greater than or equal to, less than, less than or equal to, and equal to; the path type and constraint flag are written into the corresponding path evidence record, and the path type includes attribute path and performance path.

[0015] Furthermore, the specific process for evaluating the support level of a single path constraint for path evidence that satisfies task constraints is as follows: Read attribute path evidence records and performance path evidence records; when a constraint-satisfying marker exists in a path evidence record, the corresponding path evidence record is designated as a constraint-satisfying path evidence record; and when a candidate device model has a path evidence record that satisfies the constraint marker for each task constraint record in the task constraint record set, the candidate device model is designated as a constraint-satisfying candidate device model; for each constraint-satisfying path evidence record, when the constraint comparison relationship is greater than or greater than or equal to, subtract the constraint value from the corresponding indicator value, and then divide by the sum of the absolute value of the constraint value and a preset minimum positive number to obtain the path constraint margin value; when the constraint comparison relationship is less than or less than or equal to, subtract the corresponding indicator value from the constraint value, and then divide by the sum of the absolute value of the constraint value and a preset minimum positive number to obtain the path constraint margin value; when the constraint comparison relationship is equal, set the path constraint margin value to zero; and use the sum of 1 and the path constraint margin value as the single path constraint support value.

[0016] Furthermore, the specific process for evaluating the marginal contribution of a path based on the degree of support for single-path constraints and selecting effective contributing paths is as follows: Read evidence records of paths satisfying the constraints that have the same candidate device model, constraint record number, and path type; use the maximum value among the single-path constraint support values ​​as the path type constraint support value for the corresponding candidate device model relative to the corresponding task constraint record; for each evidence record of a path satisfying the constraints, temporarily delete it from the list of evidence records of paths satisfying the constraints for the same candidate device model, constraint record number, and path type, and use the maximum value of the single-path constraint support value among the remaining evidence records of paths satisfying the constraints as the path type constraint support value after removal; when there are no remaining evidence records of paths satisfying the constraints, set the path type constraint support value after removal to zero; subtract the path type constraint support value after removal from the path type constraint support value to obtain the marginal contribution value of the evidence record of the path satisfying the constraints; when the path marginal contribution value is greater than zero, mark the evidence record of the path satisfying the constraints as an effective contributing path.

[0017] Furthermore, combining the effective contribution path and the path marginal contribution, the candidate device models are optimized and ranked to obtain the recommended candidate device models. The specific process is as follows: For candidate device models marked as meeting the constraints, the path type constraint support value of each task constraint record is read and used as the task constraint support value of the corresponding task constraint record; for each candidate device model and each task constraint record, when an effective contribution path exists, the effective contribution path with the largest path marginal contribution value is taken as the key contribution path of the corresponding task constraint record, and the path marginal contribution value of the key contribution path is taken as the constraint effective contribution value of the corresponding task constraint record; when no effective contribution path exists, the constraint effective contribution value is set to zero; for each candidate device model, the task constraint support value and constraint effective contribution value corresponding to the performance index constraint and technical attribute constraint are summarized according to the constraint index category. For each task constraint support value under the same constraint index category, minimum value extraction and summation are performed to obtain the minimum support value and total support value of the corresponding category. The effective contribution values ​​of each constraint under the same constraint index category are then summed to obtain the total effective contribution value of the corresponding category. For the task constraint record set, candidate device models marked as satisfying the constraints are sorted in descending order: minimum support value of performance constraints, total support value of performance constraints, total effective contribution value of performance paths, minimum support value of attribute constraints, total support value of attribute constraints, and total effective contribution value of attribute paths. This yields the candidate device optimization ranking result. The first candidate device model in the candidate device optimization ranking result is used as the recommended candidate device model, and a device optimization result record is generated. The device optimization result record includes: selection task number, recommended candidate device model, and recommended candidate device type.

[0018] The second aspect of this invention provides an object selection system based on a large-scale model fusion knowledge graph, comprising: a data acquisition graph construction module for collecting candidate equipment parameter data and operational performance data, preprocessing the candidate equipment parameter data and operational performance data, and constructing an equipment selection knowledge graph; a task parsing path evidence module for receiving equipment selection task information, parsing the equipment selection task information using a pre-trained language understanding and structured output model based on the Transformer architecture, filtering attribute path evidence and performance path evidence in the equipment selection knowledge graph based on the parsing results of the equipment selection task information, and determining the task constraint satisfaction of each path evidence based on the parsing results of the equipment selection task information; and a path contribution object selection module for evaluating the degree of support for single path constraints for path evidence that satisfies task constraints, evaluating the path marginal contribution based on the degree of support for single path constraints, and filtering effective contribution paths; and combining the effective contribution paths and the path marginal contribution to optimize and rank the candidate equipment models to obtain recommended candidate equipment models.

[0019] Beneficial effects

[0020] The present invention has the following beneficial effects:

[0021] (1) This invention collects candidate equipment parameter data from the equipment asset management system and operating performance data from the manufacturing execution system, and standardizes the names, unifies the units, cleans up empty fields and removes duplicate records for technical attributes and performance indicators to construct a knowledge graph for equipment selection. This enables a unified association between candidate equipment models, technical attributes, process scenarios and actual performance indicators, and reduces the impact of inconsistent field names and units of measurement in different business systems on the equipment selection results.

[0022] (2) This invention utilizes a pre-trained language understanding and structured output model based on the Transformer architecture to perform structured parsing of the target equipment type, target process scenario and task constraints in the equipment selection task information. It converts the equipment selection requirements in natural language form into task constraint records that can be matched with knowledge graph nodes and relationships, reducing the workload of manually configuring selection rules and improving the parsing adaptability of task requirements under different expression methods.

[0023] (3) The present invention constructs attribute path evidence and performance path evidence respectively, and determines whether each path evidence meets the corresponding constraints based on the task constraint record, so that the matching status of the basic technical parameters of the equipment and the actual performance of the equipment in the target process scenario can be distinguished; thereby avoiding the selection of equipment based solely on the technical parameter matching results, and improving the matching degree between the recommended equipment and the actual process application conditions.

[0024] (4) The present invention evaluates the marginal contribution of a path based on the degree of support of a single path constraint, and filters out effective contribution paths. Then, it sorts the candidate equipment models by combining the minimum support value of performance constraints, the total support value of performance constraints, the total effective contribution value of performance paths, and the support status of attribute constraints. This can identify the key paths that have a real decisive role in satisfying task constraints, avoid the problem that the basic parameter paths have an excessive influence on the optimization results when using a fixed weight superposition method, and improve the pertinence and stability of the candidate equipment sorting.

[0025] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0026] Figure 1 Flowchart of an object optimization method based on large-scale model fusion knowledge graph;

[0027] Figure 2 A structural diagram of an object optimization system based on a large model fusion knowledge graph;

[0028] Figure 3To satisfy the path constraint margin distribution diagram of the evidence for the constrained path;

[0029] Figure 4 A schematic diagram of the knowledge graph for device optimization. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. As those skilled in the art will understand, the described embodiments are only some embodiments of the present invention, and 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.

[0031] Please see Figures 1-4 This invention provides a technical solution: an object optimization method based on a large model fusion knowledge graph, such as... Figure 1 As shown, the process includes the following steps: S1, collecting candidate equipment parameter data and operational performance data, preprocessing the candidate equipment parameter data and operational performance data, and constructing an equipment selection knowledge graph; S2, receiving equipment selection task information, and using a pre-trained language understanding and structured output model based on the Transformer architecture to parse the equipment selection task information, filtering attribute path evidence and performance path evidence in the equipment selection knowledge graph based on the parsing results of the equipment selection task information, and determining the task constraint satisfaction of each path evidence based on the parsing results of the equipment selection task information; S3, evaluating the degree of support for single path constraints for path evidence that satisfies task constraints, evaluating the path marginal contribution based on the degree of support for single path constraints, and filtering effective contribution paths; combining effective contribution paths and path marginal contributions, optimizing and ranking candidate equipment models to obtain recommended candidate equipment models.

