BOM analysis system based on photovoltaic industry vertical domain large model

Abstract

The invention discloses a BOM analysis system based on a photovoltaic industry vertical domain large model, and the system comprises a BOM knowledge management module which is used for obtaining data from one or more data sources and carrying out the screening of the data; the BOM data generation module is used for converting the standardized data provided by the BOM knowledge governance module into a multi-tuple data structure, and the multi-tuple data structure comprises model design information and physical part information; the BOM polymorphic storage module is used for storing the multi-tuple data structure output by the BOM data generation module and providing a query result through a reordering algorithm; the BOM intelligent analysis module is used for analyzing the multi-tuple data of the BOM polymorphic storage module and optimizing the weight configuration of the BOM polymorphic storage module; and the BOM enhanced design module is used for generating candidate BOMs according to the multi-tuple data of the BOM polymorphic storage module, and outputting the optimal BOM to the BOM data generation module if the multi-rule evaluation meets the requirement.

Description

TECHNICAL FIELD

[0001] The present application relates to a BOM analysis system based on a photovoltaic industry vertical domain large model. BACKGROUND

[0002] In the field of photovoltaic manufacturing, the bill of materials (BOM) is the core data carrier of product life cycle management, running through key links such as research and development design, process planning, production and manufacturing, supply chain management, and post-operation and maintenance. BOM not only defines the composition structure of the product, but also carries multi-dimensional attributes such as material specifications, cost information, substitution relationship, and process path, and is the basis for enterprises to achieve fine operation and efficient collaboration.

[0003] With the acceleration of photovoltaic product iteration and the increasing demand for customization, the traditional BOM management method has become unsuitable. In the traditional BOM management method, the BOM structure is complex and has a deep hierarchy, and manual maintenance is inefficient and prone to errors; design experience is mostly dependent on individual knowledge of engineers, making it difficult to systematize and reuse; there is a serious data island between different systems (such as PLM, ERP, and MES), and the collaboration efficiency is low; the development of new product models lacks intelligent assistance, resulting in long development cycles and high trial and error costs.

[0004] Most current photovoltaic enterprises use traditional BOM management systems. The typical structure includes a data source layer and a structured storage layer. The data source layer: the design side generates an initial BOM through CAD / EDA tools, which is imported into the PLM system after engineering approval; the structured storage layer: the PLM system organizes the BOM in a tree-like hierarchical structure (parent-child relationship), supports version control, change management (ECN), and material substitution rule configuration. Although the above scheme meets the basic BOM management needs to some extent, there are still obvious shortcomings in dealing with complex design decisions, knowledge reuse, and intelligent upgrades: (1) Lack of general solutions due to high customization and sensitivity. Photovoltaic inverter BOM has significant enterprise-specific characteristics: there are significant differences in design and other aspects between different manufacturers, and a large amount of sensitive information is involved. (2) Lack of semantic understanding ability, unable to support natural language interaction. Existing systems are mostly based on structured query language or fixed report templates, and users must be familiar with field names and system operation logic. (3) Directly calling a general large model has a serious "hallucination" problem. If the vertical domain information is missing, directly using a general large language model (such as GPT series, Tongyi Qianwen, etc.) to process BOM-related tasks will result in a serious hallucination problem. Even if the vertical domain information is supplemented, the large model cannot fully understand it. (4) Knowledge sedimentation is difficult, and it is difficult to form a closed-loop iteration mechanism. Traditional systems only support static data storage and cannot automatically extract engineers' design experience. Due to the accumulation of a large amount of different forms of inventory data from project iteration, "data islands" and "experience faults" have existed for a long time.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] This invention provides a BOM analysis system based on a large vertical model of the photovoltaic industry, which solves the problems of data fragmentation, structure loss, inefficient reasoning, and unusable output in BOM management.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A BOM analysis system based on a large-scale vertical model of the photovoltaic industry, comprising: The BOM knowledge governance module is used to obtain and filter data from one or more data sources. The BOM data generation module is used to convert the standardized data provided by the BOM knowledge governance module into a tuple data structure, which includes model design information and physical part information. The BOM polymorphic storage module is used to store the tuple data structure output by the BOM data generation module and provide query results through a reordering algorithm. The BOM intelligent analysis module is used to analyze the tuple data of the BOM polymorphic storage module and optimize the weight configuration of the BOM polymorphic storage module. The BOM enhancement design module is used to generate candidate BOMs based on the tuple data of the BOM polymorphic storage module. If the multi-rule evaluation meets the requirements, the optimal BOM is output to the BOM data generation module; if the requirements are not met, the strategy is adjusted.

[0008] In some preferred embodiments, the one or more databases include relational databases, document databases, and network data; The BOM knowledge governance module includes a BOM knowledge extraction component, a BOM data processing component, and a BOM service management component. The BOM knowledge extraction component is used to extract data from the database. The BOM data processing component is used to perform data cleaning, data filtering, and data anonymization on the extracted data. The BOM service management component is used to monitor data quality and provide data service interfaces. The data service interfaces include one or more of the following: query, subscription, statistics, and lineage tracing.

