A Supply Chain Collaborative Management Method and System for Enterprise Clusters
By constructing a supply chain network map of enterprise clusters and utilizing graph neural networks and time-series forecasting models, supply chain collaborative management strategies are generated. This solves the problems of subjective evaluation and insufficient risk prediction in existing technologies, and achieves forward-looking and accurate supply chain management.
Patent Information
- Application Number
- CN202611150025.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-31
- Publication Date
- 2026-11-10
AI Technical Summary
Existing supply chain node evaluations fail to incorporate network topology information, risk management lacks the ability to predict future development trends and risk transmission patterns, and the collaborative strategy generation process fails to effectively link dynamic information with network structure characteristics, resulting in subjective evaluation results and insufficient accuracy of collaborative outcomes.
We construct a supply chain network graph of enterprise clusters, extract structural feature vectors of nodes using graph neural networks, and combine a multi-indicator comprehensive evaluation model and a time series prediction model to generate supply chain collaborative management strategies, including supply and demand matching and capacity sharing sub-modules.
It achieves objectivity in node evaluation and forward-looking risk identification, and the collaborative management strategy fully reflects the topological characteristics of the supply chain network, improving the accuracy of management decisions and dynamic adaptability.
Smart Images

Figure CN123070898A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of supply chain management technology, specifically to a supply chain collaborative management method and system for enterprise clusters. Background Technology
[0002] In the supply chain management scenario of enterprise clusters, there are multi-dimensional relationships among member enterprises, including procurement, supply, logistics and financial transactions. The supply chain network exhibits a complex topology and dynamic evolution characteristics.
[0003] Existing supply chain node evaluation methods typically rely on static indicators such as order fulfillment rate and inventory level to evaluate supply chain nodes independently. In terms of supply chain risk management, existing methods often use the results of a single collaboration as a reference and perform ex-post statistics on historical dynamic data of the supply chain. In collaborative scenarios such as supply and demand matching and capacity sharing, existing methods typically perform pairing calculations based on single node attributes.
[0004] The aforementioned existing technologies have the following shortcomings: node evaluation fails to incorporate network topology information, resulting in evaluation results that cannot objectively reflect the actual contribution of nodes in the network; risk management lacks the ability to predict future development trends and risk transmission patterns, making it difficult to identify potential risks in advance; and the collaborative strategy generation process fails to effectively link the dynamic information of the supply chain with network structure characteristics, leading to insufficient accuracy of collaborative results. These shortcomings collectively prevent enterprise clusters from obtaining objective, forward-looking supply chain collaborative management strategies that fully reflect network characteristics. Summary of the Invention
[0005] This invention provides a supply chain collaborative management method and system for enterprise clusters, which solves the technical problems in related technologies such as the difficulty in effectively extracting the characteristics of the supply chain network structure, the unclear path of enterprise risk transmission, and the lack of dynamic adaptive capability of supply chain collaborative management strategies.
[0006] This invention discloses a supply chain collaborative management method for enterprise clusters, comprising the following steps: acquiring multi-source operational data of each member enterprise in the enterprise cluster; generating a supply chain network graph with each member enterprise as a node and the procurement, supply, logistics, and financial relationships between enterprises as edges; wherein each node carries static and dynamic attributes, and each edge carries historical transaction frequency, average delivery cycle, and transaction amount; extracting structural feature vectors for each node in the supply chain network graph using a graph neural network; the graph neural network uses the adjacency structure of the supply chain network graph as the message passing path, updating the hidden features of the target node by aggregating the feature representations of neighboring nodes layer by layer. The hidden states are mapped to structural feature vectors through a linear mapping layer. The structural feature vectors of each node are input into a multi-index comprehensive evaluation model to calculate the comprehensive evaluation score of each node. The dynamic attribute time series of nodes in the supply chain network graph within the historical time window are obtained. The time series prediction model is used to predict the dynamic attribute prediction value of each node in the future prediction time domain. Based on the dynamic attribute prediction value, the risk index of each node is calculated. When the risk index exceeds the preset threshold, the risk transmission impact of downstream nodes is calculated layer by layer along the outgoing edge direction. The comprehensive evaluation score and risk transmission impact of each node are input into the collaborative strategy generation module to generate a supply chain collaborative management strategy.
