System and method for enhancing resilience of digital twin supply chain based on time-series production graph

Abstract

The application discloses a digital twin supply chain resilience improvement system and method based on a time sequence production graph, constructs a time sequence production graph with edges determined by a material consumption coefficient matrix A based on real-time transaction data, taking enterprises and products as nodes and transactions as edges, realizes real-time mapping with a physical supply chain, simulates multiple candidate propagation paths by modifying element values in A when an interruption is detected, outputs affected nodes and expected recovery time, generates a recovery scheme list containing an inventory pre-allocation vector to minimize the weighted sum of total cost and maximum backlog, and dynamically adjusts A by taking the deviation of actual recovery time from the predicted value as a feedback signal. The application enables the digital twin to have simulation deduction capability based on a causal model by introducing a production function causal correlation, flexibly simulates different interruption scenarios and recovery strategies through a parameterization mechanism, and enables the virtual model to learn from actual results and continuously optimize through a closed-loop feedback.

Description

Technical Field

[0001] This invention belongs to the field of digital twin technology for supply chains in high-end manufacturing, specifically involving a digital twin supply chain resilience enhancement system and method based on time-series production diagrams. Background Technology

[0002] Supply chain disruptions can cause enormous losses. Take automobile manufacturing as an example: a single vehicle involves thousands of suppliers, and fluctuations in the production capacity of any one supplier can trigger a chain reaction. Digital twin technology, by constructing virtual mappings, makes real-time monitoring and simulation possible. However, existing methods have three significant shortcomings:

[0003] First, there is a lack of structured modeling of causal relationships. Existing methods mostly use deep learning models (such as LSTM and random forest) for statistical prediction, which rely on data correlation rather than causal logic. They cannot accurately deduce the interruption propagation path based on the hierarchical structure of the bill of materials (BOM), resulting in large prediction biases.

[0004] Second, the disruption propagation simulation and recovery optimization are separated. Most systems separate impact assessment and recovery decision-making into two independent stages. Recovery plans are often based on simplified assumptions (such as fixed procurement quantities) and fail to make full use of real-time information on affected nodes and constraints of production causality, making it difficult to achieve optimal recovery actions in terms of cost and timeliness.

[0005] Third, digital twin models lack adaptive closed-loop optimization capabilities. Most systems use static or periodically updated model parameters, and the models are trained offline, lacking the ability to be updated online adaptively. When the physical supply chain changes, the virtual model cannot adjust in time, and the simulation results gradually deviate from the actual situation, reducing the credibility of decision-making. Summary of the Invention

[0006] To address the shortcomings of existing technologies and achieve accurate prediction of supply chain disruptions, intelligent optimization of supply chain recovery, and the formation of a self-adaptive closed loop of virtual and physical supply chain feedback, this invention adopts the following technical solution:

[0007] The digital twin supply chain resilience enhancement system based on time-series production maps includes a data acquisition module, a time-series production map construction module, an interruption propagation simulation module, a recovery optimization module, and a closed-loop feedback module.

[0008] The data acquisition module acquires real-time transaction data, inventory status data, and equipment operating status data from the physical supply chain.

[0009] The time-series production graph construction module acquires the real-time transaction data and constructs a dynamic time-series production graph that maps to the physical supply chain in real time, using enterprises and products as nodes and transactions as edges. The dynamic changes of the edges are determined by a production function, which is a material consumption coefficient matrix A for enterprises to transform input products into output products. Element a in matrix A... ij The quantity of upstream product i required to produce one unit of downstream product j;

[0010] The interruption propagation simulation module, upon detecting an interruption event, simulates multiple candidate interruption propagation paths based on the current time-series production diagram by modifying at least one element value in the material consumption coefficient matrix A, in order to generate the set of affected nodes and the estimated recovery time for each path;

[0011] The recovery optimization module obtains the set of affected nodes and the estimated recovery time. Based on the real-time inventory status data and the production function constraints, it simulates multiple candidate recovery action plans by adjusting the inventory pre-allocation vector and the procurement coefficient to generate the expected recovery result of each plan, including the expected cost, recovery time and order completion rate. Based on the expected results, it generates a sorted list of recovery action plans.

