A supply chain data sharing method and storage medium with privacy protection function

Abstract

The present invention discloses a supply chain data sharing method and storage medium with privacy protection function, which relates to the field of supply chain sharing. The sharing method comprises: obtaining entity types in the current supply chain; inputting the entity types in the current supply chain into a trained supply chain entity sharing model; the supply chain entity sharing model generates downstream entity types corresponding to the entity types in the current supply chain according to dependencies, and sorts the downstream entity types based on minimizing the sorting loss between predicted capability values ​​and true capability values; the present invention inputs the entity types in the current supply chain obtained, and a graph structure model can infer downstream entity types related to upstream entity types according to the dependencies between entity types, and predict and sort the capability values ​​of downstream entity types based on an optimization method that minimizes sorting loss, and the downstream entity type with the smallest sorting loss is regarded as the optimal shared entity and is recommended first.

Description

Technical Field

[0001] The present invention relates to a supply chain data sharing method and storage medium with a privacy protection function, and in particular to a supply chain data sharing method. Background Art

[0002] Supply chain data sharing refers to the ability of entities in different links of the supply chain (such as suppliers, manufacturers, distributors, etc.) to work together through data exchange and information sharing to optimize the overall operation of the supply chain. In traditional supply chain data sharing, each entity type will share relevant data such as inventory, production capacity, and transportation status so that upstream and downstream entities can make corresponding decisions based on the shared data.

[0003] For example, the Chinese patent document with patent publication number CN115840787A discloses a supply chain data sharing method, device, equipment and medium based on blockchain, which can solve the problems of low security, untraceability and strong coupling between multiple systems in the supply chain during data sharing. However, the above supply chain data sharing method lacks an effective recommendation and sorting optimization mechanism for upstream and downstream entities, making it difficult to accurately match the dependencies and capability values ​​between upstream and downstream entities, resulting in low supply chain efficiency. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a supply chain data sharing method, which trains a graph structure model to predict the capability value of entity types and sorts them based on the predicted capability value, thereby solving the technical problems raised in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0006] A supply chain data sharing method, comprising:

[0007] Get the entity type in the current supply chain;

[0008] Input entity types in the current supply chain into the trained supply chain entity sharing model;

[0009] The supply chain entity sharing model generates downstream entity types corresponding to entity types in the current supply chain according to dependency relationships, and sorts the downstream entity types based on minimizing the sorting loss between predicted capability values ​​and actual capability values.

[0010] In some embodiments, the training step of the supply chain entity sharing model includes:

[0011] S1. Build the graph structure of the supply chain;

[0012] S2. Determine a training set of the supply chain according to the graph structure;

[0013] S3, using the training set of the supply chain to train the graph structure model;

[0014] S4. When the graph structure model reaches the convergence condition, it is used as the supply chain entity sharing model; wherein the convergence condition is: when the change in the sorting loss is less than a preset change threshold, or when the number of iterations reaches a preset maximum number; or when the gradient change is less than a preset gradient change threshold.

[0015] In some of the embodiments, constructing a graph structure of a supply chain includes:

[0016] S1-1, obtain different entity types in the supply chain and the real capability values ​​corresponding to the entity types;

[0017] S1-2, defining different entity types and corresponding real capability values ​​as nodes of the graph structure;

[0018] S1-3, according to different entity types, determine the dependency relationship between different entity types;

[0019] S1-4, defining the dependency relationship between different entity types as the edge of the graph structure and constructing it into an adjacency matrix; the adjacency matrix represents the dependency relationship between different entity types;

[0020] In some of the embodiments, determining a training set of a supply chain according to the graph structure includes:

[0021] S2-1, creating a mapping relationship between the nodes in the graph structure and the training set; wherein the mapping relationship is: the entity type in the node is associated as the input feature, and the real capability value in the node is associated as the target label;

[0022] S2-3, combining the entity type and the corresponding real capability value into a binary array and constructing it as a training sample;

[0023] S2-4, obtaining multiple training samples and constructing a training set for the supply chain.

