Supply chain intelligent optimization decision-making system based on digital twinning

Abstract

The invention discloses a supply chain intelligent optimization decision-making system based on digital twinning, and relates to the technical field of ERP, detailed data and interaction data of each link of a supply chain are collected in a multi-channel manner through a supply chain digital twinning construction module, and a supply chain overall digital twinning model is constructed by using a graph convolutional network and a K-means clustering algorithm; the supply chain risk monitoring module monitors the operation condition of each link in real time, identifies a risk link, evaluates a risk level, and determines an association link and an association closeness degree of the risk link; the decision optimization execution module formulates and executes an optimization decision scheme by applying a support vector machine and other technologies according to the related data of the risk link and the associated link, including production plan adjustment, inventory allocation, logistics route optimization and supplier cooperation strategy change; and the visual presentation module visually displays the supply chain overall digital twinborn model and optimizes the execution condition and effect feedback of the decision scheme, the system realizes refined intelligent management of the supply chain, and the efficiency of the supply chain is improved.

Description

Technical Field

[0001] The present invention relates to the field of ERP technology, and more specifically, to a supply chain intelligent optimization decision-making system based on digital twins. Background Art

[0002] In today's era of globalization and digitalization, supply chain management is crucial to the competitiveness and market adaptability of enterprises. Traditional supply chain management methods often rely on manual experience and simple data analysis tools, which are difficult to cope with the complex dynamic changes and potential risks in modern supply chains. With the rapid development of technologies such as big data, artificial intelligence, and the Internet of Things, supply chain management has ushered in an opportunity for intelligent transformation. Although the existing supply chain management technology has improved the transparency of information and the scientific nature of decision-making to a certain extent, there are still many problems. For example, the means of data collection are limited and cannot fully cover the dynamic changes in all links of the supply chain; the risk identification and early warning capabilities are insufficient, making it difficult to timely discover potential risk points and take effective countermeasures; the decision optimization model relies too much on historical data and lacks in-depth analysis of real-time data and complex correlations; the visualization presentation method is not intuitive enough and cannot clearly show the overall operation status of the supply chain and the effect of optimized decisions. Summary of the invention

[0003] In order to overcome the problems of limited data collection, insufficient risk warning, decision-making reliance on history, and non-intuitive visualization in the prior art, the present invention designs a supply chain intelligent optimization decision-making system based on digital twins, which can effectively solve the above technical problems.

[0004] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0005] The supply chain intelligent optimization decision-making system based on digital twins includes:

[0006] The supply chain digital twin construction module is used to collect detailed data of each link in the supply chain, including enterprise production capacity, inventory status, logistics and transportation parameters, and supplier information, as well as interaction data between links, including order flow, information sharing records, and collaborative work processes; through the graph convolutional network, the supply chain links are divided into multiple link clusters based on link data; using the K-means clustering algorithm, the sub-digital twin model corresponding to each link cluster is constructed based on the interaction data, and then the overall digital twin model of the supply chain is integrated and constructed;

[0007] The supply chain risk monitoring module is used to monitor the operating status of each link in the supply chain in real time, including production progress, inventory changes and logistics tracks; identify links whose operating status deviates from the normal range as risk links, and evaluate the risk level of risk links; determine multiple related links of risk links and the degree of correlation between each related link based on the overall digital twin model of the supply chain and the risk level;

[0008] A decision optimization execution module formulates and executes an optimization decision-making plan for the risk link and its associated link according to the characteristic data and risk level of the risk link, as well as the characteristic data and association tightness of the associated link. The plan covers production plan adjustment, inventory allocation, logistics route optimization, and supplier cooperation strategy change;

[0009] The visualization presentation module is used to display the overall digital twin model of the supply chain, as well as the execution situation and effect feedback of the optimization decision-making plan.

