Digital factory full-process collaboration method and system

Abstract

The invention relates to the technical field of digital factories, in particular to a full-process collaboration method and system for a digital factory. The method comprises the following steps: obtaining production decision data, and carrying out process parameter analysis according to the production decision data to obtain process parameter data; according to the process parameter data, building a factory resource agent through a preset digital factory model to obtain factory resource agent data; performing multi-agent negotiation according to the factory resource agent data to obtain digital workshop task allocation data; and carrying out industrial Internet of Things monitoring according to the digital workshop task allocation data to obtain industrial Internet of Things real-time data so as to carry out digital factory whole-process collaborative auxiliary operation. By constructing a task allocation mechanism based on semantic comprehension and multi-agent cooperation, intelligent matching and flexible scheduling among various resources in a digital factory are realized, and the task execution efficiency and the resource utilization rate of the system are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital factories, and in particular to a full-process collaboration method and system for a digital factory. Background Art

[0002] A digital factory is a new production organization method based on the relevant data of the entire product life cycle. According to the principle of virtual manufacturing, in a virtual environment, the entire production process is planned, simulated, optimized, and reorganized. With the rapid development of industrial manufacturing towards intelligence and digitization, the traditional production scheduling and resource allocation methods based on static rules and manual decision-making are difficult to meet the actual needs of flexible manufacturing, customized production, and multi-variety and small-batch tasks. Existing digital factory management systems often adopt preset task priorities or linear scheduling algorithms. In the face of dynamic environments such as changes in resource status, task conflicts, and multi-objective optimization, they lack sufficient real-time response and collaborative decision-making capabilities, easily resulting in low resource utilization, frequent task delays, and a decline in the overall scheduling efficiency of the system. Summary of the Invention

[0003] To solve the above technical problems, the present invention proposes a full-process collaboration method and system for a digital factory to solve at least one of the above technical problems.

[0004] The present application provides a full-process collaboration method for a digital factory, including the following steps:

[0005] Step S1: Obtain production decision data, and parse process parameters according to the production decision data to obtain process parameter data;

[0006] Step S2: Construct factory resource agents through a preset digital factory model according to the process parameter data to obtain factory resource agent data;

[0007] Step S3: Conduct multi-agent negotiation according to the factory resource agent data to obtain digital workshop task allocation data;

[0008] Step S4: Conduct industrial Internet of Things monitoring according to the digital workshop task allocation data to obtain real-time industrial Internet of Things data for full-process collaborative auxiliary operations of the digital factory.

[0009] In the present invention, based on the production requirements of the MES system, ERP system or user-defined input, elements such as process routes, material requirements, equipment capabilities and time constraints are analyzed to form a structured production task model. The key resources in the factory (such as CNC machine tools, industrial robots, AGV transportation units and personnel, etc.) are abstracted into heterogeneous intelligent agents with sensing, reasoning and communication capabilities. Each intelligent agent is embedded with a parameter modeling module, a real-time status monitoring module and a scheduling interface module to realize the dynamic modeling and task response capabilities of resources. The intelligent agents can conduct multiple rounds of negotiation on task priorities, resource conflicts, path avoidance, etc. to achieve self-organizing and adaptive task collaborative scheduling. The system collects the task execution status, equipment operation parameters and environmental anomaly data in real time through industrial sensors, edge computing units and PLC controllers deployed on site, and combines data-driven models or rule engines for anomaly warning and task adjustment to realize the closed-loop control process from "task generation → distribution → execution feedback → dynamic adjustment". It realizes the full-process closed-loop collaboration from task generation to task execution in a digital factory. Compared with the traditional method that relies on centralized scheduling or rule-based allocation, constructing resource intelligent agents based on process parameters improves the dynamic matching ability between tasks and resources; using the multi-agent negotiation mechanism significantly improves the adaptive ability of the scheduling system in the face of task conflicts, resource limitations or multi-objective optimization situations; combining industrial IoT monitoring to achieve virtual-real mapping and dynamic feedback enhances the execution reliability and real-time response ability of the system. Overall, this method has higher flexibility, intelligence and system stability in a changing production environment.

[0010] Preferably, step S1 is specifically as follows:

[0011] Obtain production decision data;

[0012] Extract the process flow from the production decision data to obtain process flow data;

[0013] Analyze the process parameters according to the process flow data to obtain process parameter data, where the process parameter data includes the required equipment type, process sequence, material usage, man-hour estimation, energy demand and environmental constraints.

[0014] In the present invention, by structuring the production decision data, the transformation from a macro production plan to executable process parameters is realized. Obtaining a standardized process path through process flow extraction ensures the process continuity and controllability of task execution; introducing the extraction of multi-dimensional information such as equipment type, material usage, process sequence, man-hour estimation, etc. in the analysis process not only improves the accuracy of process modeling, but also provides a data basis for resource matching and task collaboration. Compared with the traditional manual analysis or template matching method, the present invention has stronger versatility, automation degree and engineering adaptability, and significantly improves the accuracy of process parameter generation and the early response efficiency of the scheduling system.

[0015] Preferably, step S2 is specifically as follows:

[0016] Based on the process parameter data, perform intelligent instantiation of factory resource type agents through a preset digital factory model to obtain factory resource type agent data, where the factory resource type agent data includes production line agent data, equipment agent data, personnel agent data, and material agent data;

[0017] Perform attribute configuration on the factory resource type agent data to obtain agent attribute data, where the attribute configuration includes status information configuration, willingness function configuration, and policy space configuration;

[0018] Based on the process parameter data, construct a task collaboration relationship graph for the agent attribute data to obtain factory resource agent data.

[0019] In the present invention, by combining the process parameter data with a preset digital factory model, the transformation of factory resources from static information to agent modeling is completed. At the resource type level, multiple types of agents such as production lines, equipment, personnel, and materials are respectively instantiated, realizing structured modeling of resource management; at the attribute configuration level, status information, willingness functions, and policy spaces are introduced, enabling each type of agent to have the ability of autonomous perception and response, breaking the static rule mode of traditional resource scheduling; meanwhile, based on the process parameters, a graph of the task collaboration relationship between agents is constructed to form a resource network with collaboration logic, providing a data basis for task game and scheduling optimization. Compared with traditional BOM-driven or linear scheduling models, the present invention significantly enhances the system's expression ability for resource heterogeneity, dynamics, and strategy, and has good intelligent scheduling adaptability and scalability.

[0020] Preferably, step S3 is specifically as follows:

[0021] Perform negotiation initialization based on the factory resource agent data to obtain preliminary negotiation data;

[0022] Perform task game based on the preliminary negotiation data to obtain negotiation game data;

[0023] Perform initial local optimal extraction based on the negotiation game data to obtain negotiation local optimal data;

[0024] Perform global resource conflict detection based on the negotiation local optimal data to obtain global resource conflict data;

[0025] Perform annotation limitation on the negotiation game data based on the global resource conflict data to obtain negotiation annotation agreement data;

[0026] Perform restricted local optimal extraction based on the negotiation annotation agreement data to obtain digital workshop task allocation data.

[0027] In the present invention, through the introduction of a multi-stage and multi-strategy intelligent negotiation and task game mechanism, the flexible task allocation optimization for complex resource environments is realized. Negotiation initialization ensures that various resource agents have negotiation intentions and behavioral strategies; through the multi-agent task game mechanism, candidate allocation schemes are dynamically generated, breaking through the limitations of traditional static scheduling rules; then, through the combination of local optimal extraction and global conflict detection, intelligent decoupling is achieved under the conditions of task complexity and resource conflict diversity; through the annotation limitation and restricted optimization extraction of negotiation game data, the system realizes optimal resource utilization while ensuring executability. Compared with single scheduling or heuristic allocation methods, this step has significant self-adaptability, distributed intelligence, and conflict tolerance capabilities, and can more effectively support the task scheduling requirements of digital factories under multi-objective and dynamic loads.

[0028] Preferably, the negotiation initialization is specifically as follows:

[0029] According to the factory resource agent data, an agent semantic graph mapping is performed to obtain cross-agent shared semantic data;

[0030] Perform a willingness function semantic embedding on the cross-agent shared semantic data to obtain intelligent task context data;

[0031] According to the factory resource agent data, a digital twin mirror loading is performed to obtain agent digital mirror data;

[0032] Associate and integrate the intelligent task context data and the agent digital mirror data to obtain preliminary negotiation data.

[0033] In the present invention, by integrating semantic graph mapping, willingness function embedding, and digital twin mirror technology, an intelligent negotiation foundation with semantic understanding ability and virtual-real synchronization ability is constructed. Using agent semantic graph mapping to achieve term consistency and cognitive alignment among multiple types of resource agents, the coordination obstacle between heterogeneous data is solved; through the willingness function semantic embedding of shared semantic data, an intelligent task context with context awareness ability is constructed, making the response of resource agents to tasks more personalized and strategic; loading the digital twin mirror of resource agents provides a visual and predictable virtual environment for strategy deduction and behavior simulation; through the fusion and integration of semantic context and digital mirror, preliminary negotiation data with real state mapping and semantic cognitive consistency is generated. Compared with traditional negotiation methods based on static attributes or rule driving, this step realizes cognitive synchronization, behavior rehearsal, and collaborative perception among resource agents, significantly improving the intelligence, interpretability, and prediction ability in the early stage of negotiation.

