A large-model-based intelligent scheduling and dynamic order insertion optimization system and method for MES.

The MES intelligent scheduling system based on a large model solves the problems of manual dependence and low efficiency in existing MES scheduling systems, and realizes efficient and adaptive production scheduling and order insertion optimization, thereby improving production efficiency and on-time delivery rate.

CN123070525APending Publication Date: 2026-11-10ANHUI TELECOMM PLANNING & DESIGNING
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611074754.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-20
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing MES scheduling systems suffer from problems such as heavy reliance on manual intervention, long processing time, susceptibility to errors, high subjectivity, slow response to order insertions, high cost of rescheduling, passive exception handling, high operational threshold, and inability to interact with natural language.

Method used

The system employs a large-scale model-based MES intelligent scheduling system, which integrates a data acquisition and fusion module, an intelligent scheduling optimization module, a dynamic order insertion processing module, an anomaly identification and rescheduling module, and an MES command issuance and feedback module. By combining a large language model, an industrial scheduling knowledge base, and hard rule verification, it achieves natural language interaction, global optimization, and dynamic decision-making.

Benefits of technology

It improved scheduling efficiency, shortened the response time for order insertions, increased the on-time delivery rate of orders, reduced reliance on manual labor, reduced resource waste, and achieved minute-level disturbance response and adaptive adjustment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123070525A_ABST
    Figure CN123070525A_ABST
Patent Text Reader

Abstract

This invention discloses a large-model-based intelligent scheduling and dynamic order insertion optimization system and method for MES (Manufacturing Execution System), relating to the fields of industrial informatization and intelligent manufacturing technology. It aims to address the technical pain points of traditional MES scheduling constraints, such as reliance on manual processing, delayed emergency order insertion response, passive handling of production anomalies, and high operational interaction barriers. The system includes a data acquisition and fusion module for data interaction via an industrial data bus, a large-model inference engine, an intelligent scheduling optimization module, a dynamic order insertion processing module, an anomaly identification and reordering module, and an MES command issuance and feedback module. The method generates a standardized production digital twin dataset through multi-source data cleaning and fusion. A large model, enhanced through three layers of domain fine-tuning, knowledge base mounting, and hard rule verification, completes production constraint parsing, constructs a multi-objective optimization model to generate a globally optimal schedule, significantly improving scheduling efficiency and delivery rate, reducing manual reliance, and achieving intelligent, flexible, and continuously self-evolving production scheduling.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of industrial informatization and intelligent manufacturing technology, specifically to a large-model-based MES intelligent scheduling and dynamic order insertion optimization system and method. Background Technology

[0002] Manufacturing Execution System (MES) is the core of enterprise shop floor production management. Scheduling is one of the most critical modules of MES, directly determining capacity utilization, on-time delivery rate, and production costs.

[0003] The existing MES scheduling has obvious pain points:

[0004] Complex constraints and heavy reliance on manual scheduling: Scheduling needs to consider multiple constraints such as order priority, delivery date, equipment load, changeover time, material availability, and personnel skills. Manual scheduling is time-consuming, error-prone, and highly subjective.

[0005] Slow order insertion response and high reordering costs: When an emergency order is inserted, manual recalculation, order adjustment and capacity verification take several hours, which can easily lead to delivery delays and waste of resources;

[0006] The current system is passive in handling anomalies and has low efficiency in rescheduling: When equipment fails, materials are delayed, or personnel are absent, the existing system can only issue alarms and cannot quickly reschedule globally. It relies on the experience of the scheduler and is prone to causing production line stoppages.

[0007] Unable to interact with natural language and with a high operational threshold: Dispatchers need to operate in a complex interface and cannot directly issue instructions such as inserting orders, rescheduling, or pausing using spoken language or text, resulting in high learning costs and low efficiency.

[0008] With the advancement of "intelligent transformation and digital transformation" in the manufacturing industry, enterprises urgently need a new generation of MES scheduling solutions that are intelligent, efficient, adaptive, and support natural language interaction. Summary of the Invention

[0009] To address the shortcomings of existing technologies, the present invention aims to provide a large-scale model-based intelligent scheduling and dynamic order insertion optimization system and method for MES. By leveraging the powerful natural language understanding, multi-constraint reasoning, global optimization, and dynamic decision-making capabilities of large-scale models, the system improves scheduling efficiency, shortens order insertion response time, increases on-time order delivery rate, reduces reliance on manual labor, and minimizes resource waste.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] A large-model-based MES intelligent scheduling and dynamic order insertion optimization system includes a data acquisition and fusion module that realizes data interaction through an industrial data bus, a large-model inference engine, an intelligent scheduling optimization module, a dynamic order insertion processing module, an anomaly identification and reordering module, and an MES command issuance and feedback module.

[0012] The data acquisition and fusion module is used to collect multi-source production data and perform cleaning, format unification and fusion processing, output a standardized production digital twin dataset and transmit it to the large model inference engine.

[0013] The large model inference engine is built on the basis of a general large language model, and is constructed by domain fine-tuning, mounting an industrial scheduling knowledge base and setting an industrial hard rule verification layer. It is used to receive the production digital twin dataset and complete production constraint parsing and scheduling optimization inference.

