Enterprise supply chain optimization method and device based on graph relationship

Abstract

The invention discloses an enterprise supply chain optimization method and device based on a graph relationship. The method comprises the following steps: constructing a basic graph structure based on enterprise supply chain data; clustering the nodes in the node set, simplifying the nodes conforming to the key attributes of the supply chain in the same class into representative nodes, and constructing a second supply chain graph structure based on the simplified nodes and a new supply relationship between the simplified nodes; constructing a target function, and determining a key path from the second supply chain graph structure based on the target function; the second supply chain graph structure is input into a trained graph neural network model, the graph neural network model determines the probability of forming an edge between nodes without a connection relation in the second supply chain graph structure, and an edge is added between two nodes corresponding to a probability value exceeding a preset threshold value; and taking the formed third supply chain graph structure as an optimization result of the supply chain. According to the method, key nodes and potential bottlenecks in the supply chain are identified and analyzed, and path prediction and multi-dimensional reconstruction are realized.

Description

Technical Field

[0001] The present application belongs to the field of information processing technology, and in particular, relates to a method and device for optimizing an enterprise supply chain based on a graph relationship. Background Art

[0002] In existing technologies, most companies rely on linear and static methods for supply chain optimization, which are inefficient and time-consuming when dealing with complex and dynamic environments, and are difficult to meet demand. Traditional statistical models or manual adjustments are difficult to fully capture the complex relationships between nodes and links, and the path prediction capabilities are insufficient, which limits the reconstruction and completion of multi-dimensional supply chains. Summary of the invention

[0003] The present invention provides a method and device for optimizing an enterprise supply chain based on a graph relationship, which are used to solve the above-mentioned problem or at least partially solve the above-mentioned problem.

[0004] In a first aspect, a graph-based enterprise supply chain optimization method is disclosed, the method comprising:

[0005] Step S1: Acquire enterprise production data, clean and standardize the production data, and extract enterprise supply chain data from it; construct a basic graph structure based on the enterprise supply chain data, the graph structure G = {V, E}, wherein suppliers are taken as nodes, supply relationships are taken as edges, V is a set of nodes, and E is a set of edges; the supply relationship is determined by activities between nodes, and the activities include transportation activities, production activities, and processing activities;

[0006] Step S2: Cluster the nodes in the node set so that the nodes that satisfy the supply relationship similarity exceeding the first threshold and the supply product similarity exceeding the second threshold are classified into the same category, simplify the nodes that meet the key attributes of the supply chain in the same category into representative nodes, and construct a second supply chain graph structure based on the simplified nodes and the new supply relationships between the simplified nodes;

[0007] Step S3: constructing an objective function, and determining a critical path from the second supply chain graph structure based on the objective function, wherein each node in the critical path is a critical node, and the critical node is used as a node to be optimized in the supply chain;

[0008] Step S4: Input the second supply chain graph structure into the trained graph neural network model, the graph neural network model determines the probability of forming an edge between nodes that have no connection relationship in the second supply chain graph structure, and adds an edge between two nodes corresponding to the probability value exceeding a preset threshold; the formed third supply chain graph structure is used as the optimization result of the supply chain.

[0009] Preferably, in step S1, the edges in the basic graph structure have weights, and the weights are determined based on the activity types between nodes and the distances between nodes.

[0010] Preferably, the step S2 simplifies the nodes in the same category that meet the key attributes of the supply chain into representative nodes, based on the simplified nodes and the new supply relationships between the simplified nodes, including:

[0011] The attributes of each node in the class are merged, and the merging method is one of taking the average value, weighted average, and sum of the attribute values; the node in the class whose attribute value is the smallest from the merged attribute value is taken as the node that meets the key attributes of the supply chain, that is, the representative node, and the representative node represents each node of the class; the merged attribute value is assigned to the representative node as the node feature of the representative node, and the representative node inherits the supply relationship between each node it represents and other nodes outside the class as a new supply relationship; based on the weights corresponding to the supply relationships between each node inherited by the representative node and other nodes outside the class, a new weight is established for the new supply relationship.