[0024] In some of the embodiments, using the training set of the supply chain to train the graph structure model includes:

[0025] S3-1, inputting the training set and the adjacency matrix into a graph structure model; wherein the entity type is used as a characteristic variable of the graph structure model, the true capability value is used as a target variable of the graph structure model, and the adjacency matrix represents the dependency relationship between different entity types;

[0026] S3-2, initialize the entity type of each node to obtain the embedding vector of the node;

[0027] S3-3, each current node interacts with adjacent nodes through dependencies to obtain an updated embedding vector;

[0028] The updated expression of the updated embedding vector is:

[0029]

[0030] in, represents an updated embedding vector, Nodes representing graph structures, Neural layers representing graph structure models; represents a nonlinear activation function, express The weight matrix of the layer is used to transform The embedding vector of the layer is transformed into the input vector of the next layer; Representation Node The set of adjacent nodes of The weight matrix representing the aggregation of neighboring node embedding vectors;

[0031] S3-4, obtain the prediction ability value of each node according to the updated embedding vector;

[0032] The expression of the prediction ability value is:

[0033]

[0034] in, Representation Node The predictive ability value of Representation Node In the last layer of the graph structure model, the embedding vector is represents the weight matrix of the neural layer, represents the bias term of the neural layer;

[0035] S3-5, calculate the ranking loss of each node according to the prediction ability value of each node;

[0036] The loss function of the sorting loss is:

[0037]

[0038] in, represents the sorting loss, represents the prediction capability value of node u, represents the prediction ability loss between any node u and node v, represents the real capability value of node u, Represents the true capacity loss between any node u and node v;

[0039] S3-6, through the gradient descent algorithm, minimize the sorting loss and iteratively update the weight matrix and bias terms of the model neural layer.

[0040] In some of the embodiments, the weight matrix and bias term of the neural layer of the model are iteratively updated by minimizing the ranking loss through a gradient descent algorithm, including:

[0041] S3-6-1, calculate the partial derivative of the sorting loss with respect to the weight matrix and the bias term to obtain the weight matrix gradient and the bias term gradient;

[0042] The partial derivative expression of the weight matrix gradient is:

[0043]

[0044] in, represents the weight matrix gradient, and They represent the partial derivatives of the ranking loss with respect to the prediction ability values ​​of node u and node v respectively;

[0045] The partial derivative expression of the bias term gradient is:

[0046]

[0047] in, represents the bias gradient;

[0048] S3-6-2, iteratively update the next weight matrix and the next bias term respectively by calculating the gradient of the weight matrix after partial derivative and the gradient of the bias term after partial derivative;

[0049] The update expression of the weight matrix for the next time is:

[0050]

[0051] in, represents the next weight matrix, represents the current weight matrix, Represents the learning rate, which represents the step size of the control update;

[0052] The update expression of the next bias term is:

[0053]

[0054] in, represents the next bias term, Represents the current bias term.

[0055] In some embodiments, the supply chain entity sharing model generates downstream entity types corresponding to entity types in the current supply chain according to dependency relationships, and sorts the downstream entity types based on minimizing the sorting loss between predicted capability values ​​and actual capability values, including:

[0056] C1, the set of downstream entity types output by the supply chain entity sharing model based on the input entity types and their dependencies in the current supply chain;

[0057] The expression of the downstream entity type set is:

[0058]

[0059] in, Indicates the entity type in the current supply chain, Represents the set of prediction capability values ​​of the current output downstream entity type, N represents the number of entity types; Graph represents the dependency relationships learned by the model during the training phase; Represents the model for predicting capability values ​​under the use phase;

[0060] C2, sorts the output downstream entity types according to the prediction ability value of the downstream entity types, and generates a sorted set of downstream entity types that match the upstream entity types.

[0061] The expression for the sort is:

[0062]

[0063] in, Represents the downstream entity type that matches the upstream entity type, Rank represents the sorting function of the entity type, and sort represents the sorting function of the prediction ability value.

