A Production-Stock Decision Generation Method and System Based on a Multi-Level Dynamic Coupling Model

Abstract

This invention discloses a production-inventory decision generation method and system based on a multi-layer dynamic coupling model, belonging to the field of production decision-making technology. By constructing a three-layer dynamic coupling time-delay differential equation model (material layer, production layer, and product layer), this invention fundamentally changes the traditional static, hierarchical, and passive production control model. A cross-layer coupled controller composed of a dynamic safety stock algorithm and a flexible capacity allocator enables real-time collaborative optimization and forward-looking decision-making of materials, capacity, and inventory. High-precision numerical solutions are achieved using the fourth-order Runge-Kutta method, allowing for digital extrapolation and risk prediction of future states, thereby proactively triggering precise procurement and production scheduling instructions. This effectively suppresses the "bullwhip effect," improves inventory turnover and capacity utilization, and significantly enhances the overall economic efficiency and operational resilience of the production system in the face of supply chain fluctuations and market changes.

Description

Technical Field

[0001] This invention relates to the field of production decision-making technology, and in particular to a production-stock decision generation method and system based on a multi-layer dynamic coupling model. Background Technology

[0002] Currently, the most common static control methods for production and inventory management in various manufacturing industries are based on Material Control Planning (MRP) to fix production and delivery cycles, and using a static Bill of Materials (BOM) to quantitatively describe the relationship between procurement and production consumables. Meanwhile, at the hierarchical decision-making stage, aspects such as material procurement, production planning, and inventory management operate independently, coordinating work through communication methods such as email and telephone, leveraging management experience.

[0003] The existing technical solutions have several obvious drawbacks. First, MRP has a fixed lead time, which cannot effectively cope with fluctuating production in terms of supply chain delay handling. Second, it is a passive demand response and cannot respond to changes in inventory trends in a timely manner. Third, capacity scheduling follows priorities and cannot effectively balance production line delivery rate and utilization rate. Fourth, multi-factory coordination cannot be unified and requires independent operation interspersed with manual coordination, which greatly delays decision-making time and further reduces production efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a production-stock decision generation method and system based on a multi-layer dynamic coupling model, so as to improve the technical problems of insufficient dynamic response of the supply chain, passive and lagging inventory control, rigid capacity allocation, and difficulty in multi-system coordination in the existing technology.

[0005] To achieve the above-mentioned objectives, the embodiments of the present invention provide the following technical solutions:

[0006] A production-stock decision generation method based on a multi-layer dynamic coupling model, comprising:

[0007] Collect material data, production data, and product data of optical modules to construct a multi-layer dynamic coupling model of the optical module production system; the multi-layer dynamic coupling model includes dynamic equations for the material layer, dynamic equations for the production layer, and dynamic equations for the product layer.

[0008] The dynamic safety stock algorithm and flexible capacity allocator generate the target inventory and planned capacity allocation for materials at the current time step.

[0009] Based on the target inventory of materials and the planned capacity allocation, the multi-layer dynamic coupling model is solved by the fourth-order Runge-Kutta method to generate production-inventory decisions and feed them back to the optical module production system.

[0010] Furthermore, the material data includes the purchase order placement rate, inventory level, and material consumption coefficient and production start-up rate corresponding to each optical module model;

[0011] The production data includes the maximum capacity of each production segment, standard production cycle, changeover preparation time, yield of each process segment, overall equipment efficiency, and work-in-process quantity.

[0012] The product data includes finished product inventory, demand volatility, and the proportion of each optical module model produced.

[0013] In the above-mentioned scheme, this invention solves the "bullwhip effect" and response lag problems caused by fixed parameters and hierarchical fragmentation in traditional production systems by constructing a multi-layer dynamic coupling model that integrates time-varying lead time and cross-layer feedback. Through a dynamic safety stock algorithm, product-level trends are fed back to the target material inventory in real time, realizing a shift in inventory strategy from static defense to dynamic foresight, effectively addressing demand fluctuations and supply chain disruption risks. The flexible capacity allocator, by integrating value orientation and inventory calibration, intelligently balances profit maximization and inventory health under limited capacity, overcoming the problem of rigid capacity allocation. Combined with the fourth-order Runge-Kutta method for high-precision rolling solution of the time-delay model, the production system possesses the ability to proactively simulate and control complex factors such as production delays and yield losses, ultimately forming an intelligent decision-making closed loop that can adapt to fluctuations, collaboratively optimize, and continuously learn, significantly improving supply chain resilience, resource utilization efficiency, and overall operational effectiveness.

[0014] Furthermore, the multi-layer dynamic coupling model for constructing the optical module production system includes:

[0015] Collect material data, production data, product data, and market data for optical modules, and generate a chip shortage index based on the market data;

[0016] Based on the chip shortage index and the time-varying procurement base coefficient, the time-varying procurement lead time is calculated, and combined with material data, a material-level dynamic equation is constructed.

[0017] Based on production data, calculate the input rate and completion rate at the current time step;

[0018] Based on input rate, completion rate and material data, the dynamic relationship between capacity, materials, yield and production progress in the production section of the production line is determined, and the dynamic equation of the production layer is constructed.

[0019] Obtain basic demand forecasts for optical module models, and combine these forecasts with the demand volatility of optical module models to calculate actual demand using a lifecycle function;

[0020] Based on the finished product inventory, proportion, actual demand, and completion rate of optical module models, the dynamic relationship between products and demand is determined, and the dynamic equation of the product layer is constructed through differential operations.

