Discrete manufacturing island production intelligent collaboration method and system based on OAG ontology

Abstract

The present application relates to the field of discrete manufacturing intelligence, and more particularly to a discrete manufacturing island production intelligent collaboration method and system based on an OAG ontology, comprising the following steps: real-time acquisition of cross-scene heterogeneous original data of island production full-scene, establishment of a mapping relationship, generation of a standardized mapping data set; based on the standardized mapping data set, combined with the core composition of the OAG vertical ontology, full-quantity data semantic conversion is completed, and an OAG ontology semantic information library is generated; combined with real-time production requirements, global situation analysis is completed, the OAG vertical ontology is used as a unified rule driver, a four-layer multi-agent system is started to complete collaborative verification and flexible collaborative decision-making, and standardized collaborative execution instructions are generated; the collaborative execution instructions are issued to the corresponding execution unit to complete full-link autonomous execution, and feedback data of the whole process is collected synchronously to complete closed-loop optimization. Through full-process autonomous collaboration and continuous iteration, the present application improves production efficiency and collaborative stability, and reduces scheduling cost and collaborative conflicts.

Description

Technical Field

[0001] This invention relates to the field of intelligent discrete manufacturing, and in particular to an intelligent collaborative method and system for island-type discrete manufacturing based on OAG ontology. Background Technology

[0002] Currently, discrete manufacturing industries such as automobile manufacturing are gradually transforming from traditional assembly line production models to island-based lean manufacturing models. Island-based production systems, exemplified by SAIC-GM-Wuling's world's first island-based lean intelligent manufacturing plant, consist of multiple independent unmanned island production units and possess core characteristics such as dynamic adjustment of production sequences, free combination of processes, mixed-model production, and highly flexible manufacturing. To achieve efficient operation of island-based production, existing technologies have attempted to introduce multi-agent scheduling technology, IoT sensing technology, and manufacturing execution systems to achieve localized collaborative control of unmanned island production units, automated guided vehicles, automated warehouses, and supply chain management systems. Some solutions also employ ontology modeling technology to semantically describe manufacturing resources, aiming to achieve cross-system data interoperability and command interaction.

[0003] However, the existing technological system still has the following core defects: First, in existing collaborative methods, production agents and logistics agents operate independently, lacking a global coordination and collaborative verification mechanism based on a unified semantic foundation. This easily leads to disconnects between production scheduling and logistics delivery, cross-island resource conflicts, and chaotic scheduling of multiple vehicle types on mixed lines. Second, existing decision-making schemes mostly rely on preset static rules, failing to integrate real-time data on unmanned island capacity load, logistics equipment operating status, and supply chain material supply dynamics. This results in a serious disconnect between decision-making results and actual on-site needs, and lacks a feedback data-driven closed-loop optimization mechanism, making it difficult to continuously improve collaborative efficiency. Third, the linkage between island-based manufacturing within the plant and the upstream supply chain is weak. Material inventory warnings, in-transit material tracking, and supplier replenishment instructions cannot accurately match the real-time material consumption rate of the unmanned island, easily causing material backlog at the line or production interruptions. Summary of the Invention

[0004] To address the technical deficiencies in the background technology, this invention proposes an intelligent collaborative method and system for discrete manufacturing island production based on OAG ontology, which solves the aforementioned technical problems and meets practical needs. The specific technical solution is as follows: The intelligent collaborative method for discrete manufacturing island production based on OAG ontology includes the following steps: Real-time collection of cross-scenario heterogeneous raw data across the entire island production process; establishment of a mapping relationship between the raw data and preset island production-specific OAG vertical ontology elements; generation of standardized mapping datasets. The standardized mapping dataset is input into the ontology projection rule engine, and the full data semantic transformation is completed according to the core components of the OAG vertical ontology, generating a unified OAG ontology semantic information database. Based on the OAG ontology semantic information database, global situation analysis is completed in combination with real-time production needs. With OAG vertical ontology as the unified rule driver, a four-layer multi-agent system is launched to complete collaborative verification and flexible collaborative decision-making, and standardized collaborative execution instructions are generated. The collaborative execution instructions are sent to the corresponding execution units to complete the autonomous execution of the entire chain, and the execution feedback data of the entire process is collected simultaneously to complete the closed-loop optimization.

[0005] Furthermore, the specific steps for generating the standardized mapping dataset are as follows: Through real-time message queue data pipelines and multi-source IoT sensing devices, three types of heterogeneous raw data are collected synchronously across the entire island production scenario. These three types of heterogeneous raw data include unmanned island production site data, end-to-end logistics data, and supply chain data. The preprocessing unit sequentially performs data cleaning, data deduplication, and data format unification conversion operations on three types of heterogeneous raw data, transforming unstructured data into structured data, performing structured parsing on semi-structured data, and generating standardized structured datasets. According to the preset entity element classification rules of the island-style production-specific OAG vertical ontology, each data field of the standardized structured dataset is matched to establish the mapping relationship between the original data fields and the OAG vertical ontology elements, and a standardized mapping dataset is generated.

[0006] Furthermore, the specific steps for generating a unified OAG ontology semantic information database are as follows: The ontology projection rule engine receives a standardized mapping dataset, retrieves the entity elements, business behaviors, and rule logic of the pre-defined island-style production-specific OAG vertical ontology, and loads the pre-defined standardized data and ontology semantic mapping rules. Based on the entity elements of the OAG ontology, the data corresponding to uninhabited islands, production resources, logistics units and supply chain nodes in the standardized mapping dataset are mapped to entity instances and entity associations in the ontology, generating a set of entity semantic instances. Based on OAG ontology business behavior, the corresponding action data of unmanned island production, logistics distribution and supply chain material pull in the standardized mapping dataset are mapped to business actions and process sequence in the ontology, generating a semantic instance set of business behavior. Based on the OAG ontology rule logic, the data corresponding to the uninhabited island production threshold, logistics and distribution trigger conditions, and supply chain inventory warning threshold in the standardized mapping dataset are mapped to structured rule instances in the ontology, generating a set of rule logic semantic instances. The entity semantic instance set, business behavior semantic instance set, and rule logic semantic instance set are integrated to generate a unified OAG ontology semantic information database.

[0007] Further, the specific steps for completing the global situation analysis are as follows: Receive the OAG ontology semantic information library and real-time production requirement data through the global situation analysis module, retrieve the island production industry knowledge graph, extract features from the input data, and generate production link feature datasets, logistics link feature datasets, and supply chain link feature datasets respectively; Based on the production link feature datasets, logistics link feature datasets, and supply chain link feature datasets, through the global production situation collaborative adaptation degree calculation formula, complete the quantitative calculation of the full-scenario collaborative adaptation degree, and generate the global collaborative adaptation degree quantitative result; Based on the global collaborative adaptation degree quantitative result, combined with the rule logic of the OAG ontology semantic information library, analyze the unmanned island production status, logistics system operation status, orders and production requirements, supply chain material supply status, and abnormal risks item by item, identify the core tasks, resource bottlenecks, abnormal risks, and collaborative requirements of the current island production, integrate the analysis results to generate a global production situation analysis report, and output it to the four-layer multi-agent system.

[0008] Further, the global production situation collaborative adaptation degree calculation formula is as follows: , where , is the global production situation collaborative adaptation degree, is the weight coefficient of the production link in the global collaborative adaptation degree calculation, n is the total number of unmanned island production units in the island production scenario, is the production status matching degree of the i-th unmanned island production unit, is the weight value of the i-th unmanned island production unit, is the weight coefficient of the logistics link in the global collaborative adaptation degree calculation, m is the total number of logistics execution units in the island production scenario, is the logistics status matching degree of the j-th logistics execution unit, is the weight value of the j-th logistics execution unit, is the weight coefficient of the supply chain link in the global collaborative adaptation degree calculation, p is the total number of supply chain nodes in the island production scenario, is the supply status matching degree of the k-th supply chain node, is the weight value of the k-th supply chain node.

[0009] Further, the four-layer multi-agent system includes a global decision-making agent layer, a production agent layer, a logistics agent layer, and a supply chain collaboration agent layer. The specific steps for generating a standardized collaborative execution instruction are as follows: The global decision-making intelligent agent layer receives the global production situation analysis report, retrieves the global rule logic of the OAG vertical ontology, sets the decision objectives and collaboration rules of each sub-domain intelligent agent, generates global decision instructions, and sends them to the corresponding sub-domain intelligent agents. The sub-domain intelligent agents include intelligent agents in the production intelligent agent layer, logistics intelligent agent layer, and supply chain collaboration intelligent agent layer. Each domain-specific intelligent agent receives global decision-making instructions, retrieves the corresponding domain-specific business rules from the OAG vertical ontology, combines them with the global production status analysis report, conducts independent domain-specific decisions, generates domain-specific decision-making schemes, and uploads them to the global decision-making intelligent agent layer. Based on the decision-making schemes of each domain, and combined with the global rule logic of the OAG vertical ontology, all domain decision-making schemes are collaboratively verified to detect whether there are resource conflicts, timing contradictions and rule inconsistencies. If there are problems, optimization and adjustment instructions are generated based on the OAG ontology rules and sent back to the corresponding domain intelligent agent to correct the scheme until the verification is passed. All domain-specific decision schemes that pass the verification are integrated to generate standardized collaborative execution instructions based on the unified semantics of the OAG ontology.

[0010] Furthermore, the production intelligent agent layer includes island-specific intelligent agents and inter-island collaborative intelligent agents, and the logistics intelligent agent layer includes receiving intelligent agents, warehousing intelligent agents, short-haul intelligent agents, and line-side-online intelligent agents. The specific steps for carrying out independent domain-specific decision-making are as follows: The production intelligent agent layer receives global decision-making instructions, completes the production sequence adjustment, process combination and equipment operation scheduling decisions of the island through the island-specific intelligent agent, and completes the resource allocation, cross-island process collaboration and capacity balance optimization decisions of each island through the inter-island collaborative intelligent agent, generates production domain decision-making schemes, and uploads them to the global decision-making intelligent agent layer; The logistics intelligent agent layer receives global decision-making instructions, completes unmanned receiving operation decisions through the receiving intelligent agent, completes inbound and outbound and inventory management operation decisions through the warehousing intelligent agent, completes unmanned short-haul operation decisions for materials within the factory area through the short-haul intelligent agent, and completes workstation-level automatic material calling and precise material delivery operation decisions through the line-side-online intelligent agent, generates logistics domain decision-making schemes, and uploads them to the global decision-making intelligent agent layer; The supply chain collaborative intelligent agent layer receives global decision-making instructions, completes real-time tracking and dynamic pull decisions for supplier inventory, regional distribution center warehouse inventory, and materials in transit, generates inventory warnings, supplier replenishment instructions, and supply chain domain decision-making schemes, and uploads them to the global decision-making intelligent agent layer.