[0032] Specifically, the process of collecting candidate equipment parameter data and operational performance data is as follows: Candidate equipment parameter data is collected through the equipment asset management system. This data includes: candidate equipment model, candidate equipment type, and technical attribute records. The technical attribute records include: technical attribute identifier, technical attribute name, technical attribute value, and technical attribute unit. The equipment asset management system is an information management system used for registering, maintaining, and querying equipment models, equipment classifications, equipment technical parameters, and equipment asset ledgers within the enterprise. The system stores equipment model ledgers and equipment technical parameter configuration records. The equipment model ledger includes at least candidate equipment models and candidate equipment types, and the equipment technical parameter configuration records include at least candidate equipment models, technical attribute identifiers, technical attribute names, technical attribute values, and technical attribute units. Preferably, a communication connection is established with the equipment asset management system through a pre-configured data query interface. Each candidate equipment model corresponds to one candidate equipment parameter data entry and one or more technical attribute records. The candidate equipment model uniquely identifies the equipment model to be selected, and the candidate equipment type represents the equipment category to which the candidate equipment belongs. The technical attribute identifier uniquely identifies the corresponding technical attribute, the technical attribute name represents the business meaning of the technical attribute, the technical attribute value represents the parameter content of the corresponding technical attribute, and the technical attribute unit represents the unit of measurement for the technical attribute value. For example, if the candidate equipment model is "RB-120" and the candidate equipment type is "welding robot", the corresponding technical attribute records may include: technical attribute identifier "TA001", technical attribute name "rated load", technical attribute value "120", and technical attribute unit "kg"; and technical attribute identifier "TA002", technical attribute name "repeatability", technical attribute value "0.05", and technical attribute unit "mm". When the technical attribute is a non-metric technical attribute such as communication protocol or interface type, the technical attribute value is recorded according to a preset coding rule, the technical attribute unit is empty, and it participates in subsequent task constraint matching based on equality relationships. Operational performance data is collected through equipment status records and process task records in the Manufacturing Execution System (MES). This data includes: equipment number, equipment model, process scenario number, and actual performance index records. The actual performance index records include: performance index identifier, performance index name, performance index value, and performance index unit. The MES is an information system used to collect, store, and manage production task issuance, process execution, equipment operating status, production cycle time, quality results, and process task records. The MES stores equipment status records and process task records. Equipment status records include at least the equipment number, equipment model, collection timestamp, and equipment operating status parameters. Process task records include at least the task number, equipment number, process scenario number, task start time, and task end time.Preferably, a communication connection is established with the Manufacturing Execution System (MES) through a pre-configured data query interface. Equipment status records and process task records are associated with equipment numbers, and equipment status records with timestamps between the task start and end times are selected to obtain equipment operating status data during the corresponding process task execution. Equipment status records can be written to the MES according to a preset sampling period, preferably 5 seconds. For each process task record, statistical processing is performed on the equipment operating status data during the corresponding task execution period according to preset performance index generation rules to generate actual performance index records. For example, for continuous operating parameters, the median value during task execution is preferably used to generate the corresponding performance index value; for task time-related indicators, the corresponding performance index value is generated based on the time difference between the task end time and the task start time; for pass rate indicators, the corresponding performance index value is generated based on the ratio between the number of qualified products and the total number of products during task execution. The equipment number uniquely identifies the actual equipment instance in operation; the equipment model represents the candidate equipment model to which the equipment instance belongs; and the process scenario number uniquely identifies the process application scenario corresponding to the equipment performing the process task. The performance index identifier uniquely identifies the actual performance index; the performance index name represents the business meaning of the performance index; the performance index value represents the actual operating result of the equipment in the corresponding process scenario; and the performance index unit represents the unit of measurement for the performance index value. For example, the equipment instance with equipment number "EQ-001" corresponds to equipment model "RB-120". When it performs the welding task with process scenario number "SC-W01", it can generate an actual performance index record with performance index identifier "PI001", performance index name "actual cycle time", performance index value "38.5", and performance index unit "s"; it can also generate an actual performance index record with performance index identifier "PI002", performance index name "actual positioning deviation", performance index value "0.04", and performance index unit "mm". The candidate equipment model in the candidate equipment parameter data and the equipment model in the operating performance data use the same equipment model coding rule, so that the actual performance index records of the corresponding equipment instance under different process scenarios can be associated with the candidate equipment model in the future.

[0033] In this implementation scheme, the technical parameters of candidate equipment and the operating data of the equipment in actual process scenarios are obtained through the data query interfaces of the equipment asset management system and the manufacturing execution system. The candidate equipment models and equipment instances are associated using a unified equipment model code, which can form candidate equipment parameter data and operating performance data with clear source, complete fields and traceability. At the same time, the corresponding equipment status records are filtered based on the time range of the process task, and performance indicators such as actual cycle time, positioning deviation and pass rate are extracted according to the performance index generation rules. The median is used to represent continuous operating parameters, which can reduce the impact of instantaneous fluctuations and abnormal sampling data on performance evaluation, thereby improving the reliability and applicability of the subsequent equipment selection knowledge graph construction and candidate equipment selection results.

[0034] Specifically, the process of preprocessing candidate equipment parameter data and operational performance data to construct an equipment selection knowledge graph involves: reading a pre-defined standardized dictionary of technical attributes and a standardized dictionary of performance indicators; recording the correspondence between each technical attribute name and standard technical attribute identifier, and between standard technical attribute name and standard technical attribute unit in the standardized dictionary of technical attributes; recording the correspondence between each performance indicator name and standard performance indicator identifier, and between standard performance indicator name and standard performance indicator unit in the standardized dictionary of performance indicators; wherein, the pre-defined standardized dictionary of technical attributes and the standardized dictionary of performance indicators are established and maintained according to the enterprise equipment classification specifications, equipment technical parameter coding specifications, and performance indicator definition rules of the manufacturing execution system, and are pre-stored in the standardized dictionary library. The technical attribute names in the standardized dictionary of technical attributes include fields such as rated load, rated power, repeatability accuracy, working range, communication protocol, and protection level; the performance indicator names in the standardized dictionary of performance indicators include fields such as actual cycle time, actual positioning deviation, actual response time, actual energy consumption, and actual processing pass rate. Standard technical attribute identifiers and standard performance index identifiers are used to uniquely identify standardized technical attributes and performance indexes, respectively. Standard technical attribute names and standard performance index names are used to unify the naming of data with the same business meaning from different sources. Standard technical attribute units and standard performance index units are used to unify the measurement caliber of corresponding values. For example, the technical attribute names "repeatability" and "positioning repeatability" can both correspond to the same standard technical attribute identifier; the performance index names "single-piece processing time" and "actual cycle time" can both correspond to the same standard performance index identifier. Based on the technical attribute standardization dictionary, the technical attribute identifiers, technical attribute names, technical attribute values, and technical attribute units in each technical attribute record are standardized. Specifically, for each technical attribute record, the technical attribute standardization dictionary is queried based on the technical attribute name. When the technical attribute unit and the standard technical attribute unit are both non-empty and inconsistent, the technical attribute value is converted according to the unit conversion rules between the technical attribute unit and the standard technical attribute unit, and the technical attribute identifier, technical attribute name, and standard technical attribute unit are updated to the standard technical attribute identifier, standard technical attribute name, and standard technical attribute unit. When a technical attribute name does not have a corresponding dictionary entry in the technical attribute standardization dictionary, the corresponding technical attribute record is marked as an unstandardized technical attribute record and will not participate in the subsequent construction of the equipment selection knowledge graph. When the technical attribute is a non-measuring technical attribute such as a communication protocol, interface type, or protection level, the technical attribute value is recorded according to the preset coding rules, the technical attribute unit is empty, no unit conversion is performed, and the corresponding technical attribute record is retained for subsequent equality relationship matching.Based on a standardized performance index dictionary, the performance index identifier, performance index name, performance index value, and performance index unit in each actual performance index record are standardized. Specifically, for each actual performance index record, the standardized performance index dictionary is queried based on the performance index name. When the performance index unit and the standard performance index unit are both non-empty and inconsistent, the performance index value is converted according to the unit conversion rules between the performance index unit and the standard performance index unit, and the performance index identifier, performance index name, and performance index unit are updated to the standard performance index identifier, standard performance index name, and standard performance index unit. When the performance index name does not have a corresponding dictionary entry in the standardized performance index dictionary, the corresponding actual performance index record is marked as an unstandardized performance index record and will not participate in the subsequent construction of the equipment optimization knowledge graph. Delete candidate equipment parameter data and operational performance data containing empty fields, and deduplicate duplicate technical attribute records and actual performance index records. Candidate equipment parameter data with empty fields refers to records where the candidate equipment model, candidate equipment type, technical attribute identifier, technical attribute name, or technical attribute value is empty. Operational performance data with empty fields refers to records where the equipment number, equipment model, process scenario number, performance index identifier, performance index name, or performance index value is empty. For non-measuring technical attributes, an empty technical attribute unit is not considered a deletion condition. For the same candidate equipment model, duplicate technical attribute records are identified based on the same technical attribute identifier, technical attribute value, and technical attribute unit, and the record with the latest acquisition time is retained. For the same equipment number, the same process scenario number, and the same performance index identifier, duplicate actual performance index records are identified based on the same performance index value and performance index unit, and the record with the latest corresponding process task completion time is retained. After deduplication, if a candidate equipment model still retains at least one valid technical attribute record, the candidate equipment parameter data corresponding to that candidate equipment model is retained. The candidate equipment model is used as the equipment object node, the equipment number as the equipment instance node, the equipment number and process scenario number together as the equipment operating condition node, the technical attribute identifier, technical attribute value and technical attribute unit together as the technical attribute node, and the performance index identifier, performance index value and performance index unit together as the performance index node. The model attribution relationship between the equipment object node and the equipment instance node is established, the technical attribute relationship between the equipment object node and the technical attribute node is established, the scenario operation relationship between the equipment instance node and the equipment operating condition node is established, and the performance result relationship between the equipment operating condition node and the performance index node is established. Based on the model attribution relationship, technical attribute relationship, scenario operation relationship and performance result relationship, an equipment optimization knowledge graph is constructed.The system comprises several nodes and their associated characteristics. The equipment object node uses the candidate equipment model as its unique identifier to represent the equipment model participating in the selection process. The equipment instance node uses the equipment number as its unique identifier to represent the equipment instance actually in operation. The equipment condition node uses a combination of the equipment number and the process scenario number as its unique identifier to represent the operating status of the corresponding equipment instance in the corresponding process scenario. The technical attribute node uses a combination of technical attribute identifier, technical attribute value, and technical attribute unit as its node attribute to represent the standardized technical parameters of the candidate equipment model. The performance index node uses a combination of performance index identifier, performance index value, and performance index unit as its node attribute to represent the actual operating performance of the equipment instance in the corresponding process scenario. The model attribution relationship between the equipment object node and the equipment instance node represents the candidate equipment model to which the equipment instance belongs. The technical attribute relationship between the equipment object node and the technical attribute node represents the corresponding technical attributes possessed by the candidate equipment model. The scenario operation relationship between the equipment instance node and the equipment condition node represents the process tasks executed by the equipment instance in the corresponding process scenario. The performance result relationship between the equipment condition node and the performance index node represents the actual performance index formed by the corresponding equipment instance in the corresponding process scenario. The equipment selection knowledge graph can be stored in a graph database, which records the node identifier, node type, and node attributes of each node, as well as the starting node identifier, ending node identifier, and relationship type of each relationship. This enables the technical attribute information of candidate equipment models to form a queryable association structure with the actual performance information of corresponding equipment instances under different process scenarios.

