A blockchain-based supply chain financial transaction security early warning method

Abstract

The application discloses a kind of based on blockchain supply chain financial transaction security early warning method, it is related to blockchain and supply chain financial technical field, comprising: step one, supply chain financial transaction data is collected to generate original data set;Step two, generate certificate abstract and write into blockchain, form chain on certificate record and phase fingerprint;Step three, based on financing application number constructs five-flow closed transaction unit and difference data;Step four, constructs supply chain financial heterogeneous transaction graph;Step five, extract financing creditor's rights risk subgraph and feature;Step six, the feature is input into improved FRAUDRE model and executes chain on phase constraint, five-flow decoupling and super edge modeling processing generation risk characterization data;Step seven, generates transaction security early warning result;Step eight, generate early warning evidence capsule and write into blockchain to form evidence chain.The application realizes supply chain financial transaction security early warning.

Description

Technical Field

[0001] This invention relates to the fields of blockchain and supply chain finance technology, and in particular to a blockchain-based method for early warning of supply chain finance transaction security. Background Technology

[0002] With the digitalization and platformization of supply chain finance, transaction data sources are becoming increasingly diversified, covering data from various business systems such as contracts, orders, invoices, warehouse receipts, logistics, payments, and financing. Meanwhile, the application of blockchain technology in financial data storage and traceability is deepening. Existing technologies typically determine the authenticity of transactions through simple data aggregation or rule-based verification, but the following issues still exist in complex supply chain finance scenarios: Existing technologies for risk identification in supply chain finance transactions often rely on single-dimensional data comparison or static rule matching, which struggles to simultaneously depict the complex coupling relationships between the flow of commerce, invoices, goods, funds, and ownership. This results in insufficient ability to verify transaction consistency across systems and multiple nodes. Furthermore, existing graph-based risk analysis methods typically only consider ordinary graph structure relationships, lacking the ability to model temporal features such as blockchain on-chain order, block height, and dependencies on preceding credentials, leading to an inadequate characterization of on-chain transaction consistency. In addition, traditional risk warning methods often rely on single features or simple feature combinations for judgment, failing to effectively integrate node features, edge features, and topological information, making it difficult to perform refined modeling of complex risk propagation paths at the financing and debt level.

[0003] Therefore, how to provide a blockchain-based supply chain finance transaction security early warning method is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a blockchain-based supply chain finance transaction security early warning method. This invention achieves consistent expression and traceability analysis of multi-source transactions in supply chain finance by integrating on-chain certificate phase fingerprints and five-flow closure difference data through blockchain notarization and heterogeneous transaction graph modeling. By improving the FRAUDRE model and introducing on-chain phase constraints, five-flow relationship decoupling, and hyperedge modeling mechanisms, it enhances the ability to identify cross-certificate correlation risks and abnormal fund paths, thereby achieving structured expression and refined early warning of transaction risks.

[0005] A blockchain-based supply chain finance transaction security early warning method according to an embodiment of the present invention includes the following steps: Step 1: Collect supply chain finance transaction data and generate the original supply chain finance transaction dataset; Step 2: Generate a voucher summary based on the original dataset of the supply chain finance transaction, and write the voucher summary into the blockchain to generate on-chain voucher records and on-chain voucher phase fingerprints; Step 3: Based on the financing application number and related transaction vouchers, generate a five-flow closed-loop transaction unit and generate five-flow closed-loop difference data; Step 4: Construct a supply chain finance heterogeneous transaction graph based on the on-chain certificate records, on-chain certificate phase fingerprints, and five-flow closure difference data; Step 5: Extract the financing claim risk subgraph from the heterogeneous transaction graph of supply chain finance, and generate financing claim node features, financing claim edge features, and financing claim topological features; Step Six: Input the financing claim risk subgraph, financing claim node features, financing claim edge features, and financing claim topology features into the improved FRAUDRE model, and sequentially perform on-chain phase constraint processing, five-flow relationship decoupling processing, and financing claim super-edge modeling processing to generate financing transaction security risk representation data; Step 7: Generate transaction security early warning results based on the aforementioned financing transaction security risk characterization data; Step 8: Generate an early warning evidence capsule based on the transaction security early warning result, and write the early warning evidence capsule into the blockchain to generate an on-chain transaction security early warning evidence chain.

[0006] Optionally, step one specifically includes: The data collected includes supply chain finance transaction data, such as contract data, order data, invoice data, warehouse receipt data, logistics receipt data, payment data, financing application data, accounts receivable transfer data, guarantee data, and repayment data. Extract the transaction entity information, transaction voucher information, transaction amount information, and transaction time information from the supply chain finance transaction data, and generate transaction entity field sets, transaction voucher field sets, transaction amount field sets, and transaction time field sets respectively; The transaction entity field set, transaction voucher field set, transaction amount field set, and transaction time field set are uniformly formatted and associated according to the financing application number; The contracts, orders, invoices, warehouse receipts, logistics receipts, payments, accounts receivable transfers, guarantees, and repayment records corresponding to the same financing application number are collected to form a financing transaction record group; Based on the financing transaction record group, the relationships, fund transfer relationships, and debt transfer relationships between the transaction entities are recorded to form a financing transaction data unit; The original dataset for supply chain finance transactions is generated based on the financing transaction data unit.

[0007] Optionally, step two specifically involves: Read the transaction voucher records in the original dataset of supply chain finance transactions, and generate voucher basic field groups according to financing application number, voucher type, voucher number, transaction entity, transaction amount, transaction time and associated voucher number; The basic field group of the voucher is sorted and characters are concatenated to form the voucher summary text. The voucher summary text is then hashed to generate the voucher summary corresponding to each transaction voucher. Arrange the transaction voucher records in chronological order of the business occurrence under the same financing application number, and write the previous voucher summary in the adjacent transaction voucher record into the next transaction voucher record to generate the previous voucher summary; Write the credential summary, subject signature, credential type, transaction time, previous credential summary and financing application number into the blockchain to generate an on-chain credential record containing block height, on-chain time, on-chain transaction number and subject identifier; Collect on-chain certificate records according to the financing application number, and generate an on-chain certificate sequence according to the block height, on-chain time, transaction time and previous certificate summary connection relationship; Based on the on-chain credential sequence, the block height sequence, the on-chain time sequence, the credential type sequence, the preceding digest connection state, and the main signature state are extracted to generate the on-chain credential phase fingerprint.