[0012] The closed-loop feedback module obtains the actual recovery result of the supply chain disruption, compares it with the expected recovery result of the simulation, and updates the material consumption coefficient matrix A based on the deviation of the comparison.

[0013] Furthermore, for each enterprise node i, maintain its inventory vector at time t. ; at each time step Calculate the purchase volume vector based on real-time transaction data. At the same time, estimate the enterprise's consumption vector. Consumption vector The material consumption coefficient matrix A and the output of enterprise node i in the transaction data at time t are given. The product of these factors is used, with each element in the material consumption coefficient matrix A serving as a learnable attention weight. Based on the attention mechanism, learning is constrained by constructing inventory update rules.

[0014]

[0015] in, This represents the inventory pre-allocation vector in the expected recovery result generated by the recovery optimization module, used to reserve materials needed for recovery operations or release redundant inventory. This represents the interrupt consumption priority vector generated by the interrupt propagation deduction module. This represents the physical state deviation feedback vector acquired in real time by the data acquisition module, reflecting the actual inventory fluctuations in the physical supply chain. This represents the state weight vector, which is automatically set by the virtual model validation module based on data confidence level and is used to adjust the credibility of physical feedback. This indicates element-wise multiplication.

[0016] The aforementioned inventory update rules can reflect normal business operations while also adapting to interruption scenarios and physical feedback. This allows virtual inventory to respond in real time to physical fluctuations and interruption strategies, enabling immediate execution of recovery plans, avoiding material conflicts, and significantly improving overall response efficiency.

[0017] Furthermore, the inventory pre-allocation vector It is the vector sum of the procurement increment vector, the production adjustment vector, and the inventory release vector, wherein the procurement increment vector corresponds to the quantity of materials procured from alternative suppliers, the production adjustment vector corresponds to the quantity of inventory reserved or released due to production plan adjustments, and the inventory release vector corresponds to the quantity of redundant inventory released.

[0018] Furthermore, the interrupt consumption priority vector In the event of a supply chain disruption, this is dynamically generated by the disruption propagation and deduction module based on the disruption type and the magnitude of the material consumption coefficient in the production function.

[0019] Furthermore, the interruption propagation deduction module, upon detecting an interruption event, recursively calculates the output changes of downstream product nodes based on the production function in the time-series production graph, starting from the affected node. Here, let the node... The change in output is Then downstream product nodes Input change satisfy:

[0020]

[0021] in, The element in the production function matrix A represents the node that produces one unit of downstream product. Required upstream product nodes Quantity, Indicates downstream product nodes Provides a set of all upstream product nodes for the material; the interruption propagation simulation module outputs a list of affected enterprises and products and the estimated recovery time.

[0022] This invention introduces the production function matrix A into the interruption propagation simulation module. By modifying the element values ​​in A, different interruption types (capacity loss, partial production reduction, logistics delay, etc.) are simulated, thereby simulating multiple candidate interruption propagation paths instead of a single prediction. This allows the simulation of this invention to be based on causal logic rather than data correlation, and the results are interpretable. It can accurately output the set of affected nodes and the expected recovery time, providing accurate input for recovery optimization.

[0023] Furthermore, the recovery optimization module constructs an optimization model with the objective of minimizing the weighted sum of the total cost and the maximum backlog of uncompleted orders. The objective function of this model is:

[0024]

[0025] in, This represents the unit procurement cost of alternative supplier s. This represents the quantity purchased from alternative supplier s on day d. This represents the number of orders that have not yet been completed by the end of day d. Indicates the weighting coefficient;

[0026] The quantity of goods purchased The constraints include production function constraints that ensure the procured materials meet production requirements:

[0027]

[0028] in, This represents the quantity of materials from supplier s required to produce one unit of product i. This represents the demand for product i on day d. This invention uses a production function as a constraint to construct a multi-objective optimization model (total cost + maximum backlog) to ensure that the purchased materials meet production needs.

