A freight service verification method and system based on big data analysis

Abstract

The application discloses a freight service verification method and system based on big data analysis, which comprises the following steps: collecting waybill, contract, track, fund and bill data to form a waybill verification record; mapping five-flow nodes and building relationship edges to form a five-flow relationship graph; comparing the reliable data according to rules to form a five-flow residual account book; inputting an improved R-GCN model to generate a gating coefficient and propagate the residual; aggregating the residual and performing deduction or reservation to form a propagated residual state; putting the non-returned voucher into a delay queue, writing back and updating and extracting a subgraph; recalculating the relationship convolution of the subgraph to output a verification result. The application relates to the technical field of freight service data verification, improves the freight verification accuracy and abnormal trace efficiency, and reduces the manual review cost.

Description

Technical Field

[0001] This invention relates to the field of freight business data verification technology, and in particular to a freight business verification method and system based on big data analysis. Background Technology

[0002] Current online freight business verification methods primarily combine data reporting, rule verification, and manual review. This involves collecting basic data such as waybill data, contract data, trajectory data, fund flow data, and invoice data, and combining this with vehicle qualifications, personnel qualifications, trajectory positioning, bank statements, and invoice verification data to verify the authenticity of freight transactions. Existing solutions can handle single-field comparisons, anomaly marking, result display, and manual processing, making them suitable for basic regulatory scenarios.

[0003] Existing verification methods primarily rely on separate verification of waybill, tracking, payment, and document fields, lacking a five-flow relationship diagram centered on the same waybill number. This makes it difficult to propagate and close the relationships between information flow, contract flow, transportation flow, capital flow, and document flow. When third-party trusted data sources return data late, existing solutions easily create isolated anomaly records. Subsequent appeals and supplementary documentation require re-execution of numerous verification processes, lacking an incremental recalculation mechanism based on affected waybill sub-graphs, resulting in insufficient verification efficiency, incomplete anomaly tracing, and inaccurate cross-flow anomaly identification. Therefore, providing a freight business verification method and system based on big data analysis is a problem urgently needing to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a freight business verification method and system based on big data analysis. This invention utilizes a five-flow relationship diagram, a five-flow residual ledger, and an improved R-GCN model to perform correlation verification of waybills, contracts, trajectories, fund flows, and invoice data. Combined with delay verification and appeal and supplementary evidence writing mechanisms, it has the advantages of high verification accuracy, complete anomaly tracing, and high processing efficiency.

[0005] According to an embodiment of the present invention, a freight business verification method and system based on big data analysis includes the following steps: Receive freight business reported data, collect basic data of waybill, contract data, trajectory data, fund flow data and bill data according to waybill number, and form waybill verification record; The waybill verification record is mapped to five-flow nodes, and relationship edges are established according to the subject, time, amount, trajectory and voucher fields to bind rule numbers, forming a five-flow relationship graph; Read the verification rule according to the rule number, compare the fields of the nodes at both ends of the relationship edge with the fields returned by the third-party trusted data source, and register a no-return flag when no returned field is read. Write the comparison difference and the no-return flag into the five types of residuals to form a five-stream residual ledger. The five-flow relation graph and the five-flow residual ledger are input into the improved R-GCN model. The relation convolution weight matrix is ​​read according to the relation edge type. The relation edge gating coefficients are generated based on the residual value, data source and trusted certificate status. The residual messages of adjacent nodes are weighted and propagated to obtain the relation propagation residual. Propagate residuals by aggregation relationship based on waybill number, deduct residual items whose trusted document status is returned and whose deductible document mark matches, and retain residual items whose trusted document status is not returned or whose deductible document mark does not match, thus forming waybill propagation residual status; Write the relationship edge with the trusted credential status not returned to the delayed verification queue. After the data returned by the third-party trusted data source and the appeal and supplementary evidence data are approved, update the five-flow relationship graph and the five-flow residual ledger, and extract the affected waybill graph. The affected waybill subgraph is re-input into the improved R-GCN model to perform relational convolution, and the freight business verification results are output based on the propagation residual state of the unclosed waybill.

[0006] Optionally, the operation of generating the waybill verification record includes: Parse the freight business reporting data and extract the waybill number, reporting stage marker, and field source marker; A collection line is established based on the waybill number, and the collection line is divided into a waybill basic data area, a contract data area, a trajectory data area, a fund flow data area, and a bill data area; The parsed field records are assigned to the corresponding data areas according to the field source markers; The five data areas in the same collection line are merged to form a waybill verification record carrying the waybill number, reporting stage mark, and field source mark.

[0007] Optionally, the operation of forming the five-flow relationship diagram includes: Read the waybill number, field source marker, and field record from the waybill verification record, and generate information flow node, contract flow node, transportation flow node, fund flow node, and bill flow node according to the field source marker; Extract the subject field, time field, amount field, trajectory field, loading / unloading location field, and voucher number field from the five types of nodes; Under the same waybill number, generate subject relationship edges based on the subject field value and subject correspondence table, generate time relationship edges based on the order of time fields, generate amount relationship edges based on the amount fields of contract flow nodes, fund flow nodes and bill flow nodes, generate trajectory relationship edges based on the trajectory field and loading / unloading location field, and generate voucher relationship edges based on the voucher number field value and voucher correspondence table. Register the waybill number, origin node number, destination node number, relationship edge type, and rule number for each relationship edge to form a five-flow relationship diagram.