[0008] Optionally, step three specifically includes: Read transaction vouchers with the same financing application number from the original dataset of supply chain finance transactions, and generate a set of financing application vouchers; According to the type of voucher, the set of financing application vouchers is divided into commercial flow data, invoice flow data, goods flow data, capital flow data, and ownership flow data. The commercial flow data includes contract data and order data, the invoice flow data includes invoice data, the goods flow data includes warehouse receipt data and logistics receipt data, the capital flow data includes payment data, financing application amount data, and repayment data, and the ownership flow data includes accounts receivable transfer data and guarantee data. Based on the financing application number, transaction entity, transaction amount, transaction time, transaction account, goods identifier, and ownership identifier, the fields of the commercial flow data, invoice flow data, goods flow data, capital flow data, and ownership flow data are aligned to generate a five-flow closed transaction unit; For the contract amount, order amount, invoice amount, warehouse receipt value, financing application amount, payment amount, debt transfer amount, guarantee amount, and repayment amount in the five-flow closed transaction unit, the difference is calculated according to the business closure sequence of contract to order, order to invoice, invoice to warehouse receipt value, warehouse receipt value to financing application amount, financing application amount to payment amount, payment amount to repayment amount, debt transfer amount to financing application amount, and guarantee amount to financing application amount. The comparison field, benchmark field, absolute value of the difference, direction of the difference, and source of the field are recorded to generate amount closure difference data. A consistency comparison is performed on the transaction entity, transaction time, goods identifier, transaction account, and ownership identifier in the five-flow closed transaction unit to generate entity closure difference data, time closure difference data, goods closure difference data, account closure difference data, and ownership closure difference data, respectively. The amount closure difference data, entity closure difference data, time closure difference data, goods closure difference data, account closure difference data, and ownership closure difference data are written into the five-flow closure transaction unit according to the financing application number to generate five-flow closure difference data.

[0009] Optionally, step four specifically includes: Read the on-chain certificate record, the on-chain certificate phase fingerprint, and the five-flow closure difference data, extract the transaction entity identifier, transaction account identifier, transaction certificate identifier, financing application number, and on-chain transaction number, and generate transaction entity node, transaction account node, transaction certificate node, financing application node, and on-chain certificate record node; Based on the connection relationship of the certificate summary, on-chain transaction number, block height, on-chain time and previous certificate summary, the transaction certificate node and the on-chain certificate record node are connected to generate certificate on-chain relationship edge and on-chain phase relationship edge; According to the classification of commercial flow, ticket flow, goods flow, capital flow and ownership flow, the transaction entity node, transaction account node, transaction voucher node and financing application node are connected to generate five-flow transaction relationship edges, and the five-flow closure difference data is written into the corresponding five-flow transaction relationship edges; The on-chain relationship edges, on-chain phase relationship edges, and five-flow transaction relationship edges of the vouchers are marked with relationship types to generate a heterogeneous relationship edge set. Based on the transaction subject node, transaction account node, transaction voucher node, financing application node, on-chain voucher record node, and the heterogeneous relationship edge set, a supply chain finance heterogeneous transaction graph is generated.

[0010] Optionally, step five specifically includes: Read the financing application node in the heterogeneous transaction graph of supply chain finance, and retrieve the transaction entity node, transaction voucher node, transaction account node and on-chain voucher record node connected to the financing application node according to the financing application number to generate a set of financing application associated nodes; Read the five-flow transaction relationship edges, on-chain certificate relationship edges, and on-chain phase relationship edges among the set of nodes associated with the financing application, and generate a set of edges associated with the financing application; A financing claim risk subgraph is constructed based on the set of associated nodes and the set of associated edges of the financing application. The financing claim risk subgraph includes a subject relationship subgraph, a certificate relationship subgraph, an account relationship subgraph, an on-chain phase subgraph, and a five-flow relationship subgraph. The nodes in the financing claim risk subgraph are feature-encoded to generate financing claim node features, which include node type, subject role, voucher type, account type, financing application number, transaction amount, business occurrence time, on-chain sequence, and five-flow category; The edges in the financing claims risk subgraph are feature-encoded to generate financing claims edge features. The financing claims edge features include edge type, edge direction, relationship source, amount closure difference, subject closure difference, time closure difference, account closure difference, ownership closure difference, and on-chain phase connection state. The financing debt risk subgraph is subjected to topological statistics to generate financing debt topological features, which include the number of node types, the number of edge types, the length of the five-flow path, the length of the on-chain phase path, the number of shared accounts, the number of duplicate vouchers, the number of fund return paths, and the number of debt transfer paths.

[0011] Optionally, step six specifically includes: The financing claim risk subgraph, financing claim node features, financing claim edge features, and financing claim topological features are input into the improved FRAUDRE model. The improved FRAUDRE model includes sequentially connected on-chain phase constraint units, five-flow relationship decoupling units, and financing claim superedge modeling units. The on-chain phase constraint unit reads the block height sequence, on-chain time sequence, and business occurrence time sequence from the on-chain phase relationship edge. It aligns the same transaction voucher node according to the sequence index, calculates the absolute value of the difference between the business occurrence time sequence and the block height sequence at the corresponding index position, and the absolute value of the difference between the business occurrence time and the on-chain time at the corresponding index position. It then sums the two differences with equal weight to generate the on-chain phase constraint edge strength and writes it into the transaction relationship edge. The five-flow relationship decoupling unit divides the financing and debt risk sub-graph according to the flow of commerce, invoices, goods, funds, and ownership; The cumulative values ​​of the amount closure difference, subject closure difference, time closure difference, and account closure difference within each relationship channel are calculated to form a four-dimensional channel feature vector; The absolute values ​​of the component differences between each pair of feature vectors are summed to generate the channel correlation distance, and the channel correlation distance matrix is ​​formed by filling all the channel correlation distances. A five-flow relationship decoupling representation is generated based on the channel association distance matrix; The financing claim superedge modeling unit aggregates transaction relationship edges according to the same financing application number to generate financing claim superedges, and calculates the sum of the absolute values ​​of the differences in each dimension between the five-flow relationship decoupling representation and the superedge central feature corresponding to each relationship edge to generate the relationship edge risk contribution value. The risk contribution values ​​of the relationship edges are sorted from largest to smallest, weighted, and accumulated, and then mapped to the financing application node, transaction relationship edge, and transaction path to output financing transaction security risk characterization data.

[0012] Optionally, the channel correlation distance matrix is ​​specifically: Based on the aforementioned financing and debt risk sub-graph, five relationship channels corresponding to the flow of commerce, invoices, goods, funds, and ownership are extracted; The transaction relationship edges within each relationship channel are classified and statistically analyzed to obtain the cumulative value of the amount closure difference, the cumulative value of the subject closure difference, the cumulative value of the time closure difference, and the cumulative value of the account closure difference. These values ​​are then arranged in the order of amount, subject, time, and account to construct a four-dimensional channel feature vector. For any two relational channels, calculate the difference between the values ​​of the corresponding dimensions in their four-dimensional channel feature vectors, and sum the absolute values ​​of the differences in each dimension to generate the channel association distance. Construct a set of channel association distances based on the channel association distances between all channels; Write the channel association distances between each pair of the five relation channels into the corresponding element positions of a five-row, five-column matrix to generate the initial channel association distance matrix; The five relation channels are sorted in ascending order of the sum of all channel association distances corresponding to each relation channel to generate the channel association order; The rows and columns of the channel association distance matrix are rearranged according to the channel association order to generate a rearranged channel association distance matrix.

[0013] Optionally, step seven specifically includes: Read the financing transaction security risk characterization data and extract the financing application risk characterization, transaction relationship edge risk characterization, and transaction path risk characterization; The risk representation of the financing application is decomposed into components to obtain the node risk component, the funding risk component, and the debt risk component, and a financing application risk vector is generated. The risk representation of the transaction relationship edge is divided into edge types to obtain the commercial flow risk edge, ticket flow risk edge, goods flow risk edge, capital flow risk edge and ownership flow risk edge, and a relationship edge risk set is generated. The risk characterization of the transaction path is decomposed to obtain the capital flow path, debt transfer path and transaction closed loop path, and a path risk sequence is generated. A consistency comparison is performed based on the financing application risk vector, the relationship edge risk set, and the path risk sequence to generate a risk consistency offset value. Based on the historical risk consistency offset value distribution corresponding to similar financing applications, the financing applications are classified into risk levels, and transaction security warning results are generated. The transaction security warning results include risk financing application identifiers, risk relationship edge identifiers, and risk path identifiers.

