Drug supply chain intelligent prediction method and system
By constructing a knowledge graph of the pharmaceutical supply chain and performing spectral decomposition, and combining disturbance intensity and structure density, a price fluctuation prediction model was fitted. This solved the problem of identifying the impact of pharmaceutical patent expiration events on the supply chain and predicting prices, achieving efficient prediction results.
Patent Information
- Application Number
- CN202511059592.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-30
- Publication Date
- 2025-11-14
AI Technical Summary
Existing technologies lack effective means to quantify and predict the impact of drug patent expiration events on the supply chain structure and the resulting price fluctuations. Traditional methods are unable to accurately characterize the multiple impacts of disturbance propagation paths and price change factors.
A knowledge graph of the pharmaceutical supply chain is constructed, and the supply chain entities are modeled as undirected graph structures. A feature vector space is formed through spectral decomposition. A perturbation vector is constructed and projected onto the feature vector space. By combining the perturbation strength and structure density, a price fluctuation prediction model is fitted, and prediction is made using linear regression or machine learning.
It enables accurate identification of the impact of drug patent expiration events and effective prediction of price fluctuations, improves the dynamic response and generalization capabilities of forecasts, and provides decision support.
Smart Images
Figure CN120952853A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of pharmaceutical supply chain, and more specifically, to a method and system for intelligent forecasting of pharmaceutical supply chain. Background Technology
[0002] With the continuous expansion of the global pharmaceutical market and the increasing complexity of the competitive landscape, the pharmaceutical supply chain plays a crucial role in ensuring the production, distribution, and availability of drugs at the end consumer. The pharmaceutical supply chain typically consists of multiple links, including raw material suppliers, manufacturers, and distributors, and its operational characteristics are characterized by high dependence, close connectivity, and dynamic evolution. As an important institutional arrangement in the pharmaceutical industry, the patent system, while protecting innovation, also determines the market monopoly position of certain drugs during specific periods. Therefore, once a drug patent expires, its impact on the entire supply chain structure and pricing system is sudden and far-reaching.
[0003] However, current pharmaceutical supply chain management lacks an effective means to quantify and predict the structural impact of patent expiration events on the supply chain and the resulting price fluctuations. On the one hand, the complex supply chain structure and diverse relationships between nodes make it difficult to accurately characterize the propagation path of disturbances using traditional linear models. On the other hand, price changes are often influenced by multiple factors, including supply chain structure, market behavior, and competitive response, making forecasting extremely challenging. Therefore, how to model the pharmaceutical supply chain from a graph structure perspective and identify and quantitatively analyze structural changes based on an event-driven approach is a critical issue that urgently needs to be addressed. Summary of the Invention
[0004] The technical problem to be solved by the present invention is to provide a method and system for intelligent forecasting of the pharmaceutical supply chain, so as to solve the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention adopts the following technical solution:
[0006] A method for intelligent forecasting of the pharmaceutical supply chain includes the following steps:
[0007] Construct a knowledge graph of the pharmaceutical supply chain, modeling raw material suppliers, manufacturers, and distributors as nodes, and modeling the logistics paths or contractual dependencies between nodes as weighted edges, forming an undirected graph structure;
[0008] Based on the aforementioned drug supply chain knowledge graph, a normalized Laplace matrix is constructed, and its spectral decomposition is performed to form a feature vector space.
[0009] Upon receiving a drug patent expiration event, a perturbation vector is constructed and projected onto the feature vector space to extract the affected sub-graphs due to the patent expiration.
[0010] Based on the perturbation intensity of the perturbation vector on the entire graph in historical data, and the structural density of the affected subgraph, a price fluctuation prediction model is fitted, and the price fluctuation prediction model is used to predict the price of drugs after patent expiration.
[0011] In some embodiments, the steps of constructing the normalized Laplacian matrix include:
[0012] The pharmaceutical supply chain knowledge graph is represented as an undirected weighted graph G = (V, E), where V is the set of supply chain nodes, E is the set of edges, and the weight of the edges represents the transaction frequency, contractual dependence, or material flow intensity between nodes.