[0007] Furthermore, the step of generating the supply chain network map also includes: dividing the nodes into raw material supply layer, intermediate processing layer and end demand layer according to the upstream and downstream positions of each node in the supply chain, and recording the level to which each node belongs in the supply chain network map.
[0008] Furthermore, in the step of extracting structural feature vectors using a graph neural network, the graph neural network uses the layer label to which the node belongs as the training target, adopts the cross-entropy loss function, and uses the Adam optimization algorithm for parameter update.
[0009] Furthermore, the step of extracting structural feature vectors using graph neural networks further includes: concatenating the static and dynamic attributes of nodes after independent linear mapping, as the initial input features of the nodes in the graph neural network; wherein the dimension after mapping of static attributes is the same as the dimension after mapping of dynamic attributes.
[0010] Furthermore, when aggregating neighborhood information, the graph neural network assigns different aggregation weights to first-order neighbors and second-order neighbors, with the aggregation weight of first-order neighbors being higher than that of second-order neighbors.
[0011] Furthermore, the risk index Determined by the following formula:
[0012]
[0013] in, For node indexing, For nodes The deviation between the predicted order fulfillment rate and its historical average For nodes The difference between the predicted inventory level and the safety stock threshold, For nodes The ratio of the projected capital turnover cycle to the industry benchmark. , , These are the weighting coefficients determined by offline fitting of historical data.
[0014] Furthermore, the magnitude of the risk transmission impact Determined by the following formula:
[0015]
[0016] in, For upstream node indexing, For downstream node indexing, For the edge Historical transaction frequency For the edge The transaction amount For the edge The average delivery cycle.
[0017] Furthermore, the collaborative strategy generation module includes a supply-demand matching submodule and a capacity sharing submodule. The supply-demand matching submodule takes the comprehensive evaluation score and the risk transmission impact as input, and prioritizes supply-layer nodes with high comprehensive evaluation scores and low risk transmission impact as matching objects to generate a candidate supply node list for each demand-layer node. The capacity sharing submodule takes the capacity scale attribute, inventory level prediction value, and order fulfillment rate prediction value of each node as input, identifies nodes with capacity redundancy and nodes with capacity gap, and generates a capacity sharing pairing scheme.
[0018] Furthermore, the step of calculating the comprehensive evaluation score of each node also includes: performing a moving average on the comprehensive evaluation score of each node over multiple consecutive update cycles, and using the moving average comprehensive evaluation score as the final comprehensive evaluation score; wherein, the window length of the moving average is determined according to the business cycle.
[0019] Furthermore, after each update of multi-source operational data, a recalculation process is triggered, which includes extracting structural feature vectors, calculating comprehensive evaluation scores, calculating the impact of risk transmission, and generating supply chain collaborative management strategies. The updated supply chain network map is then used to regenerate the supply chain collaborative management strategies.
[0020] This invention discloses a supply chain collaborative management system for enterprise clusters, comprising: a data acquisition module for acquiring multi-source operational data from each member enterprise in the enterprise cluster, generating a supply chain network graph with each member enterprise as a node and the procurement, supply, logistics, and financial relationships between enterprises as edges; wherein each node carries static and dynamic attributes, and each edge carries historical transaction frequency, average delivery cycle, and transaction amount; and a feature extraction module for extracting structural feature vectors from each node in the supply chain network graph using a graph neural network; the graph neural network uses the adjacency structure of the supply chain network graph as the message passing path, updating the hidden state of the target node by aggregating the feature representations of neighboring nodes layer by layer, and then... The mapping layer maps hidden states to structural feature vectors; the comprehensive evaluation module inputs the structural feature vectors of each node into a multi-index comprehensive evaluation model to calculate the comprehensive evaluation score of each node; the risk analysis module obtains the time series of dynamic attributes of nodes in the supply chain network graph within a historical time window, uses a time series prediction model to predict the dynamic attribute prediction value of each node in the future prediction time domain, and calculates the risk index of each node based on the dynamic attribute prediction value. When the risk index exceeds a preset threshold, the risk transmission impact of downstream nodes is calculated layer by layer along the outgoing edge direction; the strategy generation module inputs the comprehensive evaluation score and risk transmission impact of each node into the collaborative strategy generation module to generate a supply chain collaborative management strategy.