[0014] Optionally, step eight specifically includes: Read the transaction security warning results, extract the risk financing application identifier, risk relationship edge identifier, risk path identifier, risk level, risk triggering reason and warning time, and generate a warning result field set; Based on the aforementioned warning result field set, and by associating the on-chain certificate record, the financing claim risk subgraph identifier, the improved FRAUDRE model version identifier, and the financing transaction security risk characterization data summary, a warning evidence data package is generated; The warning evidence data packet is sorted by field and serialized by character to generate warning evidence capsule text, and the warning evidence capsule text is hashed to generate warning evidence capsule summary; Write the summary of the early warning evidence capsule, the risk financing application identifier, the early warning time, the signature entity identifier, and the storage address of the early warning evidence capsule into the blockchain to generate an on-chain early warning record; Read the block height, on-chain time, on-chain transaction number, and signature entity identifier from the on-chain warning record to generate an on-chain warning index; By associating the on-chain early warning index with the on-chain credential record according to the risk financing application identifier, credential summary and on-chain transaction number, an on-chain transaction security early warning evidence chain is generated.

[0015] The beneficial effects of this invention are: This invention constructs a blockchain-based system for storing supply chain finance transaction data and modeling multi-source heterogeneous transaction graphs. By combining on-chain certificate records, on-chain certificate phase fingerprints, and the collaborative expression of five-flow closure difference data, it addresses the challenges of dispersed data sources, difficulties in cross-system consistency verification, and complex coupling of multi-flow business relationships in supply chain finance transactions. It proposes a structured fusion strategy based on on-chain phase constraints and decoupling of five-flow relationships, significantly improving the traceability and consistency of transaction data in a blockchain environment. Furthermore, in the graph structure construction stage, a heterogeneous transaction graph and financing and debt risk subgraph modeling mechanism are introduced. Through joint encoding of node features, edge features, and topological features, the unified risk information at the financing and debt level is achieved. The model aligns expression and structure; an improved FRAUDRE model is introduced in the risk modeling stage. This model enhances the consistency of transaction timing through on-chain phase constraint units, achieves structural separation and correlation modeling among the five-flow relationship decoupling units (commercial flow, invoice flow, goods flow, capital flow, and ownership flow), and realizes the aggregated expression and risk contribution calculation of multiple transaction relationships through financing and debt hyper-edge modeling units. This effectively enhances the model's ability to identify cross-certificate correlation risks and abnormal propagation of capital paths. In the results generation stage, a risk consistency offset analysis and hierarchical early warning mechanism are constructed to achieve multi-granular characterization and structured early warning output of financing transaction risks, thereby improving the accuracy and interpretability of supply chain finance transaction security early warnings. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the overall process of a blockchain-based supply chain finance transaction security early warning method proposed in this invention; Figure 2 This is an overall structural diagram of the improved FRAUDRE model in the blockchain-based supply chain finance transaction security early warning method proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-2 A blockchain-based supply chain finance transaction security early warning method includes the following steps: Step 1: Collect supply chain finance transaction data and generate the original supply chain finance transaction dataset; Step 2: Generate voucher summaries based on the original dataset of supply chain finance transactions, and write the voucher summaries into the blockchain to generate on-chain voucher records and on-chain voucher phase fingerprints; Step 3: Based on the financing application number and related transaction vouchers, generate a five-flow closed-loop transaction unit and generate five-flow closed-loop difference data; Step 4: Construct a heterogeneous transaction graph for supply chain finance based on on-chain certificate records, on-chain certificate phase fingerprints, and five-flow closure difference data; Step 5: Extract the financing claim risk subgraph from the heterogeneous transaction graph of supply chain finance, and generate financing claim node features, financing claim edge features, and financing claim topological features; Step Six: Input the financing claim risk subgraph, financing claim node characteristics, financing claim edge characteristics, and financing claim topological characteristics into the improved FRAUDRE model, and sequentially perform on-chain phase constraint processing, five-flow relationship decoupling processing, and financing claim super-edge modeling processing to generate financing transaction security risk representation data; Step 7: Generate transaction security early warning results based on the risk characterization data of financing transactions; Step 8: Generate an early warning evidence capsule based on the transaction security early warning results, and write the early warning evidence capsule into the blockchain to generate an on-chain transaction security early warning evidence chain.

[0019] In this embodiment, step one specifically includes: Collect supply chain finance transaction data, which includes contract data, order data, invoice data, warehouse receipt data, logistics receipt data, payment data, financing application data, accounts receivable transfer data, guarantee data, and repayment data. Extract transaction entity information, transaction voucher information, transaction amount information, and transaction time information from supply chain finance transaction data, and generate transaction entity field sets, transaction voucher field sets, transaction amount field sets, and transaction time field sets respectively; The transaction entity field set, transaction voucher field set, transaction amount field set, and transaction time field set are uniformly formatted and associated according to the financing application number; The contracts, orders, invoices, warehouse receipts, logistics receipts, payments, accounts receivable transfers, guarantees, and repayment records corresponding to the same financing application number are collected to form a financing transaction record group; Based on the records of financing transaction records, the relationships between the transaction entities, the flow of funds, and the flow of claims are recorded to form financing transaction data units; The original dataset of supply chain finance transactions is generated based on the financing transaction data unit.

[0020] In this implementation, supply chain finance transaction data is exported from the business platform, the bank's core system, the electronic warehousing system, the logistics signing system, and the electronic invoice system. The exported data is first indexed according to the data source identifier, and then the company name, account name, and institutional role are merged to form a transaction entity mapping table. The voucher number, amount unit, and time format are standardized. The time field is uniformly converted into a timestamp sequence, and the amount field is uniformly converted into the same currency and the same measurement precision. Using the financing application number as the primary association key, the entity mapping table, voucher field, amount field, and time field under the same financing business are linked together, and the fund transfer relationship and debt transfer relationship are formed according to the payment account flow and debt transfer record.

[0021] In this embodiment, step two specifically involves: Read the transaction voucher records in the original dataset of supply chain finance transactions, and generate voucher basic field groups according to financing application number, voucher type, voucher number, transaction entity, transaction amount, transaction time and associated voucher number; The basic field group of the voucher is sorted and characters are concatenated to form the voucher summary text. The voucher summary text is then hashed to generate the voucher summary corresponding to each transaction voucher. Arrange the transaction voucher records in chronological order of the business occurrence under the same financing application number, and write the previous voucher summary in the adjacent transaction voucher record into the next transaction voucher record to generate the previous voucher summary; Write the credential summary, subject signature, credential type, transaction time, previous credential summary and financing application number into the blockchain to generate an on-chain credential record containing block height, on-chain time, on-chain transaction number and subject identifier; Collect on-chain certificate records according to the financing application number, and generate an on-chain certificate sequence according to the block height, on-chain time, transaction time and previous certificate summary connection relationship; Based on the on-chain credential sequence, the block height sequence, the on-chain time sequence, the credential type sequence, the preceding digest connection state, and the main signature state are extracted to generate the on-chain credential phase fingerprint.

