A steel plate supply chain coordination system based on big data

Abstract

The application discloses a kind of steel plate supply chain coordination systems based on big data, specifically relates to supply chain coordination management technical field, including the following contents: by data acquisition module, real-time acquisition multi-source data is obtained, and dynamic scheduling model is constructed using digital twin module.Based on the model, dynamic scheduling module generates and adjusts production scheduling scheme.Semantic entropy analysis module quantifies the semantic entropy change rate between adjacent versions of scheduling scheme.According to the change rate, the system coordination steady state level is inferred, and is dynamically divided into high, medium and low three steady states, respectively corresponding three scheduling scheme push types, so as to intelligently regulate the rhythm of pushing new scheme to operation terminal.The essence change of scheduling scheme is accurately perceived by the semantic entropy change rate, the interference of invalid push to operation terminal is avoided, the push frequency is matched with the system state by dynamic division, the cognitive load of operation personnel is significantly reduced, and the decision quality and coordination reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of supply chain collaborative management technology, and more specifically, to a big data-based steel plate supply chain collaborative system. Background Technology

[0002] In the collaborative production scenario of the steel plate supply chain, achieving efficient linkage across multiple links from steel mills and processing centers to logistics and distribution is the core objective. Currently, integrating the Internet of Things (IoT), enterprise resource planning (ERP), and manufacturing execution systems (MES) to build a real-time data-driven digital twin model, and then using this model for dynamic scheduling and optimization, has become a key technological path to improve the responsiveness and resource allocation efficiency of this supply chain.

[0003] Existing technologies can achieve dynamic scheduling adjustments based on real-time data updates, but this high-frequency dynamic re-optimization presents a new challenge in human-machine collaboration in practice. The operating terminal needs to continuously process the frequently generated new scheduling schemes by the system, while existing methods for managing the scheme push rhythm mostly rely on fixed time periods or triggering mechanisms based on explicit events such as order changes and equipment failures.

[0004] These methods fail to effectively identify the drastic changes in scheduling schemes at the logical structure and production intent levels. As a result, when only minor adjustments to local parameters occur, operators may receive unnecessary and frequent notifications, leading to distraction and decision fatigue. Conversely, in emergency situations involving substantial restructuring of production strategies, the intensity and timeliness of notifications may be insufficient, posing a risk of delayed critical decisions. Therefore, this invention proposes a big data-based collaborative system for the steel plate supply chain to address these issues. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides the following technical solution: A big data-based collaborative system for steel plate supply chains includes the following steps: The data acquisition module is used to collect multi-source heterogeneous data in the supply chain in real time; The digital twin module constructs and updates a dynamic scheduling digital twin model corresponding to the physical production environment based on collected multi-source heterogeneous data; The dynamic scheduling module performs optimization calculations based on a digital twin model to generate and adjust production scheduling plans; The semantic entropy analysis module is used to analyze and quantify the rate of change of semantic entropy in the logical structure and production intent of the current production scheduling scheme compared to the previous version. The collaborative control module is used to infer the system's collaborative steady-state level based on the semantic entropy change rate, and dynamically divide the scheduling plan push frequency into three types: high-frequency push, medium-frequency push, and low-frequency push according to the level, thereby controlling the rhythm of pushing the re-optimized production scheduling plan to the operation terminal.

[0006] In a preferred embodiment, multi-source heterogeneous data refers to: The data acquisition module collects real-time status data and process parameters of the production equipment through an IoT interface; Connect to external data sources via application programming interfaces (APIs) to collect raw material market data and public logistics information; Integrate order data, inventory data, and quality data through enterprise system interfaces.

[0007] In a preferred embodiment, the process of constructing a dynamic scheduling digital twin model includes: Based on the production equipment status data and process parameters collected by the data acquisition module, a real-time digital model of the production line is constructed. Based on order data, inventory data, and quality data collected by the data acquisition module, a digital model of material status in the supply chain is constructed. Based on raw material market data and logistics public information collected by the data acquisition module, a dynamic digital model of the external environment is constructed. By integrating real-time running digital models, material status digital models, and external environment dynamic digital models, a dynamic scheduling digital twin model corresponding to the physical production environment is generated, and the digital twin model is updated in real time based on multi-source heterogeneous data continuously input from the data acquisition module.

[0008] In a preferred embodiment, during fusion, the real-time running digital model, the material status digital model, and the external environment dynamic digital model are connected through a unified spatiotemporal reference data bus. Based on the flow of materials and information in the physical production process, establish a state mapping and linkage relationship between the real-time running digital model and the material status digital model; The dynamic parameters in the external environment dynamic digital model are used as boundary condition inputs and coupled to the association logic between the real-time running digital model and the material state digital model. By connecting, mapping, and coupling, a unified dynamic scheduling digital twin model that reflects the global real-time status of the physical production environment is generated.

[0009] In a preferred embodiment, the dynamic scheduling module is specifically used for: Based on the digital twin model, optimization objectives are set, including delivery time, production cost, and equipment utilization rate; Load production rules and resource capabilities as constraints; The optimization calculation engine is invoked to solve the optimization objective under constraints in order to generate an initial production scheduling plan. In response to real-time update commands from the digital twin model, the built-in optimization calculation engine is triggered to re-solve the problem, thereby dynamically adjusting the production scheduling scheme.