[0021] The beneficial effects of this invention are as follows:
[0022] This invention constructs a supply chain network graph carrying dynamic and static attributes from multi-source operational data and extracts structural feature vectors for each node using graph neural networks. This allows the node's comprehensive evaluation score to simultaneously incorporate both the node's own attributes and network location information, solving the technical problem of existing methods relying solely on static indicators and resulting in subjective node evaluations. This achieves a more objective reflection of the node's actual operational performance within the supply chain network. Furthermore, by predicting historical dynamic attribute time series and calculating the impact of risk transmission, this invention addresses the technical problem of existing methods lacking the ability to predict future development trends and risk transmission patterns, achieving a forward-looking risk identification capability for management decisions. Finally, by simultaneously incorporating comprehensive evaluation scores and risk transmission trends into the generation process of collaborative strategies for supply-demand matching and capacity sharing, this invention solves the technical problem of existing methods failing to effectively link dynamic information and network structure characteristics in collaborative analysis, achieving a technical effect where collaborative management strategies fully reflect the topological characteristics of the supply chain network. Attached Figure Description
[0023] Figure 1 This is a flowchart of a supply chain collaborative management method for enterprise clusters provided in an embodiment of the present invention. Detailed Implementation
[0024] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.
[0025] Example 1
[0026] In supply chain management for enterprise clusters, existing methods generally remain at the level of static indicators, failing to effectively mine the topological structure and dynamic correlation information of the supply chain network, resulting in insufficient objectivity in node evaluation. Furthermore, existing methods lack the ability to predict future development trends and risk transmission patterns in the supply chain, and fail to fully reflect the network characteristics of the supply chain in collaborative scenarios such as supply-demand matching and capacity sharing. This embodiment provides a supply chain collaborative management method for enterprise clusters, addressing the above problems by constructing a supply chain network map, mining dynamic correlation information, predicting risk transmission trends, and generating collaborative management strategies. This method runs on a server cluster with data storage and computing capabilities, maintaining data communication connections between the server cluster and the enterprise cluster's business systems.
[0027] A supply chain collaborative management method according to an embodiment of the present invention includes the following steps:
[0028] Step 1: Obtain multi-source operational data of the enterprise cluster and generate a supply chain network map.
[0029] Acquire multi-source operational data from each member enterprise in the enterprise cluster, including order flow data, inventory data, logistics data, contract data, and financial settlement data. Using each member enterprise as a node and the procurement, supply, logistics, and financial relationships between enterprises as edges, generate a supply chain network graph. In the supply chain network graph, each node carries the corresponding enterprise's static attributes (including enterprise type, production capacity, and geographical location) and dynamic attributes (including current inventory level, order fulfillment rate, and capital turnover cycle); each edge carries relationship attributes, including historical transaction frequency, average delivery cycle, and transaction amount.
[0030] It should be noted that the update cycle of the aforementioned dynamic attributes refers to the attribute refresh interval determined based on the data synchronization frequency of the business system. It can be a periodic refresh on a daily basis, or a real-time refresh driven by a triggering event. For example, when the order fulfillment rate of a certain node changes, the dynamic attributes of that node and the relationship attributes of its associated edges are updated immediately.
[0031] In this embodiment of the application, to more accurately reflect the hierarchical structure of the supply chain network, nodes can be hierarchically labeled when generating the supply chain network map. Specifically, based on the upstream and downstream positions of each node in the supply chain, the nodes are divided into a raw material supply layer, an intermediate processing layer, and an end-user demand layer, and the level to which each node belongs is recorded in the supply chain network map. Hierarchical labeling enables subsequent risk transmission analysis to distinguish the differences in the paths of risk propagation along the upstream and downstream directions.
[0032] Step 2: Extract the structural feature vectors of each node based on the supply chain network graph.
