BOM data multi-dimensional analysis system and analysis method based on large language model

Through a multi-dimensional analysis system based on large language model, the data silos and insufficient intelligence of the BOM analysis system are solved, and multi-dimensional impact assessment and analysis are realized, the adaptability and intelligence of the system are improved, and the digital transformation of the manufacturing industry is supported.

CN119004372APending Publication Date: 2024-11-22SHENGTAI RENHE INTELLIGENT TECH (SHENZHEN) CO LTD
View PDF 0 Cites 25 Cited by

Patent Information

Application Number
CN202411071397.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

When facing the intelligent matching of complex data and multi-dimensional impact assessment, existing BOM analysis systems have problems such as data silos, single analysis dimensions, insufficient intelligence, poor scalability and adaptability, making it difficult for enterprises to make timely, comprehensive and accurate decision making.

Method used

Using a multi-dimensional analysis system based on large language models, multi-source heterogeneous data fusion, adaptive domain knowledge graph and dynamic context enhancement prompt module, combined with deep learning and natural language processing technology, the multi-dimensional analysis of data and visual display of results are realized. .

Benefits of technology

Achieve multi-dimensional impact assessment and analysis, enhance the adaptability and scalability of the system, improve intelligence, and provide a highly intelligent, adaptive and continuously evolved BOM data analysis and impact assessment system, support intelligent manufacturing decisions, and promote Digital transformation and intelligent upgrading of manufacturing industry.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004372A_ABST
    Figure CN119004372A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-dimensional analysis method and analysis method for BOM data based on a large language model, which can solve the problem of data islands in the traditional technology by integrating multi-source data and performing BOM analysis according to the integrated multi-source data. Meanwhile, during processing, knowledge graph technology is applied to perform knowledge representation and reasoning of input data, and advanced AI technologies such as deep learning and natural language processing are combined at the same time to generate a corresponding multi-dimensional analysis prompt question and answer text; then, inputting the data into a large language model to generate a result; and finally, performing multi-dimensional analysis and result fusion to obtain a multi-modal analysis result. On the basis, deep multi-dimensional influence evaluation and analysis are realized, the adaptability and expandability of the system are enhanced, and the intelligence of the system is greatly improved through application of advanced AI technologies such as deep learning and natural language processing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing and analysis based on artificial intelligence, and specifically relates to a multidimensional analysis system and analysis method of BOM data based on a large language model. Background Art

[0002] Bill of materials (BOM) analysis and impact assessment play a key role in modern manufacturing and supply chain management. They can provide auxiliary basis for corporate decision-making, thereby helping companies adjust their operating strategies. At present, traditional BOM analysis systems mainly include: BOM modules in ERP systems, professional BOM analysis software, supply chain optimization systems based on machine learning, knowledge management and decision support systems based on knowledge graphs, traditional decision support systems, PLM (product lifecycle management) systems, industrial Internet of Things platforms, etc. In this way, companies can use the aforementioned systems to complete BOM analysis and impact assessment.

[0003] In the actual application process, although the aforementioned existing technologies have made achievements in their respective fields, they still have obvious limitations when facing complex BOM data intelligent matching and multi-dimensional impact assessment. Among them, the aforementioned existing technologies mainly have data islands (difficult to integrate data from different sources and types), single analysis dimensions, insufficient intelligence, poor scalability and adaptability, and lack of deep semantic understanding and complex reasoning capabilities. These problems make it difficult for enterprises to make timely, comprehensive and accurate judgments when facing rapidly changing market environments and complex decision-making scenarios, affecting the intelligent transformation of the manufacturing industry and the improvement of its competitiveness. Based on this, how to provide a multi-dimensional BOM data analysis system that can realize multi-source data fusion and has deep semantic understanding and complex reasoning capabilities has become a problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to provide a multidimensional analysis system and analysis method for BOM data based on a large language model, so as to solve the problems existing in the prior art, such as data islands, single analysis dimension, insufficient intelligence, poor scalability and adaptability, and lack of deep semantic understanding and complex reasoning capabilities.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, a multi-dimensional analysis system for BOM data based on a large language model is provided, comprising:

[0007] A multi-source heterogeneous data fusion module, wherein the multi-source heterogeneous data fusion module is used to obtain multiple BOM data of a target product, and perform data integration processing on each BOM data to obtain multiple processed BOM data of the same format;

[0008] An adaptive domain knowledge graph module, wherein the adaptive domain knowledge graph module stores a BOM domain knowledge graph, wherein the adaptive domain knowledge graph module is used to receive each processed BOM data transmitted by the multi-source heterogeneous data fusion module, and perform data matching on each processed BOM data based on the BOM domain knowledge graph to obtain a data matching result;

[0009] A dynamic context enhancement prompt module, the dynamic context enhancement prompt module is used to generate a multi-dimensional analysis prompt question and answer text based on the data matching result, and transmit the multi-dimensional analysis prompt question and answer text to the large language model module;

[0010] A large language model module, wherein the large language model module stores a BOM large language model, wherein the large language model module is used to input the multidimensional analysis prompt question and answer text into the BOM large language model to obtain a question and answer result corresponding to the multidimensional analysis prompt question and answer text, and transmit the question and answer result to the multidimensional analysis module;

[0011] The multidimensional analysis module is used to perform multidimensional analysis on the question and answer results to obtain multidimensional analysis results, and to fuse the multidimensional analysis results to obtain multimodal analysis results of the target product, and to visualize the multimodal analysis results, wherein the multidimensional analysis includes analysis on the cost dimension, performance dimension, safety dimension and supply chain dimension of the target product.

[0012] Based on the above disclosed content, the multidimensional analysis system of BOM data provided by the present invention, when performing BOM data analysis, first performs data integration processing on multiple BOM data of the target product through a multi-source heterogeneous data fusion module to obtain multiple processed BOM data with the same format; then, the BOM domain knowledge graph is used to perform data matching on the multiple processed BOM data, so as to enhance the understanding ability of BOM data based on the knowledge graph and expand its data range; then, the above matching results are input into the dynamic context enhancement prompt module to perform semantic understanding of the context, thereby generating input prompt information of the BOM large language model (i.e., multidimensional analysis prompt question and answer text); then, the above generated prompt information is input into the BOM large language model, and the corresponding question and answer results can be obtained; finally, based on the question and answer results, multidimensional analysis and processing of the target product is performed, and the multidimensional analysis results are fused to obtain the multimodal analysis results of the target product.

[0013] Through the above design, the multidimensional analysis system of BOM data provided by the present invention can solve the data island problem in traditional technology by integrating multi-source data and performing BOM analysis based on the integrated multi-source data; at the same time, the present invention applies knowledge graph technology to perform knowledge representation and reasoning of input data during processing, and combines advanced AI technologies such as deep learning and natural language processing to generate corresponding multidimensional analysis prompt question and answer texts; then, it is input into the large language model to generate results; finally, multidimensional analysis and result fusion are performed to obtain multimodal analysis results; based on this, the present invention realizes in-depth multidimensional impact assessment and analysis, enhances the adaptability and scalability of the system, and the application of advanced AI technologies such as deep learning and natural language processing greatly improves the intelligence of the system; thus, the present invention provides a highly intelligent, adaptive and continuously evolving BOM data analysis and impact assessment system, creating a new intelligent manufacturing decision support paradigm, thereby providing strong technical support for the digital transformation and intelligent upgrading of the manufacturing industry.

[0014] In one possible design, the dynamic context-enhanced prompt module includes: an input parser, a context analyzer, a prompt template library, a dynamic prompt generator, and a prompt validator;

[0015] An input parser, used for receiving a plurality of processed BOM data transmitted by the multi-source heterogeneous data fusion module, and extracting key information from each processed BOM data to obtain key information corresponding to each processed BOM data;

[0016] A context analyzer, which is used to use a deep learning model to perform context encoding processing on each key information to extract context semantic information of each key information, and match a prompt template corresponding to the target product from a prompt template library based on each context semantic information;

[0017] A dynamic prompt generator, used to extract historical prompt information from the historical prompt buffer area, and to extract industry knowledge of the target product from the industry rule engine;

[0018] The dynamic prompt generator is further used to generate an initial multi-dimensional analysis prompt question and answer text based on the contextual semantic information of each key information, the historical prompt information and the industry knowledge, and adopts the prompt generation strategy and the matched prompt template;

[0019] The prompt verifier is used to verify the initial multidimensional analysis prompt question and answer text, and after the verification passes, transmit the initial multidimensional analysis prompt question and answer text as the multidimensional analysis prompt question and answer text to the large language model module.

[0020] In one possible design, the dynamic context enhanced prompt module further includes: a user feedback interface, a performance evaluation unit, and a reinforcement learning optimizer;

[0021] A user feedback interface, used to obtain, in response to human-computer interaction, first feedback information of the user on the multidimensional analysis prompt question and answer text and second feedback information on the question and answer result corresponding to the multidimensional analysis prompt question and answer text, and send the first feedback information and the second feedback information to the performance evaluation unit;

[0022] a performance evaluation unit, configured to receive first feedback information and second feedback information sent by the user feedback interface, and obtain prompt template evaluation information based on the first feedback information and the second feedback information, so as to update each prompt template in the prompt template library according to the prompt template evaluation information;

[0023] A performance evaluation unit, configured to receive the question-answering result corresponding to the multi-dimensional analysis prompt question-answering text sent by the large language model module, and to perform a performance evaluation on the dynamic prompt generator based on the question-answering result to obtain an evaluation result;

[0024] A performance evaluation unit, further configured to send the evaluation result to the reinforcement learning optimizer;

[0025] A reinforcement learning optimizer is used to optimize the prompt generation strategy of the dynamic prompt generator based on the evaluation result and using a reinforcement learning strategy, so as to complete the feedback learning of the dynamic prompt generator after the optimization.

[0026] In one possible design, the multidimensional analysis module includes: a multidimensional impact analysis unit and a multimodal result interpretation and visualization unit;

[0027] A multi-dimensional impact analysis unit, used to obtain evaluation parameters, wherein the evaluation parameters include analysis dimensions;

[0028] A multi-dimensional impact analysis unit, used to match analysis models corresponding to various analysis dimensions from an analysis model library according to the evaluation parameters, and to receive the question-answering results transmitted by the large language model module, and input the question-answering results into each matched analysis model to obtain a multi-dimensional analysis result of the target product;

[0029] The multi-dimensional impact analysis unit is further used to integrate the multi-dimensional analysis results to obtain the comprehensive analysis results of the target product, and send the comprehensive analysis results to the multi-modal result interpretation and visualization unit;

[0030] The multimodal result interpretation and visualization unit is used to generate visualization data, summary text and result interpretation information corresponding to the comprehensive analysis result according to the comprehensive analysis result, and perform multimodal fusion processing on the visualization data, summary text and result interpretation information to obtain the multimodal analysis result of the target product after the multimodal fusion processing.