[0029] Furthermore, the quantity of goods purchased The constraints include a recursive constraint on the number of unfulfilled orders:

[0030]

[0031] in, This represents the total purchase volume of all suppliers on day d. This represents the total demand for all products on day d, with a backlog recursion constraint. This means that the number of unfulfilled orders at the end of the day equals the number of unfulfilled orders from the previous day plus the current day's purchases minus the current day's demand. ,and Not negative.

[0032] This invention introduces backlog recursion constraints to... By linking purchase volume and demand volume, recovery time is transformed into a calculable backlog of orders. This backlog recursive constraint model is a linear optimization problem that can be solved using solvers such as Gurobi and CPLEX.

[0033] The recovery optimization module selects the optimal recovery action plan from the solution set according to preset weights and generates the corresponding inventory pre-allocation vector. The data is sent to the time-series production graph construction module to update the inventory. This is achieved by minimizing... The system can effectively control backlog peaks, thereby accelerating recovery and reducing order backlog.

[0034] Furthermore, the quantity of goods purchased The constraints include:

[0035] Supplier capability constraints:

[0036]

[0037] in, This represents the maximum supply quantity of supplier s on day d, to constrain the purchase quantity from not exceeding the supplier's maximum supply capacity;

[0038] Order demand constraints:

[0039]

[0040] in, This indicates the minimum purchase quantity agreed upon in the contract with supplier s, in order to meet the minimum purchase quantity stipulated in the contract.

[0041] Furthermore, the closed-loop feedback module will measure the actual recovery time. Recovery time predicted by simulation The difference is used to construct the loss. The gradient of the loss function with respect to A is calculated using the chain rule. , and according to Update the element values ​​in the material consumption coefficient matrix A, where, The learning rate, which can be obtained through automatic differentiation or numerical approximation, feeds back the updated material consumption coefficient matrix A to the interruption propagation deduction module, forming a self-optimizing closed loop. In this way, the system has self-learning capabilities and can learn from actual recovery results. As actual recovery data accumulates, the recovery time prediction error gradually decreases, effectively overcoming the problem of static models deviating from physical reality, and enabling the virtual model to continuously approach physical reality, thus achieving self-optimization.

[0042] The method for enhancing the resilience of a digital twin supply chain based on time-series production maps, using the aforementioned digital twin supply chain resilience enhancement system based on time-series production maps, performs the following steps:

[0043] Step 1: Collect transaction data, inventory status, and equipment status of the physical supply chain in real time and update them synchronously to the digital twin system;

[0044] Step 2: Based on real-time collected data, a time-series production graph is dynamically constructed and maintained, with enterprises and products as nodes and transactions as edges. In the time-series production graph, the dynamic changes of the edges are determined by the production function defined by the material consumption coefficient matrix A.

[0045] Step 3: Monitor interruption events in real time. When an interruption is detected, based on the current time-series production diagram, simulate multiple candidate interruption propagation paths by modifying at least one element value in the material consumption coefficient matrix A, and generate the set of affected nodes and the estimated recovery time for each path.

[0046] Step 4: Obtain the set of affected nodes and the estimated recovery time. Based on the real-time inventory and the production function constraints, simulate multiple candidate recovery action plans by adjusting the inventory pre-allocation vector and the procurement coefficient, generate the expected recovery result for each plan, including the expected cost, recovery time and order completion rate, and generate a sorted list of recovery action plans based on the expected results.

[0047] Step 5: Continuously collect actual recovery effect data, compare it with the simulation results in real time, update the material consumption coefficient matrix A based on the comparison results, realize the self-optimization of the digital twin model, and return to step 1.

[0048] The advantages and beneficial effects of this invention are as follows:

[0049] This invention improves the accuracy of interruption propagation prediction: by introducing a production function as a causal constraint, interruption propagation prediction no longer relies on statistical correlation, but is based on layer-by-layer calculation of material consumption coefficients, which can more accurately predict the scope and degree of impact of interruptions.

[0050] This invention optimizes the effectiveness of recovery decisions: the recovery optimization module uses the production function as a constraint to accurately calculate material requirements in a multi-objective optimization model, and the generated recovery plan achieves optimal cost and recovery time, and is linked with the inventory status in real time to avoid decision conflicts;

[0051] This invention achieves adaptive optimization of digital twins: the closed-loop feedback module compares the actual recovery effect with the simulation results and dynamically adjusts the production function parameters, so that the virtual model continuously approximates physical reality and has self-learning capabilities.

