Supply chain finance risk control methods, devices, equipment and media

Abstract

This invention relates to a method, apparatus, equipment, and medium for supply chain finance risk control. The method includes: eliminating data silos across multiple business sectors such as chemicals, electronics, and logistics through cross-supply chain data integration technology to establish a dynamic risk transmission network for real-time capture of cross-chain risk events; quantifying the intensity of risk transmission based on a dynamic update mechanism of topological relationships to address the lag problem of traditional static models; and achieving precise credit control at key nodes through the generation of policy-sensitive early warning thresholds and risk path positioning. This method effectively solves the industry pain points of delayed cross-chain risk identification caused by multi-source heterogeneous data and the failure of static assessment models to cope with real-time changes, significantly improving financial institutions' ability to predict and handle cascading bad debt risks.

Description

Technical Field

[0001] This invention relates to the field of supply chain risk control technology, specifically to a supply chain finance risk control method, device, equipment, and medium. Background Technology

[0002] Supply chain finance risk control methods, as an important component of modern financial risk control systems, aim to identify, quantify, and manage credit risk, operational risk, and market risk along the supply chain through technological means. These methods integrate transaction, logistics, and capital flow information from core enterprises and upstream and downstream SMEs to construct risk early warning models, helping financial institutions reduce bad debt rates and improve capital turnover efficiency. Especially in cross-border, multi-business supply chain scenarios, risk control must consider both industry-specific differences and dynamic transmission effects, becoming a key technological support for ensuring the sound operation of supply chain finance.

[0003] However, existing technologies have significant limitations in risk control across multiple business supply chains: on the one hand, because different supply chain systems such as chemicals, electronics, and logistics adopt independent data standards and closed architectures, data silos that are difficult to connect are formed, resulting in a delay of more than 48 hours in identifying cross-chain risk transmission events (such as a strike in a certain place causing a shortage of chemical raw materials, which in turn affects the electronic product production chain); on the other hand, traditional static assessment models cannot capture the dynamic impact of real-time factors such as policy adjustments and market fluctuations on the risk transmission path, making it difficult for financial institutions to adjust their credit strategies in a timely manner, resulting in delayed risk disposal or even triggering a chain reaction of bad debt crises. Summary of the Invention

[0004] Based on this, the purpose of the present invention is to provide a supply chain finance risk control method, device, equipment and medium that can capture the risk transmission path across the supply chain in real time and dynamically respond to changes in the policy environment.

[0005] The objective of this invention is achieved through the following solution:

[0006] In a first aspect, the present invention provides a supply chain finance risk control method, comprising the following steps:

[0007] S1: Acquire multi-source heterogeneous data from multiple independent supply chains, and convert the JSON transaction flow, XML logistics data and Protobuf capital flow of the multi-source heterogeneous data into a unified vector format through a preset industry pre-trained model, generating a standardized data flow with source identifiers;

[0008] S2: Construct a dynamic risk transmission network based on standardized data flow, establish an initial graph structure based on the supply chain tags of node feature vectors and source identifiers, and update the topology in real time through time decay factor to generate a labeled dynamic graph;

[0009] S3: Based on the node labels of the labeled dynamic graph, separate the subgraphs of each independent supply chain, calculate the maximum path risk value between nodes and the cross-supply chain transmission matrix, and generate the risk transmission matrix and the maximum risk path set;

[0010] S4: Adaptive early warning is performed on the risk transmission matrix and the maximum risk path set. Dynamic early warning thresholds are generated by analyzing the statistical characteristics of the pre-stored historical risk transmission matrix and real-time policy text. When the risk value in the risk transmission matrix exceeds the dynamic early warning threshold, the key nodes in the risk transmission path are located and risk disposal instructions are generated. The risk disposal instructions are used to instruct financial institutions to adjust the credit limits of related enterprises at key nodes.

[0011] In one embodiment, S1 of the supply chain finance risk control method provided by the present invention specifically includes the following steps:

[0012] S11: Acquire multi-source heterogeneous data from multiple independent supply chains, perform field parsing on the JSON transaction stream of the multi-source heterogeneous data, extract key fields such as transaction amount, enterprise ID and payment period, and generate a structured transaction dataset;

[0013] S12: Perform multi-level tag parsing on the XML logistics data of multi-source heterogeneous data, convert GPS coordinates, cargo status and temperature and humidity monitoring values ​​into a numerical matrix, and generate a standardized logistics data matrix.

[0014] S13: Perform protocol deserialization on the Protobuf fund flow of multi-source heterogeneous data, decode the binary streams of account changes, credit limits and settlement status, and generate a time-series fund flow sequence;

[0015] S14: Perform cross-modal feature fusion processing on structured transaction datasets, standardized logistics data matrices, and time-series capital flow sequences. Through industry pre-trained models, map transaction amount features, logistics coordinate features, and capital change features to a unified vector space, and add original data source identifiers to generate standardized data streams with source identifiers.

[0016] In one embodiment, S2 of the supply chain finance risk control method provided by the present invention specifically includes the following steps:

[0017] S21: Perform similarity calculation on the node feature vectors of the standardized data stream, quantify the directional consistency between feature vectors of different enterprises through the cosine similarity algorithm, and generate an initial set of edge weights;

[0018] S22: Perform graph node labeling on the supply chain tags corresponding to the source identifiers in the standardized data stream, assign classification tags to the nodes according to the three types of source identifiers: chemical chain, electronic chain, and logistics chain, and generate a node set with classification tags;

[0019] S23: Perform graph structure construction processing on the initial set of edge weights and the set of nodes with classification labels to establish a directed topological network with enterprise entities as nodes and similarity weights as edges, and generate the initial graph structure.

[0020] S24: Perform dynamic topology update processing on the initial graph structure, introduce a time decay factor to reduce the edge weights of historical transactions exponentially, and strengthen the new connection relationships reflected by real-time data to generate a labeled dynamic graph.

[0021] In one embodiment, the adjustment formula for the edge weights of historical transaction associations in a supply chain finance risk control method provided by the present invention is as follows:

[0022]

[0023] in, To adjust the edge weights at the current time t, Initial time edge weights, Industry-specific attenuation rate, As the impact factor of real-time data, The edge weight increment is calculated for real-time data.

[0024] In one embodiment, S3 of the supply chain finance risk control method provided by the present invention specifically includes the following steps:

[0025] S31: Perform supply chain subgraph separation processing on the node labels of the tagged dynamic graph, divide the enterprise node subsets according to the chemical, electronic and logistics labels, and generate a supply chain subgraph set;

[0026] S32: Perform full-path search processing on cross-supply chain node pairs, traverse cross-chain connection paths such as from chemical subgraph to electronic subgraph and from chemical subgraph to logistics subgraph, and generate a set of potential risk paths;

[0027] S33: Perform weight decay calculation on the potential risk path set, dynamically adjust the edge weight product value based on the topological distance, and generate a set of path risk intensity values;

[0028] S34: Perform matrix processing on the set of path risk intensity values, aggregate the cross-chain risk transmission intensity according to the supply chain relationship, and generate a risk transmission matrix and the set of the maximum risk paths.

[0029] In one embodiment, the calculation formula for the path risk intensity value set of a supply chain finance risk control method provided by the present invention is as follows:

[0030]

[0031]

[0032] in, The risk intensity value for path p. The Sigmoid normalization function, Let p be the number of edges on path p. For the decay edge weight function, Let the original weight of the k-th edge be . Let be the shortest path hop count between the two ends of the k-th edge in path p. This is the distance attenuation coefficient.