[0031] In a possible design, the multimodal result interpretation and visualization unit includes: an output parsing unit, a key information extraction unit, a data classification and structuring unit, a visualization data generation unit, a summary text generation unit, an AI interpretation unit, and a multimodal display integration unit;

[0032] An output parsing unit, used to parse the comprehensive analysis result to obtain parsing information corresponding to the comprehensive analysis result, and transmit the parsing information to the key information extraction unit;

[0033] A key information extraction unit, configured to perform key information extraction processing on the parsed information to obtain a plurality of key information corresponding to the parsed information, and transmit each key information corresponding to the parsed information to a data classification and structuring unit;

[0034] A data classification and structuring unit, used to classify each key information corresponding to the parsed information to obtain a plurality of classified information, and to structure each classified information to obtain a plurality of structured classified information;

[0035] A visualization data generating unit, configured to generate visualization data of each structured classification information according to each structured classification information, and transmit the visualization data of each structured classification information to the multimodal display integration unit, wherein the visualization data of any structured classification information includes a static chart, an interactive chart and / or a network relationship map;

[0036] A summary text generation unit is used to use a natural language processing algorithm to perform text summary extraction processing on each structured classification information, obtain summary texts corresponding to each structured classification information, and transmit each summary text to the multimodal display integration unit;

[0037] The AI ​​interpretation unit is used to perform feature importance analysis on each structured classification information to obtain a feature importance analysis result of each structured classification information, and transmit the feature importance analysis result of each structured classification information as result interpretation information of each structured classification information to the multimodal display integration unit;

[0038] The multimodal display integration unit is used to perform multimodal fusion processing on the visual data, summary text and result interpretation information corresponding to each structured classification information, so as to obtain the multimodal analysis result of the target product after the multimodal fusion processing.

[0039] In a possible design, if there is a missing information problem in the multi-dimensional analysis process, the multi-dimensional impact analysis unit is used to identify the missing information, generate a missing information query question based on the missing information, and visualize the missing information query question;

[0040] A multi-dimensional impact analysis unit, configured to obtain, in response to human-computer interaction, feedback input information of a user for the missing information query question, and analyze the feedback input information to obtain an input analysis result;

[0041] A multi-dimensional impact analysis unit, used to adjust the model parameters of the analysis model corresponding to each analysis dimension based on the input analysis result to obtain an adjusted analysis model;

[0042] The multi-dimensional impact analysis unit is also used to re-input the input analysis results and the question and answer results into each adjusted analysis model to obtain a multi-dimensional analysis result of the target product.

[0043] In a possible design, the multi-dimensional impact analysis unit is used to generate a decision suggestion for the target product based on the comprehensive analysis result, and to visually display the decision suggestion; and

[0044] It is also used to update each analysis model in the analysis model library according to the comprehensive analysis results and user feedback information, so as to complete the continuous reinforcement learning of each analysis model during use.

[0045] In a possible design, the multi-source heterogeneous data fusion module includes: a data cleaning and standardization unit, an entity recognition and relationship extraction unit, a semantic integration unit, a data integration unit, and a data quality assessment unit;

[0046] A data cleaning and standardization unit is used to perform data cleaning and data standardization on each BOM data of the target product in turn to obtain a variety of standardized BOM data, and send the multiple standardized BOM data to the entity recognition and relationship extraction unit;

[0047] An entity recognition and relationship extraction unit is used to perform entity recognition and relationship extraction processing on each standardized BOM data to obtain the entity relationship of each standardized BOM data;

[0048] A semantic integration unit is used to perform semantic integration processing on each standardized BOM data based on the entity relationship of each standardized BOM data to obtain standardized BOM data after multiple semantic integration;

[0049] A data integration unit is used to perform format conversion processing on standardized BOM data after multiple semantic integration, so as to obtain multiple processed initial BOM data with the same format after the format conversion processing;

[0050] The data quality evaluation unit is used to perform quality evaluation on the multiple processed initial BOM data, and after the quality evaluation is passed, use the multiple processed initial BOM data as the processed BOM data of the target product.

[0051] In a possible design, the adaptive domain knowledge graph module includes: a knowledge graph updating unit and a feedback learning unit;

[0052] The knowledge graph updating unit is used to obtain knowledge graph update data and perform data preprocessing on the knowledge graph update data, so as to obtain named entities corresponding to the knowledge graph update data after the data preprocessing;

[0053] The knowledge graph updating unit is further used to update the BOM domain knowledge graph according to the named entities corresponding to the knowledge graph update data to obtain an updated initial BOM domain knowledge graph, and to perform knowledge integration and verification on the updated initial BOM domain knowledge graph, so as to accept the update of the BOM domain knowledge graph after the knowledge integration and verification to obtain an updated BOM domain knowledge graph, so as to use the updated BOM domain knowledge graph for data matching;

[0054] A feedback learning unit is used to perform knowledge application and reasoning processing on the updated BOM domain knowledge graph to obtain knowledge reasoning results;

[0055] A feedback learning unit is used to obtain the third feedback information of the updated BOM domain knowledge graph in the multidimensional analysis process, and evaluate the importance of each node in the updated BOM domain knowledge graph according to the third feedback information, so as to adjust the weight of each node in the updated BOM domain knowledge graph according to the importance evaluation result to obtain weight adjustment information;

[0056] The feedback learning unit is also used to obtain the question and answer results output by the large language model module, and input the knowledge reasoning results, weight adjustment results and / or question and answer results into the knowledge graph updating unit, so as to optimize the updated BOM domain knowledge graph through the knowledge graph updating unit, so as to complete the feedback learning of the updated BOM domain knowledge graph after optimization.

[0057] In a second aspect, an analysis method of the multidimensional analysis system of BOM data based on a large language model is provided based on the first aspect or any possible design of the first aspect, wherein the method comprises:

[0058] Acquire multiple BOM data of target products, and perform data integration processing on each BOM data to obtain multiple processed BOM data with the same format;

[0059] Based on the BOM domain knowledge graph, data matching is performed on each processed BOM data to obtain data matching results;

[0060] Generate a multi-dimensional analysis prompt question and answer text according to the data matching result;

[0061] Input the multidimensional analysis prompt question and answer text into the BOM large language model to obtain the question and answer result corresponding to the multidimensional analysis prompt question and answer text;

[0062] Performing multi-dimensional analysis and processing on the question and answer results to obtain multi-dimensional analysis results;

[0063] The multi-dimensional analysis results are fused to obtain a multimodal analysis result of the target product, and the multimodal analysis result is visualized, wherein the multi-dimensional analysis processing includes analysis processing of the cost dimension, analysis processing of the performance dimension, analysis processing of the safety dimension, and analysis processing of the supply chain dimension of the target product.

[0064] Beneficial effects:

[0065] (1) The multidimensional analysis system of BOM data provided by the present invention can solve the data island problem in traditional technology by integrating multi-source data and performing BOM analysis based on the integrated multi-source data. At the same time, the present invention applies knowledge graph technology to perform knowledge representation and reasoning of input data during processing, and combines advanced AI technologies such as deep learning and natural language processing to generate corresponding multidimensional analysis prompt question and answer text; then, it is input into a large language model to generate results; finally, multidimensional analysis and result fusion are performed to obtain multimodal analysis results; based on this, the present invention realizes in-depth multidimensional impact assessment and analysis, enhances the adaptability and scalability of the system, and the application of advanced AI technologies such as deep learning and natural language processing greatly improves the intelligence of the system; thus, the present invention provides a highly intelligent, adaptive and continuously evolving BOM data analysis and impact assessment system, creating a new intelligent manufacturing decision support paradigm, thereby providing strong technical support for the digital transformation and intelligent upgrading of the manufacturing industry.

[0066] (2) The efficiency and accuracy of BOM analysis can be significantly improved. The present invention shortens the BOM change analysis time by an average of 60% through automated data processing and intelligent analysis, and applies advanced machine learning algorithms during the analysis process, thereby significantly improving the accuracy of the analysis.

[0067] (3) It can realize multi-dimensional and comprehensive impact assessment. The present invention analyzes data from multiple dimensions such as cost, performance, safety, and supply chain, thereby providing a comprehensive 360-degree assessment result. At the same time, by integrating multi-source data, it can discover potential impacts that are easily overlooked by traditional methods, thereby reducing decision-making risks.

[0068] (4) Ability to enhance decision support capabilities; the present invention can provide data-driven decision recommendations to assist management in making smarter and more objective decisions; and through visualization and interactive interfaces, complex analysis results can be easily understood and operated, thereby improving decision-making efficiency.

[0069] (5) It can optimize supply chain management. Specifically, the present invention can help enterprises optimize inventory levels through accurate impact assessment, thereby reducing inventory costs, and can identify potential supply chain risks in advance, thereby improving supply chain resilience.

[0070] (6) It can promote product innovation and performance improvement. The present invention can accelerate product iteration and innovation cycles by comprehensively analyzing the impact of BOM changes on product performance. At the same time, it can output accurate evaluation results, thus helping enterprises maintain their technological leadership in the competition.

[0071] (7) It can enhance the cost control capability of enterprises; the present invention can help enterprises save costs through accurate cost impact analysis, and the long-term cost forecasting function can help enterprises formulate more reasonable pricing strategies, thereby improving profit margins.

[0072] (8) Enhance knowledge management and experience accumulation: The present invention can continuously accumulate and optimize domain knowledge through adaptive knowledge graphs; in this way, it can form unique knowledge assets for the enterprise; at the same time, it can also reduce dependence on specific experts.

[0073] (9) Improved cross-departmental collaboration efficiency: The present invention provides a unified analysis platform, which promotes collaboration among engineering, procurement, finance and other departments, reduces communication costs, and standardizes analysis processes and report formats, which can improve the consistency and efficiency of cross-departmental decision-making.

[0074] (10) Strong adaptability and wide application range: The present invention has high scalability and adaptability and can be easily expanded to different industries and product types. Through continuous learning and optimization, the system performance will continue to improve with use, thus providing long-term value to the enterprise.

[0075] (11) Enhanced enterprise risk management capabilities: Through multi-dimensional analysis, the present invention can help enterprises identify and quantify potential risks in advance, thereby improving the risk response capabilities of enterprises.