[0052] This invention enhances the overall resilience of the supply chain: through real-time sensing, accurate prediction, intelligent optimization, and adaptive closed-loop, this invention can significantly shorten the interruption recovery time, reduce order losses caused by interruptions, and improve the supply chain's ability to withstand risks in complex and ever-changing environments. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the system structure in an embodiment of the present invention.

[0054] Figure 2 This is a flowchart of the method in an embodiment of the present invention. Detailed Implementation

[0055] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0056] To achieve accurate disruption prediction and intelligent recovery optimization, and to form an adaptive closed loop of virtual-real feedback, this invention proposes a digital twin supply chain resilience enhancement system based on time-series production diagrams. This system integrates the implicit causal production functions between enterprises into the digital twin, using the time-series production diagram as the core causal model. Combined with parametric simulation and closed-loop feedback, it significantly improves the accuracy and automation level of supply chain disruption management. This invention is particularly suitable for manufacturing supply chains with complex bills of materials (BOMs), such as those in the automotive, electronics, and semiconductor industries. It is used for real-time monitoring of disruption events, prediction of disruption propagation paths, optimization of recovery action plans, and model self-optimization through closed-loop feedback, thereby enhancing supply chain resilience.

[0057] like Figure 1 As shown, the system of the present invention includes a data acquisition module, a time-series production graph construction module, an interruption propagation deduction module, a recovery optimization module, and a closed-loop feedback module.

[0058] The data acquisition module is used to acquire real-time transaction data, inventory status data, and equipment operating status data of the physical supply chain;

[0059] The data acquisition module obtains real-time data on order flow, inventory, equipment status, and logistics timeliness of the physical supply chain through IoT, ERP, or SCADA system interfaces, and updates it synchronously to the time-series production diagram construction module.

[0060] The time-series production graph construction module, connected to the data acquisition module, constructs and maintains a dynamic time-series production graph that maps to the physical supply chain in real time, based on received real-time transaction data, using enterprises and products as nodes and transactions as edges. The dynamic changes of the edges in the time-series production graph are determined by the production function, defined as the material consumption coefficient matrix A that transforms input products into output products. The material consumption coefficient matrix A is a parameter matrix pre-trained using historical transaction data, and the element a in A... ij This represents the quantity of product i required to produce one unit of product j;

[0061] The core task of the time-series production graph construction module is to learn the firm's implicit production function and construct a dynamic graph. The production function is defined as a material consumption coefficient matrix. ,in, Indicates the production of one unit of product Required products The quantity can be determined by using a time-series graph neural network combined with a dynamic inventory update model.

[0062] Specifically, for each enterprise node i, maintain its inventory vector at time t. At each time step The module calculates the purchase volume vector based on real-time transaction data. At the same time, the module needs to estimate the enterprise's consumption vector. Since consumption cannot be directly observed, the module learns through an attention mechanism. Let the output of enterprise i at time t be... (Obtained from the output transactions in the transaction data), then the baseline consumption amount Should meet However, A is unknown. Therefore, the module treats each element in A as a learnable attention weight and constrains the learning by the following inventory update rule:

[0063]

[0064] in, Let represent the inventory vector of company i at time t; This represents the purchase quantity vector of company i at time t; This represents the inventory pre-allocation vector generated by the recovery optimization module for enterprise i at time t. The recovery optimization module generates the materials reserved in the recovery plan and writes them into the virtual inventory in advance to reserve the materials needed for the recovery action or release redundant inventory. This represents the baseline consumption vector of firm i at time t, determined by the production function and output. This represents the interruption consumption priority vector of product p under the interruption scenario at time t for enterprise i. It is dynamically adjusted by the interruption propagation deduction module based on the interruption type and production function to prioritize the consumption of critical products. Its value range is... ; This represents the physical state deviation feedback vector of enterprise i at time t, which is obtained in real time by the data acquisition module. This represents the state weight vector of enterprise i at time t, which is automatically set by the virtual model verification module based on data confidence and is used to adjust the credibility of physical feedback. This represents element-wise multiplication. These terms work together to ensure that inventory updates reflect both normal business operations and can adapt to disruptions and physical feedback. The time-series production chart construction module enables virtual inventory to respond in real-time to physical fluctuations and disruption strategies. Recovery plans can be executed immediately, avoiding material conflicts and significantly improving overall response efficiency.

