Mixed integer programming optimization system for production and manufacturing resource scheduling

Abstract

The invention relates to the field of business management, in particular to a production and manufacturing resource scheduling-oriented mixed integer programming optimization system, which comprises six modules: a data acquisition module acquires industrial big data streams; the data mapping module processes the data flow and constructs a real-time data model; the dynamic modeling module is used for automatically generating a mixed integer programming model according to a built-in steel production process model library based on a real-time data model; the model solving module generates a resource scheduling scheme through an optimization engine; the instruction analysis module analyzes a specific management and operation instruction; and the scheduling recalculation module starts a new round of mixed integer programming modeling and solving according to the increment information. According to the system, information such as resources is summarized through a mixed integer programming method, a specific production instruction is generated through a solver, a new round of modeling and solving are started when an accident occurs, rapid and accurate response to resource scheduling is achieved, and the response efficiency of resource scheduling and the reliability of the system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of business management, in particular to a mixed integer programming optimization system for production and manufacturing resource scheduling. BACKGROUND

[0002] The management and operation mode of modern steel manufacturing workshops has been transformed into a large system driven by industrial big data, which is related to complex resource planning, scheduling and management. Various types of automation equipment, enterprise resource planning systems, manufacturing execution systems and Internet of Things sensors are continuously generating management and operation data streams at an unprecedented speed. When these industrial big data are used to support enterprise resource allocation and production scheduling decisions, the current management methods still have deficiencies.

[0003] Most steel enterprises still use an administrative management and manual planning method. Production planners act as core decision makers, manually complete resource allocation and scheduling through general office software such as spreadsheets according to high-level production instructions issued by the ERP system and combined with part of the real-time data provided by the workshop MES. When a key device needs to be replaced urgently, a new high-value-added order may be inserted urgently through the order management system, and the planner faces multiple fragmented information, often relying on personal experience to manually adjust scheduling and reallocate resources. This decision-making method frequently causes mismatch between the upstream and downstream of the supply chain, disrupts the original production planning, and greatly increases the operating and logistics costs of the enterprise.

[0004] In a few technology-leading enterprises or research institutions, there is another way developed for specific production management links, i.e., algorithms designed to solve specific scheduling problems, which can improve efficiency in a local range, but lack data integration capability and management application versatility at the system level. More importantly, they lack standard interfaces for integration with enterprise-level management information systems, making it difficult to adapt to the agile planning and scheduling needs of modern steel manufacturing for multi-variety, small-batch and flexible production, and unable to be promoted as a universal enterprise resource management and optimization solution covering the whole process.

[0005] In summary, the existing technology results in that the data about the resource status of the workshop is extremely rich, but the management decision system that relies on planning and deploying the production resources of the whole workshop is still poor.

[0006] Therefore, a mixed integer programming optimization system for production and manufacturing resource scheduling is proposed. SUMMARY

[0007] The purpose of the present application is to provide a mixed integer programming optimization system for production and manufacturing resource scheduling to solve the problems raised in the background art.

[0008] To achieve the above object, the application provides the following technical scheme: a mixed integer programming optimization system for production and manufacturing resource scheduling, the system module comprises: A data acquisition module: from the manufacturing execution system, resource planning and Internet of Things heterogeneous data sources, industrial big data streams are collected; A data mapping module: the industrial big data streams are fused and standardized, and a real-time data model mapping the steel manufacturing workshop is constructed; A dynamic modeling module: based on the real-time data model, a mixed integer programming model is automatically generated according to the built-in steel production process model library; the workshop state and management target in the industrial big data are automatically converted into constraint conditions and objective functions; A model solving module: the mixed integer programming model is solved by an optimization engine to generate a resource scheduling scheme; An instruction analysis module: the variable values solved in the resource scheduling scheme are reversely analyzed into management and operation instructions; A scheduling recalculation module: the incremental information containing production execution deviation and unexpected events reflected by the industrial big data stream is monitored, when the preset triggering condition is met, the incremental information is fed back to the real-time data model, and a new round of mixed integer programming modeling and solving is started.

[0009] Preferably, the implementation process of collecting industrial big data streams comprises: By analyzing the connection protocols of the manufacturing execution system, enterprise resource planning and Internet of Things platform heterogeneous data sources, the business system data interface is determined, the industrial big data streams containing equipment working condition data, material tracking data and instrument state data are collected; the industrial big data streams are processed through the industrial data standardization process for preliminary data analysis, format conversion and centralized processing.

[0010] Preferably, the specific implementation process of constructing the real-time data model accurately mapping the physical workshop comprises: The industrial big data stream after preliminary processing is converted into a standard internal management event by using a normalization rule; the real-time state of the digital object corresponding to the physical entity is updated according to the internal management event ID, the accurate mapping of industrial big data to the core generated entity is completed, and an entity data model is constructed; according to the preset production and management rules, the logical association between digital objects is established, and the management rules are materialized into explicit constraints in the entity data model; at the same time, the system continuously scans the entity data model, and automatically resolves the logical conflicts caused by multi-source delay and errors of industrial big data according to the preset strategy; a model snapshot containing the current object, state and relationship is sent through API.

[0011] Preferably, the specific implementation process of generating a mixed integer programming optimization model comprises: The model snapshot is automatically analyzed, and the work order and resource information in the model snapshot are traversed to generate binary decision variables for potential "work order-resource" pairs and continuous time variables for each process, which are stored in a variable set; the physical and logical rules in the model snapshot are scanned through the constraint library built in the dynamic modeling module, and the corresponding mathematical templates are matched to generate physical constraint conditions for steel production, and the device unavailability period is generated as a corresponding time window avoidance constraint, which are stored in a constraint set; the final optimization target is generated according to the externally input business target by selecting and parameterizing the configuration from the component library built in the dynamic modeling module; and finally, the variable set, the constraint set and the optimization target are integrated to build a complete mixed integer programming model.