[0034] Preferably, the intelligent agent semantic graph is mapped through a preset shared semantic understanding graph model. The construction steps of the shared semantic understanding graph model include:

[0035] Obtain the key term data corresponding to the factory resource intelligent agent data;

[0036] Perform term alignment based on the key term data and a preset domain ontology library to obtain term alignment data, where term alignment includes concept layer alignment, example layer alignment, and attribute layer alignment;

[0037] Extract positive and negative sample pairs according to the term alignment data to obtain positive and negative sample pair data;

[0038] Perform cross-agent semantic training on the positive and negative sample pair data to obtain a shared semantic understanding graph model, where cross-agent semantic training is jointly trained through a Transformer structure encoding network and a bidirectional gated recurrent unit encoder.

[0039] In the present invention, by constructing a shared semantic understanding graph model, semantic alignment and cognitive consistency among multiple types of factory resource intelligent agents are achieved, providing a semantic foundation guarantee for intelligent collaboration. By extracting the key terms in the factory resource intelligent agent data and combining with a preset domain ontology library, multi-level term alignment from the concept layer, instance layer to the attribute layer is completed to ensure the structural comprehensiveness and industry adaptability of semantic mapping; based on the term alignment results, positive and negative sample pairs are constructed, and a joint modeling strategy of a Transformer structure encoding network and a bidirectional gated recurrent unit encoder is introduced during the training process, taking into account the long-distance semantic modeling ability and the temporal context expression ability, effectively improving the accuracy and generalization ability of semantic representation; the generated shared semantic understanding graph model can realize the unified interpretation and dynamic mapping of task semantics among different resource intelligent agents. Compared with the traditional method relying on static keyword matching or rule mapping, this step has the technical advantages of being adaptive, having strong context understanding, and being trainable and evolvable, significantly improving the semantic collaboration efficiency and task scheduling intelligence level in a multi-agent system.

[0040] Preferably, the semantic embedding of the willingness function is specifically:

[0041] Perform attention calculation on the factory resource intelligent agent data according to the cross-agent shared semantic data to obtain intelligent agent attention weight data;

[0042] Perform weighted aggregation on the factory resource intelligent agent data according to the intelligent agent attention weight data to obtain semantic perception feature data;

[0043] Perform semantic prior distribution calculation according to the cross-agent shared semantic data to obtain semantic prior distribution data;

[0044] Obtain the current state data of the agent corresponding to the cross-agent shared semantic data, and perform conditional Gaussian model fitting on the current state data of the agent to obtain the agent observation model;

[0045] Perform Bayesian posterior inference based on the semantic prior distribution data and the agent observation model to obtain the task preference probability data;

[0046] Perform expected value weight calculation based on the semantic perception feature data and the task preference probability data to obtain the intelligent task context data.

[0047] In the present invention, attention calculation is performed on the resource agent based on the cross-agent shared semantic data to obtain its attention weight in the current semantic scenario, realizing semantic-driven feature focusing; subsequently, semantic perception features are extracted through weighted aggregation, strengthening the agent's understanding depth of the task context; combining the semantic vector to generate the prior distribution of the task preference, and fitting the observation model with the current resource state as the input, a learnable agent state-response mapping relationship is constructed by means of the conditional Gaussian model; the semantic prior and the observation model are fused through the Bayesian posterior inference method to obtain a more stable and interpretable task preference probability; the intelligent task context data is generated through the fusion of the expected weight, providing a refined and dynamic preference cognition basis for the game strategy selection. Compared with the traditional static willingness scoring method, this step has technical advantages such as strong semantic adaptability, excellent prediction ability, and high model interpretability, significantly enhancing the multi-agent's cognition and collaboration ability in the complex task environment of the digital factory scheduling system.

[0048] Preferably, the task game specifically is:

[0049] Extract the task resource candidate mapping according to the preliminary negotiation data to obtain the task resource candidate mapping data;

[0050] Perform a willingness mapping on the task resource candidate mapping data according to the factory resource agent data to obtain the game response matrix data;

[0051] Perform a scenario semantic resource response according to the game response matrix data to obtain the scenario semantic resource response data;

[0052] Send a task assistance request to the factory resource agent data according to the scenario semantic resource response data to obtain the agent role response bid data;

[0053] Perform a collaborative ability assessment according to the agent role response bid data to obtain the collaborative ability assessment data, where the collaborative ability assessment includes a temporal consistency assessment and a semantic preference conflict assessment;

[0054] Generate a temporary task group according to the collaborative ability assessment data to obtain the negotiation game data.

[0055] In the present invention, based on the preliminary negotiation data, the task-resource candidate mapping relationship is extracted to construct the task assignable range; the states and semantic preferences of the factory resource agents are fused to generate the game response matrix among multiple agents, realizing the quantitative expression of the task competition intention; through the scenario semantic resource response, the adaptability of the task assignment to the current environmental context is enhanced; through the assistance request and role response bidding for the task sub-roles, the task collaborative bidding is realized, and the temporal consistency and semantic preference conflicts among the responding agents are verified in the collaborative ability evaluation stage; through the temporary task group generation mechanism, the negotiation game result is output, and the executable collaborative relationship between the task and the resource is constructed. Different from the traditional one-to-one assignment or static auction mechanism, this step realizes the integration of intelligent technologies such as dynamic alliance and game optimization based on semantic drive, and has stronger scheduling flexibility, resource coordination and task adaptation intelligence level.

[0056] Preferably, the initial local optimal extraction is specifically as follows:

[0057] Task-resource game candidates are extracted according to the negotiation game data to obtain task-resource game candidate data;

[0058] Local optimal target annotation is performed on the task-resource game candidate data to obtain local optimal target annotation data;

[0059] Based on the local optimal target annotation data, the optimal task execution allocation is performed on the factory resource agent data to obtain the preliminary allocation object data;

[0060] Task mutual exclusion filtering is performed according to the preliminary allocation object data to obtain the negotiated local optimal data.

[0061] In the present invention, by constructing a local task assignment optimization mechanism, high-quality task screening and assignment optimization are realized based on the multi-agent task game result. Based on the negotiation game data, the task-resource game candidate pairs are extracted to clarify the resource optional range and strategy information, providing a basis for local optimal extraction; by introducing a local optimal target annotation mechanism, the key indicators (such as willingness, response delay, energy consumption priority) in the task and resource allocation process are converted into quantifiable annotation data, making the local optimization have goal orientation; then, combined with the annotation data and the state of the resource agent, the optimal matching strategy of the task assignment is executed to ensure that the preliminary result is the most suitable for the current resource configuration within the local range; through the task mutual exclusion filtering mechanism, the allocation schemes that cannot be executed simultaneously or have strategy conflicts are excluded, effectively improving the executability of the global scheduling and the resource utilization efficiency. Compared with the single-round optimal selection or fixed priority strategy in the traditional scheduling, the present invention has the technical advantages of strong local responsiveness, high strategy adaptability and strong task conflict avoidance ability, laying a key foundation for the system to achieve high-efficiency and low-interference intelligent scheduling.

[0062] Preferably, for implementing the digital factory full-process collaboration method as described above, the digital factory full-process collaboration system includes:

[0063] A process parameter analysis module for obtaining production decision data and performing process parameter analysis based on the production decision data to obtain process parameter data;

[0064] A resource intelligent agent construction module for constructing factory resource intelligent agents according to the process parameter data through a preset digital factory model to obtain factory resource intelligent agent data;

[0065] A multi-intelligent agent negotiation and task allocation module for performing multi-intelligent agent negotiation based on the factory resource intelligent agent data to obtain digital workshop task allocation data;

[0066] An industrial Internet of Things monitoring and feedback module for performing industrial Internet of Things monitoring based on the digital workshop task allocation data to obtain industrial Internet of Things real-time data for assisting in the full-process collaboration operation of the digital factory.

[0067] The beneficial effects of the present invention are as follows: By performing structured process parameter analysis on production decision data in step S1, the rapid conversion from high-level decision-making to executable processes is realized, improving the early response efficiency of the scheduling system; In step S2, based on the digital factory model, factory resource intelligent agents with state perception, willingness evaluation, and strategy response capabilities are constructed, and the collaborative relationship between resources is established through attribute configuration and collaboration graph construction, enhancing the resource expression ability and scheduling flexibility of the model; Step S3 realizes the adaptive negotiation and coalition-style task allocation among multiple intelligent agents, effectively solving the intelligent matching problem under task conflicts and resource constraints; Step S4 combines industrial Internet of Things monitoring means to synchronize the digital scheduling results to the physical execution layer, realizing virtual-real closed-loop feedback and real-time adjustment. Compared with existing systems based on static rules or linear scheduling, the present invention has stronger intelligence, semantic consistency, resource collaboration ability, and system dynamic adaptability, and can significantly improve the collaboration efficiency and flexible production ability of digital factories in complex environments. Description of the Drawings

[0068] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objectives, and advantages of the present application will become more obvious:

[0069] Figure 1 Shows the step flow chart of a digital factory full-process collaboration method of an embodiment;

[0070] Figure 2 Shows the step flow chart of a process parameter analysis method of an embodiment;

[0071] Figure 3The flowchart of the steps of a method for constructing a resource agent according to an embodiment is shown;

[0072] Figure 4 The flowchart of the steps of a method for multi-agent negotiation and task allocation according to an embodiment is shown. Detailed implementation manners

[0073] The technical method of the present invention patent will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0074] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated descriptions of them will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0075] It should be understood that although terms such as "first", "second", etc. may be used here to describe each unit, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit can be called the second unit, and similarly the second unit can be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed related items.