[0010] In a preferred embodiment, loading production rules and resource capabilities as constraints, and calling the optimization calculation engine to solve the problem, specifically involves: Extract process path rules, equipment compatibility rules, and process sequence rules from the digital twin model to serve as production rule constraints; Extract the maximum equipment capacity, material supply limit, and available man-hours from the digital twin model as resource capacity constraints; The production rule constraints and resource capacity constraints, along with the optimization objectives, are input into the optimization calculation engine. The optimization calculation engine adopts a hybrid solution strategy that includes linear programming and heuristic algorithms. Under the premise of satisfying all constraints, it performs multi-objective trade-off calculations on the optimization objective and outputs an initial production scheduling scheme that meets the requirements.

[0011] In a preferred embodiment, the hybrid solution strategy, which combines linear programming and heuristic algorithms, is specifically as follows: The first stage of linear programming solution steps: Construct a linear programming model with cost and resource utilization as the core objective functions, load constraints and perform preliminary solution to generate a feasible initial scheduling scheme skeleton for resource allocation; The second stage is a heuristic iterative optimization step: Based on the initial scheduling scheme skeleton, a search strategy based on genetic algorithm is used to iteratively optimize the process sequence and equipment assignment in order to optimize the delivery date target in the optimization objectives. Dynamic strategy switching steps: During the second stage, the optimization progress and convergence status are monitored in real time; if the convergence threshold is not reached within the preset number of iterations, the algorithm is dynamically switched to simulated annealing to continue the optimization search until an initial production scheduling scheme that meets the multi-objective trade-off requirements is output.

[0012] In a preferred embodiment, the logic for obtaining the semantic entropy change rate is as follows: From the production scheduling scheme output by the dynamic scheduling module, the optimization target weight set used to generate the scheme is extracted. The optimization target weight set includes specific values ​​of delivery option weight, production cost weight, and equipment utilization rate weight. At the same time, based on the time dependence between processes and the total buffer time, the process dominant path sequence is identified and extracted from the production scheduling scheme. The process dominant path sequence consists of consecutive process nodes with a total buffer time less than a preset time threshold. The process-dominant path sequence is combined with the optimization target weight set and encoded into a scheme feature vector; Calculate the multidimensional cosine distance between the current scheme feature vector and the scheme feature vector of the previous version; Inputting the multidimensional cosine distance into a preset linear mapping formula outputs a scalar value between 0 and 1, which is the semantic entropy change rate.

[0013] In a preferred embodiment, the linear mapping formula is configured as follows: The rate of change of semantic entropy is proportional to the multidimensional cosine distance. When the multidimensional cosine distance is at its minimum, the rate of change of semantic entropy is 0, and when the multidimensional cosine distance is at its preset maximum value, the rate of change of semantic entropy is 1.

[0014] In a preferred embodiment, the inferred cooperative steady-state level of the system refers to: The semantic entropy change rate is compared with the preset first change threshold and second change threshold. The system's coordinated steady state is divided into three levels: high, medium and low, and corresponding to the triggering of three scheduling schemes: low frequency, medium frequency and high frequency.

[0015] The technical effects and advantages of this invention are as follows: This invention introduces the semantic entropy change rate as a quantitative indicator, enabling precise perception and measurement of the degree of essential changes in the logical structure and production intent of production scheduling schemes. This technique effectively distinguishes between subtle adjustments and major restructurings during scheme iteration, thus solving the problem of mechanized push notifications caused by the inability to identify the substance of changes in existing technologies. This avoids distraction and decision-making interference caused by the operator terminal receiving high-frequency push notifications when only minor adjustments are made to the scheme's parameters. Simultaneously, it ensures that critical change information reaches the operator terminal with sufficient intensity and timeliness when production strategies undergo fundamental shifts or when responding to sudden disturbances, guaranteeing the timeliness and accuracy of decision-making.

[0016] This invention utilizes a collaborative control module to automatically infer the system's collaborative steady-state level based on the semantic entropy change rate, and dynamically categorizes the scheduling plan push frequency into high, medium, and low types accordingly, achieving adaptive intelligent adjustment of the human-machine interaction rhythm. This ensures that the intensity and frequency of information pushes match the actual dynamic level within the production system and the fluctuations in the external environment. Consequently, while maintaining the supply chain's rapid response capability to market changes and abnormal events, it significantly reduces the cognitive load and confirmation pressure on key personnel such as planners and team leaders, improves their decision-making quality in complex information environments, and enhances the reliability and execution efficiency of the entire human-machine collaborative loop.

[0017] This invention combines a data-driven dynamic scheduling optimization process with a semantic understanding-based intelligent information push and control process, constructing a complete closed-loop collaborative management mechanism encompassing physical perception, model optimization, and human-computer interaction. This mechanism not only optimizes the scheduling efficiency of physical resources such as equipment, materials, and orders, but also improves the efficiency of transmitting and receiving key instructions from the perspective of information collaboration. This dual-level optimization works together on the complex, multi-stage, long-chain steel plate supply chain system, comprehensively improving the coordination accuracy of each link in the order fulfillment process, the system's anti-interference capability, and the overall operational resilience. Attached Figure Description

[0018] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a schematic diagram of a big data-based collaborative system for steel plate supply chains, as described in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] Reference Figure 1 The following examples were obtained: Example 1: In the intelligent production environment of the modern steel plate supply chain, dynamic scheduling systems can frequently generate updated production scheduling plans by processing and optimizing massive amounts of real-time data to quickly respond to fluctuations in market demand, changes in equipment status, and material supply disturbances. However, this ability to re-optimize at a frequency of minutes or even higher presents a serious challenge in practice: operating terminals (such as team leaders and process engineers) need to continuously receive, understand, and confirm frequently changing and complex scheduling instructions.