[0022] In this implementation, the credential digest text is serialized using a unified field order, unified character encoding, and unified delimiter. Empty fields are retained by adding a null value marker to the field name, ensuring that the same credential exported from different systems forms a consistent digest. The main signature is verified using the institution certificate registered on the business platform, and a one-to-one index is established between the on-chain transaction number and the financing application number. The on-chain credential phase fingerprint uses the block height sequence and the transaction occurrence time sequence under the same financing application as the main sequence, and the previous digest connection relationship and signature verification result as the verification sequence, and is written into the local index library for use in the construction of heterogeneous transaction graphs.

[0023] In this embodiment, step three specifically includes: Read transaction vouchers with the same financing application number from the original dataset of supply chain finance transactions, and generate a set of financing application vouchers; According to the type of voucher, the set of financing application vouchers is divided into commercial flow data, invoice flow data, goods flow data, capital flow data and ownership flow data. Commercial flow data includes contract data and order data, invoice flow data includes invoice data, goods flow data includes warehouse receipt data and logistics receipt data, capital flow data includes payment data, financing application amount data and repayment data, and ownership flow data includes accounts receivable transfer data and guarantee data. Based on the financing application number, transaction entity, transaction amount, transaction time, transaction account, goods identifier, and ownership identifier, the fields of commercial flow data, invoice flow data, goods flow data, capital flow data, and ownership flow data are aligned to generate a five-flow closed transaction unit; For the contract amount, order amount, invoice amount, warehouse receipt value, financing application amount, payment amount, debt transfer amount, guarantee amount, and repayment amount in the five-flow closed transaction unit, the difference is calculated according to the business closure sequence from contract to order, order to invoice, invoice to warehouse receipt value, warehouse receipt value to financing application amount, financing application amount to payment amount, payment amount to repayment amount, debt transfer amount to financing application amount, and guarantee amount to financing application amount. The comparison field, benchmark field, absolute value of the difference, direction of the difference, and source of the field are recorded to generate amount closure difference data. A consistency comparison is performed on the transaction entity, transaction time, goods identifier, transaction account, and ownership identifier in the five-flow closed transaction unit, generating entity closure difference data, time closure difference data, goods closure difference data, account closure difference data, and ownership closure difference data, respectively. Write the amount closure difference data, entity closure difference data, time closure difference data, goods closure difference data, account closure difference data, and ownership closure difference data into the five-flow closure transaction unit according to the financing application number to generate five-flow closure difference data.

[0024] In this implementation, the five-flow closed transaction unit establishes a record index with the financing application number as the primary key. When aligning fields, the original voucher number and source system identifier are retained. The amount difference is calculated using adjacent fields in the business link, and the difference direction is recorded according to the flow of funds, goods, and claims. The comparison of entities, accounts, and ownership is performed using the unified social credit code, bank account number, and claim transfer number. Fields that cannot be matched are marked as null values ​​and written into the difference data for use by node attributes and edge attributes in the heterogeneous transaction graph of supply chain finance.

[0025] In this embodiment, step four specifically includes: Read on-chain certificate records, on-chain certificate phase fingerprints, and five-flow closure difference data; extract transaction entity identifier, transaction account identifier, transaction certificate identifier, financing application number, and on-chain transaction number; and generate transaction entity node, transaction account node, transaction certificate node, financing application node, and on-chain certificate record node. Based on the connection relationship of the certificate summary, on-chain transaction number, block height, on-chain time and previous certificate summary, the transaction certificate node and the on-chain certificate record node are connected to generate certificate on-chain relationship edge and on-chain phase relationship edge; Based on the classification of commercial flow, ticket flow, goods flow, capital flow and ownership flow, the transaction entity node, transaction account node, transaction voucher node and financing application node are connected to generate five-flow transaction relationship edges, and the five-flow closure difference data is written into the corresponding five-flow transaction relationship edges; The relationship types of the on-chain relationship edges, on-chain phase relationship edges, and five-flow transaction relationship edges are marked to generate a set of heterogeneous relationship edges. Based on the transaction entity node, transaction account node, transaction certificate node, financing application node, on-chain certificate record node, and the set of heterogeneous relationship edges, a supply chain finance heterogeneous transaction graph is generated.

[0026] In this implementation, the heterogeneous transaction graph of supply chain finance is stored in a graph database. The node number is formed by concatenating the node type, business number, and financing application number. The edge direction is recorded according to the business flow direction and the order of on-chain entry. The edge attributes use key-value pairs to store the difference field name, difference value, on-chain order, and source system identifier. Nodes and edges under the same financing application number are written to the same graph partition. When sharing entities and accounts across financing applications, the global node index is retained to ensure that both the risk subgraph of a single financing transaction and the historical relationship graph of the entity can be queried.

[0027] In this embodiment, step five specifically includes: Read the financing application node in the heterogeneous transaction graph of supply chain finance, and retrieve the transaction entity node, transaction certificate node, transaction account node and on-chain certificate record node connected to the financing application node according to the financing application number to generate a set of financing application associated nodes; Read the five-flow transaction relationship edges, the on-chain certificate relationship edges, and the on-chain phase relationship edges among the set of nodes associated with the financing application, and generate a set of edges associated with the financing application. A financing claim risk subgraph is constructed based on the set of nodes associated with financing applications and the set of edges associated with financing applications. The financing claim risk subgraph includes a subject relationship subgraph, a certificate relationship subgraph, an account relationship subgraph, an on-chain phase subgraph, and a five-flow relationship subgraph. The nodes in the financing claims risk subgraph are feature-encoded to generate financing claims node features. The financing claims node features include node type, subject role, voucher type, account type, financing application number, transaction amount, business occurrence time, on-chain sequence, and five-flow category. The edges in the financing claims risk subgraph are feature-encoded to generate financing claims edge features. The financing claims edge features include edge type, edge direction, relationship source, amount closure difference, subject closure difference, time closure difference, account closure difference, ownership closure difference, and on-chain phase connection status. Topological statistics are performed on the financing and debt risk subgraph to generate financing and debt topological features, which include the number of node types, the number of edge types, the length of the five-flow path, the length of the on-chain phase path, the number of shared accounts, the number of duplicate vouchers, the number of fund return paths, and the number of debt transfer paths.

[0028] In this implementation, the financing debt risk subgraph is formed using a graph partitioning retrieval method, with the financing application number as the partition key. After locating the financing application node in the graph database, the adjacency list is read along the five-flow transaction relationship edges, the on-chain certificate relationship edges, and the on-chain phase relationship edges. Node fields are converted into numerical sequences using a field encoding table, with missing fields retaining null value markers. Edge fields generate directional and difference codes according to the relationship type. Topology statistics are completed by the adjacency list, forming path length, number of shared nodes, and number of backflow paths. All features are bound to the financing application number to form a single financing input sample for the improved FRAUDRE model.