[0012] Preferably, the specific implementation process of solving the mixed integer programming optimization model by the optimization engine to generate a production resource scheduling scheme includes: The mixed integer programming model is solved by accessing the optimization engine; according to the steel production scheduling requirements, the solving strategy is configured and controlled, and external configuration parameters are loaded before solving; during the solving calculation process, the feasible schemes are recorded and cached; the effectiveness of the feasible schemes is judged, and after meeting the preset business requirements, the mathematical variable values in the feasible schemes are retranslated and constructed into business scheduling instructions; and the business scheduling instructions are integrated into the production resource scheduling scheme.

[0013] Preferably, the specific implementation process of parsing the production resource scheduling scheme into executable instructions and issuing it to the workshop for execution includes: The production resource scheduling scheme is received, and after format and logic checking, the production resource scheduling scheme is translated into specific production operation instructions for each process; the production operation instructions are accurately pushed to the corresponding post and device according to the preset priority and time window through the workshop task dispatching mechanism; a closed-loop management process of the work instructions is established, and for the production operation instructions that fail to be confirmed and received on time, the task is re-dispatched according to the preset business rules.

[0014] Preferably, the specific implementation process of the scheduling recalculation module includes: A tracking mechanism for production operation execution is established, a multi-channel monitoring mechanism is used to capture the workshop dynamics, unexpected execution conditions are captured, including automatic comparison of manufacturing execution system data deviation, capture of Internet of Things device abnormal reporting and reception of manual event injection, classified as workflow interruption events, and standardized and quantified, the actual impact of the workflow interruption events on the time line and resource availability of all subsequent associated steel processes is determined, the incremental industrial big data stream is re-injected into the real-time data model, and the resource scheduling review mechanism determines whether the re-planning process needs to be started according to the preset business conditions, and if the trigger condition is met, the system instructs the dynamic modeling module to obtain a new model snapshot based on the updated real-time data model, and starts a complete'modeling-solution-issuing' management cycle to generate a scheduling scheme more suitable for the current latest situation.

[0015] Compared with the prior art, the beneficial effects of the present application are: 1. In the scheduling scheme, by dynamically generating and solving the mixed integer programming model, compared with the traditional scheduling method based on heuristic rules, the global optimality of the scheduling scheme is mathematically guaranteed and a quantifiable optimization limit is provided, ensuring that under complex constraint conditions, key indicators such as resource utilization, production efficiency and cost control can reach a more optimal level in theory, significantly improving the scientificity and economic benefit of production decision-making.

[0016] 2. In terms of business variability, a logical template is used to automatically generate a mathematical model, decoupling complex physical constraints, process logic and variable business objectives into configurable and extensible modeling components, when changes occur, only the corresponding data model needs to be adjusted, and a new optimization model can be automatically generated, enhancing the system's ability to quickly adapt to business changes and long-term maintainability.

[0017] 3. In the production scheduling process, a full-process closed-loop automated management from 'data collection-model generation-solution-instruction issuance-execution tracking-dynamic recalculation' is constructed, by monitoring production deviations and unexpected events in real time, the reconstruction and solution of the model are automatically triggered, realizing rapid and accurate response to workshop dynamics, solving the pain points of information lag and inability to respond to unexpected situations in traditional static scheduling mode, ensuring that the scheduling scheme always fits the actual situation, and ensuring the smoothness and efficiency of the production process. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A method flowchart of a mixed integer programming optimization system for production and manufacturing resource scheduling is proposed for the embodiments of the present application; Figure 2 A system structure diagram of a mixed integer programming optimization system for production and manufacturing resource scheduling is proposed for the embodiments of the present application; Figure 3A flowchart for forming a mixed integer programming model is provided for the embodiments of the application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the application.

[0020] Please refer to Figures 1-3 The application relates to a mixed integer programming optimization system for production and manufacturing resource scheduling, and specific system modules are as follows. A data acquisition module acquires industrial big data streams from a manufacturing execution system, a resource plan and an Internet of Things heterogeneous data source. A data mapping module fuses and standardizes the industrial big data streams, and constructs a real-time data model of a mapped physical workshop. A dynamic modeling module automatically generates a mixed integer programming model according to an embedded model library based on the real-time data model, and automatically converts workshop states and management targets in the industrial big data into constraint conditions and an objective function of the mathematical model. A model solving module solves the mixed integer programming model through an optimization engine, and generates a resource scheduling scheme. An instruction analysis module reversely analyzes variable values solved in the resource scheduling scheme into management and operation instructions. A scheduling recalculation module monitors incremental information containing production execution deviations and unexpected events reflected by the industrial big data, and automatically feeds back the incremental information to the real-time data model and starts a new round of mixed integer programming modeling and solving when preset triggering conditions are met.

[0021] The technical solutions of the application will be described in further detail below with reference to specific embodiments. EMBODIMENT

[0022] The embodiments of the application disclose a mixed integer programming optimization system for production and manufacturing resource scheduling. In a complex environment of a job workshop of steel manufacturing, production scheduling is a very challenging task. Please refer to Figure 1 and Figure 2The specific modules of the system include: a data acquisition module: acquiring industrial big data streams from a manufacturing execution system, a resource plan and an Internet of Things heterogeneous data source; a data mapping module: processing the industrial big data streams to construct a real-time data model; a dynamic modeling module: based on the real-time data model, automatically generating a mixed integer programming model according to an embedded template library; a model solving module: solving the mixed integer programming model through an optimization engine to generate a resource scheduling scheme; an instruction analysis module: reversely analyzing the variable values solved in the resource scheduling scheme into management and operation instructions; a scheduling recalculation module: monitoring incremental information containing production execution deviations and unexpected events reflected by the industrial big data, when a preset triggering condition is met, automatically feeding back the incremental information to the real-time data model, and starting a new round of mixed integer programming modeling and solving.