[0012] Preferably, the step S3: constructs an objective function, and determines a critical path from the second supply chain graph structure based on the objective function, wherein each node in the critical path is a critical node, and the critical node is used as a node to be optimized in the supply chain, wherein:

[0013] The objective function is:

[0014]

[0015] Act={act 1 ,act 2 ,…,act num}

[0016] Where Z is the total execution cost of the entire supply chain network, AC is the minimum node subset that meets the connectivity and uniqueness conditions, and a i is the node belonging to AC in the critical path, n is the total number of nodes on the critical path, c i For node a i The corresponding execution cost includes the total cost of its execution and the corresponding activities. Act is the activity set. 1 ,act 2 ,…,act num All are activities;

[0017] Based on the objective function, a critical path method (CPM) is used to determine a critical path from the second supply chain graph structure; the critical path is a path with the greatest execution consumption from the start node to the end node of the supply chain.

[0018] Preferably, in step S4, the second supply chain graph structure is input into a trained graph neural network model, the graph neural network model determines the probability of forming an edge between nodes that do not have a connection relationship in the second supply chain graph structure, and adds an edge between two nodes corresponding to probability values ​​exceeding a preset threshold; and the formed third supply chain graph structure is used as the optimization result of the supply chain, wherein:

[0019] The graph neural network model adds an attention mechanism on the basis of the RGCN model, that is, an attention module is added between the RGCN layer and the batch normalization layer. The attention module weights the node features corresponding to each node in the second supply chain graph structure output by the RGCN layer to obtain the weights of the importance of the relationships between each node in the second supply chain graph structure, and then weights the weights with the node features corresponding to each node in the second supply chain graph structure, and inputs the obtained results into the batch normalization layer; the decoder of the graph neural network model outputs the probability of forming edges between nodes in the second supply chain graph structure that have no connection relationship.

[0020] In a second aspect, a graph-based enterprise supply chain optimization device is disclosed, the device comprising:

[0021] Initialization module: configured to obtain enterprise production data, clean and standardize the production data, and extract enterprise supply chain data from it; construct a basic graph structure based on the enterprise supply chain data, the graph structure G = {V, E}, wherein suppliers are taken as nodes, supply relationships are taken as edges, V is a set of nodes, and E is a set of edges; the supply relationship is determined by activities between nodes, and the activities include transportation activities, production activities, and processing activities;

[0022] Simplification module: configured to cluster the nodes in the node set, so that the nodes that satisfy the supply relationship similarity exceeding the first threshold and the supply product similarity exceeding the second threshold are classified into the same category, and the nodes that meet the key attributes of the supply chain in the same category are simplified into representative nodes, and the second supply chain graph structure is constructed based on the simplified nodes and the new supply relationships between the simplified nodes;

[0023] A critical path module: configured to construct an objective function, and determine a critical path from the second supply chain graph structure based on the objective function, wherein each node in the critical path is a critical node, and the critical node is used as a node to be optimized in the supply chain;

[0024] Optimization module: configured to input the second supply chain graph structure into a trained graph neural network model, the graph neural network model determines the probability of forming an edge between nodes that do not have a connection relationship in the second supply chain graph structure, and adds an edge between two nodes corresponding to probability values ​​that exceed a preset threshold; and uses the formed third supply chain graph structure as the optimization result of the supply chain.

[0025] In a third aspect, an electronic device is disclosed, the electronic device comprising:

[0026] at least one processor; and

[0027] a memory communicatively connected to the at least one processor; wherein,

[0028] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described above.

[0029] In a fourth aspect, a non-transitory computer-readable storage medium storing computer instructions is disclosed, wherein the computer instructions are used to cause the computer to execute the method as described above.

[0030] This application has the following technical effects:

[0031] Improve the efficiency and accuracy of enterprise supply chain optimization, comprehensively capture the complex relationships between nodes and links, identify and analyze key nodes and potential bottlenecks in the supply chain, and realize path prediction and multi-dimensional reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flowchart of the enterprise supply chain optimization method based on graph relations;

[0033] Figure 2 It is another process diagram of the enterprise supply chain optimization method based on graph relationship;

[0034] Figure 3 This is a sample flow chart of the executable path;

[0035] Figure 4 A flow chart of supply chain bottleneck discovery and optimization based on critical paths;

[0036] Figure 5 is the directed acyclic graph corresponding to the sample flow diagram;

[0037] Figure 6 Schematic diagram of the link prediction model RGCNP;

[0038] Figure 7 This is a comparison result diagram of link prediction accuracy;

[0039] Figure 8 This is a schematic diagram of the structure of an enterprise supply chain optimization device based on graph relations. DETAILED DESCRIPTION

[0040] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0041] like Figure 1 As shown, the present application provides a method for optimizing an enterprise supply chain based on a graph relationship, the method comprising:

