Supply chain middle platform data management system and method using big model technology

Abstract

The present invention discloses a supply chain middle-end data management system and method using large model technology, which relates to the field of large model technology. The present invention extracts supply data from supply records, cleans and pre-processes all supply data, and aligns the supply data according to timestamps; extracts supply entities and supply relationships, and constructs a knowledge graph for the supply chain middle-end; combines the constructed knowledge graph to find risk features that affect supply; constructs a virtual supply chain sandbox for each supply path, and generates a risk warning model; extracts real-time supply environment data and inputs it into the virtual supply chain sandbox of the real-time supply path, and uses the risk warning model to judge supply risks; visually marks risk paths in the virtual supply chain sandbox, extracts risk data for risk warnings, and generates a risk report.

Description

Technical Field

[0001] The present invention relates to the field of large model technology, and specifically to a supply chain middle-end data management system and method using large model technology. Background Art

[0002] In the early days, companies primarily used spreadsheets and simple databases to record and manage supply chain data, such as basic information like inventory quantities and purchase orders. Data management during this period was fragmented and isolated, with each department or business process often operating independently, and data lacking effective linkage and sharing. With the development of technologies like the internet and the Internet of Things, the amount of data generated within the supply chain has exploded. Various systems, devices, and business processes within and outside the enterprise continuously generate data, such as purchase orders, inventory records, logistics trajectories, and sales data. Supply chains involve multiple participants and diverse business systems, generating data from a wide range of sources and in a variety of formats, including structured data (such as data in databases), semi-structured data (such as XML and JSON formats), and unstructured data (such as documents, images, and videos). Different companies or departments also use different data recording methods and standards, making it difficult to directly share and integrate data. Supply chain middleware data management technologies are needed to integrate and standardize multi-source, heterogeneous data. In addition, when analyzing data in the supply chain middle platform, most of the current management is centralized and analyzed by the data center, which makes the risk prediction not specific enough. Different supply paths and lines in the supply chain may have different influencing factors, so it is crucial to carry out targeted management for different supply scenarios. Summary of the Invention

[0003] The purpose of the present invention is to provide a supply chain middle-stage data management system and method using large model technology to solve the problems raised in the prior art.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A supply chain middle platform data management method using large model technology, the method comprising the following steps:

[0006] S100: Collect supply records in the supply chain middle platform, extract supply data from the supply records, clean and pre-process all supply data, and align the supply data according to timestamps;

[0007] Furthermore, the specific steps for aligning the supply data according to the timestamp are as follows:

[0008] S101. Find the storage location of supply records in the supply chain platform, obtain different storage databases for different types of data, use the API to establish a connection with the database, collect the supply records in the supply chain platform through the connection, and extract all supply data from the supply records;

[0009] S102, collect the data intervals of adjacent data in all the extracted supply data, calculate the average value of all adjacent data intervals, use the calculated average value as the interval threshold, and use the interval threshold to judge every two adjacent data intervals collected. When L a，b > Lp, it is determined that there are missing values ​​between the supply data a and b, L a，b represents the data interval between supply data a and supply data b, Lp represents the interval threshold of supply data; when it is determined that there are missing values ​​in adjacent supply data, the average value of adjacent supply data is calculated, and the formula is: , Tc represents the average value of adjacent supply data, and the filling method is used to supplement the missing values, and the average value of adjacent supply data is used as the filling value to fill in the missing values;

[0010] Missing data can make a dataset incomplete and hinder an accurate understanding of the overall situation. Filling in missing data can make the data more complete and reflect the actual situation more comprehensively.

[0011] S103. Calculate the Z-score value of each extracted supply data using the following formula: In the formula, Z represents the Z-score value of each extracted supply data, and x represents the true value of each extracted supply data. represents the average value of all supply data extracted, Indicates the standard deviation of all extracted supply data; set the abnormal threshold to Zy. When the calculated Z-score value of the supply data is greater than the abnormal threshold Zy, the corresponding supply data is judged to be abnormal data and the abnormal data is deleted;

[0012] Anomalous data is caused by data entry errors, sensor failures, or other reasons that do not conform to normal data patterns. Removing anomalous data helps ensure data consistency and prevents individual anomalous data from misleading overall analysis. Anomalous data can also significantly interfere with model training and predictions, and may even cause model failure. Removing anomalous data can make models more robust, improving their generalization and prediction accuracy.

[0013] S104: extract the timestamps in the supply data, unify the timestamp formats of all supply data, sort all supply data according to the timestamps, and time-align the supply data according to the sorting results.

[0014] S200: Use the big model to process the supply data, extract the supply entities and supply relationships, match the supply entities and supply relationships, and build a knowledge graph for the supply chain platform;

[0015] Furthermore, the specific steps to build the knowledge graph of the supply chain platform are:

[0016] S201: Input all the supply data after preprocessing and time alignment into the big model, extract the supply entities in the supply data through the big model, set the entity label u, and use the entity label u to judge the supply data. The formula is:

[0017] ;

[0018] In the formula, P(u|v) represents the probability that the supply data v is a supply entity, v is the input supply data, u is the entity label, W is the weight, c is the bias of the large model, BERT is the large model type, and Softmax is the calculation function. A probability threshold is set, and when the calculated value is greater than the set probability threshold, it is judged to be a supply entity. The supply entities of all supply data are extracted using the BERT large model and labeled with the entity label u.