[0008] Optionally, the operation of forming the five-stream residual ledger includes: Read the edge number, edge type, starting node number, ending node number, and rule number from the five-flow relationship graph; Retrieve the corresponding verification rule according to the rule number, and determine the fields to be compared, the third-party trusted data source, and the deductible voucher marker according to the verification rule; Read the node fields corresponding to the starting node number and the ending node number, and read the fields returned by the third-party trusted data source or the flag that no data was returned; When reading the fields returned by the trusted third-party data source, the node fields are compared with the fields returned by the trusted third-party data source to obtain the comparison differences; When a "not returned" flag is read, the "not returned" flag is used as a comparison difference of the corresponding relation edge, and the trusted credential status is registered as "not returned". When the relationship edge type is a subject relationship edge, the comparison difference will be registered as a subject residual; when the relationship edge type is a time relationship edge, the comparison difference will be registered as a time residual; when the relationship edge type is an amount relationship edge, the comparison difference will be registered as an amount residual; when the relationship edge type is a trajectory relationship edge, the comparison difference will be registered as a trajectory residual; when the relationship edge type is a voucher relationship edge, the comparison difference will be registered as a negotiable instrument residual. For each residual, record the waybill number, relation edge number, residual type, residual value, data source, credible voucher status, deductible voucher mark, and closure mark to form a five-stream residual ledger.

[0009] Optionally, the operation of generating relational edge gating coefficients includes: For each relation edge in the five-flow relation graph, a corresponding gating processing record is established. The relation edge number, relation edge type, residual type, residual value, data source, and trusted credential status are recorded in the gating processing record. Based on the type of relation edge, select the corresponding relation convolution weight matrix in the improved R-GCN model, and register the matrix number of the relation convolution weight matrix in the gating processing record; Read the corresponding residual interval sequence according to the residual type, compare the residual value with the lower limit and upper limit of the interval in the residual interval sequence in turn, and write the interval level corresponding to the interval to which the residual value belongs as the residual control quantity; Read the corresponding data source level according to the data source, and write the data source level as the source control quantity; Read the corresponding document status level based on the trusted document status, and write the document status level as a document control quantity; The gate control values ​​are read according to the arrangement of the residual control quantity, source control quantity, and voucher control quantity in the gate control value sequence, and the gate control values ​​are determined as the relation edge gate control coefficients. By binding the relation edge gating coefficients with the relation edge numbers and the matrix numbers of the relation convolution weight matrix, a gating relation edge record is formed for weighted propagation in the improved R-GCN model.

[0010] Optionally, the operation of forming the waybill propagation residual state includes: Extract the relation edges from the five-flow relation graph by waybill number, extract the relation convolution weight matrix and relation edge gating coefficient from the gated relation edge record by relation edge number, and extract the residual type, residual value, reliable voucher status, deductible voucher mark and closure mark from the five-flow residual ledger. Generate adjacent node residual messages based on the starting node number, ending node number, residual type, and residual value of the relation edge; The residual messages of adjacent nodes are weighted and propagated according to the relation convolution weight matrix and relation edge gating coefficients to generate gated residual messages; the gated residual messages are aggregated according to waybill number and residual type to form relation propagation residuals. Perform residual deduction on residual items whose trusted voucher status is returned and whose deductible voucher tag corresponds to the residual type; retain residual values ​​for residual items whose trusted voucher status is not returned or whose deductible voucher tag does not correspond to the residual type. Update the closure marker based on the residual value after deduction, and summarize the residual type, relational propagation residual, deduction result and closure marker by waybill number to form the waybill propagation residual status.

[0011] Optionally, the operation of writing the relation edge with a trusted credential status of "not returned" to the delayed verification queue includes: Extract residual items from the waybill propagation residual status that are marked as unclosed and whose trusted document status is not returned, and read the waybill number, relation edge number, residual type, residual value, data source and deductible document mark corresponding to the residual item; Generate a delay verification record using the waybill number and the relationship edge number. The delay verification record records the residual type, residual value, data source, deductible voucher mark, queuing time, number of retries and queue status. Mark the queue status as pending verification and set the retry count to the initial count; After the trusted third-party data source returns data, locate the delayed verification record according to the data source and relation edge number, and fill the fields returned by the trusted third-party data source into the corresponding node and corresponding relation edge; The verification rules are reread according to the relation edge number. The fields of the nodes at both ends of the updated relation edge are compared with the fields returned by the third-party trusted data source. The residual value corresponding to the same relation edge number in the five-flow residual ledger is replaced. The trusted voucher status is updated to "returned". The closure mark and queue status are updated according to the replaced residual value.

[0012] Optionally, the operation of updating the five-flow relationship diagram and the five-flow residual ledger after the appeal and supplementary evidence data is approved includes: Extract residual items marked as unclosed from the waybill propagation residual status, and read the waybill number, relation edge number, residual type, residual value and deductible voucher mark corresponding to the residual item; Receive appeal and supplementary certificate data corresponding to the waybill number, and parse the supplementary certificate type, supplementary certificate field and review status in the appeal and supplementary certificate data; When the review status is approved, generate a supplementary certificate node number based on the waybill number and the supplementary certificate type, and register the supplementary certificate field as a supplementary certificate node; Locate the corresponding relation edge in the five-flow relation graph according to the relation edge number, and connect the supplementary proof node to the start node and end node of the corresponding relation edge to form a supplementary proof relation edge; When the type of supplementary voucher is consistent with the deductible voucher mark, update the status of the reliable voucher corresponding to the same relation edge number in the five-flow residual ledger to "returned" and register the type of supplementary voucher as the deductible voucher mark. The verification rules are reread according to the relation edge number. The supplementary verification node, the start node and end node of the corresponding relation edge are compared. The residual value corresponding to the same relation edge number in the five-flow residual ledger is replaced. The closure mark is updated according to the replaced residual value.