[0042] Step S1: Acquire enterprise production data, clean and standardize the production data, and extract enterprise supply chain data from it; construct a basic graph structure based on the enterprise supply chain data, the graph structure G = {V, E}, wherein suppliers are taken as nodes, supply relationships are taken as edges, V is a set of nodes, and E is a set of edges; the supply relationship is determined by activities between nodes, and the activities include transportation activities, production activities, and processing activities;

[0043] Step S2: Cluster the nodes in the node set so that the nodes that satisfy the supply relationship similarity exceeding the first threshold and the supply product similarity exceeding the second threshold are classified into the same category, simplify the nodes that meet the key attributes of the supply chain in the same category into representative nodes, and construct a second supply chain graph structure based on the simplified nodes and the new supply relationships between the simplified nodes;

[0044] Step S3: constructing an objective function, and determining a critical path from the second supply chain graph structure based on the objective function, wherein each node in the critical path is a critical node, and the critical node is used as a node to be optimized in the supply chain;

[0045] Step S4: Input the second supply chain graph structure into the trained graph neural network model, the graph neural network model determines the probability of forming an edge between nodes that have no connection relationship in the second supply chain graph structure, and adds an edge between two nodes corresponding to the probability value exceeding a preset threshold; the formed third supply chain graph structure is used as the optimization result of the supply chain.

[0046] In step S1, the edges in the basic graph structure have weights, and the weights are determined based on the activity types between nodes and the distances between nodes.

[0047] The step S2, clustering the nodes in the node set, includes:

[0048] Use the K clustering algorithm or the DBSCAN clustering algorithm to cluster the nodes in the node set.

[0049] In this application, the K-means model is used to realize the selective classification of known supply chains, mainly by iteratively calculating the distance between the core and the nodes to realize the selective K category distinction; the DBSCAN model is used to realize the non-selective classification of known supply chains, mainly by simulating and filtering out the noise in the data to realize the self-selection of the core and the non-selective category distinction.

[0050] The nodes in the same class that meet the key attributes of the supply chain are simplified into representative nodes. Based on the simplified nodes and the new supply relationship between the simplified nodes, in this application, the representative node is determined, for example, a node is selected as the representative node, or a new node is created, or the attributes of each node in the class are merged (such as taking the average, weighted average or sum), and the node in the class whose attribute value is the smallest from the merged attribute value is taken as the node that meets the key attributes of the supply chain, that is, the representative node; the representative node represents each node of the class; the merged attribute value is assigned to the representative node as the feature of the representative node, and the representative node inherits the supply relationship between each node it represents and other nodes outside the class as a new supply relationship; based on the weights corresponding to the supply relationship between each node inherited by the representative node and other nodes outside the class, a new weight is established for the new supply relationship.

[0051] In this application, the edges between nodes in the same class will be removed because they have been merged. The representative node will inherit the connection between the node in its corresponding original class and other nodes (nodes outside the class). The weight of the edge can be adjusted according to the merging strategy, such as taking the average weight or weighted sum of the original edges.

[0052] The step S3: constructing an objective function, and determining a critical path from the second supply chain graph structure based on the objective function, wherein each node in the critical path is a critical node, and the critical node is used as a node to be optimized in the supply chain, wherein:

[0053] The objective function is:

[0054]

[0055] Act={act 1 ,act 2 ,…,act num}

[0056] Where Z is the total execution cost of the entire supply chain network, AC is the minimum node subset that meets the connectivity and uniqueness conditions, and a i is the node belonging to AC in the critical path, n is the total number of nodes on the critical path, c i For node a i The corresponding execution cost includes the total cost of its execution and the corresponding activities. Act is the activity set.1 ,act 2 ,…,act num All are activities;

[0057] Based on the objective function, a critical path method (CPM) is used to determine a critical path from the second supply chain graph structure; the critical path is a path with the greatest execution consumption from the start node to the end node of the supply chain.

[0058] The nodes on the critical path are the key nodes in the supply chain, and are the nodes to be optimized in order to optimize the overall time consumption and efficiency of the supply chain.

[0059] In this application, if Figure 2 As shown in Figure 1, in a complex process involving multiple nodes, multiple such execution bottlenecks may be identified. When these execution bottlenecks are logically linked together to form an execution path, this path is called the critical path of the process. Obviously, each node on the critical path becomes an important target for optimization because they directly affect the ability to reduce the total execution overhead.