[0076] Obtain a task through the intelligent factory management interface, and parse the task to obtain that the task includes producing 10 high-precision customized parts and delivering them within 48 hours; the process includes cutting → welding → inspection (each piece requires 60 minutes); query resources to obtain cutting machine A (status: idle), welding arm B (currently performing a regular task and will be idle in 2 hours), welding arm C (can work overtime), inspection station C (prioritize urgent orders), personnel P1 (not on duty at night), P2 (support throughout the day); perform parameter parsing to obtain the time required for each product: 60 minutes, totaling 600 minutes, which can be split into 10 task units, each containing 3 steps.

[0077] Parse the resources to obtain the following information: Cutting machine A: idle, response score 0.92; Welding arm B: expected idle time = current time + 2h; Welding arm C: currently busy, but the strategy is set to "allow overtime at night", response score 0.85; Inspection station C: score 0.96, prioritize urgent orders; P1 is unavailable, P2 can participate in improving the welding agent score.

[0078] Perform task game scoring and sorting to obtain Cutting machine A > A2 (select A), Welding arm C (because B is still busy), Inspection station C (perform inspection operations). The system selects (A, C, C) as the minimum delay execution combination. The Internet of Things monitors that the cutting execution time slightly exceeds the expectation (21 minutes). The system dynamically adjusts the welding start time +5 minutes, and the inspection station delays 1 order. The system records and adjusts the task buffer time strategy.

[0079] Please refer to Figures 1 to 4 , this application provides a digital factory full-process collaboration method, including the following steps:

[0080] Step S1: Obtain production decision data and parse process parameters according to the production decision data to obtain process parameter data;

[0081] Specifically, collect structured data such as planned production tasks, material information, product specifications, production time windows, and delivery cycles from the enterprise's production management system (such as MES, ERP, or other task scheduling platforms). Data access can be achieved through API interfaces or OPCUA protocols. Compare the product specifications with the process standard library to automatically extract key process parameters, such as processing temperature, pressure, speed, mixing ratio, etc.; for multi-variety mixed-line production, match the process template through preset product classification rules (such as material code recognition based on the BOM structure); use a rule engine (such as Drools) to parse process constraint logic, such as "Material A and equipment B can only be used for process type C", to ensure that the generated results meet technical specifications; the process parameters are organized in JSON format, and the parameter sources and parsing paths are recorded for traceability.

[0082] Step S2: Build factory resource agents according to the process parameter data through a preset digital factory model to obtain factory resource agent data;

[0083] Specifically, the model is constructed based on the BIM and simulation model, including digital twins of virtual devices, workstations, human resources, and transportation equipment. The model is marked with a unique ID and resource tags (such as "laser cutting machine #A01"). Each type of resource is abstracted as an "Agent", which includes basic attributes (such as type, ability, status, current load) and behavior functions (such as receiving orders, executing, and feedback); the system uses the obtained process parameter data as input features and makes task assignment judgments for resource agents through built-in matching rules. The rules may include adaptation condition judgments, such as "processing temperature > 100°C" needs to match the equipment type with high-temperature tolerance ability; equipment priority strategy, among multiple optional resources, preferentially match the equipment with high resource idle degree and excellent historical success rate; production rhythm and capacity constraints, such as a certain type of processing task needs to meet the minimum output capacity per unit time at the same time. Each Agent has a state machine to manage states such as idle / ready / processing / fault, and dynamically updates. The output data structure is a list of Agent objects, including Agent-ID, adapted process list, resource capabilities (numeric or enumeration type), and current status.

[0084] Step S3: Perform multi-agent negotiation based on the factory resource agent data to obtain the digital workshop task assignment data;

[0085] Specifically, based on the multi-agent market bidding simulation mechanism, the optimal matching between tasks and resources is achieved. Each resource Agent bids according to its idle degree and task fitness. A task-resource adaptation scoring function is introduced. For example: adaptation score = α × resource load + β × device compatibility + γ × historical success rate, where α, β, and γ can be dynamically adjusted according to the factory policy. Taking the device candidate set of task T_001 as an example, the current state of device E_23 is as follows: the current resource load is 0.60, the device compatibility score is 0.90, and the historical task success rate is 0.85. The system policy weights are set as α = 0.4, β = 0.3, and γ = 0.3. In the case of multiple Agent bid responses, the system scheduler will make a priority selection based on the scoring results, assign the task to the resource with the highest adaptation score, and record the current scheduling decision process in the scheduling log to support traceability analysis and subsequent optimization. All resource Agents receive task requests and broadcast response information through a unified lightweight communication channel (such as the MQTT protocol or ROS nodes). The main scheduling Agent aggregates all response information and comprehensively considers task priorities, resource load balancing strategies (such as least load first, polling strategy, historical performance backtracking, etc.) to generate a task allocation table. During the negotiation process, the system supports a timeout retry mechanism (such as automatically resending the task broadcast if no appropriate response is received within the set time) and a failure fallback mechanism (such as introducing backup resources or entering the manual approval mode if multiple rounds of negotiation fail), ensuring the robustness and continuity of the task allocation process. The formed task allocation data structure includes: task unique identifier (Task ID), allocated resource number (Resource ID), planned start time, expected completion time, and necessary process parameters (such as processing temperature, cutting speed, robot trajectory, etc.). This structure will be used for subsequent task execution control and docking with the industrial monitoring system to achieve closed-loop collaboration in the production process.

[0086] Step S4: Perform industrial Internet of Things monitoring based on the digital workshop task allocation data to obtain real-time industrial Internet of Things data for collaborative auxiliary operations throughout the digital factory process.

[0087] Specifically, an edge computing gateway is deployed in the digital workshop to connect various field devices (including programmable logic controllers PLC, computer numerical control machines CNC, temperature control instruments, industrial robots, etc.) to the industrial Internet of Things system, realizing the standardized access of device status and task execution information. The edge gateway supports multiple industrial communication protocols (such as MQTT, OPC UA, Modbus), and the monitored data collected includes but is not limited to the following indicators: task progress, device status, processing temperature, vibration signal, energy consumption, alarm information, etc. Each indicator is associated with the assigned task to construct a task-status mapping table. The real-time data is fed back to the scheduling center. If there are delays, anomalies, or equipment downtime, a rescheduling process is triggered; the collaborative operation system dynamically adjusts resource scheduling and task priorities based on the real-time data to form a closed-loop control. All monitored data is displayed through a visualization panel, and at the same time, a RESTful API is provided for the upper-layer business system to call.

[0088] Preferably, step S1 is specifically as follows:

[0089] Step S11: Obtain production decision data;

[0090] Specifically, the production decision data can be extracted from the enterprise's ERP system, APS system, or order management platform, mainly including order ID, order type, delivery date; product model, specification parameters; customer customization requirements; inventory status and procurement plan; safety / quality-related constraints. It is synchronized regularly through a preset API or database intermediate table; if the MES system is used, it can be obtained in real time through the WebService interface; for the situation of coexistence of multiple systems, unified extraction is performed through the data center to avoid redundancy and conflicts.

[0091] Step S12: Extract the process flow from the production decision data to obtain process flow data;

[0092] Specifically, a process template library based on product categories is constructed. Each template contains a standardized processing process flow (such as: blank preparation → rough machining → heat treatment → finish machining → inspection → packaging); the template is modeled in the form of a flow chart and uses BPMN or Petri net structures to express processing nodes and their dependencies. The most appropriate process flow template is matched according to the product specification fields (such as model, size, material); for customized products, a classifier (such as random forest) trained with historical production data is used to predict the optimal process path; a conditional branch mechanism (such as whether heat treatment is required) is introduced to dynamically cut process nodes. The process flow data is represented in a directed graph structure, and each node contains a processing step ID, required resource type, process name, etc.

[0093] Step S13: Analyze the process parameters based on the process flow data to obtain process parameter data, where the process parameter data includes the required equipment type, process sequence, material usage, man-hour estimation, energy demand, and environmental constraints.

[0094] Specifically, each process node queries the equipment type that supports the process in the equipment capacity library according to its processing method field (such as laser cutting, CNC milling); if the same process is applicable to multiple devices, the priority rule is "production capacity priority > equipment availability > optimal energy consumption".

[0095] Based on the node topological sorting in the flowchart, the dependency relationship (predecessor - successor) is directly mapped to the execution order; special cases (such as multiple steps in parallel) are marked as parallel processes and processed by the task assignment module.

[0096] Extract the unit material consumption from the BOM structure; calculate the total material requirement according to the order quantity and the process loss rate; if semi-finished product conversion is involved, an additional intermediate material balance calculation module is added.

[0097] Establish a man-hour model based on empirical formulas or historical data, such as: man-hour = basic processing time + material correction coefficient + difficulty correction coefficient, and use models such as linear regression or XGBoost to dynamically calibrate by comparing multiple historical cases.

[0098] Estimate the energy consumption based on the equipment type, processing duration, and rated power; for special processes (such as heat treatment), auxiliary energy types (gas, steam, etc.) need to be added to the calculation; output a structured record in units of kWh / unit or total energy consumption / batch.

[0099] In terms of environmental constraint identification, the system can judge whether each process involves environmental control conditions, such as temperature and humidity requirements, cleanliness level (for example, electronic packaging needs to meet a cleanliness environment of 100,000 levels), etc.; at the same time, detect whether there are process nodes with high pollution, high noise, or harmful emissions. If so, mark them as tasks requiring environmental protection linkage, and record their constraint types, target values, and constraint durations.