[0014] The intelligent scheduling optimization module is used to receive the inference results of the large model, combine them with global production constraints to construct scheduling optimization targets, generate an initial global optimal scheduling plan and provide a visual display.

[0015] The dynamic order insertion processing module is used to receive and parse natural language order insertion instructions, and after completing the order insertion feasibility verification, send a reordering trigger signal to the intelligent scheduling optimization module to trigger global scheduling recalculation, or output alternative solutions for infeasible scenarios.

[0016] The anomaly identification and reordering module is used to monitor production operation data in real time and identify abnormal events. After assessing the impact of the anomaly, it sends a reordering trigger signal to the intelligent scheduling optimization module.

[0017] The MES instruction issuance and feedback module is used to receive the scheduling plan and issue scheduling execution instructions, while simultaneously feeding back the MES execution feedback data to the large model inference engine, forming a closed loop for iterative optimization of the scheduling strategy.

[0018] Furthermore, the multi-source production data collected by the data acquisition and fusion module includes the following three categories:

[0019] MES basic data includes order information, work orders, BOM, process routes, equipment ledgers, capacity calendar and changeover rules;

[0020] Real-time production dynamic data, including equipment status, overall equipment efficiency, material inventory and kitting rate, personnel scheduling and skill information, and work-in-process progress;

[0021] External collaborative data includes procurement and delivery plans, warehouse outbound plans, quality inspection results, and energy consumption data.

[0022] Furthermore, the domain fine-tuning of the large model inference engine refers to the instruction fine-tuning of the general large language model based on the enterprise's historical MES work orders, process routes, historical scheduling schemes, and exception handling cases.

[0023] The industrial scheduling knowledge base stores industry standards, enterprise process specifications, equipment capacity limits, changeover cost models, delivery priority strategies, and material substitution relationships. The large model inference engine retrieves relevant knowledge from the industrial scheduling knowledge base through vector retrieval, providing support for the inference process. The industrial hard rule verification layer is used to verify the capacity, equipment load, material completeness, and process compliance of the large model output results, filtering out invalid outputs that do not conform to actual production.

[0024] Furthermore, the intelligent scheduling optimization module uses maximizing delivery time fulfillment rate, maximizing capacity utilization, minimizing changeover time, and minimizing overall cost as the scheduling optimization objective function. The output initial scheduling plan includes work order processing sequence, equipment allocation scheme, start and completion time, changeover arrangement, and personnel allocation scheme. The visualization display includes a scheduling Gantt chart, equipment load chart, and material kitting analysis chart, and supports manual fine-tuning and confirmation of the scheduling plan.

[0025] Furthermore, the dynamic order insertion processing module supports spoken order insertion command input in both voice and text formats, and automatically parses and extracts order number, priority, delivery date, process requirements, and special constraint parameters; the order insertion feasibility verification dimensions include capacity margin, material availability status, equipment availability time, and process compatibility; the alternative solutions generated in infeasible scenarios include overtime production, outsourcing processing, and optimization suggestions for adjusting the delivery dates of other orders.

[0026] Furthermore, the anomaly identification and reordering module can identify production anomalies including equipment failure and shutdown, material shortage, delayed procurement and delivery, staff absence, quality anomalies, and energy interruptions. The anomaly identification and reordering module automatically classifies anomaly types, assesses the scope of impact, and predicts recovery time. When an anomaly affects core production capacity or order delivery dates, a global reordering is triggered, prioritizing the delivery of high-priority orders during the reordering process.

[0027] Furthermore, the MES instruction issuance and feedback module connects to an external MES system via an API interface. The issued instructions include work order scheduling instructions, work order adjustment instructions, equipment allocation instructions, personnel scheduling instructions, and line changeover instructions. The received execution feedback data includes work order start status, completion progress, abnormal alarms, and execution results. The feedback data is used to iteratively optimize the scheduling inference strategy and decision accuracy of the large model.

[0028] A method for intelligent scheduling and dynamic order insertion optimization in MES based on a large model includes the following steps:

[0029] S1. Data Acquisition and Fusion: Collect production data from multiple sources, clean and unify the format, and then fuse them to generate a standardized production digital twin dataset.

[0030] S2, Large Model Enhanced Reasoning: Input the production digital twin dataset into a general large language model that has been enhanced by domain fine-tuning, knowledge base mounting and industrial hard rule verification to complete production constraint parsing and scheduling optimization reasoning;

[0031] S3. Initial Global Scheduling Generation: Based on the inference results of the large model, combined with global production constraints, a scheduling optimization target is constructed, an initial global optimal scheduling plan is generated and visualized, and after confirmation, it is sent to the MES system for execution.

[0032] S4. Dynamic order insertion processing: During the scheduling process, if a natural language order insertion instruction is received, the order insertion constraint parameters are parsed and extracted and a feasibility check is performed. If the check passes, the global scheduling is automatically recalculated. If the check fails, an alternative solution for the infeasible scenario is generated.

[0033] S5. Anomaly Identification and Rescheduling: During the scheduling process, production operation data is monitored in real time to identify production anomalies and assess their impact. When the rescheduling trigger conditions are met, a global rescheduling is triggered to generate an adjusted scheduling plan and send it out for execution.