[0011] Furthermore, the end-to-end autonomous execution is completed by the production and logistics execution layer and the supply chain collaboration layer. Both the production and logistics execution layer and the supply chain collaboration layer have pre-completed semantic mapping with the island-type production-specific OAG vertical ontology. After receiving standardized collaborative execution instructions, the two layers complete automated production and intelligent logistics operations and supply chain material linkage operations respectively according to the execution rules corresponding to the OAG vertical ontology. They simultaneously collect production execution data, logistics execution data and supply chain execution data throughout the process, integrate them to form an end-to-end execution feedback dataset and output it to the feedback optimization layer.

[0012] Furthermore, the closed-loop optimization is performed by the feedback optimization layer. The feedback optimization layer receives the full-link execution feedback dataset and completes standardized preprocessing. Based on the preprocessed data, it dynamically optimizes the OAG vertical ontology, adjusts the collaborative rules of the four-layer multi-agent system, and sends the optimization results back to the ontology projection rule engine, the global situation analysis module, and the four-layer multi-agent system, forming a continuous optimization closed loop of data perception, analysis decision-making, and execution feedback.

[0013] The discrete manufacturing island production intelligent collaborative system based on OAG ontology includes a memory, a host computer, and a computer program stored in the memory and executable on the host computer. The computer program is configured to implement the steps of the discrete manufacturing island production intelligent collaborative method based on OAG ontology as described above.

[0014] Compared with existing technologies, the intelligent collaborative method and system for discrete manufacturing island production based on OAG ontology provided by this invention has the following advantages: This invention constructs a unique OAG (On-Agent Generic Entity) ontology for island-based production, encompassing entity elements, business behaviors, and rule logic, as a unified semantic foundation. Combined with an ontology projection rule engine, heterogeneous data is transformed into a semantic information database in real time. A four-layer multi-agent system, incorporating a global collaborative adaptability calculation formula, is employed to achieve global coordination, domain-specific decision-making, collaborative verification, and instruction generation, significantly improving the accuracy and real-time performance of collaboration across unmanned islands, logistics equipment, and supply chain nodes. Simultaneously, through a feedback optimization layer that dynamically iterates the OAG ontology and multi-agent collaborative rules, a continuous optimization loop of data perception, analysis, decision-making, and execution feedback is formed. Furthermore, blockchain, visual recognition, and autonomous driving technologies are integrated to achieve material... The system enables end-to-end traceability of box-level materials, centimeter-level detection of non-standard material frames, and robust positioning of unmanned short-haul trucks within the factory area. This comprehensively solves the technical challenges of inconsistent semantics, unbalanced collaboration among intelligent agents, disconnect between decision-making and on-site operations, and weak supply chain linkage in existing technologies. It achieves independent operation of unmanned island production units and seamless collaboration between islands, precise matching of production and logistics at the workstation level and within minutes, and real-time dynamic pulling between in-plant manufacturing and the upstream supply chain. It can adapt to the dynamic adjustment of production sequences, mixed-line production of multiple vehicle models, and free combination of cross-island processes without human intervention, significantly reducing the risk of production interruption, material backlog rate, and scheduling response delay, and improving the overall efficiency, flexibility, and intelligence level of island-based lean intelligent manufacturing. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the intelligent collaborative method for discrete manufacturing island production based on OAG ontology in this invention. Detailed Implementation

[0016] In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "middle," and "inner," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, it should be noted that unless otherwise explicitly specified and limited, the terms "installed," "connected," and "joined" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention through specific circumstances.

[0017] The embodiments of the present invention will be described below with reference to the accompanying drawings and related examples. The embodiments of the present invention are not limited to the following examples, and the present invention relates to the relevant necessary components in this technical field, which should be regarded as well-known technology in this technical field and can be known and mastered by those skilled in this technical field.

[0018] See Figure 1 This invention provides an intelligent collaborative method for discrete manufacturing island production based on OAG ontology, comprising the following steps: Step S100: Collect cross-scenario heterogeneous raw data of the entire island production scenario in real time, establish the mapping relationship between the raw data and the preset island production-specific OAG vertical ontology elements, and generate a standardized mapping dataset. Specifically, the OAG vertical ontology is a domain ontology model specifically built for the discrete manufacturing island production scenario based on the OAG standard framework. It serves as the unified semantic and rule foundation for the end-to-end collaboration of this invention. Its core components encompass entity elements, business behaviors, and rule logic across the entire island production scenario, enabling semantic unification of heterogeneous data across systems and scenarios, fundamentally solving the information silo problem in island production. Island production is a new flexible production model in the discrete manufacturing industry, distinct from traditional rigid assembly lines. SAIC-GM-Wuling's island-based lean intelligent manufacturing plant is a typical example, consisting of multiple unmanned island production units with independent and complete production capabilities. It supports dynamic adjustment of production sequences, free combination of processes, and mixed-line production of multiple varieties. Its core advantage is its ability to quickly respond to the personalized customization needs of the end market. Cross-scenario heterogeneous raw data, in this invention, specifically refers to raw data from three different business scenarios: island production, unmanned island production, intelligent logistics, and supply chain. This data has inconsistent data structures, storage formats, and communication protocols, encompassing three core data forms: structured equipment operating parameters, semi-structured process route documents, and unstructured visual inspection images.

[0019] The island-based production-specific OAG vertical ontology of this invention is built on the OAG standard framework. First, it anchors the full-scenario business needs of island-based production in automotive discrete manufacturing, deeply deconstructs the core scenario characteristics of island-based production, including all elements of "people, machines, materials, methods, and environment," unmanned island production units, end-to-end intelligent logistics, and upstream supply chain collaboration, and then completes the standardized construction of three core levels in sequence, as follows: First, the core business entities, entity master data and attributes, and relationships between entities in the entire island production scenario are identified and defined, completing the construction of the entity element layer. Next, the entire business process of island production, from material warehousing to finished product outbound, is broken down, and core business actions, process sequences, and basic business logic are standardized, completing the construction of the business behavior layer. Then, island production industry standards, on-site operation specifications, and mature business experience are integrated and transformed into structured business rules that can be parsed by machines, completing the construction of the rule logic layer. Based on this, the instantiation mapping of the ontology and the configuration of standardized data-ontology semantic mapping rules are completed. After on-site business scenario adaptability verification and logical consistency verification, a deployable OAG vertical ontology library is formed. Simultaneously, a dynamic ontology optimization mechanism based on on-site execution feedback data is established to ensure that the ontology can continuously adapt to changes in the production scenario and always meet the actual operational needs of the island production site.

[0020] Step S200: Input the standardized mapping dataset into the ontology projection rule engine, complete the semantic transformation of the entire dataset according to the core components of the OAG vertical ontology, and generate a unified OAG ontology semantic information database. Specifically, the ontology projection rule engine is the core semantic transformation execution unit of this invention. It is a rule execution system that pre-defines standardized mapping rules between OAG vertical ontology and raw data. Its core function is to transform discrete standardized business data into semantic information that conforms to the OAG ontology specification, thereby achieving accurate mapping transformation from "raw data" to "ontology semantic instance".

[0021] Step S300: Based on the OAG ontology semantic information database, and combined with real-time production needs, complete the global situation analysis. With the OAG vertical ontology as the unified rule driver, start the four-layer multi-agent system to complete collaborative verification and flexible collaborative decision-making, and generate standardized collaborative execution instructions. Specifically, the four-layer multi-agent system is a distributed decision-making system specifically built for island-type production scenarios in this invention. It includes a global decision-making agent layer, a production agent layer, a logistics agent layer, and a supply chain collaboration agent layer. Each agent is based on the unified semantic foundation of the OAG ontology, realizing independent domain-specific decision-making and global collaborative verification. It is the core carrier for achieving flexible collaborative decision-making in this invention. Standardized collaborative execution instructions refer to a set of instructions generated after collaborative verification, using the unified semantic format of the OAG ontology. These instructions clearly define the operation content, execution sequence, resource configuration, and collaborative requirements of each execution unit (such as unmanned island production equipment, AGVs, automated warehouses, etc.), ensuring consistent understanding of instructions across different systems.

[0022] Step S400: Send the collaborative execution instruction to the corresponding execution unit to complete the autonomous execution of the entire chain, and collect the execution feedback data of the entire process simultaneously to complete the closed-loop optimization.

[0023] Specifically, closed-loop optimization in this invention refers to the full-process loop mechanism that uses the feedback data from the production, logistics, and supply chain execution stages to iteratively optimize the OAG vertical ontology and multi-agent collaborative rules. This mechanism enables continuous autonomous iterative upgrades of the system's collaborative capabilities and addresses the pain point of disconnect between traditional solution decisions and on-site data.

[0024] This invention uses a dedicated OAG (Ontology, Aspect-Oriented Logic) ontology for island-based production as a unified semantic and rule foundation throughout the entire process. Through four core steps—full-scenario heterogeneous data collection and standardized mapping, ontology-driven semantic transformation, multi-agent collaborative decision-making under global situational awareness, and end-to-end execution and closed-loop optimization—it constructs a complete intelligent collaborative system for the entire island-based production chain. This fundamentally solves the core pain points of existing island-based production models, such as inconsistent cross-scenario data semantics, severe information silos, disconnect between production and logistics, lack of global coordination among multiple agents, disconnect between decision-making and on-site dynamic data, and weak supply chain linkage. It achieves a balance between independent operation of unmanned island production units and seamless inter-island collaboration, precise matching of production and logistics at the workstation and minute levels, and real-time dynamic linkage between in-plant manufacturing and the upstream supply chain. It can adapt to flexible production needs such as dynamic adjustments to production sequences, mixed-line production of multiple vehicle models, and free combination of cross-island processes without human intervention, forming a complete autonomous optimization closed loop. This fully releases the flexibility and efficiency advantages of the island-based production model, significantly reduces the risk of production interruptions, material backlog rates, and manual scheduling costs, and greatly improves the overall operational efficiency, collaborative accuracy, and intelligence level of discrete manufacturing island-based production.

[0025] In one embodiment of the present invention, the specific steps for generating the standardized mapping dataset are as follows: Step S101: Synchronously collect three types of heterogeneous raw data from the entire island production scenario through real-time message queue data pipeline and multi-source IoT sensing devices. The three types of heterogeneous raw data include unmanned island production site data, full-link logistics data, and supply chain data. Specifically, the real-time message queue data pipeline, with a Kafka cluster as its core carrier, is a real-time data transmission middleware built specifically for high-concurrency, low-latency heterogeneous data transmission across the entire island production scenario. It supports synchronous acquisition, partitioned orderly storage, and reliable forwarding of data from multiple sources. It can be configured with millisecond-level acquisition frequencies according to production needs, solving the problems of asynchronous data transmission, high latency, and easy data loss across multiple scenarios and devices. It serves as the core transmission foundation for real-time data acquisition across the entire chain. Multi-source IoT sensing devices, in this invention, are the collective term for physical terminals used for data acquisition across the entire island production scenario and covering the entire business chain. These include UWB ultra-wideband positioning devices, RFID radio frequency identification devices, high-definition industrial vision cameras, GPS / BeiDou positioning modules, equipment operation status sensors, material inventory detection sensors, and PLC data acquisition modules, respectively deployed in unmanned island production units, factory logistics channels, intelligent automated warehouses, and supply chain nodes. They serve as the physical carriers for acquiring three types of heterogeneous raw data.