[0064] The present invention uses a supply chain entity sharing model based on graph structure deep learning to input the entity types in the current supply chain. The graph structure model can infer the downstream entity types related to the upstream entity types based on the dependency relationship between the entity types, and predict and sort the capability values ​​of the downstream entity types based on the optimization method of minimizing the sorting loss. This ensures that the sorting results are highly consistent with the actual capability values, and the downstream entity types with the smallest sorting loss are regarded as the optimal shared entities and are recommended first.

[0065] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the supply chain data sharing method described in the first aspect is implemented.

[0066] Compared with the prior art, the beneficial effects of a computer-readable storage medium of the present invention are the same as the beneficial effects of the above-mentioned supply chain data sharing method, so they will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A supply chain data sharing method of the present invention;

[0068] Figure 2 A training flow chart of the supply chain entity sharing model in a supply chain data sharing method of the present invention;

[0069] Figure 3 A sorting flow chart of a supply chain entity sharing model in a supply chain data sharing method of the present invention; DETAILED DESCRIPTION

[0070] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0071] Example 1: Reference Figure 1 , this embodiment discloses a supply chain data sharing method, the sharing method is as follows:

[0072] Get the entity type in the current supply chain;

[0073] Input entity types in the current supply chain into the trained supply chain entity sharing model;

[0074] The supply chain entity sharing model generates downstream entity types corresponding to entity types in the current supply chain according to dependency relationships, and sorts the downstream entity types based on minimizing the sorting loss between predicted capability values ​​and actual capability values.

[0075] This sharing method obtains the entity types in the current supply chain and inputs them into the trained supply chain entity sharing model to generate a corresponding downstream entity type sorting set. The core idea of ​​this method is to minimize the sorting loss between the predicted capability value and the actual capability value to characterize the actual sorting of the predicted capability value in accordance with the actual capability value. The sharing method of this embodiment, based on the dependency sorting of upstream and downstream entities, can identify the optimal downstream entity type in the supply chain, and ultimately achieve the recommendation of the optimal entity type in the supply chain, ensuring the collaborative sharing of the supply chain.

[0076] Furthermore, in the sharing method, in order to ensure the data security of entity types during data sharing and avoid information leakage of entity types, an edge server authentication mechanism can be introduced. The edge server authentication ensures that each shared supply chain entity type data is authenticated and verified in real time, preventing unauthorized or illegal data access requests, and blocking the generation of illegal information flows from the source. When the downstream entity type is written into the shared search database, the edge server (supply chain data sharing platform built by supply chain users) will verify the capability value and ranking data of each entity type to ensure the integrity and legality of the data. In the access stage of the shared search database, attribute encryption technology is used to ensure secure access to data. Only users with legal authority authentication can access downstream entity types and other related data. This ensures that the data in the shared search database is only open to legal users, further enhancing the privacy protection of supply chain data.

[0077] Embodiment 2: The technical solution of Embodiment 2 is different from that of Embodiment 1 in that the training steps of the supply chain entity sharing model in Embodiment 1 are disclosed; Figure 2 , Figure 2 The training steps of the supply chain entity sharing model are shown, and the training steps include:

[0078] S1. Build the graph structure of the supply chain;

[0079] S2. Determine a training set of the supply chain according to the graph structure;

[0080] S3, using the training set of the supply chain to train the graph structure model;

[0081] S4. When the graph structure model reaches the convergence condition, it is used as the supply chain entity sharing model; wherein the convergence condition is: when the change in the sorting loss is less than a preset change threshold, or when the number of iterations reaches a preset maximum number; or when the gradient change is less than a preset gradient change threshold.

[0082] Specifically, by building a graph structure of the supply chain, determining the training set and applying a deep learning model for training, we ensure that the model learns the dependencies between entity types and the loss of their capability values. When the training is completed and convergence conditions are reached, the model can predict the capability values ​​of different entities and perform sorting optimization, thus realizing the construction of a supply chain entity sharing model and improving the accuracy of supply chain data sharing.