[0013] Optionally, the operation of extracting the affected waybill subgraph and re-performing relational convolution includes: Read the relation edge numbers whose trusted voucher status is updated to "returned", residual value is replaced, and closure mark is updated from the updated five-flow residual ledger, and generate an updated relation edge list; Locate the waybill number, start node number, and end node number of the corresponding relation edge in the five-flow relation graph according to the updated relation edge list; Using the starting node number and the ending node number as the center, extract the adjacent first-order nodes and adjacent first-order relation edges that are directly connected to the starting node number and the ending node number under the same waybill number. The updated relation edges, the supplementary proof relation edges, the adjacent first-order nodes, and the adjacent first-order relation edges are merged to form the affected waybill graph; Read the gated relationship edge records and the five-flow residual ledger according to the relationship edge number in the affected waybill subgraph, and regenerate the residual messages of adjacent nodes; The residual messages of adjacent nodes are input into the improved R-GCN model to perform relational convolution, update the relational propagation residual, deduction result and closure mark under the corresponding waybill number, and form the updated waybill propagation residual state.

[0014] A freight business verification system based on big data analysis according to an embodiment of the present invention includes the following modules: The data collection module is used to collect freight business reported data by waybill number and form waybill verification records; The five-flow mapping module is used to map waybill verification records into a five-flow relationship diagram; The residual ledger generation module is used to generate five-stream residual ledgers according to the verification rules and the fields returned by the trusted third-party data source; The R-GCN verification module has been improved to generate relational edge gating coefficients based on the five-flow relational graph and the five-flow residual ledger, perform relational convolution, and form the waybill propagation residual state; The delayed verification module is used to write the relationship edges with the trusted credential status of not returned to the delayed verification queue, and update the five-flow relationship graph and the five-flow residual ledger after the third-party trusted data source returns the data; The supplementary evidence write-back module is used to generate supplementary evidence nodes and supplementary evidence relationship edges after the appeal supplementary evidence data is approved, and to update the five-flow relationship graph and the five-flow residual ledger; The results output module is used to output the freight business verification results based on the updated waybill propagation residual status.

[0015] The beneficial effects of this invention are: (1) This invention maps the basic data of waybill, contract data, trajectory data, capital flow data and bill data into a five-flow relationship diagram, and establishes relationship edges through the subject, time, amount, trajectory and voucher fields, so that the scattered freight business data can be associated and verified under the same waybill number, which improves the consistency and integrity of cross-data source verification.

[0016] (2) This invention records the main residual, time residual, amount residual, trajectory residual and bill residual through the five-flow residual ledger, and inputs the five-flow relationship diagram and the five-flow residual ledger into the improved R-GCN model for gating propagation and residual deduction, so that individual anomalies can be transmitted and closed in the relationship, thereby improving the accuracy of anomaly identification and the ability to locate the cause of anomalies.

[0017] (3) The present invention writes the relationship edge that the trusted certificate has not been returned into the delayed verification queue, and updates the five-flow relationship graph and the five-flow residual ledger after the data returned by the third-party trusted data source and the appeal and supplementary certificate data are approved. The relationship convolution is re-executed on the affected waybill subgraph, reducing full duplicate verification, improving the verification processing efficiency and the traceability of the verification results. Attached Figure Description

[0018] 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 The flowchart shows a freight business verification method and system based on big data analysis proposed in this invention. Figure 2 This is a schematic diagram of the five-flow relationship of a freight business verification method and system based on big data analysis proposed in this invention. Figure 3 This invention presents a flowchart of the R-GCN verification and write-back process for a freight business verification method and system based on big data analysis. Detailed Implementation

[0019] 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.