[0013] Construct the adjacency matrix A and degree matrix D of the drug supply chain knowledge graph, and calculate the normalized Laplacian matrix;
[0014] The normalized Laplacian matrix is decomposed into eigenvalues to form an eigenvector space.
[0015] In some embodiments, the perturbation vector h is constructed as follows:
[0016] Define the node directly affected by the patent expiration event as the disturbance source node v. i According to v i For index i in the aforementioned drug supply chain knowledge graph, construct a perturbation vector of length |V|. Its definition is:
[0017]
[0018] Where |V| represents the total number of nodes in the supply chain graph, and j is the number of h.
[0019] In some embodiments, the step of identifying the affected subgraph includes:
[0020] Project the perturbation vector h onto the feature vector basis of the pharmaceutical supply chain knowledge graph, and calculate the response value of each node to the perturbation vector h. The response value is the inner product between the node's feature vector and the perturbation vector h.
[0021] Set a response threshold τ, and form a node set V consisting of all nodes whose response values are greater than τ. s ;
[0022] Extracting V from the pharmaceutical supply chain knowledge graph s The corresponding induced subgraph, wherein the nodes of the induced subgraph are V s The edge is the set of edges between nodes in the original graph, which is the affected subgraph that is directly or indirectly affected by the disturbance.
[0023] In some embodiments, the disturbance intensity is defined as the sum of the absolute values of the response values of all nodes.
[0024] In some embodiments, the structural density is the ratio of the actual number of edges to the maximum possible number of edges in the affected subgraph, calculated using the following formula:
[0025] ρ=|E s | / (|V s |*(|V s |-1) / 2);
[0026] Where |E s | represents the number of edges in the subgraph, |V s | represents the number of nodes in the subgraph, |V s |*(|V s |-1) / 2 represents the maximum possible number of edges in the affected subgraph.
[0027] In some embodiments, the fitting step of the price fluctuation prediction model includes:
[0028] Based on historical drug patent expiration cases, we collect the corresponding drug price change curves and graph structure features, calculate the corresponding perturbation intensity and the structural density of the affected subgraphs, and construct a training sample set.
[0029] Based on the training sample set, a linear regression model is trained with structural density and disturbance intensity as independent variables and price change magnitude in historical cases as dependent variables; or a machine learning model is trained based on the training sample set with structural density and disturbance intensity as input values and price change magnitude in historical cases as labels.
[0030] This invention also discloses a pharmaceutical supply chain intelligent forecasting system, comprising:
[0031] The graph construction module is used to build a knowledge graph of the pharmaceutical supply chain. It models raw material suppliers, manufacturers and distributors as nodes, and the logistics paths or contractual dependencies between nodes as weighted edges, forming an undirected graph structure.
[0032] The spectral decomposition module is used to construct a normalized Laplacian matrix based on the drug supply chain knowledge graph and to perform spectral decomposition on it to form a feature vector space.
[0033] The perturbation processing module is used to construct a perturbation vector after receiving a drug patent expiration event, and project it into the feature vector space to extract the affected sub-graphs due to the patent expiration.
[0034] The prediction model module is used to fit a price fluctuation prediction model based on the perturbation intensity of the perturbation vector on the whole graph in historical data and the structural density of the affected subgraph, and to use the price fluctuation prediction model to predict the price of drugs after patent expiration.
[0035] In some embodiments, the perturbation vector h is constructed as follows:
[0036] Define the node directly affected by the patent expiration event as the disturbance source node v. i According to v i For index i in the aforementioned drug supply chain knowledge graph, construct a perturbation vector of length |V|. Its definition is:
[0037]
[0038] Where |V| represents the total number of nodes in the supply chain graph, and j is the number of h.
[0039] In some embodiments, the step of identifying the affected subgraph includes:
[0040] Project the perturbation vector h onto the feature vector basis of the pharmaceutical supply chain knowledge graph, and calculate the response value of each node to the perturbation vector h. The response value is the inner product between the node's feature vector and the perturbation vector h.