[0100] Preferably, step S2 is specifically as follows:

[0101] Step S21: Instantiate the intelligent agents of factory resource types through a preset digital factory model according to the process parameter data to obtain intelligent agent data of factory resource types, where the intelligent agent data of factory resource types includes production line intelligent agent data, equipment intelligent agent data, personnel intelligent agent data, and material intelligent agent data;

[0102] Specifically, a unified resource modeling framework is established. The model contains four types of resource type metaclasses (production line, equipment, personnel, and materials), and each type contains a basic attribute template (such as resource ID, resource capacity, status interface, communication interface, etc.); all templates are defined based on JSON Schema or XML for subsequent instantiation and parsing.

[0103] Traverse the processing flow and resource type requirements involved in the process parameter data (such as "required equipment type: laser cutting machine", "required production line is an automatic packaging line"); map the matched resource types to the model templates one by one, generate agent instances as needed, and assign unique identification IDs (such as AGENT_DEVICE_LASER_001); if there are multiple candidate entities for a certain resource (such as multiple devices of the same model), batch instantiation is performed, and each agent records the corresponding physical resource number and its initial state.

[0104] Step S22: Configure the attributes of the factory resource type agent data to obtain agent attribute data, where the attribute configuration includes status information configuration, willingness function configuration, and policy space configuration;

[0105] Specifically, each agent maintains a set of dynamic state attributes to describe its real-time availability and current task situation, mainly including status representing the current running state, which can take values such as idle (idle), running (executing), maintenance (under maintenance), etc.; load representing the current task load, expressed as a percentage, for example, 0% - 100%; energy_state representing the remaining working ability, which can specifically be the remaining power, remaining executable time, or man-hour budget; the initial state is read from the MES or edge gateway, and the update period is dynamically configured from 5 seconds to 30 seconds.

[0106] Each agent has a "willingness function" to judge its willingness to accept orders under different task requests. This function consists of the following factors: current load (the lower the load, the higher the willingness); task matching degree (the degree of fit with its ability model); historical execution success rate; equipment depreciation priority (high-value equipment takes on key tasks first); an example willingness function is as follows (value range 0 - 1): w = w1·(1 - current load) + w2·task matching degree + w3·success rate - w4·equipment depreciation index, where w1 is the load weight coefficient, with a value of 0.30, w2 is the matching degree weight coefficient, with a value of 0.30, w3 is the success rate weight coefficient, with a value of 0.20, and w4 is the depreciation penalty coefficient, with a value of 0.10;

[0107] Each agent defines an executable set of policies, such as an order acceptance policy, an adjustment policy, a rejection policy, etc., and has the ability of autonomous decision-making. For example, the policy space of the device agent includes accept_task_if(load < 70% and match_score > 0.8), which means accepting a task when the load is low and the matching degree is high; delay_task_if(temperature > threshold), which means delaying a task when the device temperature exceeds the safety threshold; reject_task_if(predictive_failure = true), which means rejecting a task if the device status prediction is a potential failure. The set of policies is expressed in DSL or a rule engine and supports hot updates.

[0108] Step S23: Construct a task collaboration relationship graph for the agent attribute data according to the process parameter data to obtain the factory resource agent data.

[0109] Specifically, take the process flow chart (steps, order, dependencies) in the process parameters as the input; each processing step is used as a task node, and the node attributes include the required resource type, process ID, estimated working hours, etc.; the connection lines between the nodes represent the dependency relationship (for example, OP20 needs to wait for OP10 to complete), forming a directed task graph. Assign a candidate agent set to each task node (matched according to the resource type); establish a many-to-many mapping table between tasks and resources; record the set of collaborative resources and the sequence logic between resources in the node (such as the constraint of needing to execute tasks continuously on the same production line). Record it in the form of an adjacency list and a resource mapping table, or use a graph database (such as Neo4j); each graph node contains the bound resource type, candidate agent set, and status flag (allocated / unallocated / conflicted).

[0110] Preferably, step S3 is specifically:

[0111] Step S31: Initialize the negotiation according to the factory resource agent data to obtain the preliminary negotiation data;

[0112] Specifically, all agent entities eligible for tasks are extracted from the output factory resource agent data. The determination of task eligibility is based on the following conditions: the resource type needs to match the task requirements (such as machining type matching, resource capabilities covering the required process parameters), the current status needs to be idle or schedulable, and it is not in the state of being bound to other tasks. Before initializing the negotiation state, it includes generating a candidate resource agent set for each task; initializing the willingness function (such as the aforementioned willingness function) and status attributes of each agent; constructing a task-resource initial mapping table for all tasks and their corresponding candidate resource sets; registering a unique communication identifier for all agents participating in the negotiation, and establishing a communication channel (such as ROS nodes, MQTT topics) for information broadcasting and feedback collection among agents. Perform task adaptability analysis on all agent objects in the factory resource agent data, initialize the task candidate set based on their resource type, capability matching degree, and status parameters, construct a task-resource matching mapping relationship, and set the agent negotiation state to unlocked to obtain preliminary negotiation data.

[0113] Step S32: Conduct task games based on the preliminary negotiation data to obtain negotiation game data;

[0114] Specifically, adopt a task-oriented multi-agent game mechanism (such as discrete-time round-robin games). In each game round, each resource agent actively participates in the competition according to the willingness function and submits its task acceptance proposal. Define the game round. In each round, the agent bids on the acceptable tasks (which can be calculated based on load, willingness score, etc.); the task node evaluates the candidate bids according to the scoring function after receiving the proposal; after each round of the game, update the agent bid status and the task assignment intention table. The task response score = α × matching degree + β × availability - γ × current load + δ × priority, where α is the matching degree weight, with a value of 0.4, emphasizing the matching degree, regarded as the task "finding the right person / equipment", β is the availability weight, with a value of 0.2, availability is important but not the only decision point, γ is the load suppression coefficient, with a value of 0.3, moderately suppressing high-load resources to avoid overload, and δ is the priority enhancement weight, with a value of 0.1. If the system is sensitive to the urgency of the task, it can be adjusted to 0.2 - 0.3. After each round of the game, the system will update the bid status, task response status of the agent, and some resource conflict situations, and maintain the task assignment intention table and the task game status table. Subsequently, enter the next round of the game until the preset round is reached or the task assignment converges. The system constructs a task game model based on the preliminary negotiation data, comprehensively considers the candidate resource responses received by each task node, combines the task requirement characteristics and the current capabilities of the resources, generates a game mapping relationship between tasks and resources, and outputs and forms negotiation game data.

[0115] Step S33: Perform initial local optimal extraction based on the negotiation game data to obtain negotiation local optimal data;

[0116] Specifically, after completing the initial round of task negotiation games among multiple agents, the system obtains the willingness score data between all tasks and their candidate resource agents. The game score is a scoring metric generated by weighting the aforementioned multiple factors, reflecting the suitability of each resource for a specific task. The system will separately execute the following processing flow for each task to sort the candidate agents and sort the scores of the candidate resource set corresponding to the task; extract the preferred agent, select the resource agent with the highest score as the current preferred execution candidate for the task; record the initial mapping relationship, bind the task ID with the identifier of its preferred resource agent, and record it as the preferred resource field; construct the local optimal table, summarize the preferred matching relationships of all tasks, and construct an initial local optimal task allocation table.

[0117] Step S34: Detect global resource conflicts based on the negotiated local optimal data to obtain global resource conflict data;

[0118] Specifically, detect whether multiple tasks are assigned to the same agent (resource conflict), or whether a resource is scheduled for tasks with overlapping time. Check whether the resource ID is repeatedly allocated; if the same resource is occupied by multiple tasks simultaneously, record it as a static resource conflict event, and classify and summarize the conflict data according to the conflicting resource ID and the conflicting task group. The system further compares the time intervals occupied by all resource tasks. If it is found that there is an overlapping time interval for the same resource among multiple tasks (i.e., the task execution time ranges intersect), record it as a dynamic time conflict event. Detect resource reuse based on the negotiated local optimal data, identify conflict situations such as repeated scheduling of resource IDs or overlapping task execution times, and generate global resource conflict data.

[0119] Step S35: Annotate and limit the negotiated game data according to the global resource conflict data to obtain the negotiated annotation agreement data;

[0120] Specifically, mark the conflicts of the conflicting resources in the negotiated game data and set acceptable scheduling conditions or restrictions, such as marking the conflicting tasks as in a state of needing to renegotiate; update the policy space of the resources, such as setting the minimum interval time between tasks; attach a penalty term to the willingness function to reduce the priority of the conflicting resources in the game, such as the willingness value being equal to the original willingness value minus the conflict penalty factor term, and the conflict penalty factor term is obtained by regression calculation based on the conflict level and the historical conflict frequency, and the regression coefficient is fitted based on historical data. Feed the global resource conflict data back into the negotiated game model, and annotate and limit the relevant task agent pairs according to the conflict type, including execution order restrictions, adjustment of resource reallocation priorities, policy space contraction, etc., to generate the negotiated annotation agreement data.

[0121] Step S36: Perform restricted local optimal extraction based on the negotiated labeling agreement data to obtain digital workshop task allocation data.