[0083] Furthermore, step S1 specifically includes:

[0084] S1-1, obtain different entity types in the supply chain and the real capability values ​​corresponding to the entity types;

[0085] S1-2, defining different entity types and corresponding real capability values ​​as nodes of the graph structure; for example, suppliers and their quantified inventory levels, manufacturers and their quantified manufacturing capabilities, distributors and their quantified logistics speeds or qualitative in-transit status;

[0086] S1-3, according to different entity types, determine the dependency relationship between different entity types; such as the supply relationship between suppliers and manufacturers, and the distribution relationship between manufacturers and distributors;

[0087] S1-4, the dependency relationship between different entity types is defined as the edge of the graph structure and constructed as an adjacency matrix; the adjacency matrix represents the dependency relationship between different entity types; by storing the dependency relationship of the edge of the graph structure as an adjacency matrix that is easy for the model to receive, the matrixed dependency relationship is used to better adapt to the model input, wherein the adjacency matrix is ​​used to represent the connection relationship between nodes, for example, 0 means there is no edge, that is, there is no dependency relationship between the two entities, and 1 means there is an edge, and there is a dependency relationship between the two entities, such as a supplier supplies manufacturing components to a manufacturer.

[0088] The above steps define different entity types and their true capability values ​​in the supply chain as nodes of the graph structure. At the same time, the dependency relationship between different entity types is clearly defined as the edges in the graph structure, and the dependency relationship is expressed as an adjacency matrix that is easy for the model to identify, thereby ensuring that the dependency relationship between entity types in the supply chain is accurately represented.

[0089] Furthermore, step S2 specifically includes:

[0090] S2-1, creating a mapping relationship between the nodes in the graph structure and the training set; wherein the mapping relationship is: the entity type in the node is associated as the input feature, and the real capability value in the node is associated as the target label;

[0091] S2-3, combining the entity type and the corresponding real capability value into a binary array and constructing it as a training sample;

[0092] S2-4, obtain multiple training samples and construct a training set for the supply chain. Of course, the training set can contain multiple nodes of the same type and different types; by creating a mapping relationship between the nodes in the graph structure and the training samples in the training set, the entity type and the corresponding real capability value are combined into a training sample to form a complete training set.

[0093] Furthermore, step S3 specifically includes:

[0094] S3-1, inputting the training set and the adjacency matrix into a graph structure model; wherein the entity type is used as a characteristic variable of the graph structure model, the true capability value is used as a target variable of the graph structure model, and the adjacency matrix represents the dependency relationship between different entity types;

[0095] S3-2, initialize the entity type of each node to obtain the embedding vector of the node;

[0096] S3-3, each current node interacts with adjacent nodes through dependencies to obtain an updated embedding vector;

[0097] The updated expression of the updated embedding vector is:

[0098]

[0099] in, represents an updated embedding vector, Nodes representing graph structures, Neural layers representing graph structure models; represents a nonlinear activation function, express The weight matrix of the layer is used to transform The embedding vector of the layer is transformed into the input vector of the next layer; Representation Node The set of adjacent nodes of represents the weight matrix that aggregates the embedding vectors of neighboring nodes; specifically, represents the weight matrix used to aggregate the embedding vectors of adjacent nodes in the ground layer of the model; this weight matrix can linearly transform the embedding vectors of adjacent nodes and then aggregate them with the embedding vector of the current node, thereby updating the embedding vector of the current node; more specifically, In each neural layer, the embedding vectors received from adjacent nodes are linearly transformed and passed to the current node, which then forms part of the input of the next neural layer. This embedding vector update process, through vector transmission and aggregation, enables the embedding vector of each node to simultaneously represent the characteristics of itself and adjacent nodes.