[0010] Preferably, the supply chain digital twin construction module includes:

[0011] A data acquisition unit obtains static data, dynamic data of each link in the supply chain, and interaction data between links through multiple channels such as sensors, Internet of Things devices, and enterprise information system interfaces;

[0012] A link cluster division unit converts the data of each link into feature vectors, uses a graph convolutional network to learn the association weights between feature vectors, and divides similar links into the same link cluster according to the weights;

[0013] A model construction unit constructs interaction feature vectors according to the interaction data, uses the K-means clustering algorithm to perform clustering analysis on the interaction feature vectors within the link cluster, constructs a sub-digital twin model, and finally integrates to form an overall digital twin model of the supply chain.

[0014] Preferably, the supply chain risk monitoring module includes:

[0015] A real-time monitoring unit continuously collects real-time operation data of each link in the supply chain;

[0016] A risk identification unit trains a risk identification model based on historical operation data and a preset risk index system, inputs the real-time monitoring data into the model, identifies the risk link and determines the risk level;

[0017] An analysis unit determines the associated link according to the overall digital twin model of the supply chain, combines the risk level of the risk link, and determines the association tightness according to the preset rules.

[0018] Preferably, the decision optimization execution module includes:

[0019] An association tightness conversion unit converts the association tightness of the associated link into the corresponding risk impact degree according to specific conversion rules;

[0020] A decision generation unit uses a support vector machine, takes the characteristic data, risk level, and risk impact degree of the risk link and the associated link as inputs, trains and generates a decision model, and outputs an optimization decision-making plan for each link;

[0021] The execution control unit is responsible for implementing the optimized decision-making plan output by the decision-making generation unit, regulating the supply chain risk links and their associated links, and tracking the implementation effect.

[0022] Preferably, the process of the link cluster division unit constructing feature vectors and dividing link clusters is as follows: feature extraction and encoding are performed on the data of each link to form link feature vectors; a graph structure is constructed through a graph convolutional network, where the nodes are link feature vectors, the associated weights between the nodes are learned, and the links with high weights are clustered into the same link cluster;

[0023] The process of the model construction unit constructing a sub-digital twin model is as follows: an interaction feature vector is constructed based on the interaction data, the similarity of the interaction feature vectors within the link cluster is calculated, and through the K-means clustering algorithm, the links are merged from high to low according to the similarity to form clusters, and a sub-digital twin model is constructed.

[0024] Preferably, the process of the risk identification unit training the risk identification model is as follows: using historical operation data as input and the corresponding normal or risk labels as output to train a deep learning network model; the rule of the association analysis unit for determining the associated links and the degree of association tightness is: if the risk level of the risk link is high, the direct upstream and downstream links are determined to be highly associated, and the indirect upstream and downstream links are determined to be moderately associated; if the risk level of the risk link is low, the direct upstream and downstream links are determined to be moderately associated, and the indirect upstream and downstream links are determined to be lowly associated.

[0025] Preferably, the conversion rule of the association tightness conversion unit is: high association corresponds to a high risk impact degree, medium association corresponds to a medium risk impact degree, and low association corresponds to a low risk impact degree;

[0026] The training data of the decision-making generation unit for training the decision-making model is: various historical optimized decision-making plans executed for each link of the supply chain, as well as the corresponding link feature data and risk levels.

[0027] Preferably, the way for the link cluster division unit to learn the association weights using the graph convolutional network is: in the graph structure, the association weights between the link feature vectors are calculated through convolutional operations;

[0028] The method for the model construction unit to calculate the similarity of the interaction feature vectors is: using the Euclidean distance or cosine similarity algorithm to calculate the similarity between the interaction feature vectors of each link.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows: Data is obtained through multiple channels including sensors, Internet of Things devices, and enterprise information system interfaces, comprehensively collecting static data, dynamic data, and interaction data in all links of the supply chain, solving the problems of limited data collection means and inability to fully cover in the prior art, and providing a data basis for subsequent analysis and decision-making; Using a deep learning network model to train historical operation data, real-time monitoring the operation status of all links in the supply chain, timely identifying risk links and determining the risk level. At the same time, according to the risk level and digital twin model, determine the associated links and the degree of association tightness of the risk links, solving the problem of insufficient risk identification and early warning capabilities in the prior art, and being able to discover potential risk points more timely and accurately and take effective countermeasures; Transforming the degree of association tightness of the associated links into the degree of risk impact, combining the characteristic data, risk level, etc. of the risk links and associated links, using support vector machines to train and generate a decision-making model, outputting an optimized decision-making plan. The decision-making optimization method based on real-time data and complex association relationship analysis overcomes the limitation that the decision-making model in the prior art overly relies on historical data, can better adapt to market changes, and improve the scientificity and effectiveness of decision-making; Through the visual presentation module, display the overall digital twin model of the supply chain, as well as the implementation situation and effect feedback of the optimized decision-making plan, enabling decision-makers to intuitively understand the operation status and optimization effect of the supply chain, solving the problem that the visual presentation method in the prior art is not intuitive enough, improving the judgment and decision-making efficiency of decision-makers, and helping to quickly respond to market changes and optimize supply chain management. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary, and for those of ordinary skill in the art, without creative efforts, other implementation drawings can also be obtained based on the provided drawings.