[0021] In the above-described scheme, this invention calculates time-varying procurement lead times, enabling the material layer equations to respond to supply chain risks in real time. This solves the core defect of traditional models, which cannot adapt to external market fluctuations due to fixed lead times. In the production layer construction, by comprehensively considering multiple factors such as capacity limits, material constraints, yield losses, and production delays, the dynamics of the production line are accurately depicted, overcoming the problem of traditional production scheduling models neglecting resource coupling and process lags. The product layer integrates demand volatility and product lifecycle functions to achieve quantitative modeling of market characteristics such as rapid iteration and demand decay of optical modules. These three mutually coupled layers based on differential equations constitute a dynamic, closed-loop system model. Its core value lies in the real-time linkage and digital extrapolation of external market fluctuations, internal production constraints, and product lifecycles, thereby achieving a fundamental shift from passive response to proactive anticipation, and from local optimization to global collaboration. Ultimately, this significantly improves the system's decision-making accuracy and supply chain resilience in volatile market environments.

[0022] Furthermore, the construction of the material layer dynamic equations includes:

[0023] Based on the chip shortage index and the time-varying procurement base coefficient of each material, the corresponding time-varying procurement lead time is calculated.

[0024] Based on the time-varying procurement lead time and purchase order placement rate of each material, calculate the actual material inbound rate in the current time step;

[0025] Based on the material consumption coefficient and production start-up rate corresponding to each optical module model, the total consumption rate of all optical module models for each material is calculated by a summation function.

[0026] Based on the total consumption rate and material inbound rate of each material, the dynamic relationship between the material inventory level of each material and the inflow / outflow rate is constructed through differential operations, generating the material layer dynamic equation.

[0027] In the above solution, this invention introduces time-varying procurement lead time calculation based on chip shortage index, enabling the material inventory model to dynamically respond to external supply chain risks. This solves the problem that traditional MRP systems, which use fixed lead times, cannot adapt to market fluctuations. By combining the purchase order placement rate with the dynamic lead time to calculate the actual inbound rate, and accurately calculating the total material consumption rate based on real-time production start-up rate and consumption coefficient, a real-time and accurate characterization of material flow is achieved. Finally, by constructing a dynamic relationship between inventory level and inbound / outbound flow through differential equations, a dynamic material-level model is formed that can quantitatively reflect the supply chain delay effect and production coupling impact. This model can detect supply chain risks in advance and accurately simulate material flow status, thereby upgrading material management from static estimation based on historical experience to dynamic and forward-looking control based on real-time data and market signals. This effectively solves the problem of material shortages or backlogs caused by supply chain fluctuations and changes in production plans.

[0028] Furthermore, the processing procedure of the dynamic safety stock algorithm is as follows:

[0029] Historical demand data is acquired and fitted using a sliding window and integration operations to generate effective demand volatility.

[0030] Obtain historical data and the target inventory of each material in the previous time step, and generate cross-level compensation gain through a sequential quadratic programming algorithm;

[0031] Optimize server performance by controlling stockout costs;

[0032] A basic safety stock model is constructed based on the server level coefficient, effective demand volatility, and time-varying procurement lead time of each material.

[0033] Based on the basic safety stock model, cross-layer compensation gain, and product-layer dynamic equations, update the target inventory of each material at the current time step.

[0034] In the above scheme, this invention enables the safety stock model to continuously learn from historical data through sliding window fitting and sequential quadratic programming optimization, solving the problem that traditional static formulas cannot adapt to market trend drift due to fixed parameters. It introduces cross-layer compensation gain based on the product layer inventory change rate, and transforms downstream sales fluctuation trends into upstream inventory adjustment signals in real time through a dynamic coupling mechanism, effectively breaking the "bullwhip effect" caused by the lag in information transmission between traditional levels. At the same time, it integrates the time-varying procurement lead time and the self-optimized service coefficient into the basic safety stock calculation, realizing a dynamic quantitative response to external supply chain risks and internal service strategies. This transforms the inventory strategy from a passive defense based on historical averages to a proactive and forward-looking regulation that integrates real-time feedback, cross-layer collaboration, and continuous learning. This not only significantly reduces the risk of stockouts or backlogs caused by sudden changes in demand or supply interruptions, but also fundamentally improves the responsiveness and operational resilience of the entire supply chain system.

[0035] Furthermore, the processing procedure of the flexible capacity allocator is as follows:

[0036] Calculate the corresponding target inventory of products based on the real-time demand of each optical module model;

[0037] Obtain the standard processing time for each optical module model, and calculate the profit margin for each optical module model based on market data;

[0038] Based on the profit margin, standard processing time, and maximum capacity of each production segment for each optical module model, the value orientation of the production line is determined using the Logit model.

[0039] Based on product target inventory, production line value orientation, and inventory calibration, a flexible capacity allocator is constructed to calculate the planned capacity allocation for production segments within the production line.

[0040] In the above solution, this invention constructs a flexible capacity allocator based on the Logit model's production line value orientation, effectively solving the profit and inventory imbalance problem caused by relying on fixed priorities or empirical rules in traditional capacity allocation. First, it quantifies the "unit labor hour value" of a product by the ratio of profit margin to processing time, and uses the Logit model to convert this into a capacity allocation probability, ensuring that capacity is always prioritized for high-value, high-efficiency products, thus improving overall profitability. Second, it introduces the deviation between target inventory and actual inventory as a feedback term to dynamically adjust the capacity allocation ratio of each product, achieving rapid response to market demand fluctuations and avoiding inventory backlogs or shortages. This allows capacity allocation to simultaneously consider economic benefits and operational stability, overcoming the difficulty of achieving a dynamic balance between "maximizing profits" and "maintaining healthy inventory" in traditional scheduling. It is particularly suitable for flexible production scenarios with large demand fluctuations and rapid product iterations.