[0076] (12) Improved data value and utilization rate: The present invention fully taps the value of the enterprise’s existing data by integrating multi-source heterogeneous data, thereby improving data utilization rate. At the same time, data integration can also provide strong support for the enterprise’s digital transformation and promote a data-driven decision-making culture. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 A system block diagram of a multidimensional analysis system for BOM data based on a large language model provided by an embodiment of the present invention;

[0078] Figure 2 A structural block diagram of a dynamic context enhancement prompt module provided by an embodiment of the present invention;

[0079] Figure 3 A flowchart of multi-dimensional analysis and processing provided by an embodiment of the present invention;

[0080] Figure 4 A structural block diagram of a multimodal result interpretation and visualization unit provided in an embodiment of the present invention;

[0081] Figure 5 A structural block diagram of a multi-source heterogeneous data fusion module provided in an embodiment of the present invention;

[0082] Figure 6 A flowchart of the multi-source heterogeneous data fusion module provided in an embodiment of the present invention;

[0083] Figure 7 A structural block diagram of an adaptive domain knowledge graph module provided in an embodiment of the present invention;

[0084] Figure 8 A flow chart for feedback update of the BOM domain knowledge graph provided by an embodiment of the present invention;

[0085] Fig. 9 A flow chart of knowledge application and reasoning provided for an embodiment of the present invention;

[0086] Fig.10 A flowchart of adjusting the weight of each node in the updated BOM domain knowledge graph provided by an embodiment of the present invention;

[0087] Fig.11 A flowchart for updating the BOM domain knowledge graph based on large language model interaction provided by an embodiment of the present invention;

[0088] Fig.12A flowchart of a multidimensional analysis system for BOM data based on a large language model provided by an embodiment of the present invention;

[0089] Fig.13 A flowchart of the steps of a multidimensional analysis method of BOM data based on a large language model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0090] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0091] It should be understood that although the terms first, second, etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another unit. For example, a first unit can be referred to as a second unit, and similarly, a second unit can be referred to as a first unit without departing from the scope of the exemplary embodiments of the present invention.

[0092] It should be understood that the term "and / or" that may appear in this article is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, B exists alone, and A and B exist at the same time. The term " / and" that may appear in this article describes another type of association object relationship, indicating that two relationships may exist. For example, A / and B can represent two situations: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this article generally indicates that the previous and next associated objects are in an "or" relationship.

[0093] Example:

[0094] See also Figure 1 As shown, the multidimensional analysis system of BOM data based on the large language model provided in this embodiment is based on intelligent manufacturing and artificial intelligence technology, and at the same time deeply integrates advanced technologies in multiple related fields such as data science, knowledge engineering, visualization and decision science, to achieve efficient and accurate BOM analysis and decision support, thereby significantly improving the accuracy, efficiency and interpretability of BOM data analysis, and therefore, can provide a powerful decision support tool for manufacturing, supply chain management and other fields.

[0095] In specific applications, the aforementioned multidimensional analysis system may include, but is not limited to: a multi-source heterogeneous data fusion module, an adaptive domain knowledge graph module, a dynamic context enhancement prompt module, a large language model module, and a multidimensional analysis module; wherein the multi-source heterogeneous data fusion module is mainly used for preprocessing of input data, that is, uniformly processing structured BOM data and unstructured text descriptions, realizing data standardization and unified representation, thereby providing a consistent data basis for subsequent analysis; the adaptive domain knowledge graph module provides a BOM domain knowledge graph, which contains information such as industry standards and common component relationships. Therefore, based on the knowledge graph, the expansion of BOM data can be achieved, and at the same time, it is closely related to the upper and lower levels. The text enhancement prompt modules are configured with interactive interfaces, which can dynamically adjust the prompt content according to different types of BOM data and analysis requirements based on the knowledge graph; the large language model module uses the powerful reasoning ability of the large language model to generate the corresponding answer text for the prompt content; finally, the multidimensional analysis module provides analysis and evaluation functions covering cost, performance, environmental impact, risk, etc., and also has multimodal result interpretation and visualization functions, which can convert the evaluation results into clear and structured analysis reports, and form visualization templates (such as interactive charts, flow charts and dashboards, etc.), so as to provide users with accurate and concise analysis results to assist users in making accurate decisions.

[0096] Optionally, the following provides the working process of each of the aforementioned modules:

[0097] In this embodiment, the multi-source heterogeneous data fusion module is used to obtain multiple BOM data of the target product, and perform data integration processing on each BOM data to obtain multiple processed BOM data with the same format; in specific applications, the multi-source heterogeneous data fusion module is connected to the data input layer, and the multi-protocol data interface in the data input layer is used to receive data from different sources, wherein the aforementioned data input layer is compatible with multiple data formats, such as CSV, JSON, XML, etc., and the multiple BOM data may include but are not limited to the target product's BOM structured data, product unstructured data, industry standard data, and historical evaluation data. And so on. Of course, the specific input data type can be set according to actual use and is not limited to the above examples. At the same time, when various BOM data of the target product are input into the aforementioned multi-source heterogeneous data fusion module, the module performs data integration processing on the inputted various BOM data (which may include but is not limited to steps such as data cleaning, standardization, entity recognition and format conversion. Of course, the aforementioned processing process is described in detail below). In this way, the integration of various BOM data can be realized based on the multi-source heterogeneous data fusion module, thereby providing a consistent data basis for subsequent analysis to avoid the data island problem existing in traditional technology.

[0098] After completing the data integration of the input multiple BOM data, the adaptive domain knowledge graph module can be used to perform knowledge matching, so as to use the adaptive domain knowledge graph module to perform data expansion; that is, the adaptive domain knowledge graph module stores the BOM domain knowledge graph, wherein the adaptive domain knowledge graph module is used to receive the processed BOM data transmitted by the multi-source heterogeneous data fusion module, and based on the BOM domain knowledge graph, the processed BOM data is matched to obtain data matching results; in this embodiment, the BOM domain knowledge graph is used to store and update all data corresponding to the target product, such as industry standards, common component relationships and other information; in this way, the knowledge retrieval of the input BOM data can be performed based on the BOM domain knowledge graph, so as to retrieve the product data that has an associated relationship with the input BOM data from the aforementioned BOM domain knowledge graph, thereby expanding its data range; further, for example, a graph traversal algorithm can be used, combined with semantic similarity matching, to query the knowledge graph, so as to obtain the corresponding data matching results.

[0099] Then, the adaptive domain knowledge graph module can input the data matching results into the dynamic context enhancement prompt module so as to generate the input content of the BOM large language model based on the module, that is: the dynamic context enhancement prompt module is used to generate a multidimensional analysis prompt question and answer text based on the data matching results, and transmit the multidimensional analysis prompt question and answer text to the large language model module; in specific applications, the dynamic context enhancement prompt module mainly integrates context analysis and uses the prompt generation strategy to dynamically generate input prompts for the BOM large language model (i.e., LLM model) so as to provide accurate input prompt content for the subsequent BOM large language model.

[0100] After generating the input prompt of the LLM model, the input prompt (i.e. the aforementioned multidimensional analysis prompt question and answer text) can be input into the large language model module so as to perform data reasoning based on the BOM large language model later; optionally, the large language model module stores a BOM large language model (pre-set), wherein the large language model module is used to input the multidimensional analysis prompt question and answer text into the BOM large language model to obtain the question and answer results corresponding to the multidimensional analysis prompt question and answer text, and transmit the question and answer results to the multidimensional analysis module; in addition, in this embodiment, the LLM output result can also be evaluated, and the LLM model can be optimized according to the evaluation result.

[0101] Finally, the multidimensional analysis module can perform multidimensional analysis based on the question and answer results, and fuse the multidimensional analysis results. Finally, the fused analysis results are converted into an intuitive and easy-to-understand display form; that is, the multidimensional analysis module is used to perform multidimensional analysis and processing on the question and answer results to obtain multidimensional analysis results, and to fuse the multidimensional analysis results to obtain multimodal analysis results of the target product, and visualize the multimodal analysis results; in this embodiment, the multidimensional analysis processing is taken as an example, including the analysis and processing of the cost dimension, the analysis and processing of the performance dimension, the analysis and processing of the safety dimension, the analysis and processing of the supply chain dimension, and the analysis and processing of the environmental dimension of the target product; of course, it can be any combination of any two or more of the aforementioned multiple dimensions; and the analysis dimensions can be specifically set according to use, and are not limited to the aforementioned examples.

[0102] Based on the foregoing explanation, the multidimensional analysis system provided in this embodiment realizes in-depth BOM multidimensional impact assessment and analysis, enhances the adaptability and scalability of the system, and the application of advanced AI technologies such as deep learning and natural language processing greatly improves the intelligence of the system.

[0103] Optionally, the following provides the detailed structure and working process of each of the aforementioned modules.

[0104] First, one specific structure of the aforementioned multi-source heterogeneous data fusion module is disclosed:

[0105] In this embodiment, see Figure 5 As shown, the example multi-source heterogeneous data fusion module may include but is not limited to: a data cleaning and standardization unit, an entity recognition and relationship extraction unit, a semantic integration unit, a data integration unit, and a data quality assessment unit; wherein, see Figure 6 As shown, the specific working process of each of the above functional units is as follows:

[0106] In specific applications, the data cleaning and standardization unit is used to perform data cleaning and data standardization on each BOM data of the target product in turn, obtain multiple standardized BOM data, and send the multiple standardized BOM data to the entity recognition and relationship extraction unit; among them, for example, data cleaning can include but is not limited to removing noise data, processing missing values ​​and outliers, etc. (such as using the Z-score method to process outliers); and data standardization processing is to unify the format of data from different sources and unify the measurement units, etc. (such as converting all weight units into kilograms, or standardizing supplier names and component codes); currently, the aforementioned data cleaning and data standardization processing are commonly used methods for data preprocessing, and their principles will not be repeated.

[0107] After completing the data cleaning and standardization processing, we can enter the data analysis and integration stage, that is: the entity recognition and relationship extraction unit is used to perform entity recognition and relationship extraction processing on each standardized BOM data to obtain the entity relationship of each standardized BOM data; in this embodiment, entity recognition is to extract key entities from the data source, and relationship extraction is to analyze and extract various relationships between entities. Of course, NLP technology can be used for named entity recognition and relationship extraction, which is a commonly used technology in natural language processing, and its principle will not be repeated.

[0108] After completing entity recognition and relationship extraction, semantic integration can be performed, that is, the semantic integration unit is used to perform semantic integration processing on each standardized BOM data based on the entity relationship of each standardized BOM data, and obtain standardized BOM data after multiple semantic integration; in specific applications, the semantic integration unit is mainly to construct a unified semantic representation, so as to solve the semantic conflicts of information from different sources, that is, its main goal is to unify the meaning and relationship of concepts in different data sources and thus ensure the consistency and unambiguity of data at the semantic level; of course, it is also a commonly used technology in natural language processing, and its specific process is not described here one by one; in this way, after completing semantic integration, the semantically integrated data can be uniformly represented, and the process is shown below.

[0109] That is, the data integration unit is used to perform format conversion processing on standardized BOM data after multiple semantic integration, so as to obtain multiple processed initial BOM data with the same format after the format conversion processing; in this way, based on the above operations, the consistency and comparability of the data can be ensured.

[0110] At the same time, in this embodiment, in order to ensure the quality of data processing, a quality assessment process is also provided, namely: a data quality assessment unit is used to perform quality assessment on multiple processed initial BOM data, and after the quality assessment is passed, the multiple processed initial BOM data are used as the processed BOM data of the target product; in specific applications, a multi-dimensional quality index can be established for comprehensive assessment, so as to evaluate the integrity, consistency and accuracy of the aforementioned processed initial BOM data (that is, whether the data is complete, whether the data units are consistent, whether the data is correct, etc.); wherein, this embodiment can establish a preset quality assessment standard (such as when the data is complete, the data units are consistent and the data is completely correct, the quality assessment is passed); therefore, the quality assessment can be performed according to the quality assessment standard, so that the data is output after the quality assessment is passed; similarly, when the quality assessment fails, manual intervention is performed, that is, it is sent to the expert terminal, so that the data is adjusted with the help of experts.