[0009] In some preferred embodiments, the BOM data generation module includes a tuple data generation component, a tuple data parsing component, a data fusion component, and an output management component. The tuple data generation component receives data from the BOM knowledge governance module and generates BOM information. The BOM information includes product model and design-related information, information on the physical components constituting the product, and information related to sales regions and / or market strategies, or information related to potential product failure risks. The tuple data parsing component parses the data generated by the BOM enhanced design module, verifies the logical and format consistency of the data, filters invalid or redundant data, and maps the original data to a standardized tuple structure. The data fusion component correlates the BOM information from the tuple data generation component and the standardized tuple structure from the tuple data parsing component to construct a semantic relationship graph between BOM entities, and fuses them to generate new BOM data. The output management component includes a visual interface for users to preview or edit data, and is also used to export or transmit the processed BOM data to the BOM polymorphic storage module.

[0010] In some preferred embodiments, the BOM polymorphic storage module includes a preprocessing component for processing the data output by the BOM data generation module into module data. More preferably, the BOM polymorphic storage module includes a storage engine component for building indexes and managing data storage; the storage engine component includes: an index database submodule for providing a database system for storing and retrieving index data; a meta storage submodule for storing metadata information and supporting efficient access and management; a Neo4j graph storage submodule for persistent storage of graph data using Neo4j; and a vector storage submodule for storing high-dimensional vector data generated by the embedding model. The BOM polymorphic storage module includes a retrieval engine component, which is used to retrieve data; The BOM polymorphic storage module includes a reordering component, which is used to reorder the retrieval results.

[0011] In some preferred embodiments, the BOM intelligent analysis module includes a user interaction component, which is used to receive and parse user input requests or needs and transmit processing results. More preferably, the BOM intelligent analysis module includes an administrator agent component, which is used to analyze and understand user intent, distribute tasks to corresponding modules or components according to the analyzed user intent, perform in-depth analysis of the processed results to ensure that user needs are met, and provide feedback on the processing results. The BOM intelligent analysis module includes an expert agent component, which is used to perform in-depth analysis and evaluation of the task of the classification method; the expert agent component includes a performance specification agent, an ODM agent, and a safety regulation agent; The BOM intelligent analysis module includes an MCP component, which provides a tool library for intelligent agents to use; the MCP component includes a query engine and a sandbox component.

[0012] In some preferred embodiments, the BOM reinforcement design module is used to apply a change strategy to differentiate candidate BOMs, and the reward model of the change strategy is as follows:

[0013] In the above formula, This represents the value function of agent i in state s oriented towards the group's goal. Represents the mathematical expectation. This represents the relative reward of agent i. Indicates the discount factor. This represents the group value function of agent i when the environment transitions to a new state s' after performing an action. Indicates the weighting coefficient. This represents the average reward for the group.

[0014] In some preferred embodiments, the BOM enhancement design module includes a data generation component, which is used to generate candidate BOMs based on scenarios or user types, and to comprehensively evaluate the generated results by combining multi-dimensional rules including technology, cost and compliance, and output the optimal BOM.

[0015] In some preferred embodiments, the BOM analysis system further includes a BOM system integration module, which is used to synchronize data to external systems through mapping relationships to ensure compatibility with external systems.

[0016] In some preferred embodiments, the BOM system integration module includes a user interaction component for user interaction. The BOM system integration module includes a data search component, which is used to search and integrate data streams related to the external system; The BOM system integration module includes compatibility and adaptation components for data format conversion and expansion compatibility with external systems.

[0017] In some preferred embodiments, the BOM knowledge governance module is used to obtain inverter-related BOM data from one or more databases, and the BOM enhancement design module is used to output the optimal BOM for the inverter.

[0018] The above-mentioned solution adopted in this invention has the following advantages: The BOM analysis system based on a large vertical model of the photovoltaic industry of this invention is compatible with the data formats of multi-source heterogeneous systems such as PLM, ERP, and CRM. It not only retains key structural information such as the hierarchical relationship of materials, usage configuration and substitution rules, but also achieves efficient alignment and integration of cross-system data through unified encoding and semantic tags. It retains the tree topology and contextual dependencies of the BOM, improving the accuracy of understanding complex product structures. It introduces a multi-agent architecture based on reinforcement learning optimization, which significantly shortens the time of a single inference and improves the system response speed. Attached Figure Description

[0019] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a block diagram of a BOM analysis system according to an embodiment of the present invention.

[0021] Figure 2 This is a module diagram of the BOM knowledge governance module according to an embodiment of the present invention.

[0022] Figure 3 This is a module diagram of the BOM data generation module according to an embodiment of the present invention.

[0023] Figure 4 This is a visualization interface for a BOM data model according to an embodiment of the present invention.

[0024] Figure 5 This is a module diagram of a BOM polymorphic storage module according to an embodiment of the present invention.

[0025] Figure 6 This is a module diagram of the BOM intelligent analysis module according to an embodiment of the present invention.

[0026] Figure 7 This is a block diagram of the BOM reinforcement design module according to an embodiment of the present invention.