[0031] Figure 1 It is a structural diagram of a supply chain intelligent optimization decision-making system based on digital twins. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] The drawings are only for exemplary illustration and should not be construed as a limitation of this patent;

[0033] For better illustration of this embodiment, some components in the drawings will be omitted, enlarged, or reduced, and do not represent the dimensions of actual products;

[0034] For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0035] The technical solution of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0036] Embodiment

[0037] System Construction and Deployment

[0038] Hardware Environment Setup: Select a high-performance server as the core hardware platform for system operation, equipped with an Intel Xeon processor, sufficient memory of at least 128GB, and a large-capacity high-speed storage device such as an SSD solid-state drive, with a capacity estimated according to the data volume, starting from at least 1TB, to meet the needs of large-scale data processing and complex model operations. At the same time, configure multiple client computers for personnel in different departments, such as supply chain management, data analysis, decision-making levels, etc., to access. The configuration of the client computers should meet the basic office requirements and have a stable network connection for easy remote access to the system.

[0039] Software Environment Configuration: Install a Linux operating system on the server side, such as the CentOS or Ubuntu Server version, and utilize its stability and efficiency to provide good underlying support for system operation. Deploy Python 3.8 and above as the main development language environment because it has rich machine learning and data analysis libraries, such as TensorFlow, PyTorch, scikit-learn, pandas, numpy, etc., which are convenient for subsequent algorithm implementation and model construction. At the same time, install a relational database such as MySQL or PostgreSQL to store structured data, such as enterprise production capacity, inventory status, supplier information, etc., and match it with a Redis cache system to improve data access speed and optimize system performance. In addition, to achieve the visualization presentation function, install and configure a data visualization tool, such as Tableau Server or PowerBI Server, and integrate it into the system architecture to ensure that an intuitive and easy-to-understand visualization interface can be generated.

[0040] System Architecture Design: The system is designed using a layered architecture pattern, which is divided into a data acquisition layer, a data processing layer, a business logic layer, and an application presentation layer. The data acquisition layer is responsible for connecting to various data sources, including sensors, Internet of Things devices, and enterprise internal information system interfaces, to ensure comprehensive and real-time data collection. The data processing layer performs preprocessing operations such as cleaning, transforming, and feature extraction on the collected raw data to improve data quality and prepare for subsequent analysis and modeling. The business logic layer includes core components such as a supply chain digital twin construction module, a supply chain risk monitoring module, and a decision optimization and execution module, which use technologies such as graph convolutional networks, K-means clustering algorithms, and support vector machines to implement functions such as supply chain modeling, risk assessment, and optimization decision-making. The application presentation layer displays the overall supply chain digital twin model, risk monitoring results, and the execution status and effect feedback of the optimization decision-making plan to users through a web interface or a mobile application.