[0111] Based on the above operations, the integrity, consistency and accuracy of the various BOM data input during fusion can be guaranteed.

[0112] Furthermore, in this embodiment, a visualization and interactive exploration process is also provided, that is, a visualization tool is provided to present the aforementioned processed BOM data, thereby supporting users to perform interactive exploration and in-depth analysis with the help of the visualization tool.

[0113] Furthermore, the aforementioned multi-source heterogeneous data fusion module is also provided with a continuous update and maintenance mechanism and a data security and privacy protection mechanism; see Figure 6 As shown, the multi-source heterogeneous data fusion module is provided with an enhanced update mechanism, that is, it supports real-time or quasi-real-time data update, and only processes newly added or changed data, thereby triggering the necessary data processing flow (and timely re-performing the above-mentioned data cleaning, standardization, entity recognition and relationship extraction operations on the newly added or changed data). In this way, real-time processing of new and / or changed data can be achieved.

[0114] In addition, data security and privacy protection involves implementing access control and encryption measures during the data access stage; at the same time, data desensitization technology is used during the processing process to ensure that the output data complies with privacy protection regulations.

[0115] Therefore, through the above detailed explanation of the multi-source heterogeneous data fusion module, the present invention can effectively integrate BOM-related data from different sources through this comprehensive and flexible data fusion method, while ensuring the quality, security and privacy protection of the data, which provides a reliable and unified data foundation for subsequent BOM analysis and impact assessment, greatly improving the accuracy and efficiency of the analysis.

[0116] After completing the detailed description of the structure and workflow of the multi-source heterogeneous data fusion module, the following provides one specific structure and corresponding workflow of the dynamic context enhancement prompt module.

[0117] For specific applications, see Figure 2 As shown, the dynamic context enhanced prompt module described in the example may include but is not limited to: an input parser, a context analyzer, a prompt template library, a dynamic prompt generator and a prompt validator; wherein the working process of the above-mentioned parser, context analyzer, dynamic prompt generator and other corresponding units is:

[0118] An input parser is used to receive multiple processed BOM data transmitted by the multi-source heterogeneous data fusion module, and extract key information from each processed BOM data to obtain key information corresponding to each processed BOM data; in this embodiment, the input parser mainly uses natural language processing technology and a rule engine to parse the multiple processed BOM data inputted above, and at the same time, it can also receive analysis requirements transmitted by the user and parse them to obtain corresponding key information; that is, the input parser can realize the parsing of structured and unstructured input data; and after the parsing is completed, the key information obtained by the parsing can be transmitted to the context analyzer for semantic understanding, and the process is shown as follows.

[0119] In this embodiment, the context analyzer is used to use a deep learning model to perform context encoding processing on each key information to extract the contextual semantic information of each key information; optionally, for example, the aforementioned deep learning model can be but not limited to BERT (Bid irect iona l Encoder Representations from Transformers, a language representation model), that is, the aforementioned BERT model is used to capture deep semantic information, so as to understand the specific needs and background of the current task; at the same time, in this embodiment, a plurality of predefined prompt templates are stored in the prompt template library (a graph database can be used to store templates to support complex template relationships and fast retrieval), therefore, the context analyzer can also match the prompt template corresponding to the target product from the prompt template library based on each contextual semantic information, so as to realize customized information prompts.

[0120] After completing the semantic understanding of each key information and the matching of the prompt template, the dynamic prompt generator can be used to generate the prompt information, that is, the dynamic prompt generator is used to extract the historical prompt information from the historical prompt buffer (in this embodiment, the historical prompt buffer stores recent successful prompts, which can provide a reference for the dynamic prompt generator, thereby improving the generation efficiency), and extract the industry knowledge of the target product from the industry rule engine (the industry rule can be regarded as an intelligent knowledge base system, which stores all the data of the target product and the industry knowledge of the industry, and can provide the dynamic prompt generator with industry-related constraints and guidance, thereby enhancing the professionalism of the prompt); then Then, based on the contextual semantic information of each key information, the historical prompt information and the industry knowledge, the prompt generation strategy and the matched prompt template are used to generate the initial multidimensional analysis prompt question and answer text; optionally, the prompt word generation strategy may include but is not limited to: fixed prompt words, template-based prompt words, context sampling generation prompt words, N-Gram prompt words, adaptive prompt words (such as reinforcement learning, genetic algorithm), interactive prompt words (prompt word guidance, used for intent detection, etc.), transfer learning prompt words (such as pre-trained language model, multimodal prompt words, etc.) and other strategies. Of course, the aforementioned prompt generation strategy is a common technology for generating input words of a large language model, and its principle will not be repeated here.

[0121] After generating the input prompt information of the LLM model, in order to ensure the accuracy of the generated prompt information, this embodiment further provides a prompt validator, i.e., a prompt validator, which is used to verify the initial multidimensional analysis prompt question and answer text, and after the verification is passed, the initial multidimensional analysis prompt question and answer text is transmitted to the large language model module as the multidimensional analysis prompt question and answer text; optionally, for example, the prompt validator can, but is not limited to, use a rule engine and a machine learning model to detect the compliance and security of the prompt (i.e., the aforementioned initial multidimensional analysis prompt question and answer text), so that it meets system requirements and safety standards; further, the aforementioned rule engine can be generated based on pre-set prompt verification rules, and the machine learning model can check whether the aforementioned initial multidimensional analysis prompt question and answer text has grammatical errors and other problems.

[0122] Thus, based on the above explanation, the dynamic context enhanced prompt module can be used to generate input prompt information of the LLM model (i.e. the aforementioned multi-dimensional analysis prompt question and answer text), thereby providing accurate input data for subsequent model reasoning.

[0123] In a specific implementation, this embodiment is also provided with an external interface component, a performance evaluation unit and a reinforcement learning optimizer; so as to realize feedback learning of the prompt template library and the dynamic prompt generator, so that the two can be continuously optimized during use to achieve better use effect.

[0124] In this embodiment, the external interface component mainly includes a user feedback interface, an LLM interface adapter and an external LLM, wherein the LLM interface adapter is used to convert the verified prompts into an input format acceptable to the large language model (LLM); and the external LLM refers to the aforementioned large language model module, which stores the BOM large language model and is used to generate corresponding answer results based on the input prompt information (i.e., the multidimensional analysis prompt question and answer text) output by the LLM interface adapter.

[0125] At the same time, the user feedback interface, combined with the performance evaluation unit and the prompt template library, forms a user feedback loop to update the prompt template library based on user feedback information; specifically, the update process is as follows:

[0126] See also Figure 2 As shown, the user feedback interface is used to obtain the user's first feedback information on the multidimensional analysis prompt question and answer text and the second feedback information on the question and answer result corresponding to the multidimensional analysis prompt question and answer text in response to human-computer interaction, and send the first feedback information and the second feedback information to the performance evaluation unit; then, the performance evaluation unit is used to receive the first feedback information and the second feedback information sent by the user feedback interface, and obtain prompt template evaluation information based on the first feedback information and the second feedback information, so as to update each prompt template in the prompt template library according to the prompt template evaluation information.

[0127] Specifically, the path for the feedback loop is: user feedback interface → performance evaluation unit → prompt template library; that is, the user provides an evaluation of the system output (including an evaluation of the multidimensional analysis prompt question and answer text and an evaluation of the LLM model output) through the user feedback interface; then, this feedback evaluation is sent to the performance evaluation unit, and finally, the performance evaluation unit updates the prompt template library based on the prompt template evaluation information to improve the quality of the prompt template.

[0128] Furthermore, a feedback loop update of the dynamic prompt generator is also provided, and the update process is as follows:

[0129] The performance evaluation unit is used to receive the question and answer results corresponding to the multi-dimensional analysis prompt question and answer text sent by the large language model module (that is, the output generated by the external LLM is sent to the performance evaluation unit), and the performance evaluation unit performs performance evaluation on the dynamic prompt generator based on the question and answer results to obtain an evaluation result; then, the performance evaluation unit is also used to send the evaluation result to the reinforcement learning optimizer, so that the reinforcement learning optimizer optimizes the prompt generation strategy of the dynamic prompt generator based on the evaluation result and uses the reinforcement learning strategy to complete the feedback learning of the dynamic prompt generator after optimization; in this way, the path of the feedback loop is: external LLM→performance evaluation unit→reinforcement learning optimizer→dynamic prompt generator; based on this, the quality of the dynamic prompt generator output is evaluated through the output of the LLM, and the prompt generation strategy is optimized according to the quality evaluation, so as to improve the quality of the subsequently generated input prompts.

[0130] In addition, in this embodiment, the dynamic context enhancement prompt module may also include: an error handling unit, wherein the error handling unit is connected to the dynamic prompt generator and the prompt verifier, and through two-way interaction with the two, it can detect and handle errors that occur in the dynamic prompt generator and the prompt verifier, and provide a recovery strategy (such as restarting the process) when an error occurs; in this way, the purpose of system maintenance can be achieved, thereby improving the robustness of the system.

[0131] Therefore, through the above explanation, through the collaborative work of these components and the multi-layer feedback mechanism, the dynamic context-enhanced prompt module can generate highly customized, context-related prompts, which can significantly improve the performance of LLM in BOM data analysis and multi-dimensional impact assessment tasks. The module's adaptability, continuous optimization capability and error handling mechanism ensure that the system can cope with different industries and different types of BOM analysis needs, and continuously improve its performance and reliability with use.

[0132] Next, this embodiment provides a specific structure and workflow of a multi-dimensional analysis module.

[0133] In the specific implementation, see Figure 3 As shown, the multidimensional analysis described in the example may include, but is not limited to, modules including: a multidimensional impact analysis unit and a multimodal result interpretation and visualization unit; wherein the workflow of the aforementioned functional units is:

[0134] The multidimensional impact analysis unit is used to obtain evaluation parameters. In the present embodiment, the evaluation parameters include analysis dimensions and priorities of analysis dimensions, etc., and the analysis dimensions may include but are not limited to the cost, performance, safety, environmental impact, supply chain and other dimensions of the target product. Of course, the specific analysis dimensions may be set according to actual use and are not limited to the aforementioned examples.