[0060] like Figure 3 As shown, there are three executable process paths in the given example:

[0061] P1:e start →e1→g1→e2→e3→g2→e4→g3→g4→e8→e end

[0062] P2: e start →e1→g1→e2→e3→g2→e5→g3→g4→e8→e end

[0063] P3:e start →e1→g1→e6→e7→g4→e8→e end

[0064] from Figure 3 Based on the intuitive judgment based on the node relationship observation alone, it is possible to think that the execution overhead of execution paths P1 and P2 is higher than that of P3. However, considering special cases such as: the total execution overhead of node e6 is extremely large, the execution overhead of P3 will be much greater than that of P1 and P2.

[0065] Proof by contradiction: The path with the longest execution cost is the critical path of the process.

[0066] Assuming that there is a path with non-maximum execution cost as the critical path, according to the critical path definition given above, the total execution cost of the entire process is positively correlated with the execution cost of any activity on this path.

[0067] Take an activity node on this process. If this activity also appears in the longest execution path, the total execution cost is naturally positively correlated with its cost change. If this activity is not in the longest path, the execution cost of the assumed path is positively correlated with the activity cost, but it will not affect the total execution cost of the process determined by the longest path. This shows that the execution cost of the longest path is not affected, and the total execution cost of the process will not change, which contradicts the definition of the critical path given. In summary, the longest execution path in the process is the real critical path.

[0068] The supply chain network structure is defined as:

[0069] S=(E,A,R,C)

[0070] The event set (representing key nodes in the supply chain, such as suppliers, production nodes, distribution centers, etc.) is:

[0071] E={e start ,e 1 ,e 2 ,…,e |E|-2 ,e end}

[0072] The set of activities (representing the activities that connect event nodes in the supply chain, such as transportation, production or processing) is:

[0073] A={a 1 ,a 1 ,…,a |A|}

[0074] The mapping relationship between the E set and the A set is:

[0075] R = {r i :e a ×e b →a i |e a ,e b ∈E,a i ∈A}

[0076] The consumption set (representing the set of resource consumption or cost components corresponding to each activity) is:

[0077] C={c 1 ,c 1 ,…,c |A|}

[0078] According to the definition of execution bottleneck, for a supply chain network structure S, it is necessary to calculate and find a minimum subset AC of A that satisfies:

[0079] ① Connectivity: starting from estart, it can reach eend after passing through all elements in Ac;

[0080] ② Uniqueness: If any element ai in AC is missing, ① cannot be established;

[0081] ③When conditions ① and ② are met, the total consumption of AC is maximum.

[0082] It is hoped that a subset AC as required above can be found and optimized so that the total time consumption of AC is reduced.

[0083] Map the flowchart into a weighted directed acyclic graph. The general process is as follows: Figure 4 shown.

[0084] When modeling the above problem, a weighted directed acyclic graph (DAG) is used:

[0085] G1=(V,U)

[0086] After abstracting the gateway information, Figure 4 The directed acyclic graph corresponding to the graph shown is Figure 5 As shown. Therefore, the original problem is transformed into finding the critical path CP (Critical Path) of the weighted directed acyclic graph G.

[0087] The critical path CP is the longest path from Vstart to Vend. The sum of the total consumption on the critical path CP is the maximum execution consumption to complete the supply chain process, and the activities on the path are the execution bottlenecks of the supply chain network. This is because in the supply chain network, there are activities that can be executed in parallel. If the execution consumption of any activity on the critical path can be reduced, the total execution consumption of the entire supply chain process will be reduced, thereby improving the overall execution efficiency.

[0088] The execution cost of P1 is:

[0089] C1=c1+c2+c3+c4+c6+c11

[0090] The execution cost of P2 is:

[0091] C2=c1+c2+c3+c5+c7+c11

[0092] The execution cost of P3 is:

[0093] C3=c1+c8+c9+c10+c11

[0094] So the longest path is:

[0095] maximum(C1,C2,C3)

[0096] In this application, critical path analysis is used to identify the execution bottlenecks in the supply chain network, and the activities on the critical path are taken as the focus of optimization. The nodes and activities on the critical path directly affect the overall operating efficiency of the supply chain, so optimizing these nodes and activities is the core of improving supply chain efficiency.