[0021] This high-frequency, high-pressure interaction can easily cause key personnel to degenerate from proactive decision-makers into passive "system confirmation organs," leading to cognitive fatigue, decision delays, and even misjudgments, thereby severely weakening the necessary collaborative agility and execution reliability of the supply chain. Existing technical solutions mostly focus on the optimization efficiency and solution speed of the scheduling algorithm itself, but generally lack intelligent management of the human-machine collaborative closed-loop rhythm of "optimization-push-confirmation," failing to dynamically adjust the intensity and frequency of information pushes based on the essential changes in the scheduling scheme and the actual steady-state level of the system. In this context, this invention proposes a big data-based steel plate supply chain collaborative system, including the following steps: The data acquisition module is used to collect multi-source heterogeneous data in the supply chain in real time; The digital twin module constructs and updates a dynamic scheduling digital twin model corresponding to the physical production environment based on collected multi-source heterogeneous data; The dynamic scheduling module performs optimization calculations based on a digital twin model to generate and adjust production scheduling plans; The semantic entropy analysis module is used to analyze and quantify the rate of change of semantic entropy in the logical structure and production intent of the current production scheduling scheme compared to the previous version. The collaborative control module is used to infer the system's collaborative steady-state level based on the semantic entropy change rate, and dynamically divide the scheduling plan push frequency into three types: high-frequency push, medium-frequency push, and low-frequency push according to the level, thereby controlling the rhythm of pushing the re-optimized production scheduling plan to the operation terminal.

[0022] In one specific implementation, multi-source heterogeneous data refers to: the data acquisition module collecting real-time status data and process parameters of production equipment through an IoT interface; connecting to external data sources through an application programming interface to collect raw material market data and logistics public information; and integrating order data, inventory data, and quality data through an enterprise system interface. The implementation of the data acquisition function begins with deep interconnection with physical production equipment, utilizing various sensors and controllers deployed on rolling mills, continuous casting machines, cranes, and intelligent warehousing equipment to collect high-frequency time-series data in real-time via industrial IoT protocols. This data includes, but is not limited to, instantaneous speed of the rolls, rolling pressure, temperature distribution in different zones of the heating furnace, flow rate and pH value of cooling water, and the operating position and lifting weight of the crane. Simultaneously, predictive maintenance sensors installed on critical equipment continuously collect vibration spectrum, noise, and ultrasonic data to monitor the health status of the equipment. All these dynamic parameters captured from the physical world undergo preliminary filtering, compression, and timestamp alignment through an edge computing gateway, and are then transmitted to the data center via industrial Ethernet or a 5G private network, forming a digital mirror of the real-time operating status of the production line.

[0023] Through a series of dedicated data connectors, extensive information is acquired from external open data sources. These connectors periodically call third-party application programming interfaces (APIs) to obtain and parse dynamic market conditions for raw materials, such as spot and futures prices for iron ore and coking coal, as well as market quotations for various alloy materials. Simultaneously, they connect to public interfaces of traffic management platforms and logistics companies to capture real-time road conditions on major transportation arteries, berthing plans and estimated waiting times for ships at ports, and the scheduling dynamics of railway freight hubs. This process involves crawling and parsing unstructured web page data, as well as standardizing the returned JSON or XML format data, thereby transforming uncertainties in the external environment into quantifiable dynamic input variables.

[0024] Seamless integration of business data is achieved by interfacing with the enterprise's existing core business systems. This involves establishing an interface with the enterprise's resource planning system to periodically extract detailed specifications, quantities, delivery locations, and priorities of customer orders; interacting with the manufacturing execution system to obtain the exact location of work-in-process, current process status, labor hours consumed, and preliminary quality inspection results; communicating with the warehouse management system to synchronize real-time inventory quantities, storage location status, and batch information of raw materials, semi-finished products, and finished products; and connecting with the laboratory information management system to retrieve detailed chemical composition analysis reports, mechanical property test data, and metallographic inspection results. The key to this integration lies in resolving the differences in data models and update frequencies between different business systems. By defining unified data mapping rules and incremental synchronization mechanisms, it ensures that the real state of business activities is completely and accurately replicated in the digital space.

[0025] The massive, heterogeneous data streams converged from the three dimensions mentioned above are imported into a unified data preprocessing platform. On this platform, the raw data is cleaned according to predefined data quality rules in existing technologies, outliers and null values ​​are removed, and timestamps from different time zones are uniformly converted to Coordinated Universal Time (UTC). Subsequently, entity resolution technology is used to associate and fuse data from different sources that point to the same physical entity or logical object. For example, temperature data of a steel coil collected by an IoT sensor is associated with process card data of the same steel coil in the Manufacturing Execution System (MES) and its storage location data in the Warehouse Management System. All cleaned, transformed, and enriched data is encapsulated into event messages with a unified pattern and published to a high-throughput distributed message queue, providing a solid, consistent, and real-time data foundation for the subsequent construction of a dynamic digital twin model.

[0026] In one specific implementation, the process of constructing a dynamic scheduling digital twin model includes: constructing a real-time operating digital model of the production line based on the production equipment status data and process parameters collected by the data acquisition module; constructing a material status digital model of the supply chain based on the order data, inventory data, and quality data collected by the data acquisition module; constructing a dynamic digital model of the external environment based on the raw material market data and logistics public information collected by the data acquisition module; integrating the real-time operating digital model, the material status digital model, and the dynamic digital model of the external environment to generate a dynamic scheduling digital twin model corresponding to the physical production environment, and updating the digital twin model in real time based on the multi-source heterogeneous data continuously input by the data acquisition module.