[0020] refer to Figures 1-3 A method and system for verifying freight operations based on big data analysis, comprising the following steps: The system receives freight business reports and aggregates basic waybill data, contract data, trajectory data, fund flow data, and invoice data according to waybill number to form waybill verification records. When receiving freight business reports, the system parses the waybill number, reporting stage marker, field source marker, and field record for each report. Using the waybill number as the aggregation index, field records under the same waybill number are grouped into the same aggregation row. The aggregation row is divided into a waybill basic data area, a contract data area, a trajectory data area, a fund flow data area, and an invoice data area. The field source marker is used to determine the data area to which the field record belongs. The waybill verification record is mapped to five-flow nodes, and relationship edges with binding rule numbers are established according to the subject, time, amount, trajectory, and voucher fields to form a five-flow relationship graph. The five-flow nodes include information flow nodes, contract flow nodes, transportation flow nodes, capital flow nodes, and document flow nodes. Each five-flow node registers the node number, node type, waybill number, and field set. The relationship edges include subject relationship edges, time relationship edges, amount relationship edges, trajectory relationship edges, and voucher relationship edges. Each relationship edge registers the starting node number, ending node number, relationship edge type, and rule number. The verification rules are read according to the rule number. The fields of the nodes at both ends of the relation edge are compared with the fields returned by the third-party trusted data source. If no returned field is read, a no-return flag is registered. The comparison difference and the no-return flag are written into the five types of residuals to form a five-flow residual ledger. The verification rules register the fields to be compared, the third-party trusted data source, the field comparison method, and the deductible voucher flag. The five types of residuals include subject residuals, time residuals, amount residuals, trajectory residuals, and document residuals. The five-flow residual ledger registers the waybill number, relation edge number, residual type, residual value, data source, trusted voucher status, deductible voucher flag, and closure flag. The five-flow relationship graph and five-flow residual ledger are input into the improved R-GCN model. The relationship convolution weight matrix is ​​read according to the relationship edge type. The relationship edge gating coefficient is generated based on the residual value, data source, and trusted document status. The residual messages of adjacent nodes are weighted and propagated to obtain the relationship propagation residual. The improved R-GCN model takes the relationship edge as the processing object and configures the relationship convolution weight matrix for the main relationship edge, time relationship edge, amount relationship edge, trajectory relationship edge, and document relationship edge respectively. The residual control quantity is formed according to the residual value corresponding to the relationship edge, the source control quantity is formed according to the data source, and the document control quantity is formed according to the trusted document status. The relationship edge gating coefficient is determined by the residual control quantity, the source control quantity, and the document control quantity. The residuals are propagated by aggregation relationship according to the waybill number. Residual items with the trusted document status returned and the deductible document mark matching are deducted. Residual items with the trusted document status not returned or the deductible document mark not matching are retained, forming the waybill propagation residual status. The waybill propagation residual status records the relationship propagation residuals, deduction results and closure marks of each residual type under the same waybill number. When the residual value is zero after residual deduction, the closure mark is updated to closed. When the residual value is not zero after residual deduction, the closure mark is updated to unclosed. Write the relationship edge with the trusted credential status not returned to the delayed verification queue. After the data returned by the third-party trusted data source and the appeal and supplementary certificate data are approved, update the five-flow relationship graph and the five-flow residual ledger, and extract the affected waybill subgraph. The delayed verification queue generates delayed verification records with waybill number and relationship edge number. After the appeal and supplementary certificate data are approved, supplementary certificate nodes and supplementary certificate relationship edges are generated. The affected waybill subgraph consists of the updated relationship edge, the supplementary certificate relationship edge, the adjacent first-order node, and the adjacent first-order relationship edge. The affected waybill subgraph is re-input into the improved R-GCN model to perform relational convolution. Based on the unclosed waybill propagation residual state, the freight business verification result is output. When re-performing relational convolution, only adjacent node residual messages are generated for nodes and relation edges in the affected waybill subgraph, and the relation propagation residual, deduction result, and closure mark under the corresponding waybill number are updated. The freight business verification result includes normal verification result, abnormal waybill record, abnormal relation edge, residual type, and residual value.

[0021] In this embodiment, the operation of generating a waybill verification record includes: Parse the freight business reporting data to extract the waybill number, reporting stage marker, and field source marker. During parsing, read the field name and field value according to the message structure of the freight business reporting data. The waybill number is used to identify the freight business object, the reporting stage marker is used to distinguish between pre-shipment reporting, in-shipment reporting, and post-shipment reporting, and the field source marker is used to distinguish the data area to which the field record belongs. A collection line is established using the waybill number. Within the collection line, four data areas are defined: waybill basic data area, contract data area, trajectory data area, fund flow data area, and invoice data area. The collection line uses the waybill number as an index. Each data area within the collection line registers its data area number, field source marker, field name, and field value. The waybill basic data area stores the shipper's identifier, carrier's identifier, driver's identifier, vehicle identifier, loading location, unloading location, cargo name, and planned transportation time. The contract data area stores the contract number, contract parties, contract amount, and contract document number. The trajectory data area stores the departure time, arrival time, and trajectory point sequence. The fund flow data area stores the payer, payee, payment amount, payment time, and fund flow number. The invoice data area stores the invoice code, invoice number, invoice issuer, invoice recipient, invoice amount, and invoice issuance time. The parsed field records are assigned to the corresponding data areas according to the field source marker. When the field source marker is "waybill basic", the field record is assigned to the waybill basic data area; when the field source marker is "contract", the field record is assigned to the contract data area; when the field source marker is "track", the field record is assigned to the track data area; when the field source marker is "funds flow", the field record is assigned to the funds flow data area; when the field source marker is "invoice", the field record is assigned to the invoice data area. Five data areas in the same aggregation row are merged to form a waybill verification record carrying the waybill number, reporting stage marker, and field source marker. During the merge, the field records of the five data areas under the same waybill number are read, and the data area number, field name, field value, reporting stage marker, and field source marker are jointly registered in the waybill verification record. The waybill verification record serves as the data source for constructing the five-flow relationship diagram.