[0033] In one embodiment, step S4 of the supply chain finance risk control method provided by the present invention specifically includes the following steps:

[0034] S41: Perform statistical feature extraction on the pre-stored historical risk transmission matrix, calculate the mean and standard deviation of the 30-day sliding window as the benchmark for risk transmission intensity, and generate historical statistical features;

[0035] S42: Perform semantic parsing on the central bank's policy texts acquired in real time, quantify the policy's severity and timeliness indicators through a legal text model, and generate real-time policy risk factors;

[0036] S43: Threshold fusion processing is performed on historical statistical characteristics and real-time policy risk factors, and the risk transmission intensity benchmark and the impact of policy adjustments are weighted and integrated to generate dynamic early warning thresholds;

[0037] S44: Perform over-limit detection processing on the risk values ​​and dynamic early warning thresholds in the risk transmission matrix, locate high-risk supply chain transmission pairs that exceed the thresholds, and generate a set of high-risk transmission pairs;

[0038] S45: Perform key node location processing on the set of the largest risk paths, identify core enterprise nodes based on the risk transmission path topology, and generate a list of key nodes of risk sources.

[0039] S46: Generate credit instructions for the list of key nodes of risk sources, and generate tiered credit limit adjustment schemes by combining the location of key nodes in the supply chain, the intensity of risk transmission and the credit rules of financial institutions, and generate risk disposal instructions.

[0040] Secondly, the present invention provides a supply chain finance risk control device, which is configured with the following modules:

[0041] The multi-source heterogeneous data standardization module is used to acquire multi-source heterogeneous data from multiple independent supply chains. Through a preset industry pre-trained model, it converts the JSON transaction flow, XML logistics data, and Protobuf capital flow of multi-source heterogeneous data into a unified vector format, generating a standardized data flow with source identifiers.

[0042] The dynamic risk transmission network construction module is used to construct a dynamic risk transmission network based on standardized data flow. It establishes an initial graph structure based on the node feature vector and the supply chain label of the source identifier, and updates the topology in real time through the time decay factor to generate a labeled dynamic graph.

[0043] The supply chain subgraph separation and risk calculation module is used to separate the subgraphs of each independent supply chain based on the node labels of the labeled dynamic graph, calculate the maximum path risk value between nodes and the cross-supply chain transmission matrix, and generate the risk transmission matrix and the maximum risk path set.

[0044] The adaptive risk warning and handling instruction generation module is used to adaptively warn based on the risk transmission matrix and the maximum risk path set. It generates dynamic warning thresholds by analyzing the statistical characteristics of the pre-stored historical risk transmission matrix and real-time policy text. When the risk value in the risk transmission matrix exceeds the dynamic warning threshold, it locates the key nodes in the risk transmission path and generates risk handling instructions. The risk handling instructions are used to instruct financial institutions to adjust the credit limits of related enterprises at key nodes.

[0045] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the above-mentioned supply chain finance risk control methods.

[0046] Fourthly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-mentioned supply chain finance risk control methods.

[0047] In summary, the supply chain finance risk control method provided in this application achieves a breakthrough upgrade of the traditional risk control system through multi-dimensional technological innovation. This method effectively breaks down data barriers between independent supply chain systems such as chemicals, electronics, and logistics. By using cross-modal feature fusion technology to convert heterogeneous data into a unified vector space, it eliminates data silos. The risk transmission network built based on a dynamic topology update mechanism can capture cross-chain risk transmission paths in real time, enabling early warning of risk events. A dynamic threshold generation mechanism that integrates historical statistical features and real-time policy analysis can accurately identify high-risk transmission scenarios. The key node positioning capability of path topology analysis enables precise control of credit strategies. This method significantly improves the ability to predict cascading risks, allowing financial institutions to take intervention measures in the early stages of risk transmission and prevent the spread of systemic financial risks. Simultaneously, the policy-sensitive early warning mechanism enhances the system's adaptability to changes in the regulatory environment, enabling automatic adjustment of compliance risk control strategies. Ultimately, a complete closed loop from risk identification and quantification to disposal is formed, significantly reducing bad debt losses caused by delayed risk transmission and providing stable and reliable risk control guarantees for multi-business supply chain finance.

[0048] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0049] Figure 1 A flowchart illustrating a supply chain finance risk control method provided in this application embodiment;

[0050] Figure 2 A schematic diagram illustrating the process of generating the risk transmission matrix and the maximum risk path set provided in the embodiments of this application;

[0051] Figure 3 This is a schematic diagram of a supply chain finance risk control device provided in another embodiment of this application. Detailed Implementation

[0052] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0054] In one embodiment, such as Figure 1 As shown, a supply chain finance risk control method is provided. This embodiment illustrates the method applied to a terminal, but it is understood that the method can also be applied to a server, or to a device including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0055] S1: Acquire multi-source heterogeneous data from multiple independent supply chains, and convert the JSON transaction flow, XML logistics data, and Protobuf capital flow of the multi-source heterogeneous data into a unified vector format through a preset industry pre-trained model, generating a standardized data flow with source identifiers.

[0056] Specifically, the system collects multi-source heterogeneous data from multiple independent supply chains. Data types include JSON transaction flows, XML logistics data, and Protobuf financial flows. Data sources cover ERP systems of core supply chain enterprises and upstream and downstream SMEs, e-commerce platforms, warehousing and transportation management systems of logistics service providers, customs supervision systems in cross-border scenarios, core business systems of partner banks, third-party payment platforms, and financing management systems of financial institutions. The system employs adaptive collection methods for different data formats to ensure complete data reception. The system calls a pre-trained industry model, trained on historical heterogeneous data from various industries, which has field extraction and feature mapping capabilities. The system inputs the collected heterogeneous data into the model. The model first parses the structured fields of each data format, then performs feature engineering to generate risk-related derived features, and subsequently maps the original features and derived features into vectors of a unified dimension. The system adds source identifiers to the generated vectors, including information on the industry, supply chain, and data type, ultimately forming a standardized data stream.

[0057] S2: Construct a dynamic risk transmission network based on standardized data flow, establish an initial graph structure based on node feature vectors and supply chain labels with source identifiers, and update the topology in real time through time decay factors to generate a labeled dynamic graph.

[0058] Specifically, the system first extracts information about the supply chain participants from the data stream, treating each participant as a network node and assigning a unique identifier to each node. Simultaneously, it associates node attributes with feature vectors and source identifiers in the standardized data stream. Node attributes include basic and type information of the participant. The system constructs network edges based on the business relationships between participants, including transaction relationships, logistics collaboration relationships, financial transactions, and cross-supply chain collaboration relationships. The system calculates the initial weight of each edge based on the stability and importance of these relationships. A time decay factor is introduced, calculated based on the most recent business interaction time of the relationship, to adjust the edge weights. The system sets a fixed time window, scanning the standardized data stream within each window, updating node attributes and edge weights, deleting edges with weights below a set threshold, and adding nodes and edges corresponding to new business relationships. Finally, a labeled dynamic graph is generated, with labels containing information about the node's industry, supply chain, and type to reflect the dynamic changes in supply chain relationships.

[0059] S3: Based on the node labels of the labeled dynamic graph, separate the subgraphs of each independent supply chain, calculate the maximum path risk value between nodes and the cross-supply chain transmission matrix, and generate the risk transmission matrix and the maximum risk path set.