[0027] During integration, a unified spatiotemporal reference data bus connects the real-time running digital model, the material status digital model, and the external environment dynamic digital model. Based on the material and information flow in the physical production process, a state mapping and linkage relationship is established between the real-time running digital model and the material status digital model. Dynamic parameters in the external environment dynamic digital model are used as boundary conditions and coupled to the association logic of the real-time running digital model and the material status digital model. Through connection, mapping, and coupling, a unified dynamic scheduling digital twin model that reflects the global real-time status of the physical production environment is generated.

[0028] Specifically, by utilizing real-time collected process parameters such as the main drive motor current of the rolling mill, inter-stand tension, looper height, and cooling water pressure, combined with the speed and start / stop status feedback from the equipment controller, a digital object with physical attributes is created for each key piece of equipment. These digital objects not only mirror the current operating parameters of the equipment, but also, through embedded mechanistic models (such as heat transfer models and rolling force models) and empirical models, can extrapolate its working state and output capabilities in real time. For example, by integrating the measured temperatures of each temperature zone in the heating furnace with the real-time position of the steel plate in the furnace, the digital model can continuously predict and output the core and surface temperatures of each steel plate when it leaves the heating furnace. This key process state constitutes the initial boundary conditions for subsequent rolling processes, thus forming a high-fidelity, computable real-time digital model of the production line operation.

[0029] Simultaneously, the dynamic flow of materials in the supply chain is synchronously constructed as a digital mirror in another dimension. The steel plate specifications and delivery commitments corresponding to each customer order are transformed into specific production instruction units; the steel type, size, weight, location, and corresponding quality certificate data of each hot-rolled steel coil in the warehouse are instantiated into schedulable material objects; work-in-process is bound to specific process cards and completed quality inspection results according to its actual position on the production line. The key to this model is that it not only records the static attributes of materials, but also dynamically updates their status through real-time events (such as materials being hoisted, put into processing, and inspection completed), thereby accurately tracking the continuous changes in ownership, physical location, and quality attributes of each material from raw materials to finished products, forming a full-link, visualized digital model of material status.

[0030] Furthermore, the external dynamic environment influencing production and logistics decisions is constructed as a quantifiable model. This includes converting the fluctuation trends of iron ore price indices and coking coal spot prices into periodic adjustment parameters for the production cost curve by accessing market data sources; and converting the main road congestion index, the average speed of ships on specific routes, and the expected idle time of reserved berths into probability distribution predictions of logistics transportation segment duration and reliability by integrating traffic and port data. For example, when the model receives a forecast of persistent strong winds in the port area within the next 48 hours, it automatically increases the probability of ship berthing delays and reassesses the delivery risks of transportation routes involving that port accordingly. This part constitutes a continuously evolving dynamic digital model of the external environment.

[0031] In its specific implementation, this invention employs data modeling and simulation techniques well-known to those skilled in the art to construct the real-time digital model of the production line, the digital model of the supply chain material status, and the dynamic digital model of the external environment. The core inventive contribution of this invention lies in linking these independently constructed models using specific methods described below, and innovatively performing semantic entropy analysis and dynamic adjustment of push frequency based on the scheduling scheme generated by this fused model.

[0032] In one specific implementation, three independent digital models—the real-time operation model, the material status model, and the external environment model—are first connected to a real-time data bus with a unified spatiotemporal reference. This bus ensures that events and state changes in all models are synchronized using Coordinated Universal Time (UTC) and mapped to a unified factory geographic coordinate system and logistics route coordinate system. For example, when a batch of hot-rolled steel coils in the material status model is planned to be transported from the warehouse to the port, its planned departure time, transport vehicle identification, and scheduled route information are published to the data bus.

[0033] Based on the actual material and information flow in the physical world, a strict state mapping and causal linkage relationship are established between the above models. This is achieved by configuring a series of complex event processing rules on the data bus. Specifically, when a steel plate discarding event is triggered in the real-time running digital model, the event not only updates the work-in-process status of the model itself, but also automatically triggers the status of the corresponding steel plate entity in the material status digital model to change from "rolling" to "rolled" as a message carrying the unique identifier and completion timestamp of the steel plate, and starts its cooling timer and queuing logic for the next process. This mapping ensures absolute synchronization between production execution and material tracking.

[0034] The time-varying parameters output from the dynamic digital model of the external environment are coupled in real time to the aforementioned linkage logic as key boundary conditions. These parameters are converted into dynamic adjustments to the internal rules or constraints of the model. For example, when the external environment model predicts based on weather forecasts that the wind force in the target port area will rise to level six in the next six hours, it will output a logistics delay risk coefficient. This coefficient is not isolated data, but is coupled into the logistics scheduling rules of the material state model. It may cause the system to automatically advance the associated outbound transportation plan or recalculate the estimated time of the sea voyage, thereby dynamically affecting the predicted time of material arrival at downstream nodes.

[0035] Through the aforementioned connections, mappings, and couplings, three originally independent digital models are integrated into a unified digital twin capable of self-consistency and synchronous evolution. This twin's reflection of the global real-time state of the physical production environment is manifested in its ability to perform cross-model chain deductions for any local disturbance. For example, if a finishing mill experiences a sudden, brief malfunction (represented as a temporary decrease in capacity in the real-time operation model), it can not only immediately recalculate the machine's production queue but also trigger a reassessment of the delivery times of related orders in the material state model through linkage relationships. Furthermore, changes in delivery delay risks may, through a coupling mechanism, stimulate an assessment of external market penalty clauses. This dynamic scheduling digital twin model with global responsiveness provides a unique and reliable factual basis for subsequently calculating the semantic entropy of scheduling schemes and implementing intelligent push frequency control.