[0026] Specifically, a three-tiered data acquisition architecture is established, consisting of "edge acquisition terminal - factory transmission layer - central processing layer". The edge acquisition terminal is deployed in various unmanned islands, logistics equipment, automated warehouses, and supply chain nodes, using multi-source IoT sensing devices. The factory transmission layer adopts industrial Ethernet + 5G dual-link redundant transmission to ensure data stability. The central processing layer deploys a Kafka real-time message queue cluster and a preprocessing unit. The Kafka cluster creates independent Topic partitions for the three major business scenarios of unmanned island production, full-link logistics, and supply chain, assigns a unique producer ID to each acquisition device, and completes the initial configuration of the acquisition link.

[0027] After the system starts, each sensing device collects raw data at a preset frequency (up to milliseconds for unmanned island device operation data, seconds for logistics positioning data, and minutes for supply chain inventory data). The collected data is then encapsulated into standardized messages with timestamps, device IDs, and scene identifiers and synchronously reported to the corresponding Topic partition of the Kafka cluster. The Kafka cluster stores and forwards the multi-source data in an orderly manner according to time sequence, ensuring that the timelines of data from different scenarios and devices are fully aligned, thus solving the problem of misaligned collection times of multi-source data. At the same time, a multi-replica data backup mechanism is configured to avoid data loss due to network fluctuations or device offline, ensuring the integrity of all raw data.

[0028] The collected data strictly covers three categories of core heterogeneous raw data. The first category is production site data for unmanned islands, including PLC equipment operating parameters, process completion status, production progress, overall equipment efficiency, capacity load rate, production sequence adjustment instructions, and equipment fault alarm information for each unmanned island. The second category is end-to-end logistics data, including real-time positioning and operating status data of AGVs / AMRs, unmanned forklift operation data, automated warehouse inbound and outbound inventory data, material receiving / warehousing / distribution process data, remaining quantity data of materials in line-side warehouses, and RFID identification data of material bins. The third category is supply chain data, including upstream supplier material inventory data, off-site RDC warehouse inventory and inbound / outbound data, material transportation location and status data, supplier production scheduling and delivery plan data, and material quality inspection qualification data.

[0029] Step S102: The preprocessing unit sequentially performs data cleaning, data deduplication, and data format unification conversion operations on the three types of heterogeneous raw data to transform unstructured data into structured data, perform structured parsing on semi-structured data, and generate a standardized structured dataset. Specifically, the preprocessing unit synchronously pulls all raw data from the Kafka cluster and loads a pre-defined island-style production data cleaning rule base. The rule base clearly defines the legal value range, required fields, and format requirements for various types of data. First, it performs validation of required fields, removing invalid data due to missing key fields caused by device offline or network interruption. For non-critical missing fields, it uses the historical average of the same device and the same period and linear interpolation to complete them. Then, it uses the 3σ principle and box plot method to identify abnormal extreme values ​​caused by sensor interference and device false alarms, and removes or corrects abnormal data. Finally, it uses the moving average method to smooth the high-frequency noise data of device operating parameters, completing the data cleaning and outputting a valid dataset.

[0030] The preprocessing unit generates a unique business primary key for each cleaned data entry. The primary key is composed of "scene identifier + unique business number + timestamp". For example, the primary key for unmanned island production data is "unmanned island number + process number + production batch + timestamp", and the primary key for logistics data is "material batch number + AGV number + delivery order number + timestamp". Based on the unique business primary key, the entire dataset is traversed to identify redundant data corresponding to duplicate primary keys. Only the most recent valid data with the latest timestamp is retained, and the remaining duplicate data is deleted. The data deduplication is completed, and a simplified dataset is output.

[0031] The preprocessing unit performs a full format conversion on the deduplicated dataset according to the preset standardized data format specifications. For structured data, it unifies field names, data types, units of measurement, and numerical precision, and standardizes fields with the same meaning reported by different devices and systems. For unstructured data, it uses OCR text recognition, image feature extraction, and natural language processing technologies to convert visual inspection images and text descriptions of equipment faults into structured data with feature values ​​and key fields. For semi-structured data, it uses an XML / JSON parser to parse the process route documents and BOM lists by tags, extracts core business fields, and converts them into structured data. Finally, it generates a standardized structured dataset with completely unified format, fields, types, and precision, and outputs it to the mapping and matching unit.

[0032] Step S103: According to the preset entity element classification rules for producing exclusive OAG vertical ontology in an island-style production, match each data field of the standardized structured dataset, establish the mapping relationship between the original data fields and the OAG vertical ontology elements, and generate a standardized mapping dataset.

[0033] Specifically, the mapping and matching unit receives a standardized structured dataset, retrieves the preset classification rules for the entity elements of the OAG vertical category specific to island production, and clarifies the entity ID, attribute name, data type, association relationship, and matching rules of the four core entities of the OAG ontology (uninhabited island entity, production resource entity, logistics unit entity, and supply chain node entity) to complete the pre-mapping preparation.

[0034] The mapping and matching unit performs feature matching for each data field in the standardized structured dataset according to entity classification rules. First, it identifies the business scenario and entity type to which the data field belongs. Then, it maps the data field to the attributes of the corresponding ontology entity, clarifies the ontology entity attributes, data constraints, and relationships corresponding to each data field, establishes the mapping relationship between the original data field and OAG vertical ontology elements, and generates a unique ontology mapping identifier for each data field.

[0035] The mapping matching unit associates and encapsulates the standardized structured dataset with the corresponding ontology mapping identifier, ensuring that each piece of data has a clear ontology element affiliation, generating a standardized mapping dataset that fully conforms to the OAG ontology specification, and outputting it to the ontology projection rule engine, providing an accurate and compliant data foundation for the subsequent semantic transformation of the entire dataset.

[0036] In one embodiment of the present invention, the specific steps for generating a unified OAG ontology semantic information database are as follows: Step S201: Receive the standardized mapping dataset through the ontology projection rule engine, retrieve the entity elements, business behaviors and rule logic of the pre-set island-style production-specific OAG vertical ontology, and load the pre-set standardized data and ontology semantic mapping rules. Specifically, the ontology projection rule engine receives standardized mapping datasets through a standardized interface. First, it performs compliance checks on the datasets, including: whether the data fields have complete ontology mapping identifiers, whether the data format conforms to the attribute constraints of the OAG ontology, and whether the data value range is within the preset legal range. Abnormal data that fails the check is removed to ensure that the input data fully adapts to the ontology mapping requirements. Data sets that pass the check are stored in a temporary processing cache.

[0037] The ontology projection rule engine calls the pre-built island-style production-specific OAG vertical ontology library, fully loads the three core components of the ontology, namely entity elements, business behaviors, and rule logic, and synchronously loads the attribute definitions, relationships, and constraints corresponding to each component module to build the basic framework for ontology semantic mapping, ensuring that all subsequent mapping operations strictly follow the standardized definition of OAG vertical ontology.

[0038] The ontology projection rule engine loads the preset standardized data - ontology semantic mapping rules. These rules correspond one-to-one with the ontology element classification rules, clarifying the ontology mapping path, instantiation rules, and association binding logic for different business scenarios and data types. At the same time, based on the actual needs of the current production scenario (such as multi-model mixed-line production or emergency order insertion), the corresponding scenario-based mapping rule branches are loaded to complete the initial configuration of the ontology projection rule engine.

[0039] Step S202: Based on the entity elements of the OAG ontology, map the data corresponding to uninhabited islands, production resources, logistics units and supply chain nodes in the standardized mapping dataset to entity instances and entity association relationships in the ontology, and generate an entity semantic instance set. Specifically, the ontology projection rule engine decomposes the standardized mapping dataset into corresponding data subsets according to the four entity categories of the OAG ontology. These subsets are: uninhabited island entity dataset, production resource entity dataset, logistics unit entity dataset, and supply chain node entity dataset. Each dataset has a clear entity category mapping identifier to ensure that the data corresponds one-to-one with the entity category.

[0040] For each categorized dataset, the ontology projection rule engine generates a unique entity instance for each specific business data according to the attribute definition of the corresponding entity element, assigns a globally unique ontology instance ID to each instance, assigns the value of the data field to the corresponding attribute of the instance, and verifies whether the attribute value conforms to the data type and value range constraints defined by the ontology; for example, the "No. 1 welding unmanned island, number of equipment 12, design capacity 30 JPH, current load rate 75%" in the dataset is mapped to a unique instance under "unmanned island entity", which fully inherits all the attribute definitions of the unmanned island entity.

[0041] The ontology projection rule engine binds corresponding relationships to the generated entity instances according to the entity association rules defined in the OAG ontology, thus constructing a complete entity association network. For example, it binds the subordinate associations of "workstation instances" and "equipment instances" to the "unmanned island instance", the usage association of "workstation instances" and the supply association of "supplier instances" to the "material instance", and the delivery association of "automatic warehouse instance" and "workstation instance" to the "AGV instance", thereby realizing semantic associations between entities, rather than isolated instance objects.

[0042] Step S203: Based on the OAG ontology business behavior, map the corresponding action data of unmanned island production, logistics distribution and supply chain material pull in the standardized mapping dataset to business actions and process sequence in the ontology, and generate a set of business behavior semantic instances. Specifically, the ontology projection rule engine decomposes the standardized mapping dataset into corresponding dynamic action data subsets according to the four major business behavior categories of the OAG ontology. These subsets are: uninhabited island production behavior dataset, full-link logistics behavior dataset, supply chain collaboration behavior dataset, and global scheduling behavior dataset. Each dataset has a clear business behavior mapping identifier, timestamp, and execution entity ID.

[0043] For each business behavior dataset, the ontology projection rule engine decomposes the execution subject, triggering conditions, input parameters, and output results of the business action according to the definition of the corresponding business behavior, and maps the discrete action data to the standardized business action unit defined in the ontology. For example, the action data of "AGV05 delivers material batch B001 from the automated warehouse A to workstation No. 3 of the No. 2 assembly unmanned island" is mapped to the standardized business action unit "logistics scheduling - flexible online" in the ontology, which clarifies the execution subject, start and end nodes, operation objects, and timing requirements of the action.

[0044] The ontology projection rule engine, based on the business process sequence rules defined in the OAG ontology, links standardized business action units together according to business logic and time sequence, encapsulating them into complete business behavior instances. It clarifies the preconditions, subsequent actions, time constraints, and abnormal branches of the business process. For example, a series of logistics actions, such as "material receiving - warehousing - storage - short-haul - online delivery", are linked together in time sequence into a business behavior instance of "end-to-end material delivery", fully inheriting the process rules and time constraints of logistics behavior in the ontology.

[0045] The ontology projection rule engine integrates all generated business behavior instances and stores them in a structured manner according to business scenarios, execution subjects, and process types. At the same time, it binds a corresponding entity instance to each business behavior instance to realize the semantic association between dynamic behavior and static entity, generate a complete set of business behavior semantic instances, and outputs them to the intermediate processing module of the ontology projection rule engine.