[0019] S202. Analyze all supply data using the large model. Based on the labeled supply entities, output a fixed supply structure [entity-non-entity data-entity]. In the fixed supply structure, when there are two supply entities in the supply data and a non-entity data exists between them, the corresponding non-entity data is determined to be a supply relationship. Match the adjacent entities of each supply relationship with the extracted supply entities to generate a supply line [entity 1-supply relationship-entity 2], where entity 1 is the supply end and entity 2 is the receiving end.

[0020] S203. After analyzing all supply data and generating all supply lines, graph embedding is used to convert the extracted supply entities into low-dimensional vectors. Similarity analysis is performed on the entities in the supply lines. Entities 1 in two supply lines are extracted respectively. The vector values ​​of the two entities 1 are input into the Manhattan distance formula. The distance between the two entities 1 is calculated as D(s1 i , s1 j ), s1 i Represents entity 1, s1 in supply line i j Represents entity 1 in supply line j, and the similarity is calculated using the distance between two entities 1. The formula is:

[0021] ;

[0022] In the formula, Sim represents the similarity between the two entities 1 in supply lines i and j. When the calculated similarity Sim = 1, the two entities 1 in supply lines i and j are judged to be the same, and the two supply lines are merged into [entity 1 - (supply relationship - entity 2) i 、(Supply relationship-entity 2) j ]; Use the same method to judge entity 2, and judge entity 1 and entity 2 in all supply lines in turn, merge the supply lines with the same entity to eliminate the ambiguity of the graph;

[0023] S204. Graph the supply lines after eliminating graph ambiguity, use entities as graph nodes, and supply relationships as edges to construct a knowledge graph for the supply chain middle platform.

[0024] Because data in the supply chain comes from a wide range of sources and formats, building a knowledge graph for the supply chain middleware using big models can effectively integrate and process multi-source heterogeneous data. Whether it's structured data from internal enterprise systems or unstructured data from external channels like social media and industry reports, big models can transform it into a unified representation for comprehensive analysis, providing enterprises with a more comprehensive view of the supply chain.

[0025] When building a knowledge graph, the large model can accurately match the extracted supply entities and supply relationships, ensuring that the nodes and edges in the knowledge graph accurately reflect the actual situation in the supply chain. This helps reduce incorrect connections and false relationships, and improves the accuracy and reliability of the knowledge graph.

[0026] When building a knowledge graph, if ambiguity is not eliminated, identical entities will form identical nodes, adding unnecessary nodes to the knowledge graph, increasing the amount of data analysis, and resulting in a less clear knowledge graph. Eliminating ambiguity makes supply chain data more consistent and clear, and allows different departments and personnel to have a more unified understanding of the data. This helps reduce communication costs and misunderstandings, and improves the efficiency and collaboration of supply chain management. Clear and consistent data also facilitates data analysis and decision support, providing businesses with a more reliable basis.

[0027] S300: Collect supply data from historical supply records in the supply chain platform when anomalies occur, and use the constructed knowledge graph to find risk features that affect supply;

[0028] Furthermore, the specific steps for finding risk characteristics that affect supply by combining the constructed knowledge graph are as follows:

[0029] S301. Collect supply data when anomalies occur in historical supply records in the supply chain middle platform, compare all collected supply data when anomalies occur with supply data during normal supply, and judge the supply data that are different after comparison as suspicious data; in the knowledge graph of the supply chain middle platform, all the judged suspicious data are matched in the supply path, and when the suspicious data exists continuously in the supply path in the knowledge graph, the corresponding continuously existing suspicious data is judged as a risk feature; conversely, the suspicious data that exists alone on one edge of the supply path is judged as a non-risk feature; all risk features in the supply data are obtained through judgment.

[0030] S400: Differentiate each supply path in the knowledge graph, build a virtual supply chain sandbox for each supply path, collect supply environment data and risk characteristics of each supply path, and generate a risk warning model;

[0031] Furthermore, the specific steps for generating a risk warning model are as follows:

[0032] S401. Differentiate all supply paths in the knowledge graph using a graph traversal algorithm, collect supply environment data and entity supply attributes for each supply path, combine the collected supply environment data and supply attributes, and construct a virtual supply sandbox for each supply path using a discrete time model.

[0033] S402. Draw a curve graph using the supply environment data and risk characteristics in each supply path. Use the supply environment data as an independent variable and the risk characteristics as a variable. The drawn curve graph is segmented to obtain a function polynomial of the supply environment data and risk characteristics. The formula is:

[0034] ;

[0035] In the formula, Tf represents the risk characteristics, H represents the supply environment data, G represents the total number of segments in the curve graph, and f(H) k The expression of the k-th segment curve is represented, M represents the constant term; the function polynomial of each risk feature and supply environment data is calculated in turn;

[0036] The average value and standard deviation of the risk characteristics of historical supply anomalies are collected, and the threshold value of each risk characteristic is obtained by subtracting the standard deviation from the average value. The early warning inequality for each risk characteristic is constructed as follows: , the early warning inequalities of all risk characteristics are integrated to obtain the risk early warning model.