[0041] Furthermore, the generation of production-stock decisions includes:

[0042] Set the initial state vector and the total number of time steps; the initial state vector includes the initial inventory level of the material layer, the initial work-in-process quantity of the production layer, and the initial finished goods inventory of the product layer.

[0043] Based on the multi-layer dynamic coupling model, the first intermediate slope estimate, second intermediate slope estimate, third intermediate slope estimate and fourth intermediate slope estimate of different layers at the current time step are calculated by the classical fourth-order Runge-Kutta method.

[0044] Based on the first, second, third, and fourth intermediate slope estimates of different layers, the initial state vector is updated to obtain the state vector.

[0045] Based on state vectors, target material inventory, and planned capacity allocation, production-inventory decisions are generated.

[0046] In the above scheme, this invention employs the classic fourth-order Runge-Kutta method (RK4), a high-precision numerical integration technique, to iteratively solve complex differential equations containing time delays and cross-layer coupling terms. By calculating four weighted intermediate slopes, it accurately predicts the future state of the system, significantly improving the simulation realism of state evolution under multiple constraints (such as supply chain delays, production bottlenecks, and demand fluctuations). This provides a reliable quantitative forecast for intelligent decision-making and fundamentally solves the model distortion and decision bias problems caused by traditional simplified algorithms (such as the Euler method) neglecting higher-order terms and time delay effects. It ensures the scientificity and effectiveness of production planning and inventory control instructions, thereby mathematically guaranteeing the reliability of the entire intelligent management and control system's decision-making in dynamic and complex environments.

[0047] A production-stock decision generation system based on a multi-layer dynamic coupling model includes:

[0048] The data acquisition module is used to collect material data, production data, and product data of the optical module.

[0049] The calculation module is used to generate the target inventory of materials and the planned capacity allocation for the current time step through a dynamic safety stock algorithm and a flexible capacity allocator.

[0050] The model building module is used to construct a multi-layered dynamic coupling model of the optical module production system based on material data, production data, and product data.

[0051] The solver module is used to solve the multi-layer dynamic coupling model based on the target inventory of materials and the planned capacity allocation, using the fourth-order Runge-Kutta method, to generate production-inventory decisions and feed them back to the optical module production system.

[0052] Furthermore, it also includes a stability and convergence judgment module, which is used to perform steady-state and convergence judgments on the optical module production system.

[0053] In the aforementioned scheme, the system integrates dynamic modeling, intelligent algorithms, and high-precision numerical solutions into a closed-loop decision-making process, achieving a fundamental shift in production and inventory management from static fragmentation to dynamic collaboration. Its core value lies in its real-time response to supply chain fluctuations and demand changes through time-varying parameters and cross-layer feedback. It utilizes coupled models for forward-looking simulation and deduction, and leverages a self-learning mechanism to continuously optimize decision parameters, thereby effectively suppressing the "bullwhip effect" and significantly improving inventory turnover, capacity utilization, and order delivery timeliness. Ultimately, it forms a production control system that can adapt to environmental changes, intelligently balance multiple objectives, and continuously evolve, providing highly resilient and efficient intelligent decision support for flexible manufacturing scenarios involving multiple varieties and small batches. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of the method in an embodiment of the present invention;

[0056] Figure 2 This is a system structure diagram in an embodiment of the present invention. Detailed Implementation

[0057] 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0058] Please see Figure 1 This embodiment provides a production-stock decision generation method based on a multi-layer dynamic coupling model. Figure 1The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not impose any limitations on this.

[0059] A production-stock decision generation method based on a multi-layer dynamic coupling model includes:

[0060] S1. Collect material data, production data, and product data of the optical module, and construct a multi-layer dynamic coupling model of the optical module production system; the multi-layer dynamic coupling model includes material layer dynamic equations, production layer dynamic equations, and product layer dynamic equations.

[0061] Specifically, for system dynamics, the core of basic modeling is the relationship between stock and flow, which can be simplified as: Stock = Inflow rate - Outflow rate. In the production process of an optical module manufacturing system, its basic hierarchical logic can be simplified to a material layer, a production layer, and a final product layer. By dividing the system by physical location and functional hierarchy, the "flattening distortion" of traditional Material Requirements Planning (MRP) can be further avoided.

[0062] Therefore, S1 includes:

[0063] S1-1. Collect material data, production data, product data, and market data for the optical module. Based on the market data, generate a chip shortage index. ;

[0064] The material data includes each material Purchase orders Inventory levels and optical module model Corresponding material consumption coefficient and production start-up rate ;

[0065] The production data includes data from each production segment. The equipment parameters, i.e., maximum capacity. Standard production cycle Preparation time for model change Process segment yield Overall equipment efficiency and work-in-process quantity ;

[0066] The product data includes various optical module models. Finished goods inventory Demand volatility and in the production section Medium-sized optical module models proportion ;

[0067] Chip shortage index The value range is (0, 100), and it is dynamically updated based on real-time market data. For example, assuming the current chip delivery time is 8 weeks and the historical average delivery time is 4 weeks, the delivery time extension ratio R is calculated as 8 / 4 = 2. Similarly, if R <= 1, then the chip shortage index... If R>=3, then the chip shortage index is 0. The value is 100. If it falls between (1, 3), the chip shortage index can be generated through linear interpolation. ;

[0068] S1-2. Based on the chip shortage index and the time-varying procurement base coefficient, calculate the time-varying procurement lead time, and construct the material layer dynamic equation by combining material data.