[0046] Step S204: Based on the OAG ontology rule logic, map the data corresponding to the uninhabited island production threshold, logistics and distribution triggering conditions, and supply chain inventory early warning threshold in the standardized mapping dataset to structured rule instances in the ontology, and generate a set of rule logic semantic instances. Specifically, the ontology projection rule engine decomposes the standardized mapping dataset into corresponding rule class data subsets according to the four rule types of the OAG ontology. These subsets are: uninhabited island production rule dataset, logistics collaboration rule dataset, supply chain linkage rule dataset, and anomaly handling rule dataset. Each dataset contains clear rule thresholds, judgment conditions, and execution actions.

[0047] For each rule dataset, the ontology projection rule engine decomposes the rule's triggering conditions, judgment formulas, threshold parameters, execution actions, and constraint range according to the definition of the corresponding rule logic, transforming the unstructured rule description into a machine-parsable structured logical expression. For example, the rule "trigger delivery when the remaining quantity of materials at the line is less than 50%" is parsed into a structured logical expression "trigger the material delivery instruction for the corresponding workstation when the remaining quantity of materials at the line / the full quantity of materials < 0.5", which fully conforms to the rule definition specifications of the OAG ontology.

[0048] The ontology projection rule engine generates a unique rule instance for each structured rule logic, assigns a globally unique rule ID, clarifies the applicable scenarios, associated entities, and associated business behaviors of the rule, and binds the rule instance to the corresponding entity instance and business behavior instance; for example, the "automatic material requisition rule instance" is bound to the corresponding workstation entity instance and the line-side delivery business behavior instance, realizing the semantic constraints of the rule on business objects and business processes.

[0049] The ontology projection rule engine integrates all generated rule instances, stores them in a structured manner according to rule type and applicable scenario, verifies the logical consistency between rules to avoid rule conflicts, generates a complete set of rule logical semantic instances, and outputs them to the ontology projection rule engine intermediate processing module.

[0050] Step S205: Integrate the entity semantic instance set, business behavior semantic instance set, and rule logic semantic instance set to generate a unified OAG ontology semantic information database.

[0051] Specifically, the ontology projection rule engine fully integrates the entity semantic instance set, business behavior semantic instance set, and rule logic semantic instance set. With entity instances as the core anchor, it deeply binds business behavior instances and rule logic instances with their corresponding entity instances, constructing a complete semantic network of "static entity-dynamic behavior-constraint rule". This ensures that the relationship between all semantic instances fully conforms to the definition of OAG vertical ontology, with no logical conflicts or broken relationships.

[0052] The ontology projection rule engine will integrate the complete semantic network and store it in a structured manner according to the standard architecture of OAG ontology, building a unified OAG ontology semantic information database. At the same time, it will establish a multi-dimensional indexing system to support fast retrieval and retrieval according to entity type, business scenario, rule type, and time dimension, ensuring that subsequent global situation analysis and multi-agent decision-making can retrieve the corresponding semantic information in real time and accurately.

[0053] The ontology projection rule engine generates an OAG ontology semantic information database and outputs it synchronously to the global situation analysis module and the four-layer multi-agent system. At the same time, it establishes a real-time incremental update mechanism. When a new standardized mapping dataset is input, the semantic information database can be incrementally updated in real time to ensure that the database is always synchronized with the actual production on site.

[0054] In one embodiment of the present invention, the specific steps for completing the global situation analysis are as follows: Step S301: Receive the OAG ontology semantic information database and real-time production demand data through the global situation analysis module, retrieve the island production industry knowledge graph, extract features from the input data, and generate production link feature datasets, logistics link feature datasets and supply chain link feature datasets respectively. Specifically, the global situation analysis module is the core analysis unit developed by this invention for collaborative management and control of the entire island production scenario. It is the core hub connecting the OAG ontology semantic information database and the multi-agent decision-making layer. It has built-in standardized feature engineering rules, collaborative adaptation quantification models, multi-dimensional situation analysis logic and industry knowledge graph calling interfaces. It can deeply integrate static ontology semantic information with dynamic real-time production needs to complete the accurate judgment of the entire production situation. It provides a unique and standardized quantitative basis and analysis support for subsequent four-layer multi-agent collaborative decision-making. Unlike the situation analysis module of general manufacturing scenarios, it can be fully adapted to the flexible characteristics of island production, such as multiple varieties, small batches and dynamic adjustments. The island manufacturing industry knowledge graph is a structured knowledge system built on industry standards, process specifications, mature management experience, and best historical cases of island lean manufacturing in automotive discrete manufacturing. It is stored in a triple structure of "entity-relationship-attribute". The core knowledge includes island manufacturing process routes, multi-model mixed-line production management specifications, intelligent logistics scheduling rules, supply chain collaboration standards, equipment anomaly handling experience, bottleneck optimization solutions, etc. It serves as the industry experience foundation for global situation analysis and can effectively improve the industry adaptability, judgment accuracy, and anomaly identification foresight of situation analysis.

[0055] The global situation analysis module receives OAG ontology semantic information database and real-time production demand data from the production planning system (including order delivery dates, priorities, order insertion instructions, user personalized customization requirements, multi-model mixed-line production plans, etc.) through a standardized semantic interface. First, it performs compliance verification on the received data, including: whether the semantic information conforms to the OAG ontology specification, whether the real-time production demand data has complete order identifiers and timestamps, and whether there are any missing or abnormal extreme values ​​in the data. Invalid data that fails the verification is removed to ensure the integrity and compliance of the input data.

[0056] The global situation analysis module retrieves the pre-built knowledge graph of the island-type production industry, loads industry standards, process specifications, anomaly judgment rules, and bottleneck identification experience related to situation analysis; at the same time, it loads the preset feature engineering rules bound to OAG ontology elements, clarifies the core feature fields, data standardization methods, and feature selection logic of the three major links, and completes the pre-configuration of feature extraction.

[0057] The global situational analysis module decomposes the OAG ontology semantic information database according to three major business scenarios: production, logistics, and supply chain. Based on feature engineering rules, it performs full feature extraction: For semantic information related to production on the unmanned island, it extracts core features such as capacity load, production progress, and equipment status to generate a production link feature dataset; for semantic information related to the entire logistics chain, it extracts core features such as equipment operating status, inventory level, and delivery on-time rate to generate a logistics link feature dataset; for semantic information related to supply chain collaboration, it extracts core features such as supplier inventory, in-transit material status, and delivery cycle to generate a supply chain link feature dataset. During the feature extraction process, all feature fields are subjected to min-max standardization, mapping all feature values ​​uniformly to the 0-1 range to ensure dimensional consistency in subsequent quantitative calculations.

[0058] The global situation analysis module performs consistency verification on the generated feature datasets of the three stages, verifying the binding relationship between feature fields and OAG ontology elements, the range of feature values, and the time axis alignment of the datasets, to ensure that the three datasets are synchronized in time and semantically unified. After the verification is passed, the feature datasets of the three stages are output to the quantization calculation unit to complete the entire feature extraction process.

[0059] Step S302: Based on the production process feature dataset, logistics process feature dataset, and supply chain process feature dataset, the global production situation collaboration adaptation degree calculation formula is used to complete the quantitative calculation of the full-scenario collaboration adaptation degree and generate the global collaboration adaptation degree quantification result. Specifically, the quantification unit classifies the quantification results of collaborative adaptation based on preset grading rules: 0.8-1 is an excellent collaborative state, 0.6-0.8 is a good collaborative state, 0.4-0.6 is a general collaborative state, and below 0.4 is a poor collaborative state. At the same time, it generates the sub-item collaborative adaptation results of the three major stages, clarifying the degree of influence of each stage on the global collaborative state. Finally, the global collaborative adaptation quantification results, sub-item results, and grading labels are synchronously output to the multi-dimensional situation analysis unit.

[0060] Step S303: Based on the global collaborative adaptation quantification results and combined with the rule logic of the OAG ontology semantic information database, analyze the production status, logistics system operation status, order and production demand, supply chain material supply status, and abnormal risks of the unmanned island item by item. Identify the core tasks, resource bottlenecks, abnormal risks, and collaborative needs of the current island production, integrate the analysis results to generate a global production situation analysis report, and output it to the four-layer multi-agent system.

[0061] Specifically, the multi-dimensional situational analysis unit receives the quantified results of collaborative adaptation and, combined with the rule logic of the OAG ontology semantic information database and industry knowledge graph, conducts item-by-item analysis across five dimensions: First, unmanned island production status analysis, clarifying the capacity surplus / overload status, production schedule deviation, equipment health status, and process execution status of each unmanned island; second, logistics system operation status analysis, clarifying the load status of logistics equipment, inventory health, delivery on-time rate, line-side material support capability, and path congestion risk; third, order and production demand analysis, clarifying the priority, delivery cycle, process requirements, production schedule matching degree of each order, as well as the dynamic demand for order insertion and mixed-line production; fourth, supply chain material supply status analysis, clarifying the inventory level, in-transit status, and supply guarantee capability of each material, and predicting the risk of material shortage; fifth, anomaly risk analysis, based on the anomaly handling rules of the OAG ontology, identifying potential anomaly risks such as equipment failure, material shortage, logistics congestion, and delivery delay, and predicting the degree of impact of anomalies on overall production.

[0062] Based on multi-dimensional item-by-item analysis results and combined with the quantification results of collaborative adaptation, the multi-dimensional situational analysis unit accurately identifies the core bottlenecks of current island production (such as overloaded unmanned island capacity, congested logistics paths, and shortages of core materials), high-priority abnormal risks, and core collaborative needs such as cross-island resource allocation, logistics and distribution adjustments, and supply chain replenishment required to meet real-time production demands. At the same time, it combines historical best cases in the industry knowledge graph to match preliminary optimization directions for the identified bottlenecks and risks, providing a reference for subsequent multi-agent decision-making.

[0063] The multi-dimensional situation analysis unit integrates the results of collaborative adaptation quantification, multi-dimensional analysis, bottleneck and risk identification, and collaborative demand assessment. Based on the unified semantics of the OAG ontology, it generates a standardized global production situation analysis report. The report adopts a structured format, clearly marking the ontology entities, business rules, and timestamps corresponding to each analysis content, ensuring that it can be directly parsed and invoked by the four-layer multi-agent system without semantic bias.

[0064] The multi-dimensional situation analysis unit generates a global production situation analysis report, which is then output to the global decision-making agent layer of the four-layer multi-agent system through a standardized collaborative interface, serving as the core input for collaborative decision-making. At the same time, a real-time incremental update mechanism for the report is established. When the OAG ontology semantic information database and real-time production demand data are updated, the recalculation of the situation analysis and the incremental update of the report can be triggered in real time, ensuring that the analysis results are always synchronized with the actual production on site.

[0065] It should be noted that the formula for calculating the global production situation coordination and adaptability is as follows: , in, , The overall production situation coordination and adaptability is used to comprehensively and quantitatively evaluate the degree of coordination and matching, resource carrying capacity and order demand response capability of the three core links of production, logistics and supply chain in the entire island production scenario. It is the core judgment basis for subsequent four-layer multi-agent to formulate collaborative decision-making schemes.