[0033] For each node in the supply chain network graph, a graph neural network (GNN) is used to extract structural feature vectors. The input to the GNN consists of the static and dynamic attributes of each node in the supply chain network graph, as well as the relational attributes of its neighbors. The output is the structural feature vector of each node, which encodes a combination of the node's own attributes and its structural position in the network. The GNN uses the adjacency structure of the supply chain network graph as the message passing path, updating the hidden state of the target node by aggregating the feature representations of neighboring nodes layer by layer. Finally, a linear mapping layer maps the hidden state of each node into a fixed-dimensional structural feature vector. During training, node classification is used as the supervision signal, and the hierarchical label of each node is used as the training objective. The cross-entropy loss function is employed, and the Adam optimization algorithm is used for parameter updates.
[0034] It should be noted that the aforementioned neighborhood refers to the first-order neighbor nodes directly connected to the target node in the supply chain network graph, as well as the second-order neighbor nodes directly connected to the first-order neighbor nodes. When aggregating neighborhood information, GNN assigns different aggregation weights to first-order and second-order neighbors, with the aggregation weight of first-order neighbors being higher than that of second-order neighbors, in order to preserve the dominant role of proximity relationships in node features.
[0035] In this embodiment, to enable the structural feature vector to simultaneously capture both the static and dynamic parts of node attributes, the static and dynamic attributes of a node can be concatenated after undergoing independent linear mappings, serving as the initial input features for the GNN node. The dimension after the static attribute mapping is the same as the dimension after the dynamic attribute mapping, and the dimension of the concatenated initial input feature is twice that of the single mapping, thus ensuring that the two types of attributes have equal representational weights in the initial features.
[0036] Step 3: Calculate the comprehensive evaluation score of each node based on the structural feature vector of each node.
[0037] The structural feature vectors of each node are input into a multi-index comprehensive evaluation model to calculate the comprehensive evaluation score for each node. The multi-index comprehensive evaluation model is a linear regression model; its input layer receives the structural feature vectors of each node, and its output layer outputs the comprehensive evaluation score scalar for the corresponding node. The comprehensive evaluation score reflects the overall operational performance and collaborative contribution of each node in the supply chain network. During training, manually labeled node operational quality ratings from historical periods are used as the supervision objective, and the mean squared error loss function is employed, with gradient descent used for parameter updates.
[0038] It should be noted that the comprehensive evaluation score mentioned above refers to the scalar value obtained by multiplying the components of the structural feature vector corresponding to performance capability, financial health, network centrality, and response speed by their respective evaluation weights and then summing them. Each evaluation weight is determined offline using the analytic hierarchy process (AHP) based on historical data and is updated periodically with the business cycle.
[0039] In this embodiment, to avoid interference from dynamic attribute fluctuations at a single time point on the comprehensive evaluation score, a moving average can be applied to the comprehensive evaluation score of each node over multiple consecutive update cycles, and the moving average score is used as the final comprehensive evaluation score. The window length of the moving average is determined according to the business cycle, so that the final comprehensive evaluation score can reflect the stable operating level of the node over a period of time, rather than the instantaneous state at a single moment.
[0040] Step 4: Based on historical dynamic data of the supply chain network map, predict the risk transmission trend of each node.
[0041] This study obtains the dynamic attribute sequences of nodes and edge relationship attribute sequences at each moment within a historical time window of the supply chain network graph. The time series of dynamic attributes of each node are input into a time series prediction model, which employs a Long Short-Term Memory (LSTM) network. The input layer receives the time series of dynamic attributes of each node within the historical time window, and the output layer maps the hidden states to predicted dynamic attribute values for each node in the future prediction time domain via a linear mapping layer. During training, the known true values of future dynamic attributes from historical data are used as the supervision objective, employing a mean squared error loss function and the Adam optimization algorithm for parameter updates. Based on the predicted dynamic attribute values of each node, the risk index of each node in the future prediction time domain is calculated.
[0042] Furthermore, the risk index Determined by the following formula:
[0043]
[0044] in, For node indexing, For nodes The risk index, For nodes The deviation between the predicted order fulfillment rate and its historical average For nodes The difference between the predicted inventory level and the safety stock threshold, For nodes The ratio of the projected capital turnover cycle to the industry benchmark. , , The corresponding weighting coefficients are determined by offline fitting of historical data.