[0022] In this embodiment, the operation of forming the five-flow relationship diagram includes: Read the waybill number, field source marker, and field record from the waybill verification record, and generate information flow nodes, contract flow nodes, transportation flow nodes, fund flow nodes, and bill flow nodes according to the field source marker. When generating nodes, generate node numbers according to the waybill number, field source marker, and node type. Compile the field records corresponding to the waybill basic data area into information flow nodes, the field records corresponding to the contract data area into contract flow nodes, the field records corresponding to the trajectory data area into transportation flow nodes, the field records corresponding to the fund flow data area into fund flow nodes, and the field records corresponding to the bill data area into bill flow nodes. Extract the following fields from five types of nodes: subject field, time field, amount field, trajectory field, loading / unloading location field, and voucher number field. The subject field includes cargo owner identifier, carrier identifier, driver identifier, vehicle identifier, payer, payee, invoice issuer, and invoice recipient. The time field includes order creation time, order acceptance time, departure time, arrival time, payment time, and invoice issuance time. The amount field includes contract amount, payment amount, and invoice amount. The trajectory field includes the sequence of trajectory points. The loading / unloading location field includes loading location and unloading location. The voucher number field includes waybill number, contract number, fund flow number, invoice code, and invoice number. Under the same waybill number, entity relationship edges are generated based on the entity field value and entity correspondence table; time relationship edges are generated based on the order of time fields; amount relationship edges are generated based on the amount fields of contract flow nodes, fund flow nodes, and bill flow nodes; trajectory relationship edges are generated based on the trajectory field and loading / unloading location field; and document relationship edges are generated based on the document number field value and document correspondence table. The entity correspondence table records the entity field name, corresponding entity field name, and allowed correspondence status. Entity relationship edges are generated when entity field values ​​are the same or the allowed correspondence status is valid. Time fields are arranged from first to last according to their time values, and adjacent time field nodes are connected to generate time relationship edges. Amount relationship edges connect corresponding nodes according to the verification direction of contract amount, payment amount, and invoice amount. Trajectory relationship edges connect the node belonging to the trajectory field with the node belonging to the loading location field and the node belonging to the unloading location field. The document correspondence table records the document number field name, corresponding document number field name, and allowed correspondence status. Document relationship edges are generated when document number field values ​​are the same or the allowed correspondence status is valid. Record the waybill number, origin node number, destination node number, relationship edge type, and rule number for each relationship edge to form a five-flow relationship graph. Relationship edge types include subject relationship edges, time relationship edges, amount relationship edges, trajectory relationship edges, and voucher relationship edges. The rule number is determined by the relationship edge type and the field to be compared. The five-flow relationship graph is stored in a node table and a relationship edge table.

[0023] In this embodiment, the operations for forming the five-stream residual ledger include: Read the relation edge number, relation edge type, start node number, end node number, and rule number from the five-flow relation graph; use the relation edge number as the residual generation index, and process the relation edges one by one according to the relation edge registration order in the five-flow relation graph; Retrieve the corresponding verification rule according to the rule number, and determine the fields to be compared, the third-party trusted data source, and the deductible voucher mark according to the verification rule; the verification rule registers the rule number, relationship edge type, fields to be compared, third-party trusted data source name, field comparison method, and deductible voucher mark; the field comparison methods include field consistency comparison, time sequence comparison, amount difference comparison, trajectory location comparison, and voucher number comparison. Read the node fields corresponding to the starting node number and the ending node number, and read the fields returned by the third-party trusted data source or the non-return flag; the third-party trusted data source includes vehicle qualification data source, personnel qualification data source, trajectory data source, bank transaction data source and invoice verification data source; when a field returned by the third-party trusted data source is read, the trusted credential status is registered as returned; when no field returned by the third-party trusted data source is read, the trusted credential status is registered as not returned; When reading fields returned by a trusted third-party data source, the node fields are compared with the fields returned by the trusted third-party data source to obtain the comparison differences; field consistency comparison is used for the main field, and the comparison result forms the main field difference; time sequence comparison is used for the time field, and the comparison result forms the time field difference; amount difference comparison is used for the amount field, and the comparison result forms the amount field difference; trajectory location comparison is used for the trajectory field and the loading / unloading location field, and the comparison result forms the trajectory field difference; voucher number comparison is used for the voucher number field, and the comparison result forms the voucher field difference; When a non-return flag is read, the non-return flag is used as the comparison difference of the corresponding relation edge, and the trusted credential status is registered as non-return; the non-return flag is bound to the relation edge number, and the relation edge corresponding to the non-return flag enters the delayed verification queue in subsequent steps; When the relationship edge type is a subject relationship edge, the comparison difference is recorded as a subject residual; when the relationship edge type is a time relationship edge, the comparison difference is recorded as a time residual; when the relationship edge type is a monetary relationship edge, the comparison difference is recorded as a monetary residual; when the relationship edge type is a trajectory relationship edge, the comparison difference is recorded as a trajectory residual; when the relationship edge type is a voucher relationship edge, the comparison difference is recorded as a document residual. The subject residual, time residual, monetary residual, trajectory residual, and document residual are all bound to the waybill number and the relationship edge number. For each residual, record the waybill number, relation edge number, residual type, residual value, data source, credible document status, deductible document flag, and closure flag to form a five-flow residual ledger. When the residual value is zero, the closure flag is recorded as closed; when the residual value is not zero or the comparison difference is marked as not returned, the closure flag is recorded as unclosed; the data source is recorded as freight business reported data or data returned by a third-party credible data source.