[0041] Set a response threshold τ, and form a node set V consisting of all nodes whose response values are greater than τ. s ;
[0042] Extracting V from the pharmaceutical supply chain knowledge graph s The corresponding induced subgraph, wherein the nodes of the induced subgraph are V s The edge is the set of edges between nodes in the original graph, which is the affected subgraph that is directly or indirectly affected by the disturbance.
[0043] The advantages of this invention compared to existing technologies lie in its intelligent prediction method for the pharmaceutical supply chain. By modeling the pharmaceutical supply chain as an undirected weighted graph structure, and utilizing spectral graph theory to construct a normalized Laplace matrix and perform spectral decomposition, a feature vector space is introduced. This allows for the accurate mathematical expression and analysis of the propagation path and impact range of disturbances in the supply chain when facing pharmaceutical patent expiration events. Constructing disturbance vectors and projecting them onto the feature vector space enables precise identification of key nodes affected by patent expiration events and their constituent subgraphs, effectively overcoming the inaccuracies of traditional methods in identifying the direct impact range. Furthermore, by defining two metrics—disturbance intensity and structural density—the breadth of the disturbance propagation effect and the connectivity of the affected subgraphs are systematically characterized, providing a quantitative input basis for subsequent price prediction models. By collecting historical patent expiration cases and their corresponding price changes and spectral response data, a training sample set is constructed and fitted using linear regression or machine learning models, achieving effective prediction of price fluctuations with high interpretability and adaptability. Compared to methods that rely on expert experience or single-dimensional statistics, this solution can work together from two dimensions—supply chain structure and data-driven approaches—to improve the dynamic response and generalization capabilities of forecasts, thereby providing decision support for drug distribution management, market regulation, and generic drug launch strategies. Attached Figure Description
[0044] Figure 1 This is the overall flowchart of the present invention;
[0045] Figure 2 This is a flowchart of the spectrum construction and spectral decomposition process of this invention;
[0046] Figure 3 This is a flowchart of the perturbation propagation and subgraph extraction process of the present invention;
[0047] Figure 4 This is a flowchart of the calculation and model prediction of disturbance intensity and structural density in this invention. Detailed Implementation
[0048] The specific embodiments of the present invention will now be described with reference to the accompanying drawings.
[0049] like Figure 1 The flowchart shown is a method of the present invention, including:
[0050] Construct a knowledge graph of the pharmaceutical supply chain, modeling raw material suppliers, manufacturers, and distributors as nodes, and modeling the logistics paths or contractual dependencies between nodes as weighted edges, forming an undirected graph structure;
[0051] Based on the aforementioned drug supply chain knowledge graph, a normalized Laplace matrix is constructed, and its spectral decomposition is performed to form a feature vector space.
[0052] Upon receiving a drug patent expiration event, a perturbation vector is constructed and projected onto the feature vector space to extract the affected sub-graphs due to the patent expiration.
[0053] Based on the perturbation vector's perturbation intensity on the entire graph from historical data, and the structural density of the affected subgraphs, a price fluctuation prediction model is fitted, and this model is used to predict drug prices after patent expiration. More specifically:
[0054] To comprehensively capture the complex relationships within the pharmaceutical supply chain, it is first necessary to model the entities and their interactions as a graph structure. Specifically, raw material suppliers, manufacturers, and distributors in the supply chain are abstracted as nodes in the graph, while the logistical paths or contractual dependencies between nodes are modeled as weighted edges, forming an undirected graph structure. The motivation for this modeling approach is that the pharmaceutical supply chain is a highly interconnected network, and the strength of relationships between nodes, such as transaction frequency or material flow, is crucial to the dynamic response of the supply chain. Weighted edges quantify the strength of these relationships, thus providing a structured data foundation for subsequent disturbance analysis and price forecasting.
[0055] In implementation, assuming the supply chain comprises N entities, the graph G is defined as G = (V, E), where V is the set of nodes representing entities in the supply chain, such as raw material supplier A, manufacturer B, and distributor C; E is the set of edges representing relationships between nodes, such as the logistics path from supplier A to manufacturer B for supplying raw materials. Edge weights can be determined based on actual business data; for example, weights could be the average annual transaction frequency, contract amount, or material transportation volume between nodes. For instance, if a supplier supplies 1000 units of raw materials to a manufacturer monthly, the edge weight could be set to 1000. If the relationship is solely based on contractual dependencies, the weight can be normalized based on contract amount or transaction frequency.