[0023] Further, the industrial big data streams are acquired from a manufacturing execution system, a resource plan and an Internet of Things heterogeneous data source; the above data acquisition module is corresponding; the specific implementation process includes: The system analyzes the interface and identifies the protocol of the heterogeneous data source. For a manufacturing execution system (MES), it is connected to the back-end database of the MES through ODBC or JDBC protocol. By analyzing the database mode, the key data tables related to production scheduling are identified, such as "production order table", which contains order number, steel grade, target weight, process path, planned start / end time, etc.; and "material tracking table", which records the flow state and timestamp of ladle, tundish, billet or coil between stations in real time. For an enterprise resource planning (ERP) system, by analyzing the business application program interface provided or parsing the intermediate document, customer orders, raw material procurement and inventory information and finished product inventory level are obtained. For the Internet of Things (IoT) platform, through OPC-UA, ModbusTCP and MQTT industrial communication protocols, the IoT data gateway is connected, and the key measurement points are identified, such as temperature sensor data and temperature data in the converter and vibration and current data of key motors.

[0024] After interface analysis, the module initiates multi-mode data collection. For transaction-based systems such as MES and ERP, a "pull" mode is adopted, and a preset frequency is used to execute SQL queries to actively pull data updates. For sources such as IoT platforms and PLCs that generate high-frequency time-series data, the system adopts a "push" mode and establishes a "publish / subscribe" relationship through protocols. Ultimately, all raw industrial big data streams collected through multi-mode collection and in different formats are imported into the industrial data standardization process for centralized processing. First, the data streams obtained from different protocols are parsed into unified, machine-readable memory objects. Second, format conversion is performed to handle the heterogeneity of data in units, types, and encodings. For example, Celsius (℃) and Fahrenheit (℉) temperature values reported by different sensors are uniformly converted to the international unit of Kelvin (K); and time stamps in different formats from different information systems are all converted to a unified, time zone-based UTC standard time format. Finally, the standardized data after parsing and conversion is centrally processed. For example, a work order status change record from MES and a furnace temperature overrun alarm from an IoT platform are both converted into event messages with a unified JSON format.

[0025] By systematically analyzing and accessing all relevant data sources from MES, ERP, and IoT, and using a "push-pull combination" multi-mode collection method, an unprecedented industrial big data stream covering the entire spectrum from operation to management to physical execution is constructed. This enables scheduling decisions to be based on real-time, complete insights into the production panorama rather than on one-sided or outdated information.

[0026] Further, the industrial big data stream is fused and standardized to construct a real-time data model that maps the physical workshop; the corresponding data mapping module includes the following specific implementation processes: The industrial big data stream after preliminary processing is converted into standard internal management events using preset canonicalization rules. For example, when a PLC of a steel ladle transport overhead crane reports that it has reached the coordinates of the continuous casting area, and the status of the steel ladle in the MES is updated to "has been lifted to the continuous casting machine", the canonicalization rule engine will trigger the generation of a standardized internal management event, defined as "steel ladle arrival at continuous casting station event". The event object will include a unique event ID, event type, timestamp, and key business load information. By establishing a canonicalization knowledge base containing hundreds of rules, the system can automatically translate massive, heterogeneous underlying data points into a stream of discrete, clearly business-meaning "management events" that can be uniformly understood by upper-layer applications.

[0027] Subsequently, the system updates the internal state of the digital object corresponding to the physical entity in real time according to the ID of the internal management event and the load information thereof, thereby completing the accurate mapping of industrial big data to the core management entity and constructing an entity data model. In the system, each core production or logistics unit, such as a converter, a ladle, a continuous casting machine, a heating furnace, a billet, or a production order, is abstracted as a "digital object" with clear attributes and states. When the "ladle arrives at the continuous casting station event" is triggered, the system retrieves the digital objects of the specified ladle and continuous casting machine in the event load, updates the position attribute of the ladle digital object, and updates the state attribute thereof from "in transit" to "waiting for casting". At the same time, the state attribute of the continuous casting machine digital object is updated to "ladle is in place".

[0028] While updating the state of a single digital object, the system establishes dynamic logical associations between digital objects according to preset production and management rules, i.e., abstract management regulations and process requirements are materialized as computable explicit constraints in the model. For example, there is a process rule of "molten steel overheating degree limit" in steel production, and the system materializes this rule as a logical association between the ladle digital object and the steel grade digital object. When the state of a ladle object is updated, the system will immediately query the liquidus temperature of the associated steel grade object according to the steel grade mark, calculate the real-time temperature difference, and if the threshold is exceeded, a constraint violation attribute of the ladle object will be marked.

[0029] In addition, due to the multi-source and heterogeneous characteristics of industrial big data and the inevitability of delays, conflicts, and errors between data, the system continuously and automatically scans the entire real-time data model to ensure the logical consistency of the model according to preset conflict resolution strategies. For example, if the MES system reports that a billet has been put into storage, but the laser scanning sensor in the storage area has not detected the ID of the billet within the preset 2-minute time window, the system will determine this as a logical conflict. First, the state of the billet digital object is marked as "state inconsistency"; second, the system automatically queries the crane system log to try to find supporting evidence; if it still cannot be confirmed, a low-priority to-do task is generated and pushed to the warehouse attendant for manual confirmation. During this period, any scheduling decision based on the billet will be suspended or given a very low priority. Finally, the real-time data model composed of a large number of digital objects, their real-time states, and complex logical associations between them can be called by external systems in the form of "model snapshot" through standardized API interfaces.