[0027] Figure 8 This is a flowchart illustrating the BOM (Bill of Materials) reinforcement design according to an embodiment of the present invention.

[0028] Figure 9 This is a module diagram of the BOM system integration module according to an embodiment of the present invention. Detailed Implementation

[0029] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art. It should be noted that the description of these embodiments is for the purpose of aiding understanding the present invention, but does not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0030] The embodiments involve a novel BOM (Bill of Material) data model that incorporates multimodal data, uses tuples instead of tables, preserves hierarchical structure and semantic relationships, and is better adapted to large language model inference and reinforcement learning algorithm training. The embodiments also involve an automated material processing tool for converting existing and newly generated data, compatible with various current software systems. Furthermore, the embodiments involve an intelligent agent for the photovoltaic field, introducing a vRag database to store novel BOM data. A multi-agent collaborative workflow accelerates the business analysis process, enabling intelligent BOM analysis, similar solution recommendation, and comprehensive evaluation, significantly improving design efficiency and accuracy.

[0031] The embodiment provides a BOM analysis system based on a large vertical model of the photovoltaic industry. It solves the problems of data fragmentation, structure loss, inefficient reasoning, and unusable output in traditional BOM management by combining data modeling with a multi-agent collaborative architecture.

[0032] 1. Construct a standardized, semantically rich BOM data model to break down system barriers. The example design utilizes a BOM data structure optimized for generative AI, integrating a metadata management system to support unified descriptions of material attributes, hierarchical relationships, configuration rules, process constraints, and other multi-dimensional information. This model is compatible with the data interfaces of mainstream PLM / ERP systems, enabling rapid access and structured import without altering the existing IT architecture, significantly improving data availability and model training efficiency.

[0033] 2. Storing complete BOM polymorphic information enhances the understanding and reasoning capabilities of large models. By introducing graph structure encoding and hierarchical representation learning mechanisms, the implementation example fully preserves the tree-like topology and semantic dependencies of the BOM during the embedding process, avoiding information loss. This enables large models not only to "read" the BOM content but also to "understand" its internal logic, providing reliable support for subsequent tasks such as similarity matching and change impact analysis.

[0034] 3. An optimized multi-agent workflow is employed to improve analysis efficiency and output quality. The example constructs a multi-agent collaborative system based on reinforcement learning tuning, decomposing the BOM analysis task into multiple sub-tasks such as "parse → compare → recommend → verify," which are executed in parallel by specialized agents. Compared to serial processing by a single model, this significantly shortens inference time and compensates for the limitations of individual models in vertical domain knowledge through task division. Each agent's output is returned in structured JSON or standardized API format, which can be seamlessly embedded into existing enterprise R&D processes and information systems.

[0035] 4. Supports customized recommendations and evaluations for new models, promoting the formation of a knowledge loop. Based on the traditional historical BOM library and design experience library, the system can automatically identify similar product solutions, generate preliminary BOM suggestions for new models, and conduct multi-dimensional evaluations such as cost, manufacturability, and supply chain risks. As usage frequency increases, the system continuously accumulates feedback data, driving iterative optimization of the model and gradually building a sustainably evolving super BOM intelligent agent.

[0036] Figure 1 The overall architecture of BOM analysis system 2, based on a large-scale vertical model of the photovoltaic industry, is shown. See also... Figure 1 As shown, the BOM analysis system 2 includes: a BOM knowledge governance module 21, used to acquire and filter data from one or more data sources 1 (such as original BOM data, local multi-source data, network data, etc.); a BOM data generation module 22, used to convert the standardized data provided by the BOM knowledge governance module 21 into a tuple data structure, which includes model design information and physical part information; a BOM polymorphic storage module 23, used to store the tuple data structure output by the BOM data generation module 22 and provide query results through a reordering algorithm; and a BOM intelligent... Analysis module 24 analyzes the tuple data of BOM polymorphic storage module 23 and optimizes the weight configuration of BOM polymorphic storage module 23; BOM enhancement design module 25 generates candidate BOMs based on the tuple data of BOM polymorphic storage module 23. If the multi-rule evaluation meets the requirements, the optimal BOM is output to BOM data generation module 22; otherwise, the strategy is adjusted; and BOM system integration module 26 synchronizes data to external system 3, such as PLM system, MySQL database, ERP system, etc., through mapping relationships to ensure compatibility with external systems. The above data sources mainly target data outside the system, used to improve the accuracy of BOM design and automatically check database quality.

[0037] Module 21 of the BOM (Bill of Materials) Knowledge Governance module is responsible for data collection and cleaning. (Refer to...) Figure 2As shown, the BOM knowledge governance module 21 connects to data source 1 to collect raw data, which comes from relational databases, local multi-source data (document databases), and web data (obtained by web crawlers). This module 21 automatically acquires and filters data, providing an interface for subsequent BOM data generation module 22 to call. The BOM knowledge governance module 21 includes a BOM knowledge extraction component 211, a BOM data processing component 212, and a BOM service management component 213.