[0069] Specifically, for a production line, changes in material inventory can be simply understood as: Inventory = Procurement Inbound - Production Consumption. However, there are delays and delivery times in the process of procuring materials (for example, the amount of material received at any given time is determined by the order at that time). Secondly, production consumption is closely related to the process, for example, the production of optical module models is related to the production start-up rate, materials, and consumption coefficients.

[0070] Therefore, S1-2 includes:

[0071] S1-2-1. Based on the chip shortage index and the time-varying procurement base coefficient of each material, calculate the corresponding time-varying procurement lead time. The corresponding formula is:

[0072] ;

[0073] in, Indicates material The corresponding time-varying procurement base coefficient.

[0074] S1-2-2. Based on the time-varying procurement lead time and purchase order placement rate of each material, calculate the actual material inbound rate in the current time step.

[0075] S1-2-3. Based on the material consumption coefficient and production start-up rate corresponding to the optical module model of each material, the total consumption rate of all optical module models for each material is calculated by the summation function.

[0076] S1-2-4. Based on the total consumption rate and material inbound rate of each material, the dynamic relationship between the material inventory level of each material and the inflow / outflow rate is constructed through differential operations, generating the material layer dynamic equation.

[0077] The formula corresponding to the dynamic equation of the material layer is:

[0078] ;

[0079] in, Indicates the current time step Production start-up rate This indicates the material requirements for all optical module models. Total consumption rate Indicates material Material receiving rate This represents the differentiation operation.

[0080] S1-3. Based on production data, calculate the input rate and completion rate at the current time step; based on the input rate, completion rate, and material data, determine the dynamic relationship between the production capacity, materials, yield, and production progress of the production segment in the production line, and construct the dynamic equation of the production layer.

[0081] Specifically, the input rate and completion rate The corresponding formula is:

[0082] ;

[0083] ;

[0084] in, Describes the minimum value function. Indicates production section Material consumption coefficient of required materials Indicates production section Processing delay days Indicates production section Total number of intermediate processes Indicates production section The product of the yield rates of all processes in the process.

[0085] For the production layer, the basic physical rule can be simply expressed as: Work-in-process turnover rate = Production input rate - Finished output rate. However, in actual production, capacity is often limited by equipment production limits, including both yield and material constraints. Since the actual yield varies across different processes, the overall yield equals the product of the yields of each process. Furthermore, the production process is significantly affected by human factors; for example, unforeseen events such as material loading / unloading or sudden power outages can cause production delays. Therefore, the formula corresponding to the dynamic equation of the production layer is:

[0086] ;

[0087] ;

[0088] in, Indicates production section The number of processing delay days.

[0089] S1-4. Obtain basic demand forecasts for optical module models, and combine these forecasts with the demand volatility of optical module models to calculate actual demand using a lifecycle function. The corresponding formula is:

[0090] ;

[0091] ;

[0092] in, Indicates the optical module model at the current time step. The lifecycle function, Indicates the optical module model at the current time step. Basic demand forecasting, This represents the standardized random disturbance term. Represents the natural constant. , These represent the optical module models. The obsolescence rate coefficient and time to market. The obsolescence rate coefficient is generally derived from historical data and periodic fitting. In this embodiment, It is a stochastic process with a mean of 0 and a standard deviation of 1, describing all stochastic factors that are not captured by the multi-layer dynamic coupling model, such as the temporary advance or postponement of a project for an end customer; or the order transfer caused by a sudden quality incident of a competitor; or small-scale, short-term market sentiment fluctuations.

[0093] S1-5. Based on the finished product inventory, proportion, actual demand and completion rate of optical module models, determine the dynamic relationship between products and demand, and construct the product layer dynamic equation through differential operations.

[0094] Specifically, for finished products, inventory changes can be understood as: Inventory = Completed Production and Inbound - Actual Outbound. However, issues such as model matching during the production process (e.g., the proportion of different models produced in the production segment), product cycle decay (decreased demand) during the sales process, and demand fluctuations can significantly impact inventory changes. Therefore, the formula corresponding to the product-level dynamic equation is:

[0095] ;

[0096] Indicates production section Production optical module models The allocation coefficient.

[0097] S2. Generate the target inventory of materials and planned capacity allocation for the current time step through the dynamic safety stock algorithm and the flexible capacity allocator.

[0098] Specifically, the traditional safety stock formula It is static, single-layered, and lagging, and for optical module production, it is often affected by many factors such as rapid iteration, urgent demand, and unpredictable chip price fluctuations. Among them, Indicates the volatility of the product. Indicates the safety service coefficient. The standard deviation of demand. Indicates the replenishment cycle.

[0099] To transmit product-level inventory trends to the material level in real time and achieve proactive inventory adjustments, the core innovation lies in introducing the product inventory change rate as a feedback item. Therefore, the processing procedure of the dynamic safety stock algorithm is as follows:

[0100] S2-1-1. Obtain historical demand data and fit it using a sliding window and integration operation to generate the effective demand volatility. The corresponding formula is:

[0101] ;

[0102] in, Indicates material effective demand volatility, This indicates the size of the sliding window (value is 7). This indicates the integration operation. Indicates the material The rate of demand.