[0060] Specifically, the system traverses all nodes in the dynamic graph, filters out nodes with the same supply chain information and the internal edges between nodes, forming subgraphs for each supply chain. It also extracts and stores cross-supply chain edges separately. The system validates the separated subgraphs using metrics including the number of nodes and edge density. Subgraphs that fail validation are marked and added to a separate monitoring queue. The system calculates the initial risk value for each node. This risk value is generated based on the node's business feature vector using a pre-defined model, reflecting the node's inherent risk level.

[0061] Preferably, the system can employ an improved algorithm to calculate the maximum path risk value between nodes within a subgraph. The algorithm uses edge weights as transmission coefficients, and the path risk value is the sum of the products of the node risk value and the corresponding edge weight on the path. It then filters out the paths with the highest risk values ​​between each pair of nodes, forming a set of paths with maximum risk. The system constructs a cross-supply chain transmission matrix, with dimensions consistent with the total number of supply chains. Matrix elements are calculated using the weights of cross-supply chain related edges, node risk values, and industry correlation coefficients to reflect the intensity of risk transmission between different supply chains, ultimately generating a risk transmission matrix and a set of paths with maximum risk.

[0062] S4: Adaptive early warning is performed on the risk transmission matrix and the maximum risk path set. Dynamic early warning thresholds are generated by analyzing the statistical characteristics of the pre-stored historical risk transmission matrix and real-time policy text. When the risk value in the risk transmission matrix exceeds the dynamic early warning threshold, the key nodes in the risk transmission path are located and risk disposal instructions are generated. The risk disposal instructions are used to instruct financial institutions to adjust the credit limits of related enterprises at key nodes.

[0063] Specifically, the system extracts a pre-stored historical risk transmission matrix and uses a sliding window to calculate the statistical characteristics of historical data, including mean, variance, and high-risk quantiles. The system acquires real-time policy texts from industry policies, cross-border trade policies, and financial regulatory policies issued by regulatory agencies. It calls a pre-trained model to parse the policy texts, extracting the industries involved and adjustment requirements, calculating the impact factors of the policies on the supply chains of each industry, and generating dynamic early warning thresholds by combining historical statistical characteristics, impact factors, and the risk preference coefficients of financial institutions. When the system detects that the risk value in the risk transmission matrix exceeds the dynamic early warning threshold, it initiates a key node location process, using an improved algorithm to calculate the risk transmission contribution of each node and selecting the nodes with the highest contribution ranking as key nodes. Based on this, the system generates risk disposal instructions in a standardized format, including instruction identifier, risk level, key node information, credit adjustment requirements, and execution time limit. The system pushes the instructions to the financial institution's risk control system through an encrypted channel, receives the execution results returned by the financial institution, and stores the execution results in association with the updated results of the risk transmission matrix, forming a closed loop for risk disposal and providing a basis for financial institutions to adjust their credit strategies.

[0064] In summary, the supply chain finance risk control method provided in this application achieves a breakthrough upgrade of the traditional risk control system through multi-dimensional technological innovation. This method effectively breaks down data barriers between independent supply chain systems such as chemicals, electronics, and logistics. By using cross-modal feature fusion technology to convert heterogeneous data into a unified vector space, it eliminates data silos. The risk transmission network built based on a dynamic topology update mechanism can capture cross-chain risk transmission paths in real time, enabling early warning of risk events. A dynamic threshold generation mechanism that integrates historical statistical features and real-time policy analysis can accurately identify high-risk transmission scenarios. The key node positioning capability of path topology analysis enables precise control of credit strategies. This method significantly improves the ability to predict cascading risks, allowing financial institutions to take intervention measures in the early stages of risk transmission and prevent the spread of systemic financial risks. Simultaneously, the policy-sensitive early warning mechanism enhances the system's adaptability to changes in the regulatory environment, enabling automatic adjustment of compliance risk control strategies. Ultimately, a complete closed loop from risk identification and quantification to disposal is formed, significantly reducing bad debt losses caused by delayed risk transmission and providing stable and reliable risk control guarantees for multi-business supply chain finance.

[0065] In one embodiment, S1 of the supply chain finance risk control method provided by the present invention specifically includes the following steps:

[0066] S11: Acquire multi-source heterogeneous data from multiple independent supply chains, perform field parsing on the JSON transaction stream of the multi-source heterogeneous data, extract key fields such as transaction amount, enterprise ID and payment period, and generate a structured transaction dataset.

[0067] Specifically, for JSON transaction streams from multi-source heterogeneous data, the system initiates a field parsing process, calls a JSON parser, and parses the JSON transaction stream layer by layer according to preset field mapping rules to locate and extract three key fields: transaction amount, enterprise ID, and payment period. During the parsing process, the system performs data validity checks on the extracted fields, verifying whether the data type of the transaction amount field conforms to the numerical specifications, confirming whether the encoding format of the enterprise ID field is consistent with the preset enterprise identification rules, and verifying whether the time representation of the payment period field meets the date format requirements.

[0068] For data with missing fields or abnormal formatting encountered during parsing, the system marks it as pending data and temporarily stores it in the abnormal data buffer, while recording the type and location of the abnormality. After completing the extraction and verification of key fields, the system organizes the valid data into a structured table format according to a fixed field order of "Enterprise ID-Transaction Amount-Payment Period," generating a structured transaction dataset. This dataset uses rows as individual transaction records and columns corresponding to the extracted key fields, supporting data retrieval by enterprise ID or transaction time.

[0069] S12: Perform multi-level tag parsing on the XML logistics data from multiple heterogeneous sources, convert GPS coordinates, cargo status, and temperature and humidity monitoring values ​​into a numerical matrix, and generate a standardized logistics data matrix.

[0070] Specifically, the system employs a multi-level tag parsing mechanism for XML logistics data from multi-source heterogeneous data. The system loads an XML parser and, based on the tag hierarchy of the XML logistics data, traverses layer by layer starting from the top-level tag, sequentially parsing sub-tags containing GPS coordinates, cargo status, and temperature and humidity monitoring values. For GPS coordinate data, the system converts the original latitude and longitude text descriptions into a unified coordinate numerical format, ensuring that the coordinate accuracy meets preset standards. For cargo status data, the system converts the text-based cargo status descriptions into corresponding numerical codes according to a preset status mapping table, achieving a quantitative representation of cargo status. For temperature and humidity monitoring values, the system performs range verification on the original monitoring data, removing outliers exceeding reasonable monitoring ranges while retaining the original numerical accuracy. After completing the parsing and conversion of these three types of data, the system constructs a two-dimensional numerical matrix with each logistics record as a row and GPS coordinates (longitude and latitude), cargo status codes, temperature values, and humidity values ​​as columns, generating a standardized logistics data matrix. Each row in the matrix corresponds to the core monitoring data of a single logistics record, and each column corresponds to a type of quantified logistics feature. The matrix dimensions are dynamically adjusted according to the number of logistics records to meet the matrix operation requirements for subsequent feature calculations.

[0071] S13: Perform protocol deserialization processing on the Protobuf fund flow of multi-source heterogeneous data, decode the binary stream of account changes, credit limit and settlement status, and generate a time-series fund flow sequence.

[0072] Specifically, the system performs protocol deserialization processing on Protobuf fund flows from multi-source heterogeneous data. The system loads the protocol definition file corresponding to the Protobuf fund flow, clarifying the storage structure and encoding rules for three types of data in the binary stream: account changes, credit limits, and settlement status. Based on the protocol definition, the system can call the Protobuf deserialization interface to decode the binary stream of fund flows, sequentially extracting the amount and direction information of account changes, the current and available values ​​of the credit limit, and the text description of the settlement status. During the decoding process, the system performs integrity checks on the decoded data, confirming whether the account change amount and credit limit values ​​conform to the numerical range of the fund data, and checking for any ambiguous statements in the settlement status description.