[0038] like Figure 2 As shown, the BOM knowledge extraction component 211 is used to extract data from the database and is a core component of the BOM knowledge governance module 21. The BOM knowledge governance module 21 includes the following sub-modules: document database extraction, which extracts content from various local document databases and supports user uploads; relational database extraction, which extracts content from relational databases; and a web crawler engine, which extracts content from online information sources such as industry standard websites and competitor company websites. The BOM data processing component 212 is used to clean, filter, and anonymize the extracted data to ensure data quality. The BOM data processing component 212 includes the following three sub-modules: a data cleaning sub-module, used for deduplication, format standardization, null value handling, outlier detection, and error correction; a data filtering sub-module, used for business rule filtering, timeliness filtering, and tiered scoring; and a data anonymization sub-module, used for anonymizing personal information and protecting sensitive information. These three sub-modules form a pipeline processing flow. BOM Service Management Component 213 is used for data quality monitoring and provides data service interfaces, offering business-oriented service capabilities. It is the main interface layer for the system, and the data service interfaces include one or more of query, subscription, statistics, and lineage tracing. BOM Service Management Component 213 includes the following sub-modules: a data quality monitoring sub-module, which monitors integrity, accuracy, consistency, and timeliness, and generates monitoring logs; a query service sub-module, used for querying data; a subscription service sub-module, used for subscribing to updates; a statistics service sub-module, used for generating statistical logs; and a lineage tracing sub-module, used for tracing data sources, change history, and personnel involved in data changes.

[0039] The BOM data generation module 22 is responsible for assembling data into a new data model. As the core data modeling component of the new BOM analysis system 2, it transforms the standardized data provided by the knowledge governance module into a multi-dimensional data structure containing model design information, physical part information, sales market information, and fault risk information. This provides a rich data foundation for the subsequent BOM polymorphic storage module 23, BOM intelligent analysis module 24, and BOM enhanced design module 25. (Refer to...) Figure 3As shown, the BOM data generation module 22 includes a tuple data generation component 221, a tuple data parsing component 222, a data fusion component 223, and an output management component 224.

[0040] like Figure 3 As shown, the tuple data generation component 221 receives data from the BOM knowledge governance module 21 and generates BOM information. The BOM information includes product model and design-related information, as well as information on the physical components constituting the product. The BOM information also includes information related to sales regions and / or market strategies, or information related to potential product failure risks. The tuple data generation component 221 includes the following sub-modules: model design information generation, used to generate product model and design-related information; physical component information generation, used to generate information on the physical components constituting the product; sales market information generation, used to generate BOM information related to sales regions and market strategies; and failure risk information generation, used to generate data related to potential product failures and risks.

[0041] like Figure 3 As shown, the tuple data parsing component 222 is used to parse the data generated by the BOM reinforcement design module 25, verify the consistency of the data in logic and format, filter invalid or redundant data, and map the original data (which can be the data generated by the BOM reinforcement design module 25) into a normalized tuple structure. The tuple data parsing component 222 includes the following sub-modules: a consistency verification sub-module, used to verify the consistency of the data (BOM data output by the BOM reinforcement design module 25) in logic and format; a data filtering sub-module, used to filter invalid or redundant data according to rules; and a tuple information mapping sub-module, used to map the original data (BOM data output by the BOM reinforcement design module 25) into a normalized tuple structure.

[0042] like Figure 3 As shown, the data fusion component 223 is used to correlate BOM information from the tuple data generation component 221 and the standardized tuple structure from the tuple data parsing component 222, constructing a semantic relationship graph between BOM entities and fusing them to generate new BOM data. The data fusion component 223 includes the following sub-modules: tuple association, used to correlate tuple data from different sources; and relationship graph construction, used to construct a semantic relationship graph between BOM entities. Specifically, an entity in a material BOM may contain multiple elements, sub-BOM material numbers, design drawings, related documentation, business-related work orders, and other information.

[0043] like Figure 3As shown, the output management component 224 includes a visual interface for users to preview or edit data. The output management component 224 is also used to export or transfer the processed BOM data to the BOM polymorphic storage module 23. The output management component 224 includes the following sub-modules: a data preview and modification sub-module, used to provide a visual interface for users to preview and edit data; and a data export sub-module, used to export the processed BOM data to the BOM polymorphic storage module 23.

[0044] Figure 4 This illustrates a visualized inverter BOM data model generated by the BOM data generation module 22. (Refer to...) Figure 4 As shown, the BOM model of this inverter includes the inverter's IGBT module, PCBA module, etc., and is associated with thermal risk management that needs to be paid attention to in the module, component aging information, market information, etc., and includes multimodal data such as installation manual, maintenance functions, product appearance pictures, internal structure pictures, installation demonstration videos and fault diagnosis videos.

[0045] The BOM polymorphic storage module 23 is responsible for storing data and dynamically adjusting weights. As the core storage and retrieval engine of the BOM analysis system 2, it utilizes the Neo4j graph database and VRAG technology to achieve efficient storage, hybrid retrieval, and intelligent re-ranking of multimodal tuple data. This module supports multimodal data storage and provides users with the most accurate query results through a configurable re-ranking algorithm.