[0037] When supplying in the supply chain, there are many supply paths, and there are continuous supply edges in the supply paths. The risk characteristics and supply environment data in them are different. Generating function polynomials for different supply paths and supply edges and constructing a risk warning model can not only warn of risks in advance, but also locate and mark the locations where risks exist; different function polynomials can be used to judge the risks of different supply paths and supply edges in a distributed manner, making risk warnings more relevant and detailed, and increasing the generalization of the system.

[0038] S500: When performing real-time supply within the supply chain platform, extract the supply entities of the real-time supply and map them in the knowledge graph to obtain the real-time supply path. Extract the real-time supply environment data and input it into the virtual supply chain sandbox of the real-time supply path. Use the risk warning model to judge the supply risk.

[0039] Furthermore, the specific steps for using the risk warning model to judge supply risks are as follows:

[0040] S501. When real-time supply is carried out in the supply chain middle platform, the supply entity of the real-time supply is extracted and matched in the knowledge graph to obtain the real-time supply path; the supply environment data in the real-time supply path is extracted, and the supply is simulated using the virtual supply sandbox in the real-time supply path. When the simulated supply reaches the corresponding edge, the real-time supply environment data in the corresponding edge is extracted and input into the risk warning model. When an inequality is satisfied, it is judged that there is a risk in the supply path on this edge, and a risk warning is issued.

[0041] Using a virtual supply sandbox for simulated supply, you can proactively identify risks based on supply environment data along the supply path, providing early warnings and avoiding the financial losses and increased costs associated with supply anomalies during real-world supply. The virtual supply sandbox also identifies specific risk locations, allowing users to clearly identify them and develop appropriate response plans.

[0042] S600: When it is determined that there is a risk in real-time supply, a risk warning is issued, the risk path is visually marked in the virtual supply chain sandbox, the risk data of the risk warning is extracted, and a risk report is generated.

[0043] Furthermore, the specific steps for generating a risk report are:

[0044] S601. When it is determined that there is a risk in real-time supply and a risk warning is issued in advance, the risk path is marked with different colors in the virtual supply sandbox, the number of risk features that meet the inequality in the risk path is calculated, and the marking color is calculated using the number of risk features as the color weight. The supply edge position with risk, the risk features that meet the inequality, and the supply environment data with risk are extracted in the virtual supply sandbox, and a risk report is generated using the extracted information.

[0045] The supply chain middle-end data management system that applies big model technology includes a data collection module, a knowledge graph construction module, a risk warning model construction module, a virtual sandbox warning module, and a report generation module.

[0046] The data collection module is used to collect supply records in the supply chain platform, extract supply data from the supply records, clean and pre-process all supply data, and align the supply data according to timestamps;

[0047] The knowledge graph construction module is used to process supply data using a large model, extract supply entities and supply relationships, match supply entities and supply relationships, and construct a knowledge graph for the supply chain platform;

[0048] The risk warning model building module is used to collect supply data when anomalies occur in historical supply records in the supply chain platform, extract risk characteristics that affect supply; collect supply environment data and risk characteristics of each supply path, and generate a risk warning model;

[0049] The virtual sandbox early warning module is used to collect supply environment data in real time and input it into the risk early warning model, using the virtual supply sandbox to simulate supply and determine whether there is risk in each supply path;

[0050] The report generation module is used to issue a risk warning when it is determined that there is a risk in real-time supply, visually mark the risk path in the virtual supply chain sandbox, extract risk data of the risk warning, and generate a risk report.

[0051] The knowledge graph construction module includes a supply entity unit, a supply relationship unit, an ambiguity elimination unit, and a graph construction unit;

[0052] The supply entity unit is used to analyze the supply data using the BERT large model and extract the supply entity;

[0053] The supply relationship unit is used to construct a fixed structure, judge the output fixed structure, and obtain the supply relationship;

[0054] The ambiguity elimination unit is used to calculate the similarity of entities in different supply lines and merge the same entities;

[0055] The graph construction unit is used to graph the supply lines after eliminating graph ambiguity, using entities as graph nodes and supply relationships as edges to construct a knowledge graph for the supply chain middle platform.

[0056] The risk warning model construction module includes a risk feature extraction unit, a function construction unit and a model construction unit;

[0057] The risk feature extraction unit is used to collect supply data when anomalies occur in historical supply records in the supply chain platform, and to find risk features that affect supply in combination with the constructed knowledge graph;

[0058] The function construction unit is used to analyze the collected supply environment data and risk characteristics, draw a curve graph, and generate a function polynomial;

[0059] The model building unit is used to merge the function polynomials of all risk characteristics to build a risk warning model.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] 1. When constructing a knowledge graph, the present invention's large model can accurately match extracted supply entities and supply relationships, ensuring that the nodes and edges in the knowledge graph accurately reflect the actual situation in the supply chain. This helps reduce incorrect connections and false relationships, and improves the accuracy and reliability of the knowledge graph.