[0041] The supply chain intelligent optimization decision-making system based on digital twin, please refer to Figure 1 , including:

[0042] A supply chain digital twin construction module, which is used to collect detailed data of each link in the supply chain, including enterprise production capacity, inventory status, logistics transportation parameters, and supplier information, as well as interaction data between each link, including order flow, information sharing records, and collaborative operation processes; through a graph convolutional network, divide each link in the supply chain into multiple link clusters according to the link data; use the K-means clustering algorithm to build sub-digital twin models corresponding to each link cluster based on the interaction data, and then integrate and build the overall supply chain digital twin model;

[0043] A supply chain risk monitoring module, which is used to monitor the operation status of each link in the supply chain in real time, including production progress, inventory changes, and logistics tracks; determine the links with operating status deviating from the normal range as risk links, and evaluate the risk levels of the risk links; based on the overall supply chain digital twin model and the risk levels, determine multiple associated links of the risk links and the degree of association tightness of each associated link;

[0044] A decision optimization and execution module, which formulates and executes an optimization decision-making plan for the risk link and its associated links according to the characteristic data and risk levels of the risk link, as well as the characteristic data and degree of association tightness of the associated links. The plan covers production plan adjustment, inventory allocation, logistics route optimization, and supplier cooperation strategy change;

[0045] The visualization presentation module is used to display the overall supply chain digital twin model, as well as the execution status and effect feedback of the optimization decision-making plan.

[0046] The supply chain digital twin construction module includes:

[0047] The data acquisition unit obtains static data, dynamic data of each link in the supply chain, and interaction data between links through multiple channels such as sensors, Internet of Things devices, and enterprise information system interfaces;

[0048] The link cluster division unit converts the data of each link into feature vectors, uses a graph convolutional network to learn the correlation weights between the feature vectors, and divides similar links into the same link cluster according to the weights;

[0049] The model construction unit constructs interaction feature vectors according to the interaction data, uses the K-means clustering algorithm to perform clustering analysis on the interaction feature vectors within the link cluster, constructs sub digital twin models, and finally integrates them to form an overall digital twin model of the supply chain.

[0050] Comprehensively collect data of each link in the supply chain through multiple channels. Use sensors and Internet of Things devices to obtain the operating parameters of equipment on the production line in real time, such as temperature, pressure, production speed, etc., the location and status information of logistics transport vehicles, such as speed, driving route, cargo loading conditions, etc., and warehouse environment data, such as temperature, humidity, inventory quantity, etc. At the same time, deeply integrate with enterprise internal information systems, such as ERP, CRM, WMS, etc., to obtain static data and dynamic data such as enterprise production capacity, inventory status, sales orders, customer information, supplier information, etc. In addition, collect interaction data between each link, including order flow, the entire process path from customer order placement to supplier order receipt, information sharing records, historical records of communication and collaboration, data exchange between different links, and collaborative operation processes, such as the connection process between production and logistics, the collaborative plan between suppliers and producers, etc., to comprehensively reflect the actual operation of the supply chain.

[0051] Clean a large amount of collected raw data, remove duplicate, incorrect, and missing data records to ensure the accuracy and integrity of the data. Perform standardization processing on different types of data to unify the data format and measurement unit. For example, unify time data into a standard time format, convert inventory quantity data from different sources into a unified measurement unit, perform encoding processing on text data, and convert unstructured text information into structured feature vectors for subsequent algorithm processing. At the same time, perform normalization or standardization operations on the data to make the data fall within a suitable numerical range, improving the efficiency and accuracy of model training.

[0052] Convert the data of each link after preprocessing into feature vectors. Each feature vector contains multiple dimensions, such as key indicators like production capacity, inventory turnover rate, and logistics response time. Use a graph convolutional network to construct a graph structure, where nodes represent the feature vectors of each link. Through graph convolutional operations, learn the association weights between nodes to quantify the degree of mutual influence between different links. Based on the learned weights, divide similar links into the same link cluster. For example, divide the closely related processing and assembly links in the production process into one cluster, and divide the main line transportation and distribution links in logistics transportation into another cluster, ensuring that the links within the same cluster have high similarity and relevance in terms of function and business process.

[0053] Construct interaction feature vectors based on interaction data to reflect the interaction relationships between link clusters in aspects such as information flow, logistics flow, and capital flow. For example, the raw material supply frequency and quantity between suppliers and producers, and the goods transportation arrangements between producers and logistics providers. Use the K-means clustering algorithm to perform clustering analysis on the interaction feature vectors within the link cluster, combine and merge the links with similar interaction patterns according to the similarity to form clusters. For each cluster, construct a sub-digital twin model to describe in detail the operating mechanism, mutual relationships of the links within the cluster, and the interaction methods with external links, ensuring that the sub-model can accurately map the operation of the corresponding part in the actual supply chain.