[0103] S2-1-2. Obtain historical data and the target inventory of each material in the previous time step, and generate cross-layer compensation gain through a sequential quadratic programming algorithm; historical data is a dataset of material layer, product layer, production layer and some market forecast data recorded in the historical time step.

[0104] Specifically, according to the formula:

[0105] ;

[0106] ;

[0107] ;

[0108] Construct the objective function .in, Indicates the first intermediate variable. Indicates the second intermediate variable. , Indicates the weighting coefficient ( , (sum of 1) Indicates material Cross-layer compensation gain The target inventory of materials below Indicates the optical module model Market demand, This represents the maximum value function.

[0109] Reinitialize a cross-layer compensation gain The guessed value is used to compensate for the cross-layer gain using the Sequential Quadratic Programming (SQP) algorithm. Perform iterative updates, calculate the corresponding objective function value, and search in the direction that makes the objective function value decrease until the cross-layer compensation gain that minimizes the objective function value is found. .

[0110] S2-1-3. Optimize server level coefficients based on stockout cost; stockout cost optimization mainly involves module models. ,materials The relevant factor is a coefficient derived from a comprehensive analysis of factors such as gross profit margin, order volume, product lifecycle, and chip shortage index, which determines the optical module model. ,materials The process is called out-of-stock cost optimization. Initially, a simple cost estimate can be used to obtain a baseline, and then this parameter can be iterated using historical data.

[0111] S2-1-4. Based on the server level coefficient, effective demand volatility, and time-varying procurement lead time of each material, construct a basic safety stock model;

[0112] S2-1-5. Based on the basic safety stock model, cross-layer compensation gain, and product layer dynamic equation, update the target inventory of each material at the current time step.

[0113] The formula for updating the target inventory of each material at the current time step is:

[0114] ;

[0115] in, Indicates the updated material Target inventory of materials Indicates material The corresponding server level coefficient, This represents the basic safety stock model.

[0116] like A value less than 0 indicates a decrease in the inventory of that material, and a corresponding increase is required. Conversely, it decreases. .

[0117] In traditional models, production lines are allocated fixed amounts based on project model and total capacity, but this ignores the impact of product profit differences and inventory response. For production line operations, capacity is limited within a certain timeframe, making it crucial to combine inventory levels with value orientation. Therefore, in this embodiment, a Logit model is chosen to assign value weights, and a flexible capacity allocator is constructed in conjunction with an inventory model (inventory calibration).

[0118] The processing procedure of the flexible capacity allocator is as follows:

[0119] S2-2-1. Calculate the corresponding target inventory of products based on the real-time demand of each optical module model. ;

[0120] In this embodiment, to accurately calculate the target inventory for each optical module model, a machine learning model was introduced to perform in-depth processing and prediction of real-time demand. High-performance algorithms such as XGBoost and LightGBM can be selected for the machine learning model, whose core advantage lies in its ability to integrate and learn complex nonlinear relationships between diverse features. In addition to the basic real-time demand sequence, the model can also incorporate richer external features, such as market seasonal fluctuation cycles, known promotional activity plans, and chip shortage indices reflecting supply chain conditions, thereby constructing a multi-dimensional, high-information-density feature space.

[0121] The training and prediction data for the machine learning model are sourced from multiple channels, including historical sales data from within the company, real-time new and customer orders, and long-term demand forecasts provided by customers. This multi-source data fusion ensures that the model can both grasp long-term patterns and respond nimbly to short-term changes.

[0122] S2-2-2 Obtain the standard processing time for each optical module model Based on market data, the profit margin of each optical module model was calculated. ; This indicates the model number of an optical module being manufactured. Standard processing time for all required processes (including front and back ends) To manufacture an optical module model The sum of the processing times for all required steps.

[0123] S2-2-3, Based on the profit margin, standard processing time, and production stages of each optical module model Maximum production capacity, using the Logit model to determine the production line's value orientation. The corresponding formula is:

[0124] ;

[0125] in, This represents an exponential function with the natural constant as its base. This represents the product strategy weight (in this embodiment, the value is 0.5).

[0126] S2-2-4. Based on product target inventory, production line value orientation, and inventory calibration, construct a flexible capacity allocator to calculate the planned capacity allocation for each production segment within the production line. Planned capacity allocation refers to prioritizing the allocation of products with high profit per unit of labor hour.

[0127] The formula corresponding to the flexible capacity allocator is:

[0128] ;

[0129] in, This represents the inventory adjustment coefficient. Indicates inventory calibration; if If the value is greater than 0, production will be increased; otherwise, production will be decreased.

[0130] S3. Based on the target inventory of materials and the planned capacity allocation, the multi-layer dynamic coupling model is solved by the fourth-order Runge-Kutta method to generate production-inventory decisions and feed them back to the optical module production system.

[0131] Specifically, the existing optical module production system first collects multi-source production data and initializes the model, selecting parameter benchmarks or dynamically updating them depending on whether it is the first run; then it collects new data and adaptively adjusts the calculation step size, using numerical methods (such as the Runge-Kutta method) to iteratively solve the material-production-product problem; after verifying stability, physical rationality and actual deviation, the solution results are transformed into decision instructions such as procurement and production scheduling and sent to the execution system to drive the physical production line to run, and its response data is collected again, forming a dynamic cycle of continuous feedback optimization.