[0035] like Figure 4 The diagram illustrates the knowledge graph for equipment selection. It shows the node structure and relationships within the knowledge graph. Equipment object nodes are constructed based on candidate equipment models, and connected to corresponding equipment instance nodes by model affiliation, thus distinguishing different actually operational equipment of the same model. Equipment object nodes are also connected to technical attribute nodes via technical attribute relationships, recording the technical attribute identifiers, values, and units of the candidate equipment models, such as rated load and repeatability accuracy. Equipment instance nodes are further connected to equipment condition nodes via scenario operation relationships. These equipment condition nodes, determined by both the equipment number and the process scenario number, characterize the operational status of a specific equipment instance within a specific process scenario. Equipment condition nodes are connected to performance indicator nodes via performance result relationships, recording the performance indicator identifiers, values, and units of the corresponding equipment instance within that process scenario, such as actual cycle time and actual positioning deviation. Thus, the graph establishes a correlation between the basic technical attributes of candidate equipment models and the actual operational performance of equipment instances within specific process scenarios, providing a data foundation for subsequent screening of attribute path evidence and performance path evidence based on task constraints.

[0036] In this implementation plan, a standardized dictionary is used to unify the names, identifiers, and units of measurement of the technical attributes and actual performance indicators of candidate equipment. Empty fields, unstandardized records, and duplicate records are processed to reduce the impact of differences in data caliber between different systems on subsequent selection. At the same time, an equipment selection knowledge graph is constructed based on equipment objects, equipment instances, equipment operating conditions, technical attributes, and performance indicators. This enables a traceable association between the basic parameters of candidate equipment models and the actual operating performance of equipment instances in different process scenarios, providing a structured data foundation for subsequent screening of path evidence based on task constraints and implementation of candidate equipment selection.

[0037] Specifically, the process of receiving equipment selection task information and parsing it using a pre-trained language understanding and structured output model based on the Transformer architecture is as follows: The equipment selection task information includes a selection task number and a task requirement description. The equipment selection task information is created by the equipment selection management terminal and read through the task receiving interface. The selection task number uniquely identifies the current equipment selection task. The task requirement description is natural language text describing the category of the candidate equipment, the target process scenario, and the equipment capability requirements. The task requirement description may include equipment type terms, process scenario terms, technical attribute constraint terms, performance index constraint terms, constraint values, constraint units, and comparison relation terms. For example, the task requirement description might be: "For welding tasks, select a welding robot with a rated load of not less than 100kg, a repeatability of not more than 0.05mm, and an actual cycle time of not more than 45s." Here, "welding robot" corresponds to the target equipment type, "welding task" is used to identify the target process scenario number, and "rated load not less than 100kg," "repeatability of not more than 0.05mm," and "actual cycle time of not more than 45s" correspond to different task constraints. The task requirement description is input into a pre-trained language understanding and structured output model based on the Transformer architecture, resulting in a set of target equipment type, target process scenario number, and task constraint records. Each task constraint record includes: constraint record number, constraint index category, constraint index name, constraint value, constraint unit, and constraint comparison relationship. The constraint index categories include technical attribute constraints and performance index constraints. The pre-trained language understanding and structured output model based on the Transformer architecture is trained and fine-tuned using task requirement descriptions from historical equipment selection tasks and corresponding labeled target equipment type, target process scenario number, and task constraint records. The labeled task constraint records are generated according to a preset field structure, enabling the model to learn the correspondence between equipment type terms, process scenario terms, constraint index terms, constraint values, constraint units, and comparison relationship terms. The pre-trained language understanding and structured output model based on the Transformer architecture includes a text segmentation and positional encoding unit, a Transformer contextual semantic encoding unit, and a preset field structure output unit. The text segmentation and positional encoding unit is used to convert the task requirement description into a sequence of text tags and corresponding positional encoding sequences. The Transformer contextual semantic encoding unit is used to extract the contextual semantic association features between each text tag. The preset field structure output unit is used to output the target equipment type, target process scenario number, and task constraint record set according to the preset field structure.The preset field structure includes at least the following fields: Target Equipment Type, Target Process Scenario Number, Constraint Record Number, Constraint Index Category, Constraint Index Name, Constraint Value, Constraint Unit, and Constraint Comparison Relationship. The Target Equipment Type field uses the same classification name or coding rule as the candidate equipment type, and the Target Process Scenario Number field uses the same coding rule as the process scenario number in the equipment selection knowledge graph, to enable subsequent equipment type matching and process scenario matching. For multiple task constraints included in the same task requirement description, corresponding constraint record numbers are generated according to the order in which each task constraint appears in the task requirement description. When the constraint index category is a technical attribute constraint, the corresponding constraint index is used to limit the basic technical parameters of the candidate equipment model. When the constraint index category is a performance index constraint, the corresponding constraint index is used to limit the actual operating performance of the equipment instance under the target process scenario. The constraint comparison relationships include greater than, greater than or equal to, less than, less than or equal to, and equal to; among them, "higher than," "greater than," and "exceeding" correspond to greater than; "not lower than," "not less than," and "not less than" correspond to greater than or equal to; "lower than" and "less than" correspond to less than; "not higher than," "not greater than," and "not exceeding" correspond to less than or equal to; and "is" and "equal to" correspond to equal. When "adopted" is followed by non-metric technical attribute values ​​such as communication protocol, interface type, or protection level, it corresponds to equal. Based on the constraint index category of the task constraint record, the corresponding technical attribute standardization dictionary or performance index standardization dictionary is queried to obtain the standard index identifier, standard index name, and standard index unit of the corresponding category. The constraint index name, constraint value, and constraint unit are standardized, and the corresponding standard index identifier is used as the constraint index identifier of the task constraint record. Specifically, when the constraint index category is a technical attribute constraint, the standard dictionary of technical attributes is queried based on the constraint index name. When the constraint unit and the standard technical attribute unit are both not empty and inconsistent, the constraint value is converted according to the unit conversion rules between the constraint unit and the standard technical attribute unit, and the constraint index name and constraint unit are updated to the standard technical attribute name and standard technical attribute unit. The standard technical attribute identifier is used as the constraint index identifier of the task constraint record. Specifically, the corresponding standard technical attribute identifier, standard technical attribute name, and standard technical attribute unit are obtained from the standard dictionary of technical attributes based on the constraint index name. When the technical attribute constraint corresponds to non-measuring technical attributes such as communication protocol, interface type, or protection level, the constraint value is recorded according to the same preset coding rules as the technical attribute value, the constraint unit is empty, and the constraint comparison relationship is equal. In this case, no unit conversion is performed.When the constraint index category is a performance index constraint, the performance index standardization dictionary is queried based on the constraint index name. If the constraint unit and the standard performance index unit are both non-empty and inconsistent, the constraint value is converted according to the unit conversion rules between the constraint unit and the standard performance index unit. The constraint index name and constraint unit are then updated to the standard performance index name and standard performance index unit, and the standard performance index identifier is used as the constraint index identifier for the task constraint record. Specifically, the corresponding standard performance index identifier, standard performance index name, and standard performance index unit are obtained from the performance index standardization dictionary based on the constraint index name. After standardization, each task constraint record includes at least the constraint record number, constraint index category, standardized constraint index name, constraint value, standardized constraint unit, constraint comparison relationship, and constraint index identifier. The constraint index identifier is used to match the technical attribute identifier or performance index identifier in the equipment optimization knowledge graph.