[0066] This is the weighting coefficient of the production process in the global collaborative adaptability calculation, representing the importance ratio of the production process in the global collaborative assessment. During normal steady-state production, it is assigned a default value of 0.4, the highest weight among the three processes, aligning with the "production-centric" business logic of discrete manufacturing. For urgent orders or multi-model mixed-line production, it can be increased to 0.5-0.6 to prioritize the rational allocation of production resources on the unmanned island. During equipment maintenance or low-load production, it can be decreased to 0.2-0.3, focusing on logistics and supply chain inventory optimization. The basic fixed value is pre-set in the rule logic module of the OAG vertical ontology and directly retrieved from the OAG ontology semantic information library. The global decision-making agent automatically adjusts it based on real-time production needs and order priorities, with the adjustment rules pre-set in the OAG ontology rule library.

[0067] n is the total number of unmanned island production units in the island production scenario. It is assigned a positive integer value, which is exactly the same as the number of unmanned islands actually deployed in the factory. For example, if a factory deploys 8 unmanned islands, then n=8. The production status matching degree of the i-th unmanned island production unit is used to quantitatively evaluate the degree of matching between the production capacity, operating status and current production demand of a single unmanned island. It has a fixed value range of [0,1], with higher values ​​indicating better matching between the unmanned island and production demand. It is calculated by weighting three core sub-indicators, using the following formula: , in, The capacity utilization rate matching degree of the uninhabited island is set to [0,1]. The optimal load range is 60%-80%, and the value is higher the closer it is to this range. This represents the production progress achievement rate of the uninhabited island, with a value of [0,1]. It is the ratio of actual progress to planned progress, and is set to 1 if it exceeds 100%. The equipment availability rate of the uninhabited island is defined as [0,1], which is the ratio of the number of normally operating devices to the total number of devices.

[0068] The weight value of the i-th unmanned island production unit represents the importance of the unmanned island in the overall production process. The fixed value range is [0,1]. Unmanned islands in core processes have higher weights, while those in auxiliary processes have lower weights. The weights of the unmanned islands are assigned according to their process priority in the vehicle production process. For example, unmanned islands in final assembly are assigned a weight of 0.9-1.0, unmanned islands in stamping are assigned a weight of 0.7-0.8, and unmanned islands in auxiliary processes are assigned a weight of 0.3-0.6. The weight values ​​of all unmanned islands on the same production line must be strongly correlated with the processes in the production process, and there is no fixed total constraint.

[0069] This is the weighting coefficient of the logistics link in the global collaborative adaptability calculation, representing the importance ratio of the end-to-end intelligent logistics link in the global collaborative evaluation. The default value is 0.35 during normal steady-state production, matching the core requirement of "deep linkage between logistics and production" in island production. When large quantities of materials arrive in a concentrated manner or production is changed, it can be increased to 0.4-0.5 to prioritize logistics path optimization and accurate material delivery. When low-capacity intermittent production is carried out, it can be decreased to 0.2-0.25 to focus on logistics cost optimization. The basic fixed value is pre-set in the rule logic module of the OAG vertical ontology and directly retrieved from the semantic information database of the OAG ontology. The logistics intelligent agent layer proposes an adjustment request based on logistics load and delivery needs, and the adjustment is completed after verification by the global decision-making intelligent agent.

[0070] m represents the total number of logistics execution units in the island-style production scenario. It is assigned a positive integer value and is counted by logistics functional units, not by individual equipment. For example, the receiving unit, the automated warehouse storage unit, the factory short-haul unit, and the line-side delivery unit are each one execution unit, totaling m=4.

[0071] The logistics status matching degree of the j-th logistics execution unit is used to quantitatively evaluate the degree of matching between the operating status, delivery capacity, and production rhythm of a single logistics unit and the unmanned island. It has a fixed value range of [0,1], with higher values ​​indicating better matching between the logistics unit and the production rhythm. It is calculated by weighting three core sub-indicators, using the following formula: , in, This represents the on-time delivery rate of the logistics unit, with a value of [0,1], which is the ratio of the number of on-time delivered orders to the total number of delivered orders. The equipment availability rate of this logistics unit, with a value of [0,1], is the ratio of the number of normally operating devices to the total number of devices. This is the inventory turnover rate of the logistics unit, with a value of [0,1]. It is calculated as the ratio of actual turnover rate to optimal turnover rate, and is set to 1 if it exceeds 100%.

[0072] The weight value is the j-th logistics execution unit, representing the importance of the logistics unit in the whole-chain logistics system. It is assigned according to the priority of the unmanned island served by the logistics unit. For example, the line-side delivery unit directly serves the final assembly unmanned island and is assigned a value of 0.9-1.0; the factory short-haul unit is assigned a value of 0.7-0.8; and the receiving and storage unit is assigned a value of 0.6-0.7. The weight value is strongly tied to the importance of the unmanned island served and there is no fixed total constraint.

[0073] This is the weighting coefficient of the supply chain link in the global collaboration adaptability calculation, representing the importance ratio of the upstream supply chain collaboration link in the global collaboration assessment. The default value is 0.25 during normal steady-state production, matching the business logic of "in-plant production as the core and external supply as the guarantee" of the OEM supply chain. For new model mass production or key material switching, it can be adjusted to 0.3-0.4 to prioritize the stability of supply chain material supply. During steady-state mass production with sufficient materials, it can be adjusted to 0.15-0.2 to focus on supply chain inventory cost optimization. The basic fixed value is pre-set in the rule logic module of the OAG vertical ontology and directly retrieved from the OAG ontology semantic information database. The supply chain collaboration intelligent agent layer proposes an adjustment request based on the material supply status and replenishment needs, and the adjustment is completed after verification by the global decision-making intelligent agent.

[0074] p represents the total number of supply chain nodes in the island-style production scenario. It is assigned a positive integer value and counted as independent nodes based on the first-tier suppliers of core materials and off-site RDC warehouses. For example, if there are 5 core component suppliers and 2 off-site RDC warehouses, the total p = 7.

[0075] The supply status matching degree of the k-th supply chain node is used to quantitatively evaluate the degree of matching between the material supply capacity, delivery capacity and the on-site production needs of a single supply chain node. It has a fixed value range of [0,1], with higher values ​​indicating better matching between the material supply and on-site production needs of that supply chain node. It is calculated by weighting three core sub-indicators, using the following formula: , in, This represents the on-time delivery rate of materials at this supply chain node, with a value of [0,1], which is the ratio of on-time delivery batches to total delivery batches. This represents the material inventory adequacy ratio of this supply chain node, with a value of [0,1]. It is the ratio of actual inventory to safety stock required for production, and is set to 1 if it exceeds 100%. This represents the on-time delivery rate of materials in transit at this supply chain node, with a value of [0,1], calculated as the ratio of expected on-time delivery batches to total on-time delivery batches.

[0076] The weight value of the k-th supply chain node represents the importance of the materials supplied by that supply chain node to the factory's production. The weight is assigned according to the importance level of the supplied materials. For example, power system and body core component suppliers are assigned a weight of 0.9-1.0, interior component suppliers are assigned a weight of 0.6-0.8, and general standard component suppliers are assigned a weight of 0.3-0.5. The weight value is strongly tied to the production importance of the supplied materials and there is no fixed total constraint.

[0077] The above formula is a quantitative evaluation model for end-to-end collaborative capabilities specifically designed for island-style production scenarios. It addresses the core pain points of existing island-style production technologies, such as "collaboration status relying on manual experience judgment, lack of unified quantitative standards, inability to accurately assess cross-stage matching degree, and disconnect between decision-making and on-site data." Its core value lies in: The standardized mathematical model transforms the operational status of the three heterogeneous links of production, logistics, and supply chain into a single quantitative indicator that can be directly recognized by machines and used for automatic decision-making by multiple agents. It takes into account the dual requirements of "independent operation of unmanned islands" and "global collaboration across the entire chain" in island-type production. It not only realizes the differentiated evaluation of different units within a single link, but also completes the global quantification of the collaborative capabilities across the entire scenario. All input data comes from the OAG ontology semantic information database, and the entire calculation is based on a unified semantic foundation, which completely avoids the distortion of calculation results caused by cross-system data semantic deviation.

[0078] In one embodiment of the present invention, the four-layer multi-agent system includes a global decision-making agent layer, a production agent layer, a logistics agent layer, and a supply chain collaboration agent layer. The specific steps for generating standardized collaborative execution instructions are as follows: Step S304: Receive the global production situation analysis report through the global decision-making intelligent agent layer, retrieve the global rule logic of the OAG vertical ontology, set the decision objectives and collaboration rules of each domain intelligent agent, generate global decision instructions, and send them to the corresponding domain intelligent agents. The domain intelligent agents include intelligent agents in the production intelligent agent layer, logistics intelligent agent layer, and supply chain collaboration intelligent agent layer. Specifically, the four-layer multi-agent system is a distributed collaborative decision-making architecture specifically built for the scenario characteristics of island-style production with multiple varieties, small batches, and high flexibility. It is the core decision-making carrier of the entire collaborative method and is divided into four layers: global decision-making agent layer, production agent layer, logistics agent layer, and supply chain collaborative agent layer. Unlike the isolated multi-agent system in general industrial scenarios, all layers of this system operate on the unified semantic and rule foundation of OAG vertical ontology. It supports independent distributed decision-making of agents in each domain and realizes cross-domain global collaboration, which completely solves the core pain point of imbalance between independent operation of agents and global collaboration in existing technologies.

[0079] The global decision-making intelligent agent layer receives global production status analysis reports through standardized semantic interfaces, synchronously retrieves global rule logic from the OAG vertical ontology library, including full-link resource allocation rules, cross-domain collaborative timing rules, global rules for anomaly handling, and order priority control rules, and loads the optimal scheduling cases from the island production industry knowledge graph to complete pre-decision preparation.

[0080] The global decision-making intelligent agent layer, based on the core tasks, resource bottlenecks, abnormal risks, and collaborative needs in the global production situation analysis report, and combined with the priority and delivery requirements of real-time production orders, sets the core objectives for global decision-making, including the objectives of ensuring order delivery, balancing the capacity of unmanned islands, accurately adapting logistics and distribution, and ensuring continuous supply of materials in the supply chain. At the same time, based on the global rules of the OAG ontology, it sets hard constraint rules for cross-domain collaboration, including the rule that production resources on unmanned islands cannot conflict, the rule that logistics and distribution timing is strongly matched with production progress, and the rule that supply chain replenishment and line-side material consumption are strongly linked, thus clarifying the decision-making boundaries of each sub-domain intelligent agent.

[0081] The global decision-making agent layer sets the decision-making objectives, collaborative rules, constraints, and priority requirements, and encapsulates them in a standardized manner based on the unified semantics of the OAG ontology. It binds the corresponding ontology rule instance ID to each constraint in the instruction, generating standardized global decision-making instructions to ensure that each domain agent has a completely consistent understanding of the instructions without semantic deviation.