[0135] After obtaining the evaluation parameters, the BOM analysis of the target product can be performed, that is, the multidimensional impact analysis unit is used to match the analysis model corresponding to each analysis dimension from the analysis model library according to the evaluation parameters, and to receive the question and answer results transmitted by the large language model module, and input the question and answer results into each matched analysis model to obtain the multidimensional analysis result of the target product; in this embodiment, the multidimensional evaluation is a composite state, which may include but is not limited to the following dimensions of evaluation in parallel: environmental impact analysis (the evaluation indicators are energy consumption, material recyclable bins, carbon footprint, etc., and the evaluation method is the life cycle assessment (LCA) model), cost analysis (the evaluation indicators are material cost, production cost, etc.), Cost, operating cost, and the evaluation method is activity-based costing (ABC) method), performance evaluation (evaluation indicators are product performance parameters, reliability, service life, etc., and the evaluation method is to use multi-objective optimization algorithm and finite element analysis (FEA) tool), supply chain impact assessment (evaluation indicators are supplier changes, delivery time, inventory level, etc., and the evaluation method is to use supply chain simulation model and risk assessment matrix), compliance impact assessment (evaluation indicators are compliance with industry standards, degree of satisfaction of regulatory requirements, etc., and the evaluation method is automatic inspection and compliance scoring by rule engine), etc.; of course, the evaluation of each of the above dimensions is set with a corresponding analysis model library, and is stored in the analysis model library, and can be matched according to the dimension when used.

[0136] After completing the multi-dimensional analysis of the BOM of the target product, a comprehensive evaluation can be performed, that is, the multi-dimensional impact analysis unit is also used to integrate the multi-dimensional analysis results to obtain the comprehensive analysis results of the target product, and send the comprehensive analysis results to the multi-modal result interpretation and visualization unit; in specific applications, the multi-criteria decision analysis (MCDA) method can be applied to integrate the multi-dimensional evaluation results to obtain the comprehensive analysis results; finally, it can be transmitted to the multi-modal interpretation and visualization unit so as to obtain the multi-modal analysis results based on this.

[0137] That is: the multimodal result interpretation and visualization unit is used to generate visualization data, summary text and result interpretation information corresponding to the comprehensive analysis result according to the comprehensive analysis result, and perform multimodal fusion processing on the visualization data, summary text and result interpretation information, so as to obtain the multimodal analysis result of the target product after the multimodal fusion processing; in this embodiment, the multimodal result interpretation and visualization unit converts the comprehensive analysis result into an intuitive and easy-to-understand form, which integrates a variety of visualization technologies, charts, network diagrams, heat maps, etc., and has functions such as summary generation and result interpretation, and can also realize interactive data exploration functions; in this way, the final analysis results can be provided to users clearly and concisely, making it easier for users to view.

[0138] Furthermore, the following provides a specific structure of a multimodal result interpretation and visualization unit:

[0139] See also Figure 4 As shown, the example multimodal result interpretation and visualization unit may include, but is not limited to: an output parsing unit, a key information extraction unit, a data classification and structuring unit, a visualization data generation unit, a summary text generation unit, an AI interpretation unit, and a multimodal display integration unit; wherein the working process of each of the aforementioned units is:

[0140] In this embodiment, the working process of the multimodal result interpretation and visualization unit can be divided into a data processing stage, a visualization generation stage, a result integration and display stage, and a user interaction and optimization stage; among them, for the data processing stage, it is: an output parsing unit, which is used to parse the comprehensive analysis results, obtain the parsing information corresponding to the comprehensive analysis results, and transmit the parsing information to the key information extraction unit; in specific applications, for example, it can be but not limited to using natural language processing (NLP) technology to identify key sentences and paragraphs in the comprehensive analysis results, and then applying named entity recognition (NER) technology to extract important entities and attributes in the key sentences and paragraphs.

[0141] Next, the key information extraction unit is used to perform key information extraction processing on the parsed information to obtain multiple key information corresponding to the parsed information, and transmit each key information corresponding to the parsed information to the data classification and structuring unit; wherein the key information extraction unit extracts core data points and information from the parsed text, that is, uses a keyword extraction algorithm to identify keywords in the aforementioned analysis, and applies text summarization technology to refine core information, and finally, keywords and refined text summaries can be used to form multiple key information.

[0142] Finally, the data classification and structuring unit is used to classify the key information corresponding to the analyzed information to obtain multiple classified information, and to structure each classified information to obtain multiple structured classified information; in this embodiment, the data classification and structuring unit is mainly used to classify the extracted information and convert it into a structured format, where, for example, a multi-dimensional classification system can be designed, including impact type, time scale, etc., and then combined with a machine learning algorithm (such as a decision tree) to automatically classify the data, and finally, the unstructured text can be converted into structured data (i.e., Json or CSV).

[0143] Through the above operations, data processing can be completed, and then the visualization generation stage can be entered. The process is as follows:

[0144] A visualization data generation unit is used to generate visualization data of each structured classification information according to each structured classification information, and transmit the visualization data of each structured classification information to the multimodal display integration unit, wherein the visualization data of any structured classification information includes static charts, interactive charts and / or network relationship maps; in this embodiment, examples can be but are not limited to matching corresponding visualization methods according to data characteristics and analysis objectives of each structured classification information, such as using a decision tree algorithm to recommend visualization types according to data types and analysis requirements.

[0145] Optionally, static charts may include but are not limited to bar charts, line charts, pie charts, etc., and data visualization libraries (such as Matplotlib or Plotly) may be used to generate high-quality charts, and automatic layout and style optimization may be implemented at the same time to ensure that the charts are beautiful; interactive charts are dynamic visualization charts that allow users to interact and explore, and they may utilize Web technologies (such as D3.js or ECharts) to allow users to perform interactive operations such as zooming and filtering; finally, network relationship graphs are used to visualize relationship networks and dependency structures, and they may use force-directed graph algorithms to display network structures, and implement node clustering and edge binding techniques to optimize the visualization of large-scale networks.

[0146] At the same time, this embodiment also has the functions of intelligent summary generation and result AI interpretation, so that it can be combined with the aforementioned visualization data to form multi-modal visualization fusion data; that is, the summary text generation unit (i.e. Figure 4 The intelligent summary generation in the AI ​​explanation unit is used to extract and process the text summaries of each structured classification information using a natural language processing algorithm, obtain the summary text corresponding to each structured classification information, and transmit each summary text to the multimodal display integration unit; and the AI ​​interpretation unit (i.e. Figure 4 The explanatory AI technology in the invention is used to perform feature importance analysis on each structured classification information to obtain the feature importance analysis result of each structured classification information, and transmit the feature importance analysis result of each structured classification information as the result explanatory information of each structured classification information to the multimodal display integration unit.

[0147] In this embodiment, the summary text generation unit mainly uses extractive and generative text summarization techniques and utilizes the natural language generation (NLG) model to create a fluent summary text; while the AI ​​interpretation unit uses a visual decision tree to display the reasoning process, that is, to implement feature importance analysis and explain the influence of different factors.

[0148] After completing the aforementioned visualization data generation, intelligent summary generation and result AI interpretation, the result integration and display stage can be entered, that is: the multimodal display integration unit is used to perform multimodal fusion processing on the visualization data, summary text and result interpretation information corresponding to each structured classification information, so as to obtain the multimodal analysis result of the target product after the multimodal fusion processing; in this embodiment, the aforementioned summary text, visualization data, and result interpretation information are integrated into a unified display interface to obtain the multimodal analysis result; further, in this embodiment, the multimodal display integration unit is designed with a modular display template, which supports the flexible combination of different types of content and has a responsive function, which can ensure good display effects on different devices.

[0149] Furthermore, the example multimodal result interpretation and visualization unit also includes an interactive dashboard, that is, the multimodal analysis results generated above are displayed on the interactive dashboard; wherein, the interactive dashboard provides a unified user interface, integrates all visualization and analysis results, allows users to freely combine and layout different visualization components, and supports linkage interaction between components and multi-dimensional data exploration.

[0150] After completing the result integration and display, you can enter the user interaction and optimization stage, see Figure 4 As shown, the multimodal result interpretation and visualization unit may also include: a user interaction layer, which is used to process user input and interaction requests (by setting up event monitoring and response mechanisms), and then collect user feedback (i.e., collect and analyze user feedback on visualization and interpretation). Finally, dynamic adjustment and optimization can be performed based on the aforementioned information, that is, visualization and interpretation can be optimized based on user feedback and usage data. In this embodiment, for example, user preferences and behavior patterns can be analyzed based on, but not limited to, machine learning algorithms, so as to dynamically adjust visualization recommendations and display strategies, so as to continuously optimize data processing and visualization generation processes.

[0151] Of course, the above process is cyclical and iterative, in which user feedback and dynamic adjustments will continuously affect the data processing and visualization generation process, forming a closed loop of continuous optimization; the entire process aims to transform complex BOM analysis and impact assessment results into an intuitive, easy-to-understand and highly interactive form, thereby enhancing users' understanding and insight into the data.

[0152] In the specific implementation, in the multi-dimensional analysis and processing process, this embodiment also provides an interactive multi-dimensional impact assessment process, that is, when there is insufficient information, an interactive dialogue method is used to interact with the user to obtain the missing information, so as to dynamically adjust the assessment process. The specific process is as follows:

[0153] See also Figure 3As shown, if there is a problem of missing information in the multi-dimensional analysis process, the multi-dimensional impact analysis unit is used to identify the missing information, generate a missing information query question based on the missing information, and visualize the missing information query question; then, the multi-dimensional impact analysis unit is used to obtain the user's feedback input information for the missing information query question in response to human-computer interaction, and parse the feedback input information to obtain an input parsing result; then, the multi-dimensional impact analysis unit is used to adjust the model parameters of the analysis model corresponding to each analysis dimension based on the input parsing result to obtain an adjusted analysis model; finally, the multi-dimensional impact analysis unit is also used to re-input the input parsing result and the question and answer result into each adjusted analysis model to obtain a multi-dimensional analysis result of the target product.

[0154] That is to say, when key information is found to be missing during the multi-dimensional analysis and processing process or the initialization process, the information gap can be identified, and targeted queries can be generated, and then the user interaction state can be carried out to collect the necessary information; and user interaction is a sub-state machine, which mainly includes generating questions (that is, generating clear and specific questions based on the identified information needs), waiting for responses (presenting questions to users and waiting for user input), parsing responses (processing information provided by users) and information updating process (integrating new information into the evaluation process); in this way, through multiple rounds of questions and answers, it can be ensured that comprehensive and accurate information is collected; finally, after the new information is integrated into the evaluation process, dynamic adjustment can be triggered, that is, updating the evaluation parameters (that is, adjusting the parameters of the evaluation model according to the new information) and re-planning the evaluation path (that is, re-planning the subsequent evaluation steps according to the new parameters and information) so as to return to the multi-dimensional analysis and processing process.

[0155] Of course, in this embodiment, the multidimensional impact analysis unit is also used to generate decision recommendations for the target product based on the comprehensive analysis results, and to visualize the decision recommendations; wherein the visualized decision support report includes sensitivity analysis and scenario simulation results, thereby helping users make more accurate decisions.

[0156] In addition, in practice, this multidimensional impact analysis unit also supports continuous learning, see Figure 3 As shown, based on the comprehensive analysis results and user feedback information, each analysis model in the analysis model library is updated to complete the continuous reinforcement learning of each analysis model during use; in layman's terms, the evaluation model is continuously optimized based on the evaluation results and user feedback, and the knowledge base is recorded to record the evaluation experience and best practices for future reference.