[0046] Reference Figure 5 As shown, the BOM polymorphic storage module 23 includes a preprocessing component 231, which processes the data output by the BOM data generation module 22 into module data. The preprocessing component 231 includes the following sub-modules: a data verification sub-module, used to verify and clean the input data to ensure the integrity and correctness of the data; and an embedding model, used to convert text or other types of data into vector form for easier subsequent processing.

[0047] like Figure 5 As shown, the BOM polymorphic storage module 23 includes a storage engine component 232, which is used to establish indexes and manage data storage. The storage engine component 232 includes the following sub-modules: an index database, which is a database system providing fast lookup capabilities for storing and retrieving index data; a meta storage sub-module, used to store metadata information, supporting efficient access and management; a Neo4j graph storage sub-module, used for persistent storage of graph data using Neo4j, facilitating complex relationship queries; and a vector storage sub-module, used to store high-dimensional vector data generated by the embedding model.

[0048] like Figure 5As shown, the BOM polymorphic storage module 23 includes a retrieval engine component 233, which is used to retrieve data. The retrieval engine component 233 includes the following sub-modules: a strategy configuration sub-module, used to set different retrieval strategies and parameters according to business needs; and a hybrid retrieval sub-module, used to combine multiple retrieval methods (such as keywords, vector similarity, etc.) to provide comprehensive retrieval services. The BOM intelligent analysis module 24, the BOM enhanced design module 25, or the BOM system integration module 26 can retrieve the required data through the retrieval engine component 233.

[0049] like Figure 5 As shown, the BOM polymorphic storage module 23 includes a reordering component 234, which is used to reorder the search results to improve the acceptance rate. The BOM polymorphic storage module 23 includes the following sub-modules: a reordering model sub-module, used to apply machine learning or other methods to optimize the sorting of search results; a data scheduling sub-module, used to control and optimize the data flow to improve processing efficiency and response speed; and a result return sub-module, used to return the processed and sorted final results to the requester, which can be the BOM intelligent analysis module 24, the BOM enhanced design module 25, or the BOM system integration module 26.

[0050] The BOM Intelligent Analysis Module 24 is responsible for analyzing and updating BOM data. As the core intelligent engine of the BOM Analysis System 2, it employs a multi-agent workflow architecture to evaluate various business needs, including ODM, safety regulations, and performance specifications, deploying dedicated intelligent agents for in-depth analysis. This module features a feedback mechanism that continuously optimizes the weight configuration of the polymorphic storage module based on user feedback.

[0051] Reference Figure 6 As shown, the BOM intelligent analysis module 24 includes a user interaction component 241, which receives and parses user input requests or needs and transmits processing results. The user interaction component 241 includes the following sub-modules: a security verification sub-module, used to ensure the security of user input and prevent malicious data attacks; a requirement input sub-module, used to receive and parse user requests or requirement inputs; and a summary output sub-module, used to transmit the system processing results to the next node in a concise and clear manner, saving tokens.

[0052] Reference Figure 6As shown, the BOM intelligent analysis module 24 includes an administrator agent component 242. This administrator agent analyzes and understands user intent, distributes tasks to appropriate modules or components based on the analyzed user intent, performs in-depth analysis of the processed results to ensure user needs are met, and provides feedback on the processing results. The administrator agent component 242 includes the following sub-modules: an intent recognition sub-module, used to analyze and understand user intent to better meet their needs; a task distribution sub-module, used to assign tasks to suitable modules or components based on the analyzed user intent; a result analysis sub-module, used to perform in-depth analysis of the processed results to ensure user needs are met; and a result feedback sub-module, used to provide detailed processing result feedback to the user or other system components (such as the policy configuration sub-module of the BOM polymorphic storage module 23).

[0053] Reference Figure 6 As shown, the BOM intelligent analysis module 24 includes an expert agent component 243, which serves as an expandable group of specialized intelligent agents for in-depth analysis and evaluation of the tasks assigned. The expert agent component 243 may include a performance specification agent, an ODM agent, and a safety compliance agent. The expert agent component 243 includes the following sub-modules: a performance specification agent, an agent focused on performance-related issues such as performance evaluation and optimization suggestions; an ODM agent, an agent handling tasks related to operational data management, such as database operation optimization; and a safety compliance agent, an agent ensuring all operations comply with safety specifications and standards.

[0054] The BOM intelligent analysis module 24 includes an MCP component 244, which provides a toolkit for agents to use. The MCP component 244 may include a query engine and a sandbox component. The query engine provides powerful query capabilities, supporting complex query requirements. The sandbox component provides an isolated testing environment for safely testing new features or code.

[0055] Reference Figure 7 and Figure 8 The BOM Enhancement Design Module 25 is responsible for intelligently recommending and designing new BOMs. Specifically, based on the GRPO (Group Relative Policy Optimization) algorithm, an intelligent BOM design assistant is designed, which can automatically generate the optimal bill of materials according to user needs. The system analyzes historical BOM data, learns design patterns, and generates BOM configurations that meet user requirements.