[0034] S6. Command Issuance and Closed-Loop Optimization: The scheduling plan is issued to the MES system for execution, the execution feedback data of MES is collected and fed back to the large model, and the scheduling inference strategy is iteratively optimized.

[0035] Furthermore, in step S2, the domain fine-tuning is the instruction fine-tuning of the general large language model based on the enterprise's historical MES scheduling schemes and anomaly handling cases; the knowledge base mounting is the access to a vector knowledge base storing industrial scheduling rules and experience, and the real-time recall of relevant knowledge during the reasoning process; the industrial hard rule verification is the capacity verification, equipment load verification, material completeness verification and process compliance verification of the large model reasoning results.

[0036] Furthermore, in step S3, the scheduling optimization objectives include maximizing delivery time fulfillment rate, maximizing capacity utilization, minimizing changeover time, and minimizing overall cost; the generated initial scheduling plan includes work order processing sequence, equipment allocation, start and finish times, changeover arrangements, and personnel allocation schemes; the visualization display formats include Gantt charts, equipment load charts, and material kitting analysis charts, supporting manual fine-tuning and confirmation of the scheduling plan.

[0037] When this large-model-based MES intelligent scheduling and dynamic order insertion optimization system is running, the data acquisition and fusion module first performs 3σ criterion denoising, min-max normalization, and confidence-weighted fusion on three types of production data to generate a standardized production digital twin dataset, providing a unified and reliable data foundation for end-to-end computation. The large-model inference engine adopts a three-layer enhanced architecture of "domain fine-tuning + vector knowledge base + industrial hard rule verification layer" to complete unstructured constraint parsing, domain knowledge retrieval, and physical constraint fallback, balancing inference flexibility and industrial feasibility.

[0038] After receiving structured constraints, the intelligent scheduling optimization module constructs a linear weighted multi-objective function based on delivery date, capacity, line changeover, and cost. It then combines genetic algorithms and heuristic rules to achieve global optimization, outputting an initial scheduling plan and supporting visual manual fine-tuning. Dynamic order insertion processing and anomaly identification and reordering form a dual dynamic branch, respectively parsing natural language order insertion instructions and identifying production anomalies. After multi-dimensional verification and impact assessment, a global reordering is triggered, increasing the weight of high-priority orders to achieve adaptive adjustment and achieving minute-level disturbance response.

[0039] Finally, the MES instruction issuance and feedback module completes the issuance of scheduling instructions and the collection of execution data. Through deviation calculation, real production cases are added to the sample library, and incremental fine-tuning is performed on the large model to continuously improve the accuracy of scheduling inference and form a self-evolving optimization loop.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] 1. Significantly improved scheduling efficiency: Manual scheduling is now automatically generated within 10 minutes, reducing the time required for manual scheduling from several hours;

[0042] 2. Rapid Order Response: Urgent orders are processed within 10 minutes for parsing, verification, reordering, and distribution, improving on-time delivery rate by 20%.

[0043] 3. Intelligent anomaly handling: Automatic identification and rapid rescheduling of production anomalies reduce unplanned downtime by 40% and increase capacity utilization by 15%;

[0044] 4. Significantly reduced reliance on manual scheduling: Reduces manual scheduling, adjustment, and verification work by more than 70%, reduces dependence on experienced dispatchers, and lowers labor costs; Attached Figure Description

[0045] Figure 1 This is a flowchart of the workflow of a large-model-based MES intelligent scheduling and dynamic order insertion optimization system of the present invention. Detailed Implementation

[0046] The technical solutions of the present invention will now be described in detail with reference to the accompanying drawings.

[0047] like Figure 1 As shown, a large-model-based MES intelligent scheduling and dynamic order insertion optimization system includes a data acquisition and fusion module that realizes data interaction through an industrial data bus, a large-model inference engine, an intelligent scheduling optimization module, a dynamic order insertion processing module, an anomaly identification and reordering module, and an MES command issuance and feedback module.

[0048] The data acquisition and fusion module is used to collect multi-source production data and perform cleaning, format unification and fusion processing, output a standardized production digital twin dataset and transmit it to the large model inference engine.

[0049] The large model inference engine is built on the basis of a general large language model, and is constructed by domain fine-tuning, mounting an industrial scheduling knowledge base and setting an industrial hard rule verification layer. It is used to receive the production digital twin dataset and complete production constraint parsing and scheduling optimization inference.

[0050] The intelligent scheduling optimization module is used to receive the inference results of the large model, combine them with global production constraints to construct scheduling optimization targets, generate an initial global optimal scheduling plan and provide a visual display.

[0051] The dynamic order insertion processing module is used to receive and parse natural language order insertion instructions, and after completing the order insertion feasibility verification, send a reordering trigger signal to the intelligent scheduling optimization module to trigger global scheduling recalculation, or output alternative solutions for infeasible scenarios.

[0052] The anomaly identification and reordering module is used to monitor production operation data in real time and identify abnormal events. After assessing the impact of the anomaly, it sends a reordering trigger signal to the intelligent scheduling optimization module.

[0053] The MES instruction issuance and feedback module is used to receive the scheduling plan and issue scheduling execution instructions, while simultaneously feeding back the MES execution feedback data to the large model inference engine, forming a closed loop for iterative optimization of the scheduling strategy.