[0030] By pre-implanting these rules as logical associations between digital objects, the constraints are "materialized" and "explicitized" when generating the model snapshot, reducing the modeling threshold and cycle, and enabling the model to quickly adapt to changes in business rules, thereby improving the agility of the entire system.

[0031] Further, based on the real-time data model, a mixed integer programming model is automatically generated from a built-in model library; a plant state and a management target in the industrial big data are automatically converted into constraint conditions and an objective function of the mathematical model; corresponding to the dynamic modeling module, reference is made to Figure 3 , and a specific implementation process includes: The dynamic modeling module automatically analyzes the model snapshot and traverses all work orders and their associated process information and state information of all available resources in the snapshot. Based on the traversal result, the module automatically generates a binary decision variable for each potential "process-resource" pair and adds constraints for deciding which available resource actually performs the process in subsequent solving, and by means of the constraints, a large number of invalid combinations are explicitly excluded, and the "feasible solution space" that needs to be explored by the solver is reduced. For example, for the "billet heating" process of work order WO-089, if the model snapshot shows that heating furnace H1 and heating furnace H2 are both in an available state and the process capacity matches, the system automatically generates binary variables X WO−089,heat,H1 and X WO−089,heat,H2 and adds constraints, and the sum of the two is 1, which can decide which heating furnace actually performs the process in subsequent solving. At the same time, for each process, the system generates continuous time variables S WO−089,heat and E WO−089,heat representing the start and end times. After generating the decision variables, the dynamic modeling module uses its built-in highly structured constraint library to perform a second deep scan of the model snapshot, automatically match and generate linear constraint conditions. The constraint template library contains a standard mathematical expression set of common physical and logical rules in steel production. When the module scans the physical and logical rules between the digital objects established by the data mapping module in the model snapshot, the corresponding mathematical template is triggered. For example, when the "refining" process in a work order is identified to have a process path association that must precede the "continuous casting" process, the module calls the "precedence relationship template" to automatically generate constraints. In particular, for the explicitly marked device unavailable time period in the model snapshot, such as the planned maintenance of continuous casting machine-C2 between 2025-06-21T08:00:00Z and 2025-06-21T10:00:00Z, the module matches the "time window avoidance template" to generate corresponding constraints for all processes that can be allocated to the continuous casting machine-C2, forcing the end time to be earlier than 08:00:00 or the start time to be later than 10:00:00, thereby "shielding" this unavailable time in mathematics. Subsequently, the system selects and parameterizes the configuration according to the business target externally input, representing the current business focus, such as "minimize maximum completion time" or "minimize total production cost", from the target function template library built in the dynamic modeling module, to generate the final optimization target.

[0032] Finally, the dynamic modeling module integrates and formats all the decision variables, all the linear constraints and the final determined optimization objective generated in the previous steps, and builds a structured, grammatically correct, mixed integer programming model file which can be directly read and solved by standard optimization solvers.

[0033] Through the "template matching" mechanism, the production scheduling is realized from static offline planning to dynamic online optimization, which greatly improves the agility of the system to adapt to business changes, so that the optimization model can keep pace with the evolution of the physical factory, improve production efficiency and reduce scheduling cost.

[0034] Further, the resource scheduling scheme is generated by solving the mixed integer programming model through the optimization engine; the corresponding model solving module, the specific implementation process includes: The model solving module solves the mixed integer programming model by accessing multiple optimization engines such as Gurobi, CPLEX or Xpress. Before the solution starts, configuration parameters are loaded first, which allow the scheduling manager to flexibly configure and adjust the solution strategy according to the specific business needs and urgency. For example, the manager first configures a longer upper limit of the solution time, such as 30 minutes, then, in response to an emergency re-scheduling due to a sudden failure of the hot rolling mill, the manager sets the upper limit of the solution time to be extremely short, 90 seconds, and lets the solver focus on quickly finding any feasible solution that can alleviate the production interruption, rather than pursuing theoretical optimality. These configuration parameters are dynamically transmitted through API, so that the solution process can be highly adaptable to different business situations.

[0035] In the process of solving, the system continuously monitors the running state of the optimization engine, records the output of the phased feasible solution, and displays the evolution process of the solution in the Gantt chart in real time. Managers can intuitively see the generation of these solutions, quickly compare and select the most favorable countermeasures, enhance the agility and emergency response capability of decision-making, and make the solving process transparent, reducing the calculation time in emergency situations. In the process of searching for the optimal solution, modern MIP solvers usually find multiple integer feasible solutions of acceptable quality but not optimal, for example, the solver finds a scheduling solution with a maximum completion time of 42 hours at 20 seconds, and a better solution with a completion time of 40.5 hours at 60 seconds. The system will store these solutions together with their corresponding objective function values, and when the solving process of the optimization engine is terminated, the system will obtain the final solving result. Subsequently, the system will first perform a validity judgment at the business level, which is based on pre-set business requirements, for example, checking whether the scheduling solution causes the order of a key customer to be delayed for more than 2 days, or whether it makes the continuous operation furnace number of a continuous casting machine lower than the economic furnace number. Only when the solution meets these basic business-level hard requirements will the system accept the solution, if not, the system may try to adjust the solving parameters and initiate a new round of solving based on pre-set rules, or report the problem to the human decision maker.