[0073] After successful verification, the system adds a collection timestamp to each decoded fund data entry and sorts the fund data according to the order of the timestamps, forming a time-based sequence structure to generate a time-series fund flow sequence. Each data point in this sequence includes a timestamp, account change information, credit limit information, and settlement status information, and supports data slicing by time interval to reflect the dynamic changes in fund flow.

[0074] S14: Perform cross-modal feature fusion processing on structured transaction datasets, standardized logistics data matrices, and time-series capital flow sequences. Through industry pre-trained models, map transaction amount features, logistics coordinate features, and capital change features to a unified vector space, and add original data source identifiers to generate standardized data streams with source identifiers.

[0075] Specifically, the system performs cross-modal feature fusion processing on structured transaction datasets, standardized logistics data matrices, and time-series capital flow sequences. The system calls an industry-pre-trained model to first normalize the transaction amount features in the structured transaction dataset, mapping them to feature values ​​within a fixed range. For the logistics coordinate features in the standardized logistics data matrix, a coordinate projection algorithm is used to convert them into low-dimensional feature vectors. For the capital change features in the time-series capital flow sequence, a time-series feature extraction method is used to extract trend and fluctuation features. Subsequently, the industry-pre-trained model maps the processed transaction amount features, logistics coordinate features, and capital change features to a unified high-dimensional vector space, achieving spatial alignment of different modal features. After feature mapping, the system adds an original data source identifier to each fused feature vector. This identifier includes the supply chain number to which the original data belongs, the data type (transaction / logistics / capital), and the data collection batch information, ensuring that the feature vector is traceable to the original data.

[0076] In one embodiment, S2 of the supply chain finance risk control method provided by the present invention specifically includes the following steps:

[0077] S21: Perform similarity calculation on the node feature vectors of the standardized data stream, quantify the directional consistency between feature vectors of different enterprises using the cosine similarity algorithm, and generate an initial set of edge weights.

[0078] Specifically, the system extracts node feature vectors for each enterprise from a standardized data stream with source identifiers. These vectors consist of cross-modal fusion features of transaction amount, logistics coordinates, and capital changes. The system then uses a cosine similarity algorithm to calculate the similarity between the node feature vectors of any two enterprises, following the formula:

[0079]

[0080] in, and These represent the node feature vectors of the two companies, respectively. Represents the dot product of two vectors. and Let them represent the magnitudes of the two vectors, respectively. The result is the similarity value of the feature vectors of the two companies. The system performs range calibration on the calculated similarity value, correcting values ​​that exceed the 0 to 1 range to within that range, and eliminating outliers caused by calculation errors.

[0081] Preferably, the system combines each pair of enterprise identifiers into an enterprise ID pair, and associates the enterprise ID pairs with their corresponding similarity values ​​to form initial edge weight data units. All initial edge weight data units are organized according to a fixed structure of "enterprise ID pair - similarity weight" to generate an initial edge weight set. This set is stored in the data storage module in key-value pair format, where the key is the enterprise ID pair and the value is the corresponding similarity weight. The system supports querying all edge weight data associated with a given enterprise by its enterprise ID.

[0082] S22: Perform graph node labeling on the supply chain tags corresponding to the source identifiers in the standardized data stream. Assign classification tags to the nodes according to the three types of source identifiers: chemical chain, electronic chain, and logistics chain, and generate a node set with classification tags.

[0083] Specifically, the system parses the source identifier corresponding to the feature vector of each node in the standardized data stream, extracts the supply chain number contained in the source identifier, and this supply chain number is directly associated with the original supply chain affiliation of the multi-source heterogeneous data. The system performs label allocation according to preset supply chain label mapping rules, and the mapping process follows the rules:

[0084]

[0085] in, Represents category tags, The supply chain number in the source identifier represents the source. Represents the mapping function. When the supply chain number... When conforming to chemical chain coding rules, the mapping function Output "Chemical Chain" category label; when supply chain number When conforming to the electronic chain coding rules, the mapping function Output "Electronic Chain" category label; when the supply chain number... When conforming to the logistics chain coding rules, the mapping function Output the "Logistics Chain" category label.

[0086] The system will classify the tags The system integrates the data with corresponding enterprise node information, which includes the enterprise ID and node feature vector, to form a complete node data unit. All node data units are sorted in ascending order by enterprise ID, generating a node set with category labels. This node set is stored in a structured table format, with columns for enterprise ID, node feature vector, and category label. The system supports filtering all node data within the same supply chain using category labels.

[0087] S23: Perform graph structure construction processing on the initial set of edge weights and the set of nodes with classification labels to establish a directed topological network with enterprise entities as nodes and similarity weights as edges, and generate the initial graph structure.

[0088] Specifically, the system loads the initial set of edge weights and the set of nodes with classification labels, initiating the graph structure construction process. The system first defines each enterprise entity in the node set with classification labels as a node in the graph structure. Node attributes include enterprise ID, node feature vector, and classification label, and each node is assigned a unique node identifier. The system determines the connection relationships between nodes based on the enterprise IDs in the initial set of edge weights, and simultaneously defines the direction of the edges, with the determination of edge direction following certain rules. ,in Represents the direction of the edge. This represents the business association relationship corresponding to the enterprise ID. Represents the direction function. When business relationships... When establishing a transaction initiation and reception relationship, the direction function... Define the direction of the edge as the transaction initiating enterprise node pointing to the transaction receiving enterprise node; when business relationships... When there is a logistics consignment and service relationship, the direction function Define the direction of the edges as from the logistics consignor node to the logistics service provider node. Set the weight of each edge to the similarity weight of the corresponding enterprise ID pair in the initial edge weight set.

[0089] The system constructs a directed topology network based on the "node-edge-weight" association logic. The node set consists of all enterprise nodes in the node set with classification labels, the edge set consists of the connections between all enterprise ID pairs in the initial edge weight set, and the edge weight set consists of the corresponding similarity values. The system stores the constructed directed topology network in a graph database, generating an initial graph structure. This structure includes a node list, an edge list, and an edge weight list. The system supports node attribute queries, edge weight queries, and path traversal operations on this structure, providing an initial network framework for subsequent dynamic topology updates.

[0090] S24: Perform dynamic topology update processing on the initial graph structure, introduce a time decay factor to reduce the edge weights of historical transactions exponentially, and strengthen the new connection relationships reflected by real-time data to generate a labeled dynamic graph.

[0091] Specifically, the system retrieves the initial graph structure from the graph database and obtains the initial weights of all edges. and initial time Simultaneously, it collects standardized data streams updated in real time, extracts node feature vectors corresponding to newly added business associations, and calculates real-time edge weight increments using a cosine similarity algorithm. The system introduces a time decay factor and adjusts edge weights according to a formula. The adjustment formula for edge weights associated with historical transactions is as follows:

[0092]

[0093] in, To adjust the edge weights at the current time t, Initial time edge weights, Industry-specific attenuation rate, As the impact factor of real-time data, The edge weight increment is calculated for real-time data when adding new business associations. Take a positive value when the service connection is interrupted. Take the negative value. The system performs weight decay calculations on all historical edges in the initial graph structure and updates the edge weights to the new values. Simultaneously, add new edges to the node pairs corresponding to the newly added business associations, and set the weight of the new edges to [value missing]. For node pairs whose business connections are interrupted, configure... The system reduces the weight of the corresponding edge for negative values, and deletes the edge when its weight falls below a preset threshold. The system integrates the adjusted node set, edge set, edge weight set, and node classification labels to generate a labeled dynamic graph. This dynamic graph reflects changes in the relationships between nodes in real time. The system stores this graph in a graph database, overwriting the initial graph structure, providing real-time topology network support for subsequent risk transmission analysis.