[0054] Furthermore, the multi-source production data collected by the data acquisition and fusion module includes the following three categories:

[0055] MES basic data includes order information, work orders, BOM, process routes, equipment ledgers, capacity calendar and changeover rules;

[0056] Real-time production dynamic data, including equipment status, overall equipment efficiency, material inventory and kitting rate, personnel scheduling and skill information, and work-in-process progress;

[0057] External collaborative data includes procurement and delivery plans, warehouse outbound plans, quality inspection results, and energy consumption data.

[0058] Furthermore, the domain fine-tuning of the large model inference engine refers to the instruction fine-tuning of the general large language model based on the enterprise's historical MES work orders, process routes, historical scheduling schemes, and exception handling cases.

[0059] The industrial scheduling knowledge base stores industry standards, enterprise process specifications, equipment capacity limits, changeover cost models, delivery priority strategies, and material substitution relationships. The large model inference engine retrieves relevant knowledge from the industrial scheduling knowledge base through vector retrieval, providing support for the inference process. The industrial hard rule verification layer is used to verify the capacity, equipment load, material completeness, and process compliance of the large model output results, filtering out invalid outputs that do not conform to actual production.

[0060] Furthermore, the intelligent scheduling optimization module uses maximizing delivery time fulfillment rate, maximizing capacity utilization, minimizing changeover time, and minimizing overall cost as the scheduling optimization objective function. The output initial scheduling plan includes work order processing sequence, equipment allocation scheme, start and completion time, changeover arrangement, and personnel allocation scheme. The visualization display includes a scheduling Gantt chart, equipment load chart, and material kitting analysis chart, and supports manual fine-tuning and confirmation of the scheduling plan.

[0061] Furthermore, the dynamic order insertion processing module supports spoken order insertion command input in both voice and text formats, and automatically parses and extracts order number, priority, delivery date, process requirements, and special constraint parameters; the order insertion feasibility verification dimensions include capacity margin, material availability status, equipment availability time, and process compatibility; the alternative solutions generated in infeasible scenarios include overtime production, outsourcing processing, and optimization suggestions for adjusting the delivery dates of other orders.

[0062] Furthermore, the anomaly identification and reordering module can identify production anomalies including equipment failure and shutdown, material shortage, delayed procurement and delivery, staff absence, quality anomalies, and energy interruptions. The anomaly identification and reordering module automatically classifies anomaly types, assesses the scope of impact, and predicts recovery time. When an anomaly affects core production capacity or order delivery dates, a global reordering is triggered, prioritizing the delivery of high-priority orders during the reordering process.

[0063] Furthermore, the MES instruction issuance and feedback module connects to an external MES system via an API interface. The issued instructions include work order scheduling instructions, work order adjustment instructions, equipment allocation instructions, personnel scheduling instructions, and line changeover instructions. The received execution feedback data includes work order start status, completion progress, abnormal alarms, and execution results. The feedback data is used to iteratively optimize the scheduling inference strategy and decision accuracy of the large model.

[0064] A method for intelligent scheduling and dynamic order insertion optimization in MES based on a large model includes the following steps:

[0065] S1. Data Acquisition and Fusion: Collect production data from multiple sources, clean and unify the format, and then fuse them to generate a standardized production digital twin dataset.

[0066] Production data is collected in parallel across three types of data sources, covering all dimensions:

[0067] MES basic data: Static basic data such as order information, work orders, BOM material list, process route, equipment ledger, capacity calendar, and changeover rules are extracted from the enterprise's MES system. The data update cycle is once a day.

[0068] Real-time production dynamic data: Dynamic data such as equipment status, overall equipment efficiency (OEE), material inventory and kitting rate, personnel scheduling and skill information, and work-in-process (WIP) progress are collected through industrial IoT gateways at a frequency of once per minute.

[0069] External collaborative data: Through interfaces with ERP, WMS, and QMS systems, the system integrates with procurement and delivery plans, warehouse outbound plans, quality inspection results, and energy consumption data, with a data synchronization cycle of once per hour.

[0070] Outlier identification and removal are performed on the collected raw data. The 3σ criterion is used to identify outliers in the numerical data.

[0071] ;

[0072] In the formula: For the first The numerical values ​​of the original data; This is the arithmetic mean of the data samples in this category; This represents the standard deviation of this type of data sample.

[0073] If the data satisfies the above formula, it is determined to be an outlier, and the linear interpolation result of adjacent data in the same time period is used to replace it to ensure data continuity.

[0074] Perform min-max normalization on numerical data with different dimensions and value ranges, mapping to... Range, unified data format:

[0075] ;

[0076] In the formula: The normalized value; The original data values; It is the minimum value in this type of data sample; This is the maximum value among the data samples of this type.

[0077] At the same time, non-numerical data (such as equipment status and process name) are uniformly encoded and mapped to convert them into structured enumeration values.

[0078] The three types of data sources are aligned in time dimension based on timestamps, and spatial dimension association is performed using device / work order as the primary key. A standardized production digital twin dataset is generated through a confidence-weighted fusion algorithm. :

[0079] ;

[0080] In the formula: These are the merged data indicator values; The total number of data sources participating in the integration; For the first The confidence weights of each data source satisfy the following conditions: Based on the accuracy and real-time performance of the data source; For the first Standardized values ​​of metrics corresponding to each data source. The final generated production digital twin dataset. It includes a basic data subset, a real-time dynamic subset, and an external collaborative subset, which serve as inputs for large model inference.