[0062] 2. By eliminating ambiguity in the knowledge graph, this invention makes supply chain data more consistent and clear, and allows for a more unified understanding of the data across departments and personnel. This helps reduce communication costs and misunderstandings, and improves the efficiency and synergy of supply chain management. Clear and consistent data also facilitates data analysis and decision support, providing businesses with a more reliable basis.

[0063] 3. The present invention generates function polynomials for different supply paths and supply edges, and constructs a risk warning model that can not only warn of risks in advance, but also locate and mark the locations where risks exist; and different function polynomials can judge the risks of different supply paths and supply edges in a distributed manner, making risk warnings more tailored and detailed, and increasing the generalization of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 This is a module distribution diagram of the supply chain middle-end data management system using the large model technology of the present invention;

[0065] Figure 2 A schematic diagram of the steps of the supply chain middle platform data management method using the large model technology of the present invention;

[0066] Figure 3 This is the knowledge graph of the supply chain middle-end data management method using big model technology in the present invention. DETAILED DESCRIPTION

[0067] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0068] Example: Figure 1-Figure 3 As shown, the present invention provides a technical solution.

[0069] A supply chain middle platform data management method using large model technology, the method comprising the following steps:

[0070] S100: Collect supply records in the supply chain middle platform, extract supply data from the supply records, clean and pre-process all supply data, and align the supply data according to timestamps;

[0071] The specific steps to align supply data based on timestamps are:

[0072] S101. Find the storage location of supply records in the supply chain platform, obtain different storage databases for different types of data, use the API to establish a connection with the database, collect the supply records in the supply chain platform through the connection, and extract all supply data from the supply records;

[0073] S102, collect the data intervals of adjacent data in all the extracted supply data, calculate the average value of all adjacent data intervals, use the calculated average value as the interval threshold, and use the interval threshold to judge every two adjacent data intervals collected. When L a，b > Lp, it is determined that there are missing values ​​between the supply data a and b, L a，b represents the data interval between supply data a and supply data b, Lp represents the interval threshold of supply data; when it is determined that there are missing values ​​in adjacent supply data, the average value of adjacent supply data is calculated, and the formula is: , Tc represents the average value of adjacent supply data, and the filling method is used to supplement the missing values, and the average value of adjacent supply data is used as the filling value to fill in the missing values;

[0074] Missing data can make a dataset incomplete and hinder an accurate understanding of the overall situation. Filling in missing data can make the data more complete and reflect the actual situation more comprehensively.

[0075] S103. Calculate the Z-score value of each extracted supply data using the following formula: In the formula, Z represents the Z-score value of each extracted supply data, and x represents the true value of each extracted supply data. represents the average value of all supply data extracted, Indicates the standard deviation of all extracted supply data; set the abnormal threshold to Zy. When the calculated Z-score value of the supply data is greater than the abnormal threshold Zy, the corresponding supply data is judged to be abnormal data and the abnormal data is deleted;

[0076] Anomalous data is caused by data entry errors, sensor failures, or other reasons that do not conform to normal data patterns. Removing anomalous data helps ensure data consistency and prevents individual anomalous data from misleading overall analysis. Anomalous data can also significantly interfere with model training and predictions, and may even cause model failure. Removing anomalous data can make models more robust, improving their generalization and prediction accuracy.

[0077] S104: extract the timestamps in the supply data, unify the timestamp formats of all supply data, sort all supply data according to the timestamps, and time-align the supply data according to the sorting results.

[0078] S200: Use the big model to process the supply data, extract the supply entities and supply relationships, match the supply entities and supply relationships, and build a knowledge graph for the supply chain platform;

[0079] The specific steps to build the knowledge graph of the supply chain platform are:

[0080] S201: Input all the supply data after preprocessing and time alignment into the big model, extract the supply entities in the supply data through the big model, set the entity label u, and use the entity label u to judge the supply data. The formula is:

[0081] ;

[0082] In the formula, P(u|v) represents the probability that the supply data v is a supply entity, v is the input supply data, u is the entity label, W is the weight, c is the bias of the large model, BERT is the large model type, and Softmax is the calculation function. A probability threshold is set, and when the calculated value is greater than the set probability threshold, it is judged to be a supply entity. The supply entities of all supply data are extracted using the BERT large model and labeled with the entity label u.

[0083] S202. Analyze all supply data using the large model. Based on the labeled supply entities, output a fixed supply structure [entity-non-entity data-entity]. In the fixed supply structure, when there are two supply entities in the supply data and a non-entity data exists between them, the corresponding non-entity data is determined to be a supply relationship. Match the adjacent entities of each supply relationship with the extracted supply entities to generate a supply line [entity 1-supply relationship-entity 2], where entity 1 is the supply end and entity 2 is the receiving end.