[0122] Specifically, the local optimal solution is recalculated under the conflict constraint to form the most conflict-free task allocation table: unavailable resources are excluded or new scheduling windows are set; the optimal allocation is reconstructed based on the updated willingness function and policy rules, that is, the weight calculation of the minimized resource status, task priority and policy matching degree is calculated, the corresponding resource status weight is 0.5, the corresponding task priority weight is 0.3, and the corresponding policy matching degree weight is 0.2; the system uses heuristic matching based on task priority sorting, maximum weight matching algorithm or local graph search algorithm to combine and screen candidate task-resource pairs to ensure that the results retain the original task intention as much as possible under the premise of meeting the scheduling constraints, taking into account executability and resource balance, and outputting conflict-free and time-reasonable task→resource binding relationships. Based on the negotiated and annotated agreement data, the task redistribution optimization operation under restricted conditions is performed, and the local optimal task acceptance plan is re-extracted based on resource usage restrictions and time constraints to obtain digital workshop task allocation data.

[0123] Preferably, the negotiation initialization is specifically as follows:

[0124] Perform agent semantic graph mapping based on factory resource agent data to obtain cross-agent shared semantic data;

[0125] Specifically, define the factory resource ontology, including classes: equipment class, personnel class, production line class, material class, and class attributes: capability, status, region, communication interface, etc.; the system converts each resource agent (such as equipment, personnel, etc.) into a triple, maps its basic attributes (such as resource ID, capability set, current status, region) to nodes in the graph, and connects and builds a graph structure through the above attribute relationships. For example, use RDF / OWL semantic structure to describe the type relationship and upstream and downstream dependency between agents; map each agent data (such as equipment ID, capability set, current status) to a node in the graph; establish the following edge-to-edge relationships: hasCapability (capable), locatedIn (location), connectedTo (resource interface communication association), isCandidateFor (candidate for a task). Generate cross-agent shared semantic data: use graph databases (such as Neo4j, GraphDB) to store and query; support SPARQL semantic queries to identify agent combinations that meet the conditions for collaboration. A semantic graph is constructed based on the factory resource agent data, and semantic class relationships including equipment, personnel, production lines and materials are defined. Semantic mapping is performed through attributes such as resource capabilities, spatial distribution, and communication interfaces to obtain cross-agent shared semantic data that is used to support multi-agent understanding and collaboration.

[0126] Perform semantic embedding of the willingness function on the cross-agent shared semantic data to obtain intelligent task context data;

[0127] Specifically, use knowledge representation learning methods (such as TransE, R-GCN) to vectorize the entities and relationships in the graph; the embedding goal is to construct a low-dimensional vector space for expressing the collaborative willingness state of each agent. Map the willingness function of each agent (such as weighted calculation according to load, matching degree, energy consumption, task priority) into a function item embedding; the system embeds this willingness function as a "willingness expression vector" into the semantic vector space, so that each agent forms a willingness intensity point and a resource adaptation tendency vector in the semantic space, expressing its potential scheduling tendency in the multi-task collaboration scenario. To achieve context matching between resources and tasks, the system also embeds task requirements (including processing accuracy requirements, energy consumption limits, execution order dependencies, process special constraints, etc.) into the same semantic vector space through feature extraction and label encoding to form a task semantic embedding representation. Calculate the context matching degree of agent-task by calculating the cosine similarity or Euclidean distance. Based on the cross-agent shared semantic data, adopt a semantic embedding mechanism to vectorize and model the willingness functions of each agent, construct a low-dimensional feature space representing the task context understanding ability, and form intelligent task context data that can be used by agents to judge task priorities and acceptance willingness.

[0128] Perform digital twin mirror loading based on the intelligent agent data of the factory resources to obtain intelligent agent digital mirror data;

[0129] Specifically, load the corresponding digital twin mirror model for each physical resource agent, including real-time operating status (state machine / sensor values); historical usage records (failure rate, task success rate); prediction information (estimated remaining life, energy consumption model); data can be pulled from edge devices or historical databases using OPCUA, MQTT protocols. Each mirror object adopts a unified structure: {entity ID, current state, historical trajectory, predictable parameters}; a visualization interface (such as Unity, Cesium) can be used to optionally display the twin state.

[0130] Associate and integrate the intelligent task context data and the intelligent agent digital mirror data to obtain preliminary negotiation data.

[0131] Specifically, merge the semantic embedding (behavior tendency) and digital twin mirror (real-time ability) of the same agent; establish a hybrid structure, including the current state; collaborative context score; ability-task matching degree; scheduling recommendation weight. Use a weighted fusion formula to generate a preliminary negotiation score: is the willingness weight coefficient, with a value of 0.4, is the state weight coefficient, with a value of 0.3, is the reliability weight coefficient, with a value of 0.3. Generate an initial collaboration candidate matrix for each task agent and mark the candidate level. Perform feature-level and policy-level association integration on the intelligent task context data and the agent digital mirror data, fuse its task understanding ability, current resource status, and behavior prediction ability, and construct preliminary negotiation data for use in the game negotiation process to support local optimal calculation and conflict detection. The system constructs a task-resource scoring matrix based on the above scoring results, where each element represents the collaboration fitness between a task and a resource agent. According to the matrix results, sort the resource agents that can be matched for each task, and divide the candidate levels according to the score levels. For example, level A: score greater than 0.85, is a high-priority candidate; level B: score between 0.7 and 0.85, is an optional candidate; level C: lower than 0.7, is a secondary candidate or alternative resource.

[0132] Preferably, the agent semantic graph mapping is performed through a preset shared semantic understanding graph model. The construction steps of the shared semantic understanding graph model include:

[0133] Obtain the key term data corresponding to the factory resource agent data;

[0134] Specifically, extract the key information fields from the structured and semi-structured data of the factory resource agent, such as type fields (such as injection molding machine, process section); ability fields (such as maximum load, supported material type); status fields (such as running, pre-maintenance). Sort the structured data field names and values using TF-IDF; use NER (named entity recognition) to extract term phrases for descriptive fields (such as equipment description, task annotation); filter out low-frequency words and general words, and retain professional terms. Uniformly encode Chinese and English terms (such as convert to pinyin-underscore style: chu_kou_leng_que_xi_tong); retain the original context, location, and entity ID mapping of the terms. Obtain the term information involved in the factory resource agent data, including resource type, ability parameters, status labels, and attribute descriptions, etc., and extract the key term data using keyword extraction and named entity recognition methods to provide the basic input for subsequent semantic alignment.

[0135] Perform term alignment based on the key term data and the preset domain ontology library to obtain term alignment data, where the term alignment includes concept layer alignment, example layer alignment, and attribute layer alignment;

[0136] Specifically, an ontology knowledge base for the field of intelligent manufacturing is constructed, including a concept layer, such as equipment class, process class, and personnel class; an attribute layer, such as power, precision, number of stations, and displacement range; and an instance layer, such as specific model equipment, personnel roles, and material batches. The concept layer alignment is to use word embedding similarity (Word2Vec, FastText) and edit distance to determine whether they belong to the same concept category. For example, CNC machining equipment and numerically controlled machine tools will be aligned to the equipment class concept node; the instance layer alignment is to cluster based on task context and equipment task historical behavior similarity. Feature encoding is performed based on the task context semantic vector (such as task description, instruction path) and equipment historical behavior data, and an unsupervised clustering algorithm (such as KMeans or DBSCAN) is used to analyze the behavior similarity of multiple devices / personnel, and then determine whether they can be classified as the same instance object; the attribute layer alignment is to compare the semantic similarity between the term and the ontology attribute name (such as the alignment of "maximum force" and "rated thrust"), calculate the semantic similarity between the attribute items in the term and the standard attributes in the ontology (using models such as BERT, SBERT, etc.), and determine whether an attribute alignment relationship is formed. For example, maximum force and rated thrust will be recognized as semantically equivalent attributes; if there are differences in attribute units (such as N and kN), they will be uniformly converted to the standard unit and normalized (such as maximum value normalization or standard deviation normalization). Align the key term data with the preset ontology library in the field of intelligent manufacturing. The term alignment includes concept layer alignment, instance layer alignment, and attribute layer alignment, and uses semantic similarity calculation and attribute normalization processing methods to generate structured term alignment data.

[0137] Extract positive and negative sample pairs according to the term alignment data to obtain positive and negative sample pair data;

[0138] Specifically, the positive sample generation rule: if the matching degree between the term and the ontology item > threshold (such as 0.85), it is marked as a positive sample; the term group with successful concept alignment is used as a positive pair combination; the attribute names and units that are the same or completely match after conversion are used as positive attribute alignment pairs;

[0139] The negative sample generation rule: extract from terms with different semantic categories or context mismatches; control the positive and negative ratio (such as 1:3) to improve the training generalization ability; exclude phrases with similar spellings but opposite meanings (such as internal cooling system vs cooling shell);

[0140] Construct a semantic alignment sample set based on the term alignment data. The samples include matching term pairs as positive sample pairs, and term pairs with inconsistent concepts and context disconnection as negative sample pairs, forming positive and negative sample pair data that can be used for semantic relationship learning.

[0141] Perform cross-agent semantic training on positive and negative sample pair data to obtain a shared semantic understanding graph model, where the cross-agent semantic training is jointly trained through a Transformer structure encoding network and a bidirectional gated recurrent unit encoder.