[0081] S2, Large Model Enhanced Reasoning: Input the production digital twin dataset into a general large language model that has been enhanced by domain fine-tuning, knowledge base mounting and industrial hard rule verification to complete production constraint parsing and scheduling optimization reasoning;

[0082] Furthermore, in step S2, the domain fine-tuning is the instruction fine-tuning of the general large language model based on the enterprise's historical MES scheduling schemes and anomaly handling cases; the knowledge base mounting is the access to a vector knowledge base storing industrial scheduling rules and experience, and the real-time recall of relevant knowledge during the reasoning process; the industrial hard rule verification is the capacity verification, equipment load verification, material completeness verification and process compliance verification of the large model reasoning results.

[0083] To produce digital twin datasets Input a domain-fine-tuned general-purpose large language model, which performs semantic parsing of unstructured constraints and outputs a structured set of constraints. The domain fine-tuning adopts the instruction fine-tuning method. The training samples are the enterprise's historical MES work orders, process routes, historical scheduling schemes and anomaly handling cases. The loss function of the fine-tuning process adopts cross-entropy loss, and the constraint model output conforms to the industrial scheduling semantic specification.

[0084] During the inference process, the large model transforms the current scheduling scenario into query vectors. By using a cosine similarity algorithm, the top-K knowledge entries with the highest matching degree are retrieved from the industrial scheduling knowledge base, providing domain knowledge support for the reasoning process.

[0085] ;

[0086] In the formula: For query vector and the first The cosine similarity of the knowledge base vectors ranges from -1 to 1, with higher values ​​indicating stronger matching. This is the query vector for the current scheduling scenario; For the first in the knowledge base Vector representation of knowledge points; , Let be the magnitudes of the query vector and the knowledge base vector, respectively.

[0087] The knowledge required for recalls includes industry standards, enterprise process specifications, equipment capacity limits, changeover cost models, delivery priority strategies, and material substitution relationships.

[0088] After the large model outputs preliminary inference results, they are input into the industrial hard rule verification layer, where four types of mandatory verification are executed sequentially, with the corresponding mathematical constraints as follows:

[0089] Equipment capacity constraints:

[0090] ;

[0091] In the formula: To be assigned to the device A collection of work orders; For work orders In the equipment Processing time; For equipment The maximum available capacity duration within the planned period.

[0092] Process sequence constraints:

[0093] ;

[0094] In the formula: For work orders The The start time of each process step; For work orders The Completion time of each process step. Material kitting constraints: In the formula: This is the set of work orders that commence within the planned period; For work orders Production quantity; For work orders per unit of product to materials Consumption amount; For materials The total current available inventory. Process compliance constraints: In the formula: For equipment Supported process set; For work orders Required process set; symbols This indicates an inclusion relationship, meaning the equipment must cover all the process requirements of the work order.

[0095] If the inference result violates any hard rule, the verification fails, and the process returns to the large model to re-infer until all output results satisfy the hard rule constraints.

[0096] S3. Initial Global Scheduling Generation: Based on the inference results of the large model, combined with global production constraints, a scheduling optimization target is constructed, an initial global optimal scheduling plan is generated and visualized, and after confirmation, it is sent to the MES system for execution.

[0097] Furthermore, in step S3, the scheduling optimization objectives include maximizing delivery time fulfillment rate, maximizing capacity utilization, minimizing changeover time, and minimizing overall cost; the generated initial scheduling plan includes work order processing sequence, equipment allocation, start and finish times, changeover arrangements, and personnel allocation schemes; the visualization display formats include Gantt charts, equipment load charts, and material kitting analysis charts, supporting manual fine-tuning and confirmation of the scheduling plan.

[0098] Based on the constraint set of the large model analysis, a multi-objective scheduling optimization model is constructed. The multi-objective problem is transformed into a single-objective optimization problem using a linear weighted summation method. The overall objective function is:

[0099] ;

[0100] In the formula: To comprehensively optimize the target value, the smaller the value, the better the scheduling scheme; The weighting coefficients for the four sub-objectives—delivery time, capacity, line changeover, and cost—are respectively, satisfying the following conditions: It can be adjusted according to the company's production strategy; These are four sub-objective functions, specifically defined as follows:

[0101] Sub-objective 1: Maximize on-time delivery rate

[0102] ;

[0103] In the formula: This represents the total number of work orders within the planned period. For work orders The priority coefficient of the work order is positively correlated with the priority of the work order; For work orders The actual completion time; For work orders The required delivery time.

[0104] Sub-objective 2: Maximize capacity utilization

[0105] ;

[0106] In the formula: This represents the total number of equipment in the production workshop. Equipment within the planned period Total idle time; For equipment within the planned period Total available time.

[0107] Sub-objective 3: Minimize line switching time

[0108] ;

[0109] In the formula: This represents the total number of product categories. For equipment From the product Switch to product The time required for line switching; For 0-1 decision variables, if the equipment The adjacent processing task above is the product. Subsequent products If the result is 1, then take 1; otherwise, take 0.