[0036] In one specific implementation, the activation of the dynamic scheduling function begins with the structured extraction of all input elements for optimization calculations from a unified dynamic scheduling digital twin model. The optimization objectives are specifically quantified into computable mathematical expressions: the delivery time objective is expressed as minimizing the weighted sum of the deviations between the actual completion times and promised delivery times of all orders, with higher-priority orders assigned higher penalty weights; the production cost objective is decomposed into a linear combination of raw material consumption, energy consumption, and labor costs, with coefficients linked in real-time to market prices provided by the external environment sub-model in the digital twin model; and the equipment utilization objective is expressed as maximizing the fit between the planned load rate and the ideal load curve of key units (such as the finishing mill). These objectives are collectively constructed into a multi-objective function requiring trade-offs.

[0037] A series of mandatory production rules and resource capacity constraints are derived from the same digital twin model. The production rule constraints are specifically defined as follows: process path rules mandate that each type and specification of steel plate must be processed according to its predefined process sequence (e.g., heating, descaling, rough rolling, finish rolling, laminar cooling, coiling), without skipping or reversing the order; equipment compatibility rules ensure that specific product specifications can only be assigned to units with corresponding technical parameters (e.g., roll length, maximum rolling force); and process sequence rules specify the equipment adjustment time necessary for specification switching between consecutive production batches, or the minimum and maximum waiting time windows between preceding and following processes required to ensure product quality. Resource capacity constraints are quantified as: the maximum available machine hours for each piece of equipment within a specified scheduling period, excluding the time spent on planned maintenance and predictive repairs; the daily supply ceiling for various raw materials and auxiliary materials determined based on procurement arrival plans and inventory levels; and the number of workers of different skill levels that can be configured in each work shift.

[0038] The formally defined optimization objective and multi-dimensional constraints are fully input into a dedicated optimization engine. The engine's initial solution process aims to find a feasible solution that strictly satisfies all process and resource constraints. The initial production scheduling output not only includes a precise timetable of when and on which equipment each production instruction will begin and end, but also details the material batches consumed by each process, the estimated energy cost, and the associated quality control points. The solution is presented in a machine-readable and human-intuitive data structure, such as a complete report containing detailed cost estimates.

[0039] The core dynamism of this scheduling function lies in its continuous response to real-time changes in the digital twin model. When the digital twin model detects a change in the physical world's state and issues an update instruction—for example, the material status sub-model reports a 24-hour delay in the arrival of a batch of critical alloy materials, or the real-time running sub-model warns of an abnormal bearing temperature in a straightening machine requiring early intervention and maintenance—it immediately triggers a new optimization solution cycle. The response mechanism is not simply a complete recalculation, but rather first assesses the scope and urgency of the change event to determine whether to quickly fix a local solution or initiate a global replanning. For example, for a brief failure of a single piece of equipment, the optimization engine may only reorder and reassign the processes involving that equipment within its affected time window; while for a raw material shortage event affecting the entire system, it may trigger a regeneration of the entire scheduling cycle, prioritizing the material allocation for high-priority orders. Through this continuous closed-loop interaction with the digital twin model, production scheduling can agilely adapt to fluctuations in the internal and external environment, always maintaining its timeliness and executability.

[0040] In one specific implementation, the hybrid solution strategy, which incorporates linear programming and heuristic algorithms, is as follows: The first stage of linear programming solution steps: Construct a linear programming model with cost and resource utilization as the core objective functions, load constraints and perform preliminary solution to generate a feasible initial scheduling scheme skeleton for resource allocation; The second stage is a heuristic iterative optimization step: Based on the initial scheduling scheme skeleton, a search strategy based on genetic algorithm is used to iteratively optimize the process sequence and equipment assignment in order to optimize the delivery date target in the optimization objectives. Dynamic strategy switching steps: During the second stage, the optimization progress and convergence status are monitored in real time; if the convergence threshold is not reached within the preset number of iterations, the algorithm is dynamically switched to simulated annealing to continue the optimization search until an initial production scheduling scheme that meets the multi-objective trade-off requirements is output.

[0041] Specifically, the first stage of the hybrid solution strategy focuses on constructing a basic framework for a feasible and cost-effective resource allocation scheme. This stage is accomplished by establishing a linear programming model. The model construction follows mature linear programming modeling methods in operations research, and its objective function aims to quantify and optimize production costs and resource utilization. Specifically, the cost minimization objective can be achieved by constructing a linear function with raw material consumption, energy usage, and labor costs as variables; the resource utilization maximization objective can be transformed into minimizing the idle time or load imbalance of key equipment. Those skilled in the art will understand that, based on the above-mentioned objectives, there are various equivalent mathematical function forms that can be used to construct this objective function, all of which fall within the scope of existing technology and are applicable to the implementation of this invention. The model's constraints precisely incorporate resource capacity limitations extracted from the digital twin model, such as the total available labor hours of all equipment, the maximum daily supply of various materials, and the basic requirement constraint that all orders must be fulfilled. By solving this linear programming model, a globally optimal allocation scheme at the resource level can be obtained. This scheme determines the proportion of equipment capacity to be occupied by each order category, the approximate material consumption plan, and the cost baseline, forming a macro-economical and resource-feasible scheduling scheme framework. However, the precise timing of the processes and the details of equipment assignment have not yet been determined.