[0084] S203. After analyzing all supply data and generating all supply lines, graph embedding is used to convert the extracted supply entities into low-dimensional vectors. Similarity analysis is performed on the entities in the supply lines. Entities 1 in two supply lines are extracted respectively. The vector values ​​of the two entities 1 are input into the Manhattan distance formula. The distance between the two entities 1 is calculated as D(s1 i , s1 j ), s1 i Represents entity 1, s1 in supply line i j Represents entity 1 in supply line j, and the similarity is calculated using the distance between two entities 1. The formula is:

[0085] ;

[0086] In the formula, Sim represents the similarity between the two entities 1 in supply lines i and j. When the calculated similarity Sim = 1, the two entities 1 in supply lines i and j are judged to be the same, and the two supply lines are merged into [entity 1 - (supply relationship - entity 2) i 、(Supply relationship-entity 2) j ]; Use the same method to judge entity 2, and judge entity 1 and entity 2 in all supply lines in turn, merge the supply lines with the same entity to eliminate the ambiguity of the graph;

[0087] S204. Graph the supply lines after eliminating graph ambiguity, use entities as graph nodes, and supply relationships as edges to construct a knowledge graph for the supply chain middle platform.

[0088] Because data in the supply chain comes from a wide range of sources and formats, building a knowledge graph for the supply chain middleware using big models can effectively integrate and process this multi-source, heterogeneous data. Whether it's structured data from internal enterprise systems or unstructured data from external channels like social media and industry reports, big models can transform it into a unified representation for comprehensive analysis, providing enterprises with a more comprehensive view of the supply chain.

[0089] When building a knowledge graph, the large model can accurately match the extracted supply entities and supply relationships, ensuring that the nodes and edges in the knowledge graph accurately reflect the actual situation in the supply chain. This helps reduce incorrect connections and false relationships, and improves the accuracy and reliability of the knowledge graph.

[0090] When building a knowledge graph, if ambiguity is not eliminated, identical entities will form identical nodes, adding unnecessary nodes to the knowledge graph, increasing the amount of data analysis, and resulting in a less clear knowledge graph. Eliminating ambiguity makes supply chain data more consistent and clear, and allows different departments and personnel to have a more unified understanding of the data. This helps reduce communication costs and misunderstandings, and improves the efficiency and collaboration of supply chain management. Clear and consistent data also facilitates data analysis and decision support, providing businesses with a more reliable basis.

[0091] S300: Collect supply data from historical supply records in the supply chain platform when anomalies occur, and use the constructed knowledge graph to find risk features that affect supply;

[0092] The specific steps for finding risk features that affect supply based on the constructed knowledge graph are as follows:

[0093] S301. Collect supply data when anomalies occur in historical supply records in the supply chain middle platform, compare all collected supply data when anomalies occur with supply data during normal supply, and judge the supply data that are different after comparison as suspicious data; in the knowledge graph of the supply chain middle platform, all the judged suspicious data are matched in the supply path, and when the suspicious data exists continuously in the supply path in the knowledge graph, the corresponding continuously existing suspicious data is judged as a risk feature; conversely, the suspicious data that exists alone on one edge of the supply path is judged as a non-risk feature; all risk features in the supply data are obtained through judgment.

[0094] S400: Differentiate each supply path in the knowledge graph, build a virtual supply chain sandbox for each supply path, collect supply environment data and risk characteristics of each supply path, and generate a risk warning model;

[0095] The specific steps to generate a risk warning model are:

[0096] S401. Differentiate all supply paths in the knowledge graph using a graph traversal algorithm, collect supply environment data and entity supply attributes for each supply path, combine the collected supply environment data and supply attributes, and construct a virtual supply sandbox for each supply path using a discrete time model.

[0097] S402. Draw a curve graph using the supply environment data and risk characteristics in each supply path. Use the supply environment data as an independent variable and the risk characteristics as a variable. The drawn curve graph is segmented to obtain a function polynomial of the supply environment data and risk characteristics. The formula is:

[0098] ;

[0099] In the formula, Tf represents the risk characteristics, H represents the supply environment data, G represents the total number of segments in the curve graph, and f(H) k The expression of the k-th segment curve is represented, M represents the constant term; the function polynomial of each risk feature and supply environment data is calculated in turn;

[0100] The average value and standard deviation of the risk characteristics of historical supply anomalies are collected, and the threshold value of each risk characteristic is obtained by subtracting the standard deviation from the average value. The early warning inequality for each risk characteristic is constructed as follows: , the early warning inequalities of all risk characteristics are integrated to obtain the risk early warning model.

[0101] When supplying in the supply chain, there are many supply paths, and there are continuous supply edges in the supply paths. The risk characteristics and supply environment data in them are different. Generating function polynomials for different supply paths and supply edges and constructing a risk warning model can not only warn of risks in advance, but also locate and mark the locations where risks exist; different function polynomials can be used to judge the risks of different supply paths and supply edges in a distributed manner, making risk warnings more relevant and detailed, and increasing the generalization of the system.

[0102] S500: When performing real-time supply within the supply chain platform, extract the supply entities of the real-time supply and map them in the knowledge graph to obtain the real-time supply path. Extract the real-time supply environment data and input it into the virtual supply chain sandbox of the real-time supply path. Use the risk warning model to judge the supply risk.