[0094] In one embodiment, S3 of the supply chain finance risk control method provided by the present invention specifically includes the following steps:

[0095] S31: Perform supply chain subgraph separation processing on the node labels of the tagged dynamic graph, divide the enterprise node subsets according to the chemical, electronic and logistics labels, and generate a supply chain subgraph set.

[0096] Specifically, the system reads all node data from the labeled dynamic graph and extracts the corresponding category label for each node, which is one of three assigned labels: chemical, electronics, and logistics. The system groups the nodes according to label type, filtering out all enterprise nodes labeled "chemical" to form a chemical node subset; filtering out all enterprise nodes labeled "electronics" to form an electronics node subset; and filtering out all enterprise nodes labeled "logistics" to form a logistics node subset. For each node subset, the system extracts all edge connections and corresponding edge weights between nodes within that subset from the labeled dynamic graph, constructing an independent subgraph structure: the chemical node subset, along with its corresponding edges and weights, forms a chemical subgraph; the electronics node subset, along with its corresponding edges and weights, forms an electronics subgraph; and the logistics node subset, along with its corresponding edges and weights, forms a logistics subgraph. Furthermore, the system integrates the chemical, electronics, and logistics subgraphs to generate a supply chain subgraph set. Each subgraph in this set contains a node list, an edge list, and an edge weight list, and the node labels within each subgraph are unified, supporting path analysis within the subgraph and cross-subgraph association queries.

[0097] S32: Perform full-path search processing on cross-supply chain node pairs, traverse cross-chain connection paths such as from the chemical subgraph to the electronic subgraph and from the chemical subgraph to the logistics subgraph, and generate a set of potential risk paths.

[0098] Specifically, the system identifies cross-supply chain node pairs from the supply chain subgraph set. These are node pairs where the source node belongs to a different subgraph than the target node, including three types of cross-chain node pairs: chemical subgraph node and electronics subgraph node, chemical subgraph node and logistics subgraph node, and electronics subgraph node and logistics subgraph node. The system employs a depth-first search algorithm to perform a full path search for each type of cross-chain node pair: starting from the source node and ending at the target node, it traverses all possible edge connection paths between the source and target subgraphs. Paths are allowed to contain nodes from other subgraphs as intermediate nodes, but the paths must be continuous and free of loops. During the search, the system records the node sequence, edge sequence, and corresponding edge weights for each path, eliminating paths whose length exceeds the preset maximum number of edges (the maximum number of edges is determined based on the number of supply chain levels). The system then organizes all eligible cross-chain paths to generate a set of potential risk paths. Each path in this set contains a source node identifier, a target node identifier, a node sequence, an edge sequence, and a list of edge weights.

[0099] S33: Perform weight decay calculation on the set of potential risk paths, dynamically adjust the edge weight product value based on the topological distance, and generate a set of path risk intensity values.

[0100] Specifically, the system loads a set of potential risk paths and extracts the edge sequence and corresponding edge weights for each path p. Simultaneously calculate the shortest path hop count between the two endpoints of each edge in the path. , This refers to the minimum number of edge connections between two nodes in a labeled dynamic graph. The system first calculates the decaying edge weight of each edge according to the formula. The formula is:

[0101]

[0102] in, Let the weight of the k-th edge be the decaying edge weight. Let the original weight of the k-th edge be . Let be the number of hops in the shortest path between the two endpoints of the k-th edge. This is the distance attenuation coefficient. Determined by the average node spacing in the supply chain network, it is used to adjust the degree of influence of distance on edge weight decay.

[0103] Furthermore, the system performs a product operation on the weights of all decaying edges in path p, and then normalizes the result using the Sigmoid function. After processing, the risk intensity value PathRisk(p) for path p is obtained, and the calculation formula is as follows:

[0104]

[0105] Wherein, PathRisk(p) is the risk intensity value of path p. Sigmoid normalization function This is used to map risk intensity values ​​to a range of 0 to 1. Let be the number of edges on path p, and ∏ represent the number of edges on path k from 1 to 1. of Perform a series of multiplication operations. The system performs the above calculations on each path in the potential risk path set, associates the path identifier with the corresponding risk intensity value, and generates a set of path risk intensity values.

[0106] S34: Perform matrix processing on the set of path risk intensity values, aggregate the cross-chain risk transmission intensity according to the supply chain relationship, and generate a risk transmission matrix and the set of the maximum risk paths.

[0107] Specifically, the system groups the path risk intensity value set according to supply chain pairs. These supply chain pairs include three categories: chemical-electronics, chemical-logistics, and electronics-logistics. Each supply chain pair contains all cross-chain paths where the source subgraph is the former and the target subgraph is the latter. The system calculates the average risk intensity value of all paths under each type of supply chain pair and uses this average value as the element value of the corresponding position in the risk transmission matrix to construct the risk transmission matrix: matrix rows correspond to source supply chain types, matrix columns correspond to target supply chain types, and matrix element values ​​represent the average cross-chain risk transmission intensity of the corresponding source-target supply chain pair. Simultaneously, the system sorts the path risk intensity values ​​under each type of supply chain pair in descending order and filters the top-ranked paths. The number of filtered paths is determined based on the risk monitoring accuracy requirements; these paths have risk intensity values ​​higher than the average intensity value of the same type of supply chain pair. The system integrates the filtered paths with the risk transmission matrix to generate a risk transmission matrix and a set of maximum risk paths. The set of maximum risk paths includes path identifiers, source-target supply chain pairs, risk intensity values, and path structure information, and supports queries by supply chain pair or risk intensity value.

[0108] In one embodiment, step S4 of the supply chain finance risk control method provided by the present invention specifically includes the following steps:

[0109] S41: Perform statistical feature extraction on the pre-stored historical risk transmission matrix, calculate the mean and standard deviation of the 30-day sliding window as the benchmark for risk transmission intensity, and generate historical statistical features.

[0110] Specifically, the system calls the historical database interface to retrieve a pre-stored risk transmission matrix sequence and sorts the matrix data in ascending order by timestamp. The system sets a sliding window length of 30 days and calculates the mean and standard deviation for all matrix elements within each window. Preferably, the mean calculation formula is:

[0111]

[0112] in, This represents the average intensity of risk transmission within the window. Represents the total number of source supply chains. Represents the total number of target supply chains. Represents the number of time steps within the window. This represents the risk transmission value from source supply chain i to target supply chain j at time t. The standard deviation is calculated using the following formula:

[0113]

[0114] in, This represents the standard deviation of the risk transmission intensity within the window. The system will assign the corresponding standard deviation to each sliding window. and The timestamp of the end of the associated window is compiled into historical statistical features.

[0115] S42: Perform semantic parsing on the central bank's policy texts acquired in real time, quantify the policy's severity and timeliness indicators through a legal text model, and generate real-time policy risk factors.