[0056] In practice, the first step is to collect supply chain-related data, including supplier directories, manufacturer information, distributor networks, and transaction records between them. This data can be obtained through enterprise resource planning systems or supply chain management software. Then, using graph processing tools, such as Python's NetworkX library, the data is transformed into an undirected weighted graph structure and stored as an adjacency matrix for subsequent calculations. For example, in the simplest case, assuming the supply chain contains three nodes: supplier S1, manufacturer M1, and distributor D1, with transaction frequencies of 10 times per month from S1 to M1 and 15 times per month from M1 to D1, the adjacency matrix A would be:
[0057] A = [[0,10,0],
[0058] [10,0,15],
[0059] [0,15,0]];
[0060] In this way, the complete topology of the supply chain can be represented.
[0061] like Figure 2 As shown, after constructing the supply chain graph, it needs to be mathematically processed to extract key features. This is specifically achieved by constructing a normalized Laplacian matrix and performing spectral decomposition. The Laplacian matrix can capture the topological properties of the graph, and its normalized form can eliminate the influence of differences in node degree, making the analysis results more universal. Spectral decomposition further transforms the graph structure into an eigenvector space, facilitating the analysis of the propagation of perturbation effects.
[0062] The specific implementation steps are as follows: First, based on the graph G=(V,E), generate the adjacency matrix A, where A[i,j] represents the edge weight between node i and node j. For an undirected graph, A is a symmetric matrix. Next, calculate the degree matrix D, which is a diagonal matrix, where D[i,i] represents the degree of node i, i.e., the sum of the weights of all edges connected to it. The normalized Laplacian matrix L is defined as:
[0063] L = ID -1 / 2 *A*D -1 / 2 ;
[0064] Where I is the identity matrix, and D -1 / 2 Let L be the inverse square root of the degree matrix D. After calculating L, use a linear algebra library, such as the linalg module of NumPy, to perform eigenvalue decomposition, obtaining a set of eigenvalues and eigenvectors, forming the eigenvector space. The normalized Laplacian matrix reflects the global and local connectivity of the graph, while the eigenvector space provides a low-dimensional representation of the graph, facilitating the analysis of the propagation path of disturbances in the supply chain. For example, a pharmaceutical supply chain contains 100 nodes. After calculating its normalized Laplacian matrix, 100 eigenvectors are obtained. The first few eigenvectors usually correspond to the main connected components of the graph and can be used to identify key nodes or subnetworks in the supply chain.
[0065] In the implementation, assuming that the adjacency matrix A of the supply chain graph has already been generated, the Python code snippet is as follows:
[0066] import numpy as np
[0067] import networkx as nx
[0068] G = nx.Graph() # Create an undirected graph
[0069] #Add nodes and weighted edges, example data
[0070] G.add_weighted_edges_from([(0,1,10),(1,2,15)])
[0071] A = nx.adjacency_matrix(G).toarray() # Adjacency matrix
[0072] D = np.diag([sum(row)for row in A]) # Degree matrix
[0073] D_inv_sqrt = np.diag(1.0 / np.sqrt(np.diag(D))) # Inverse square root of the degree matrix
[0074] L = np.eye(len(A)) - D_inv_sqrt@A@D_inv_sqrt # Normalized Laplacian matrix eigenvalues, eigenvectors = np.linalg.eigh(L) # Eigenvalue decomposition
[0075] Based on the above calculations, the feature vector space can be used for subsequent perturbation vector projection and subgraph extraction.
[0076] When a drug patent expires, it is necessary to quantify its direct impact on the supply chain. In this embodiment of the invention, this is achieved by constructing a perturbation vector. Patent expiration typically leads to changes in the production costs or market competition landscape of the relevant drug, thereby affecting certain nodes in the supply chain, such as manufacturers. Therefore, the perturbation vector is designed to identify these directly affected nodes.