[0103] The specific steps for using the risk early warning model to judge supply risks are as follows:

[0104] S501. When real-time supply is carried out in the supply chain middle platform, the supply entity of the real-time supply is extracted and matched in the knowledge graph to obtain the real-time supply path; the supply environment data in the real-time supply path is extracted, and the supply is simulated using the virtual supply sandbox in the real-time supply path. When the simulated supply reaches the corresponding edge, the real-time supply environment data in the corresponding edge is extracted and input into the risk warning model. When an inequality is satisfied, it is judged that there is a risk in the supply path on this edge, and a risk warning is issued.

[0105] Using a virtual supply sandbox for simulated supply, you can proactively identify risks based on supply environment data along the supply path, providing early warnings and avoiding the financial losses and increased costs associated with supply anomalies during real-world supply. The virtual supply sandbox also identifies specific risk locations, allowing users to clearly identify them and develop appropriate response plans.

[0106] S600: When it is determined that there is a risk in real-time supply, a risk warning is issued, the risk path is visually marked in the virtual supply chain sandbox, the risk data of the risk warning is extracted, and a risk report is generated.

[0107] The specific steps to generate a risk report are:

[0108] S601. When it is determined that there is a risk in real-time supply and a risk warning is issued in advance, the risk path is marked with different colors in the virtual supply sandbox, the number of risk features that meet the inequality in the risk path is calculated, and the mark color is calculated using the number of risk features as the color weight. The minimum risk mark color data value is set to Bs. The formula is: , Se represents the calculated risk mark color value, and R represents the number of risk features; in the virtual supply sandbox, the supply edge positions with risks, the risk features that satisfy the inequality, and the supply environment data with risks are extracted, and a risk report is generated using the extracted information.

[0109] The supply chain middle-end data management system that applies big model technology includes a data collection module, a knowledge graph construction module, a risk warning model construction module, a virtual sandbox warning module, and a report generation module.

[0110] The data collection module is used to collect supply records in the supply chain platform, extract supply data from the supply records, clean and pre-process all supply data, and align the supply data according to timestamps;

[0111] The knowledge graph construction module is used to process supply data using a large model, extract supply entities and supply relationships, match supply entities and supply relationships, and construct a knowledge graph for the supply chain platform;

[0112] The risk warning model building module is used to collect supply data when anomalies occur in historical supply records in the supply chain platform, extract risk characteristics that affect supply; collect supply environment data and risk characteristics of each supply path, and generate a risk warning model;

[0113] The virtual sandbox early warning module is used to collect supply environment data in real time and input it into the risk early warning model, using the virtual supply sandbox to simulate supply and determine whether there is risk in each supply path;

[0114] The report generation module is used to issue a risk warning when it is determined that there is a risk in real-time supply, visually mark the risk path in the virtual supply chain sandbox, extract risk data of the risk warning, and generate a risk report.

[0115] The knowledge graph construction module includes a supply entity unit, a supply relationship unit, an ambiguity elimination unit, and a graph construction unit;

[0116] The supply entity unit is used to analyze the supply data using the BERT large model and extract the supply entity;

[0117] The supply relationship unit is used to construct a fixed structure, judge the output fixed structure, and obtain the supply relationship;

[0118] The ambiguity elimination unit is used to calculate the similarity of entities in different supply lines and merge the same entities;

[0119] The graph construction unit is used to graph the supply lines after eliminating graph ambiguity, using entities as graph nodes and supply relationships as edges to construct a knowledge graph for the supply chain middle platform.

[0120] The risk warning model construction module includes a risk feature extraction unit, a function construction unit and a model construction unit;

[0121] The risk feature extraction unit is used to collect supply data when anomalies occur in historical supply records in the supply chain platform, and to find risk features that affect supply in combination with the constructed knowledge graph;

[0122] The function construction unit is used to analyze the collected supply environment data and risk characteristics, draw a curve graph, and generate a function polynomial;

[0123] The model building unit is used to merge the function polynomials of all risk characteristics to build a risk warning model.

[0124] Example: Now analyze the supply chain middle platform data of a certain enterprise. First, collect a series of supply data and pre-process them to obtain complete supply data;

[0125] The entities extracted from the enterprise supply data using the big model are supply side 1 and 2, transit side 3, processing side 4 and 5, and sales side 6 and 7;

[0126] The extracted supply relationships are 1→4, 1→3, 2→3, 3→4, 3→5, 4→6, 4→7, and 5→7. It is judged that there are ambiguities in the supply lines 1→4 and 1→3, 3→4 and 3→5, and 4→6 and 4→7. Entities are merged and the knowledge graph is constructed. Figure 3 As shown;

[0127] For example: the supply path is: 1→3→4→6; the supply edge is each edge →;

[0128] Examples of risk features extracted from the knowledge graph are 2→3 (time), 1→4 (output), and 5→7 (time); the corresponding supply environment data are 2→3 (transportation route length), 1→4 (equipment), and 5→7 (enterprise turnover);

[0129] Calculate the function polynomial of the above distance and construct a risk warning model. The numerical value of each data (excluding the unit) is used in the calculation. For example, the function polynomial of 2→3 is t1=2.5×Y+4; Y represents 2→3 (the length of the transportation route). Let the obtained risk warning model be , t1 represents 2→3 (time), t2 represents 5→7 (time), and CH represents 1→4 (output);

[0130] Using the virtual supply sandbox simulation, the real-time 2→3 time is 30, which indicates that there is a risk and issues an early warning.