[0142] Specifically, use the Transformer structure encoder to extract term context semantic features; parallel the Bi-GRU (bidirectional gated recurrent unit) structure to capture word sequence order features; the model training adopts a triplet loss function: Loss = max(0, margin + d(anchor, positive) - d(anchor, negative)), where Loss is the triplet loss value, max is the maximization function, margin is the set minimum semantic discrimination threshold, d(x, y) is the distance between vectors x and y in the embedding space (such as the Euclidean distance), anchor is the encoding vector of the current target term, positive is a synonymous term with a similar meaning to anchor, and negative is a term that has no relation or semantic difference with anchor. Set the minimum discrimination threshold margin (such as 0.5) to achieve the effect of semantic contrast learning; convert the term text into a Token sequence, input it into the Transformer to extract the semantic feature vector; at the same time, input it into the Bi-GRU encoder to extract context relevance; the joint output after splicing is sent to the fully connected layer for training and classification. After optimization, the distance between synonymous terms in the embedding space is as close as possible; the distance between heteronymous terms is widened, with strong semantic discrimination ability; after training, the model can be used for automatic classification of new terms and semantic label recommendation. Perform cross-agent semantic training on positive and negative sample pair data, adopt a joint semantic representation model composed of a Transformer structure encoding network and a bidirectional gated recurrent unit (Bi-GRU), and optimize and train through a triplet loss function to generate a shared semantic understanding graph model, which is used to establish unified semantic understanding and context expression capabilities among multiple agents.

[0143] Preferably, the semantic embedding of the willingness function is specifically as follows:

[0144] Calculate the attention of the factory resource agent data according to the cross-agent shared semantic data to obtain the agent attention weight data;

[0145] Specifically, input A: factory resource agent data (attributes of each agent, task adaptation labels, etc.); input B: shared semantic understanding graph embedding (term-level semantic vectors, upstream and downstream dependency information). For each task objective, calculate the attention scores of all candidate agents in the context of the semantic graph, and adopt the following scoring function: Where A t t eniwhere Softmax is Q i is the attribute vector of the i-th agent, K is the set of key vectors from the semantic graph embedding, T is the transpose symbol, and d k is the vector dimension scaling coefficient; the output result is the agent attention weight matrix corresponding to each task. The attention score represents the priority of each agent under the current task and semantic context. The system can implement an agent matching evaluation mechanism based on semantic enhancement, improving the dynamic adaptation ability and collaborative efficiency between tasks and resources.

[0146] Weighted aggregation is performed on the factory resource agent data according to the agent attention weight data to obtain semantic perception feature data;

[0147] Specifically, the attention weight is used as the weighting coefficient to perform a linear combination of the original attribute vectors of each agent, obtaining a new feature representation that integrates semantic perception ability. The feature dimensions include the ability dimension (productivity, precision, etc.); the state dimension (availability, health); the semantic dimension (task relevance score). Based on the agent attention weight data, weighted aggregation is performed on the factory resource agent attribute vectors, integrating the semantic attention information of the task context, and generating semantic perception feature data with context understanding ability.

[0148] Semantic prior distribution calculation is performed according to the cross-agent shared semantic data to obtain semantic prior distribution data;

[0149] Specifically, a joint frequency modeling is performed on the adaptation between task types and historical task agents; a multinomial distribution prior from task types to agent categories is constructed. The system collects the matching relationships between task types and participating resource agents in historical task assignment records, and jointly models the co-occurrence frequencies between task semantic tags and resource semantic tags to obtain the empirical adaptation degree of tasks to resource categories. Based on the above frequency statistics results, a conditional probability distribution model from task types to resource categories is constructed, forming a multinomial distribution structure, P(RT│TT) = f(TT, HC), where P(RT│TT) is the conditional probability distribution, that is, the semantic prior distribution data, RT is the resource type, TT is the task type, f(TT, HC) is the co-occurrence frequency mapping function that maps the task type TT and the historical co-occurrence data HC into a function structure of conditional probability (such as normalized frequency, Laplace smoothing, etc.), and HC is the historical co-occurrence frequency data; to enhance the stability and robustness of the model and prevent the zero-probability problem caused by sample sparsity, the system introduces Dirichlet prior estimation for the above multinomial distribution. By assigning non-zero prior smoothing factors to each resource category, the model has a certain generalization ability for unseen task-resource combinations. The system adopts a hierarchical Bayesian modeling structure, with the task type as the upper-level topic distribution of the model and the agent ability category as the lower-level resource response distribution, forming the following two-level modeling framework. Upper-level modeling: the semantic topic distribution of task categories; lower-level modeling: the resource ability response distribution under various tasks. Combining with the Dirichlet distribution to form a conditional resource tendency model, and then outputting the semantic prior distribution data, that is, the probability mapping set of resource categories corresponding to each task type.

[0150] Obtain the current state data of the agent corresponding to the cross-agent shared semantic data, and perform conditional Gaussian model fitting on the current state data of the agent to obtain the agent observation model.

[0151] Specifically, the state characteristic data of the factory resource agent includes, but is not limited to, the following indicators: load rate, remaining energy, reliability score, fault prediction value, etc.; and is conditionally combined with the semantic vector (such as the task type and the state variable form a joint distribution). The system uses a conditional Gaussian mixture model (CGMM) for modeling, and its structure is as follows: conditional variable (X), such as the task semantic embedding vector; response variable (Y), such as the vector composed of resource state variables; model objective, such as setting to learn P(Y∣X), that is, the probability distribution of the resource state Y under the condition of the task semantics X. The EM algorithm, a classic algorithm, is used for parameter estimation in the modeling process, including the following steps: E step (expectation step), estimating the posterior probability that the sample belongs to each Gaussian component under the current parameters; M step (maximization step): based on the results of the E step, updating the mean vector, covariance matrix, and mixing coefficient of each Gaussian distribution; iterating in a loop until the convergence condition is met to obtain the agent observation model.

[0152] Perform Bayesian posterior inference based on semantic prior distribution data and the agent observation model to obtain task preference probability data;

[0153] Specifically, based on the semantic prior distribution data and the agent observation model, use the Bayesian posterior inference mechanism to estimate the task preference probability data of each agent for different task types, which is used to express the context adaptation tendency of resources to tasks. The semantic prior distribution data represents the task preference trend of various tasks in the semantic space, denoted as P(T k ), where represents the set of task types; the agent observation model is a state observation model based on conditional Gaussian distribution, which is used to estimate the likelihood probability of the current state O k of the agent under the given task condition T j , denoted as P(O j ∣ T k ), where O j represents the agent state vector; the system then applies Bayes' theorem to calculate the posterior probability of the agent for the task type T k , and the expression is as follows: P(T k ∣ O j ) is the preference probability (posterior) of the agent for the task type T k , P(O j ∣ T k ) is the likelihood of the agent state occurring under the given task type, P(T k ) is the semantic prior probability of the task type, l is the candidate task type index, O j is the current state of the agent, and T k is the task type; this inference process jointly models the current state of the agent and the task prior features, outputs the preference probability distribution of each agent with respect to various tasks, and generates the task preference probability data.

[0154] Calculate the expected value weight according to the semantic perception feature data and the task preference probability data to obtain the intelligent task context data.

[0155] Specifically, the semantic perception feature vector (extracted by the previous encoding module and used to represent the semantic feature vector of each resource agent in the semantic embedding space) is used as the feature weight; the task preference probability (generated by the semantic prior modeling module and used to express the preference probability distribution of the current task type for each resource type) is used as the importance factor; and the weighted combination of the expected feature vectors of each agent is performed. The semantic perception feature data and the task preference probability data are jointly weighted and calculated to construct a task expected response function, that is, intelligent task context = semantic perception feature data × task preference probability data, and the intelligent task context data expressing the response ability of the agent in the current task context is obtained.

[0156] Preferably, the task game specifically is:

[0157] Extract the task resource candidate mapping according to the preliminary negotiation data to obtain the task resource candidate mapping data;

[0158] Specifically, the task context vector (used to represent the semantic requirements of the task in the current negotiation scenario) from the preliminary negotiation data and the agent semantic features (used to represent the ability expression and semantic features of each resource agent). For each task, match based on the cosine similarity between the task context vector and the semantic vectors of each agent; set a threshold (such as similarity > 0.7) as the candidate resource admission condition. Record task ID → candidate resource list to form a task resource mapping table.

[0159] Perform a willingness mapping on the task resource candidate mapping data according to the factory resource agent data to obtain the game response matrix data;

[0160] Specifically, each agent calculates its own task acceptance willingness value according to its state (load, health, etc.) and preference parameters. Construct a willingness score table for all task-candidate resource combinations to form a matrix of task × agent; for each candidate task-agent pair (T i , A j ), the system constructs a willingness score function according to the following three types of indicators: M(i,j) = S cap (i,j) + S sem (i,j) ― D load (j), where M(i,j) is the willingness score between the candidate task and the resource agent, i is the task number, j is the agent number, S cap (i,j) is the matching score between the process required by the i-th task and the ability of the j-th resource, S sem (i,j) is the similarity score (such as cosine similarity) between the semantic label of the i-th task and the semantic embedding of the j-th resource, D load(j) The willingness discount term generated for the current load rate of the j-th agent can be directly set as the load removal coefficient or calculated by setting weights for the load coefficient; the system integrates the scoring tables of all tasks into a two-dimensional response matrix. Based on the factory resource agent data, the willingness function matching is performed on the task-resource candidate mapping data to construct the game response matrix data between tasks and agents, which is used to describe the response intensity of agents to each task.