[0056] Each BOM is treated as a state in reinforcement learning, and its state value can be calculated based on different tuples in the super BOM data model.

[0057] Using a group of intelligent agents, we can derive differentiated changes to the existing BOM based on user demand characteristics. This can be considered as a change strategy based on this state.

[0058] The reward model for this change strategy is obtained based on the Agent with different tuple dimensions, as shown in the following formula:

[0059] In the above formula, V_i^group(s) represents the group-oriented value function of agent i in state s; it measures the expected long-term benefit (possibly including trade-offs between the agent's own interests and those of others) that agent i will bring to the entire group after adopting a certain strategy starting from the current state s. This is not the individual value function V(s) in traditional RL, but an extended version that incorporates a group perspective.

[0060] E[] represents the mathematical expectation, which is the average of random variables (such as environmental transitions, action choices, and the behavior of other agents). The expectation in the first term is usually the expectation of the next state s′ and the subsequent trajectory; the second term may be the expectation over multiple time steps or multiple agents.

[0061] R_i^relative represents the relative reward for agent i. Its purpose is to encourage agents to exceed the group average, and it is often used to introduce competition, avoid the "free-rider" problem, or incentivize contribution in cooperation.

[0062] Γ represents the discount factor, 0 ≤ γ < 1, used to weigh the importance of current rewards against future rewards. It is a standard parameter in reinforcement learning, ensuring convergence of infinite horizon rewards.

[0063] V_i^group(s') represents the group value function of agent i after the environment transitions to a new state s' following the execution of an action. Together with the first term, it forms the structure of the Bellman Equation: current reward + discounted future value.

[0064] A represents the scalar hyperparameter, which controls the strength of the relative reward. If α=0, it degenerates into the standard group value function; if α>0, it explicitly encourages agents to obtain rewards higher than the group average, enhancing individual motivation or competitiveness.

[0065] This represents the group average reward. It serves as a baseline to measure whether an individual's performance is better than the group's average.

[0066] Finally, the generated candidate BOMs will be evaluated to determine whether they meet the requirements, such as... Figure 8 As shown.

[0067] like Figure 7 As shown, the BOM enhancement design module 25 includes a data generation component 254. This component generates candidate BOMs based on scenarios or user types, comprehensively evaluates the generated results using multi-dimensional rules including technology, cost, and compliance, and outputs the optimal BOM. This optimal BOM can be fed back to the BOM data generation module 22. The data generation component 254 standardizes its output data, including the following sub-modules: an application differentiation strategy sub-module, used to apply differentiated generation strategies based on scenarios or user types; and a multi-rule evaluation sub-module, used to comprehensively evaluate the generated results using multi-dimensional rules (technology, cost, compliance, etc.).

[0068] like Figure 7 As shown, the BOM enhanced design module 25 includes a preprocessing component 251 for processing data obtained from the BOM polymorphic storage module 23. The preprocessing component 251 includes the following sub-modules: a user requirement input sub-module for receiving and structuring the original requirement information submitted by the user; a requirement feature extraction sub-module for extracting key features (such as functions, parameters, constraints, etc.) from the user requirements; and a security verification sub-module for performing legality and security checks on the input content to prevent malicious or invalid input.

[0069] like Figure 7 As shown, the BOM-enhanced design module 25 includes an intelligent agent swarm strategy model component 252, which serves as an extensible learning agent swarm. The intelligent agent swarm strategy model component 252 includes the following sub-modules: a performance specification agent, used to analyze and generate agents that meet performance indicators (such as power consumption, speed, load, etc.); and a safety compliance agent, used to ensure that the design scheme complies with safety specifications and industry standards (such as electrical and mechanical safety).

[0070] like Figure 7 As shown, the BOM reinforcement design module 25 includes a reward model component 253, which serves as a variable evaluation agent group. This reward model component 253 may include a cost agent for evaluating solution costs and providing cost optimization-oriented reward signals.

[0071] like Figure 7As shown, the BOM reinforcement design module 25 includes a reference model component 255, which is a tool library for the evaluation agent to use. The reference model component 255 includes the following sub-modules: a similar BOM retrieval sub-module, used to retrieve BOMs with similar structure or function based on historical data as a reference; and a similar requirement retrieval sub-module, used to retrieve similar user requirements and their solutions from the past to assist in current decision-making.

[0072] The BOM system integration module 26 is responsible for synchronizing data with external systems through mapping relationships. This module serves as the core integration hub between the new BOM analysis system 2 and traditional enterprise information systems, handling data synchronization, business process integration, and real-time collaboration with MySQL databases, PLM systems, and ERP systems. This module ensures that the new BOM analysis system 2 can seamlessly integrate into the existing enterprise IT ecosystem, guaranteeing business continuity.

[0073] Reference Figure 9 As shown, the BOM system integration module 26 includes a user interaction component 261 for user interaction, such as handling cross-external system operations. The user interaction component 261 includes the following sub-modules: a security verification sub-module for verifying input content to prevent unauthorized access or malicious input; a requirement input sub-module for receiving and parsing requirement information as the starting point for module processing; and a result feedback sub-module for returning the system processing results to the user in a visual or structured form.