[0082] The global decision-making intelligent agent layer uses the OAG ontology standardized collaborative interface to distribute global decision-making instructions to the corresponding domain intelligent agents, including the production intelligent agent layer, logistics intelligent agent layer, and supply chain collaborative intelligent agent layer. At the same time, it distributes a global production status analysis report, records the instruction distribution timestamp and receipt receipt, and ensures that the instructions are delivered completely and on time, providing a unified input basis for subsequent domain decision-making.

[0083] Step S305: Each domain-specific intelligent agent receives the global decision-making instruction, retrieves the exclusive business rules of the corresponding domain in the OAG vertical ontology, combines the global production status analysis report, carries out independent domain-specific decision-making, generates a domain-specific decision-making scheme, and uploads it to the global decision-making intelligent agent layer. Specifically, each domain-specific intelligent agent receives global decision-making instructions and global production status analysis reports, completes instruction compliance verification, and confirms the semantic integrity and rule consistency of the instructions. After the verification is passed, it retrieves the exclusive business rules of the corresponding domain in the OAG vertical ontology. For example, the production intelligent agent layer retrieves the unmanned island production rules and cross-island process collaboration rules, the logistics intelligent agent layer retrieves the logistics distribution rules and path optimization rules, and the supply chain collaboration intelligent agent layer retrieves the material pull rules and inventory early warning rules to complete the pre-configuration of domain-specific decisions.

[0084] Each domain-specific intelligent agent takes the global decision-making objective as its core constraint and, in conjunction with the domain's operational status, bottlenecks, and risks in the global production situation analysis report, makes independent decisions based on the OAG ontology's domain-specific rules: the production intelligent agent layer completes decisions on adjusting the production sequence of the unmanned island, combining processes, allocating resources across islands, and optimizing capacity balance; the logistics intelligent agent layer completes decisions on the entire logistics operation process, including material receiving, warehousing management, short-distance transportation within the factory, and workstation-level delivery; and the supply chain collaboration intelligent agent layer completes collaborative decisions on tracking off-site materials, dynamic pull, inventory warning, and supplier replenishment. The entire decision-making process follows the global collaboration rules and does not exceed the decision boundaries.

[0085] Each domain-specific agent organizes the decision results into a standardized domain-specific decision scheme. The scheme clearly defines the list of execution actions, resource requirements, execution sequence, key nodes, and contingency plans for handling anomalies in its domain. At the same time, each item in the scheme is bound to the corresponding OAG ontology instance ID and rule identifier to complete semantic encapsulation and ensure that the scheme can be directly parsed and verified by the global decision-making agent layer.

[0086] Each domain-specific agent uploads its generated domain-specific decision-making scheme to the global decision-making agent layer through the OAG ontology standardized collaboration interface. Simultaneously, it uploads the timestamp of the scheme generation, the rule basis, and the verification instructions. After receiving all the domain-specific schemes, the global decision-making agent layer completes the scheme integrity verification, providing complete input data for subsequent collaborative verification.

[0087] Step S306: Based on each domain decision scheme, and combined with the global rule logic of the OAG vertical ontology, perform collaborative verification on all domain decision schemes to detect whether there are resource conflicts, timing contradictions and rule inconsistencies. If there are problems, generate optimization and adjustment instructions based on the OAG ontology rules and send them back to the corresponding domain agent for scheme correction until the verification is passed. Specifically, the global decision-making intelligent agent layer loads the global collaborative verification rules of the OAG vertical ontology, preprocesses all received domain-specific decision schemes, decomposes the resource requirements, timing arrangements, execution actions, and rule basis in the schemes, extracts the core verification elements, and establishes the resource-timing correlation matrix of the entire scheme, providing a data foundation for collaborative verification.

[0088] The global decision-making intelligent agent layer, based on the global rules of the OAG ontology, performs item-by-item verification on all domain-specific decision-making schemes in three dimensions: First, resource conflict verification, detecting whether multiple domain-specific schemes simultaneously occupy the same production resources, logistics equipment, or material inventory, such as two orders simultaneously applying for the same unmanned island's capacity at the same time, or two delivery tasks simultaneously occupying the same AGV equipment; Second, timing contradiction verification, detecting whether the execution timing of the domain-specific schemes matches, such as the logistics delivery timing not matching the unmanned island's production progress, or the supply chain replenishment timing not matching the line-side material consumption timing; Third, rule non-compliance verification, detecting whether the domain-specific schemes comply with the global rules of the OAG ontology, whether they break the constraints of the global decision-making instructions, and whether they violate industry production standards.

[0089] The global decision-making agent layer summarizes the verification results. If all solutions pass the verification, it directly enters the solution integration stage. If conflicts, contradictions, or rule inconsistencies are detected, it accurately locates the domain agent to which the problem belongs, the cause of the problem, and the corresponding rule basis. Based on the global rules of the OAG ontology, it generates targeted optimization and adjustment instructions, clarifying the optimization direction, constraints, and correction time limit.

[0090] The global decision-making agent layer sends optimization and adjustment instructions back to the corresponding domain agents. The domain agents then revise their domain-specific decision schemes based on these instructions, regenerate compliant schemes, and upload them again. The global decision-making agent layer performs a second verification on the revised schemes, repeating the verification-revision process until all domain-specific decision schemes pass collaborative verification, ensuring that the final schemes have no cross-domain conflicts, timing inconsistencies, or rule violations. If a scheme fails collaborative verification within the preset iteration limit or time window, the global decision-making agent layer is triggered to change the decision objective or generate an alert for manual collaborative handling.

[0091] Step S307: Integrate all the domain decision schemes that have passed the verification to generate standardized collaborative execution instructions based on the unified semantics of the OAG ontology.

[0092] Specifically, the global decision-making intelligent agent layer integrates all verified domain decision-making schemes. First, based on the timeline of the global production plan, it aligns the timing of the three domain schemes of production, logistics, and supply chain to ensure that the timing of the actions in each link is fully matched. At the same time, it coordinates the allocation of resources across the entire chain, clarifies the resource configuration, usage time, and collaboration requirements of each execution unit, and avoids the problem of uneven resource allocation.

[0093] The global decision-making intelligent agent layer standardizes and encapsulates the entire solution content after time-series alignment and resource matching based on the unified semantics of OAG vertical ontology. It binds the corresponding ontology entity instance, business behavior rules and execution standards to each execution action, generating the basic framework of end-to-end collaborative execution instructions, ensuring that there are no semantic deviations or ambiguities in the transmission of instructions throughout the entire production, logistics and supply chain process.

[0094] The global decision-making intelligent agent layer standardizes and breaks down the full-link execution instructions according to the execution units, and generates corresponding targeted execution sub-instructions for each execution unit. The sub-instructions clearly define the operation content, execution sequence, acceptance criteria, and anomaly reporting path of the unit, while retaining the semantic association with the full-link instructions, ensuring that the independent execution of each unit is completely consistent with global collaboration.

[0095] The global decision-making intelligent agent layer generates the final standardized collaborative execution instructions, which are then distributed to the corresponding execution units in the production and logistics execution layer and the supply chain collaboration layer through standardized interfaces. At the same time, the instructions are fully and synchronously stored in the OAG ontology semantic information database as a benchmark for subsequent execution feedback and closed-loop optimization, thus completing the entire collaborative decision-making process.

[0096] It should be noted that the production intelligent agent layer includes island-specific intelligent agents and inter-island collaborative intelligent agents, and the logistics intelligent agent layer includes receiving intelligent agents, warehousing intelligent agents, short-haul intelligent agents, and line-side-to-line intelligent agents. The specific steps for carrying out independent domain-specific decision-making are as follows: Step S3051: The production intelligent agent layer receives global decision-making instructions, completes the production sequence adjustment, process combination and equipment operation scheduling decisions of the island through the island-specific intelligent agent, completes the resource allocation, cross-island process collaboration and capacity balance optimization decisions of each island through the inter-island collaborative intelligent agent, generates production domain decision-making schemes, and uploads them to the global decision-making intelligent agent layer. Specifically, the production intelligence layer receives global decision-making instructions and global production status analysis reports from the global decision-making intelligence layer through the OAG ontology standardized collaboration interface, completes instruction compliance verification, and confirms the semantic integrity, rule consistency, and decision boundary clarity of the instructions. After the verification is passed, it retrieves the corresponding production domain-specific rules from the OAG vertical ontology library, including unmanned island production rules, cross-island process collaboration rules, capacity balancing optimization rules, and equipment scheduling specifications. At the same time, it loads the attributes, relationships, and constraints of the corresponding unmanned island entities to complete the pre-configuration of domain-specific decision-making.

[0097] Each uninhabited island has its own dedicated intelligent agent, which takes the global decision-making objective as the core constraint and combines the corresponding uninhabited island's capacity load, production progress, equipment status, and process requirements from the global production situation analysis report to carry out autonomous decision-making for the island: Based on order priority and delivery cycle, it completes the dynamic adjustment of the island's production sequence and the decision of free combination of processes; Based on the island's equipment operating status and capacity load, it completes the decision-making of equipment operation scheduling, process cycle optimization, and emergency switching in case of failure; The decision-making process follows the production rules of the OAG ontology throughout, does not exceed the boundaries of the global decision-making instructions, and generates single-island execution sub-plans.

[0098] The inter-island collaborative agent receives single-island execution sub-plans generated by the dedicated agents of all unmanned islands. Combining these with the overall production situation analysis report on the plant's overall capacity bottlenecks, resource gaps, and cross-island process requirements, it conducts cross-island collaborative decisions: Based on the capacity load of each unmanned island, it dynamically allocates production orders and resources among multiple unmanned islands to achieve balanced optimization of the plant's overall capacity; based on the process requirements of cross-island process flow, it completes the timing matching, material flow, and resource collaboration decisions for cross-island processes, adapting to the flexible requirements of multi-model mixed-line production and free combination of cross-island processes; and it optimizes the cross-island consistency of each single-island execution sub-plan to generate cross-island collaborative sub-plans.

[0099] The production intelligence layer integrates all single-island execution sub-schemes and cross-island collaborative sub-schemes to achieve time-series alignment and resource matching throughout the entire production process, ensuring that there are no resource conflicts or time-series contradictions within a single island or between islands. The integrated scheme is then standardized and encapsulated based on the unified semantics of the OAG vertical ontology, binding each execution action in the scheme with the corresponding ontology entity instance ID, business rule identifier, and constraint conditions, generating a standardized production domain decision scheme, ensuring that the scheme can be directly parsed and verified by the global decision intelligence layer.

[0100] The production intelligence layer uploads the generated production domain decision-making schemes to the global decision-making intelligence layer through the OAG ontology standardized collaboration interface, and simultaneously uploads the timestamp of the scheme generation, the rule basis, and the single-island and cross-island decision details, thus completing the entire process of production domain decision-making.