[0042] After obtaining the resource allocation framework, the solution process enters the second stage, which involves using a genetic algorithm to perform refined iterative optimization of the precise ordering of processes and equipment assignment. This stage focuses on the delivery date target as the core optimization object. First, the scheduling problem is encoded as individual chromosomes in the genetic algorithm. Chromosome genes can represent the order of processes or the pairing relationship between processes and equipment. The initial population consists of several feasible solutions generated based on the framework from the first stage. In each generation of evolution, the algorithm exchanges some process sequences in different individual chromosomes through crossover operations and randomly adjusts the positions of certain processes or assigns equipment through mutation operations, thereby exploring new solution spaces. The fitness function calculates the total delay time or number of delayed orders based on the scheduling scheme represented by each individual. Individuals with higher fitness (i.e., schemes with fewer delays) have a greater probability of being selected to reproduce the next generation. Through multiple generations of selection, crossover, and mutation, the algorithm can gradually evolve a detailed scheduling scheme that significantly improves on-time order delivery performance while satisfying the given resource allocation.

[0043] However, the search process of genetic algorithms may get stuck in local optima and stagnate. To address this challenge, this strategy incorporates a dynamic strategy switching step. During this process, the improvement in the optimal fitness of the genetic algorithm population over multiple generations is continuously monitored. If the improvement in optimal fitness falls below a very small threshold for more than a preset number of generations (e.g., 50 generations), the search is considered to have plateaued. At this point, the solution process automatically switches from the genetic algorithm to simulated annealing for further optimization. Simulated annealing uses the current best solution obtained by the genetic algorithm as the initial solution and generates new solutions by randomly perturbing the current solution, such as randomly swapping the processing order of two steps. According to the Metropolis criterion specific to simulated annealing, even if the new solution is worse than the current solution (i.e., worse delivery time performance), there is a certain probability that it will be accepted. This acceptance probability decreases as a parameter called "temperature" gradually decreases. This mechanism allows the algorithm to explore widely outside local optima in the early stages and gradually converge to a better solution in the later stages. This dynamic switching mechanism ensures the robustness of the optimization search and avoids the inability to find a satisfactory solution due to the limitations of a single algorithm.

[0044] Through the aforementioned phased, dynamically switching hybrid solution process, a production scheduling scheme that satisfies multiple objective trade-offs can be output. This scheme not only inherits the macro-level advantages of the first-stage linear programming model in resource utilization and cost control, but also incorporates the second-stage heuristic algorithm for deep optimization of delivery time targets at the process-level scheduling. The output scheme is a complete and executable set of scheduling instructions that includes the start and end times of each process on specific equipment, logical relationships between processes, a bill of materials, and expected costs and delivery performance. This scheme serves as the input basis for semantic entropy analysis and subsequent dynamic control, and its generation quality directly affects the efficiency of the entire collaborative process.

[0045] In one specific implementation, the acquisition of semantic entropy change rate begins with a deep analysis of the production scheduling scheme output by the dynamic scheduling module to extract its inherent optimization preferences and structural critical paths. First, the set of optimization target weights used to generate the scheme is extracted from the metadata accompanying the scheduling scheme. The specific values ​​of this weight set, such as a delivery option weight of 0.5, a production cost weight of 0.3, and an equipment utilization rate weight of 0.2, are set by the planner through a configuration interface before each scheduling optimization, or dynamically calculated by the upper-level decision support logic based on the current business strategy (such as a rush delivery season or a cost control month). This set of weights clearly quantifies the relative importance of each performance indicator in this optimization. Simultaneously, the process network of the scheduling scheme itself is analyzed, calculating the sum of the buffer time between processes on each path (i.e., the difference between the earliest and latest start times of a process). Continuous process nodes with a total buffer time less than a preset time threshold (e.g., 4 hours) are identified as the process-dominant path sequence. This path is the most strained and least flexible chain in the scheduling, and its changes directly reflect the core changes in the production logic. By analyzing a specific scheduling scheme, it is possible to extract a dominant path sequence such as "heating furnace A -> roughing mill R2 -> finishing mill F5 -> laminar cooling line C1".

[0046] The extracted discretized information is encoded into machine-measurable and comparable numerical feature vectors. For the process-dominant path sequence, one-hot encoding or embedding techniques based on process and equipment numbers are used to convert it into a set of fixed-length numerical vectors. For example, in a workshop containing 10 key pieces of equipment, "Heating Furnace A" might be encoded as [1,0,0,0,0,0,0,0,0,0], while "Roughing Mill R2" might be encoded as [0,0,1,0,0,0,0,0,0,0]. The encoding of the entire path sequence is the sequential concatenation of these vectors. Meanwhile, the optimization target weight set itself constitutes a three-dimensional vector, such as [0.5,0.3,0.2]. Finally, the path encoding vector and the weight vector are concatenated to form a high-dimensional scheme feature vector representing the logical structure and optimization intent of the scheduling scheme.

[0047] The degree of difference between the current scheduling scheme and the previous version is calculated using a vector space approach. Specifically, the feature vectors of the scheme generated at the current time and the scheme generated in the previous period are obtained, and the cosine value of the angle between these two high-dimensional vectors is calculated. The closer the cosine value is to 1, the more consistent the directions of the two vectors, meaning the schemes are more similar in structure and intent; conversely, the closer the value is to 1, the greater the difference. To obtain an intuitive distance metric, the formula "distance = 1 - cosine similarity" is typically used to calculate the multidimensional cosine distance. This distance value is a scalar between 0 and 2, where 0 represents that the feature vectors of the two schemes are completely in the same direction, meaning there is no semantic change.