[0100] S3-4, obtain the prediction ability value of each node according to the updated embedding vector;

[0101] The expression of the prediction ability value is:

[0102]

[0103] in, Representation Node The predictive ability value of Representation Node In the last layer of the graph structure model, the embedding vector is represents the weight matrix of the neural layer, represents the bias term of the neural layer;

[0104] S3-5, calculate the ranking loss of each node according to the prediction ability value of each node;

[0105] The loss function of the sorting loss is:

[0106]

[0107] in, represents the sorting loss, represents the prediction capability value of node u, represents the prediction ability loss between any node u and node v, represents the real capability value of node u, Represents the true capacity loss between any node u and node v;

[0108] Specifically, during the model training process, the downstream entity with the highest score is selected based on the predicted capability value of the downstream entity. Therefore, when predicting the capability value, the predicted capability value must not only be accurate, but also be correctly sorted so that the entity type with the highest predicted capability value is given priority. In the supply chain, the dependencies between different entity types determine the upstream and downstream relationships between them. In order to ensure the selection of the appropriate downstream entity, the model needs to reflect the dependencies between entities in the capability score sorting. Therefore, the sorting loss can be used as the content of the graph structure model optimization.

[0109] It should be noted that when using ranking loss, there is no need to add an additional sorting step in the middle. The ranking loss itself has an embedded learning mechanism for node sorting. During the training process, the model automatically adjusts the predicted ability scores of the nodes by minimizing the ranking loss so that the ranking of these scores is consistent. Sorting loss is essentially an optimization based on the relative order between the predicted scores and the true scores of the nodes during model training. Through this mechanism, the model learns how to sort the ability scores of the nodes, and at the final output, the model sorts according to the predicted ability scores, thereby generating a score-based sorted set of downstream entity types. Therefore, when using ranking loss, the model already implies the sorting process, and only needs to sort the predicted scores at output.

[0110] S3-6, through the gradient descent algorithm, minimize the sorting loss and iteratively update the weight matrix and bias terms of the model neural layer.

[0111] In the above steps, the model updates the embedding vector of each node based on the mapping relationship between entity type and true capability value, the edge in the dependency graph structure, and updates it based on the information of neighboring nodes, and finally calculates the predicted capability value of each node. Therefore, the model predicts the capability value by continuously updating the embedding vector and dependency relationship of the node, providing data support for sorting in the supply chain.

[0112] Furthermore, step S3-6 also includes:

[0113] S3-6-1, calculate the partial derivative of the sorting loss with respect to the weight matrix and the bias term to obtain the weight matrix gradient and the bias term gradient;

[0114] The partial derivative expression of the weight matrix gradient is:

[0115]

[0116] in, represents the weight matrix gradient, and They represent the partial derivatives of the ranking loss with respect to the prediction ability values ​​of node u and node v respectively;

[0117] The partial derivative expression of the bias term gradient is:

[0118]

[0119] in, represents the bias gradient;

[0120] Specifically, the weight matrix It is used to embed the vector The matrix for linear transformation. It determines how to convert the node embedding vector of the last layer into a prediction value. Since the prediction value is obtained by linearly transforming the embedding vector, each column of the weight matrix is ​​related to a different dimension of the embedding vector; therefore, when taking the partial derivative of the weight matrix for the ranking loss, it is necessary to consider the contribution of each embedding vector to the prediction score. Specifically, the gradient depends not only on the partial derivative of the prediction score, but also on the corresponding node embedding vector. and .

[0121] S3-6-2, iteratively update the next weight matrix and the next bias term respectively by calculating the gradient of the weight matrix after partial derivative and the gradient of the bias term after partial derivative;

[0122] The update expression of the weight matrix for the next time is:

[0123]

[0124] in, represents the next weight matrix, represents the current weight matrix, Represents the learning rate, which represents the step size of the control update;

[0125] The update expression of the next bias term is:

[0126]

[0127] in, represents the next bias term, Represents the current bias term.

[0128] Based on the gradient descent algorithm, the model can minimize the sorting loss function and continuously update the weight matrix and bias terms of the neural layer. This ensures that the model is gradually optimized during the iteration process, and the gap between the predicted capability value and the actual capability value of each node is gradually narrowed, so that the model can accurately predict and sort the nodes based on the sorting loss, ensuring that the capability values ​​of different entity types in the supply chain are reasonably sorted, and then sharing the optimal downstream entity type.