[0029] In this embodiment, step six specifically includes: The risk subgraph of financing claims, the node characteristics of financing claims, the edge characteristics of financing claims, and the topological characteristics of financing claims are input into the improved FRAUDRE model. The improved FRAUDRE model includes sequentially connected on-chain phase constraint units, five-flow relationship decoupling units, and financing claims superedge modeling units. The on-chain phase constraint unit reads the block height sequence, on-chain time sequence, and business occurrence time sequence from the on-chain phase relationship edge. It aligns the nodes of the same transaction certificate according to the sequence index, calculates the absolute value of the difference between the business occurrence time sequence and the block height sequence at the corresponding index position, and the absolute value of the difference between the business occurrence time and the on-chain time at the corresponding index position. It then sums the two differences with equal weight to generate the on-chain phase constraint edge strength and writes it into the transaction relationship edge. The five-flow relationship decoupling unit divides the financing and debt risk sub-graph according to the flow of commerce, invoices, goods, funds, and ownership; The cumulative values ​​of the amount closure difference, subject closure difference, time closure difference, and account closure difference within each relationship channel are calculated to form a four-dimensional channel feature vector; The absolute values ​​of the component differences between each pair of feature vectors are summed to generate the channel correlation distance, and the channel correlation distance matrix is ​​formed by filling all the channel correlation distances. Five-flow relationship decoupling representation generated based on channel association distance matrix; The financing claim superedge modeling unit aggregates transaction relationship edges according to the same financing application number to generate financing claim superedges, and calculates the sum of the absolute values ​​of the differences in each dimension between the five-flow relationship decoupling representation and the superedge central feature corresponding to each relationship edge to generate the relationship edge risk contribution value. After sorting the risk contribution values ​​of the relationship edges from largest to smallest, the values ​​are weighted and accumulated, and then mapped to the financing application node, transaction relationship edge, and transaction path to output financing transaction security risk characterization data.

[0030] In this implementation, the strength of the on-chain phase constraint edge is generated by the graph computing engine at the edge attribute layer and written into the weight field of the transaction relationship edge. This weight field is based on the corresponding difference between the block height sequence, the on-chain time sequence, and the business occurrence time sequence. The five-flow relationship decoupling representation consists of the vector results output by the five types of relationship channels, and participates in the subsequent hyperedge modeling as channel-level weights. The financing debt hyperedge modeling unit aggregates multiple types of transaction relationship edges under the same financing application number and calculates the edge weight summary result inside the hyperedge to form a risk contribution value that is mapped to the corresponding financing application node and path node to generate risk representation data. The improved FRAUDRE model is constructed based on the graph structure risk identification framework of the FRAUDRE model. Both models use the transaction relationship graph as the input carrier and achieve risk representation learning through the joint expression of node features, edge features and topological features. Furthermore, a multi-layer structure is used to perform hierarchical analysis of abnormal patterns in the graph. The improvements are made by introducing an on-chain phase constraint unit, a five-flow relationship decoupling unit, and a financing and debt super-edge modeling unit. The on-chain phase constraint unit integrates the block height sequence and the on-chain time sequence to form a time consistency constraint. The five-flow relationship decoupling unit divides the transaction relationship into commercial flow, ticket flow, goods flow, capital flow, and ownership flow and constructs a channel association matrix. The financing and debt super-edge modeling unit aggregates multiple types of transaction relationship edges at the granularity of financing applications to form a super-edge structure.

[0031] The above improvements enable the model to simultaneously characterize the blockchain's time consistency features, the coupling relationship of the supply chain's multi-flow structure, and the financing-level multi-relationship aggregation structure, thereby improving its ability to express cross-certificate correlation anomalies and fund path deviations, and making the risk characterization results of financing transactions more stable and reliable in terms of structural integrity and correlation consistency.

[0032] In this embodiment, the channel correlation distance matrix is ​​specifically as follows: Based on the financing and debt risk sub-graph, extract five relationship channels corresponding to the flow of commerce, invoices, goods, funds, and ownership. The transaction relationship edges within each relationship channel are classified and statistically analyzed to obtain the cumulative value of the amount closure difference, the cumulative value of the subject closure difference, the cumulative value of the time closure difference, and the cumulative value of the account closure difference. These values ​​are then arranged in the order of amount, subject, time, and account to construct a four-dimensional channel feature vector. For any two relational channels, calculate the difference between the values ​​of the corresponding dimensions in their four-dimensional channel feature vectors, and sum the absolute values ​​of the differences in each dimension to generate the channel association distance. Construct a set of channel association distances based on the channel association distances between all channels; Write the channel association distances between each pair of the five relation channels into the corresponding element positions of a five-row, five-column matrix to generate the initial channel association distance matrix; The five relation channels are sorted in ascending order of the sum of all channel association distances corresponding to each relation channel to generate the channel association order; The rows and columns of the channel association distance matrix are rearranged according to the channel association order to generate the rearranged channel association distance matrix.

[0033] In this implementation, the construction of the channel association distance matrix is ​​performed by the graph calculation module. The five-flow relationship channels correspond to independent subgraph data structures. Each subgraph aggregates transaction relationship edges according to the financing application number, and extracts the amount closure difference, subject closure difference, time closure difference, and account closure difference from the edge attributes to form a four-dimensional feature vector. The channel distance calculation is performed by accumulating the absolute value of the difference in the vector space dimension by dimension, and the result is written into a two-dimensional matrix storage structure. The matrix rearrangement process is completed by synchronously replacing rows and columns based on the sorting index. The rearrangement result is used for the channel weight adjustment in the subsequent five-flow relationship decoupling representation.

[0034] In this embodiment, step seven specifically includes: Read the risk characterization data of financing transactions and extract the risk characterization of financing applications, risk characterization of transaction relationship edges, and risk characterization of transaction paths; The risk representation of financing applications is decomposed into components to obtain node risk components, funding risk components, and debt risk components, and a financing application risk vector is generated. The risk representation of transaction relationships is classified by edge type to obtain commercial flow risk edge, ticket flow risk edge, goods flow risk edge, capital flow risk edge and ownership flow risk edge, and a set of relationship edge risks is generated. The risk representation of the transaction path is decomposed to obtain the capital flow path, the debt transfer path and the transaction closed loop path, and a path risk sequence is generated. Consistency comparison is performed based on the risk vector of financing application, the risk set of relation edge, and the risk sequence of path to generate a risk consistency offset value. Based on the distribution of historical risk consistency offset values ​​corresponding to similar financing applications, financing applications are classified into risk levels, and transaction security warning results are generated. The transaction security warning results include risk financing application identifiers, risk relationship edge identifiers, and risk path identifiers.

[0035] In this implementation, the risk representation data of financing transactions is output by the unified risk fusion module and then structured and parsed. A risk index is established according to the financing application number. The risk representation of financing application, risk representation of transaction relationship edge, and risk representation of transaction path are vectorized and normalized. The risk consistency offset value is obtained by calculating the sum of the differences between the corresponding dimensions of the node risk component, the relationship edge risk component, and the path risk component under the same financing application number. The magnitude of the offset value is used as the basis for risk level classification, and a set of early warning indicators for the corresponding financing application and related transaction relationship and path is generated.