[0024] In this embodiment, the operation of generating relational edge gating coefficients includes: For each relation edge in the five-flow relationship diagram, a corresponding gated processing record is established. This record records the relation edge number, relation edge type, residual type, residual value, data source, and trusted document status. Gated processing records are generated sequentially according to relation edge numbers, with one relation edge number corresponding to one gated processing record. When generating a gated processing record, the relation edge number and relation edge type are first read from the five-flow relationship diagram. Then, the residual type, residual value, data source, and trusted document status under the same relation edge number are read from the five-flow residual ledger. After reading, these fields are arranged into a gated input field group. This gated input field group is used for the subsequent generation of residual control quantities, source control quantities, and document control quantities. Based on the relationship edge type, the corresponding relationship convolution weight matrix is ​​selected in the improved R-GCN model, and the matrix number of the relationship convolution weight matrix is ​​registered in the gating processing record. The improved R-GCN model pre-configures a relationship convolution weight matrix index table, which registers the mapping relationship between the main relationship edge, time relationship edge, amount relationship edge, trajectory relationship edge, and voucher relationship edge and the corresponding matrix number. When processing relationship edges, the relationship edge type in the gating processing record is read, the matrix number corresponding to the same relationship edge type is located in the relationship convolution weight matrix index table, and the matrix number is registered in the gating processing record. Based on the residual type, the corresponding residual interval sequence is read. The residual value is compared sequentially with the lower and upper limits of the interval in the residual interval sequence. The interval level corresponding to the interval to which the residual value belongs is written as the residual control quantity. The residual interval sequence is established separately for main residuals, time residuals, amount residuals, trajectory residuals, and document residuals. Each residual interval sequence includes low residual intervals, medium residual intervals, and high residual intervals. When processing residual values, the residual type in the gating processing record is read first, and then the residual interval sequence of the corresponding residual type is retrieved. Starting from the first interval, the residual value is compared item by item with the lower and upper limits of the interval. When the residual value falls into the low residual interval, the low residual level is written as the residual control quantity. When the residual value falls into the medium residual interval, the medium residual level is written as the residual control quantity. When the residual value falls into the high residual interval, the high residual level is written as the residual control quantity. When the residual value is a non-returned flag, the delayed residual level is written as the residual control quantity. The corresponding data source level is read based on the data source, and the data source level is written as the source control quantity. The data source level is generated according to the data source field. When the data source is freight business reported data, the reporting source level is written as the source control quantity. When the data source is data returned by a third-party trusted data source, the trusted source level is written as the source control quantity. When the data source is appeal and supplementary evidence data, the supplementary evidence source level is written as the source control quantity. After the source control quantity is written into the gating processing record, it participates in the gating value location together with the residual control quantity. Based on the trusted document status, read the corresponding document status level and write the document status level as the document control quantity; when the trusted document status is returned, write the returned level as the document control quantity; when the trusted document status is not returned, write the not returned level as the document control quantity; the trusted document status is determined by the same relation edge number in the five-flow residual ledger, and the document control quantity is used to distinguish whether the relation edge participates in residual deduction or enters the delayed verification queue in this round of relation convolution; The gate control value is read according to its position in the gate control value sequence based on the residual control quantity, source control quantity, and voucher control quantity. The gate control value is then determined as the relational edge gate control coefficient. The gate control value sequence is arranged in the order of residual control quantity, source control quantity, and voucher control quantity. When reading the gate control value, the first level position is determined by the residual control quantity, the second level position by the source control quantity, and the third level position by the voucher control quantity. The three levels of positions jointly locate a gate control value. After locating the gate control value, the gate control value is written into the relational edge gate control coefficient field in the gate control processing record. The gating coefficients of the relation edges are bound to the relation edge numbers and the matrix numbers of the relation convolution weight matrix to form gated relation edge records for weighted propagation in the improved R-GCN model. Each gated relation edge record includes the relation edge number, relation edge type, matrix number, relation edge gating coefficient, residual type, residual value, trusted document status, and deductible document flag. After generation, the gated relation edge records are archived according to the waybill number and used as the edge weight input when the improved R-GCN model performs relation convolution.

[0025] In this embodiment, the operations for forming the waybill propagation residual state include: Extract relation edges from the five-flow relationship graph by waybill number, and extract relation convolution weight matrix and relation edge gating coefficient from the gated relation edge records by relation edge number. Extract residual type, residual value, reliable voucher status, deductible voucher marker, and closure marker from the five-flow residual ledger. During extraction, the waybill number is used as the processing scope of the relation convolution in this round, and only relation edges, gated relation edge records, and five-flow residual ledger records under the same waybill number are extracted. The relation edge number is used to associate the five-flow relationship graph, gated relation edge records, and five-flow residual ledger to ensure that the relation convolution weight matrix, relation edge gating coefficient, and residual terms under the same relation edge number participate in the same round of propagation. Adjacent node residual messages are generated based on the starting node number, ending node number, residual type, and residual value of the relation edge. When generating adjacent node residual messages, the node field set corresponding to the starting node number and the node field set corresponding to the ending node number are read, and the starting node field set, ending node field set, relation edge type, residual type, and residual value are combined into an adjacent node residual message. The adjacent node residual message retains the starting node number and ending node number to record the propagation path of the residual in the five-flow relation graph. The residual messages of adjacent nodes are weighted and propagated according to the relation convolution weight matrix and the relation edge gating coefficients to generate gated residual messages. During weighted propagation, the residual messages of adjacent nodes are fed into the relation convolution weight matrix corresponding to the relation edge type to obtain the relation convolution output. Then, the relation edge gating coefficients in the gated relation edge record are read and loaded into the relation convolution output to form gated residual messages. The gated residual messages register the relation edge number, residual type, gated residual value and propagation path. Gated residual messages are aggregated by waybill number and residual type to form relational propagation residuals. During aggregation, gated residual messages corresponding to the main residual under the same waybill number are grouped into the main residual set, gated residual messages corresponding to the time residual are grouped into the time residual set, gated residual messages corresponding to the amount residual are grouped into the amount residual set, gated residual messages corresponding to the trajectory residual are grouped into the trajectory residual set, and gated residual messages corresponding to the bill residual are grouped into the bill residual set. After each residual set is aggregated, a relational propagation residual of the corresponding residual type is formed. For residual items where the trusted voucher status is "returned" and the deductible voucher marker corresponds to the residual type, residual deduction is performed. For residual items where the trusted voucher status is "not returned" or the deductible voucher marker does not correspond to the residual type, the residual value is retained. When performing residual deduction, the residual value corresponding to the same relation edge number in the five-stream residual ledger is read, and then the relation propagation residual is read. The relation propagation residual is used as the basis for deduction to update the residual value. When the trusted voucher status is "returned" and the deductible voucher marker corresponds to the residual type, the residual item enters the deduction queue. When the trusted voucher status is "not returned", the residual item enters the retention queue. When the deductible voucher marker does not correspond to the residual type, the residual item enters the retention queue. The closure marker is updated based on the residual value after deduction. The residual type, relationship propagation residual, deduction result, and closure marker are summarized by waybill number to form the waybill propagation residual status. When the residual value is reduced to zero, the closure marker is updated to closed; when the residual value is still not zero after deduction, the closure marker is updated to unclosed; residual items in the retention queue remain unclosed. The waybill propagation residual status includes waybill number, residual type, relationship propagation residual, residual value before deduction, residual value after deduction, deduction result, closure marker, and propagation path.