[0097] To this end, the optimization method proposed in this application will give priority to reducing the resource consumption or execution time of each activity on the critical path, ensuring that bottleneck activities no longer become a constraint on the supply chain. By optimizing these key activities, the total execution consumption of the supply chain can be significantly reduced, making the operation of the entire network more efficient. At the same time, for activities on non-critical paths, based on their floating time, resources and arrangements are flexibly adjusted to ensure that while optimizing the critical path, the operating stability of other process nodes is not affected.

[0098] During the implementation process, Figure 4 and Figure 5 The supply chain network structure shown in the figure maps the flow chart into a weighted directed acyclic graph (DAG). This graph model helps to intuitively display the relationship between activities and their corresponding consumption, thereby identifying the real critical path and optimizing it. The optimized path will significantly improve the overall execution efficiency of the supply chain, reduce the total execution overhead, and ultimately provide higher controllability and efficiency for supply chain management.

[0099] Furthermore, in step S4, the second supply chain graph structure is input into the trained graph neural network model, the graph neural network model determines the probability of forming an edge between nodes that do not have a connection relationship in the second supply chain graph structure, and adds an edge between two nodes corresponding to probability values ​​exceeding a preset threshold; the formed third supply chain graph structure is used as the optimization result of the supply chain, wherein:

[0100] The graph neural network model adds an attention mechanism on the basis of the RGCN model, that is, an attention layer is added between the RGCN layer and the batch normalization layer. The attention module weights the node features corresponding to each node in the second supply chain graph structure output by the RGCN layer to obtain the weights of the importance of the relationships between each node in the second supply chain graph structure, and then weights the weights with the node features corresponding to each node in the second supply chain graph structure, and inputs the obtained results into the batch normalization layer; the decoder of the graph neural network model outputs the probability of forming edges between nodes in the second supply chain graph structure that have no connection relationship.

[0101] In this application, an attention mechanism is introduced on the basis of RGCN to weight the aggregation operation of nodes to improve the model's expressiveness on heterogeneous relationships. The model structure includes one-hot encoding the data and mapping it to the hidden layer dimension through a fully connected layer; then, two layers of RGCN are used to aggregate the node information, and an attention layer is added to achieve edge weighting, and a batch normalization layer (BN layer), a ReLU activation layer, and a DropOut layer are used to improve the robustness of the model and prevent overfitting. Link prediction is achieved through the decoder to complete the missing links.

[0102] In this application, supply chain optimization is performed based on the RGCNP model, such as Figure 6 The architecture shown is based on the RGCN idea. First, the dataset is one-hot encoded. After being processed by a fully connected layer, the node information is aggregated through the RGCN model, and the aggregated edges are weighted through the attention mechanism. Then, it passes through the BN, Relu and DropOut layers, passes through the intermediate state model twice, and passes through the final decoder to achieve the final link prediction effect and realize the purpose of supply chain link prediction.

[0103] This application constructs the pre-processed enterprise data into a graph structure, in which nodes represent key business nodes or participants in the enterprise, and edges represent the relationship between these nodes; then, the node features are encoded into vector form using graph embedding technology and input into the graph neural network (RGCN) model for training; through multi-layer information aggregation, the model learns the complex relationship between nodes and their neighborhoods. Then, the trained RGCN model is used for link prediction to calculate the probability of forming edges between unconnected nodes; based on the prediction results, the missing links are identified and completed to ensure the integrity and continuity of the supply chain. Finally, the completed supply chain structure is analyzed and optimized as a whole to improve the efficiency of enterprise resource allocation and the operational capabilities of the supply chain.

[0104] The present application provides a specific embodiment of a method for optimizing an enterprise supply chain based on graph relationships.

[0105] Use FB15k-237 and wn18 datasets and DataCo datasets

[0106] Dataset Description

[0107]

[0108] Indicator definition:

[0109] Use hits@n and accuracy evaluation indicators as indicators. Hit means the average proportion of triples ranked less than n in link prediction. Accuracy represents whether the prediction of the relationship type is consistent with the label. The formula is as follows:

[0110]

[0111] TP represents the total number of samples predicted correctly, TN corresponds to the total number of samples predicted correctly, and R is the total sample size. The higher the accuracy, the more accurate the model link prediction.