[0036] In this embodiment, step eight specifically includes: Read the transaction security warning results, extract the risk financing application identifier, risk relationship edge identifier, risk path identifier, risk level, risk triggering reason and warning time, and generate a warning result field set; Based on the early warning result field set, and associated on-chain voucher records, financing claim risk subgraph identifiers, improved FRAUDRE model version identifiers, and financing transaction security risk characterization data summaries, an early warning evidence data package is generated; The warning evidence data packet is sorted by field and serialized by character to generate the warning evidence capsule text. The warning evidence capsule text is then hashed to generate the warning evidence capsule summary. Write the summary of the early warning evidence capsule, the risk financing application identifier, the early warning time, the signature entity identifier, and the storage address of the early warning evidence capsule into the blockchain to generate an on-chain early warning record; Read the block height, on-chain time, on-chain transaction number, and signature entity identifier from the on-chain warning record to generate an on-chain warning index; By associating the on-chain early warning index with the on-chain certificate record according to the risk financing application identifier, certificate summary and on-chain transaction number, an on-chain transaction security early warning evidence chain is generated.

[0037] In this implementation, the warning evidence capsule text is stored in the business platform's evidence library as an encrypted file, and the storage address of the warning evidence capsule is recorded as the file index address. The blockchain node only writes the warning evidence capsule summary, risk financing application identifier, warning time, and signature entity identifier. During verification, the capsule text corresponding to the file index address is read and the summary is recalculated. The recalculated summary is compared with the on-chain summary for consistency. The on-chain warning index establishes an association table according to the financing application number, certificate summary, and on-chain transaction number to form a traceable warning evidence chain.

[0038] Example 1: To verify the feasibility of this invention in practice, it was applied to a supply chain finance platform for automotive parts in East China. This platform serves a core OEM (Original Equipment Manufacturer) and its upstream Tier 1 and Tier 2 suppliers, offering services including accounts receivable financing, order financing, warehouse receipt financing, and factoring. In its daily operations, the platform integrates with the core enterprise's procurement system, supplier order system, electronic invoice system, warehouse management system, logistics receipt system, bank payment system, and factoring system. Since a single financing transaction often involves multiple stages such as contract confirmation, order generation, invoice issuance, warehouse receipt registration, logistics receipt, payment receipt, accounts receivable transfer, and financing application, the business data is scattered across multiple systems. Relying solely on single-document verification or manual rule-based review makes it difficult to promptly identify issues such as invoice-order mismatches, discrepancies between warehouse receipt pledges and logistics receipts, financing applications preceding the formation of underlying transaction documents, abnormal return of payment funds, and duplicate financing of the same accounts receivable.

[0039] In this implementation scenario, the platform first collects supply chain finance transaction data to generate a raw dataset. The collected data includes contract data, order data, invoice data, warehouse receipt data, logistics receipt data, payment data, financing application data, accounts receivable transfer data, guarantee data, and repayment data. The system uses the financing application number as the primary key to group contracts, orders, invoices, warehouse receipts, logistics, payments, debt transfers, guarantees, and repayment records under the same financing transaction into financing transaction data units. The system also standardizes the format of the enterprise name, unified social credit code, bank account, voucher number, amount, and time fields. For data from different systems, the system retains the source system identifier and original voucher number to ensure that the original business records can be located during subsequent risk tracing.

[0040] Subsequently, the platform generates a document summary for each transaction document and writes the document summary, subject signature, document type, transaction time, preceding document summary, and financing application number into the blockchain, generating an on-chain document record. The system further generates an on-chain document phase fingerprint based on block height, on-chain time, transaction time, and the connection relationship with preceding document summaries. The on-chain document phase fingerprint reflects the formation order and connection status of various documents in the same financing transaction on the blockchain, such as whether there is time reversal, on-chain delay, or preceding summary breakage among order documents, invoice documents, warehouse receipt documents, logistics receipt documents, payment documents, and financing application documents. In this way, the platform not only saves document summaries but also forms an on-chain temporal structure that can be recognized by the model.

[0041] In the five-flow closure analysis phase, the system categorizes transaction documents into commercial flow, invoice flow, goods flow, capital flow, and ownership flow based on the financing application number. Commercial flow corresponds to contracts and orders; invoice flow to invoices; goods flow to warehouse receipts and logistics receipts; capital flow to payments, financing application amounts, and repayments; and ownership flow to accounts receivable transfers and guarantees. The system calculates the differences between contract amounts, order amounts, invoice amounts, warehouse receipt values, financing application amounts, payment amounts, debt transfer amounts, guarantee amounts, and repayment amounts according to the order of business closure, and records the absolute value, direction, and source of the differences. Simultaneously, the system performs consistency comparisons on the transaction entities, transaction occurrence time, goods identifiers, transaction accounts, and ownership identifiers, generating five-flow closure difference data. This data is directly written into the edge attributes of the heterogeneous transaction graph in supply chain finance, enabling the graph structure to express the degree of business closure, rather than merely indicating whether a transaction relationship exists between entities.

[0042] In the graph construction phase, the system uses transaction entities, transaction accounts, transaction certificates, financing applications, and on-chain certificate records as nodes, and on-chain certificate relationships, on-chain phase relationships, and five-flow transaction relationships as edges to generate a heterogeneous transaction graph for supply chain finance. Starting with the financing application node, the system extracts the entity nodes, certificate nodes, account nodes, on-chain certificate record nodes, and corresponding relationship edges related to the financing application, forming a financing receivables risk subgraph. This risk subgraph simultaneously includes on-chain certificate phase relationships, five-flows closure difference relationships, fund flow relationships, and receivables transfer relationships. The system further generates financing receivables node features, financing receivables edge features, and financing receivables topological features, and inputs them into the improved FRAUDRE model.

[0043] During the model identification phase, the improved FRAUDRE model first performs on-chain phase constraint processing, writing the offset relationship between the transaction occurrence time sequence, block height sequence, and on-chain time sequence into the transaction relationship edge strength.

[0044] To verify the application effect of the present invention, the platform selected six consecutive months of anonymized business data for trial operation, and the experimental results are shown in Table 1.

[0045] Table 1. Data on the Trial Operation Results of Supply Chain Finance Transaction Security Early Warning System

[0046] As shown in Table 1, during the six-month trial period, 166 risk financing applications were ultimately confirmed manually, 87 were identified using the original rules, resulting in a coverage rate of 52.4%; the present invention identified 145, achieving a coverage rate of 87.3%. The present invention issued a total of 181 warnings, of which 145 were confirmed as genuine risks through manual review, resulting in a warning accuracy rate of 80.1%. Among the confirmed risks, the present invention identified 51 cases of fund repatriation risk and 27 cases of debt idling risk, demonstrating its ability to identify abnormal fund paths across multiple entities and abnormal accounts receivable transfers. The average advance warning time was 6.5 days, indicating that the present invention can provide early transaction security alerts before loan approval, repayment tracking, or manual verification.