[0161] Perform scenario semantic resource response according to the game response matrix data to obtain the scenario semantic resource response data;

[0162] Specifically, introduce task context information (such as execution time, material location, dependent tasks) into the agent response analysis; construct a multi-dimensional semantic context feature representation. If the task context conflicts with the resource physical environment (such as remote scheduling and remote resources), then adjust the response value; if there is an inconsistency between the physical environment where the agent is located and the task context (such as long resource distance, response time delay, previous task not completed, etc.), then the system introduces two scenario correction factors, the environmental consistency factor, which is used to measure the accessibility of resources in terms of spatial geographical location and job unit matching degree; the temporal accessibility factor, which is used to judge whether the current resource can respond and complete the task within the task required time. Add a scenario weight correction term, for example, response value = willingness value × environmental consistency factor × temporal accessibility factor. On the basis of the game response matrix, combine the task context scenario information to perform semantic correction and weighted processing on the agent response behavior to form the scenario semantic resource response data that integrates environmental constraints and execution feasibility.

[0163] Make a task assistance request for the factory resource agent data according to the scenario semantic resource response data to obtain the agent role response bid data;

[0164] Specifically, the system divides roles for each task according to process characteristics and process requirements, sets multiple roles for each task, including the main executor (undertaking the main processing tasks), the assistant (performing auxiliary operations such as clamping and supporting), the transporter (responsible for workpiece handling and transfer), the quality inspector (responsible for quality inspection of task results), etc.; each role type will be mapped to a specific resource category. For example: main executor → processing equipment or core operators; transporter → logistics robots or AGV cars; quality inspector → vision inspection equipment or quality engineering personnel. Each task broadcasts a task execution request (based on semantic response ranking) to its candidate resources; each agent returns whether to bid and the bid willingness value according to its own role strategy. For the scenario semantic resource response data, the task role semantic vector is matched with the semantic response scores of the candidate resources to select the current responsive candidate resource set. For each task role, the system sends a task assistance request (broadcast request) to its candidate resource agents. The resource agents that receive the request will make a bidding decision based on the following information: their current status (idle, busy, maintenance); the role order-taking rules in the current policy space (such as role priority, scheduling window); the willingness function score (indicating the response degree of the resource in the current semantic task). The response result includes two parts: the bid response flag indicating whether the resource participates in the role task; the bid willingness value indicating the score output according to its willingness function. The system records the mapping relationship between all task roles and responsive resources as structured agent role response bidding data, in the form of task ID – role type – candidate resource ID – bid willingness value.

[0165] Based on the agent role response bidding data, the collaborative ability is evaluated to obtain the collaborative ability evaluation data, where the collaborative ability evaluation includes temporal consistency evaluation and semantic preference conflict evaluation;

[0166] Specifically, the system judges whether there are overlapping tasks based on the planned execution window of the tasks responded by the bidding agents; time interval overlapping detection is performed based on the task schedule and the resource available window. If there is an overlapping part, it is regarded as a time conflict task pair. The system will reconcile according to the task priority or the game score, or temporarily exclude some bids. For each resource agent, its bidding willingness for different tasks may be relatively high at the same time. If the resource physically cannot execute multiple high-preference tasks concurrently, it is necessary to judge whether there is a semantic preference conflict in its bids. The system introduces a mutual exclusion coefficient for evaluation, defined as: EC(A k ) = ∑ i,j,i<j (w ki ·w kj ·δ ij ), where EC(A k ) is the agent mutual exclusion coefficient, i is the task number, j is the task number different from i, w kiThe bidding willingness value of the agent for task T i is w kj The bidding willingness value of the agent for task T j is δ ij is the task conflict indication factor. If there is a time overlap or resource exclusivity conflict between task T i and T j , then δ ij = 1, otherwise it is 0. The collaborative ability of the agent role's response bidding data is evaluated. The evaluation includes the timing consistency evaluation of task execution and the semantic conflict evaluation between task preferences, and is used to judge whether each resource has the feasibility and priority to participate in the current task game. The system will perform task allocation reconciliation based on the threshold of EC(A k ). For example, if EC(A k ) > θ, where θ is the reconciliation threshold, configured between [0.2, 0.6] and determined according to the system's tolerance for resource concurrency ability, it can be set. Then the system can perform priority sorting, score reduction or partial elimination on its conflicting bidding tasks; if EC(A k ) ≈ 0, it means that there is no obvious conflict in the current task combination of this agent, and it can directly participate in collaborative optimization.

[0167] Generate a temporary task group based on the collaborative ability evaluation data to obtain negotiation game data.

[0168] Specifically, to construct a resource collaboration plan that meets the task execution requirements, the system, based on the obtained collaborative ability evaluation data, identifies resource role combinations that pass the feasibility verification, and generates a temporary execution group for the task accordingly; the temporary task group is used to provide a set of resources with execution ability and coordination ability during the task negotiation and game stages. At least one main execution resource needs to be satisfied in the combination; each task involves several supporting roles (such as handling, assembly, inspection, etc.). The system requires that the actual allocation number of auxiliary roles in the task group is not less than the set coverage threshold (such as 80%) of the total number of required roles. If there are multiple combinations, they are sorted by the evaluation score (weighted calculation of indicators such as resource willingness, timing consistency, and load balance, or directly obtained based on resource willingness), and the Top-N candidate solutions are selected as the negotiation output, including task ID, resource combination list (annotating the role types of each resource), collaborative ability evaluation score, and role coverage rate and executable status mark. Construct a temporary task group that meets the task requirements based on the collaborative ability evaluation data. The task group includes a main execution resource and multiple auxiliary role resources, forming negotiation game data for negotiation decision-making.

[0169] Preferably, the initial local optimal extraction is specifically as follows:

[0170] Perform task resource game candidate extraction based on negotiation game data to obtain task resource game candidate data;

[0171] Specifically, from the negotiation game data, each task corresponds to multiple resource combinations (temporary task groups). Retain the top N high-scoring candidate groups (e.g., N = 5) for each task; each candidate group includes resource composition, role configuration, and evaluation scores. Extract and sort the resource response candidate solutions for each task according to the negotiation game data, and retain the high-score task resource combinations that meet the threshold requirements to obtain task resource game candidate data for optimal target selection.

[0172] Perform local optimal target annotation on the task resource game candidate data to obtain local optimal target annotation data;

[0173] Specifically, in the game candidate stage, the system has formed matching candidate relationships between multiple resource combination schemes (teams) and each task. To select the combination with the best performance from them, the system presets a multi-factor weighted objective function to score each resource combination scheme and accordingly annotate the optimal candidate, and calculate as where f(team) is the resource combination score value, representing the comprehensive score of the resource combination scheme under the current task. The higher the value, the better, is the collaborative scoring weight, which determines the degree of importance the system attaches to the collaborative ability between resources (such as role complementarity and parallelism). The value is 0.4, indicating that the collaborative ability of the team directly affects the stability and execution efficiency of the overall scheduling. In multi-resource collaborative tasks, the quality of the collaborative relationship takes precedence over individual performance, s tea m is the collaborative score value, representing the degree of collaboration within the combination in terms of semantics, function, scheduling relationship, etc., is the load balancing weight, representing the degree of importance attached to the balance of load distribution. The value is 0.2, indicating that load balancing helps to avoid resource bottlenecks, but in some high-priority tasks, the balance can be sacrificed appropriately, so the weight is secondary, Var load is the load variance, representing the variance of the current load of each agent within the combination. The smaller the value, the more balanced, is the energy utilization rate weight, representing the degree of importance the system attaches to the average energy efficiency of the combination. The value is 0.2. Energy conservation consideration is an auxiliary indicator, unless the weight needs to be increased in the green manufacturing scenario, E eff is the energy utilization rate, the inverse ratio of the energy consumption required to execute the current task. The larger the value, the more energy-efficient, is the historical success rate weight, representing the weight of the historical success experience of each resource in the combination for executing this type of task. The value is 0.2. The success rate reflects reliability, and its importance is on par with energy efficiency. It is suitable for experience-driven scheduling scenarios, R historyThe historical success rate refers to the average success rate of each resource in the combination performing similar tasks historically. Each task is tagged with a locally optimal candidate label; if the scores of multiple groups are close to the preset tolerance threshold (±ε), they are marked as tied for the best. The candidate data for the task-resource game is scored and sorted according to the preset objective function, and the resource combination plan with the best score or tied for the best is extracted, labeled as the locally optimal objective, and the locally optimal objective labeled data is obtained.

[0174] Based on the locally optimal objective labeled data, the task execution of the factory resource agent data is optimally allocated to obtain the preliminary allocation object data;

[0175] Specifically, the agents in the locally optimal combination are marked as task allocation candidates; the scheduling status of these agents is updated to pending, and the task ID is recorded. The system supports the temporary participation of the same resource in multiple tasks, but conflict filtering is required; a pre-allocation task list is maintained for each resource. The system supports the temporary reuse of resources, that is, it allows the same resource agent to participate in multiple task combinations in the preliminary stage, but it needs to be screened out or the scheduling optimized in the subsequent conflict detection and resource exclusivity evaluation. Therefore, each resource agent will maintain the following allocation-related fields: Agent ID: the unique identifier of the resource; PreAssignmentList: the set of multiple task IDs currently pre-allocated to the resource; CurrentDispatchStatus: set to "Pending". According to the locally optimal objective labeled data, the agents in the optimal task-resource combination are marked as objects to be scheduled, and their pre-allocated tasks are recorded to form the preliminary allocation object data, preparing for task conflict handling.

[0176] Based on the preliminary allocation object data, task mutual exclusion filtering is performed to obtain the negotiated locally optimal data.