[0036] Once the solving result is confirmed to be business-effective, the system starts the reverse translation and construction process, re-translating and constructing the pure mathematical variable values in the solving result into scheduling instructions with clear business meaning. For example, the optimal solution returned by the solver contains X WO−089,heat,H2 =1, S WO−089,heat =2025-06-21T14:00:00Z and E WO−089,heat =2025-06-21T16:30:00Z. The system translates it into a specific, human-readable business scheduling instruction: "Heating process of order WO-089 starts at 14:00:00 on June 21, 2025, on heating furnace-H2, and is expected to be completed at 16:30:00". The system will perform such translation for all decision variables with non-zero values in the model, and finally integrate all generated atomic business scheduling instructions according to the order number, equipment and time sequence to form a complete and logically clear production resource scheduling solution.

[0037] By introducing an externally configurable solving strategy, the system can flexibly adjust the solving goal according to whether it is for regular planning or emergency response, strike a balance between computing resources and business needs, and enhance the practicality and adaptability of the optimization tool in real industrial environments.

[0038] Further, the variable values solved in the resource scheduling scheme are reversely analyzed into management and operation instructions; a corresponding instruction analysis module, the specific implementation process includes: The instruction analysis module first receives the structured production resource scheduling scheme and immediately performs format and logic verification. Format verification ensures that the scheme file conforms to the predefined syntax specification, and logic verification is the last line of defense before execution, avoiding the issuance of obviously incorrect or outdated instructions.

[0039] After verification, the translation engine of the system core starts, accurately translating the macro description in the production resource scheduling scheme facing the planning level into atomic instructions facing specific execution objects, which cannot be further divided. This translation process has a high context awareness capability. For example, a record in the scheduling scheme "Steel billet SB-901 of work order WO-115, perform heating process in heating furnace-H1 on 2025-06-21T18:00:00Z", will be translated into multiple different atomic instructions: for the control system of the heating furnace, the system will issue a machine-readable instruction set containing the heating curve parameters corresponding to the target steel grade X80 through OPC-UA or Profinet protocol; for the crane operator in the loading area, he will receive a human-readable instruction on his handheld terminal: "Task: Steel billet hoisting, please hoist the steel billet SB-901 located at storage location A-05-3 to the furnace loading area of heating furnace-H1 before 17:50"; for the heating furnace operator station, the human-machine interface will display: "18:00 will start heating work order WO-115, please confirm that the furnace door is closed and all parameters are normal".

[0040] Subsequently, through the workshop task dispatching mechanism, the system pushes these atomic instructions to the corresponding posts or devices according to the preset priority and time window. This module also contains a rigorous work instruction closed-loop management process, and each issued instruction is attached with an expected "confirmation receipt" timestamp, for example, requiring the crane operator to confirm receipt of the instruction 10 minutes before the task starts. The system will continuously track the status of the instruction, and if no confirmation signal is received from the operator terminal or device PLC within 10 minutes, the system will automatically trigger the reassignment of the task. For example, if the operator of crane-TC1 does not confirm the task of hoisting the steel ladle, the system will first query the status of crane-TC2, and if it is idle and nearby, the system will automatically assign the task to the operator of crane-TC2, and send a "task has been automatically reassigned" notification to the monitoring interface of the workshop dispatch supervisor.

[0041] By automatically translating macro plans into context-specific atomic instructions for different execution objects and accurately pushing them in a digital way, it ensures that the intentions of the optimization scheme can be accurately conveyed to each operation unit, ensuring the executability of the scheme.

[0042] Further, the monitoring of the incremental information containing production execution deviation and unexpected events reflected by industrial big data meets the preset triggering condition, and the incremental information is automatically fed back to the real-time data model, and a new round of mixed integer programming modeling and solving is started; the specific implementation process of the scheduling recalculation module includes: The scheduling recalculation module establishes a comprehensive and multi-channel production operation execution tracking mechanism. First, it automatically and frequently compares the data deviation in the manufacturing execution system (MES), for example, the system will continuously compare the “scheduled arrival time at the refining station” set for a ladle in the current effective scheduling scheme, such as 14:30, with the “actual arrival time” actually input by the operator or automatically reported by the position sensor in the MES, such as 14:45. Any deviation exceeding the preset threshold, such as 5 minutes, will be identified as an incremental information. Secondly, it captures abnormal events reported by Internet of Things (IoT) devices in real time, for example, the liquid level sensor of the tundish of the continuous casting machine suddenly drops below the safety threshold, or the vibration frequency of the main motor of the hot rolling mill exceeds the warning line, which will be captured as high-priority incremental information. Finally, the mechanism also allows manual event injection, for example, when the shift scheduler inputs an information “No. 5 ladle needs to be reworked due to abnormal composition, with a delay of 1 hour” through the HMI interface, the information is also injected into the system as a standardized incremental information.

[0043] When the above unexpected execution condition occurs, the system first classifies it as a standardized “workflow interruption event” and performs quantitative analysis to determine the actual impact of this interruption on the time line and resource availability of all subsequent related tasks and the overall project. Taking the “No. 5 ladle rework delay of 1 hour” as an example, the system not only records this delay, but also immediately deduces the chain reaction of this event using its internally maintained real-time data model: since No. 5 ladle occupies a certain refining station, the refining task of No. 6 ladle originally planned to use the station must be delayed; and the delay of No. 6 ladle will also disrupt the subsequent continuous casting machine casting plan it supplies. The system solidifies this series of quantifiable impacts into structured data and injects it back into the real-time data model as incremental industrial big data, updates the model to reflect the latest plant reality containing this interruption event, and stores the incremental information for summary and classification through machine learning algorithms. When it happens again, it identifies and automatically generates possible root cause hypotheses, generates suggestions through verification, and submits them to the administrator, reducing unplanned downtime and production interruption times, making the entire production system run more smoothly and more predictable, thereby reducing the huge cost caused by interruptions.