[0054] Integrate each sub-digital twin model according to its actual position and interaction relationship in the supply chain to construct a complete overall digital twin model of the supply chain. In this model, it includes the basic information and operating status of each link, as well as the complex associations and collaborative operation processes between links, forming a dynamic, real-time, and accurate virtual mapping of the supply chain, providing a model basis for subsequent risk monitoring and decision optimization.

[0055] The supply chain risk monitoring module includes:

[0056] A real-time monitoring unit that continuously collects real-time operation data of each link in the supply chain;

[0057] A risk identification unit that trains a risk identification model based on historical operation data and a preset risk index system, inputs the real-time monitoring data into the model, identifies the risk links and determines the risk levels;

[0058] An analysis unit that determines the associated links according to the overall digital twin model of the supply chain and combines the risk levels of the risk links, and determines the degree of association tightness according to the preset rules.

[0059] The real-time monitoring unit continuously collects the real-time operation data of each link in the supply chain, including production progress, such as the number of products completed on the production line, production stages, etc., inventory changes, such as increases and decreases in inventory quantity, inventory turnover rate, etc., and logistics trajectories, such as the real-time location of transport vehicles, estimated arrival time, etc. These real-time data are transmitted to the system through the data acquisition interface, and after preprocessing, they are updated to the overall digital twin model of the supply chain to ensure that the model can reflect the actual operation status of the supply chain in real time.

[0060] Based on historical operation data and a preset risk index system, the risk identification unit uses a deep learning network model for training. It takes the normal operation samples and risk event samples in the historical data as inputs and the corresponding normal or risk labels as outputs, enabling the model to learn the potential risk feature patterns in the data. When real-time monitoring data is input, the model can automatically identify whether there are risk links and determine their risk levels, such as high, medium, and low levels. For example, if the failure rate of equipment in a certain production link suddenly rises beyond the normal range and has a high similarity with the characteristics of historical risk events, it is identified as a high-risk link.

[0061] According to the overall digital twin model of the supply chain and in combination with the risk levels of risk links, the analysis unit determines their associated links. For links with a high risk level, it determines the directly upstream and downstream links as highly associated and the indirectly upstream and downstream links as moderately associated; for links with a low risk level, it determines the directly upstream and downstream links as moderately associated and the indirectly upstream and downstream links as lowly associated. For example, if a high-risk equipment failure occurs in a certain production link, its upstream raw material supply link and downstream product assembly link are determined as highly associated, while the raw material procurement link of the more upstream supplier and the logistics distribution link of the more downstream are determined as moderately associated.

[0062] The decision optimization execution module includes:

[0063] The association tightness conversion unit converts the association tightness of associated links into corresponding risk impact degrees according to specific conversion rules;

[0064] The decision generation unit uses a support vector machine. Taking the feature data, risk levels, and risk impact degrees of risk links and associated links as inputs, it trains and generates a decision model and outputs an optimized decision plan for each link;

[0065] The execution control unit is responsible for implementing the optimized decision plan output by the decision generation unit, regulating the risk links and their associated links in the supply chain, and tracking the execution effect.

[0066] The correlation intensity conversion unit converts the correlation intensity (high, medium, low) of the correlation link into the corresponding risk impact level (high, medium, low) according to specific conversion rules. The decision-making generation unit collects the characteristic data of the risk link and the correlation link, such as production efficiency, inventory level, logistics cost, etc., risk levels, and risk impact levels as input samples. Combining various historical optimization decision-making plans and their effect evaluation data executed on each link of the supply chain, it uses a support vector machine to train and generate a decision-making model. This model can output optimization decision-making plans for each link according to the current input multi-dimensional data, covering production plan adjustments, such as adjusting production task volume, priority, inventory allocation, such as increasing or decreasing inventory reserves, optimizing inventory layout, logistics route optimization, such as re-planning transportation routes, adjusting transportation methods, and supplier cooperation strategy changes, such as adjusting procurement volume, changing suppliers, etc.