[0026] In this embodiment, the operation of writing the relation edge with a trusted credential status of "not returned" to the delayed verification queue includes: Extract residual items from the waybill propagation residual state that are marked as unclosed and whose trusted document status is not returned. Read the waybill number, relation edge number, residual type, residual value, data source, and deductible document flag corresponding to the residual item. During extraction, traverse the residual items in the waybill propagation residual state, filter the residual items marked as unclosed, and then select the residual items whose trusted document status is not returned from the filter results to obtain the delayed verification object. The delayed verification object retains the original waybill number and relation edge number to avoid the inability to locate the relation edges in the original five-flow relation graph after the delayed verification returns. Delayed verification records are generated using waybill numbers and relation edge numbers. These records record residual type, residual value, data source, deductible voucher flag, enqueue time, retries, and queue status. The record number is generated by combining the waybill number and relation edge number. The enqueue time records the time when the delayed verification object enters the delayed verification queue. The retries record the number of times a trusted third-party data source is invoked. The queue status records the processing stage of the delayed verification object. Mark the queue status as pending verification and set the retry count to the initial count. When the initial count is zero and the queue status is pending verification, the delayed verification record waits to call the third-party trusted data source again. When the queue status is updated to verification in progress, the delayed verification record is calling the third-party trusted data source. When the queue status is updated to verified, the fields returned by the third-party trusted data source have been filled into the corresponding nodes and corresponding relationship edges. After the trusted third-party data source returns data, the delayed verification record is located according to the data source and relation edge number. The fields returned by the trusted third-party data source are then added to the corresponding node and the corresponding relation edge. The data returned by the trusted third-party data source includes the returned field name, returned field value, and returned time. After locating the delayed verification record, the starting node number and ending node number are read according to the relation edge number. The returned field values ​​are then added to the node corresponding to the starting node number, the node corresponding to the ending node number, or the corresponding relation edge. After the addition is completed, the "not returned" flag is replaced with the field returned by the trusted third-party data source. The verification rules are reread according to the relation edge number. The fields of the nodes at both ends of the updated relation edge are compared with the fields returned by the third-party trusted data source. The residual value corresponding to the same relation edge number in the five-flow residual ledger is replaced, and the trusted document status is updated to "returned". The closure flag and queue status are updated according to the replaced residual value. After the re-comparison, a new comparison difference is obtained, which is converted into a new residual value. The new residual value replaces the original residual value in the five-flow residual ledger. The trusted document status is updated from "not returned" to "returned". When the new residual value is zero, the closure flag is updated to "closed". When the new residual value is not zero, the closure flag remains "unclosed". The queue status is updated to "verified".