[0044] Subsequently, the resource scheduling review mechanism within the system automatically determines whether a re-planning process needs to be initiated based on this updated reality according to pre-set, highly customized business conditions. These trigger conditions are the embodiment of business logic, for example, the rule can be "trigger when the unavailability time of any key resource exceeds 30 minutes", or "trigger when the expected completion time of any high-priority order is delayed by more than 2 hours". In the aforementioned case of ladle rework, if the resulting continuous casting machine plan delay of 1 hour meets the trigger condition, the system will immediately automatically instruct the dynamic modeling module to obtain a new model snapshot from the updated real-time data model. Based on this snapshot containing the latest disturbance information, the system will automatically start a complete "modeling-solution-issuing" management cycle to generate a new scheduling scheme that is more suitable for the current latest situation and is globally optimized, and issue it for execution, so that the entire production system can quickly recover to the optimal running track after encountering a disturbance.

[0045] By establishing a complete closed loop of "monitoring-quantification-feedback-re-calculation", the scheduling system is transformed from a static "scheduling tool" to a dynamic "production control system". Through continuous, data-driven re-calculation cycles, it is ensured that at every point in time during the production process, the optimal decision based on the current latest information and global trade-off is made to guide the operation of the workshop. This brings significant and lasting efficiency improvement and cost savings to the enterprise.

[0046] The real value of this entire process is to build a production execution system with strong resilience and self-healing ability, which can actively digest production disturbances and ensure that production activities are always running efficiently in a near-optimal state in a complex and changing industrial environment, thereby achieving a breakthrough improvement in efficiency, cost and delivery capability. The highly automated complex optimization modeling and solving process significantly reduces the application threshold and gives the production system unprecedented business agility, enabling it to quickly respond to various changes. Embodiments

[0047] In a steel production workshop, the specific implementation of the data acquisition and data mapping module is as follows: The "sales order" in the ERP system triggers the generation of production instructions. The data collection module pulls a new production order data {order ID: SN20240521-001, steel grade: Q235B, target quality: 120.5, unit: ton, process flow: BOF#2->LF#1->CCM#3} through the API call to the ERP at a preset 30-minute cycle. The order is passed down to the MES system, which decomposes it into specific heat plan. When the No. 2 converter starts to smelt for the order and prepares to tap, the workshop scheduler updates the status on the MES interface. The SQL task of the data collection module polls the back-end database of the MES at a frequency of 1 minute. The returned result is the status: 'preparing to tap', ladle number: Ladle-101. The workshop action officially starts, and the No. 2 converter starts to pour the molten steel into the ladle Ladle-101 below. This process generates a large amount of high-frequency sensor data: converter inclination sensor: {label: BOF2 inclination, value: 95.5, unit: degree, timestamp: 2024-05-21T10:30:05.123Z}, ladle weighing sensor: {label: BOF2 weighing value, value: 120.8, unit: T, timestamp: 2024-05-21T10:32:15.456Z}, temperature probe: {label: BOF2 temperature probe, value: 1650, unit: C, timestamp: 2024-05-21T10:32:20.789Z}.

[0048] The above heterogeneous data streams are converged into the "industrial data standardization process" for centralized processing. The ERP order, MES status, and PLC sensor data are parsed into unified internal memory objects. The unit is converted: temperature 1650°C is converted to 1923.15K; the weight unit is unified to T. The timestamp is converted to the UTC standard time format with time zone and retains millisecond precision. The paradigm rule engine starts to work. It receives data that the converter inclination is greater than 90 degrees and the MES status is "preparing to tap". The engine triggers rule #27: "converter tapping starts", and generates a standardized event in JSON format.

[0049] Transportation data collection and event generation: Ladle-101 is lifted by Crane-5 to CCM-3. Crane PLC and laser scanner in CCM area continuously push data. Crane PLC data: {Tag: Crane-5 position, XY coordinates, Value: {x: 450, y: 800}, Timestamp: 2024-05-21T10:38:40.850Z} (coordinates enter CCM area). CCM area scanner: {Tag: CCM-3 identification, Value: Ladle-101, Timestamp: 2024-05-21T10:38:40.900Z}. Rule engine fuses the above two pieces of information, generates event type: Ladle arrival at CCM station event. System retrieves the digital object of ladle with ID Ladle-101. Its internal state is updated: state attribute changes from 'in transportation' to'ready for casting'; location attribute is updated to 'CCM#3'. Meanwhile, the system retrieves the digital object of CCM with ID CCM#3. Its internal state is updated: state attribute changes to 'ladle ready'. The system establishes a temporary waiting logical association between the two digital objects of Ladle-101 and CCM#3. The system automatically performs consistency check. It finds that the MES system, due to network delay, still has the state of Ladle-101 as 'tapped', not 'arrived at CCM'. The system determines that there is a logical conflict, marks the consistency attribute of Ladle-101 digital object as 'warning: MES lag', but the scheduling decision prioritizes the more real-time PLC / IoT data. After resolving the conflict, the system sends a model snapshot containing all current objects, states and relationships through API. Embodiment

[0050] In a steel production workshop, referring to Figure 3 , the specific implementation of the dynamic modeling module is as follows: The model snapshot is deeply analyzed and converted from text format to dynamic internal data structure that can be queried at high speed in memory. The order list in the snapshot is parsed, and the data is filled into the "order registry table" in memory. The process route set in the snapshot is parsed, and the data is filled into the "process path registry table". The resource list in the snapshot is parsed, and the data is filled into the "resource registry table". The association between objects in the snapshot is parsed, and the data is filled into the "constraint association registry table". The module traverses each process in the "order registry table" and queries the "constraint association registry table" to obtain all compatible resources. For each valid "process-resource" pair, the system generates a binary decision variable X(path id, resource id) from the data pair (path id, resource id) and stores it in the "variable set". The module traverses all processes in the "order registry table" again. For each path id, the system generates two continuous time variables S(path id) (start time) and E(path id) (end time) and adds them to the "variable set".