[0074] like Figure 9 As shown, the BOM system integration module 26 includes a data search component 262, which is used to search and integrate data streams related to external systems. The data search component 262 includes the following sub-modules: a transaction management sub-module, used to manage transaction consistency during data query and operation processes to ensure data integrity; a log monitoring sub-module, used to record and monitor key operation logs during the data search process for easy auditing and problem tracking; and a monitoring notification sub-module, used to automatically trigger alarms or notification mechanisms when anomalies or critical events occur.

[0075] like Figure 9 As shown, the BOM system integration module 26 includes a compatibility and adaptation component 263 for data format conversion and to extend compatibility with external systems. The compatibility and adaptation component 263 includes the following sub-modules: a relational database conversion sub-module for data format conversion with external relational databases; a PLM system conversion sub-module for data format conversion with Product Lifecycle Management (PLM) systems; and an ERP system conversion sub-module for data conversion with Requirements Planning (RP) systems.

[0076] In this embodiment, the data layer ultimately generates new BOM data. Raw data is collected from the BOM knowledge governance module 21, and the new BOM data is generated by the BOM data generation module 22. Finally, it is stored in the BOM polymorphic storage module 23 in the form of VRag. The application layer provides three applications: data analysis, reinforcement design, and external IT system synchronization. Users can call these applications according to their specific needs. Specifically, the BOM intelligent analysis module 24 calls the BOM polymorphic data storage module, taking BOM-related information as input and outputting a multi-dimensional BOM analysis report, while also updating the policy configuration of the BOM polymorphic data storage module. The BOM reinforcement design module 25 calls the BOM polymorphic data storage module, taking the user's natural language requirements as input and outputting a new BOM, which is then sent to the data generation module. The BOM system integration module 26 calls the BOM polymorphic data storage module, taking synchronization tasks as input and outputting corresponding IT system requests.

[0077] The embodiment designs a data model suitable for generative AI, employing a knowledge graph and generating tuple data using Neo4j; it utilizes a BOM polymorphic storage module 23, enabling rapid integration of new models and leveraging the improvements brought by large model updates; the BOM reinforcement design module 25 uses GRPO to fill in additional data; and the BOM intelligent analysis module 24 employs multi-agent inference to accelerate reasoning, shortening inference time and utilizing more vertical domain knowledge to generate results. The BOM analysis system 2 in this embodiment is compatible with data formats from multiple heterogeneous systems such as PLM, ERP, and CRM. It not only preserves key structural information such as material hierarchy, usage configuration, and substitution rules, but also achieves efficient alignment and integration of cross-system data through unified encoding and semantic tags. This effectively breaks down data barriers between traditional systems, providing a high-quality, understandable input foundation for large models. Traditional methods often lose key structural information such as parent-child relationships and assembly paths when converting BOMs into model input due to text truncation or flattening. This embodiment employs graph structure encoding and hierarchical representation learning techniques to accurately preserve the tree-like topology and contextual dependencies of the BOM during vectorization, significantly improving the accuracy of large models in understanding complex product structures, fully preserving BOM polymorphic information, and avoiding information loss during embedding. Unlike general solutions that rely on a single model for sequential processing, this implementation introduces a multi-agent architecture optimized based on reinforcement learning. Tasks such as BOM parsing, similarity comparison, alternative recommendation, and compliance verification are decomposed and executed in parallel by specialized agents. This mechanism significantly reduces the time required for a single inference iteration, improving system response speed. Simultaneously, by assigning roles, it compensates for the limitations of individual models in terms of domain-specific knowledge, enhancing the professionalism and credibility of the output results. The analysis results of this implementation are output in standardized JSON, API interfaces, or lightweight data packages, containing complete bills of materials, recommendation criteria, risk warnings, and other information. These results can be directly imported into PLM systems for version comparison or integrated into ERP systems for cost simulation. This solves the problems of "loose, uncontrollable, and difficult-to-integrate" output from general large models, truly achieving deep integration of AI capabilities with the enterprise's existing IT ecosystem.

[0078] As indicated in this specification and claims, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, and these steps and elements do not constitute an exclusive list; the method or apparatus may also include other steps or elements. The term "and / or" as used herein includes any combination of one or more of the associated listed items.

[0079] It should be noted that, unless otherwise specified, when a feature is referred to as "fixed" or "connected" to another feature, it can be directly fixed or connected to the other feature, or indirectly fixed or connected to the other feature. Furthermore, the descriptions of "up," "down," "left," and "right" used in this invention are only relative to the relative positional relationships of the various components of the invention in the accompanying drawings.