[0045] It should be noted that the aforementioned risk transmission trend refers to the risk transmission impact on downstream nodes, calculated layer by layer along the outgoing edges of that node in the supply chain network diagram, when the risk index of a certain node exceeds a preset threshold. Risk transmission impact. Determined by the following formula:
[0046]
[0047] in, For upstream node indexing, For downstream node indexing, For the edge Historical transaction frequency For the edge The transaction amount For the edge Average delivery cycle Quantified upstream nodes Risks to downstream nodes The potential impact strength.
[0048] In this embodiment of the application, in order to improve the accuracy of risk transmission trend prediction, when predicting the time series of dynamic attributes of each node, the structural information of the supply chain network graph can also be introduced into the time series prediction process. Specifically, at each prediction time step, the structural feature vector of each node at the current time is concatenated with the dynamic attribute time series features of the corresponding node, and used together as the input of the time series prediction model at that time step, so that the prediction result includes both time series trend information and network structure information.
[0049] Step 5: Generate a supply chain collaborative management strategy based on the comprehensive evaluation score and risk transmission trend.
[0050] The comprehensive evaluation scores and risk transmission trends of each node are input into the collaborative strategy generation module to generate a supply chain collaborative management strategy. The collaborative strategy generation module includes a supply and demand matching submodule and a capacity sharing submodule.
[0051] The supply-demand matching submodule takes the comprehensive evaluation score and the risk transmission impact as input to perform matching calculations between supply-layer nodes and demand-layer nodes. The matching calculation prioritizes supply-layer nodes with high comprehensive evaluation scores and low risk transmission impact, generating an ordered list of candidate supply nodes for each demand-layer node.
[0052] The capacity sharing submodule takes the capacity scale attributes of each node, the current inventory level forecast, and the order fulfillment rate forecast as input to identify nodes with capacity redundancy and nodes with capacity gaps in the future forecast time domain, and generates a capacity sharing pairing scheme. In the capacity sharing pairing scheme, each pair of paired nodes includes shared capacity, shared time window, and protocol priority.
[0053] The collaborative strategy generation module merges the candidate supply node list with the capacity sharing pairing scheme and outputs the supply chain collaborative management strategy. The supply chain collaborative management strategy is presented in the form of a structured report, which includes operational suggestions and risk warning information for each node.
[0054] In this embodiment of the application, in order to enable the collaborative management strategy to dynamically respond to real-time changes in the supply chain network, after each update of operational data, the recalculation process of steps 2 to 5 can be triggered. Based on the updated supply chain network map, the structural feature vectors, comprehensive evaluation scores, risk transmission trends and supply chain collaborative management strategies of each node are regenerated to ensure the consistency between the supply chain collaborative management strategy and the actual operational status.
[0055] In this embodiment, by constructing a supply chain network graph carrying dynamic attributes from multi-source operational data of enterprise clusters, and using graph neural networks to extract structural feature vectors of each node, node evaluation simultaneously incorporates both the node's own attributes and its network location information, overcoming the problem of subjective evaluation caused by existing methods relying solely on static indicators. By predicting historical dynamic attribute time series and calculating the impact of risk transmission, management decisions are equipped with the ability to predict future development trends and risk transmission patterns, overcoming the shortcomings of existing methods that rely on single cooperation results and lack foresight. By simultaneously incorporating comprehensive evaluation scores and risk transmission trends into the collaborative strategy generation process of supply and demand matching and capacity sharing, the collaborative results fully reflect the network topology characteristics of the supply chain, overcoming the problem of existing methods failing to effectively correlate dynamic information in collaborative analysis.
[0056] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.