[0065] To learn A, the module defines a loss function that penalizes inventory overdrafts and encourages reasonable consumption. After training, A becomes the learned production function. The time-series production graph construction module ultimately outputs a dynamic graph structure where nodes include all enterprises and products, edges represent transactions, and each edge is associated with a coefficient in the production function. This graph evolves over time, reflecting the real-time state of the physical supply chain.

[0066] Furthermore, the inventory pre-allocation vector It is the vector sum of the procurement increment vector, production adjustment vector, and inventory release vector, where the procurement increment vector corresponds to the quantity of materials purchased from alternative suppliers, the production adjustment vector corresponds to the quantity of inventory reserved or released due to production plan adjustments, and the inventory release vector corresponds to the quantity of redundant inventory released.

[0067] Interrupt Consumption Priority Vector It is dynamically generated by the interruption propagation deduction module based on the interruption type and the magnitude of the material consumption coefficient in the production function.

[0068] Physical state deviation feedback vector The data is calculated from sensor data from IoT devices and ERP systems accessed in real time by the data acquisition module, so as to correct the virtual inventory in real time, achieve virtual-real synchronization, and reflect the actual inventory fluctuations of the physical supply chain.

[0069] The interruption propagation simulation module is connected to the time-series production graph construction module. It is used to monitor interruption events in real time. When an interruption event is detected, it parameterizes the time-series production graph based on the current time-series production graph by modifying at least one element value in the material consumption coefficient matrix A (for example, setting the capacity coefficient of the affected node to 0), simulates multiple candidate interruption propagation paths, generates the set of affected nodes and the expected recovery time for each path, and sends the simulation results to the recovery optimization module.

[0070] When an interruption event is detected, the interruption propagation deduction module recursively calculates the output changes of downstream nodes based on the production function in the time-series production graph, starting from the affected node. Here, we assume the product node... The change in output is Then downstream single product node Input change satisfy:

[0071]

[0072] in, The element in the production function matrix A represents the production of one unit of product. Required products The quantity, upstream Indicates to product node Provides a set of all upstream product nodes for the material; the interruption propagation simulation module outputs a list of affected companies and products and the estimated recovery time.

[0073] This invention introduces the production function matrix A into the interruption propagation simulation module of a digital twin system for the first time. By modifying the element values ​​in A, it simulates different interruption types (capacity loss, partial production reduction, logistics delay, etc.). This invention can simulate multiple candidate interruption propagation paths instead of a single prediction, so that the inference of this invention can be based on causal logic rather than data correlation. The results are interpretable and can accurately output the set of affected nodes and the expected recovery time, providing accurate input for recovery optimization.

[0074] The recovery optimization module, connected to the interruption propagation simulation module and the time-series production graph construction module, receives the set of affected nodes and the estimated recovery time. Based on real-time inventory data and production function constraints, it simulates multiple candidate recovery action plans by adjusting the inventory pre-allocation vector and purchasing coefficients. It generates the expected cost, recovery time, and order completion rate for each plan and generates a ranked list of recovery action plans based on the expected results. This list is then sent to the time-series production graph construction module to update the inventory pre-allocation vector in the time-series production graph in real time (writing the materials reserved for the recovery plan into the virtual model). Finally, it executes the physical recovery plan (such as adding materials to the alternatives). (The supplier places a purchase order); the recovery plan and prediction results are recorded for closed-loop comparison (predicted values ​​are retained). The recovery plan is split into three parallel actions, decoupling the decision-making, execution, and learning threads so that they do not block each other, ensuring the data integrity required for closed-loop feedback and improving the real-time performance and reliability of the system. This invention directly uses the "affected nodes and expected recovery time" output by the interruption propagation deduction as the input of the recovery optimization module, thereby linking the recovery decision with the real-time interruption, automatically balancing cost and recovery speed, effectively shortening the average recovery time and reducing recovery costs. The output inventory pre-allocation vector is used to update the virtual inventory in real time.