[0048] The calculated multidimensional cosine distance is input into a predefined linear mapping formula, which normalizes and transforms it into a more interpretable semantic entropy change rate. A specific form of this linear mapping formula is: Semantic Entropy Change Rate = Multidimensional Cosine Distance / D_max. Here, D_max is a predefined maximum distance reference value, which can be set based on the maximum distance observed in historical data or the theoretically possible maximum value (e.g., 2). This formula satisfies the condition that when the multidimensional cosine distance is 0, the semantic entropy change rate is 0; when the distance reaches D_max, the change rate is 1. For example, if D_max is set to 2, and the calculated current distance is 0.8, then the semantic entropy change rate is 0.4. This scalar value between 0 and 1 ultimately quantifies the drastic change in the core logic and decision-making intent of the production scheduling scheme during continuous optimization iterations, providing a key input for subsequent intelligent control.

[0049] In one specific implementation, the collaborative control function begins with the real-time reception and parsing of continuously incoming semantic entropy change rates. This change rate, as a quantifiable scalar, directly reflects the drastic degree of change in the production scheduling plan at the logical and intentional levels. To transform this abstract metric into actionable control instructions, judgments are made based on two pre-defined thresholds with clear physical meaning. The setting of the first and second change thresholds is not arbitrary but is determined jointly based on statistical analysis of historical production operation data, domain expert experience, and empirical research on operator cognitive load. For example, analysis of hundreds of scheduling change events over the past year revealed that when the semantic entropy change rate is below 0.2, plan adjustments are mostly minor tweaks to local parameters, and operators generally report no pressure in understanding and following up; when the change rate is between 0.2 and 0.5, the plan typically undergoes a moderate degree of restructuring, requiring operators to invest moderate attention in review; when the change rate exceeds 0.5, it often corresponds to a major shift in production strategy or an emergency rescheduling to address unforeseen events, demanding extremely high cognitive and decision-making speed from operators. Based on this analysis, the first change threshold can be set to 0.2 and the second change threshold can be set to 0.5, which can be used as an objective benchmark for classifying the cooperative steady-state level of the system.

[0050] The semantic entropy change rate calculated in real time is automatically compared with the two thresholds mentioned above. This comparison process is continuous and dynamic; a comparison logic is triggered immediately each time a new scheduling scheme is generated and its corresponding change rate is calculated. The purpose of the comparison is to classify continuous numerical change rates into discrete steady-state levels with clear regulatory implications. Specifically, if the current semantic entropy change rate is less than the first change threshold (e.g., 0.2), it is inferred that the entire "scheduling-execution" collaborative loop is in a high collaborative steady-state level. This level indicates that the production plan is relatively stable, the differences between successive iterations of the scheme are minor, and the internal and external environmental fluctuations of the system are small. Based on this inference, low-frequency push-type regulatory commands will be automatically triggered.

[0051] When the rate of change of semantic entropy falls between the first and second change thresholds (e.g., greater than or equal to 0.2 and less than 0.5), the system is inferred to be in a medium-level cooperative steady state. This typically indicates a noticeable disturbance, such as adjustments to the priority of multiple orders or slight fluctuations in the efficiency of some equipment, leading to structural adjustments to the scheduling scheme with a certain impact, but without triggering a complete refactoring. For this level, a medium-frequency push-type control command will be triggered.

[0052] If the rate of change of semantic entropy is greater than or equal to the second change threshold (e.g., 0.5), the system is inferred to be in a low cooperative steady-state level, i.e., a highly dynamic and highly uncertain state. This may be due to sudden critical equipment failure, disruption of the core raw material supply chain, or important emergency orders, resulting in a fundamental change in the scheduling scheme. At this level, high-frequency push-type control commands will be triggered.

[0053] The type of push notification triggered by different steady-state levels directly determines the rhythm and form in which the re-optimized production scheduling plan is sent to the operator terminal. For high-frequency push notifications, the system may push the latest scheduling plan to relevant operators at a frequency of one minute or even higher, supplemented by prominent visual and auditory cues to ensure close follow-up on rapidly evolving situations. For medium-frequency push notifications, the push frequency may be reduced to once every 5 to 10 minutes, with a brief change summary attached to the push notification, prompting operators to pay attention to key changes. For low-frequency push notifications, the push frequency may be further reduced to once every 15 minutes or longer, or even allow operators to actively pull the latest plan at their convenience, thereby minimizing unnecessary interference to personnel in stable working environments. Through this step-by-step dynamic adjustment based on quantitative indicators, the system effectively manages and optimizes the load and efficiency of human-machine interaction while ensuring scheduling agility.

[0054] The above algorithms or formulas are all dimensionless and numerical calculations, and the results are obtained by software simulation based on a large amount of collected data to obtain the most recent real-world results. The preset parameters are set by those skilled in the art according to the actual situation.

[0055] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0056] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0057] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0058] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A big data-based collaborative system for steel plate supply chains, characterized in that, Includes the following steps: The data acquisition module is used to collect multi-source heterogeneous data in the supply chain in real time; The digital twin module constructs and updates a dynamic scheduling digital twin model corresponding to the physical production environment based on collected multi-source heterogeneous data; The dynamic scheduling module performs optimization calculations based on a digital twin model to generate and adjust production scheduling plans; The semantic entropy analysis module is used to analyze and quantify the rate of change of semantic entropy in the logical structure and production intent of the current production scheduling scheme compared to the previous version. The collaborative control module is used to infer the system's collaborative steady-state level based on the semantic entropy change rate, and dynamically divide the scheduling plan push frequency into three types: high-frequency push, medium-frequency push, and low-frequency push according to the level, thereby controlling the rhythm of pushing the re-optimized production scheduling plan to the operation terminal.