[0038] In this implementation scheme, a pre-trained language understanding and structured output model based on the Transformer architecture is used to parse the equipment selection task requirements in natural language form into target equipment type, target process scenario number, and task constraint record. Furthermore, a standardized dictionary is used to unify the names, identifiers, and units of measurement for constraint indicators, enabling equipment capability requirements with different expressions and measurement standards to be converted into structured constraints that can be directly matched with the equipment selection knowledge graph. Simultaneously, by distinguishing between technical attribute constraints and performance indicator constraints and identifying corresponding comparison relationships, the completeness and standardization of task constraint parsing are improved, providing a foundation for accurately screening attribute path evidence and performance path evidence that meet the equipment selection requirements.

[0039] Specifically, based on the analysis results of equipment selection task information, the process of filtering attribute path evidence and performance path evidence in the equipment selection knowledge graph is as follows: Candidate equipment models with the same type as the target equipment are selected from the equipment selection knowledge graph to obtain a candidate equipment set. Specifically, the equipment object nodes and their corresponding candidate equipment types are read from the equipment selection knowledge graph, and the candidate equipment models corresponding to the equipment object nodes with the same type as the target equipment are determined as candidate equipment models. The candidate equipment set includes one or more candidate equipment models. The target equipment type and candidate equipment types use the same classification name or coding rules to ensure the accuracy of the equipment type matching results. Each candidate equipment model in the candidate equipment set is obtained, and for each candidate equipment model in the candidate equipment set, its associated technical attribute nodes, equipment instance nodes, equipment operating condition nodes, and performance indicator nodes are read one by one. Corresponding path evidence filtering processing is then performed on each task constraint record in the task constraint record set. For task constraint records where the constraint index category is technical attribute constraint, the technical attribute record is read. When the technical attribute identifier in the technical attribute record is the same as the constraint index identifier in the task constraint record, an attribute path evidence record is established between the candidate equipment model, technical attribute identifier, technical attribute value, and constraint record number. Specifically, for the current candidate equipment model, technical attribute nodes with technical attribute relationships to the corresponding equipment object nodes are read from the equipment optimization knowledge graph. The technical attribute identifier, technical attribute value, and technical attribute unit in the corresponding technical attribute node are obtained, and the technical attribute identifier is matched with the constraint index identifier corresponding to the current technical attribute constraint. The attribute path evidence record is used to represent the evidence path connecting the candidate equipment model to the technical attribute node that meets the identifier matching condition through the technical attribute relationship. The attribute path evidence record may also include technical attribute units and path type, with the path type marked as attribute path. The technical attribute unit is used to confirm that the technical attribute value and the constraint value use the same measurement caliber in the subsequent task constraint satisfaction determination process. For example, if the candidate equipment model is "RB-120", its technical attribute record includes the technical attribute identifier "TA001", the technical attribute name "rated load", the technical attribute value "120", and the technical attribute unit "kg". When the constraint index identifier corresponding to the technical attribute constraint is "TA001", an attribute path evidence record is established between the candidate equipment model "RB-120", the technical attribute identifier "TA001", the technical attribute value "120", and the corresponding constraint record number.For task constraint records with performance index constraints as the constraint category, operational performance data is read if the equipment model is the same as the candidate equipment model and the process scenario number is the same as the target process scenario number. When the performance index identifier in the actual performance index record is the same as the constraint index identifier in the task constraint record, a performance path evidence record is established between the candidate equipment model, equipment number, process scenario number, performance index identifier, performance index value, and constraint record number. Specifically, for the current candidate equipment model, equipment instance nodes with model affiliation relationships with the corresponding equipment object nodes are read from the equipment optimization knowledge graph, and the equipment operating condition nodes corresponding to each equipment instance node are read. The process scenario number in the equipment operating condition node is matched with the target process scenario number, and the performance index nodes with performance result relationships with the matched equipment operating condition nodes are read. The performance path evidence record is used to characterize the evidence path from the candidate equipment model to the performance index node that meets the identifier matching condition via the equipment instance node and the equipment operating condition node. The equipment number in the performance path evidence record is used to distinguish different actually operational equipment instances under the same candidate equipment model, and the process scenario number is used to limit the process application scenario to which the performance index value belongs. Performance path evidence records may also include performance metric units and path types, with the path type labeled as a performance path. For multiple equipment instances of the same candidate equipment model, or performance metric records generated by the same equipment instance under different process scenarios, corresponding performance path evidence records should be established separately. This enables subsequent task constraint satisfaction determination and path marginal contribution assessment based on performance path evidence from different equipment instances and different process scenarios. For example, the equipment instance with equipment number "EQ-001" corresponds to the candidate equipment model "RB-120", and its actual performance index record under process scenario number "SC-W01" includes the performance index identifier "PI001", the performance index name "actual cycle time", the performance index value "38.5", and the performance index unit "s". When the target process scenario number is "SC-W01", and the constraint index identifier corresponding to the performance index constraint is "PI001", a performance path evidence record is established between the candidate equipment model "RB-120", equipment number "EQ-001", process scenario number "SC-W01", performance index identifier "PI001", performance index value "38.5", and the corresponding constraint record number. The attribute path evidence record and the performance path evidence record are then associated with the corresponding task constraint records. Among them, the attribute path evidence record is associated with the task constraint record of the corresponding technical attribute through the constraint record number, and the performance path evidence record is associated with the task constraint record of the corresponding performance index through the constraint record number. The association relationship is used to enable the subsequent processing to read the corresponding attribute path evidence record or performance path evidence record based on each task constraint record, and to determine whether the corresponding path evidence meets the task constraints.

[0040] In this implementation plan, technical attribute path evidence and performance path evidence are screened from the equipment selection knowledge graph based on the target equipment type, target process scenario number, and constraint indicator identifier. This ensures that the basic technical parameters of candidate equipment and the actual operating performance of different equipment instances in the target process scenario correspond to specific task constraints. At the same time, by distinguishing the performance records of different equipment instances and different operating scenarios through equipment number and process scenario number, the performance data of non-target scenarios or different equipment instances can be avoided from being mixed into the screening results. This provides a reliable path evidence basis for accurately determining the satisfaction of task constraints and implementing candidate equipment selection.