[0075] The recovery optimization module aims to minimize the weighted sum of total cost and the maximum backlog of unfulfilled orders. It constructs a multi-objective optimization model to solve for the optimal recovery action plan, resulting in a list of recovery plans (including a pre-allocated inventory vector). Taking supplier disruption as an example, recovery actions include emergency procurement from alternative suppliers. The supplier selection set for equipment is S, and the decision variables are... To determine the purchase quantity from supplier s on day d, the objective function is to minimize the weighted sum of total purchase cost and maximum backlog.

[0076]

[0077] in, This represents the unit procurement cost of alternative supplier s. This represents the quantity purchased from alternative supplier s on day d. This represents the number of unfulfilled orders (i.e., backlog) at the end of day d. Indicates the weighting coefficient. Represent decision variables;

[0078] The constraints include:

[0079] (1) Production function constraints:

[0080]

[0081] in, This represents the quantity of materials from supplier s required to produce one unit of product i. This represents the demand for product i on day d. This invention uses a production function as a constraint to construct a multi-objective optimization model (total cost + maximum backlog) to ensure that the purchased materials meet production needs.

[0082] (2) Supplier capability constraints:

[0083]

[0084] in, This represents the maximum supply quantity of supplier s on day d, to constrain the purchase quantity from not exceeding the supplier's maximum supply capacity;

[0085] (3) Order demand constraints:

[0086]

[0087] in, This indicates the minimum purchase quantity agreed upon in the contract with supplier s, in order to meet the minimum purchase quantity stipulated in the contract;

[0088] (4) Backlog recursion constraint:

[0089]

[0090] in, This represents the total purchase volume of all suppliers on day d. This represents the total demand for all products on day d, with a backlog recursion constraint. This means that the number of unfulfilled orders at the end of the day equals the number of unfulfilled orders from the previous day plus the current day's purchases minus the current day's demand. ,and Not negative.

[0091] This invention introduces backlog recursion constraints to... By linking purchase volume and demand volume, recovery time is transformed into a calculable backlog of orders. This backlog recursive constraint model is a linear optimization problem that can be solved using solvers such as Gurobi and CPLEX.

[0092] The recovery optimization module selects the optimal recovery action plan from the solution set according to preset weights and generates the corresponding inventory pre-allocation vector. The data is sent to the time-series production graph construction module to update the inventory. This is achieved by minimizing... The system can effectively control backlog peaks, thereby accelerating recovery and reducing order backlog.

[0093] The closed-loop feedback module, connected to the time-series production diagram construction module, is used to continuously collect performance data such as order completion rate and real-time recovery time during the actual recovery process, compare them with the simulation results in real time, calculate the deviation, and update the material consumption coefficient matrix A based on the deviation to achieve self-optimization of the digital twin model.

[0094] The closed-loop feedback module uses the difference between the actual recovery time and the simulation-predicted recovery time as a feedback signal. Based on this feedback signal, the element values ​​in the material consumption coefficient matrix A are dynamically adjusted using the gradient descent method, and the updated A is fed back to the interruption propagation deduction module to form a self-optimizing closed loop.

[0095] The closed-loop feedback module continuously collects performance data during the actual recovery process, such as order completion rate and actual recovery time. After the recovery is completed, the actual recovery time will be compared with the recovery time predicted by the simulation. The comparison is performed, and the deviation is calculated. In this embodiment, the closed-loop feedback module uses gradient descent to update the production function matrix A. Let the loss function be... The gradient of the loss function with respect to A is calculated using the chain rule. , and according to Update, in which, The learning rate is the gradient, which can be obtained through automatic differentiation or numerical approximation. This approach endows the system with self-learning capabilities, enabling it to learn from actual recovery results. As actual recovery data accumulates, the recovery time prediction error gradually decreases, effectively overcoming the problem of static models deviating from physical reality. This allows the virtual model to continuously approximate physical reality, achieving self-optimization.