[0157] Based on the above explanation, the entire multidimensional analysis process is as follows: (1) evaluation initialization (including initialization of evaluation parameters); (2) multidimensional evaluation; (3) insufficient information processing; (4) user interaction; (5) dynamic adjustment; (6) comprehensive analysis; (7) result generation (including subsequent visual data generation, intelligent summary and AI interpretation results, etc.); (8) continuous learning. This process is cyclical and iterative, in which: steps (2) to (5) may be repeated multiple times until sufficient information is collected and the evaluation of all dimensions is completed; and insufficient information processing and user interaction (steps (3) and (4)) may be triggered at any stage of evaluation initialization or multidimensional evaluation; similarly, dynamic adjustment (step (5)) may occur at any time during the multidimensional evaluation; finally, continuous learning runs through the entire process, constantly learning and optimizing from the evaluation results and user interactions.

[0158] Therefore, through the above detailed explanation of the multidimensional analysis module, the present invention can comprehensively and accurately evaluate the various impacts of BOM changes through this interactive, multidimensional impact assessment process; at the same time, its dynamic adjustment capability enables it to adapt to new situations that arise during the assessment process, and the continuous learning mechanism ensures that the accuracy and efficiency of the assessment will continue to improve over time; in addition, the design of user interaction enables the system to actively obtain necessary input when information is insufficient, thereby providing more accurate assessment results; based on this, this provides enterprises with a powerful and flexible decision support tool, which helps to make more informed decisions in complex manufacturing and supply chain environments.

[0159] Finally, this embodiment also designs a graph update mechanism for the BOM domain knowledge graph of the adaptive domain knowledge graph module; it realizes dynamic updating, intelligent reasoning and continuous optimization of knowledge.

[0160] See also Figure 7 , Figure 8 to Figure 11 As shown, the example adaptive domain knowledge graph module may include but is not limited to: a knowledge graph update unit and a feedback learning unit, wherein the workflow of each of the aforementioned units is:

[0161] In this embodiment, the knowledge graph update unit is used to obtain knowledge graph update data and perform data preprocessing on the knowledge graph update data to obtain named entities corresponding to the knowledge graph update data after data preprocessing; wherein, in specific applications, the knowledge graph update data can be new information from various sources, including but not limited to industry standard update data, new product specification data, market change data, etc., and data preprocessing can include but is not limited to data source integration, ontology model matching, and entity recognition and relationship extraction.

[0162] See also Figure 8As shown, data source integration is to standardize the knowledge graph update data, that is, to collect and standardize data from different channels, among which, examples can include but are not limited to the use of ETL algorithm (a data processing technology that involves the process of extracting data from the data source, converting data and loading it into the target system) to integrate the data source, so as to ensure data quality; at the same time, ontology model matching is to map the standardized knowledge graph update data to the BOM domain knowledge graph; in this embodiment, examples can include but are not limited to the use of semantic matching algorithms to achieve the mapping of the aforementioned data and the knowledge graph; finally, entity recognition and relationship extraction are to identify key entities and the relationship between entities from the aforementioned mapped data, so as to carry out knowledge integration and verification based on this later; optionally, examples can include but are not limited to the use of named entity recognition (NER) and relationship extraction models to complete the aforementioned operations.

[0163] After completing the above operations, you can enter the knowledge integration and verification processing stage, see Figure 7 As shown, the knowledge graph updating unit is also used to update the BOM domain knowledge graph according to the named entities corresponding to the knowledge graph update data, obtain an updated initial BOM domain knowledge graph, and perform knowledge integration and verification on the updated initial BOM domain knowledge graph, so as to accept the update of the BOM domain knowledge graph after the knowledge integration and verification, and obtain an updated BOM domain knowledge graph, so as to use the updated BOM domain knowledge graph for data matching.

[0164] In this embodiment, updating the BOM domain knowledge graph is an updating of the graph structure, that is, integrating the newly identified entities and relationships into the existing BOM domain knowledge graph (for example, graph database operations can be used to efficiently insert, update or delete nodes and edges), and the knowledge integration and verification processing mainly includes: consistency checking, knowledge verification, conflict resolution, expert review and automatic merging.

[0165] Optionally, the consistency check is to check whether the newly added knowledge in the updated initial BOM domain knowledge graph conflicts with the existing knowledge. Logical rules and constraint checks can be used to identify potential inconsistencies. At the same time, knowledge verification is to verify the accuracy and reliability of the newly added knowledge in the updated initial BOM domain knowledge graph. It can combine statistical analysis and preset rules to evaluate the credibility of the new knowledge, so that when the credibility assessment result is greater than a threshold, the knowledge verification is passed. After completing the knowledge verification, conflict resolution can be performed, that is, a confidence-based conflict resolution mechanism (human judgment can be introduced when necessary) is used to resolve potential conflicts between new and old knowledge. Finally, expert review and automatic merging can be performed, that is, based on the verification results, it is decided whether to intervene manually. Among them, updates with high risks or high uncertainties are output to the review terminal for review by the reviewers at the review terminal, and others are automatically merged (that is, the update of the graph is accepted to obtain an updated BOM domain knowledge graph).

[0166] In this embodiment, the merged knowledge graph is a richer, newer, and more complex knowledge network; it not only contains the latest information, but also maintains the intrinsic connections between knowledge, and can provide a more comprehensive and accurate knowledge basis for subsequent queries, analysis, and decision-making. This dynamic update feature enables the knowledge graph to continuously adapt to and reflect the latest status of domain knowledge, and can provide continuous optimization support for BOM data analysis and multi-dimensional impact assessment.

[0167] In this way, based on the above operations, the update of the BOM domain knowledge graph can be completed. Then, when the updated BOM domain knowledge graph is used, feedback learning can be performed, that is, entering the feedback learning update stage.

[0168] In this embodiment, the feedback learning of the updated BOM domain knowledge graph is mainly divided into three aspects: knowledge application and reasoning stage, LLM interaction stage and continuous optimization stage. The three are carried out in parallel. The process is as follows:

[0169] A feedback learning unit is used to perform knowledge application and reasoning processing on the updated BOM domain knowledge graph to obtain knowledge reasoning results; to obtain third feedback information of the updated BOM domain knowledge graph in the multidimensional analysis process, and based on the third feedback information, evaluate the importance of each node in the updated BOM domain knowledge graph, so as to adjust the weight of each node in the updated BOM domain knowledge graph according to the importance evaluation result to obtain weight adjustment information; and is also used to obtain the question and answer results output by the large language model module, and input the knowledge reasoning results, weight adjustment results and / or question and answer results into the knowledge graph updating unit, so as to optimize the updated BOM domain knowledge graph through the knowledge graph updating unit, so as to complete the feedback learning of the updated BOM domain knowledge graph after optimization.

[0170] Optional, see Fig. 9 As shown in the figure, knowledge application and reasoning processing mainly includes the following steps: graph embedding update, knowledge reasoning, complex query processing, uncertainty reasoning and new knowledge generation; among them, graph embedding update is to update the vector representation of the BOM domain knowledge graph after the above update (graph embedding algorithms (such as TransE, RotatE, etc.) can be used to generate low-dimensional representations of entities and relationships); knowledge reasoning is to perform complex reasoning based on the updated knowledge graph (its implementation method is: combining symbolic reasoning and statistical reasoning, and supporting multi-hop reasoning and rule reasoning); then, complex query processing can be performed, that is, using efficient graph traversal algorithms and query optimization techniques to process complex queries of multiple designed entities and relationships; then, after completing the complex query, uncertain reasoning can be performed, that is, using probabilistic graph models and fuzzy logic to realize reasoning processing under uncertain conditions; specifically, it is to add a connection from uncertainty reasoning to graph structure update, create a path connected to the knowledge injection LLM, add an arrow pointing to visualization and interactive exploration, establish a connection with feedback learning, and so on.

[0171] In addition, new knowledge generation uses techniques such as inductive logic programming (ILP) to derive new rules and relationships from existing knowledge to obtain new implicit knowledge. This new implicit knowledge can then be used to update the knowledge graph status so that the knowledge graph can be updated again later.

[0172] After completing the description of knowledge application and reasoning processing, this embodiment describes the detailed process of the LLM interaction stage:

[0173] See also Fig.11As shown, the processing process of the example LLM interaction stage may include but is not limited to: LLM interaction interface, knowledge injection into LLM, LLM output alignment, and knowledge supplement and correction; in this embodiment, the LLM interaction interface is used to realize the two-way communication between the aforementioned updated BOM domain knowledge graph and LLM, which provides a standardized data interaction format to ensure efficient knowledge transmission; at the same time, the knowledge injection into LLM provides relevant graph knowledge to the LLM model to enhance its understanding and generation capabilities (i.e., designing a context injection strategy to convert graph knowledge into prompts or contexts that LLM can understand); and the LLM output alignment uses a semantic alignment algorithm to adjust the output of LLM to conform to the structure and semantics of the knowledge graph, thereby ensuring that the output of the LL model is consistent with the knowledge graph; finally, knowledge supplement and correction uses the LLM model output (i.e., the aforementioned question and answer results) to expand and correct the knowledge graph.

[0174] Similarly, see Fig.10 As shown in the figure, the continuous optimization process can be: feedback learning, user feedback and importance assessment, knowledge forgetting mechanism, retrieval strategy optimization and adaptive learning; among them, feedback learning is based on the feedback collection mechanism set up, and uses machine learning algorithms to analyze feedback to guide graph optimization, that is, to obtain the third feedback information of the updated BOM domain knowledge graph in the multidimensional analysis process, and based on this, to carry out continuous optimization; specifically, according to the aforementioned third feedback information, and using data mining technology to analyze user behavior, to evaluate the importance of knowledge; then, based on the frequency of use, timeliness and user feedback, dynamically adjust the weight of knowledge (that is, obtain the aforementioned weight adjustment information), that is, remove or reduce outdated or irrelevant knowledge; finally, the aforementioned weight adjustment information can be sent to the knowledge graph update unit to update the graph again.

[0175] At the same time, adaptive learning refers to automatically adjusting learning strategies and parameters according to system performance and environmental changes, and then using reinforcement learning algorithms to continuously adjust and optimize retrieval strategies, and input the optimized retrieval strategies into the knowledge application and reasoning process, thereby optimizing the knowledge reasoning process.

[0176] Of course, there is a key update path for the above graph update, namely:

[0177] Graph embedding update -> new knowledge generation -> new state of knowledge graph -> new data input -> graph structure update -> feedback learning -> graph embedding update; this key path describes the cyclic process of dynamic update and self-improvement of the knowledge graph, that is, graph embedding update converts the knowledge graph into a low-dimensional vector representation to capture the semantic information of entities and relationships; new knowledge generation is based on the updated graph embedding, and the system infers new implicit knowledge or relationships; the new state of the knowledge graph is to merge the newly generated knowledge to form an updated and expanded knowledge graph; new data input is the introduction of new external information or internally generated new knowledge into the system; and graph structure update is to adjust the structure of the knowledge graph accordingly according to the newly input data; then, feedback learning is carried out, that is, analyzing the performance of the updated graph and collecting feedback on system operation and user interaction; finally, returning to graph embedding update again to start a new cycle.