[0132] However, simply following the iterative process of existing optical module production systems will inevitably perpetuate the shortcomings of traditional solutions. First, model parameters often remain static, failing to respond to real-time fluctuations in the supply chain and demand, leading to a disconnect between production plans and actual conditions. Second, the solution processes at each level are isolated, lacking cross-level real-time feedback and collaboration mechanisms, leaving material, production, and inventory management fragmented. Third, the conversion from model state to production instructions relies on fixed rules or human experience, failing to achieve true intelligent optimization in the decision-making process. Although the entire system possesses a closed-loop data acquisition and execution mechanism, it remains essentially a passive and slow system due to the lack of a core controller with intelligent compensation and trend prediction capabilities, unable to achieve forward-looking control. Therefore, this embodiment proposes to use the classic fourth-order Runge-Kutta method (RK4) to solve the three-layer model of the production system, utilizing time-varying parameters (such as time-varying procurement lead time). Production delays With its online learning mechanism, it can automatically adapt to supply chain delays and changes in market demand, dynamically mitigating the bullwhip effect. Through a cross-layer coupling controller, it achieves real-time linkage and global optimization of material consumption, capacity allocation, and inventory status, completely breaking down hierarchical decision-making barriers and enabling the output of forward-looking intelligent decisions that take into account profits, inventory, and delivery rates.

[0133] The classic fourth-order Runge-Kutta method requires historical state values ​​to handle time delay terms, thus necessitating the use of a cache to record historical states. Furthermore, this is particularly relevant for general differential equations. The basic iterative formula for RK4 is:

[0134] ;

[0135] ;

[0136] ;

[0137] ;

[0138] ;

[0139] in, Indicates a time step. This represents the function on the right-hand side (in this example, the dynamic equation of the material layer-production layer-product layer). , , , These represent the estimated slope values ​​between the first and fourth intervals, respectively. This indicates the time step used for numerical integration (simulation from time step). Advance to time step (time interval) , These represent the system at time steps. and time step The state vector.

[0140] Therefore, S3 includes:

[0141] S3-1. Set the initial state vector and the total number of time steps, based on the time step size. Proceed with the advancement; initial state vector The expression is ; This indicates the initial inventory level of the material layer. This indicates the initial work-in-process quantity at the production level. This represents the initial finished goods inventory at the product level. This represents the transpose of the matrix.

[0142] The fourth-order Runge-Kutta method requires historical state values ​​to process time delay terms. Therefore, it is necessary to set up a cache to record historical states, using the state vector in the previous time step as the historical state value of the current time step, and the historical state value of the first time step as the initial state vector.

[0143] S3-2. Based on the multi-layer dynamic coupling model, the first intermediate slope estimate, second intermediate slope estimate, third intermediate slope estimate and fourth intermediate slope estimate of different layers at the current time step are calculated by the classical fourth-order Runge-Kutta method.

[0144] Specifically, the dynamic equations for the material layer, production layer, and product layer are transformed respectively. This is demonstrated using one material, one production segment, and one product as examples, according to the formula:

[0145] ;

[0146] ;

[0147] ;

[0148] Perform the conversion. This represents the material layer conversion function. This represents the production layer conversion function. This represents the product layer conversion function. , These represent the material consumption coefficient and production start-up rate for this optical module model produced on the production line, respectively. This indicates the time-varying procurement lead time for this optical module model produced in this production section. This indicates the derivative of the inventory level of this optical module model produced on the production line. Indicates the rate of demand. This represents the allocation coefficient.

[0149] Assuming that investing at this point is equivalent to starting up, , , The corresponding formulas are as follows:

[0150] ;

[0151] ;

[0152] ;

[0153] The fourth-order Runge-Kutta method was used to calculate the estimated values ​​of the first to fourth intermediate slopes for the material layer, production layer, and product layer, respectively. Indicates the completion rate. This represents the yield rate. The second intermediate slope estimate refers to the value at time step [missing information]. The calculation is performed at point 3, and the third intermediate slope estimate refers to the value at time step 4. The calculation is performed at point 4, and the fourth intermediate slope estimate refers to the value at time step 5. Calculations are performed at that location.

[0154] According to the formula:

[0155] ;

[0156] ;

[0157] ;

[0158] Calculate the first intermediate slope estimate of the material layer. The first intermediate slope estimate of the production layer First intermediate slope estimate of the product layer . , , These represent the inventory level, initial work-in-process quantity, and finished goods inventory at the current time step, respectively.

[0159] Then, according to the formula:

[0160] ;

[0161] Calculate the first intermediate state vector and order , , , The second intermediate slope estimate is calculated using the same method as the first intermediate slope estimate, i.e.:

[0162] ;

[0163] ;

[0164] ;

[0165] The second intermediate slope estimate of the material layer was obtained. The second intermediate slope estimate of the production layer And the second intermediate slope estimate of the product layer . , , These represent the inventory level, initial work-in-process quantity, and finished goods inventory corresponding to the first intermediate state vector, respectively.

[0166] According to the formula:

[0167] ;

[0168] Calculate the third intermediate state vector and order , , , Then, the third intermediate slope estimate is calculated using the same method as the first intermediate slope estimate.

[0169] According to the formula:

[0170] ;

[0171] Calculate the fourth intermediate state vector and order , , , Then, the third intermediate slope estimate is calculated using the same method as the first intermediate slope estimate.

[0172] The calculation of the third / fourth intermediate slope estimate is the same as that of the first / second intermediate slope estimate, so it will not be described in detail.

[0173] Since the time delay term needs to be searched in the historical cache based on the intermediate time node of the current time step, if the intermediate value cannot be found, it can be fitted by linear interpolation or the nearest value can be substituted into the calculation.