[0101] Step S3052: The logistics intelligent agent layer receives global decision instructions, completes unmanned receiving operation decisions through the receiving intelligent agent, completes inbound and outbound and inventory management operation decisions for the material bin-level automated warehouse through the warehousing intelligent agent, completes unmanned short-haul operation decisions for materials within the factory area through the short-haul intelligent agent, and completes workstation-level automatic material calling and precise material delivery operation decisions through the line-side-online intelligent agent, generates a logistics domain decision scheme, and uploads it to the global decision intelligent agent layer; Specifically, the logistics intelligent agent layer receives global decision-making instructions and global production status analysis reports from the global decision-making intelligent agent layer through the OAG ontology standardized collaborative interface. It simultaneously obtains the production rhythm, process sequence, and material requirements in the production domain decision-making scheme. After completing the compliance verification of the instruction and demand data, it retrieves the corresponding logistics domain-specific rules from the OAG vertical ontology library, including receiving management rules, warehousing control rules, short-haul dispatch rules, and line-side delivery rules. At the same time, it loads the attributes, relationships, and constraints of the corresponding logistics unit entities to complete the pre-configuration of domain decision-making.

[0102] The receiving intelligent agent completes the entire receiving process decision-making based on the supply chain incoming material plan and production material requirements: Based on the fusion rules of OAG ontology and visual recognition technology, it generates execution instructions for automatic barcode scanning and identification, visual quality inspection, and missed scan verification of supplier incoming materials; Based on the attributes of incoming materials and the urgency of production needs, it generates decision-making schemes for warehouse reservation and warehouse location allocation; It clarifies the execution sequence, equipment scheduling, and verification standards of receiving operations, and generates receiving operation sub-schemes to achieve unmanned and automated decision-making throughout the entire incoming material receiving process.

[0103] Based on receiving operation sub-plans and production material distribution needs, the warehouse intelligence agent completes the full-process control decisions of the bin-level automated warehouse: based on the material inbound and outbound requirements and first-in-first-out rules, it generates execution instructions for automated warehouse inbound and outbound operations and bin scheduling; based on material attributes and turnover frequency, it completes decisions on warehouse location optimization allocation, dynamic inventory counting, and inventory health management; it clarifies the execution sequence of warehousing operations, equipment scheduling, and inventory control requirements, generates warehousing operation sub-plans, and achieves refined bin-level control of materials.

[0104] The short-haul intelligent agent completes the entire process decision-making for short-haul material transportation within the factory area based on warehousing operation sub-schemes and line-side material requirements: Based on the fusion rules of OAG (On-Action Vehicle) and autonomous driving technology, it generates execution instructions for short-haul route planning, multi-vehicle collaborative scheduling, and dynamic obstacle avoidance; Based on the production rhythm and material demand sequence, it completes the timing matching and capacity allocation decisions for short-haul operations, ensuring that material transportation is fully aligned with the warehousing and production rhythm; It clarifies the execution sequence, vehicle scheduling, and route requirements of short-haul operations, generating short-haul operation sub-schemes to solve the challenges of short-haul positioning and scheduling in dynamic occlusion and open environments within the factory area.

[0105] The line-side intelligent agent, based on the production domain decision-making scheme, the workstation production rhythm, and the remaining material quantity at the line side, completes the entire process decision-making for material delivery: based on the integration rules of the OAG body and automatic material calling technology, it generates automatic material calling trigger rules and delivery sequence matching schemes for unmanned / manned positions; based on the workstation material demand and line-side inventory level, it completes the decision-making for AGV delivery task issuance, path planning, and precise docking, achieving precise adaptation of material delivery and workstation production rhythm at the workstation level and minute level; it clarifies the execution sequence, equipment scheduling, and delivery requirements of the delivery operation, and generates line-side delivery operation sub-schemes.

[0106] The logistics intelligent agent layer integrates four major operational sub-solutions: receiving, warehousing, short-haul transportation, and line-side deployment. Based on the production sequence of the production domain decision-making scheme, it completes the timing alignment and rhythm matching of the entire logistics chain, ensuring that logistics operations are completely synchronized with production needs. The integrated scheme is standardized and encapsulated based on the unified semantics of the OAG vertical ontology, binding each logistics operation to the corresponding ontology entity instance ID, business rule identifier, and production demand association relationship, generating a standardized logistics domain decision-making scheme. Through the OAG ontology standardized collaborative interface, the scheme is uploaded to the global decision-making intelligent agent layer, simultaneously uploading the timestamp of scheme generation, rule basis, and full-chain operation details, completing the entire process of logistics domain decision-making.

[0107] Step S3053: The supply chain collaborative intelligent agent layer receives global decision-making instructions, completes real-time tracking and dynamic pull decisions for supplier inventory, regional distribution center warehouse inventory, and materials in transit, generates inventory warnings, supplier replenishment instructions, and supply chain domain decision-making schemes, and uploads them to the global decision-making intelligent agent layer.

[0108] Specifically, the supply chain collaborative intelligent agent layer receives global decision-making instructions and global production status analysis reports issued by the global decision-making intelligent agent layer through the OAG ontology standardized collaborative interface. Simultaneously, it obtains the material consumption plan for the production domain decision-making scheme and the inventory requirements for the logistics domain decision-making scheme. After completing the compliance verification of the instruction and demand data, it retrieves the corresponding supply chain domain-specific rules from the OAG vertical ontology library, including material pull rules, inventory warning rules, and supplier collaboration rules. At the same time, it loads the attributes, relationships, and constraints of the corresponding supply chain node entities to complete the pre-configuration of domain-specific decisions.

[0109] The supply chain collaborative intelligent agent layer is based on the supply chain node rules of the OAG ontology. It connects with the upstream supplier management system, the off-site RDC warehouse management system, and the in-transit logistics control system. It collects full status data of supplier inventory, off-site RDC warehouse inventory, and in-transit materials in real time, completes the standardized preprocessing and semantic mapping of data, and generates a full-chain material status ledger of the supply chain to accurately grasp the full life cycle status of all production materials in terms of inventory, in-transit, and delivery.

[0110] The supply chain collaborative intelligent agent layer conducts supply chain collaborative decision-making based on the factory's production material consumption plan, line-side inventory levels, and end-to-end material status ledger: Based on the factory's production rhythm and material consumption rate, it generates dynamic pull instructions for supplier materials, realizing real-time linkage between factory demand and supplier production; Based on safety stock rules, it identifies material inventory shortage risks, generates inventory warning information and supplier replenishment instructions, and clarifies the replenishment quantity, delivery cycle, and arrival sequence; Based on the status of materials in transit, it completes the matching and optimization of arrival sequence with factory production needs, avoiding the risks of material backlog or shortage.

[0111] The supply chain collaborative intelligent agent layer integrates the end-to-end material tracking scheme, material dynamic pull instructions, inventory early warning rules, and supplier replenishment instructions to achieve time-series alignment with the factory's production and logistics plans. The integrated scheme is standardized and encapsulated based on the unified semantics of the OAG vertical ontology, and each execution action is bound to the corresponding ontology entity instance ID, business rule identifier, and factory demand association relationship, generating a standardized supply chain domain decision-making scheme to ensure that the scheme can be directly parsed and verified by the global decision-making intelligent agent layer.

[0112] The supply chain collaborative intelligent agent layer uploads the generated supply chain domain decision-making schemes to the global decision-making intelligent agent layer through the OAG ontology standardized collaborative interface, and simultaneously uploads the timestamp of the scheme generation, the rule basis, and the full-link material status details, thus completing the entire process of supply chain domain decision-making.

[0113] It should be noted that the autonomous execution of the entire chain is completed by the production and logistics execution layer and the supply chain collaboration layer. Both the production and logistics execution layer and the supply chain collaboration layer have completed semantic mapping with the island-type production-specific OAG vertical ontology in advance. After receiving standardized collaborative execution instructions, the two layers complete automated production and intelligent logistics operations and supply chain material linkage operations respectively according to the execution rules corresponding to the OAG vertical ontology. They simultaneously collect production execution data, logistics execution data and supply chain execution data of the entire process, integrate them to form a full-chain execution feedback dataset and output it to the feedback optimization layer.

[0114] Specifically, the production and logistics execution layer and the supply chain collaboration layer complete initialization during system startup. All equipment, systems, materials, and workstations undergo semantic mapping of entities, attributes, and rules with the island-based production's dedicated OAG vertical ontology, ensuring that all execution units possess a unified command understanding capability. The two execution layers receive standardized collaborative execution commands from the four-layer multi-agent system, automatically parsing the command content based on the OAG ontology's semantics. This includes all execution information such as production sequences, process arrangements, equipment scheduling, material delivery timing, delivery quantities, supply chain replenishment requirements, and collaboration timing. The unmanned island production units in the production and logistics execution layer automatically complete automated production operations such as process combination, production sequence adjustment, equipment start-up and shutdown, and processing and assembly according to the command requirements, strictly adhering to the production rules and process constraints in the OAG ontology. The logistics unit completes the entire logistics process, including material receiving, warehouse scheduling, AGV delivery, line-side material replenishment, and empty container recycling, ensuring that the logistics rhythm matches and is precisely synchronized with the unmanned island's production pace in real time. The supply chain collaboration layer automatically triggers supply chain-driven operations such as supplier replenishment, RDC warehouse outbound shipments, in-transit material tracking, and arrival confirmation based on in-plant production consumption and material demand, achieving real-time linkage between in-plant demand and external supply. Throughout the entire production, logistics, and supply chain operation process, real-time production execution data, logistics execution data, and supply chain execution data are collected, including equipment status, production progress, material consumption, delivery results, inventory changes, and supplier response status. The collected multi-source execution data is cleaned, aligned, and structured, integrated into a unified end-to-end execution feedback dataset, and output to the feedback optimization layer, providing a data foundation for subsequent closed-loop iterations.

[0115] It should be noted that the closed-loop optimization is performed by the feedback optimization layer. The feedback optimization layer receives the full-link execution feedback dataset and completes standardized preprocessing. Based on the preprocessed data, it dynamically optimizes the OAG vertical ontology, adjusts the collaborative rules of the four-layer multi-agent system, and sends the optimization results back to the ontology projection rule engine, the global situation analysis module, and the four-layer multi-agent system, forming a continuous optimization closed loop of data perception, analysis decision-making, and execution feedback.

[0116] Specifically, the feedback optimization layer receives end-to-end execution feedback datasets from the production and logistics execution layer and the supply chain collaboration layer. It cleans, deduplicates, standardizes the format, and removes outliers to generate a standardized preprocessed dataset suitable for optimization analysis. Based on this standardized dataset, it statistically evaluates indicators such as production achievement rate, logistics delivery accuracy, supply chain response timeliness, anomaly rate, resource utilization, and collaboration conflict frequency, generating execution performance evaluation results. According to the execution performance and actual business data, it automatically updates the entity relationships, business process sequences, rule logic thresholds, and parameter configurations of the OAG vertical ontology, ensuring consistency between the ontology semantic model and actual on-site conditions. Based on the execution data and evaluation results, it optimizes the decision rules, collaboration interface protocols, conflict verification logic, resource allocation strategies, and timing matching mechanisms of the four-layer multi-agent system, improving agent collaboration accuracy. The optimized OAG ontology rules are then fed back to the ontology projection rule engine, the optimized analysis logic is fed back to the global situation analysis module, and the optimized collaboration strategy is fed back to the four-layer multi-agent system, completing the end-to-end update. The updated models and rules are directly used for the next round of data collection, semantic transformation, global analysis, intelligent decision-making and autonomous execution, enabling the system to continuously improve its collaborative accuracy, response speed and adaptability during continuous operation, forming a complete, autonomous and sustainable evolutionary optimization loop.