[0047] In actual business reviews, the platform discovered that some risky transactions had invoice amounts matching order amounts, and the companies' credit status was within the normal range, so the original rules did not trigger warnings. However, this invention, through on-chain certificate phase fingerprinting, discovered that the financing application's on-chain time was earlier than the warehouse receipt generation time. Through five-flow closure difference data, it found an anomaly in the sequence of logistics receipt time and payment arrival time. Furthermore, it identified a path in the financing receivables over-edge data where the same bank account was used for cyclical payments and receipts between multiple suppliers, thus marking these transactions as high-risk in advance. Additionally, some suppliers submitted accounts receivable financing applications using different invoice numbers. Single-invoice reviews could not directly detect the risk of duplicate financing. This invention identifies receivables ownership conflicts through ownership flow channels and receivables transfer paths, and writes this risk into a warning evidence capsule, facilitating risk control personnel to trace the original evidence in the on-chain certificate records.

[0048] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A blockchain-based method for early warning of supply chain finance transaction security, characterized in that, Includes the following steps: Step 1: Collect supply chain finance transaction data and generate the original supply chain finance transaction dataset; Step 2: Generate a voucher summary based on the original dataset of the supply chain finance transaction, and write the voucher summary into the blockchain to generate on-chain voucher records and on-chain voucher phase fingerprints; Step 3: Based on the financing application number and related transaction vouchers, generate a five-flow closed-loop transaction unit and generate five-flow closed-loop difference data; Step 4: Construct a supply chain finance heterogeneous transaction graph based on the on-chain certificate records, on-chain certificate phase fingerprints, and five-flow closure difference data; Step 5: Extract the financing claim risk subgraph from the heterogeneous transaction graph of supply chain finance, and generate financing claim node features, financing claim edge features, and financing claim topological features; Step Six: Input the financing claim risk subgraph, financing claim node features, financing claim edge features, and financing claim topology features into the improved FRAUDRE model, and sequentially perform on-chain phase constraint processing, five-flow relationship decoupling processing, and financing claim super-edge modeling processing to generate financing transaction security risk representation data; Step 7: Generate transaction security early warning results based on the aforementioned financing transaction security risk characterization data; Step 8: Generate an early warning evidence capsule based on the transaction security early warning result, and write the early warning evidence capsule into the blockchain to generate an on-chain transaction security early warning evidence chain.

2. The blockchain-based supply chain finance transaction security early warning method according to claim 1, characterized in that, Step one specifically involves: The data collected includes supply chain finance transaction data, such as contract data, order data, invoice data, warehouse receipt data, logistics receipt data, payment data, financing application data, accounts receivable transfer data, guarantee data, and repayment data. Extract the transaction entity information, transaction voucher information, transaction amount information, and transaction time information from the supply chain finance transaction data, and generate transaction entity field sets, transaction voucher field sets, transaction amount field sets, and transaction time field sets respectively; The transaction entity field set, transaction voucher field set, transaction amount field set, and transaction time field set are uniformly formatted and associated according to the financing application number; The contracts, orders, invoices, warehouse receipts, logistics receipts, payments, accounts receivable transfers, guarantees, and repayment records corresponding to the same financing application number are collected to form a financing transaction record group; Based on the financing transaction record group, the relationships, fund transfer relationships, and debt transfer relationships between the transaction entities are recorded to form a financing transaction data unit; The original dataset for supply chain finance transactions is generated based on the financing transaction data unit.

3. The blockchain-based supply chain finance transaction security early warning method according to claim 1, characterized in that, Step two specifically involves: Read the transaction voucher records in the original dataset of the supply chain finance transactions, and generate a voucher basic field group according to the financing application number, voucher type, voucher number, transaction entity, transaction amount, transaction time and associated voucher number; The basic field group of the voucher is sorted and characters are concatenated to form the voucher summary text. The voucher summary text is then hashed to generate the voucher summary corresponding to each transaction voucher. The transaction voucher records are arranged in chronological order of the business occurrence under the same financing application number, and the previous voucher summary in the adjacent transaction voucher records is written into the next transaction voucher record to generate the previous voucher summary; Write the certificate summary, subject signature, certificate type, transaction occurrence time, previous certificate summary and financing application number into the blockchain to generate an on-chain certificate record containing block height, on-chain time, on-chain transaction number and the written subject identifier; The on-chain certificate records are collected according to the financing application number, and an on-chain certificate sequence is generated according to the block height, on-chain time, transaction time and connection relationship of the previous certificate summary; Based on the on-chain credential sequence, the block height sequence, the on-chain time sequence, the credential type sequence, the preceding digest connection status, and the main signature status are extracted to generate an on-chain credential phase fingerprint.

4. The blockchain-based supply chain finance transaction security early warning method according to claim 1, characterized in that, Step three specifically involves: Read transaction vouchers with the same financing application number from the original dataset of supply chain finance transactions, and generate a set of financing application vouchers; According to the type of voucher, the set of financing application vouchers is divided into commercial flow data, invoice flow data, goods flow data, capital flow data, and ownership flow data. The commercial flow data includes contract data and order data, the invoice flow data includes invoice data, the goods flow data includes warehouse receipt data and logistics receipt data, the capital flow data includes payment data, financing application amount data, and repayment data, and the ownership flow data includes accounts receivable transfer data and guarantee data. Based on the financing application number, transaction entity, transaction amount, transaction time, transaction account, goods identifier, and ownership identifier, the fields of the commercial flow data, invoice flow data, goods flow data, capital flow data, and ownership flow data are aligned to generate a five-flow closed transaction unit; For the contract amount, order amount, invoice amount, warehouse receipt value, financing application amount, payment amount, debt transfer amount, guarantee amount, and repayment amount in the five-flow closed transaction unit, the difference is calculated according to the business closure sequence of contract to order, order to invoice, invoice to warehouse receipt value, warehouse receipt value to financing application amount, financing application amount to payment amount, payment amount to repayment amount, debt transfer amount to financing application amount, and guarantee amount to financing application amount. The comparison field, benchmark field, absolute value of the difference, direction of the difference, and source of the field are recorded to generate amount closure difference data. A consistency comparison is performed on the transaction entity, transaction time, goods identifier, transaction account, and ownership identifier in the five-flow closed transaction unit to generate entity closure difference data, time closure difference data, goods closure difference data, account closure difference data, and ownership closure difference data, respectively. The amount closure difference data, entity closure difference data, time closure difference data, goods closure difference data, account closure difference data, and ownership closure difference data are written into the five-flow closure transaction unit according to the financing application number to generate five-flow closure difference data.

5. The blockchain-based supply chain finance transaction security early warning method according to claim 1, characterized in that, Step four specifically involves: Read the on-chain certificate record, the on-chain certificate phase fingerprint, and the five-flow closure difference data, extract the transaction entity identifier, transaction account identifier, transaction certificate identifier, financing application number, and on-chain transaction number, and generate transaction entity node, transaction account node, transaction certificate node, financing application node, and on-chain certificate record node; Based on the connection relationship of the certificate summary, on-chain transaction number, block height, on-chain time and previous certificate summary, the transaction certificate node and the on-chain certificate record node are connected to generate certificate on-chain relationship edge and on-chain phase relationship edge; According to the classification of commercial flow, ticket flow, goods flow, capital flow and ownership flow, the transaction entity node, transaction account node, transaction voucher node and financing application node are connected to generate five-flow transaction relationship edges, and the five-flow closure difference data is written into the corresponding five-flow transaction relationship edges; The on-chain relationship edges, on-chain phase relationship edges, and five-flow transaction relationship edges of the vouchers are marked with relationship types to generate a heterogeneous relationship edge set. Based on the transaction subject node, transaction account node, transaction voucher node, financing application node, on-chain voucher record node, and the heterogeneous relationship edge set, a supply chain finance heterogeneous transaction graph is generated.