[0041] Specifically, the process of determining the satisfaction of task constraints for each path evidence based on the analysis results of equipment selection task information is as follows: For each attribute path evidence record and performance path evidence record, the technical attribute value or performance index value is matched with the constraint value according to the constraint comparison relationship in the corresponding task constraint record; when the matching judgment result satisfies the constraint comparison relationship, a constraint-satisfied mark is generated; when the matching judgment result does not satisfy the constraint comparison relationship, a constraint-unsatisfied mark is generated; wherein, the constraint comparison relationship includes greater than, greater than or equal to, less than, less than or equal to, and equal to. Specifically, a constraint-satisfied flag is generated when the constraint comparison relationship is greater than, and the technical attribute value or performance index value is greater than the constraint value; a constraint-satisfied flag is generated when the constraint comparison relationship is greater than or equal to, and the technical attribute value or performance index value is greater than or equal to the constraint value; a constraint-satisfied flag is generated when the constraint comparison relationship is less than, and the technical attribute value or performance index value is less than the constraint value; a constraint-satisfied flag is generated when the constraint comparison relationship is less than or equal to, and the technical attribute value or performance index value is less than or equal to the constraint value; a constraint-satisfied flag is generated when the constraint comparison relationship is equal, and the technical attribute value or performance index value is the same as the constraint value; otherwise, a constraint-unsatisfied flag is generated. For each path evidence record, the corresponding task constraint record is read based on the constraint record number in the path evidence record to obtain the corresponding constraint index category, constraint index identifier, constraint value, constraint unit, and constraint comparison relationship. For attribute path evidence records, the technical attribute value is read as the comparison index value; for performance path evidence records, the performance index value is read as the comparison index value. Since the technical attribute values, performance index values, and constraint values ​​have already undergone standardized processing in the preceding steps according to the corresponding standardized dictionary in terms of index name, index identifier, and unit of measurement, the index values ​​to be compared and the constraint values ​​have the same index meaning and measurement scope. For numerical indicators such as rated load, repeatability accuracy, actual cycle time, and actual positioning deviation, the index values ​​to be compared and the constraint values ​​are converted into numerical formats before comparison. For non-metric technical attributes such as communication protocols, interface types, and protection levels, both technical attribute values ​​and constraint values ​​are recorded using the same preset coding rules, and the constraint comparison relationship is equal; when the corresponding codes are the same, a constraint-satisfied mark is generated; otherwise, a constraint-unsatisfied mark is generated. When the constraint comparison relationship is equal and the corresponding index is a numerical index, the index values ​​to be compared and the constraint values ​​are processed to the same precision according to the preset numerical precision of the corresponding standard unit before equality comparison is performed; for example, for positioning accuracy indicators with a standard unit of mm, the preset number of decimal places are used for unified processing to avoid the influence of differences in decimal representation from different data sources on the determination of equality.For example, for the technical attribute constraint of "rated load not less than 100kg", if the technical attribute value in the attribute path evidence record is 120kg, a constraint satisfaction mark is generated; for the performance index constraint of "actual cycle time less than 45s", if the performance index value in the performance path evidence record is 38.5s, a constraint satisfaction mark is generated; for the technical attribute constraint of "communication protocol adopts EtherCAT", if the technical attribute value is the same as the preset code corresponding to the constraint value, a constraint satisfaction mark is generated. The path type and constraint mark are written into the corresponding path evidence record. The path type includes attribute path and performance path. Specifically, the path type of the attribute path evidence record is written as attribute path, and the path type of the performance path evidence record is written as performance path. The constraint satisfaction mark indicates that the corresponding path evidence can support the corresponding task constraint record, and the constraint non-satisfaction mark indicates that the corresponding path evidence cannot support the corresponding task constraint record. Subsequently, based on the candidate equipment model, constraint record number, path type, and constraint mark, constraint-satisfied path evidence records are screened, and the degree of path constraint support and path marginal contribution are evaluated.

[0042] In this implementation plan, by comparing the standardized numerical values ​​or codes of technical attribute path evidence and performance path evidence based on the comparison relationships in the task constraint records, and writing the judgment results into the corresponding path evidence records, it is possible to accurately distinguish between path evidence that meets and does not meet specific selection requirements. At the same time, by unifying the measurement caliber and accuracy of numerical indicators and using code consistency judgment for non-measurement technical attributes, the impact of differences in units, numerical representations, and textual descriptions on the constraint judgment results can be reduced, providing a reliable foundation for subsequent screening of effective paths and implementation of candidate equipment selection.

[0043] Specifically, the process for evaluating the degree of support for single-path constraints for path evidence that satisfies task constraints is as follows: Read attribute path evidence records and performance path evidence records; when a constraint-satisfying marker exists in a path evidence record, the corresponding path evidence record is designated as a constraint-satisfying path evidence record; and when a candidate device model has a path evidence record that satisfies the constraint marker for each task constraint record in the task constraint record set, the candidate device model is designated as a constraint-satisfying candidate device model. For each path evidence record, its associated constraint record number is read, and the constraint comparison relationship and constraint value of the corresponding task constraint record are obtained based on the constraint record number; the technical attribute values ​​in the attribute path evidence record or the performance index values ​​in the performance path evidence record are used as the corresponding index values. A candidate device model having a path evidence record that satisfies the constraint mark for a specific task constraint record means that the candidate device model has at least one attribute path evidence record or performance path evidence record associated with that task constraint record and written with the constraint mark. If a candidate device model does not have a path evidence record that satisfies the constraint mark for any task constraint record in the task constraint record set, the candidate device model will not be marked as a constraint-satisfied candidate device model and will not participate in subsequent candidate device selection and ranking. Since the corresponding indicator values ​​and constraint values ​​in the constraint-satisfied path evidence records have undergone unified processing of indicator identification, indicator names, and units of measurement, the corresponding indicator values ​​and constraint values ​​have the same business meaning and measurement scope, and can be directly used for subsequent path constraint margin value evaluation. For each evidence record satisfying the constraint path, when the constraint comparison relationship is greater than or greater than or equal to, the corresponding indicator value is subtracted from the constraint value, and then divided by the sum of the absolute value of the constraint value and a preset minimum positive number to obtain the path constraint margin value; when the constraint comparison relationship is less than or less than or equal to, the constraint value is subtracted from the corresponding indicator value, and then divided by the sum of the absolute value of the constraint value and a preset minimum positive number to obtain the path constraint margin value; when the constraint comparison relationship is equal, the path constraint margin value is set to zero; and the sum of 1 and the path constraint margin value is used as the single-path constraint support value. The preset minimum positive number is a pre-defined value greater than zero and less than one, used to avoid division by zero when the constraint value is zero, and is preferably set to... The path constraint margin value is used to characterize the margin by which the corresponding indicator value satisfies the constraint value. The larger the path constraint margin value, the stronger the support of the corresponding path evidence for the task constraint. When the constraint comparison relationship is greater than or greater than or equal to, the greater the extent to which the corresponding indicator value exceeds the constraint value, the larger the path constraint margin value; when the constraint comparison relationship is less than or less than or equal to, the greater the extent to which the corresponding indicator value is lower than the constraint value, the larger the path constraint margin value. For equality relationships, satisfying the constraint only indicates that the corresponding indicator value is consistent with the constraint value, and does not distinguish the strength of support based on the degree of numerical deviation, therefore the path constraint margin value is set to zero. For example, for the task constraint "rated load not less than 100kg", if the technical attribute value corresponding to the candidate equipment model is 120kg, then the path constraint margin value is (120-100)÷(|100|+ For the task constraint of "actual cycle time not exceeding 45s", if the performance index of the equipment instance in the target process scenario is 38.5s, then the path constraint margin is (45 - 38.5) ÷ (|45| + ...). The value of 1 in the single-path constraint support value is used to indicate the basic support level that the path evidence has met the corresponding task constraints, and the path constraint margin value is used to characterize the additional support level on top of the basic support level; thus, different dimensional indicators form comparable single-path constraint support levels after constraint value normalization.

[0044] In this embodiment, Table 1 is a data table of path constraint margin values. The preset minimum positive number is... The table details the candidate device model, path type, constraint record number, constraint index name, constraint comparison relationship, constraint value, corresponding index value, and path constraint margin value for five evidence records that satisfy the constraints.

[0045] Table 1. Path constraint margin values ​​data table

[0046]

[0047] like Figure 3The figure shows the distribution of path constraint margins for evidence that satisfies the constraints. In the figure, the horizontal axis represents the path constraint margin value of evidence records that satisfy the constraints, the vertical axis represents different evidence records that satisfy the constraints, the length of the bars reflects the satisfaction margin of the corresponding indicator value relative to the task constraint value, and the right side of the bars is labeled with the corresponding single-path constraint support value. The attribute path evidence records P1 have a rated load constraint margin of 0.200 and P2 have a repeatability accuracy constraint margin of approximately 0.200, indicating that candidate equipment model RB-120 has a large margin of compliance under these two technical attribute constraints, with a corresponding single-path constraint support value of 1.200. The performance path evidence records P3 and P4 have actual cycle time constraint margins of 0.144 and 0.089, respectively, indicating that equipment numbers EQ-001 and EQ-002 both meet the actual cycle time requirements in their respective process scenarios. However, EQ-001's actual cycle time is closer to the optimal level, resulting in a higher single-path constraint support value. Attribute path evidence record P5 corresponds to the communication protocol equality constraint; because the corresponding indicator value is consistent with the constraint value, its path constraint margin is zero, but it still has a basic single-path constraint support value of 1.000. This figure visually demonstrates the differences in the compliance margin and support level of different attribute paths and performance paths for task constraints, providing a data foundation for subsequent path marginal contribution assessment.