[0178] In addition, this embodiment also provides visualization and interactive exploration, which is used to provide an intuitive interface to support users in exploring and verifying knowledge graphs. Specifically, the visualization and interactive exploration supports browsing and querying of large-scale knowledge graphs, and can realize multi-dimensional knowledge screening and focusing functions; at the same time, a knowledge editing interface is also provided to allow users to directly contribute and correct knowledge.

[0179] In summary, the entire process is cyclical and iterative, in which new data inputs continuously trigger the entire process, allowing the knowledge graph to be continuously updated and optimized. At the same time, the LLM interaction and continuous optimization phases run through the entire process, continuously improving the performance and accuracy of the system.

[0180] Therefore, through this comprehensive and dynamic adaptive domain knowledge graph, the system can continuously accumulate, verify and optimize the expertise related to BOM analysis and impact assessment. It not only provides rich domain context for LLM and enhances the system's understanding and reasoning capabilities, but also through its adaptive learning mechanism, enables the entire system to continuously improve its performance and accuracy as it is used; based on this, this dynamic and adaptive knowledge representation method provides a strong knowledge foundation for complex BOM data analysis and decision support, enabling the system to better adapt to the specific needs of different industries and scenarios while maintaining the timeliness and accuracy of knowledge.

[0181] Based on the above description, the overall workflow of the system provided in this embodiment is provided as follows:

[0182] See also Fig.12 As shown, the overall workflow is:

[0183] (1) Data processing, including multi-source data input, data preprocessing and fusion, and data quality assessment.

[0184] (2) Knowledge management, i.e., querying and updating the knowledge graph.

[0185] (3) LLM interaction, which includes dynamic prompt generation and LLM processing, that is, using the LLM model for deep semantic understanding and complex reasoning to generate answer text corresponding to the input prompt.

[0186] (4) Analysis and decision-making, including multidimensional impact assessment, decision recommendation generation, and result interpretation and visualization.

[0187] (5) User interaction and feedback, including: user interaction, user feedback, continuous optimization, analysis model update and visualization optimization, etc.

[0188] In this way, the system has the following characteristics: (1) Adaptability: Through multiple feedback loops, the system can automatically adjust and optimize each processing link; (2) Scalability: The modular design allows easy integration of new data sources, analysis models or visualization technologies; (3) Real-time: Supports real-time data processing and analysis, and quickly responds to user needs; (4) Intelligent interaction: Actively interact with users when necessary to obtain additional information or clarify requirements; (5) Continuous learning: Continuously learn from user feedback and system operation data to improve performance and accuracy.

[0189] Therefore, through the above detailed explanation of the multidimensional analysis system, the present invention can efficiently process complex BOM data, provide in-depth multidimensional impact assessment, and present the results in an intuitive and easy-to-understand manner. The adaptive characteristics and continuous optimization mechanism of the system ensure that it can continuously learn and improve to adapt to changing business needs and user expectations; this comprehensive approach provides a powerful decision support tool for manufacturing and supply chain management, which can help enterprises make smarter and more efficient decisions in complex environments.

[0190] In a possible design, the second aspect of this embodiment provides an application example of the first aspect of the embodiment:

[0191] The following is a detailed application example of this system in the electric vehicle manufacturing industry, showing how the system can analyze and evaluate the multi-dimensional impact of powertrain BOM changes.

[0192] Application background: An electric vehicle manufacturer plans to replace key components of the battery management system (BMS) in its best-selling models to improve overall performance and safety; this change will affect multiple aspects of the power system, including cost, performance, supply chain, and environmental impact. The manufacturer needs to fully evaluate the impact of this change to make an informed decision.

[0193] System application process:

[0194] 1. Data input and preprocessing; the input data is: detailed BOM data of the current BMS (CSV format); technical specifications of the new BMS components (PDF document); relevant supplier information (JSON format); historical performance and cost data (exported from a relational database).

[0195] Preprocessing steps: a) Data cleaning: Use regular expressions to process non-standard formats; Apply the Z-score method to detect and process outliers; b) Data standardization: Unify measurement units (such as converting all weight units to kilograms); Standardize supplier names and component codes; Data fusion, fuse BOM data with supplier information; Data quality assessment: Completeness: Check the proportion of non-null values ​​in key fields; Consistency: Cross-validate overlapping information from different data sources; Timeliness: Check the last update time of the data.

[0196] 2. Knowledge graph query and update; Knowledge graph query: use graph traversal algorithm to query related knowledge; Knowledge graph update:

[0197] Update the graph based on new data.

[0198] 3. LLM interaction and analysis; Dynamic prompt generation: The system uses templates and contextual information to generate prompts: LLM processing, using batch processing mechanism to improve efficiency: Output evaluation; Use ROUGE scoring to evaluate LLM output quality.

[0199] 4. Multi-dimensional impact assessment; the system combines LLM output and professional models to conduct in-depth multi-dimensional impact assessment: a) Cost impact: direct cost changes: use the total cost of ownership (TCO) model; long-term cost savings: apply Monte Carlo simulation to predict long-term costs; pricing impact: use the price elasticity model to evaluate the potential impact on sales; b) Performance impact: range impact: use energy density and efficiency models; charging speed: apply electrochemical models to simulate charging curves; vehicle performance: use vehicle dynamics models to evaluate acceleration performance changes; c) Safety impact: extreme condition performance: use finite element analysis (FEA) to simulate performance under extreme temperatures; battery life and stability: apply cycle life prediction models; safety accident risks: use fault tree analysis (FTA) to evaluate system reliability; d) Supply chain impact: supplier reliability: apply supplier scoring models; supply chain risks: use Monte Carlo simulation to evaluate supply disruption risks; inventory costs: apply the economic order quantity (EOQ) model to optimize inventory; e) Environmental impact: carbon footprint calculation: use life cycle assessment (LCA) methods; energy efficiency: apply energy flow analysis models; recycling impact: use material flow analysis (MFA) to evaluate recyclability.

[0200] 5. Result visualization and decision support; the system generates comprehensive visual reports: a) Interactive dashboard: Use D3.js to create dynamic dashboards that allow real-time data updates and interactions; b) Cost-Benefit Analysis Chart: Apply Net Present Value (NPV) calculation to visualize the return on investment for different time frames; c) Supply Chain Network Diagram: Generate a network diagram using the force-directed graph algorithm: d) Performance Improvement Heat Map: Use a custom color scale to show performance changes: e) Decision Tree: Generate a decision tree using the CART algorithm and visualize it with D3.js.

[0201] 6. Dealing with uncertainty: The system uses the following methods to deal with insufficient information or uncertainty: a) Sensitivity analysis: Conduct sensitivity analysis on key parameters to assess their impact on the final results; b) Fuzzy logic: Use fuzzy logic for reasoning when data is uncertain; c) Bayesian update: Use Bayesian methods to update model parameters as new information is obtained

[0202] 7 System outputs and recommendations: Based on the comprehensive analysis, the system generates the following key outputs and recommendations:

[0203] (1) It is recommended to replace BMS components, which is expected to bring 17.3% performance improvement and 12.5% ​​long-term cost savings (NPV calculation based on a 5-year period).

[0204] (2) The supply chain risk score is 0.68 (out of 1), and the following mitigation strategies are recommended; sign a long-term contract with the newly selected high-performance BMS manufacturer to lock in price and supply; maintain 3 months of existing BMS component inventory as a buffer, and the optimal inventory quantity calculated according to the EOQ model is 1,245 units.

[0205] (3) The new BMS is expected to improve vehicle safety. Fault tree analysis shows that the probability of critical failures is reduced by 35%, and it is expected that warranty costs can be reduced by approximately RMB 6.2 million over 5 years.

[0206] (4) It is recommended that the replacement plan be implemented in three phases: I. Pilot program for high-end models (third quarter); II. Promotion of mid-range models (first quarter of the following year); and II. Update of the entire model line (fourth quarter of the following year).

[0207] (5) Environmental impact analysis shows that although carbon emissions will increase by 5.8% in the initial production phase, energy efficiency improvements during the use phase will offset this impact within 18 months. It is recommended that this be included in the marketing strategy.

[0208] Based on the above application examples, manufacturers can make strategic decisions based on comprehensive and in-depth data analysis, effectively balancing multiple key factors such as technological innovation, financial performance, operational efficiency and sustainable development, laying a solid foundation for the company's long-term competitiveness.

[0209] See also Fig.13 As shown, in a possible design, the third aspect of this embodiment provides an analysis method for the multidimensional analysis system of BOM data based on the large language model described in the first aspect of the embodiment, wherein the operating steps of the method may be but are not limited to the following steps S1 to S6.

[0210] S1. Acquire multiple BOM data of target products, and perform data integration processing on each BOM data to obtain multiple processed BOM data with the same format.

[0211] S2. Based on the BOM domain knowledge graph, perform data matching on each processed BOM data to obtain data matching results.

[0212] S3. Generate a multi-dimensional analysis prompt question and answer text based on the data matching results.

[0213] S4. Input the multidimensional analysis prompt question and answer text into the BOM large language model to obtain the question and answer result corresponding to the multidimensional analysis prompt question and answer text.

[0214] S5. Perform multi-dimensional analysis and processing on the question and answer results to obtain multi-dimensional analysis results.

[0215] S6. Fuse the multi-dimensional analysis results to obtain the multimodal analysis results of the target product, and visualize the multimodal analysis results, wherein the multi-dimensional analysis processing includes the analysis processing of the cost dimension, the analysis processing of the performance dimension, the analysis processing of the safety dimension, and the analysis processing of the supply chain dimension of the target product.

[0216] The working process, working details and technical effects of the method provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0217] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multidimensional analysis system for BOM data based on a large language model, characterized in that: include: A multi-source heterogeneous data fusion module, wherein the multi-source heterogeneous data fusion module is used to obtain multiple BOM data of a target product, and perform data integration processing on each BOM data to obtain multiple processed BOM data of the same format; An adaptive domain knowledge graph module, wherein the adaptive domain knowledge graph module stores a BOM domain knowledge graph, wherein the adaptive domain knowledge graph module is used to receive each processed BOM data transmitted by the multi-source heterogeneous data fusion module, and perform data matching on each processed BOM data based on the BOM domain knowledge graph to obtain a data matching result; A dynamic context enhancement prompt module, the dynamic context enhancement prompt module is used to generate a multi-dimensional analysis prompt question and answer text based on the data matching result, and transmit the multi-dimensional analysis prompt question and answer text to the large language model module; A large language model module, wherein the large language model module stores a BOM large language model, wherein the large language model module is used to input the multidimensional analysis prompt question and answer text into the BOM large language model to obtain a question and answer result corresponding to the multidimensional analysis prompt question and answer text, and transmit the question and answer result to the multidimensional analysis module; The multidimensional analysis module is used to perform multidimensional analysis on the question and answer results to obtain multidimensional analysis results, and to fuse the multidimensional analysis results to obtain multimodal analysis results of the target product, and to visualize the multimodal analysis results, wherein the multidimensional analysis includes analysis on the cost dimension, performance dimension, safety dimension and supply chain dimension of the target product.