2. The steel plate supply chain collaborative system based on big data according to claim 1, characterized in that, Multi-source heterogeneous data refers to: The data acquisition module collects real-time status data and process parameters of the production equipment through an IoT interface; Connect to external data sources via application programming interfaces (APIs) to collect raw material market data and public logistics information; Integrate order data, inventory data, and quality data through enterprise system interfaces.

3. The steel plate supply chain collaborative system based on big data according to claim 2, characterized in that, The process of building a dynamic scheduling digital twin model includes: Based on the production equipment status data and process parameters collected by the data acquisition module, a real-time digital model of the production line is constructed. Based on order data, inventory data, and quality data collected by the data acquisition module, a digital model of material status in the supply chain is constructed. Based on raw material market data and logistics public information collected by the data acquisition module, a dynamic digital model of the external environment is constructed. By integrating real-time running digital models, material status digital models, and external environment dynamic digital models, a dynamic scheduling digital twin model corresponding to the physical production environment is generated, and the digital twin model is updated in real time based on multi-source heterogeneous data continuously input from the data acquisition module.

4. The steel plate supply chain collaborative system based on big data according to claim 3, characterized in that, During integration, a unified spatiotemporal reference data bus is used to connect the real-time running digital model, the material status digital model, and the external environment dynamic digital model. Based on the flow of materials and information in the physical production process, establish a state mapping and linkage relationship between the real-time running digital model and the material status digital model; The dynamic parameters in the external environment dynamic digital model are used as boundary condition inputs and coupled to the association logic between the real-time running digital model and the material state digital model. By connecting, mapping, and coupling, a unified dynamic scheduling digital twin model that reflects the global real-time status of the physical production environment is generated.

5. A big data-based steel plate supply chain collaborative system according to claim 4, characterized in that, The dynamic scheduling module is specifically used for: Based on the digital twin model, optimization objectives are set, including delivery time, production cost, and equipment utilization rate; Load production rules and resource capabilities as constraints; The optimization calculation engine is invoked to solve the optimization objective under constraints in order to generate an initial production scheduling plan. In response to real-time update commands from the digital twin model, the built-in optimization calculation engine is triggered to re-solve the problem, thereby dynamically adjusting the production scheduling scheme.

6. A big data-based steel plate supply chain collaborative system according to claim 5, characterized in that, The process of loading production rules and resource capabilities as constraints, and calling the optimization calculation engine to solve the problem, is as follows: Extract process path rules, equipment compatibility rules, and process sequence rules from the digital twin model to serve as production rule constraints; Extract the maximum equipment capacity, material supply limit, and available man-hours from the digital twin model as resource capacity constraints; The production rule constraints and resource capacity constraints, along with the optimization objectives, are input into the optimization calculation engine. The optimization calculation engine adopts a hybrid solution strategy that includes linear programming and heuristic algorithms. Under the premise of satisfying all constraints, it performs multi-objective trade-off calculations on the optimization objective and outputs an initial production scheduling scheme that meets the requirements.

7. A big data-based steel plate supply chain collaborative system according to claim 6, characterized in that, The hybrid solution strategy, which combines linear programming and heuristic algorithms, is as follows: The first stage of linear programming solution steps: Construct a linear programming model with cost and resource utilization as the core objective functions, load constraints and perform preliminary solution to generate a feasible initial scheduling scheme skeleton for resource allocation; The second stage is a heuristic iterative optimization step: Based on the initial scheduling scheme skeleton, a search strategy based on genetic algorithm is used to iteratively optimize the process sequence and equipment assignment in order to optimize the delivery date target in the optimization objectives. Dynamic strategy switching steps: During the second phase, monitor the optimization progress and convergence status in real time; If the convergence threshold is not reached within the preset number of iterations, the algorithm will be dynamically switched to simulated annealing to continue the optimization search until an initial production scheduling scheme that meets the multi-objective trade-off requirements is output.

8. A big data-based steel plate supply chain collaborative system according to claim 7, characterized in that, The logic for obtaining the semantic entropy change rate is as follows: From the production scheduling scheme output by the dynamic scheduling module, the optimization target weight set used to generate the scheme is extracted. The optimization target weight set includes specific values ​​of delivery option weight, production cost weight, and equipment utilization rate weight. At the same time, based on the temporal dependency relationship between processes and the total buffer time, the process dominant path sequence is identified and extracted from the production scheduling scheme. The process dominant path sequence consists of consecutive process nodes with a total buffer time less than a preset time threshold. The process-dominant path sequence is combined with the optimization target weight set and encoded into a scheme feature vector; Calculate the multidimensional cosine distance between the current scheme feature vector and the scheme feature vector of the previous version; Inputting the multidimensional cosine distance into a preset linear mapping formula outputs a scalar value between 0 and 1, which is the semantic entropy change rate.

9. A big data-based steel plate supply chain collaborative system according to claim 8, characterized in that, The linear mapping formula is configured as follows: The rate of change of semantic entropy is proportional to the multidimensional cosine distance. When the multidimensional cosine distance is at its minimum, the rate of change of semantic entropy is 0, and when the multidimensional cosine distance is at its preset maximum value, the rate of change of semantic entropy is 1.

10. A big data-based steel plate supply chain collaborative system according to claim 9, characterized in that, The inference of the cooperative steady-state level of the system refers to: The semantic entropy change rate is compared with the preset first change threshold and second change threshold. The system's coordinated steady state is divided into three levels: high, medium and low, and corresponding to the triggering of three scheduling schemes: low frequency, medium frequency and high frequency.

Images