[0048] In this implementation scheme, candidate equipment models that do not meet any task constraints are first screened out, and then the degree of support for single-path constraints is evaluated based on the satisfaction margin of the corresponding index value relative to the constraint value. This allows different candidate equipment and different types of constraints to be compared under a unified standard. At the same time, the strength of a path's support for task constraints is characterized by a combination of basic support and additional support, which can highlight paths with a larger constraint satisfaction margin and provide a reliable basis for subsequent assessment of path marginal contribution and ranking of candidate equipment.

[0049] Specifically, the process of evaluating the marginal contribution of a path based on the degree of support for single-path constraints and selecting effective contributing paths is as follows: Evidence records of satisfying constraints for candidate equipment models, constraint record numbers, and path types are read. The maximum value among the single-path constraint support values ​​is taken as the path type constraint support value for the corresponding candidate equipment model for the corresponding task constraint record. Evidence records of satisfying constraints for candidate equipment models, constraint record numbers, and path types are grouped into the same path evidence group. The same path evidence group is used to characterize the constraint support evidence provided by the same candidate equipment model for the same task constraint record through the same type of path. When the path type is an attribute path, the corresponding path evidence group includes constraint attribute path evidence records of satisfying constraint attribute constraints for the same candidate equipment model for the same technical attribute constraint; when the path type is a performance path, the corresponding path evidence group includes constraint performance path evidence records of satisfying constraint performance constraints for the same candidate equipment model for the same performance index constraint. The path type constraint support value is used to characterize the overall support degree of the corresponding path evidence group for the corresponding task constraint record; the maximum value among the single-path constraint support values ​​is used as the path type constraint support value, so that the path evidence with the largest satisfaction margin is the main supporting path under that path type. When only one constraint-satisfied path evidence record exists in the corresponding path evidence group, the single-path constraint support value of that record is used as the corresponding path type constraint support value, and that record is identified as the primary supporting path for that path type. For each constraint-satisfied path evidence record, it is temporarily deleted from the list of constraint-satisfied path evidence records of the same candidate device model, constraint record number, and path type. The maximum single-path constraint support value among the remaining constraint-satisfied path evidence records is used as the path type constraint support value after removal. When no constraint-satisfied path evidence records remain, the path type constraint support value after removal is set to zero. Temporary deletion means excluding the current constraint-satisfied path evidence record from the corresponding path evidence group only when evaluating its path marginal contribution value. After evaluating the path marginal contribution value of the current constraint-satisfied path evidence record, it is restored, and the same process is performed on the next record to avoid the evaluation order of different path evidence records affecting the path marginal contribution value. When multiple path constraint support values ​​are the same and all are the maximum value in the same path constraint support group, temporarily deleting any one of the path constraint support values ​​will still result in the same maximum value among the remaining path constraint support values. In this case, the path type constraint support value after the removal of the current path constraint support value is equal to the path type constraint support value.The path type constraint support value is obtained by subtracting the path type constraint support value after removal from the path type constraint support value. When the path marginal contribution value is greater than zero, the path evidence record that satisfies the constraint is marked as a valid contributing path. The path marginal contribution value characterizes the extent to which the overall support level of the corresponding path evidence group decreases after the temporary deletion of the currently satisfying path evidence record. The larger the path marginal contribution value, the stronger the support effect of the currently satisfying path evidence record on the corresponding task constraint record. Before comparing the path marginal contribution value with zero, the path type constraint support value and the path type constraint support value after removal are processed with the same precision according to a preset number of decimal places. When the processed path marginal contribution value is greater than zero, the corresponding path evidence record that satisfies the constraint is marked as a valid contributing path; when the processed path marginal contribution value is less than or equal to zero, it is not marked as a valid contributing path. For example, if there are two constraint-satisfied performance path evidence records with single-path constraint support values ​​of 1.111 and 1.044 respectively for the same candidate equipment model, the same actual cycle time constraint record, and the same performance path type, then the path type constraint support value is 1.111. After temporarily deleting the constraint-satisfied performance path evidence record with a single-path constraint support value of 1.111, the path type constraint support value after removal is 1.044, and the corresponding path marginal contribution value is 0.067. Therefore, this constraint-satisfied performance path evidence record is marked as a valid contributing path. After temporarily deleting the constraint-satisfied performance path evidence record with a single-path constraint support value of 1.044, the path type constraint support value after removal is still 1.111, and the corresponding path marginal contribution value is zero. Therefore, it is not marked as a valid contributing path. When there is only one constraint-satisfied path evidence record in the same path evidence group, after temporarily deleting this constraint-satisfied path evidence record, there are no remaining constraint-satisfied path evidence records, and the path type constraint support value after removal is zero. In this case, the path marginal contribution value of this constraint-satisfied path evidence record is equal to the corresponding path type constraint support value. In this embodiment, path evidence is not the physical transmission path, but rather associated evidence used to characterize whether a candidate device model meets the corresponding task constraints. Grouping path evidence with the same candidate device model, constraint record number, and path type into the same group ensures that comparisons are made only within the same device model, selection requirement, and capability category, avoiding interference between different task constraints or different path types. Using the maximum value among single-path constraint support values ​​as the path type constraint support value ensures that the path evidence with the largest margin characterizes the main supporting capabilities of the candidate device model in the corresponding capability dimension, preventing the simple accumulation of similar duplicate path evidence and thus amplifying device capabilities.By temporarily deleting the current path evidence and comparing the path type constraint support values ​​before and after deletion, it can be determined whether the path evidence can be replaced by other similar paths. If the overall support level decreases after deletion, the path's marginal contribution value increases, indicating that the path has an irreplaceable supporting role for the corresponding task constraints, and thus it is regarded as an effective contributing path.

[0050] In this implementation scheme, by determining the main supporting paths under the same candidate device model, the same task constraint record, and the same path type, and evaluating the marginal contribution of the paths based on the change in the overall support level before and after the temporary removal of a path, it is possible to identify effective contributing paths that have a substantial effect on the satisfaction of task constraints. At the same time, for groups containing only evidence of a single path, the overall support level is directly represented by the single path constraint support value of that path, ensuring the consistency of the evaluation process under different path evidence levels, thereby avoiding redundant or alternative paths from having an excessive impact on the candidate device selection results.