2. According to claim 1, a multidimensional analysis system for BOM data based on a large language model is characterized in that: The dynamic context enhanced prompt module includes: an input parser, a context analyzer, a prompt template library, a dynamic prompt generator and a prompt validator; An input parser, used for receiving a plurality of processed BOM data transmitted by the multi-source heterogeneous data fusion module, and extracting key information from each processed BOM data to obtain key information corresponding to each processed BOM data; A context analyzer, which is used to use a deep learning model to perform context encoding processing on each key information to extract context semantic information of each key information, and match a prompt template corresponding to the target product from a prompt template library based on each context semantic information; A dynamic prompt generator, used to extract historical prompt information from the historical prompt buffer area, and to extract industry knowledge of the target product from the industry rule engine; The dynamic prompt generator is further used to generate an initial multi-dimensional analysis prompt question and answer text based on the contextual semantic information of each key information, the historical prompt information and the industry knowledge, and adopts the prompt generation strategy and the matched prompt template; The prompt verifier is used to verify the initial multidimensional analysis prompt question and answer text, and after the verification passes, transmit the initial multidimensional analysis prompt question and answer text as the multidimensional analysis prompt question and answer text to the large language model module.

3. The multidimensional analysis system of BOM data based on a large language model according to claim 2, characterized in that: The dynamic context enhancement prompt module also includes: a user feedback interface, a performance evaluation unit and a reinforcement learning optimizer; A user feedback interface, used to obtain, in response to human-computer interaction, first feedback information of the user on the multidimensional analysis prompt question and answer text and second feedback information on the question and answer result corresponding to the multidimensional analysis prompt question and answer text, and send the first feedback information and the second feedback information to the performance evaluation unit; a performance evaluation unit, configured to receive first feedback information and second feedback information sent by the user feedback interface, and obtain prompt template evaluation information based on the first feedback information and the second feedback information, so as to update each prompt template in the prompt template library according to the prompt template evaluation information; A performance evaluation unit, configured to receive the question-answering result corresponding to the multi-dimensional analysis prompt question-answering text sent by the large language model module, and to perform a performance evaluation on the dynamic prompt generator based on the question-answering result to obtain an evaluation result; A performance evaluation unit, further configured to send the evaluation result to the reinforcement learning optimizer; A reinforcement learning optimizer is used to optimize the prompt generation strategy of the dynamic prompt generator based on the evaluation result and using a reinforcement learning strategy, so as to complete the feedback learning of the dynamic prompt generator after the optimization.

4. The multidimensional analysis system of BOM data based on a large language model according to claim 1, characterized in that: The multidimensional analysis module includes: a multidimensional impact analysis unit and a multimodal result interpretation and visualization unit; A multi-dimensional impact analysis unit, used to obtain evaluation parameters, wherein the evaluation parameters include analysis dimensions; A multi-dimensional impact analysis unit, used to match analysis models corresponding to various analysis dimensions from an analysis model library according to the evaluation parameters, and to receive the question-answering results transmitted by the large language model module, and input the question-answering results into each matched analysis model to obtain a multi-dimensional analysis result of the target product; The multi-dimensional impact analysis unit is further used to integrate the multi-dimensional analysis results to obtain the comprehensive analysis results of the target product, and send the comprehensive analysis results to the multi-modal result interpretation and visualization unit; The multimodal result interpretation and visualization unit is used to generate visualization data, summary text and result interpretation information corresponding to the comprehensive analysis result according to the comprehensive analysis result, and perform multimodal fusion processing on the visualization data, summary text and result interpretation information to obtain the multimodal analysis result of the target product after the multimodal fusion processing.

5. The multidimensional analysis system of BOM data based on a large language model according to claim 4 is characterized in that: The multimodal result interpretation and visualization unit includes: an output parsing unit, a key information extraction unit, a data classification and structuring unit, a visualization data generation unit, a summary text generation unit, an AI interpretation unit, and a multimodal display integration unit; An output parsing unit, used to parse the comprehensive analysis result to obtain parsing information corresponding to the comprehensive analysis result, and transmit the parsing information to the key information extraction unit; A key information extraction unit, configured to perform key information extraction processing on the parsed information to obtain a plurality of key information corresponding to the parsed information, and transmit each key information corresponding to the parsed information to a data classification and structuring unit; A data classification and structuring unit, used to classify each key information corresponding to the parsed information to obtain a plurality of classified information, and to structure each classified information to obtain a plurality of structured classified information; A visualization data generating unit, configured to generate visualization data of each structured classification information according to each structured classification information, and transmit the visualization data of each structured classification information to the multimodal display integration unit, wherein the visualization data of any structured classification information includes a static chart, an interactive chart and / or a network relationship map; A summary text generation unit is used to use a natural language processing algorithm to perform text summary extraction processing on each structured classification information, obtain summary texts corresponding to each structured classification information, and transmit each summary text to the multimodal display integration unit; The AI ​​interpretation unit is used to perform feature importance analysis on each structured classification information to obtain a feature importance analysis result of each structured classification information, and transmit the feature importance analysis result of each structured classification information as result interpretation information of each structured classification information to the multimodal display integration unit; The multimodal display integration unit is used to perform multimodal fusion processing on the visual data, summary text and result interpretation information corresponding to each structured classification information, so as to obtain the multimodal analysis result of the target product after the multimodal fusion processing.

6. The multidimensional analysis system of BOM data based on a large language model according to claim 4, characterized in that: If there is a missing information problem in the multi-dimensional analysis process, the multi-dimensional impact analysis unit is used to identify the missing information, generate a missing information query question based on the missing information, and visualize the missing information query question; A multi-dimensional impact analysis unit, configured to obtain, in response to human-computer interaction, feedback input information of a user for the missing information query question, and analyze the feedback input information to obtain an input analysis result; A multi-dimensional impact analysis unit, used to adjust the model parameters of the analysis model corresponding to each analysis dimension based on the input analysis result to obtain an adjusted analysis model; The multi-dimensional impact analysis unit is also used to re-input the input analysis results and the question and answer results into each adjusted analysis model to obtain a multi-dimensional analysis result of the target product.

7. The multidimensional analysis system of BOM data based on a large language model according to claim 4, characterized in that: The multidimensional impact analysis unit is used to generate a decision suggestion for the target product based on the comprehensive analysis result, and to visualize the decision suggestion; as well as It is also used to update each analysis model in the analysis model library according to the comprehensive analysis results and user feedback information, so as to complete the continuous reinforcement learning of each analysis model during use.

8. The multidimensional analysis system of BOM data based on a large language model according to claim 1, characterized in that: The multi-source heterogeneous data fusion module includes: a data cleaning and standardization unit, an entity recognition and relationship extraction unit, a semantic integration unit, a data integration unit and a data quality assessment unit; A data cleaning and standardization unit is used to perform data cleaning and data standardization on each BOM data of the target product in turn to obtain a variety of standardized BOM data, and send the multiple standardized BOM data to the entity recognition and relationship extraction unit; An entity recognition and relationship extraction unit is used to perform entity recognition and relationship extraction processing on each standardized BOM data to obtain the entity relationship of each standardized BOM data; A semantic integration unit is used to perform semantic integration processing on each standardized BOM data based on the entity relationship of each standardized BOM data to obtain standardized BOM data after multiple semantic integration; A data integration unit is used to perform format conversion processing on standardized BOM data after multiple semantic integration, so as to obtain multiple processed initial BOM data with the same format after the format conversion processing; The data quality evaluation unit is used to perform quality evaluation on the multiple processed initial BOM data, and after the quality evaluation is passed, use the multiple processed initial BOM data as the processed BOM data of the target product.

9. The multidimensional analysis system of BOM data based on a large language model according to claim 1, characterized in that: The adaptive domain knowledge graph module includes: a knowledge graph updating unit and a feedback learning unit; The knowledge graph updating unit is used to obtain knowledge graph update data and perform data preprocessing on the knowledge graph update data, so as to obtain named entities corresponding to the knowledge graph update data after the data preprocessing; The knowledge graph updating unit is further used to update the BOM domain knowledge graph according to the named entities corresponding to the knowledge graph update data to obtain an updated initial BOM domain knowledge graph, and to perform knowledge integration and verification on the updated initial BOM domain knowledge graph, so as to accept the update of the BOM domain knowledge graph after the knowledge integration and verification to obtain an updated BOM domain knowledge graph, so as to use the updated BOM domain knowledge graph for data matching; A feedback learning unit is used to perform knowledge application and reasoning processing on the updated BOM domain knowledge graph to obtain knowledge reasoning results; A feedback learning unit is used to obtain the third feedback information of the updated BOM domain knowledge graph in the multidimensional analysis process, and evaluate the importance of each node in the updated BOM domain knowledge graph according to the third feedback information, so as to adjust the weight of each node in the updated BOM domain knowledge graph according to the importance evaluation result to obtain weight adjustment information; The feedback learning unit is also used to obtain the question and answer results output by the large language model module, and input the knowledge reasoning results, weight adjustment results and / or question and answer results into the knowledge graph updating unit, so as to optimize the updated BOM domain knowledge graph through the knowledge graph updating unit, so as to complete the feedback learning of the updated BOM domain knowledge graph after optimization.

10. An analysis method of a multidimensional analysis system of BOM data based on a large language model according to any one of claims 1 to 9, characterized in that: include: Acquire multiple BOM data of target products, and perform data integration processing on each BOM data to obtain multiple processed BOM data with the same format; Based on the BOM domain knowledge graph, data matching is performed on each processed BOM data to obtain data matching results; Generate a multi-dimensional analysis prompt question and answer text according to the data matching result; Input the multidimensional analysis prompt question and answer text into the BOM large language model to obtain the question and answer result corresponding to the multidimensional analysis prompt question and answer text; Performing multi-dimensional analysis and processing on the question and answer results to obtain multi-dimensional analysis results; The multi-dimensional analysis results are fused to obtain a multimodal analysis result of the target product, and the multimodal analysis result is visualized, wherein the multi-dimensional analysis processing includes analysis processing of the cost dimension, analysis processing of the performance dimension, analysis processing of the safety dimension, and analysis processing of the supply chain dimension of the target product.

Citation Information

Cited By

  • Model distillation method and system based on teacher model and situational reasoning

    CN119539011A

  • Fusion inference system of large language model and knowledge graph

    CN119670885A

  • Software detection method and device based on software bill of material file

    CN119783109A

  • Symbol logical reasoning optimization system based on large language model

    CN119849632A

  • Material data quality evaluation method based on large model Agent

    CN119993341A