[0129] refer to Figure 3 , Figure 3 The steps of sorting downstream entity types by the supply chain entity sharing model are shown; specifically, the supply chain entity sharing model generates downstream entity types corresponding to entity types in the current supply chain according to dependency relationships, and sorts downstream entity types based on minimizing the sorting loss between predicted capability values ​​and actual capability values, including:

[0130] C1, the set of downstream entity types output by the supply chain entity sharing model based on the input entity types and their dependencies in the current supply chain;

[0131] The expression of the downstream entity type set is:

[0132]

[0133] in, Indicates the entity type in the current supply chain, Represents the set of prediction capability values ​​of the current output downstream entity type, N represents the number of entity types; Graph represents the dependency relationships learned by the model during the training phase; It represents a model that calculates the prediction capability value of downstream entity types according to the input upstream entity types and their dependency graph structure in the use phase;

[0134] Specifically, the upstream entity types in the current supply chain are input into the trained model. The trained model infers the downstream entity types related to the upstream entity types based on the dependencies between entity types in the graph structure learned in the training phase, and generates predictive capability values ​​for these downstream entity types.

[0135] C2, sorts the output downstream entity types according to the prediction ability value of the downstream entity types, and generates a sorted set of downstream entity types that match the upstream entity types.

[0136] The expression for the sort is:

[0137]

[0138] in, Represents the downstream entity type that matches the upstream entity type, Rank represents the sorting function of the entity type, and sort represents the sorting function of the prediction ability value; the sorting expression represents sorting all downstream entity types according to the set of prediction ability values ​​of the downstream entity types to obtain an ordered list of entity types.

[0139] In the present invention, the first entity of the sorted downstream entity type can be regarded as the optimal shared entity. For example, the current manufacturer can use the first distributor as the optimal distributor to match. Specifically, the sorting loss is used to optimize the predictive capability value of the graph structure model, so that the sorting of the predictive capability values ​​of different entities is as close to the actual situation as possible; entities with high capability values ​​are ranked in front, and entities with low capability values ​​are ranked in the back, forming the downstream entity type sorting set. Specifically for the downstream entity type sorting set, the minimum sorting loss means that the sorting set is closer to the actual sorting. If the sorting loss is the smallest, it means that the sorting of the predicted capability value output by the model is highly consistent with the actual capability value sorting. In other words, the model ensures that entities with high capability values ​​are shared first by minimizing the sorting loss, so the optimal downstream entity type will be at the front in the sorting.

[0140] In summary, the present invention uses a supply chain entity sharing model based on graph structure deep learning to input the entity types in the current supply chain. The graph structure model can infer the downstream entity types related to the upstream entity types based on the dependency relationship between the entity types, and predict and sort the capability values ​​of the downstream entity types based on the optimization method of minimizing the sorting loss. This ensures that the sorting results are highly consistent with the actual capability values, and the downstream entity types with the smallest sorting loss are regarded as the optimal shared entities and are recommended first.

[0141] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) means.

[0142] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., DVD ), or semiconductor media. The semiconductor media may be a solid state drive.