[0110] Sub-objective 4: Minimize overall cost

[0111] ;

[0112] In the formula: For equipment The processing cost per unit time; The overall cost per unit of line change time; For work orders Total material cost.

[0113] The large model combines genetic algorithms and heuristic rules to solve the above multi-objective optimization model and outputs an initial global optimal scheduling plan, including: work order processing sequence, equipment allocation scheme, start and finish times of each process, line changeover arrangement, and personnel allocation scheme.

[0114] The system automatically generates three types of visualization charts based on the scheduling plan:

[0115] 1. Scheduling Gantt Chart: With time as the horizontal axis and equipment as the vertical axis, it displays the processing time and progress of each work order;

[0116] 2. Equipment Load Diagram: Displays the load rate distribution of each piece of equipment and identifies bottleneck equipment;

[0117] 3. Material availability analysis chart: Displays the material availability status and material shortage warnings for each work order.

[0118] Schedulers can manually fine-tune the scheduling plan by dragging and dropping. After the fine-tuning, hard rule verification is automatically triggered, and the plan can only be confirmed to take effect after the verification is passed.

[0119] S4. Dynamic order insertion processing: During the scheduling process, if a natural language order insertion instruction is received, the order insertion constraint parameters are parsed and extracted and a feasibility check is performed. If the check passes, the global scheduling is automatically recalculated. If the check fails, an alternative solution for the infeasible scenario is generated.

[0120] It supports both voice and text input for spoken order insertion commands. Through the named entity recognition capability of the large model, it automatically parses and extracts key parameters for order insertion from the commands, including: order number, product type, production quantity, priority level, required delivery date, special process requirements, and constraints, and outputs a structured set of order insertion parameters.

[0121] The feasibility of the order insertion request was verified from four dimensions:

[0122] (1) Verification of capacity margin

[0123] Calculate the remaining available capacity of each piece of equipment:

[0124] ;

[0125] In the formula: For equipment The remaining available capacity duration; For equipment within the planned period Total available time; For equipment The total processing time for the work orders that have already been scheduled.

[0126] If the processing time of the inserted work order on the corresponding equipment is less than or equal to the remaining capacity of all associated equipment, then the capacity dimension verification passes.

[0127] (2) Verification of material completeness

[0128] Calculate the completeness rate of materials corresponding to the supplementary work order:

[0129] ;

[0130] In the formula: For materials The completeness rate; For materials Current available inventory; The production quantity for the inserted work order; For inserting work orders, the unit product is paired with the material. Consumption amount.

[0131] If the completeness rate of all related materials If so, the material dimension verification passes.

[0132] (3) Equipment availability period verification

[0133] Verify whether there is a conflict between the planned processing time of the inserted work order and the existing schedule of the equipment. If there is an idle time period that can be inserted, the equipment time period dimension verification passes.

[0134] (4) Process compatibility verification

[0135] Verify whether the process requirements of the inserted product match the process capabilities of the existing equipment. If the matching degree reaches the threshold, the process dimension verification is passed.

[0136] If all four dimensions pass, the insertion of an order is deemed feasible; otherwise, it is deemed infeasible.

[0137] After the insertion work order passes the verification, a rescheduling trigger signal is sent to the intelligent scheduling optimization module. The priority coefficient of the insertion work order is automatically increased, the multi-objective scheduling optimization model is called again to perform a global solution, and an updated scheduling plan is generated and sent for execution.

[0138] If inserting an order is not feasible, automatically generate three alternative solutions, calculate the corresponding costs, and output the optimal suggestion:

[0139] 1. Overtime production plan: The cost calculation formula is as follows:

[0140] ;

[0141] in For the required overtime hours, Cost per unit of overtime;

[0142] 2. Outsourcing processing plan: The cost calculation formula is as follows:

[0143] ;

[0144] in This refers to the cost of outsourcing processing per unit of product.

[0145] 3. Delivery date adjustment plan: The cost calculation formula is as follows:

[0146] ;

[0147] in The cost of breach of contract per unit of delay time. This refers to the delivery time that needs to be adjusted.

[0148] The system outputs a cost comparison of the three solutions and recommends the solution with the lowest cost.

[0149] S5. Anomaly Identification and Rescheduling: During the scheduling process, production operation data is monitored in real time to identify production anomalies and assess their impact. When the rescheduling trigger conditions are met, a global rescheduling is triggered to generate an adjusted scheduling plan and send it out for execution.

[0150] Real-time collection of production operation data, comparison with normal baseline thresholds, and automatic identification of six types of production anomalies: equipment failure and shutdown, material shortage, delayed procurement and delivery, staff absence, quality anomalies, and energy interruption, and automatic classification and labeling of anomaly types.

[0151] Assess the scope and extent of the impact of the identified anomalies:

[0152] (1) Assessment of capacity loss

[0153] ;

[0154] In the formula: This refers to the amount of production capacity lost due to the anomaly. This refers to the estimated duration of recovery from an abnormal situation. For abnormally associated devices Productivity per unit time.

[0155] (2) Delivery delay assessment

[0156] ;

[0157] In the formula: For work orders The estimated delay time; Work orders affected by the anomaly The estimated completion time; For work orders The original planned delivery date.