[0131] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A supply chain middle platform data management method using large model technology, characterized by: The method comprises the following steps: S100: Collect supply records in the supply chain middle platform, extract supply data from the supply records, clean and pre-process all supply data, and align the supply data according to timestamps; S200: Use the big model to process the supply data, extract the supply entities and supply relationships, match the supply entities and supply relationships, and build a knowledge graph for the supply chain platform; The specific steps to build the knowledge graph of the supply chain platform are: S201: Input all the supply data after preprocessing and time alignment into the big model, extract the supply entities in the supply data through the big model, set the entity label u, and use the entity label u to judge the supply data. The formula is: ； In the formula, P(u|v) represents the probability that the supply data v is a supply entity, v is the input supply data, u is the entity label, W is the weight, c is the bias of the large model, BERT is the large model type, and Softmax is the calculation function. A probability threshold is set, and when the calculated value is greater than the set probability threshold, it is judged to be a supply entity. The supply entities of all supply data are extracted using the BERT large model and labeled with the entity label u. S202. Analyze all supply data using the large model. Based on the labeled supply entities, output a fixed supply structure [entity-non-entity data-entity]. In the fixed supply structure, when there are two supply entities in the supply data and a non-entity data exists between them, the corresponding non-entity data is determined to be a supply relationship. Match the adjacent entities of each supply relationship with the extracted supply entities to generate a supply line [entity 1-supply relationship-entity 2], where entity 1 is the supply end and entity 2 is the receiving end. S203. After analyzing all supply data and generating all supply lines, graph embedding is used to convert the extracted supply entities into low-dimensional vectors. Similarity analysis is performed on the entities in the supply lines. Entities 1 in two supply lines are extracted respectively. The vector values ​​of the two entities 1 are input into the Manhattan distance formula. The distance between the two entities 1 is calculated as D(s1 i , s1 j ), s1 i Represents entity 1, s1 in supply line i j Represents entity 1 in supply line j, and the similarity is calculated using the distance between two entities 1. The formula is: ； In the formula, Sim represents the similarity between the two entities 1 in supply lines i and j. When the calculated similarity Sim = 1, the two entities 1 in supply lines i and j are judged to be the same, and the two supply lines are merged into [entity 1 - (supply relationship - entity 2) i 、(Supply relationship-entity 2) j ]; Use the same method to judge entity 2, and judge entity 1 and entity 2 in all supply lines in turn, merge the supply lines with the same entity to eliminate the ambiguity of the graph; S204. Graph the supply lines after eliminating the ambiguity in the graph, using entities as graph nodes and supply relationships as edges to construct a knowledge graph for the supply chain middle platform. S300: Collect supply data from historical supply records in the supply chain platform when anomalies occur, and use the constructed knowledge graph to find risk features that affect supply; The specific steps for finding risk features that affect supply based on the constructed knowledge graph are as follows: S301. Collect supply data from historical supply records in the supply chain platform when anomalies occur, compare all collected supply data when anomalies occur with supply data during normal supply, and determine that the supply data that differs after comparison is suspicious data; match all determined suspicious data with the supply path in the knowledge graph of the supply chain platform. When suspicious data exists continuously in the supply path in the knowledge graph, the corresponding continuously existing suspicious data is determined to be a risk feature; conversely, suspicious data that exists alone on one edge of the supply path is determined to be a non-risk feature. After this determination, all risk features in the supply data are obtained; S400: Differentiate each supply path in the knowledge graph, build a virtual supply chain sandbox for each supply path, collect supply environment data and risk characteristics of each supply path, and generate a risk warning model; The specific steps to generate a risk warning model are: S401. Differentiate all supply paths in the knowledge graph using a graph traversal algorithm, collect supply environment data and entity supply attributes for each supply path, combine the collected supply environment data and supply attributes, and construct a virtual supply sandbox for each supply path using a discrete time model. S402. Draw a curve graph using the supply environment data and risk characteristics in each supply path. Use the supply environment data as an independent variable and the risk characteristics as a variable. The drawn curve graph is segmented to obtain a function polynomial of the supply environment data and risk characteristics. The formula is: ； In the formula, Tf represents the risk characteristics, H represents the supply environment data, G represents the total number of segments in the curve graph, and f(H) k The expression of the k-th segment curve is represented, M represents the constant term; the function polynomial of each risk feature and supply environment data is calculated in turn; The average value and standard deviation of the risk characteristics of historical supply anomalies are collected, and the threshold value of each risk characteristic is obtained by subtracting the standard deviation from the average value. The early warning inequality for each risk characteristic is constructed as follows: , the early warning inequalities of all risk characteristics are integrated to obtain the risk early warning model; S500: When performing real-time supply within the supply chain platform, extract the supply entities of the real-time supply and map them in the knowledge graph to obtain the real-time supply path. Extract the real-time supply environment data and input it into the virtual supply chain sandbox of the real-time supply path. Use the risk warning model to judge the supply risk. S600: When it is determined that there is a risk in real-time supply, a risk warning is issued, the risk path is visually marked in the virtual supply chain sandbox, the risk data of the risk warning is extracted, and a risk report is generated.