[0143] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A supply chain data sharing method, characterized in that: include: Get the entity type in the current supply chain; Input entity types in the current supply chain into the trained supply chain entity sharing model; The supply chain entity sharing model generates downstream entity types corresponding to entity types in the current supply chain according to dependency relationships, and sorts the downstream entity types based on minimizing the sorting loss between the predicted capability value and the actual capability value; The training steps of the supply chain entity sharing model include: S1. Build the graph structure of the supply chain; S2. Determine a training set of the supply chain according to the graph structure; S3, using the training set of the supply chain to train the graph structure model; S4. When the graph structure model reaches the convergence condition, it is used as the supply chain entity sharing model; wherein the convergence condition is: when the change in the sorting loss is less than a preset change threshold, or when the number of iterations reaches a preset maximum number; or when the gradient change is less than a preset gradient change threshold; Constructing the graph structure of the supply chain, including: S1-1, obtain different entity types in the supply chain and the real capability values ​​corresponding to the entity types; S1-2, defining different entity types and corresponding real capability values ​​as nodes of the graph structure; S1-3, according to different entity types, determine the dependency relationship between different entity types; S1-4, defining the dependency relationship between different entity types as the edge of the graph structure and constructing it into an adjacency matrix; the adjacency matrix represents the dependency relationship between different entity types; According to the graph structure, a training set of the supply chain is determined, including: S2-1, creating a mapping relationship between the nodes in the graph structure and the training set; wherein the mapping relationship is: the entity type in the node is associated as the input feature, and the real capability value in the node is associated as the target label; S2-3, combining the entity type and the corresponding real capability value into a binary array and constructing it as a training sample; S2-4, obtaining a plurality of training samples and constructing a training set of the supply chain; The training set of the supply chain is used to train the graph structure model, including: S3-1, inputting the training set and the adjacency matrix into a graph structure model; wherein the entity type is used as a characteristic variable of the graph structure model, the true capability value is used as a target variable of the graph structure model, and the adjacency matrix represents the dependency relationship between different entity types; S3-2, initialize the entity type of each node to obtain the embedding vector of the node; S3-3, each current node interacts with adjacent nodes through dependencies to obtain an updated embedding vector; The updated expression of the updated embedding vector is: ； in, represents an updated embedding vector, Nodes representing graph structures, Neural layers representing graph structure models; represents a nonlinear activation function, express The weight matrix of the layer is used to transform The embedding vector of the layer is transformed into the input vector of the next layer; Representation Node The set of adjacent nodes of The weight matrix representing the aggregation of neighboring node embedding vectors; S3-4, obtain the prediction ability value of each node according to the updated embedding vector; The expression of the prediction ability value is: ； in, Representation Node The predictive ability value of Representation Node In the last layer of the graph structure model, the embedding vector is represents the weight matrix of the neural layer, represents the bias term of the neural layer; S3-5, calculate the ranking loss of each node according to the prediction ability value of each node; The loss function of the sorting loss is: ； in, represents the sorting loss, represents the prediction capability value of node u, represents the prediction ability loss between any node u and node v, represents the real capability value of node u, Represents the true capacity loss between any node u and node v; S3-6, through the gradient descent algorithm, minimize the sorting loss and iteratively update the weight matrix and bias term of the model neural layer; Through the gradient descent algorithm, the sorting loss is minimized and the weight matrix and bias items of the model neural layer are iteratively updated, including: S3-6-1, calculate the partial derivative of the sorting loss with respect to the weight matrix and the bias term to obtain the weight matrix gradient and the bias term gradient; The partial derivative expression of the weight matrix gradient is: ； in, represents the weight matrix gradient, and They represent the partial derivatives of the ranking loss with respect to the prediction ability values ​​of node u and node v respectively; The partial derivative expression of the bias term gradient is: ； in, represents the bias gradient; S3-6-2, iteratively update the next weight matrix and the next bias term respectively by calculating the gradient of the weight matrix after partial derivative and the gradient of the bias term after partial derivative; The update expression of the weight matrix for the next time is: ； in, represents the next weight matrix, represents the current weight matrix, Represents the learning rate, which represents the step size of the control update; The update expression of the next bias term is: ； in, represents the next bias term, Represents the current bias item; The supply chain entity sharing model generates downstream entity types corresponding to entity types in the current supply chain according to dependency relationships, and sorts the downstream entity types based on minimizing the sorting loss between predicted capability values ​​and actual capability values, including: C1, the set of downstream entity types output by the supply chain entity sharing model based on the input entity types and their dependencies in the current supply chain; The expression of the downstream entity type set is: ； in, Indicates the entity type in the current supply chain, Represents the set of prediction capability values ​​of the current output downstream entity type, N represents the number of entity types; Graph represents the dependency relationships learned by the model during the training phase; Represents the model for predicting capability values ​​under the use phase; C2, sort the output downstream entity types according to the prediction ability value of the downstream entity types, and generate a sorted set of downstream entity types that match the upstream entity types; The expression for the sort is: ； in, Represents the downstream entity type that matches the upstream entity type, Rank represents the sorting function of the entity type, and sort represents the sorting function of the prediction ability value.

2. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the supply chain data sharing method described in claim 1 is implemented.

Images