[0096] Based on the above system, the present invention also provides a method for improving the resilience of a digital twin supply chain based on a time-series production graph, such as... Figure 2 As shown, the following steps are executed in a real-time loop:

[0097] Step 1: Collect transaction data, inventory status, and equipment status of the physical supply chain in real time and update them synchronously to the digital twin system;

[0098] Step 2: Based on real-time data, dynamically construct and maintain a time-series production graph with enterprises and products as nodes and transactions as edges. The dynamics of the edges in the time-series production graph are determined by the production function defined by the material consumption coefficient matrix A.

[0099] Step 3: Monitor interruption events in real time. When an interruption is detected, based on the current production timeline, simulate multiple candidate interruption propagation paths by modifying at least one element value in the material consumption coefficient matrix A, and generate the set of affected nodes and the estimated recovery time for each path.

[0100] Step 4: Based on real-time inventory and the production function constraints, simulate multiple candidate recovery action plans by adjusting the inventory pre-allocation vector and purchasing coefficient, generate the expected cost, recovery time and order completion rate of each plan, and generate a sorted list of recovery action plans based on the expected results;

[0101] Step 5: Continuously collect actual recovery effect data, compare it with the simulation results in real time, and update the material consumption coefficient matrix A based on the comparison results to achieve self-optimization of the digital twin model.

[0102] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A digital twin supply chain resilience enhancement system based on time-series production maps, comprising a data acquisition module, a time-series production map construction module, an interruption propagation simulation module, a recovery optimization module, and a closed-loop feedback module, characterized in that: The data acquisition module acquires real-time transaction data and inventory status data from the physical supply chain. The time-series production graph construction module acquires the real-time transaction data and constructs a dynamic time-series production graph that maps to the physical supply chain, with enterprises and products as nodes and transactions as edges. The dynamic changes of the edges are determined by the production function, which is a material consumption coefficient matrix for enterprises to transform input products into output products. The elements in the matrix are the quantities of upstream products required to produce one unit of downstream product. The interruption propagation simulation module, upon detecting an interruption event, simulates candidate interruption propagation paths by modifying at least one element value in the material consumption coefficient matrix based on the current time-series production diagram, in order to generate the set of affected nodes and the expected recovery time corresponding to the path. The recovery optimization module obtains the set of affected nodes and the estimated recovery time. Based on the inventory status data and the production function constraints, it simulates candidate recovery action plans by adjusting the coefficients of inventory pre-allocation and procurement to generate the expected recovery result of the plan. The closed-loop feedback module obtains the actual recovery result of the supply chain disruption, compares it with the expected recovery result of the simulation, and updates the material consumption coefficient matrix based on the deviation of the comparison.

2. The digital twin supply chain resilience enhancement system based on time-series production diagrams according to claim 1, characterized in that: For each enterprise node i, maintain its inventory vector at time t. ; at each time step Calculate the purchase volume vector based on real-time transaction data. At the same time, estimate the enterprise's consumption vector. Consumption vector The material consumption coefficient matrix and the output of enterprise node i at time t in the transaction data. The product of these factors, along with the elements in the material consumption coefficient matrix, are used as learnable attention weights. Based on the attention mechanism, learning is constrained by constructing inventory update rules. in, This represents the inventory pre-allocation vector in the expected recovery result generated by the recovery optimization module. This represents the interrupt consumption priority vector generated by the interrupt propagation deduction module. This represents the physical state deviation feedback vector obtained by the data acquisition module, reflecting the actual inventory fluctuations in the physical supply chain. This represents the state weight vector, used to adjust the reliability of physical feedback; This indicates element-wise multiplication.

3. The digital twin supply chain resilience enhancement system based on time-series production diagrams according to claim 2, characterized in that: The inventory pre-allocation vector It is the vector sum of the procurement increment vector, the production adjustment vector, and the inventory release vector, wherein the procurement increment vector corresponds to the quantity of materials procured from alternative suppliers, the production adjustment vector corresponds to the quantity of inventory reserved or released due to production plan adjustments, and the inventory release vector corresponds to the quantity of redundant inventory released.