[0174] S3-3. Based on the first, second, third, and fourth intermediate slope estimates from different layers, update the initial state vector to obtain the state vector. ;

[0175] The formula for updating the initial state vector is:

[0176] ;

[0177] ;

[0178] ;

[0179] ;

[0180] in, , , These represent the predicted inventory levels, predicted work-in-process inventory, and predicted finished goods inventory, respectively, and correspond to the actual quantities of materials produced, work-in-process, and finished goods. , , , These represent the first, second, third, and fourth intermediate slope estimates of the material layer, respectively. , , , These represent the first, second, third, and fourth intermediate slope estimates for the production layer, respectively. , , , These represent the first, second, third, and fourth intermediate slope estimates for the product layer, respectively.

[0181] S3-4. Based on state vectors, target material inventory, and planned capacity allocation, generate production-inventory decisions.

[0182] Specifically, a judgment is made on the state vector and the target inventory of materials. If the predicted inventory level is... Less than the target inventory of materials This will trigger a purchase order. This generates the corresponding production instructions.

[0183] Then, adjust the input rate according to the planned capacity allocation to obtain the updated input rate. .

[0184] During the update of the input rate, the system first receives planned capacity allocation instructions from the flexible capacity allocator as the initial basis for adjustments. Then, the production execution controller verifies the constraints of these instructions, checking whether material availability, real-time equipment capacity, and personnel process conditions meet the planned requirements and identifying potential conflicts. The real-time dynamic adjuster fine-tunes the instructions based on the production line's immediate status (such as actual OEE and material availability deviation), continuously collecting key production line status data (such as actual equipment OEE, material availability deviation, and work-in-process queue) and quantifying these statuses into instantaneous constraint coefficients. This dynamically corrects the received planned capacity instructions to match the production line's current instantaneous throughput capacity, avoiding overload or idling.

[0185] Next, the exception handling and instruction shaping module addresses sudden anomalies (such as equipment failures and emergency order insertions). It quickly identifies and categorizes these anomalies, then triggers a pre-defined handling rule base (e.g., initiating task rerouting in case of equipment failure, and performing rapid simulation and slot optimization in case of emergency order insertion) to generate response strategies. Based on this, the module "compiles" the logical-level production instructions into structured parameters executable by downstream systems. This includes generating specific work order numbers, process sequences, material delivery lists, and process documents with time and resource bindings, ultimately outputting conflict-free executable work order data. This executable work order data is aggregated and mapped to obtain updated input rates, which are then sent to the Manufacturing Execution System (MES) and Warehouse Management System (WMS) to drive the physical production line to perform specific operations.

[0186] Finally, by collecting execution feedback in real time (such as actual OEE and kitting rate) and immediately feeding this data back to the verification and adjustment process, a rapid adaptive closed loop for the next cycle of instructions is formed, thereby ensuring that the intelligent plan can continuously adapt to the dynamic changes in the real production environment.

[0187] Based on the updated input rate, the completion rate is updated using the same method as in S1-3. Update finished goods inventory. Integrate production orders, updated input rates, and updated completion rates, feeding them back to the optical module production system so that the optical module production system can produce according to purchase orders. Production is carried out in accordance with the requirements.

[0188] In addition, steady-state and convergence assessments of the optical module production system are required, therefore S3 also includes S3-5:

[0189] If the state vector at the current time step does not change with time, that is, its derivative approaches zero, then:

[0190] ;

[0191] Then the physical rationality needs to be checked. For example, no negative inventory:

[0192] ;

[0193] If negative inventory exists, the operator should be prompted to restart the equipment, and the background alarm log should be triggered to check the actual inventory status of the materials.

[0194] Then steady-state calculations are needed, comparing the state vector with the actual system data and calculating the deviation (difference) between the two. If the deviation between the two... If the value is greater than 10%, return to S3-2 and trigger the parameter recalculation. Represents absolute value. Representing the limit symbol, , , These represent material deviation, production deviation, and product deviation, respectively.

[0195] like Figure 2 As shown, a production-stock decision generation system based on a multi-layer dynamic coupling model includes:

[0196] The data acquisition module is used to collect material data, production data, and product data of the optical module.

[0197] The calculation module is used to generate the target inventory of materials and the planned capacity allocation for the current time step through a dynamic safety stock algorithm and a flexible capacity allocator.

[0198] The model building module is used to construct a multi-layered dynamic coupling model of the optical module production system based on material data, production data, and product data.

[0199] The solver module is used to solve the multi-layer dynamic coupling model based on the target inventory of materials and the planned capacity allocation, using the fourth-order Runge-Kutta method, to generate production-inventory decisions and feed them back to the optical module production system.

[0200] It also includes a stability and convergence judgment module, which is used to judge the steady state and convergence of the optical module production system.

[0201] In summary, this invention fundamentally changes the traditional static, hierarchical, and passive production control model by constructing a three-layer dynamic coupled time-delay differential equation model encompassing the material layer, production layer, and product layer. A cross-layer coupled controller, composed of a dynamic safety stock algorithm and a flexible capacity allocator, enables real-time collaborative optimization and forward-looking decision-making for materials, capacity, and inventory. High-precision numerical solutions using the fourth-order Runge-Kutta method allow for digital extrapolation and risk prediction of the system's future state, proactively triggering precise procurement and production scheduling instructions. This effectively suppresses the "bullwhip effect," improves inventory turnover and capacity utilization, and significantly enhances the overall economic efficiency and operational resilience of the production system in the face of supply chain fluctuations and market changes.