[0077] Specifically, assuming the patent expiration event directly affects node v i For example, if a manufacturer can produce generic drugs due to patent expiration, the disturbance vector h is defined as a vector of length |V|, where |V| is the total number of nodes in the supply chain graph. The element h[j] of vector h indicates whether node j is a disturbance source, defined as follows: if j equals i, i.e., a disturbance source node, then h[j] = 1; otherwise, h[j] = 0. This definition method can clearly identify the disturbance source and facilitates subsequent mathematical projection operations.
[0078] For example, suppose a supply chain contains 5 nodes, numbered 0 to 4. Patent expiration affects node 2, a certain manufacturer. Then the disturbance vector is h = [0,0,1,0,0]. In implementation, the number of the affected node can be obtained through an event-triggered mechanism. For example, through patent expiration notifications in the supply chain management system, the node ID of the relevant manufacturer can be automatically identified, and the corresponding disturbance vector can be generated.
[0079] like Figure 3As shown, the effects of patent expiration events are not limited to the directly affected nodes, but also propagate to other nodes, such as suppliers or distributors, through the supply chain network. To identify these sets of affected nodes and their subgraph structures, the perturbation vectors need to be projected onto the feature vector space, and the subgraphs need to be extracted based on the response values.
[0080] The specific implementation steps are as follows: First, project the perturbation vector h onto the eigenvector basis of the normalized Laplacian matrix, and calculate the response value of each node.
[0081] The response value is defined as the inner product of the eigenvector corresponding to the node and the perturbation vector h, i.e., r[j] = j ,h>, where u j Let be the feature vector of the j-th node. Next, set a response threshold τ, for example, τ = 0.1. The specific value can be adjusted according to the scale of the supply chain and experience. Nodes with response values greater than τ are grouped into a set V. s Finally, extract V from the original image G. s The corresponding induced subgraph, i.e., the node set is V s The edge set is V in the original graph. s The affected subgraph is obtained by considering all edges between nodes. In the above steps, the feature vector space captures the global propagation characteristics of the graph, while the inner product operation quantifies the sensitivity of each node to disturbances. Threshold filtering allows focusing on nodes that are more significantly affected. For example, a supply chain graph contains 10 nodes. The disturbance vector h, after projection, yields response values [0.05, 0.3, 0.8, 0.2, 0.01, 0, 0.4, 0.1, 0.02, 0.15]. Setting τ = 0.1, then V... s The induced subgraph contains nodes 1, 2, 3, 6, and 9, and the corresponding subgraph contains these nodes and the edges between them.
[0082] like Figure 4 As shown, to quantify the overall impact of patent expiration events on the entire supply chain, the disturbance intensity is defined as the sum of the absolute values of the responses of all nodes, i.e., S = ∑|r[j]|, where |r[j] is the response value of node j. This indicator reflects the propagation range and intensity of the disturbance in the supply chain; a larger value indicates a more widespread impact. The complexity of the supply chain means that the response of a single node is insufficient to reflect the global effect, and the disturbance intensity provides a global perspective.
[0083] To further analyze the connectivity of the affected subgraph, the structure density ρ is defined as the actual number of edges in the subgraph |E s |and the maximum possible number of edges|V s |(|V s The ratio of |-1) / 2 is given by the formula:
[0084] ρ=|E s | / (|V s |(|V s |-1) / 2).
[0085] The motivation behind this formula is that structural density reflects the connectivity within a subgraph; higher density indicates stronger dependencies between affected nodes, which may lead to more significant price fluctuations.
[0086] As mentioned above, disturbance intensity measures the structural responsiveness triggered by a patent expiration event across the entire supply chain graph, reflecting the potential for the disturbance to propagate throughout the graph. Structural density, on the other hand, measures the degree of connectivity of the affected subgraphs. The higher the degree of connectivity, the greater the correlation between the impact of the patent expiration event and the supply chain, and the more likely it is to cause price fluctuations.
[0087] Based on disturbance intensity and structure density, a price fluctuation prediction model is constructed to predict drug price changes after patent expiration. Specific steps include data collection, model training, and predictive application.