[0067] The execution control unit is responsible for implementing the optimization decision-making plan output by the decision-making generation unit. Through interfaces with enterprise internal execution systems, such as the manufacturing execution system MES, logistics scheduling system, etc., it issues decision-making instructions to the corresponding links to regulate the risk links and their correlation links in the supply chain. At the same time, it continuously tracks the execution effect, collects feedback data during the execution process, such as production efficiency improvement, inventory cost changes, logistics on-time delivery rate, etc., and feeds these data back to the system for further optimizing the decision-making model, forming a closed-loop decision-making optimization execution process to continuously improve the accuracy and effectiveness of the decision-making.

[0068] The system provides a user interaction interface to support multi-role user login and operation. Supply chain managers can view risk monitoring results, optimization decision-making plans and their execution situations, and conduct manual intervention or adjust decision-making parameters for abnormal situations. Data analysts can deeply mine data, view detailed data reports and analysis results to provide data support for decision-making optimization. Decision-making leaders can quickly understand the overall operation status and key indicators of the supply chain through the overview interface to assist in formulating enterprise strategic plans. At the same time, the system supports users to customize reports and views to meet the personalized needs of different users and improve user experience and work efficiency.

[0069] Regularly evaluate the accuracy and timeliness of the overall digital twin model of the supply chain. According to the actual business changes and data accumulation situation, update and optimize the model. For example, when an enterprise introduces new production technologies, logistics partners or adjusts business processes, timely adjust the corresponding links and parameters in the model to ensure that the model can truly reflect the actual operation of the supply chain. At the same time, optimize the algorithm and data structure of the model to improve the operation efficiency and calculation speed of the model to meet the needs of large-scale data processing and real-time analysis.

[0070] Continuously pay attention to the latest research results in the fields of machine learning, data mining, etc., and combine the characteristics of supply chain operations to optimize and improve algorithms such as graph convolutional networks, K-means clustering algorithms, and support vector machines used in the system. For example, by adjusting the hyperparameters of the algorithms, improving the loss function, introducing new training techniques, etc., to enhance the performance and generalization ability of the algorithms, making them more accurate and efficient in risk identification, decision optimization, etc.

[0071] Monitor and analyze the overall performance of the system, including indicators such as server resource utilization, data transfer speed, response time, etc. According to the performance analysis results, optimize the system architecture and resource configuration, such as adjusting the server load balancing strategy, optimizing the database index and query statements, upgrading the network bandwidth, etc., to ensure that the system can still operate stably and efficiently under high-concurrency access and large-data-volume processing conditions. At the same time, regularly scan and repair the security vulnerabilities of the system to ensure the information security of the system and prevent data leakage and malicious attacks.

[0072] The process of the link cluster division unit constructing feature vectors and dividing link clusters is as follows: Extract and encode the features of each link's data to form link feature vectors; Construct a graph structure through a graph convolutional network, where the nodes are link feature vectors, learn the association weights between nodes, and cluster the links with high weights into the same link cluster;

[0073] The process of the model construction unit constructing a sub-digital twin model is as follows: Construct interaction feature vectors based on interaction data, calculate the similarity of the interaction feature vectors within the link cluster, and through the K-means clustering algorithm, merge the links from high to low similarity to form clusters, and construct a sub-digital twin model.

[0074] The process of the risk identification unit training the risk identification model is as follows: Use historical operation data as input and the corresponding normal or risk labels as output to train a deep learning network model; The rule of the association analysis unit for determining the associated links and the degree of association tightness is: If the risk level of the risk link is high, directly determine the upstream and downstream links as highly associated, and the indirect upstream and downstream links as moderately associated; If the risk level of the risk link is low, directly determine the upstream and downstream links as moderately associated, and the indirect upstream and downstream links as lowly associated.

[0075] The conversion rule of the degree of association tightness conversion unit is: High association corresponds to a high risk impact degree, moderate association corresponds to a medium risk impact degree, and low association corresponds to a low risk impact degree;

[0076] The training data for the decision generation unit to train the decision model is: Various historical optimization decision-making schemes executed for each link in the supply chain, as well as the corresponding link feature data and risk levels.