Claims
1. A supply chain collaborative management method for enterprise clusters, characterized in that, Includes the following steps: The system acquires multi-source operational data from each member enterprise in the enterprise cluster, uses each member enterprise as a node and the procurement, supply, logistics and financial transactions between enterprises as edges, and generates a supply chain network graph. Each node carries static and dynamic attributes, and each edge carries historical transaction frequency, average delivery cycle and transaction amount. For each node in the supply chain network graph, a graph neural network is used to extract structural feature vectors. The graph neural network uses the adjacency structure of the supply chain network graph as the message passing path, and updates the hidden state of the target node by aggregating the feature representations of neighboring nodes layer by layer. The hidden state is then mapped to a structural feature vector through a linear mapping layer. Input the structural feature vectors of each node into the multi-index comprehensive evaluation model to calculate the comprehensive evaluation score of each node; Obtain the time series of dynamic attributes of nodes in the supply chain network map within a historical time window, use a time series prediction model to predict the dynamic attribute prediction value of each node in the future prediction time domain, and calculate the risk index of each node based on the dynamic attribute prediction value. When the risk index exceeds the preset threshold, calculate the risk transmission impact of downstream nodes layer by layer along the outgoing edge direction. The comprehensive evaluation scores of each node and the impact of risk transmission are input into the collaborative strategy generation module to generate a supply chain collaborative management strategy.
2. The supply chain collaborative management method for enterprise clusters according to claim 1, characterized in that, The steps for generating the supply chain network map also include: dividing the nodes into raw material supply layer, intermediate processing layer and end demand layer according to the upstream and downstream positions of each node in the supply chain, and recording the level to which each node belongs in the supply chain network map.
3. The supply chain collaborative management method for enterprise clusters according to claim 2, characterized in that, In the step of extracting structural feature vectors using a graph neural network, the graph neural network uses the layer label to which the node belongs as the training target, adopts the cross-entropy loss function, and uses the Adam optimization algorithm to update the parameters.
4. The supply chain collaborative management method for enterprise clusters according to claim 3, characterized in that, The step of extracting structural feature vectors using graph neural networks further includes: concatenating the static and dynamic attributes of a node after performing independent linear mappings, which serve as the initial input features of the node in the graph neural network; wherein the dimension after mapping the static attributes is the same as the dimension after mapping the dynamic attributes.
5. The supply chain collaborative management method for enterprise clusters according to claim 4, characterized in that, When aggregating neighborhood information, the graph neural network assigns different aggregation weights to first-order neighbors and second-order neighbors, with the aggregation weight of first-order neighbors being higher than that of second-order neighbors.
6. The supply chain collaborative management method for enterprise clusters according to claim 5, characterized in that, The risk index Determined by the following formula: in, For node indexing, For nodes The deviation between the predicted order fulfillment rate and its historical average For nodes The difference between the predicted inventory level and the safety stock threshold, For nodes The ratio of the projected capital turnover cycle to the industry benchmark. , , These are the weighting coefficients determined by offline fitting of historical data.
7. A supply chain collaborative management method for enterprise clusters according to claim 6, characterized in that, The impact of risk transmission Determined by the following formula: in, For upstream node indexing, For downstream node indexing, For the edge Historical transaction frequency For the edge The transaction amount For the edge The average delivery cycle.
8. A supply chain collaborative management method for enterprise clusters according to claim 7, characterized in that, The collaborative strategy generation module includes a supply and demand matching submodule and a capacity sharing submodule. The supply and demand matching submodule takes the comprehensive evaluation score and the risk transmission impact as input, and prioritizes supply layer nodes with high comprehensive evaluation scores and low risk transmission impact as matching objects to generate a candidate supply node list for each demand layer node. The capacity sharing submodule takes the capacity scale attribute, inventory level forecast, and order fulfillment rate forecast of each node as input, identifies nodes with capacity redundancy and nodes with capacity gap, and generates a capacity sharing pairing scheme.
9. A supply chain collaborative management method for enterprise clusters according to claim 8, characterized in that, The step of calculating the comprehensive evaluation score of each node further includes: performing a moving average on the comprehensive evaluation score of each node over multiple consecutive update cycles, and using the moving average comprehensive evaluation score as the final comprehensive evaluation score; wherein, the window length of the moving average is determined according to the business cycle.
10. A supply chain collaborative management method for enterprise clusters according to claim 9, characterized in that, After each update of multi-source operational data, a recalculation process is triggered, including structural feature vector extraction, comprehensive evaluation score calculation, risk transmission impact calculation, and supply chain collaborative management strategy generation. The supply chain collaborative management strategy is then regenerated based on the updated supply chain network map.