[0088] First, data on historical drug patent expiration cases is collected, including price change curves, supply chain graph structure, perturbation strength, and the structure density of the affected subgraph. For example, if the price of a drug drops by 20% within 6 months after its patent expires, the corresponding perturbation strength is 2.5 and the structure density is 0.75, thus forming a training sample. The sample set should cover multiple drug types and market environments to ensure the model's generalization ability.
[0089] Next, a prediction model is trained based on the training sample set. A simple approach is to use a linear regression model with disturbance intensity and structural density as independent variables and price change magnitude as the dependent variable, fitting the model f(S,ρ)=w1S+w2ρ+b, where w1, w2 and b are model parameters.
[0090] Linear regression offers the advantage of strong interpretability, making it suitable for preliminary analysis. However, to improve prediction accuracy, machine learning models such as random forests or neural networks can be employed. In some embodiments, a three-layer fully connected neural network can be constructed. The input layer receives the perturbation strength and structure density (2D input), the hidden layer contains 64 neurons using the ReLU activation function, and the output layer predicts the magnitude of price changes (1D output). The training process uses the mean squared error loss function, the Adam optimizer, a batch size of 32, and 100 training epochs.
[0091] Training code example, using Python's TensorFlow:
[0092]
[0093]
[0094] After training, the model is used to predict price changes for new patent expiration events. For example, if an event corresponds to a disturbance strength of 2.0 and a structure density of 0.8, the model predicts a 15% price decrease.
[0095] This invention also discloses a system corresponding to the above method, comprising:
[0096] The graph construction module builds a knowledge graph of the pharmaceutical supply chain, modeling raw material suppliers, manufacturers, and distributors as nodes, and the logistics paths or contractual dependencies between nodes as weighted edges, forming an undirected graph structure.
[0097] The spectral decomposition module is used to construct a normalized Laplacian matrix based on the drug supply chain knowledge graph and to perform spectral decomposition on it to form a feature vector space.
[0098] The perturbation processing module is used to construct a perturbation vector after receiving a drug patent expiration event, and project it into the feature vector space to extract the affected sub-graphs due to the patent expiration.
[0099] The prediction model module is used to fit a price fluctuation prediction model based on the perturbation intensity of the perturbation vector on the whole graph in historical data and the structural density of the affected subgraph, and to use the price fluctuation prediction model to predict the price of drugs after patent expiration.
[0100] Through the above implementation methods, the present invention realizes a complete process from supply chain modeling to price forecasting, and provides an efficient tool for pharmaceutical supply chain management by combining graph analysis and machine learning technologies.
[0101] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for intelligent forecasting of the pharmaceutical supply chain, characterized in that, Includes the following steps: Construct a knowledge graph of the pharmaceutical supply chain, modeling raw material suppliers, manufacturers, and distributors as nodes, and modeling the logistics paths or contractual dependencies between nodes as weighted edges, forming an undirected graph structure; Based on the aforementioned drug supply chain knowledge graph, a normalized Laplace matrix is constructed, and its spectral decomposition is performed to form a feature vector space. Upon receiving a drug patent expiration event, a perturbation vector is constructed and projected onto the feature vector space to extract the affected sub-graphs due to the patent expiration. Based on the perturbation intensity of the perturbation vector on the entire graph in historical data, and the structural density of the affected subgraph, a price fluctuation prediction model is fitted, and the price fluctuation prediction model is used to predict the price of drugs after patent expiration.
2. The intelligent forecasting method for the pharmaceutical supply chain according to claim 1, characterized in that, The steps to construct a normalized Laplacian matrix include: The pharmaceutical supply chain knowledge graph is represented as an undirected weighted graph G = (V, E), where V is the set of supply chain nodes, E is the set of edges, and the weight of the edges represents the transaction frequency, contractual dependence, or material flow intensity between nodes. Construct the adjacency matrix A and degree matrix D of the drug supply chain knowledge graph, and calculate the normalized Laplacian matrix; The normalized Laplacian matrix is decomposed into eigenvalues to form an eigenvector space.