[0077] The method by which the link cluster division unit uses a graph convolutional network to learn the association weights is as follows: in the graph structure, the association weights between the feature vectors of each link are calculated through convolutional operations;

[0078] The method by which the model construction unit calculates the similarity of the interaction feature vectors is as follows: the similarity between the interaction feature vectors of each link is calculated using the Euclidean distance or cosine similarity algorithm.

[0079] In a specific implementation, an electronic manufacturing enterprise faced problems such as difficulties in coordinating upstream and downstream of the supply chain, serious inventory backlogs, and delays in logistics distribution, resulting in low production efficiency and decreased customer satisfaction. After introducing this system, by comprehensively collecting data on each link of the supply chain, an accurate digital twin model was constructed. During real-time monitoring, it was found that there was a risk of equipment failure in a key component production link. The system quickly identified and evaluated the risk level as high, and determined that its associated links included the upstream raw material supplier and the downstream assembly production line. The decision optimization execution module generated an optimized decision plan based on the risk impact degree and the feature data of each link: adjusted the production plan, allocated the production task of this component to the standby production line; at the same time, notified the supplier to prepare goods in advance and optimized the logistics distribution route to ensure the timely supply of raw materials. The execution control unit issued instructions to the corresponding links, and the tracking of the execution effect showed that the inventory backlog situation was alleviated, the on-time rate of logistics distribution was improved, and the customer satisfaction was significantly enhanced.

[0080] A clothing enterprise is affected by seasonal and market demand fluctuations, and risks such as inaccurate demand forecasting, slow inventory turnover, and delayed supplier deliveries often occur in the supply chain. After applying this system, using its powerful data collection and analysis capabilities, a digital twin model covering links such as design, procurement, production, logistics, and sales was constructed. During the monitoring process, it was found that the delivery cycle of a certain supplier showed an extended trend. The system evaluated it as a medium risk and determined that its associated links included production scheduling and inventory management. The decision generation unit comprehensively considered factors such as the risk impact degree, the inventory levels and production capacities of each link, and output an optimized decision plan: adjusted the production schedule, and gave priority to processing orders with sufficient inventory; at the same time, negotiated with the supplier to increase temporary purchase orders and optimized the logistics transportation method to speed up the transportation of raw materials. After implementation, the production progress and inventory changes were tracked in real time through the visualization interface, and the results showed that the inventory turnover rate was increased, the production delay situation was effectively controlled, the on-time delivery rate of the enterprise was improved, and the market response ability was significantly enhanced.

[0081] The same or similar reference numerals correspond to the same or similar components;

[0082] The terms used to describe the positional relationship in the drawings are for illustrative purposes only and should not be construed as a limitation of this patent;

[0083] Obviously, the above-mentioned embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation manners here. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention.

Claims

1. A digital-twin-based intelligent optimization decision-making system for supply chains, characterized in that, Including: A supply chain digital twin construction module, which is used to collect detailed data of each link in the supply chain, including enterprise production capacity, inventory status, logistics transportation parameters, and supplier information, as well as interaction data between each link, including order flow, information sharing records, and collaborative operation processes; through a graph convolutional network, each link in the supply chain is divided into multiple link clusters according to the link data; using the K-means clustering algorithm, sub-digital twin models corresponding to each link cluster are constructed based on the interaction data, and then the overall digital twin model of the supply chain is integrated and constructed; A supply chain risk monitoring module, which is used to monitor the operation status of each link in the supply chain in real time, including production progress, inventory changes, and logistics trajectories; determine the links with abnormal operation status as risk links and evaluate the risk levels of the risk links; Based on the overall digital twin model of the supply chain and the risk level, determine multiple associated links of the risk link and the degree of association tightness of each associated link; A decision optimization execution module, which formulates and executes an optimization decision plan for the risk link and its associated links according to the characteristic data and risk level of the risk link, as well as the characteristic data and degree of association tightness of the associated links. The plan covers production plan adjustment, inventory allocation, logistics route optimization, and supplier cooperation strategy change; A visualization presentation module, which is used to display the overall digital twin model of the supply chain, as well as the execution situation and effect feedback of the optimization decision plan.