[0051] Specifically, the process of optimizing and ranking candidate equipment models by combining effective contribution paths and path marginal contributions to obtain recommended candidate equipment models is as follows: For candidate equipment models marked as meeting the constraints, the path type constraint support value of each task constraint record is read and used as the task constraint support value of the corresponding task constraint record; specifically, for task constraint records where the constraint index category is technical attribute constraint, the path type constraint support value of the corresponding attribute path evidence group is read and used as the task constraint support value of that task constraint record; for task constraint records where the constraint index category is performance index constraint, the path type constraint support value of the corresponding performance path evidence group is read and used as the task constraint support value of that task constraint record. The task constraint support value is used to characterize the degree of support of the candidate equipment model for the corresponding single task constraint. For each candidate device model and each task constraint record, when a valid contribution path exists, the valid contribution path with the largest marginal contribution value is selected as the critical contribution path for the corresponding task constraint record, and the marginal contribution value of the critical contribution path is used as the constraint valid contribution value for the corresponding task constraint record. When no valid contribution path exists, the constraint valid contribution value is set to zero. Specifically, when multiple valid contribution paths with the same maximum marginal contribution value exist for a corresponding task constraint record, one of these valid contribution paths can be selected as the critical contribution path according to the device number order. The constraint valid contribution value characterizes the degree of irreplaceable support that the current candidate device model has for the corresponding task constraint. When no valid contribution path exists, it indicates that although the corresponding task constraint has evidence of satisfying the constraint path, this evidence can be replaced by other path evidence in the same group, and therefore the constraint valid contribution value is set to zero. For each candidate device model, the task constraint support value and effective contribution value corresponding to performance index constraints and technical attribute constraints are summarized according to the constraint index category. For each task constraint support value under the same constraint index category, minimum value extraction and summation are performed to obtain the minimum constraint support value and total constraint support value for the corresponding category. The effective contribution values ​​for each constraint under the same constraint index category are then summed to obtain the total effective contribution value for the corresponding category. Specifically: the task constraint record for the constraint index category of performance index constraint is read, and the minimum value among the task constraint support values ​​is taken as the minimum performance constraint support value; the support values ​​of each task constraint are added to obtain the total performance constraint support value; the effective contribution values ​​of each constraint are added to obtain the total effective contribution value for performance paths. The minimum performance constraint support value is used to characterize the lowest support level of the candidate device model among various performance index constraints, avoiding the masking effect of insufficient support for a single performance index by higher support levels for other performance indices; the total performance constraint support value is used to characterize the overall support level of the candidate device model for all performance index constraints; and the total effective contribution value for performance paths is used to characterize the cumulative irreplaceable support level of critical performance paths among various performance index constraints.Read the task constraint records whose constraint index category is technical attribute constraint, and take the minimum value among the support values ​​of each task constraint as the minimum support value of the attribute constraint; add up the support values ​​of each task constraint to obtain the total support value of the attribute constraint; add up the effective contribution values ​​of each constraint to obtain the total effective contribution value of the attribute path; whereby, the minimum support value of the attribute constraint is used to characterize the minimum support level of the candidate equipment model in each technical attribute constraint; the total support value of the attribute constraint is used to characterize the overall support level of the candidate equipment model for all technical attribute constraints; and the total effective contribution value of the attribute path is used to characterize the cumulative irreplaceable support level of the key attribute path in each technical attribute constraint. When the task constraint record set contains both performance indicator constraints and technical attribute constraints, determine the minimum support value of the performance constraint, the total support value of the performance constraint, the total effective contribution value of the performance path, the minimum support value of the attribute constraint, the total support value of the attribute constraint, and the total effective contribution value of the attribute path. When the task constraint record set contains only performance indicator constraints, determine only the minimum support value of the performance constraint, the total support value of the performance constraint, and the total effective contribution value of the performance path. When the task constraint record set contains only technical attribute constraints, determine only the minimum support value of the attribute constraint, the total support value of the attribute constraint, and the total effective contribution value of the attribute path. For the task constraint record set, candidate equipment models marked as satisfying the constraints are sorted in descending order: minimum support value of performance constraints, total support value of performance constraints, total effective contribution value of performance paths, minimum support value of attribute constraints, total support value of attribute constraints, and total effective contribution value of attribute paths. This yields the optimal ranking result for candidate equipment. Specifically, when the task constraint record set contains both performance index constraints and technical attribute constraints, the ranking is further determined by the following order: minimum support value of performance constraints, total support value of performance constraints, total effective contribution value of performance paths, minimum support value of attribute constraints, total support value of attribute constraints, and total effective contribution value of attribute paths. The process involves several steps: First, sorting the task constraint records in lexicographical descending order. When the task constraint record set contains only performance indicator constraints, sorting is performed in descending order based on the minimum support value of the performance constraint, the total support value of the performance constraint, and the total effective contribution value of the performance path. Second, when the task constraint record set contains only technical attribute constraints, sorting is performed in descending order based on the minimum support value of the attribute constraint, the total support value of the attribute constraint, and the total effective contribution value of the attribute path. Lexicographical descending sorting means first comparing the previous sorting indicator; if the previous sorting indicators are different, the candidate device model with the larger previous sorting indicator is ranked first; if the previous sorting indicators are the same, the next sorting indicator is compared. When all corresponding sorting indicators are the same, the candidate device models are preferably sorted according to a preset coding order to ensure that the candidate device selection ranking result has a unique first candidate device model. The first candidate device model in the candidate device selection ranking result is used as the recommended candidate device model, and a device selection result record is generated. The device selection result record includes: the selection task number, the recommended candidate device model, and the recommended candidate device type.Specifically, the candidate device type is retrieved from the corresponding candidate device parameter data based on the recommended candidate device model, and used as the recommended candidate device type. The device selection result record may also include the candidate device selection ranking result, which is used to show the ranking position of the candidate device models that satisfy each constraint. When there is no candidate device model that satisfies the constraint, a "no recommendation result" mark is generated, and the record number of the unsatisfied constraint and the reason for the missing candidate are output. No recommended candidate device model is generated.

[0052] In this implementation plan, by separately evaluating the minimum support level, overall support level, and irreplaceable support level of the critical path for each performance index constraint and technical attribute constraint of the candidate equipment model, and by adopting the corresponding lexicographical order sorting rule according to the task constraint category, it is possible to avoid the insufficient support of a single constraint being masked by other high support indicators. At the same time, by combining the critical path with the prominent marginal contribution of the path that has a substantial effect on the specific task constraint, and by eliminating the situation of tied sorting through the preset coding order, the relevance, stability and traceability of the candidate equipment selection results can be improved.

[0053] Reference Figure 2 As shown, the second aspect of the present invention provides an object selection system based on a large model fusion knowledge graph, applied to the aforementioned object selection method based on a large model fusion knowledge graph, comprising: a data acquisition graph construction module, used to collect candidate device parameter data and operating performance data, preprocess the candidate device parameter data and operating performance data, and construct an device selection knowledge graph; a task parsing path evidence module, used to receive device selection task information, and use a pre-trained language understanding and structured output model based on the Transformer architecture to parse the device selection task information, and based on the parsing results of the device selection task information, filter attribute path evidence and performance path evidence in the device selection knowledge graph, and determine the task constraint satisfaction of each path evidence in combination with the parsing results of the device selection task information; a path contribution object selection module, used to evaluate the degree of support for single path constraints for path evidence that satisfies task constraints, evaluate the path marginal contribution based on the degree of support for single path constraints, and filter effective contribution paths; and combine the effective contribution paths and the path marginal contribution to optimize and rank the candidate device models to obtain recommended candidate device models.

[0054] In this implementation plan, the technical parameters and actual operating performance of candidate equipment are structurally correlated with the equipment selection task constraints through the collaborative processing of the data acquisition map construction module, the task analysis path evidence module, and the path contribution object selection module. Furthermore, by distinguishing between attribute paths and performance paths and evaluating the degree of support for path constraints and marginal contributions, the candidate equipment models are ranked in a targeted manner, thereby improving the matching, stability, and traceability between the equipment selection results and the target process scenario and specific capability requirements.

[0055] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0056] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. As those skilled in the art will understand, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An object optimization method based on large-scale model fusion knowledge graph, characterized in that, Includes the following steps: S1. Collect candidate equipment parameter data and operating performance data, preprocess the candidate equipment parameter data and operating performance data, and construct a knowledge graph for equipment optimization. S2 receives equipment selection task information and uses a pre-trained language understanding and structured output model based on the Transformer architecture to parse the equipment selection task information. Based on the parsing results of the equipment selection task information, it filters attribute path evidence and performance path evidence in the equipment optimization knowledge graph and determines the task constraint satisfaction of each path evidence in combination with the parsing results of the equipment selection task information. S3. For path evidence that satisfies task constraints, evaluate the degree of support for single path constraints, assess the marginal contribution of the path based on the degree of support for single path constraints, and screen effective contributing paths; combine effective contributing paths and path marginal contributions to optimize and rank candidate device models to obtain recommended candidate device models.

2. The object optimization method based on large-model fusion knowledge graph as described in claim 1, characterized in that, The specific process for collecting candidate device parameter data and operational performance data is as follows: Candidate equipment parameter data is collected through the equipment asset management system. The candidate equipment parameter data includes: candidate equipment model, candidate equipment type and technical attribute record. The technical attribute record includes: technical attribute identifier, technical attribute name, technical attribute value and technical attribute unit. Operational performance data is collected through equipment status records and process task records in the manufacturing execution system. The operational performance data includes: equipment number, equipment model, process scenario number, and actual performance index record. The actual performance index record includes: performance index identifier, performance index name, performance index value, and performance index unit.