3. The intelligent forecasting method for the pharmaceutical supply chain according to claim 1, characterized in that, The process of constructing the perturbation vector h is as follows: Define the node directly affected by the patent expiration event as the disturbance source node v. i According to v i For index i in the aforementioned drug supply chain knowledge graph, construct a perturbation vector of length |V|. Its definition is: Where |V| represents the total number of nodes in the supply chain graph, and j is the number of h.
4. The intelligent forecasting method for the pharmaceutical supply chain according to claim 1, characterized in that, The steps to identify the affected subgraph include: Project the perturbation vector h onto the feature vector basis of the pharmaceutical supply chain knowledge graph, and calculate the response value of each node to the perturbation vector h. The response value is the inner product between the node's feature vector and the perturbation vector h. Set a response threshold τ, and form a node set V consisting of all nodes whose response values are greater than τ. s ; Extracting V from the pharmaceutical supply chain knowledge graph s The corresponding induced subgraph, wherein the nodes of the induced subgraph are V s The edge is the set of edges between nodes in the original graph, which is the affected subgraph that is directly or indirectly affected by the disturbance.
5. The intelligent forecasting method for the pharmaceutical supply chain according to claim 4, characterized in that, The disturbance intensity is defined as the sum of the absolute values of the response values of all nodes.
6. The intelligent forecasting method for the pharmaceutical supply chain according to claim 1, characterized in that, The structural density is the ratio of the actual number of edges to the maximum possible number of edges in the affected subgraph, and the calculation formula is: ρ=|E s | / (|V s |*(|V s |-1) / 2); Where |E s | represents the number of edges in the subgraph, |V s | represents the number of nodes in the subgraph, |V s |*(|V s |-1) / 2 represents the maximum possible number of edges in the affected subgraph.
7. The intelligent forecasting method for the pharmaceutical supply chain according to claim 1, characterized in that, The fitting steps of the price fluctuation prediction model include: Based on historical drug patent expiration cases, we collect the corresponding drug price change curves and graph structure features, calculate the corresponding perturbation intensity and the structural density of the affected subgraphs, and construct a training sample set. Based on the training sample set, a linear regression model is trained with structural density and disturbance intensity as independent variables and price change magnitude in historical cases as dependent variables; or a machine learning model is trained based on the training sample set with structural density and disturbance intensity as input values and price change magnitude in historical cases as labels.
8. A system for implementing the intelligent forecasting method for the pharmaceutical supply chain as described in claim 1, characterized in that, include: The graph construction module is used to build a knowledge graph of the pharmaceutical supply chain. It models raw material suppliers, manufacturers and distributors as nodes, and the logistics paths or contractual dependencies between nodes as weighted edges, forming an undirected graph structure. The spectral decomposition module is used to construct a normalized Laplacian matrix based on the drug supply chain knowledge graph and to perform spectral decomposition on it to form a feature vector space. The perturbation processing module is used to construct a perturbation vector after receiving a drug patent expiration event, and project it into the feature vector space to extract the affected sub-graphs due to the patent expiration. The prediction model module is used to fit a price fluctuation prediction model based on the perturbation intensity of the perturbation vector on the whole graph in historical data and the structural density of the affected subgraph, and to use the price fluctuation prediction model to predict the price of drugs after patent expiration.
9. The system according to claim 8, characterized in that, The process of constructing the perturbation vector h is as follows: Define the node directly affected by the patent expiration event as the disturbance source node v. i According to v i For index i in the aforementioned drug supply chain knowledge graph, construct a perturbation vector of length |V|. Its definition is: Where |V| represents the total number of nodes in the supply chain graph, and j is the number of h.
10. The system according to claim 8, characterized in that, The steps to identify the affected subgraph include: Project the perturbation vector h onto the feature vector basis of the pharmaceutical supply chain knowledge graph, and calculate the response value of each node to the perturbation vector h. The response value is the inner product between the node's feature vector and the perturbation vector h. Set a response threshold τ, and form a node set V consisting of all nodes whose response values are greater than τ. s ; Extracting V from the pharmaceutical supply chain knowledge graph s The corresponding induced subgraph, wherein the nodes of the induced subgraph are V s The edge is the set of edges between nodes in the original graph, which is the affected subgraph that is directly or indirectly affected by the disturbance.