4. The digital twin supply chain resilience enhancement system based on time-series production diagrams according to claim 2, characterized in that: The interruption consumption priority vector In the event of a supply chain disruption, this is dynamically generated by the disruption propagation and deduction module based on the disruption type and the magnitude of the material consumption coefficient in the production function.

5. The digital twin supply chain resilience enhancement system based on time-series production diagrams according to claim 1, characterized in that: The interruption propagation and deduction module, upon detecting an interruption event, recursively calculates the output changes of downstream product nodes based on the production function in the time-series production graph, starting from the affected node. Here, let the node... The change in output is Then downstream product nodes Input change satisfy: in, The element in the production function matrix A represents the node that produces one unit of downstream product. Required upstream product nodes Quantity, Indicates downstream product nodes Provides a set of all upstream product nodes for the material; the interruption propagation simulation module outputs a list of affected enterprises and products and the estimated recovery time.

6. The digital twin supply chain resilience enhancement system based on time-series production diagrams according to claim 1, characterized in that: The recovery optimization module constructs an optimization model with the objective of minimizing the weighted sum of total cost and the maximum backlog of uncompleted orders. Its objective function is: in, This represents the unit procurement cost of alternative supplier s. This represents the quantity purchased from alternative supplier s on day d. This represents the number of orders that have not yet been completed by the end of day d. Indicates the weighting coefficient; The quantity of goods purchased The constraints include production function constraints that ensure the procured materials meet production requirements: Where i is the product index and d is the day index. This represents the quantity of materials from supplier s required to produce one unit of product i. This represents the demand for product i on day d.

7. The digital twin supply chain resilience enhancement system based on time-series production diagrams according to claim 6, characterized in that: The quantity of goods purchased The constraints include a recursive constraint on the number of unfulfilled orders: in, This represents the total purchase volume of all suppliers on day d. This represents the total demand for all products on day d, with a backlog recursion constraint. This means that the number of unfulfilled orders at the end of the day equals the number of unfulfilled orders from the previous day plus the current day's purchases minus the current day's demand. ,and Not negative.

8. The digital twin supply chain resilience enhancement system based on time-series production diagrams according to claim 6, characterized in that: The quantity of goods purchased The constraints include: Supplier capability constraints: in, This represents the maximum supply from supplier s on day d; Order demand constraints: in, This indicates the minimum purchase quantity agreed upon in the contract with supplier s.

9. The digital twin supply chain resilience enhancement system based on time-series production diagrams according to claim 6, characterized in that: The closed-loop feedback module will record the actual recovery time. Recovery time predicted by simulation The difference is used to construct the loss. The gradient of the loss function with respect to A is calculated using the chain rule. , and according to Update the element values ​​in the material consumption coefficient matrix A, where, The updated material consumption coefficient matrix A is fed back to the interruption propagation deduction module as the learning rate.

10. A method for improving the resilience of a digital twin supply chain based on time-series production diagrams, characterized by: Using the digital twin supply chain resilience enhancement system based on time-series production diagrams as described in any one of claims 1 to 9, the following steps are performed: Step 1: Collect transaction data and inventory status of the physical supply chain; Step 2: Based on the collected data, dynamically construct and maintain a time-series production graph with enterprises and products as nodes and transactions as edges. In the time-series production graph, the dynamic changes of the edges are determined by the production function defined by the material consumption coefficient matrix. Step 3: When an interruption is detected, based on the current time-series production diagram, by modifying at least one element value in the material consumption coefficient matrix, the candidate interruption propagation path is simulated, and the set of affected nodes and the expected recovery time corresponding to the path are generated. Step 4: Obtain the set of affected nodes and the estimated recovery time. Based on the inventory and the production function constraints, simulate candidate recovery action plans by adjusting the inventory pre-allocation and procurement coefficients, and generate the expected recovery results of the plan. Step 5: Continuously collect actual recovery effect data, compare it with the simulation results in real time, update the material consumption coefficient matrix based on the comparison results, and return to Step 1.

Images