3. The object optimization method based on large-scale model fusion knowledge graph as described in claim 2, characterized in that, The specific process of preprocessing candidate device parameter data and operational performance data to construct a device optimization knowledge graph is as follows: Read the preset technical attribute standardization dictionary and performance index standardization dictionary; based on the technical attribute standardization dictionary, standardize the technical attribute identifier, technical attribute name, technical attribute value and technical attribute unit in each technical attribute record; Based on a standardized dictionary of performance indicators, the performance indicator identifiers, names, values, and units in each actual performance indicator record are standardized. Delete candidate device parameter data and operating performance data that contain empty fields, and remove duplicate technical attribute records and actual performance index records; The candidate equipment model is used as the equipment object node, the equipment number as the equipment instance node, the equipment number and process scenario number together as the equipment operating condition node, the technical attribute identifier, technical attribute value and technical attribute unit together as the technical attribute node, and the performance index identifier, performance index value and performance index unit together as the performance index node. The model attribution relationship between the equipment object node and the equipment instance node is established, the technical attribute relationship between the equipment object node and the technical attribute node is established, the scenario operation relationship between the equipment instance node and the equipment operating condition node is established, the performance result relationship between the equipment operating condition node and the performance index node is established, and an equipment optimization knowledge graph is constructed.

4. The object optimization method based on large-model fusion knowledge graph as described in claim 1, characterized in that, The specific process of receiving device selection task information and parsing the device selection task information using a pre-trained language understanding and structured output model based on the Transformer architecture is as follows: Receive equipment selection task information, including: selection task number and task requirement description; input the task requirement description into a pre-trained language understanding and structured output model based on the Transformer architecture to obtain the target equipment type, target process scenario number, and task constraint record set. Each task constraint record includes: constraint record number, constraint index category, constraint index name, constraint value, constraint unit, and constraint comparison relationship; constraint index categories include technical attribute constraints and performance index constraints. Based on the constraint index category of the task constraint record, query the corresponding technical attribute standardization dictionary or performance index standardization dictionary to obtain the standard index identifier, standard index name, and standard index unit of the corresponding category. Standardize the constraint index name, constraint value, and constraint unit, and use the corresponding standard index identifier as the constraint index identifier of the task constraint record.

5. The object optimization method based on large model fusion knowledge graph according to claim 2, characterized in that, The specific process for filtering attribute path evidence and performance path evidence in the equipment optimization knowledge graph based on the analysis results of equipment selection task information is as follows: The candidate equipment set is obtained by filtering candidate equipment models that are the same type as the target equipment from the equipment selection knowledge graph; Obtain the model of each candidate device in the candidate device set; for the task constraint record whose constraint index category is technical attribute constraint, read the technical attribute record; when the technical attribute identifier in the technical attribute record is the same as the constraint index identifier in the task constraint record, establish an attribute path evidence record between the candidate device model, technical attribute identifier, technical attribute value and constraint record number. For task constraint records with performance index constraints as the constraint index category, read the operating performance data where the equipment model is the same as the candidate equipment model and the process scenario number is the same as the target process scenario number. When the performance index identifier in the actual performance index record is the same as the constraint index identifier in the task constraint record, establish a performance path evidence record between the candidate equipment model, equipment number, process scenario number, performance index identifier, performance index value and constraint record number. Associate the attribute path evidence record and the performance path evidence record with the corresponding task constraint record, respectively.

6. The object optimization method based on large model fusion knowledge graph according to claim 5, characterized in that, The specific process for determining the satisfaction of task constraints for each path's evidence based on the analysis results of equipment selection task information is as follows: For each attribute path evidence record and performance path evidence record, the technical attribute value or performance indicator value is matched with the constraint value according to the constraint comparison relationship in the corresponding task constraint record. When the matching judgment result satisfies the constraint comparison relationship, a constraint-satisfied flag is generated; when the matching judgment result does not satisfy the constraint comparison relationship, a constraint-unsatisfied flag is generated. The constraint comparison relationship includes greater than, greater than or equal to, less than, less than or equal to, and equal to. Write the path type and constraint flags into the corresponding path evidence record. The path type includes attribute path and performance path.

7. The object optimization method based on large model fusion knowledge graph according to claim 6, characterized in that, The specific process for evaluating the degree of support for single-path constraints based on path evidence that satisfies task constraints is as follows: Read attribute path evidence records and performance path evidence records; when there is a constraint-satisfying marker in the path evidence record, the corresponding path evidence record is used as a constraint-satisfying path evidence record; and when there is a path evidence record that satisfies the constraint marker for each task constraint record in the task constraint record set for the candidate device model, the candidate device model is marked as a constraint-satisfying candidate device model. For each evidence record that satisfies the constraint path, when the constraint comparison relationship is greater than or greater than or equal to, the corresponding indicator value is subtracted from the constraint value, and then divided by the sum of the absolute value of the constraint value and the preset minimum positive number to obtain the path constraint margin value; when the constraint comparison relationship is less than or less than or equal to, the constraint value is subtracted from the corresponding indicator value, and then divided by the sum of the absolute value of the constraint value and the preset minimum positive number to obtain the path constraint margin value; when the constraint comparison relationship is equal, the path constraint margin value is set to zero; and the sum of 1 and the path constraint margin value is used as the single path constraint support value.

8. The object optimization method based on large model fusion knowledge graph according to claim 7, characterized in that, The specific process of evaluating the marginal contribution of a path based on the support level of a single path constraint and selecting effective contributing paths is as follows: Read the evidence records of constrained paths that are identical in terms of candidate device model, constraint record number, and path type, and take the maximum value among the single path constraint support values ​​as the path type constraint support value for the corresponding candidate device model for the corresponding task constraint record; For each constraint-satisfying path evidence record, the constraint-satisfying path evidence record is temporarily deleted from the constraint-satisfying path evidence records of the same candidate device model, the same constraint record number, and the same path type, and the maximum single-path constraint support value in the remaining constraint-satisfying path evidence records is used as the path type constraint support value after removal; When there are no remaining evidence records of paths that satisfy the constraints, the support value for the path type constraint after removal will be set to zero. Subtracting the path type constraint support value after removal from the path type constraint support value yields the path marginal contribution value of the path evidence record that satisfies the constraint; when the path marginal contribution value is greater than zero, the path evidence record that satisfies the constraint is marked as a valid contributing path.

9. The object optimization method based on large model fusion knowledge graph according to claim 8, characterized in that, The specific process of combining effective contribution paths and path marginal contributions to optimize and rank candidate device models to obtain recommended candidate device models is as follows: For candidate device models that are marked as satisfying the constraints, read the path type constraint support value of each task constraint record and use it as the task constraint support value of the corresponding task constraint record. For each candidate device model and each task constraint record, when a valid contribution path exists, the valid contribution path with the largest marginal contribution value is taken as the key contribution path of the corresponding task constraint record, and the marginal contribution value of the key contribution path is taken as the constraint valid contribution value of the corresponding task constraint record; when no valid contribution path exists, the constraint valid contribution value is set to zero. For each candidate device model, the task constraint support value and effective contribution value corresponding to the performance index constraint and technical attribute constraint are summarized according to the constraint index category. The minimum value is extracted and summed for each task constraint support value under the same constraint index category to obtain the minimum constraint support value and total constraint support value of the corresponding category. The effective contribution value of each constraint under the same constraint index category is summed to obtain the total effective contribution value of the path of the corresponding category. For the task constraint record set, the candidate device models marked as satisfying the constraints are sorted in descending order according to the minimum support value of performance constraints, the total support value of performance constraints, the total effective contribution value of performance paths, the minimum support value of attribute constraints, the total support value of attribute constraints, and the total effective contribution value of attribute paths, to obtain the candidate device optimization ranking result; The first candidate device model in the candidate device selection ranking results is taken as the recommended candidate device model, and a device selection result record is generated.

10. An object optimization system based on a large-scale model fusion knowledge graph, characterized in that, include: The data acquisition graph construction module is used to collect candidate equipment parameter data and operating performance data, preprocess the candidate equipment parameter data and operating performance data, and construct a knowledge graph for equipment optimization. The task parsing path evidence module is used to receive equipment selection task information and use a pre-trained language understanding and structured output model based on the Transformer architecture to parse the equipment selection task information. Based on the parsing results of the equipment selection task information, attribute path evidence and performance path evidence are filtered in the equipment optimization knowledge graph. The task constraint satisfaction of each path evidence is determined by combining the parsing results of the equipment selection task information. The path contribution object selection module is used to evaluate the degree of support for single path constraints for path evidence that meets task constraints, evaluate the marginal contribution of the path based on the degree of support for single path constraints, and screen effective contributing paths; combining the effective contributing paths and the marginal contribution of the paths, the candidate device models are selected and ranked to obtain recommended candidate device models.

Citation Information

Patent Citations

  • Knowledge graph enhancement-based black and grey production management large model optimization method, knowledge graph data structure and knowledge graph enhancement-based black and grey production management large model optimization device

    CN120822636A