[0051] For each record containing the (path A, path B) sequence, the module extracts the time variable E(path A) and S(path B) of the corresponding process. These data are fed into the "precedence relationship template" to generate the mathematical constraint E(path A) <= S(path B) and added to the "constraint set". For each resource id, the module finds all the processes that can be assigned to it, i.e. all the possibilities of X(path i, resource id) being 1. For any two process paths i and j that can use the resource, the module extracts their decision variables and time variables and feeds these variable data into the "resource mutual exclusion template" to generate a set of constraints that ensure they cannot overlap in time. This newly generated set of constraints is added to the "constraint set". For each resource with a planned downtime, the module finds all the processes that can be assigned to it. For each such process, the module extracts its time variable S(path id), E(path id) and the downtime data of the resource. These data are fed into the "time window avoidance template" to generate the constraint E(path id) <= downtime start time or S(path id) >= downtime start time and added to the "constraint set".

[0052] The module receives an externally inputted service objective instruction. The instruction triggers the module to select the corresponding mathematical expression template from the "objective function template library". The parameters in the template, including the unit processing cost, the unit energy consumption cost, and the delay penalty coefficient, drive the module to go back to the "resource registry" and "work order registry" in the memory to extract these specific numerical values. The extracted cost, time, and other parameter values, together with the relevant decision variables X(path, resource) and completion time variables in the "variable set", are fed into the selected template. The template combines these data and variables into a complete, parameterized mathematical expression that performs a weighted summation of the processing cost, processing time, delay time, and penalty cost of all processes. The integrated "variable set", "constraint set", and final "objective function expression" are integrated and packaged to form a complete structure of a mixed integer programming model file that can be directly read and solved by a solver. Embodiment

[0053] In a steel production workshop, the specific implementation of the model solving and instruction analysis module is as follows: The model solving module starts, fuses the MIP model with the solving strategy configuration file containing parameters such as time limit and optimal gap into a "solving task package". This data package is transmitted and submitted to an external high-performance optimization engine. The optimization engine starts calculation, establishes a bidirectional data flow with the solving module, and continuously transmits the solving state and any intermediate feasible solution found as a "phase solution" data stream back. The solving module receives and caches these phase solutions, and updates the currently known optimal solution in real time. When the solving terminates, the optimization engine transmits the final "original mathematical solution" data package back to the solving module. This data package is immediately sent to the business validity verification process, and compared with the preset business hard rules delivery period and equipment constraints. After verification, the data is marked as "business valid solution". The "business valid solution" data stream flows into the reverse translation engine. The engine traverses the variable values in the solution, and converts them from pure mathematical expressions (such as X = 1, S = timestamp) to structured business instruction fragments. All fragments are integrated, and finally form a "production resource scheduling scheme" document.

[0054] The instruction parsing module receives the scheduling scheme. After format and logic verification, the scheme data flows into the translation engine. The engine translates each macro scheduling instruction into a set of indivisible atomic instructions for different execution objects, including machine-readable instructions, human-readable instructions, and HMI display instructions. These atomic instruction sets are routed to different communication pipelines according to the target, priority, and time by the dispatch mechanism, and the data is accurately pushed to the target device or operator terminal through the corresponding industrial protocol. The "acknowledgement of receipt" signal sent by the execution end is transmitted back to the parsing module as a data stream, updating the instruction state. If the acknowledgement signal is not received within the specified time, an internal timeout event is triggered. This event drives the system to automatically query the real-time data model to find alternative resources, and generates new, reassigned atomic instruction sets. The new instruction sets are pushed to the alternative execution objects, and a "task automatic reassignment" notification data is generated and sent to the upper-level monitoring system, completing the closed-loop management of instruction execution. Embodiments

[0055] In a steel production workshop, the specific implementation of the scheduling recalculation module is as follows: The scheduling recalculation module is started and establishes a multi-channel data monitoring process that runs continuously and receives incremental information from data sources in parallel. The module continuously compares the "planned event timestamp" in the scheme with the "actual event timestamp" reported by the MES at a high frequency. When the difference between the two exceeds the preset threshold, a "deviation data packet" containing the deviation type, the object ID involved, and the deviation value is generated. The module subscribes to and receives "abnormal event messages" from IoT / PLC in real time. These messages contain event source ID, event type, severity level, and related measurement values, and are directly captured. The unexpected event information submitted by the operator or dispatcher through the HMI is encapsulated into a standardized "manual injection data packet" by the system.

[0056] All the "deviation data packets", "abnormal event messages" and "manual injection data packets" captured above are sent to a classifier and classified into standardized "workflow interruption event" data objects. The "workflow interruption event" data object is sent to the real-time data model as a query parameter. The module uses the existing objects and their associated relationships in the model to perform impact deduction calculations. This calculation process generates a new data structure, namely a "quantitative impact set", which contains the new time of all subsequent tasks affected by the event and the new unavailable time window of related resources. The real-time data model consumes this data and updates the state and attributes of the corresponding digital objects in the internal, enabling it to instantly reflect the latest shop floor reality containing the interruption event.