6. The blockchain-based supply chain finance transaction security early warning method according to claim 1, characterized in that, Step five specifically involves: Read the financing application node in the heterogeneous transaction graph of supply chain finance, and retrieve the transaction entity node, transaction voucher node, transaction account node and on-chain voucher record node connected to the financing application node according to the financing application number to generate a set of financing application associated nodes; Read the five-flow transaction relationship edges, on-chain certificate relationship edges, and on-chain phase relationship edges among the set of nodes associated with the financing application, and generate a set of edges associated with the financing application; A financing claim risk subgraph is constructed based on the set of associated nodes and the set of associated edges of the financing application. The financing claim risk subgraph includes a subject relationship subgraph, a certificate relationship subgraph, an account relationship subgraph, an on-chain phase subgraph, and a five-flow relationship subgraph. The nodes in the financing claim risk subgraph are feature-encoded to generate financing claim node features, which include node type, subject role, voucher type, account type, financing application number, transaction amount, business occurrence time, on-chain sequence, and five-flow category; The edges in the financing claims risk subgraph are feature-encoded to generate financing claims edge features. The financing claims edge features include edge type, edge direction, relationship source, amount closure difference, subject closure difference, time closure difference, account closure difference, ownership closure difference, and on-chain phase connection state. The financing debt risk subgraph is subjected to topological statistics to generate financing debt topological features, which include the number of node types, the number of edge types, the length of the five-flow path, the length of the on-chain phase path, the number of shared accounts, the number of duplicate vouchers, the number of fund return paths, and the number of debt transfer paths.

7. The blockchain-based supply chain finance transaction security early warning method according to claim 1, characterized in that, Step six specifically involves: The financing claim risk subgraph, financing claim node features, financing claim edge features, and financing claim topological features are input into the improved FRAUDRE model. The improved FRAUDRE model includes sequentially connected on-chain phase constraint units, five-flow relationship decoupling units, and financing claim superedge modeling units. The on-chain phase constraint unit reads the block height sequence, on-chain time sequence, and business occurrence time sequence from the on-chain phase relationship edge. It aligns the same transaction voucher node according to the sequence index, calculates the absolute value of the difference between the business occurrence time sequence and the block height sequence at the corresponding index position, and the absolute value of the difference between the business occurrence time and the on-chain time at the corresponding index position. It then sums the two differences with equal weight to generate the on-chain phase constraint edge strength and writes it into the transaction relationship edge. The five-flow relationship decoupling unit divides the financing and debt risk sub-graph according to the flow of commerce, invoices, goods, funds, and ownership; The cumulative values ​​of the amount closure difference, subject closure difference, time closure difference, and account closure difference within each relationship channel are calculated to form a four-dimensional channel feature vector; The absolute values ​​of the component differences between each pair of feature vectors are summed to generate the channel correlation distance, and the channel correlation distance matrix is ​​formed by filling all the channel correlation distances. A five-flow relationship decoupling representation is generated based on the channel association distance matrix; The financing claim superedge modeling unit aggregates transaction relationship edges according to the same financing application number to generate financing claim superedges, and calculates the sum of the absolute values ​​of the differences in each dimension between the five-flow relationship decoupling representation and the superedge central feature corresponding to each relationship edge to generate the relationship edge risk contribution value. The risk contribution values ​​of the relationship edges are sorted from largest to smallest, weighted, and accumulated, and then mapped to the financing application node, transaction relationship edge, and transaction path to output financing transaction security risk characterization data.

8. A blockchain-based supply chain finance transaction security early warning method according to claim 7, characterized in that, The channel correlation distance matrix is ​​specifically as follows: Based on the aforementioned financing and debt risk sub-graph, five relationship channels corresponding to the flow of commerce, invoices, goods, funds, and ownership are extracted; The transaction relationship edges within each relationship channel are classified and statistically analyzed to obtain the cumulative value of the amount closure difference, the cumulative value of the subject closure difference, the cumulative value of the time closure difference, and the cumulative value of the account closure difference. These values ​​are then arranged in the order of amount, subject, time, and account to construct a four-dimensional channel feature vector. For any two relational channels, calculate the difference between the values ​​of the corresponding dimensions in their four-dimensional channel feature vectors, and sum the absolute values ​​of the differences in each dimension to generate the channel association distance. Construct a set of channel association distances based on the channel association distances between all channels; Write the channel association distances between each pair of the five relation channels into the corresponding element positions of a five-row, five-column matrix to generate the initial channel association distance matrix; The five relation channels are sorted in ascending order of the sum of all channel association distances corresponding to each relation channel to generate the channel association order; The rows and columns of the channel association distance matrix are rearranged according to the channel association order to generate a rearranged channel association distance matrix.

9. A blockchain-based supply chain finance transaction security early warning method according to claim 1, characterized in that, Step seven specifically involves: Read the financing transaction security risk characterization data and extract the financing application risk characterization, transaction relationship edge risk characterization, and transaction path risk characterization; The risk representation of the financing application is decomposed into components to obtain the node risk component, the funding risk component, and the debt risk component, and a financing application risk vector is generated. The risk representation of the transaction relationship edge is divided into edge types to obtain the commercial flow risk edge, ticket flow risk edge, goods flow risk edge, capital flow risk edge and ownership flow risk edge, and a relationship edge risk set is generated. The risk characterization of the transaction path is decomposed to obtain the capital flow path, debt transfer path and transaction closed loop path, and a path risk sequence is generated. A consistency comparison is performed based on the financing application risk vector, the relationship edge risk set, and the path risk sequence to generate a risk consistency offset value. Based on the historical risk consistency offset value distribution corresponding to similar financing applications, the financing applications are classified into risk levels, and transaction security warning results are generated. The transaction security warning results include risk financing application identifiers, risk relationship edge identifiers, and risk path identifiers.

10. A blockchain-based supply chain finance transaction security early warning method according to claim 1, characterized in that, Step eight specifically involves: Read the transaction security warning results, extract the risk financing application identifier, risk relationship edge identifier, risk path identifier, risk level, risk triggering reason and warning time, and generate a warning result field set; Based on the aforementioned warning result field set, and by associating the on-chain certificate record, the financing claim risk subgraph identifier, the improved FRAUDRE model version identifier, and the financing transaction security risk characterization data summary, a warning evidence data package is generated; The warning evidence data packet is sorted by field and serialized by character to generate warning evidence capsule text, and the warning evidence capsule text is hashed to generate warning evidence capsule summary; Write the summary of the early warning evidence capsule, the risk financing application identifier, the early warning time, the signature entity identifier, and the storage address of the early warning evidence capsule into the blockchain to generate an on-chain early warning record; Read the block height, on-chain time, on-chain transaction number, and signature entity identifier from the on-chain warning record to generate an on-chain warning index; By associating the on-chain early warning index with the on-chain credential record according to the risk financing application identifier, credential summary and on-chain transaction number, an on-chain transaction security early warning evidence chain is generated.

Images