[0117] The present invention also provides an intelligent collaborative system for discrete manufacturing island production based on OAG ontology, including a memory, a host computer, and a computer program stored in the memory and executable on the host computer. The computer program is configured to implement the steps of the intelligent collaborative method for discrete manufacturing island production based on OAG ontology as described above.

[0118] This invention constructs an intelligent collaborative method for discrete manufacturing island production based on OAG vertical ontology, incorporating the entire production, logistics, and supply chain links into a unified semantic framework. This enables dynamic optimization and closed-loop iteration of multi-agent collaborative rules, effectively solving technical challenges in traditional discrete manufacturing such as lagging production scheduling, inaccurate logistics and distribution, slow supply chain response, frequent collaborative conflicts, and rigid system rules that hinder adaptive upgrades. It significantly improves the collaborative efficiency, scheduling accuracy, supply chain response speed, and system self-evolution capability of the entire discrete manufacturing process, greatly reducing the cost of manual intervention, the incidence of collaborative conflicts, and the risk of production anomalies. This enables the entire discrete manufacturing island production system to possess core capabilities such as autonomous data perception, intelligent analysis and decision-making, precise autonomous execution, and continuous closed-loop optimization, comprehensively enhancing the intelligence, precision, efficiency, and long-term stable operation of discrete manufacturing.

[0119] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent collaborative production in discrete island manufacturing based on OAG ontology, characterized in that, Includes the following steps: Real-time collection of cross-scenario heterogeneous raw data across the entire island production process; establishment of a mapping relationship between the raw data and preset island production-specific OAG vertical ontology elements; generation of standardized mapping datasets. The standardized mapping dataset is input into the ontology projection rule engine, and the full data semantic transformation is completed according to the core components of the OAG vertical ontology, generating a unified OAG ontology semantic information database. Based on the OAG ontology semantic information database, global situation analysis is completed in combination with real-time production needs. With OAG vertical ontology as the unified rule driver, a four-layer multi-agent system is launched to complete collaborative verification and flexible collaborative decision-making, and standardized collaborative execution instructions are generated. The collaborative execution instructions are sent to the corresponding execution units to complete the autonomous execution of the entire chain, and the execution feedback data of the entire process is collected simultaneously to complete the closed-loop optimization.

2. The intelligent collaborative method for discrete manufacturing island production based on OAG ontology according to claim 1, characterized in that, The specific steps for generating the standardized mapping dataset are as follows: Through real-time message queue data pipelines and multi-source IoT sensing devices, three types of heterogeneous raw data are collected synchronously across the entire island production scenario. These three types of heterogeneous raw data include unmanned island production site data, end-to-end logistics data, and supply chain data. The preprocessing unit sequentially performs data cleaning, data deduplication, and data format unification conversion operations on three types of heterogeneous raw data, transforming unstructured data into structured data, performing structured parsing on semi-structured data, and generating standardized structured datasets. According to the preset entity element classification rules of the island-style production-specific OAG vertical ontology, each data field of the standardized structured dataset is matched to establish the mapping relationship between the original data fields and the OAG vertical ontology elements, and a standardized mapping dataset is generated.

3. The intelligent collaborative method for discrete manufacturing island production based on OAG ontology according to claim 1, characterized in that, The specific steps for generating a unified OAG ontology semantic information database are as follows: The ontology projection rule engine receives a standardized mapping dataset, retrieves the entity elements, business behaviors, and rule logic of the pre-defined island-style production-specific OAG vertical ontology, and loads the pre-defined standardized data and ontology semantic mapping rules. Based on the entity elements of the OAG ontology, the data corresponding to uninhabited islands, production resources, logistics units and supply chain nodes in the standardized mapping dataset are mapped to entity instances and entity associations in the ontology, generating a set of entity semantic instances. Based on OAG ontology business behavior, the corresponding action data of unmanned island production, logistics distribution and supply chain material pull in the standardized mapping dataset are mapped to business actions and process sequence in the ontology, generating a semantic instance set of business behavior. Based on the OAG ontology rule logic, the data corresponding to the uninhabited island production threshold, logistics and distribution trigger conditions, and supply chain inventory warning threshold in the standardized mapping dataset are mapped to structured rule instances in the ontology, generating a set of rule logic semantic instances. The entity semantic instance set, business behavior semantic instance set, and rule logic semantic instance set are integrated to generate a unified OAG ontology semantic information database.

4. The intelligent collaborative method for discrete manufacturing island production based on OAG ontology according to claim 1, characterized in that, The specific steps for completing the global situation analysis are as follows: The global situation analysis module receives the OAG ontology semantic information database and real-time production demand data, retrieves the island production industry knowledge graph, extracts features from the input data, and generates production link feature datasets, logistics link feature datasets, and supply chain link feature datasets respectively. Based on the feature datasets of the production process, logistics process, and supply chain process, the global production situation collaboration adaptation degree calculation formula is used to complete the quantitative calculation of the collaboration adaptation degree of the entire scenario and generate the global collaboration adaptation degree quantification result. Based on the quantification results of global collaborative adaptation, and combined with the rule logic of the OAG ontology semantic information database, the production status, logistics system operation status, order and production demand, supply chain material supply status, and abnormal risks of the unmanned island are analyzed item by item. The core tasks, resource bottlenecks, abnormal risks and collaborative needs of the current island production are identified, and the analysis results are integrated to generate a global production situation analysis report, which is then output to the four-layer multi-agent system.

5. The intelligent collaborative method for discrete manufacturing island production based on OAG ontology according to claim 4, characterized in that, The formula for calculating the global production situation coordination and adaptability is as follows: ， Among them, , is the global production situation collaborative adaptation degree, is the weight coefficient of the production link in the calculation of the global collaborative adaptation degree, n is the total number of unmanned island production units in the island production scenario, is the production status matching degree of the i-th unmanned island production unit, is the weight value of the i-th unmanned island production unit, is the weight coefficient of the logistics link in the calculation of the global collaborative adaptation degree, m is the total number of logistics execution units in the island production scenario, is the logistics status matching degree of the j-th logistics execution unit, is the weight value of the j-th logistics execution unit, is the weight coefficient of the supply chain link in the calculation of the global collaborative adaptation degree, p is the total number of supply chain nodes in the island production scenario, is the supply status matching degree of the k-th supply chain node, is the weight value of the k-th supply chain node.

6. The intelligent collaborative method for discrete manufacturing island production based on OAG ontology according to claim 4, characterized in that, The four-layer multi-agent system includes a global decision-making agent layer, a production agent layer, a logistics agent layer, and a supply chain collaboration agent layer. The specific steps for generating standardized collaborative execution instructions are as follows: The global decision-making intelligent agent layer receives the global production situation analysis report, retrieves the global rule logic of the OAG vertical ontology, sets the decision objectives and collaboration rules of each sub-domain intelligent agent, generates global decision instructions, and sends them to the corresponding sub-domain intelligent agents. The sub-domain intelligent agents include intelligent agents in the production intelligent agent layer, logistics intelligent agent layer, and supply chain collaboration intelligent agent layer. Each domain-specific intelligent agent receives global decision-making instructions, retrieves the corresponding domain-specific business rules from the OAG vertical ontology, combines them with the global production status analysis report, conducts independent domain-specific decisions, generates domain-specific decision-making schemes, and uploads them to the global decision-making intelligent agent layer. Based on the decision-making schemes of each domain, and combined with the global rule logic of the OAG vertical ontology, all domain decision-making schemes are collaboratively verified to detect whether there are resource conflicts, timing contradictions and rule inconsistencies. If there are problems, optimization and adjustment instructions are generated based on the OAG ontology rules and sent back to the corresponding domain intelligent agent to correct the scheme until the verification is passed. All domain-specific decision schemes that pass the verification are integrated to generate standardized collaborative execution instructions based on the unified semantics of the OAG ontology.

7. The intelligent collaborative method for discrete manufacturing island production based on OAG ontology according to claim 6, characterized in that, The production intelligence layer includes island-specific intelligence agents and inter-island collaborative intelligence agents. The logistics intelligence layer includes receiving intelligence agents, warehousing intelligence agents, short-haul intelligence agents, and line-side-to-line intelligence agents. The specific steps for conducting independent domain-specific decision-making are as follows: The production intelligent agent layer receives global decision-making instructions, completes the production sequence adjustment, process combination and equipment operation scheduling decisions of the island through the island-specific intelligent agent, and completes the resource allocation, cross-island process collaboration and capacity balance optimization decisions of each island through the inter-island collaborative intelligent agent, generates production domain decision-making schemes, and uploads them to the global decision-making intelligent agent layer; The logistics intelligent agent layer receives global decision-making instructions, completes unmanned receiving operation decisions through the receiving intelligent agent, completes inbound and outbound and inventory management operation decisions through the warehousing intelligent agent, completes unmanned short-haul operation decisions for materials within the factory area through the short-haul intelligent agent, and completes workstation-level automatic material calling and precise material delivery operation decisions through the line-side-online intelligent agent, generates logistics domain decision-making schemes, and uploads them to the global decision-making intelligent agent layer; The supply chain collaborative intelligent agent layer receives global decision-making instructions, completes real-time tracking and dynamic pull decisions for supplier inventory, regional distribution center warehouse inventory, and materials in transit, generates inventory warnings, supplier replenishment instructions, and supply chain domain decision-making schemes, and uploads them to the global decision-making intelligent agent layer.

8. The intelligent collaborative method for discrete manufacturing island production based on OAG ontology according to claim 1, characterized in that, The end-to-end autonomous execution is completed by the production and logistics execution layer and the supply chain collaboration layer. Both the production and logistics execution layer and the supply chain collaboration layer have pre-completed semantic mapping with the island-type production-specific OAG vertical ontology. After receiving standardized collaborative execution instructions, the two layers complete automated production and intelligent logistics operations and supply chain material linkage operations respectively according to the execution rules corresponding to the OAG vertical ontology. They simultaneously collect production execution data, logistics execution data and supply chain execution data throughout the process, integrate them to form an end-to-end execution feedback dataset and output it to the feedback optimization layer.

9. The intelligent collaborative method for discrete manufacturing island production based on OAG ontology according to claim 8, characterized in that, The closed-loop optimization is performed by the feedback optimization layer. The feedback optimization layer receives the full-link execution feedback dataset and completes standardized preprocessing. Based on the preprocessed data, it dynamically optimizes the OAG vertical ontology, adjusts the collaborative rules of the four-layer multi-agent system, and sends the optimization results back to the ontology projection rule engine, the global situation analysis module, and the four-layer multi-agent system, forming a continuous optimization closed loop of data perception, analysis decision-making, and execution feedback.

10. A discrete manufacturing island-type intelligent collaborative system based on OAG ontology, characterized in that, It includes a memory, a host computer, and a computer program stored in the memory and executable on the host computer, the computer program being configured to implement the steps of the intelligent collaborative method for discrete manufacturing island production based on OAG ontology as described in any one of claims 1 to 9.

Images