[0112] Comparison method selection

[0113] a) The GCN model effectively integrates graph structure features through convolution technology

[0114] b) GAT model: Combined with the Attention mechanism to process the edge weights during fusion

[0115] c) RGCN model: a heterogeneous network model improved based on GCN

[0116] The Hits@n (H@n) evaluation method is used to evaluate the model. In this method, for each triple sample, a large number of negative samples are generated by randomly replacing its head vertex or tail vertex. All samples are then scored and sorted in ascending order according to the score to evaluate the prediction performance of the model. At the same time, compared with the baseline model, the accuracy evaluation indicators under each data set are as follows:

[0117] Comparison of model accuracy indicators in various data sets

[0118]

[0119] The comparison results are as follows Figure 7 shown.

[0120] Through experimental verification on the DataCo dataset and the traditional datasets FB15k-237 and WN18, the research results show that the R-GCN model with enhanced supplementary attention mechanism performs better than the traditional R-GCN in predicting missing links and achieves higher link prediction accuracy. In addition, giving different attention weights to different types of relationships in heterogeneous graphs and taking them into comprehensive consideration is beneficial to improving the overall accuracy and reliability of link prediction.

[0121] Specifically, experiments conducted on multiple datasets show that the RGCNP model performs best in each test dataset compared to the baseline model. On the FB15k-237, WN18, and DataCo datasets, the performance of the RGCNP model is 18.35%, 10.22%, and 25.79% higher than the three baseline models of RGCN, GCN, and GAT, respectively. The above findings fully demonstrate the effectiveness and superiority of the RGCNP model in handling complex link prediction tasks.

[0122] like Figure 8As shown, the present application provides an enterprise supply chain optimization device based on graph relationship, the device comprising:

[0123] Initialization module: configured to obtain enterprise production data, clean and standardize the production data, and extract enterprise supply chain data from it; construct a basic graph structure based on the enterprise supply chain data, the graph structure G = {V, E}, wherein suppliers are taken as nodes, supply relationships are taken as edges, V is a set of nodes, and E is a set of edges; the supply relationship is determined by activities between nodes, and the activities include transportation activities, production activities, and processing activities;

[0124] Simplification module: configured to cluster the nodes in the node set, so that the nodes that satisfy the supply relationship similarity exceeding the first threshold and the supply product similarity exceeding the second threshold are classified into the same category, and the nodes that meet the key attributes of the supply chain in the same category are simplified into representative nodes, and the second supply chain graph structure is constructed based on the simplified nodes and the new supply relationships between the simplified nodes;

[0125] A critical path module: configured to construct an objective function, and determine a critical path from the second supply chain graph structure based on the objective function, wherein each node in the critical path is a critical node, and the critical node is used as a node to be optimized in the supply chain;

[0126] Optimization module: configured to input the second supply chain graph structure into a trained graph neural network model, the graph neural network model determines the probability of forming an edge between nodes that do not have a connection relationship in the second supply chain graph structure, and adds an edge between two nodes corresponding to probability values ​​that exceed a preset threshold; and uses the formed third supply chain graph structure as the optimization result of the supply chain.

[0127] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application is described in detail with reference to the above embodiments, a person skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some or all of the technical features can be replaced by equivalents, and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A graph-based enterprise supply chain optimization method, characterized in that: The method comprises the following steps: Step S1: Acquire enterprise production data, clean and standardize the production data, and extract enterprise supply chain data from it; construct a basic graph structure based on the enterprise supply chain data, the graph structure G = {V, E}, wherein suppliers are taken as nodes, supply relationships are taken as edges, V is a set of nodes, and E is a set of edges; the supply relationship is determined by activities between nodes, and the activities include transportation activities, production activities, and processing activities; Step S2: Cluster the nodes in the node set so that the nodes that satisfy the supply relationship similarity exceeding the first threshold and the supply product similarity exceeding the second threshold are classified into the same category, simplify the nodes that meet the key attributes of the supply chain in the same category into representative nodes, and construct a second supply chain graph structure based on the simplified nodes and the new supply relationships between the simplified nodes; Step S3: constructing an objective function, and determining a critical path from the second supply chain graph structure based on the objective function, wherein each node in the critical path is a critical node, and the critical node is used as a node to be optimized in the supply chain; Step S4: Input the second supply chain graph structure into the trained graph neural network model, the graph neural network model determines the probability of forming an edge between nodes that have no connection relationship in the second supply chain graph structure, and adds an edge between two nodes corresponding to the probability value exceeding a preset threshold; the formed third supply chain graph structure is used as the optimization result of the supply chain.

2. The method according to claim 1, characterized in that In step S1, the edges in the basic graph structure have weights, and the weights are determined based on the activity types between nodes and the distances between nodes.