[0080] The above embodiments are merely illustrative of the technical concept and features of the present invention, and are preferred embodiments. Their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly, and they should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made according to the principles of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A BOM analysis system based on a large-scale vertical model of the photovoltaic industry, characterized in that, include: The BOM knowledge governance module is used to obtain and filter data from one or more data sources. The BOM data generation module is used to convert the standardized data provided by the BOM knowledge governance module into a tuple data structure, which includes model design information and physical part information. The BOM polymorphic storage module is used to store the tuple data structure output by the BOM data generation module and provide query results through a reordering algorithm. The BOM intelligent analysis module is used to analyze the tuple data of the BOM polymorphic storage module and optimize the weight configuration of the BOM polymorphic storage module. The BOM enhancement design module is used to generate candidate BOMs based on the tuple data of the BOM polymorphic storage module. If the multi-rule evaluation meets the requirements, the optimal BOM is output to the BOM data generation module; if the requirements are not met, the strategy is adjusted.

2. The BOM analysis system according to claim 1, characterized in that, The one or more databases include relational databases, document databases, and network data; The BOM knowledge governance module includes a BOM knowledge extraction component, a BOM data processing component, and a BOM service management component. The BOM knowledge extraction component is used to extract data from the database. The BOM data processing component is used to perform data cleaning, data filtering, and data anonymization on the extracted data. The BOM service management component is used to monitor data quality and provide data service interfaces. The data service interfaces include one or more of the following: query, subscription, statistics, and lineage tracing.

3. The BOM analysis system according to claim 1, characterized in that, The BOM data generation module includes a tuple data generation component, a tuple data parsing component, a data fusion component, and an output management component. The tuple data generation component receives data from the BOM knowledge governance module and generates BOM information. This BOM information includes product model and design-related information, information on the physical components constituting the product, and information related to sales regions and / or market strategies, or information related to potential product failure risks. The tuple data parsing component parses the data generated by the BOM enhanced design module, verifies the logical and format consistency of the data, filters invalid or redundant data, and maps the original data to a standardized tuple structure. The data fusion component correlates the BOM information from the tuple data generation component and the standardized tuple structure from the tuple data parsing component to construct a semantic relationship graph between BOM entities, fusing them to generate new BOM data. The output management component includes a visual interface for users to preview or edit data, and is also used to export or transmit the processed BOM data to the BOM polymorphic storage module.

4. The BOM analysis system according to claim 1, characterized in that, The BOM polymorphic storage module includes a preprocessing component, which is used to process the data output by the BOM data generation module into module data. The BOM polymorphic storage module includes a storage engine component, which is used to establish indexes and manage data storage; the storage engine component includes an index database submodule, which is used to provide a database system for storing and retrieving index data; The meta storage submodule is used to store metadata information and supports efficient access and management; the Neo4j graph storage submodule is used for persistent storage of graph data using Neo4j; and the vector storage submodule is used to store high-dimensional vector data generated by the embedding model. The BOM polymorphic storage module includes a retrieval engine component, which is used to retrieve data; The BOM polymorphic storage module includes a reordering component, which is used to reorder the retrieval results.

5. The BOM analysis system according to claim 1, characterized in that, The BOM intelligent analysis module includes a user interaction component, which is used to receive and parse user input requests or needs and transmit processing results. The BOM intelligent analysis module includes an administrator agent component, which is used to analyze and understand user intent, distribute tasks to the corresponding modules or components according to the analyzed user intent, conduct in-depth analysis of the processed results to ensure that user needs are met, and provide feedback on the processing results. The BOM intelligent analysis module includes an expert agent component, which is used to perform in-depth analysis and evaluation of the task of the classification method; the expert agent component includes a performance specification agent, an ODM agent, and a safety regulation agent; The BOM intelligent analysis module includes an MCP component, which provides a tool library for intelligent agents to use; the MCP component includes a query engine and a sandbox component.

6. The BOM analysis system according to claim 1, characterized in that, The BOM enhancement design module is used to apply change strategies to differentiate candidate BOMs. The reward model for the change strategy is as follows: ； In the above formula, This represents the value function of agent i in state s oriented towards the group's goal. Represents the mathematical expectation. This represents the relative reward of agent i. Indicates the discount factor. This represents the group value function of agent i when the environment transitions to a new state s' after performing an action. Indicates the weighting coefficient. This represents the average reward for the group.

7. The BOM analysis system according to claim 1 or 6, characterized in that, The BOM enhancement design module includes a data generation component, which generates candidate BOMs based on scenarios or user types. The generated results are then comprehensively evaluated using multi-dimensional rules, including those related to technology, cost, and compliance, to output the optimal BOM.

8. The BOM analysis system according to claim 1, characterized in that, The BOM analysis system also includes a BOM system integration module, which is used to synchronize data to external systems through mapping relationships to ensure compatibility with external systems.

9. The BOM analysis system according to claim 8, characterized in that, The BOM system integration module includes a user interaction component for users to interact with. The BOM system integration module includes a data search component, which is used to search and integrate data streams related to the external system; The BOM system integration module includes compatibility and adaptation components for data format conversion and expansion compatibility with external systems.

10. The BOM analysis system according to any one of claims 1 to 9, characterized in that, The BOM knowledge governance module is used to obtain inverter-related BOM data from one or more databases, and the BOM enhancement design module is used to output the optimal BOM for the inverter.

Images