[0202] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0203] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0204] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention 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 the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A production-stock decision generation method based on a multi-layer dynamic coupling model, characterized in that, include: Material data, production data, and product data of optical modules are collected to construct a multi-layer dynamic coupling model of the optical module production system. The multi-layer dynamic coupling model includes dynamic equations for the material layer, production layer, and product layer. The production data includes the maximum capacity of each production segment, standard production cycle, changeover preparation time, yield of each process segment, overall equipment efficiency, and work-in-process quantity. The product data includes finished product inventory, demand volatility, and the proportion of each optical module model produced. The dynamic safety stock algorithm and flexible capacity allocator generate the target inventory and planned capacity allocation for materials at the current time step. Based on the target inventory of materials and the planned capacity allocation, the multi-layer dynamic coupling model is solved by the fourth-order Runge-Kutta method to generate production-inventory decisions and feed them back to the optical module production system; The multi-layer dynamic coupling model for constructing the optical module production system includes: Collect material data, production data, product data, and market data for optical modules, and generate a chip shortage index based on the market data; Based on the chip shortage index and the time-varying procurement base coefficient, the time-varying procurement lead time is calculated, and combined with material data, a material-level dynamic equation is constructed. Based on production data, calculate the input rate and completion rate at the current time step; Based on input rate, completion rate and material data, the dynamic relationship between capacity, materials, yield and production progress in the production section of the production line is determined, and the dynamic equation of the production layer is constructed. Obtain basic demand forecasts for optical module models, and combine these forecasts with the demand volatility of optical module models to calculate actual demand using a lifecycle function; Based on the finished product inventory, proportion, actual demand and completion rate of optical module models, the dynamic relationship between products and demand is determined, and the product-level dynamic equation is constructed through differential operations. The processing procedure of the dynamic safety stock algorithm is as follows: Historical demand data is acquired and fitted using a sliding window and integration operations to generate effective demand volatility. Obtain historical data and the target inventory of each material in the previous time step, and generate cross-level compensation gain through a sequential quadratic programming algorithm; Optimize server performance by controlling stockout costs; A basic safety stock model is constructed based on the server level coefficient, effective demand volatility, and time-varying procurement lead time of each material. Based on the basic safety stock model, cross-layer compensation gain, and product-layer dynamic equations, update the target inventory of each material at the current time step. The processing procedure of the flexible capacity allocator is as follows: Calculate the corresponding target inventory of products based on the real-time demand of each optical module model; Obtain the standard processing time for each optical module model, and calculate the profit margin for each optical module model based on market data; Based on the profit margin, standard processing time, and maximum capacity of each production segment for each optical module model, the value orientation of the production line is determined using the Logit model. Based on product target inventory, production line value orientation, and inventory calibration, a flexible capacity allocator is constructed to calculate the planned capacity allocation for production segments within the production line.

2. The production-stock decision generation method based on a multi-layer dynamic coupling model according to claim 1, characterized in that, The material data includes the purchase order placement rate, inventory level, material consumption coefficient and production start-up rate for each material, as well as the corresponding optical module model.

3. The production-stock decision generation method based on a multi-layer dynamic coupling model according to claim 2, characterized in that, The construction of the material layer dynamic equations includes: Based on the chip shortage index and the time-varying procurement base coefficient of each material, the corresponding time-varying procurement lead time is calculated. Based on the time-varying procurement lead time and purchase order placement rate of each material, calculate the actual material inbound rate in the current time step; Based on the material consumption coefficient and production start-up rate corresponding to each optical module model, the total consumption rate of all optical module models for each material is calculated by a summation function. Based on the total consumption rate and material inbound rate of each material, the dynamic relationship between the material inventory level of each material and the inflow / outflow rate is constructed through differential operations, generating the material layer dynamic equation.

4. The production-stock decision generation method based on a multi-layer dynamic coupling model according to claim 2, characterized in that, The generation of production-stock decisions includes: Set the initial state vector and the total number of time steps; the initial state vector includes the initial inventory level of the material layer, the initial work-in-process quantity of the production layer, and the initial finished goods inventory of the product layer. Based on the multi-layer dynamic coupling model, the first, second, third, and fourth intermediate slope estimates of different layers at the current time step are calculated using the classical fourth-order Runge-Kutta method. Based on the first, second, third, and fourth intermediate slope estimates of different layers, the initial state vector is updated to obtain the state vector. Based on state vectors, target material inventory, and planned capacity allocation, production-inventory decisions are generated.

5. A production-stock decision generation system based on a multi-layer dynamic coupling model, used to implement the production-stock decision generation method based on a multi-layer dynamic coupling model as described in any one of claims 1 to 4, characterized in that, include: The data acquisition module is used to collect material data, production data, and product data of the optical module. The calculation module is used to generate the target inventory of materials and the planned capacity allocation for the current time step through a dynamic safety stock algorithm and a flexible capacity allocator. The model building module is used to construct a multi-layered dynamic coupling model of the optical module production system based on material data, production data, and product data. The solver module is used to solve the multi-layer dynamic coupling model based on the target inventory of materials and the planned capacity allocation, using the fourth-order Runge-Kutta method, to generate production-inventory decisions and feed them back to the optical module production system.

6. The production-stock decision generation system based on a multi-layer dynamic coupling model according to claim 5, characterized in that, It also includes a stability and convergence judgment module, which is used to judge the steady state and convergence of the optical module production system.

Images