3. The method according to claim 1, characterized in that The step S2 simplifies the nodes in the same category that meet the key attributes of the supply chain into representative nodes, based on the simplified nodes and the new supply relationships between the simplified nodes, including: The attributes of each node in the class are merged, and the merging method is one of taking the average value, weighted average, and sum of the attribute values; the node in the class whose attribute value is the smallest from the merged attribute value is taken as the node that meets the key attributes of the supply chain, that is, the representative node, and the representative node represents each node of the class; the merged attribute value is assigned to the representative node as the node feature of the representative node, and the representative node inherits the supply relationship between each node it represents and other nodes outside the class as a new supply relationship; based on the weights corresponding to the supply relationships between each node inherited by the representative node and other nodes outside the class, a new weight is established for the new supply relationship.

4. The method according to claim 1, characterized in that The step S3: constructing an objective function, and determining a critical path from the second supply chain graph structure based on the objective function, wherein each node in the critical path is a critical node, and the critical node is used as a node to be optimized in the supply chain, wherein: The objective function is: Act={act1,act2,…,act num } Where Z is the total execution cost of the entire supply chain network, AC is the minimum node subset that meets the connectivity and uniqueness conditions, and a i is the node belonging to AC in the critical path, n is the total number of nodes on the critical path, c i For node a i The corresponding execution cost includes the total cost of its execution and the corresponding activities. Act is the set of activities, act1, act2,…, act num All are activities; Based on the objective function, a critical path method (CPM) is used to determine a critical path from the second supply chain graph structure; the critical path is a path with the greatest execution consumption from the start node to the end node of the supply chain.

5. The method according to claim 1, characterized in that The step S4: inputting the second supply chain graph structure into the trained graph neural network model, the graph neural network model determines the probability of forming an edge between nodes that do not have a connection relationship in the second supply chain graph structure, and adds an edge between two nodes corresponding to probability values ​​exceeding a preset threshold; The third supply chain graph structure is taken as the optimization result of the supply chain, where: The graph neural network model adds an attention mechanism on the basis of the RGCN model, that is, an attention module is added between the RGCN layer and the batch normalization layer. The attention module weights the node features corresponding to each node in the second supply chain graph structure output by the RGCN layer to obtain the weights of the importance of the relationships between each node in the second supply chain graph structure, and then weights the weights with the node features corresponding to each node in the second supply chain graph structure, and inputs the obtained results into the batch normalization layer; the decoder of the graph neural network model outputs the probability of forming edges between nodes in the second supply chain graph structure that have no connection relationship.

6. An enterprise supply chain optimization device based on graph relations, characterized in that: The device comprises: Initialization module: configured to obtain enterprise production data, clean and standardize the production data, and extract enterprise supply chain data from it; construct a basic graph structure based on the enterprise supply chain data, the graph structure G = {V, E}, wherein suppliers are taken as nodes, supply relationships are taken as edges, V is a set of nodes, and E is a set of edges; the supply relationship is determined by activities between nodes, and the activities include transportation activities, production activities, and processing activities; Simplification module: configured to cluster the nodes in the node set, so that the nodes that satisfy the supply relationship similarity exceeding the first threshold and the supply product similarity exceeding the second threshold are classified into the same category, and the nodes that meet the key attributes of the supply chain in the same category are simplified into representative nodes, and the second supply chain graph structure is constructed based on the simplified nodes and the new supply relationships between the simplified nodes; A critical path module: configured to construct an objective function, and determine a critical path from the second supply chain graph structure based on the objective function, wherein each node in the critical path is a critical node, and the critical node is used as a node to be optimized in the supply chain; Optimization module: configured to input the second supply chain graph structure into a trained graph neural network model, the graph neural network model determines the probability of forming an edge between nodes that do not have a connection relationship in the second supply chain graph structure, and adds an edge between two nodes corresponding to probability values ​​that exceed a preset threshold; and uses the formed third supply chain graph structure as the optimization result of the supply chain.

7. A computer-readable storage medium, wherein a plurality of instructions are stored in the storage medium; the plurality of instructions are used for a processor to load and execute the method as claimed in any one of claims 1 to 5.

8. An electronic device, characterized in that: The electronic device comprises: A processor, which is used to execute multiple instructions; A memory for storing a plurality of instructions; The plurality of instructions are used to be stored in the memory and loaded and executed by the processor according to any one of claims 1 to 5.

Images