[0158] Set a reordering trigger threshold. A reordering trigger signal will be sent to the intelligent scheduling optimization module when any of the following conditions are met:

[0159] 1. Core equipment failure, expected to last more than 30 minutes;

[0160] 2. The estimated delay time for high-priority orders exceeds a preset threshold;

[0161] 3. Material shortages prevent three or more work orders from starting normally.

[0162] After triggering a global reordering, the intelligent scheduling optimization module adjusts the objective function weights, amplifies the delivery weight coefficients of high-priority work orders, prioritizes the delivery of high-priority orders during the reordering process, re-solves the multi-objective optimization model, generates an adjusted scheduling plan, and sends it to the MES system for execution.

[0163] S6. Command Issuance and Closed-Loop Optimization: The scheduling plan is issued to the MES system for execution, the execution feedback data of MES is collected and fed back to the large model, and the scheduling inference strategy is iteratively optimized.

[0164] The final confirmed scheduling plan is transformed into structured instructions that the MES system can recognize, including: work order scheduling instructions, work order adjustment instructions, equipment allocation instructions, personnel scheduling instructions, and line changeover instructions. These instructions are then sent to external MES systems through standard API interfaces to drive production execution.

[0165] The system collects execution feedback data from the MES system in real time, including work order start status, completion progress, abnormal alarms, and final execution results, and then feeds the data back to the large model inference engine.

[0166] Calculate the deviation rate between the planned schedule and the actual execution to evaluate the accuracy of the scheduling:

[0167] ;

[0168] In the formula: For scheduling accuracy, the range of values ​​is... A higher value indicates higher scheduling accuracy. For work orders The actual completion time; For work orders The planned completion time.

[0169] Execution feedback data, bias data, and anomaly handling results are added to the domain fine-tuning sample library, and incremental fine-tuning of the large model is performed regularly to optimize the model loss function.

[0170] ;

[0171] In the formula: Fine-tune the total loss function for the model; These are the weighting coefficients for accuracy loss and constraint loss; The prediction loss corresponding to the scheduling accuracy; The penalties and losses corresponding to violations of hard rules.

[0172] Through a continuous feedback-iteration mechanism, the scheduling inference strategy and decision-making accuracy of the large model are continuously improved, forming a complete optimization loop.

[0173] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. These embodiments are merely descriptions of preferred embodiments and are not intended to limit the scope or concept of the invention. The specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. Such combinations, as long as they do not violate the spirit of the present invention, should also be considered as part of this disclosure. To avoid unnecessary repetition, the present invention will not further describe the various possible combinations.

[0174] This invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this invention and without departing from the design idea of ​​this invention, all modifications and improvements made by those skilled in the art to the technical solutions of this invention should fall within the protection scope of this invention. The technical content for which protection is sought in this invention has been fully described in the claims.

Claims

1. A large-model-based MES intelligent scheduling and dynamic order insertion optimization system, characterized in that, It includes a data acquisition and fusion module that enables data interaction through an industrial data bus, a large model inference engine, an intelligent scheduling optimization module, a dynamic order insertion processing module, an anomaly identification and reordering module, and an MES command issuance and feedback module. The data acquisition and fusion module is used to collect multi-source production data and perform cleaning, format unification and fusion processing, output a standardized production digital twin dataset and transmit it to the large model inference engine. The large model inference engine is built on the basis of a general large language model, and is constructed by domain fine-tuning, mounting an industrial scheduling knowledge base and setting an industrial hard rule verification layer. It is used to receive the production digital twin dataset and complete production constraint parsing and scheduling optimization inference. The intelligent scheduling optimization module is used to receive the inference results of the large model, combine them with global production constraints to construct scheduling optimization targets, generate an initial global optimal scheduling plan and provide a visual display. The dynamic order insertion processing module is used to receive and parse natural language order insertion instructions, and after completing the order insertion feasibility verification, send a reordering trigger signal to the intelligent scheduling optimization module to trigger global scheduling recalculation, or output alternative solutions for infeasible scenarios. The anomaly identification and reordering module is used to monitor production operation data in real time and identify abnormal events. After assessing the impact of the anomaly, it sends a reordering trigger signal to the intelligent scheduling optimization module. The MES instruction issuance and feedback module is used to receive the scheduling plan and issue scheduling execution instructions, while simultaneously feeding back the MES execution feedback data to the large model inference engine, forming a closed loop for iterative optimization of the scheduling strategy.

2. The MES intelligent scheduling and dynamic order insertion optimization system based on a large model according to claim 1, characterized in that, The multi-source production data collected by the data acquisition and fusion module includes the following three categories: MES basic data includes order information, work orders, BOM, process routes, equipment ledgers, capacity calendar and changeover rules; Real-time production dynamic data, including equipment status, overall equipment efficiency, material inventory and kitting rate, personnel scheduling and skill information, and work-in-process progress; External collaborative data includes procurement and delivery plans, warehouse outbound plans, quality inspection results, and energy consumption data.