[0057] The key indicators in the "quantitative impact set", namely the key resource delay duration and the expected delay of high-priority orders, are extracted and compared with the preset "recalculation trigger rule library" in the system. If the comparison result meets any rule, the review mechanism will generate a "start recalculation" instruction signal. After receiving the instruction, the dynamic modeling module immediately requests and obtains a new "model snapshot" from the updated real-time data model. This new snapshot containing the latest disturbance information is sent to the complete "modeling-solution-instruction analysis" management cycle as initial input data, thereby starting a new round of global scheduling optimization.

[0058] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A mixed integer programming optimization system oriented to production manufacturing resource scheduling, characterized in that, Comprise: Data acquisition module: from manufacturing execution system, resource planning and Internet of Things heterogeneous data sources, industrial big data stream acquisition; Data mapping module: the industrial big data stream is fused and standardized, and the real-time data model of mapping steel manufacturing workshop is constructed; Dynamic modeling module: based on the real-time data model, the mixed integer programming model is automatically generated according to the built-in steel production process model library; the workshop state and management target in industrial big data are automatically converted into constraint conditions and objective function; Model solving module: the mixed integer programming model is solved by optimization engine to generate resource scheduling scheme; Instruction analysis module: the variable value solved in the resource scheduling scheme is reversely analyzed into management and operation instruction; Scheduling recalculation module: monitor the incremental information containing production execution deviation and unexpected event reflected by industrial big data stream, when the preset trigger condition is met, the incremental information is fed back to the real-time data model, and a new round of mixed integer programming modeling and solving is started.

2. The mixed integer programming optimization system for production manufacturing resource scheduling according to claim 1, wherein, The specific construction process of the data acquisition module comprises: By analyzing the connection protocol of manufacturing execution system, enterprise resource planning and Internet of Things platform heterogeneous data source, the business system data interface is determined, the industrial big data stream containing equipment working condition data, material tracking data and instrument state data is collected; the industrial big data stream is processed through industrial data standardization process, and the data is analyzed, the format is converted and the data is processed.

3. The mixed integer programming optimization system for production manufacturing resource scheduling according to claim 1, wherein, The specific construction process of the data mapping module comprises: The industrial big data stream after preliminary processing is converted into standard internal management event by using normalization rule; the real-time state of digital object corresponding to physical entity is updated according to the internal management event ID, the accurate mapping of core generated entity in industrial big data is completed, the entity data model is constructed; according to the preset production and management rule, the logical association between digital objects is established, the management rules are embodied as explicit constraints in the entity data model; at the same time, the system continuously scans the entity data model, and automatically resolves the logical conflict caused by multi-source delay and error of industrial big data according to the preset strategy; the model snapshot containing current object, state and relationship is sent through API.

4. The mixed integer programming optimization system for production manufacturing resource scheduling of claim 1, wherein, The specific construction process of the dynamic modeling module comprises: The model snapshot is automatically analyzed, and the work order and resource information in the model snapshot are traversed, the binary decision variable is generated for potential "work order-resource” pairing, and the continuous time variable is generated for each process, which are stored in variable set; the physical and logical rules in the model snapshot are scanned through the constraint library built in the dynamic modeling module, and the corresponding mathematical template is matched, the steel production physical constraint condition is generated, and the device unavailable period is generated as the corresponding time window avoidance constraint, which are stored in constraint set; according to the generated business target transmitted from outside, the final optimization target is generated by selecting and parameterizing configuration from the component library built in the dynamic modeling module; finally, the variable set, constraint set and optimization target are integrated to construct the mixed integer programming model with complete structure.

5. The mixed integer programming optimization system for production manufacturing resource scheduling according to claim 1, wherein, The specific construction process of the model solving module comprises: Solving the mixed integer programming model by the access optimization engine; according to the scheduling requirements of steel production, configuring and regulating the solving strategy, loading external configuration parameters before solving; in the solving process, record and cache the phased feasible scheme; judge the effectiveness of the feasible scheme, and after meeting the preset business requirements, retranslate and construct the mathematical variable values in the feasible scheme into business scheduling instructions; the business scheduling instructions are integrated into the production resource scheduling scheme.

6. The mixed integer programming optimization system for production manufacturing resource scheduling according to claim 1, wherein, The specific construction process of the instruction analysis module includes: Receiving the production resource scheduling scheme, after format and logic checking, translating the production resource scheduling scheme into specific production operation instructions for each process; through the workshop task allocation mechanism, according to the preset priority and time window, the production operation instructions are accurately pushed to the corresponding post and equipment; a closed-loop management process of work instructions is established, and for the production operation instructions that cannot be confirmed in time, according to the preset business rules, the task is redistributed.

7. The mixed integer programming optimization system for production manufacturing resource scheduling according to claim 1, wherein, The specific construction process of the scheduling recalculation module includes: Establishing a tracking mechanism for production operation execution, capturing the workshop dynamics through a multi-channel monitoring mechanism, and classifying unexpected execution conditions, including automatically comparing the data deviation of the manufacturing execution system, capturing Internet of Things device abnormal reports, and receiving manual event injection, as workflow interruption events, and standardizing and quantifying them to determine the actual impact of the workflow interruption events on the timelines and resource availability of all subsequent associated steel processes; as an incremental industrial big data stream, re-inject the real-time data model; the resource scheduling review mechanism will determine whether it needs to start the re-planning process according to the preset business conditions, and if the trigger conditions are met, the system will instruct the dynamic modeling module to obtain a new model snapshot based on the updated real-time data model, and start a complete "modeling-solution-issuing" management cycle to generate a scheduling scheme that is more suitable for the current latest situation.

Images