2. The digital twin-based intelligent optimization decision-making system for supply chain according to claim 1, wherein The supply chain digital twin construction module includes: A data collection unit, which obtains static data, dynamic data of each link in the supply chain, and interaction data between links through multiple channels such as sensors, Internet of Things devices, and enterprise information system interfaces; A link cluster division unit, which converts each link data into a feature vector, uses a graph convolutional network to learn the association weights between feature vectors, and divides similar links into the same link cluster according to the weights; A model construction unit, which constructs an interaction feature vector according to the interaction data, performs clustering analysis on the interaction feature vectors within the link cluster using the K-means clustering algorithm, constructs a sub-digital twin model, and finally integrates and forms the overall digital twin model of the supply chain.

3. The digital twin-based intelligent optimization decision-making system for supply chain according to claim 2, wherein The supply chain risk monitoring module includes: A real-time monitoring unit, which continuously collects real-time operation data of each link in the supply chain; A risk identification unit, which trains a risk identification model based on historical operation data and a preset risk index system, inputs the real-time monitoring data into the model, identifies the risk links and determines the risk levels; An analysis unit, which determines the associated links according to the overall digital twin model of the supply chain and the risk level of the risk link, and determines the degree of association tightness according to the preset rules.

4. The digital-twin-based intelligent optimization decision-making system for supply chain according to claim 3, wherein The decision optimization execution module includes: An association tightness conversion unit, which converts the degree of association tightness of the associated links into the corresponding risk impact degree according to specific conversion rules; A decision generation unit, which uses a support vector machine, takes the characteristic data, risk level, and risk impact degree of the risk link and the associated links as inputs, trains and generates a decision model, and outputs an optimization decision plan for each link; The execution control unit is responsible for implementing the optimized decision-making scheme output by the decision-making generation unit, regulating the supply chain risk links and their associated links, and tracking the implementation effect.

5. The digital-twin-based intelligent optimization decision-making system for supply chain according to claim 4, wherein The process of the link cluster division unit constructing the feature vector and dividing the link clusters is as follows: extracting and encoding the features of each link's data to form a link feature vector; constructing a graph structure through a graph convolutional network, where the nodes are the link feature vectors, learning the association weights between the nodes, and clustering the links with high weights into the same link cluster; The process of the model construction unit constructing the sub-digital twin model is as follows: constructing an interaction feature vector based on the interaction data, calculating the similarity of the interaction feature vectors within the link cluster, and through the K-means clustering algorithm, merging the links from high to low similarity to form clusters and constructing the sub-digital twin model.

6. The digital-twin-based intelligent optimization decision-making system for supply chain according to claim 5, wherein The process of the risk identification unit training the risk identification model is as follows: using the historical operation data as the input and the corresponding normal or risk labels as the output to train the deep learning network model; the rule of the association analysis unit determining the associated links and the degree of association tightness is: if the risk level of the risk link is high, the direct upstream and downstream links are determined to be highly associated, and the indirect upstream and downstream links are determined to be moderately associated; If the risk level of the risk link is low, the direct upstream and downstream links are determined to be moderately associated, and the indirect upstream and downstream links are determined to be lowly associated.

7. The digital-twin-based intelligent optimization decision-making system for supply chain according to claim 6, wherein, The conversion rule of the association tightness conversion unit is: high association corresponds to a high risk impact degree, moderate association corresponds to a medium risk impact degree, and low association corresponds to a low risk impact degree; The training data of the decision-making generation unit for training the decision model is: various historical optimized decision-making schemes executed for each link of the supply chain, as well as the corresponding link feature data and risk levels.

8. The digital twin-based intelligent optimization decision-making system for supply chain according to claim 7, wherein, The way the link cluster division unit uses the graph convolutional network to learn the association weights is: in the graph structure, calculating the association weights between each link feature vector through convolution operations; The method used by the model construction unit to calculate the similarity of the interaction feature vectors is: using the Euclidean distance or cosine similarity algorithm to calculate the similarity between each link's interaction feature vectors.

Images