[0116] Specifically, the system obtains policy texts released by the central bank in real time through an encrypted data interface, performs text preprocessing to remove punctuation, redundant line breaks, and non-policy-related expressions, and standardizes the text encoding format. The system then loads a pre-trained legal text model to perform semantic parsing on the preprocessed policy text, extracting restrictive clauses, regulatory requirements, and effective dates related to supply chain finance. The system quantifies policy stringency indicators. This indicator reflects the degree of restriction imposed by policies on supply chain financing activities, and its value ranges from 0 to 1; it quantifies the timeliness of policies. This indicator reflects the overlap between the policy's effective period and the current time, and its value ranges from 0 to 1. The system generates the real-time policy risk factor using the following formula. :

[0117]

[0118] in, The severity level is determined by the policy type; for example, credit control policies have a higher severity level. Support policies correspond to lower levels The system will It is associated with the policy release time and policy number and stored in the policy factor database.

[0119] S43: Threshold fusion processing is performed on historical statistical characteristics and real-time policy risk factors, and the risk transmission intensity benchmark and the impact of policy adjustments are weighted and integrated to generate dynamic early warning thresholds.

[0120] Specifically, the system retrieves the latest 30-day sliding window data from the historical statistical feature database. and Retrieve currently effective real-time policy risk factors from the policy factor database. The system first calculates the initial threshold based on the risk transmission intensity benchmark, using the following formula:

[0121]

[0122] in, This is a risk sensitivity coefficient, set by financial institutions based on their own risk preferences. The system then performs threshold fusion processing using a weighted formula to generate dynamic early warning thresholds. :

[0123]

[0124] in, The baseline weighting coefficient, ranging from 0 to 1, is used to balance the impact of historical data baselines and real-time policies. It is reduced when the policy impact is significant. Increase when the policy has a weak impact The system for Apply range constraints to ensure that it does not fall below the preset minimum warning threshold, and then... Store in the threshold management library.

[0125] S44: Perform over-limit detection processing on the risk values ​​and dynamic early warning thresholds in the risk transmission matrix, locate high-risk supply chain transmission pairs that exceed the thresholds, and generate a set of high-risk transmission pairs.

[0126] Specifically, the system retrieves the real-time generated risk transmission matrix from the risk transmission matrix library and the current dynamic early warning threshold from the threshold management library. The system iterates through the risk transmission matrix row by row, processing each element in the matrix. (i.e., the risk transmission value from source supply chain i to target supply chain j) is used for over-limit detection to determine... Is it greater than .like Greater than The system marks the supply chain pair (source supply chain i, target supply chain j) corresponding to this element as a high-risk transmission pair, and records... The system collects all marked high-risk transmission pairs, including specific numerical values, detection times, and corresponding risk path identifiers. The values ​​are sorted from largest to smallest to generate a set of high-risk transmission pairs. The system synchronizes this set to the risk warning module, triggering an early warning, and simultaneously stores it in the risk event database to define the risk range for subsequent key node location.

[0127] S45: Perform key node location processing on the set of the largest risk paths, identify core enterprise nodes based on the risk transmission path topology, and generate a list of key nodes of risk sources.

[0128] Specifically, the system retrieves all path data associated with high-risk transmission pairs from the maximum-risk path set library. Each path data includes a node sequence, edge weights, and path risk intensity values. The system calculates the betweenness centrality of each node based on the path topology, using the following formula:

[0129]

[0130] in, Represents the betweenness centrality of node v. This represents the total number of shortest paths from node s to node t. This represents the number of shortest paths from node s to node t via node v. The system filters nodes with betweenness centrality greater than a preset threshold and further confirms whether these nodes are core enterprises in the supply chain (such as enterprises with a dominant position in the industry chain) by using the enterprise type identifier in the node attributes. The system collects core enterprise nodes that meet the criteria, records their supply chain name, associated risk path identifier, and risk contribution percentage, generates a list of key risk source nodes, stores it in the key node database, and associates it with the corresponding high-risk transmission pair information.

[0131] S46: Generate credit instructions for the list of key nodes of risk sources, and generate tiered credit limit adjustment schemes by combining the location of key nodes in the supply chain, the intensity of risk transmission and the credit rules of financial institutions, and generate risk disposal instructions.

[0132] Specifically, the system extracts the risk transmission value of each node's location in the supply chain (core layer / supporting layer) and its associated path from the list of key risk source nodes. The system retrieves benchmark credit adjustment coefficients corresponding to different supply chain levels from the credit rule database of financial institutions. It then generates a tiered credit limit adjustment scheme using a formula.

[0133]

[0134] in, This represents the adjusted credit limit. The node represents the current credit limit of the corresponding enterprise. This represents the adjustment coefficient, with a smaller value corresponding to the core layer nodes. The corresponding nodes in the supporting layer are relatively large. , This represents the percentage of risk exceeding the limit. Based on the adjustment plan, the system generates risk disposal instructions, which include the key node enterprise ID, the adjusted credit limit, the execution start time, the subsequent review cycle, and the conditions for the instructions to take effect. The system uses a symmetric encryption algorithm to encrypt the instructions and sends them to the credit management system through a dedicated communication channel for financial institutions. Simultaneously, it records the instruction sending time, reception status, and encryption key identifier, forming a closed-loop record of risk disposal, which is stored in the disposal log database.

[0135] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0136] Based on the same inventive concept, this application also provides a supply chain finance risk control device for implementing the supply chain finance risk control method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more supply chain finance risk control device embodiments provided below can be found in the limitations of the supply chain finance risk control method described above, and will not be repeated here.

[0137] Preferably, such as Figure 3 As shown, the present invention provides a supply chain finance risk control device 500, which is configured with the following modules:

[0138] The multi-source heterogeneous data standardization module 510 is used to acquire multi-source heterogeneous data from multiple independent supply chains. Through a preset industry pre-trained model, it converts the JSON transaction flow, XML logistics data and Protobuf capital flow of the multi-source heterogeneous data into a unified vector format, generating a standardized data flow with source identifiers.

[0139] The dynamic risk transmission network construction module 520 is used to construct a dynamic risk transmission network based on standardized data flow. It establishes an initial graph structure based on the node feature vector and the supply chain label of the source identifier, and updates the topology relationship in real time through the time decay factor to generate a labeled dynamic graph.

[0140] The supply chain subgraph separation and risk calculation module 530 is used to separate the subgraphs of each independent supply chain based on the node labels of the labeled dynamic graph, calculate the maximum path risk value between nodes and the cross-supply chain transmission matrix, and generate the risk transmission matrix and the maximum risk path set.

[0141] The adaptive risk warning and handling instruction generation module 540 is used to adaptively warn the risk transmission matrix and the maximum risk path set. It generates a dynamic warning threshold by analyzing the statistical characteristics of the pre-stored historical risk transmission matrix and real-time policy text. When the risk value in the risk transmission matrix exceeds the dynamic warning threshold, it locates the key nodes in the risk transmission path and generates a risk handling instruction. The risk handling instruction is used to instruct financial institutions to adjust the credit limit of related enterprises at the key nodes.

[0142] Preferably, the multi-source heterogeneous data standardization module 510 provided in this application is configured with the following units:

[0143] The SON transaction flow parsing unit is used to acquire multi-source heterogeneous data from multiple independent supply chains, perform field parsing on the JSON transaction flow of the multi-source heterogeneous data, extract key fields such as transaction amount, enterprise ID and payment period, and generate a structured transaction dataset.

[0144] The XML logistics data conversion unit is used to perform multi-level tag parsing processing on multi-source heterogeneous XML logistics data, converting GPS coordinates, cargo status and temperature and humidity monitoring values ​​into a numerical matrix, and generating a standardized logistics data matrix.