2. The supply chain mid-stage data management method using large model technology according to claim 1 is characterized by: The specific steps of aligning the supply data according to the timestamp in S100 are: S101. Find the storage location of supply records in the supply chain platform, obtain different storage databases for different types of data, use the API to establish a connection with the database, collect the supply records in the supply chain platform through the connection, and extract all supply data from the supply records; S102, collect the data intervals of adjacent data in all the extracted supply data, calculate the average value of all adjacent data intervals, use the calculated average value as the interval threshold, and use the interval threshold to judge every two adjacent data intervals collected. When L a，b > Lp, it is determined that there are missing values ​​between the supply data a and b, L a，b represents the data interval between supply data a and supply data b, Lp represents the interval threshold of supply data; when it is determined that there are missing values ​​in adjacent supply data, the average value of adjacent supply data is calculated, and the formula is: , Tc represents the average value of adjacent supply data, and the filling method is used to supplement the missing values, and the average value of adjacent supply data is used as the filling value to fill in the missing values; S103. Calculate the Z-score value of each extracted supply data using the following formula: In the formula, Z represents the Z-score value of each extracted supply data, and x represents the true value of each extracted supply data. represents the average value of all supply data extracted, represents the standard deviation of all supply data extracted; Set the abnormal threshold to Zy. When the Z-score value of the calculated supply data is greater than the abnormal threshold Zy, the corresponding supply data is judged to be abnormal data and the abnormal data is deleted. S104: extract the timestamps in the supply data, unify the timestamp formats of all supply data, sort all supply data according to the timestamps, and time-align the supply data according to the sorting results.

3. The supply chain middle platform data management method using large model technology according to claim 2 is characterized by: The specific steps of using the risk warning model to judge the supply risk in S500 are: S501. When real-time supply is carried out in the supply chain middle platform, the supply entity of the real-time supply is extracted and matched in the knowledge graph to obtain the real-time supply path; the supply environment data in the real-time supply path is extracted, and the supply is simulated using the virtual supply sandbox in the real-time supply path. When the simulated supply reaches the corresponding edge, the real-time supply environment data in the corresponding edge is extracted and input into the risk warning model. When an inequality is satisfied, it is judged that there is a risk in the supply path on this edge, and a risk warning is issued.

4. The supply chain mid-stage data management method using large model technology according to claim 3 is characterized by: The specific steps of generating the risk report in S600 are: S601. When it is determined that there is a risk in real-time supply and a risk warning is issued in advance, the risk path is marked with different colors in the virtual supply sandbox, the number of risk features that meet the inequality in the risk path is calculated, and the marking color is calculated using the number of risk features as the color weight. The supply edge position with risk, the risk features that meet the inequality, and the supply environment data with risk are extracted in the virtual supply sandbox, and a risk report is generated using the extracted information.

5. The supply chain middle platform data management system using large model technology is characterized by: The supply chain middle platform data management system includes data collection module, knowledge graph construction module, risk warning model construction module, virtual sandbox warning module and report generation module; The data collection module is used to collect supply records in the supply chain platform, extract supply data from the supply records, clean and pre-process all supply data, and align the supply data according to timestamps; The knowledge graph construction module is used to process supply data using a large model, extract supply entities and supply relationships, match supply entities and supply relationships, and construct a knowledge graph for the supply chain platform; The knowledge graph construction module includes a supply entity unit, a supply relationship unit, an ambiguity elimination unit and a graph construction unit; The supply entity unit is used to analyze the supply data using the BERT large model and extract the supply entity; The supply relationship unit is used to construct a fixed structure, judge the output fixed structure, and obtain the supply relationship; The ambiguity elimination unit is used to calculate the similarity of entities in different supply lines and merge the same entities; The graph construction unit is used to graph the supply lines after eliminating the ambiguity of the graph, using entities as graph nodes and supply relationships as edges to construct a knowledge graph of the supply chain middle platform; The risk warning model building module is used to collect supply data when anomalies occur in historical supply records in the supply chain platform and extract risk features that affect supply; Collect supply environment data and risk characteristics of each supply path and generate a risk warning model; The risk warning model construction module includes a risk feature extraction unit, a function construction unit and a model construction unit; The risk feature extraction unit is used to collect supply data when anomalies occur in historical supply records in the supply chain platform, and to find risk features that affect supply in combination with the constructed knowledge graph; The function construction unit is used to analyze the collected supply environment data and risk characteristics, draw a curve graph, and generate a function polynomial; The model building unit is used to combine the function polynomials of all risk characteristics to build a risk warning model; The virtual sandbox early warning module is used to collect supply environment data in real time and input it into the risk early warning model, using the virtual supply sandbox to simulate supply and determine whether there is risk in each supply path; The report generation module is used to issue a risk warning when it is determined that there is a risk in real-time supply, visually mark the risk path in the virtual supply chain sandbox, extract risk data of the risk warning, and generate a risk report.

Images