3. The MES intelligent scheduling and dynamic order insertion optimization system based on a large model according to claim 1, characterized in that, The domain fine-tuning of the large model inference engine refers to the fine-tuning of instructions on the general large language model based on the enterprise's historical MES work orders, process routes, historical scheduling schemes and exception handling cases. The industrial scheduling knowledge base stores industry standards, enterprise process specifications, equipment capacity limits, changeover cost models, delivery priority strategies, and material substitution relationships. The large model inference engine retrieves relevant knowledge from the industrial scheduling knowledge base through vector retrieval, providing support for the inference process. The industrial hard rule verification layer is used to verify the capacity, equipment load, material completeness, and process compliance of the large model output results, filtering out invalid outputs that do not conform to actual production.

4. The MES intelligent scheduling and dynamic order insertion optimization system based on a large model according to claim 1, characterized in that, The intelligent scheduling optimization module uses maximizing delivery time fulfillment rate, maximizing capacity utilization, minimizing changeover time, and minimizing overall cost as the scheduling optimization objective function. The output initial scheduling plan includes work order processing sequence, equipment allocation scheme, start and completion time, changeover arrangement, and personnel allocation scheme. The visualization display includes a scheduling Gantt chart, equipment load chart, and material kitting analysis chart, and supports manual fine-tuning and confirmation of the scheduling plan.

5. The MES intelligent scheduling and dynamic order insertion optimization system based on a large model according to claim 1, characterized in that, The dynamic order insertion processing module supports spoken order insertion commands in both voice and text formats, and automatically parses and extracts order numbers, priorities, delivery dates, process requirements, and special constraint parameters. The order insertion feasibility verification dimensions include capacity margin, material availability, equipment availability time, and process compatibility. In infeasible scenarios, the generated alternative solutions include overtime production, outsourcing, and optimization suggestions for adjusting the delivery dates of other orders.

6. The MES intelligent scheduling and dynamic order insertion optimization system based on a large model according to claim 1, characterized in that, The anomaly identification and reordering module can identify production anomalies including equipment failure and downtime, material shortages, delayed procurement and delivery, staff absences, quality anomalies, and energy interruptions. The anomaly identification and reordering module automatically classifies anomaly types, assesses the scope of impact, and predicts recovery time. When an anomaly affects core production capacity or order delivery dates, a global reordering is triggered, prioritizing the delivery of high-priority orders during the reordering process.

7. The MES intelligent scheduling and dynamic order insertion optimization system based on a large model according to claim 1, characterized in that, The MES instruction issuance and feedback module connects to an external MES system via an API interface. The issued instructions include work order scheduling instructions, work order adjustment instructions, equipment allocation instructions, personnel scheduling instructions, and line changeover instructions. The received execution feedback data includes work order start status, completion progress, abnormal alarms, and execution results. The feedback data is used to iteratively optimize the scheduling inference strategy and decision accuracy of the large model.

8. A method for intelligent scheduling and dynamic order insertion optimization in MES based on a large model, characterized in that, Includes the following steps: S1. Data Acquisition and Fusion: Collect production data from multiple sources, clean and unify the format, and then fuse them to generate a standardized production digital twin dataset. S2, Large Model Enhanced Reasoning: Input the production digital twin dataset into a general large language model that has been enhanced by domain fine-tuning, knowledge base mounting and industrial hard rule verification to complete production constraint parsing and scheduling optimization reasoning; S3. Initial Global Scheduling Generation: Based on the inference results of the large model, combined with global production constraints, a scheduling optimization target is constructed, an initial global optimal scheduling plan is generated and visualized, and after confirmation, it is sent to the MES system for execution. S4. Dynamic order insertion processing: During the scheduling process, if a natural language order insertion instruction is received, the order insertion constraint parameters are parsed and extracted and a feasibility check is performed. If the check passes, the global scheduling is automatically recalculated. If the check fails, an alternative solution for the infeasible scenario is generated. S5. Anomaly Identification and Rescheduling: During the scheduling process, production operation data is monitored in real time to identify production anomalies and assess their impact. When the rescheduling trigger conditions are met, a global rescheduling is triggered to generate an adjusted scheduling plan and send it out for execution. S6. Command Issuance and Closed-Loop Optimization: The scheduling plan is issued to the MES system for execution, the execution feedback data of MES is collected and fed back to the large model, and the scheduling inference strategy is iteratively optimized.

9. The MES intelligent scheduling and dynamic order insertion optimization method based on a large model according to claim 8, characterized in that, In step S2, the domain fine-tuning is the instruction fine-tuning of the general large language model based on the enterprise's historical MES scheduling schemes and anomaly handling cases; the knowledge base mounting is the access to a vector knowledge base storing industrial scheduling rules and experience, and the real-time recall of relevant knowledge during the reasoning process; the industrial hard rule verification is the capacity verification, equipment load verification, material completeness verification and process compliance verification of the large model reasoning results.

10. The MES intelligent scheduling and dynamic order insertion optimization method based on a large model according to claim 8, characterized in that, In step S3, the scheduling optimization objectives include maximizing delivery time fulfillment rate, maximizing capacity utilization, minimizing changeover time, and minimizing overall cost; the generated initial scheduling plan includes work order processing sequence, equipment allocation, start and finish times, changeover arrangements, and personnel allocation schemes; The visualization formats include Gantt charts, equipment load diagrams, and material kitting analysis diagrams, which support manual fine-tuning and confirmation of the scheduling plan.