[0145] The Protobuf Funds Flow Deserialization Unit is used to perform protocol deserialization processing on Protobuf funds flows of multi-source heterogeneous data, decode the binary streams of account changes, credit limits and settlement status, and generate time-series funds flow sequences.

[0146] The cross-modal feature fusion and standardization unit is used to perform cross-modal feature fusion processing on structured transaction datasets, standardized logistics data matrices, and time-series capital flow sequences. Through industry pre-trained models, transaction amount features, logistics coordinate features, and capital change features are mapped to a unified vector space, and original data source identifiers are added to generate standardized data streams with source identifiers.

[0147] Preferably, the dynamic risk transmission network construction module 520 provided in this application is configured with the following units:

[0148] The node feature similarity calculation unit is used to perform similarity calculation on the node feature vectors of the standardized data stream. It quantifies the directional consistency between feature vectors of different enterprises through the cosine similarity algorithm and generates an initial set of edge weights.

[0149] The node classification labeling unit is used to perform graph node labeling processing on the supply chain labels corresponding to the source identifiers in the standardized data stream. It assigns classification labels to nodes according to the three types of source identifiers: chemical chain, electronic chain, and logistics chain, and generates a node set with classification labels.

[0150] The initial graph structure building unit is used to process the initial edge weight set and the node set with classification labels to build a graph structure, establish a directed topological network with enterprise entities as nodes and similarity weights as edges, and generate the initial graph structure.

[0151] The dynamic topology update unit is used to perform dynamic topology update processing on the initial graph structure. It introduces a time decay factor to reduce the edge weights of historical transaction associations exponentially, while strengthening the new connection relationships reflected in real-time data, and generating a labeled dynamic graph.

[0152] Preferably, the supply chain sub-graph separation and risk calculation module 530 provided in this application is configured with the following units:

[0153] The supply chain subgraph separation unit is used to separate the node labels of the labeled dynamic graph into supply chain subgraphs. It divides the enterprise node subsets according to the chemical, electronic and logistics labels and generates a set of supply chain subgraphs.

[0154] The cross-chain potential risk path search unit is used to perform full path search processing on cross-supply chain node pairs, traversing cross-chain connection paths such as chemical subgraph to electronic subgraph and chemical subgraph to logistics subgraph, and generating a set of potential risk paths;

[0155] The path risk intensity calculation unit is used to calculate the weight decay of the potential risk path set, dynamically adjust the edge weight product value based on the topological distance, and generate a set of path risk intensity values.

[0156] The risk transmission matrix and maximum path generation unit are used to matrix-process the set of path risk intensity values, aggregate cross-chain risk transmission intensity according to supply chain relationships, and generate the risk transmission matrix and the maximum risk path set.

[0157] Preferably, the adaptive risk warning and handling instruction generation module 540 provided in this application is configured with the following units:

[0158] The historical risk statistical feature extraction unit is used to extract statistical features from the pre-stored historical risk transmission matrix, calculate the mean and standard deviation of the 30-day sliding window as the benchmark for risk transmission intensity, and generate historical statistical features.

[0159] The real-time policy text semantic parsing unit is used to perform semantic parsing on the central bank's policy texts acquired in real time, quantify the policy severity and timeliness indicators through legal text models, and generate real-time policy risk factors.

[0160] The dynamic early warning threshold fusion generation unit is used to perform threshold fusion processing on historical statistical characteristics and real-time policy risk factors, and to weight and integrate the risk transmission intensity benchmark and the impact of policy adjustments to generate dynamic early warning thresholds.

[0161] The risk value over-limit detection and high-risk transmission pair location unit is used to perform over-limit detection processing on the risk values ​​and dynamic early warning thresholds in the risk transmission matrix, locate high-risk supply chain transmission pairs that exceed the thresholds, and generate a set of high-risk transmission pairs.

[0162] The critical node location unit for the maximum risk path is used to locate critical nodes in the set of maximum risk paths, identify core enterprise nodes based on the risk transmission path topology, and generate a list of critical nodes of risk sources.

[0163] The risk disposal instruction generation unit is used to generate credit instructions from the list of key nodes of risk sources. It combines the location of the key nodes in the supply chain, the intensity of risk transmission, and the credit rules of financial institutions to generate a tiered credit limit adjustment plan and generate risk disposal instructions.

[0164] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described supply chain finance risk control method.

[0165] In one embodiment, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described supply chain finance risk control method.

[0166] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0167] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0168] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A supply chain finance risk control method, characterized in that, Includes the following steps: S1: Acquire multi-source heterogeneous data from multiple independent supply chains, and convert the JSON transaction flow, XML logistics data and Protobuf capital flow of the multi-source heterogeneous data into a unified vector format through a preset industry pre-trained model to generate a standardized data flow with source identifiers; S2: Based on the standardized data flow, a dynamic risk transmission network is constructed. An initial graph structure is established based on the node feature vector and the supply chain label of the source identifier. The topology relationship is updated in real time through the time decay factor to generate a labeled dynamic graph. S3: Based on the node labels of the labeled dynamic graph, separate the subgraphs of each independent supply chain, calculate the maximum path risk value between nodes and the cross-supply chain transmission matrix, and generate the risk transmission matrix and the maximum risk path set; S4: Adaptive early warning is performed on the risk transmission matrix and the maximum risk path set. A dynamic early warning threshold is generated by analyzing the statistical characteristics of the pre-stored historical risk transmission matrix and real-time policy text. When the risk value in the risk transmission matrix exceeds the dynamic early warning threshold, the key node in the risk transmission path is located and a risk disposal instruction is generated. The risk disposal instruction is used to instruct financial institutions to adjust the credit limit of the related enterprises of the key node. The labeled animated image is obtained through the following steps: S21: Perform similarity calculation on the node feature vectors of the standardized data stream, quantify the directional consistency between feature vectors of different enterprises using the cosine similarity algorithm, and generate an initial set of edge weights; S22: Perform graph node labeling processing on the supply chain tags corresponding to the source identifiers in the standardized data stream, and assign classification tags to the nodes according to the three types of source identifiers: chemical chain, electronic chain, and logistics chain, to generate a node set with classification tags; S23: Perform graph structure construction processing on the initial edge weight set and the node set with classification labels to establish a directed topological network with enterprise entities as nodes and similarity weights as edges, and generate the initial graph structure; S24: Perform dynamic topology update processing on the initial graph structure, introduce a time decay factor to exponentially reduce the edge weights of historical transaction associations, and strengthen the new connections reflected in real-time data to generate a labeled dynamic graph. The adjustment formula for the edge weights of historical transaction associations is as follows: ； in, To adjust the edge weights at the current time t, Initial time edge weights, Industry-specific attenuation rate, As the impact factor of real-time data, The edge weight increment is calculated for real-time data; The risk transmission matrix and the maximum risk path set are obtained through the following steps: S31: Perform supply chain subgraph separation processing on the node labels of the labeled dynamic graph, divide the enterprise node subsets according to the chemical, electronic and logistics labels, and generate a supply chain subgraph set; S32: Perform full-path search processing on cross-supply chain node pairs, traverse cross-chain connection paths such as from chemical subgraph to electronic subgraph and from chemical subgraph to logistics subgraph, and generate a set of potential risk paths; S33: Perform weight decay calculation on the potential risk path set, dynamically adjust the edge weight product value based on the topological distance, and generate a path risk intensity value set; wherein, the calculation formula for the path risk intensity value set is: ； ； in, Let p be the risk intensity value. The Sigmoid normalization function, Let p be the number of edges on path p. For the decay edge weight function, Here is the original weight of the k-th edge. Let be the shortest path hop count between the two ends of the k-th edge in path p. This is the distance attenuation coefficient; S34: Perform matrix processing on the set of path risk intensity values, aggregate the cross-chain risk transmission intensity according to the supply chain relationship, and generate a risk transmission matrix and the set of the maximum risk paths; The risk handling instruction is obtained through the following steps: S41: Perform statistical feature extraction on the pre-stored historical risk transmission matrix, calculate the mean and standard deviation of the 30-day sliding window as the benchmark for risk transmission intensity, and generate historical statistical features; S42: Perform semantic parsing on the central bank's policy texts acquired in real time, quantify the policy's severity and timeliness indicators through a legal text model, and generate real-time policy risk factors; S43: Perform threshold fusion processing on the historical statistical characteristics and the real-time policy risk factors, and weightedly integrate the risk transmission intensity benchmark and the impact of policy adjustments to generate dynamic early warning thresholds; S44: Perform over-limit detection processing on the risk values ​​in the risk transmission matrix and the dynamic early warning threshold, locate high-risk supply chain transmission pairs that exceed the threshold, and generate a set of high-risk transmission pairs; S45: Perform key node localization processing on the set of the largest risk paths, identify core enterprise nodes based on the risk transmission path topology, and generate a list of key nodes of risk sources. S46: Generate credit instructions for the list of key nodes of the risk source, and generate a tiered credit limit adjustment plan by combining the location of the key nodes in the supply chain, the intensity of risk transmission and the credit rules of financial institutions, and generate risk disposal instructions.