[0177] Specifically, the following two types of conflicts are checked: resource time conflicts include agents participating in multiple tasks at the same time; functional role conflicts include the same role cannot be executed repeatedly (e.g., one person cannot control two processes simultaneously). The conflict task groups are sorted by priority, and the high-priority tasks are retained; the conflict tasks are returned to the candidate group, triggering a rollback game label. Task mutual exclusion detection is performed according to the preliminary allocation object data to identify the scheduling conflict tasks caused by resource overlap or role conflict, and conflict filtering is performed through task priority and resource exclusivity rules to retain the conflict-free task-resource combination, generating the negotiated locally optimal data.

[0178] Specifically, the system performs a scheduling conflict detection operation based on the task resource binding information recorded in the preliminary allocation object data. The conflict types mainly include the following two categories. Resource time conflict means that the same resource agent is occupied by multiple tasks simultaneously within an overlapping time period. For example, the same device is called by multiple tasks during the same time period. Functional role conflict means that the same agent undertakes multiple mutually exclusive role tasks simultaneously. For example, the same operator cannot act as the master controller to execute two different processes at the same time. To resolve the above conflicts, the system calculates the urgency level of each conflicting task based on its deadline or remaining schedulable time, and performs a preliminary sorting from high to low according to the urgency. Tasks with a high urgency are preferentially ranked at the front. Deadline ≤ 2 hours → high urgency, deadline ≤ 4 hours → medium urgency, deadline > 4 hours → low urgency; Based on the first-round sorting result, the system checks the scheduling priority level of the tasks (such as level 1 to level 5), and inserts the tasks with a high priority level to the front of the same urgency segment to ensure that critical tasks are executed first. Among the tasks with a high urgency, tasks at level 1 are ranked at the front, tasks at level 3 are ranked in the middle, and tasks at level 5 are ranked at the end; After the first two levels of sorting, the system will also fine-tune the time window looseness of the tasks within the same priority level segment. For tasks with a tight time window (small time elasticity), they will be ranked in front of tasks with a loose window to reduce the risk of scheduling failure. Delayable time ≤ 30 minutes → tight time window → one position forward, delayable time ≥ 2 hours → loose time window → one position backward. Preferentially retain the task resource combinations with a relatively high scheduling importance and a small conflict impact; For tasks with a relatively low score or non-critical tasks among the conflicting tasks, the system removes them from the current scheduling set and returns them to the original task resource candidate set. The returned task combination will be set to the game rollback state, and the system can re-enter it into the candidate pool for task reallocation in subsequent scheduling rounds. This process is recorded with a game rollback flag inside the model and is linked to the game state management module. The system outputs the conflict-free task resource binding set after retention, that is, the negotiated locally optimal data, as the basic input for the next resource conflict detection or final allocation plan confirmation.

[0179] Preferably, for implementing the digital factory full-process collaboration method as described above, the digital factory full-process collaboration system includes:

[0180] A process parameter analysis module, configured to obtain production decision data and perform process parameter analysis based on the production decision data to obtain process parameter data;

[0181] A resource agent construction module, configured to construct factory resource agents based on the process parameter data through a preset digital factory model to obtain factory resource agent data;

[0182] The multi-agent negotiation and task allocation module is used to perform multi-agent negotiation based on the factory resource agent data to obtain the digital workshop task allocation data;

[0183] The industrial Internet of Things monitoring and feedback module is used to perform industrial Internet of Things monitoring based on the digital workshop task allocation data to obtain the real-time industrial Internet of Things data for full-process collaborative auxiliary operation of the digital factory.

[0184] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended application documents rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be encompassed by the present invention.

[0185] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A full-process collaboration method for a digital factory, characterized in that, It includes the following steps: Step S1: Obtain production decision data, and perform process parameter analysis based on the production decision data to obtain process parameter data; Step S2: Construct factory resource agents according to the process parameter data through a preset digital factory model to obtain factory resource agent data; Step S3: Conduct multi-agent negotiation based on the factory resource agent data to obtain digital workshop task allocation data; Step S4: Conduct industrial Internet of Things monitoring based on the digital workshop task allocation data to obtain real-time industrial Internet of Things data for digital factory full-process collaborative auxiliary operations.

2. The method according to claim 1, wherein Specifically, step S1 is as follows: Obtain production decision data; Extract the process flow from the production decision data to obtain process flow data; Perform process parameter analysis based on the process flow data to obtain process parameter data, where the process parameter data includes required equipment types, process sequences, material usage, man-hour estimates, energy requirements, and environmental constraints.

3. The method according to claim 1, wherein Specifically, step S2 is as follows: Instantiate factory resource type agents according to the process parameter data through a preset digital factory model to obtain factory resource type agent data, where the factory resource type agent data includes production line agent data, equipment agent data, personnel agent data, and material agent data; Configure the attributes of the factory resource type agent data to obtain agent attribute data, where the attribute configuration includes status information configuration, willingness function configuration, and policy space configuration; Construct a task collaboration relationship graph based on the process parameter data for the agent attribute data to obtain factory resource agent data.

4. The method according to claim 1, wherein Specifically, step S3 is as follows: Perform negotiation initialization based on the factory resource agent data to obtain preliminary negotiation data; Conduct task games based on the preliminary negotiation data to obtain negotiation game data; Extract the initial local optimum based on the negotiation game data to obtain negotiation local optimum data; Detect global resource conflicts based on the negotiation local optimum data to obtain global resource conflict data; Annotate and limit the negotiation game data based on the global resource conflict data to obtain negotiation annotation agreement data; Extract the restricted local optimum based on the negotiation annotation agreement data to obtain digital workshop task allocation data.

5. The method according to claim 4, wherein Among them, the negotiation initialization is specifically as follows: Perform agent semantic graph mapping based on the factory resource agent data to obtain cross-agent shared semantic data; Perform willingness function semantic embedding on the cross-agent shared semantic data to obtain intelligent task context data; Load the digital twin mirror based on the factory resource agent data to obtain agent digital mirror data; Associate and integrate the intelligent task context data and the agent digital mirror data to obtain preliminary negotiation data.

6. The method according to claim 5, characterized in that, Among them, the agent semantic graph mapping is performed through a preset shared semantic understanding graph model. The construction steps of the shared semantic understanding graph model include: Obtain the key term data corresponding to the factory resource agent data; Align the terms based on the key term data and a preset domain ontology library to obtain term alignment data, where the term alignment includes concept layer alignment, example layer alignment, and attribute layer alignment; Extract positive and negative sample pairs according to the term alignment data to obtain positive and negative sample pair data; Perform cross-agent semantic training on the positive and negative sample pair data to obtain a shared semantic understanding graph model, where the cross-agent semantic training is jointly trained through a Transformer structure encoding network and a bidirectional gated recurrent unit encoder.

7. The method according to claim 5, characterized in that Among them, the semantic embedding of the willingness function is specifically: Calculate the attention of the factory resource agent data according to the cross-agent shared semantic data to obtain the agent attention weight data; Perform weighted aggregation on the factory resource agent data according to the agent attention weight data to obtain the semantic perception feature data; Calculate the semantic prior distribution according to the cross-agent shared semantic data to obtain the semantic prior distribution data; Obtain the current state data of the agent corresponding to the cross-agent shared semantic data, and fit the current state data of the agent with a conditional Gaussian model to obtain the agent observation model; Perform Bayesian posterior inference according to the semantic prior distribution data and the agent observation model to obtain the task preference probability data; Calculate the expected value weight according to the semantic perception feature data and the task preference probability data to obtain the intelligent task context data.

8. The method according to claim 4, wherein Among them, the task game is specifically: Extract the task resource candidate mapping according to the preliminary negotiation data to obtain the task resource candidate mapping data; Perform a willingness mapping on the task resource candidate mapping data according to the factory resource agent data to obtain the game response matrix data; Perform a scenario semantic resource response according to the game response matrix data to obtain the scenario semantic resource response data; Make a task assistance request to the factory resource agent data according to the scenario semantic resource response data to obtain the agent role response bid data; Perform a collaborative ability evaluation according to the agent role response bid data to obtain the collaborative ability evaluation data, where the collaborative ability evaluation includes a timing consistency evaluation and a semantic preference conflict evaluation; Generate a temporary task group according to the collaborative ability evaluation data to obtain the negotiation game data.

9. The method according to claim 4, wherein Among them, the initial local optimum extraction is specifically: Extract the task resource game candidates according to the negotiation game data to obtain the task resource game candidate data; Perform local optimum target annotation on the task resource game candidate data to obtain the local optimum target annotation data; Perform an optimal task execution allocation on the factory resource agent data according to the local optimum target annotation data to obtain the preliminary allocation object data; Perform task mutual exclusion filtering according to the preliminary allocation object data to obtain the negotiated local optimum data.

10. A full-process collaborative system for a digital factory, characterized in that, For implementing the digital factory full-process collaboration method as described in claim 1, the digital factory full-process collaboration system includes: A process parameter analysis module, configured to obtain production decision data and perform process parameter analysis according to the production decision data to obtain process parameter data; A resource agent construction module, configured to construct a factory resource agent according to the process parameter data through a preset digital factory model to obtain factory resource agent data; A multi-agent negotiation and task allocation module, configured to perform multi-agent negotiation according to the factory resource agent data to obtain digital workshop task allocation data; Industrial Internet of Things Monitoring and Feedback Module, which is used to perform industrial Internet of Things monitoring based on the digital workshop task allocation data, obtain real-time industrial Internet of Things data, and carry out collaborative auxiliary operations for the entire process of digital factory.

Images