2. The method according to claim 1, characterized in that, S1 includes: S11: Obtain multi-source heterogeneous data from multiple independent supply chains, perform field parsing processing on the JSON transaction stream of the multi-source heterogeneous data, extract key fields such as transaction amount, enterprise ID and payment period, and generate a structured transaction dataset; S12: Perform multi-level tag parsing on the XML logistics data of the multi-source heterogeneous data, convert the GPS coordinates, cargo status and temperature and humidity monitoring values ​​into a numerical matrix, and generate a standardized logistics data matrix. S13: Perform protocol deserialization processing on the Protobuf fund flow of the multi-source heterogeneous data, decode the binary streams of account changes, credit limits and settlement status, and generate a time-series fund flow sequence; S14: Perform cross-modal feature fusion processing on the structured transaction dataset, the standardized logistics data matrix, and the time-series capital flow sequence. Map the transaction amount features, logistics coordinate features, and capital change features to a unified vector space through an industry pre-trained model, and add the original data source identifier to generate a standardized data stream with source identifier.

3. A supply chain finance risk control device, characterized in that, The device includes: The multi-source heterogeneous data standardization module is used to acquire multi-source heterogeneous data from multiple independent supply chains. Through a preset industry pre-trained model, the JSON transaction flow, XML logistics data and Protobuf capital flow of the multi-source heterogeneous data are converted into a unified vector format to generate a standardized data flow with source identifiers. The dynamic risk transmission network construction module is used to construct a dynamic risk transmission network based on the standardized data flow. It establishes an initial graph structure based on the supply chain tags of node feature vectors and source identifiers, and updates the topology in real time through a time decay factor to generate a labeled dynamic graph. The supply chain subgraph separation and risk calculation module is used to separate the subgraphs of each independent supply chain based on the node labels of the labeled dynamic graph, calculate the maximum path risk value between nodes and the cross-supply chain transmission matrix, and generate the risk transmission matrix and the maximum risk path set. An adaptive risk warning and handling instruction generation module is used to adaptively warn the risk transmission matrix and the maximum risk path set. It generates a dynamic warning threshold by using the statistical characteristics of the pre-stored historical risk transmission matrix and real-time policy text parsing. When the risk value in the risk transmission matrix exceeds the dynamic warning threshold, it locates the key node in the risk transmission path and generates a risk handling instruction. The risk handling instruction is used to instruct financial institutions to adjust the credit limit of the related enterprises of the key node. The labeled animated image is obtained through the following steps: The node feature vectors of the standardized data stream are processed by similarity calculation. The cosine similarity algorithm is used to quantify the directional consistency between feature vectors of different enterprises and generate an initial set of edge weights. The supply chain tags corresponding to the source identifiers in the standardized data stream are labeled with graph nodes. Classification tags are assigned to the nodes according to the three types of source identifiers: chemical chain, electronic chain, and logistics chain, generating a node set with classification tags. The initial edge weight set and the node set with classification labels are processed to construct a graph structure, and a directed topological network with enterprise entities as nodes and similarity weights as edges is established to generate the initial graph structure. The initial graph structure undergoes dynamic topology update processing. A time decay factor is introduced to exponentially reduce the edge weights associated with historical transactions, while simultaneously strengthening new connections reflected in real-time data, generating a labeled dynamic graph. The adjustment formula for the edge weights associated with historical transactions is as follows: ； in, To adjust the edge weights at the current time t, Initial time edge weights, Industry-specific attenuation rate, As the impact factor of real-time data, The edge weight increment is calculated for real-time data; The risk transmission matrix and the maximum risk path set are obtained through the following steps: The node labels of the tagged dynamic graph are subjected to supply chain subgraph separation processing. Enterprise node subsets are divided according to chemical, electronic and logistics labels to generate a supply chain subgraph set. A full-path search is performed on cross-supply chain node pairs, traversing cross-chain connection paths such as from the chemical subgraph to the electronics subgraph and from the chemical subgraph to the logistics subgraph, to generate a set of potential risk paths. Weight decay is calculated for the set of potential risk paths, and the edge weight product is dynamically adjusted based on the topological distance to generate a set of path risk intensity values; wherein, the calculation formula for the set of path risk intensity values ​​is: ； ； in, Let p be the risk intensity value. The Sigmoid normalization function, Let p be the number of edges on path p. For the decay edge weight function, Here is the original weight of the k-th edge. Let be the shortest path hop count between the two ends of the k-th edge in path p. This is the distance attenuation coefficient; The set of path risk intensity values ​​is matrixed, and the cross-chain risk transmission intensity is aggregated according to the supply chain relationship to generate a risk transmission matrix and a set of maximum risk paths; The risk handling instruction is obtained through the following steps: The pre-stored historical risk transmission matrix is ​​subjected to statistical feature extraction processing. The mean and standard deviation of the 30-day sliding window are calculated as the benchmark for risk transmission intensity to generate historical statistical features. Semantic analysis is performed on the central bank's policy texts acquired in real time. The severity and timeliness of the policies are quantified through a legal text model to generate real-time policy risk factors. The historical statistical features and the real-time policy risk factors are subjected to threshold fusion processing, and the risk transmission intensity benchmark and the impact of policy adjustments are weighted and integrated to generate dynamic early warning thresholds; The risk values ​​in the risk transmission matrix and the dynamic early warning threshold are subjected to over-limit detection processing to locate high-risk supply chain transmission pairs that exceed the threshold and generate a set of high-risk transmission pairs. The set of maximum risk paths is processed to locate key nodes, and core enterprise nodes are identified based on the risk transmission path topology to generate a list of key nodes of risk sources. The list of key nodes of the risk source is processed to generate credit instructions. Based on the location of the key nodes in the supply chain, the intensity of risk transmission and the credit rules of financial institutions, a tiered credit limit adjustment plan is generated, and a risk disposal instruction is generated.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 2.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 2.

Images