[0027] In this implementation, the operation of updating the five-flow relationship diagram and the five-flow residual ledger after the appeal and supplementary evidence data is approved includes: Extract residual items marked as unclosed from the waybill propagation residual status, and read the corresponding waybill number, relation edge number, residual type, residual value, and deductible voucher mark for each residual item; after extracting the residual items marked as unclosed, establish appeal and supplementary certificate association records according to the residual type; the appeal and supplementary certificate association records store the waybill number, relation edge number, residual type, residual value, and deductible voucher mark, which are used for association and positioning after receiving appeal and supplementary certificate data; Receive appeal and supplementary certificate data corresponding to the waybill number, and parse the supplementary certificate type, supplementary certificate field, and review status in the appeal and supplementary certificate data; the appeal and supplementary certificate data includes the waybill number, supplementary certificate type, supplementary certificate field, upload time, and review status; the supplementary certificate type is used to compare with the deductible certificate mark; the supplementary certificate field is used to supplement the missing or abnormal fields in the original relation edge; the review status is used to determine whether the supplementary certificate data is included in the five-flow relation graph; When the review status is "approved", a supplementary certificate node number is generated based on the waybill number and the supplementary certificate type, and the supplementary certificate field is registered as a supplementary certificate node. When generating a supplementary certificate node, the waybill number, supplementary certificate type, and supplementary certificate field are read and combined into a supplementary certificate node. The supplementary certificate node saves the supplementary certificate node number, waybill number, supplementary certificate type, supplementary certificate field, and review status. When the review status is "not approved", no supplementary certificate node is generated and the five-flow relationship diagram is not updated. Locate the corresponding relationship edge in the five-flow relationship diagram according to the relationship edge number, and connect the supplementary certificate node to the start node and end node of the corresponding relationship edge to form a supplementary certificate relationship edge; after locating the corresponding relationship edge, read the start node number and end node number; establish a supplementary certificate relationship edge from the supplementary certificate node to the node corresponding to the start node number, and establish a supplementary certificate relationship edge from the supplementary certificate node to the node corresponding to the end node number; register the supplementary certificate relationship edge with the supplementary certificate node number, relationship edge number, start node number, end node number, supplementary certificate type and rule number; When the type of supplementary voucher is consistent with the deductible voucher mark, the status of the trusted voucher corresponding to the same relation edge number in the five-flow residual ledger is updated to "returned", and the type of supplementary voucher is registered as the deductible voucher mark. When the type of supplementary voucher is consistent with the deductible voucher mark, the supplementary voucher node is allowed to participate in residual deduction as a trusted voucher. When the type of supplementary voucher is inconsistent with the deductible voucher mark, the supplementary voucher node is retained in the five-flow relation graph, but does not participate in the deduction of the current residual item. The verification rules are reread according to the relation edge number. The supplementary verification node, the start node and end node of the corresponding relation edge are compared. The residual value corresponding to the same relation edge number in the five-flow residual ledger is replaced. The closure mark is updated according to the replaced residual value. During the re-comparison, the supplementary verification field in the supplementary verification node is used as a supplementary verification field other than the field returned by the third-party trusted data source. When the supplementary verification field meets the verification rules with the start node field and the end node field, the residual value is replaced with zero. When the supplementary verification field does not meet the verification rules with the start node field and the end node field, the residual value is replaced with the supplementary verification comparison difference. When the replaced residual value is zero, the closure mark is updated to closed. When the replaced residual value is not zero, the closure mark is updated to unclosed.

[0028] In this embodiment, the operation of extracting the affected waybill subgraph and re-performing relational convolution includes: Read the relation edge numbers whose trusted document status has been updated to "returned", residual value has been replaced, and closure mark has been updated from the updated five-flow residual ledger, and generate an updated relation edge list; when generating the updated relation edge list, read the records of status changes in the current round in the five-flow residual ledger, merge the records of trusted document status changes, residual value replacement records, and closure mark changes, and remove duplicates according to relation edge numbers to obtain the updated relation edge list; According to the updated relation edge list, locate the waybill number, start node number and end node number of the corresponding relation edge in the five-flow relation graph; during the location, read the relation edge number in the updated relation edge list one by one, search the relation edge corresponding to the same relation edge number in the five-flow relation graph, and extract the waybill number, start node number and end node number. The waybill number is used as the extraction range of the affected waybill subgraph. Centered on the origin node number and the destination node number, extract the adjacent first-order nodes and adjacent first-order relation edges that are directly connected to the origin node number and the destination node number under the same waybill number; the adjacent first-order nodes include the nodes directly connected to the origin node number, the nodes directly connected to the destination node number, and the supplementary certificate node; the adjacent first-order relation edges include the relation edges connecting the origin node number and the adjacent first-order nodes, the relation edges connecting the destination node number and the adjacent first-order nodes, and the supplementary certificate relation edges. The updated relation edges, supplementary proof relation edges, adjacent first-order nodes, and adjacent first-order relation edges are merged to form an affected waybill subgraph. During merging, the updated relation edges are first added to the subgraph edge set, then the supplementary proof relation edges are added to the subgraph edge set, and then the adjacent first-order relation edges are added to the subgraph edge set. The corresponding nodes are extracted from the subgraph edge set to form a subgraph node set. The subgraph node set is deduplicated by node number, and the subgraph edge set is deduplicated by relation edge number to obtain the affected waybill subgraph. Read the gated relation edge records and five-flow residual ledger according to the relation edge number in the affected waybill subgraph, and regenerate the adjacent node residual message; when regenerating the adjacent node residual message, only process the relation edge number in the affected waybill subgraph, and do not read the relation edges under the same waybill number that have not entered the affected waybill subgraph; the regenerated adjacent node residual message carries the updated residual value, the updated trusted document status, and the updated closure mark. The residual messages from adjacent nodes are input into the improved R-GCN model to perform relational convolution, updating the relational propagation residuals, deduction results, and closure markers under the corresponding waybill number, forming the updated waybill propagation residual state. After re-performing the relational convolution, the updated relational propagation residuals replace the old ones, the updated deduction results replace the old ones, and the updated closure markers replace the old ones; the updated waybill propagation residual state serves as the basis for outputting the freight business verification results.

[0029] A freight business verification system based on big data analysis according to an embodiment of the present invention includes the following modules: The data collection module is used to collect freight business reported data by waybill number and form waybill verification records; The five-flow mapping module is used to map waybill verification records into a five-flow relationship diagram; The residual ledger generation module is used to generate five-stream residual ledgers according to the verification rules and the fields returned by the trusted third-party data source; The R-GCN verification module has been improved to generate relational edge gating coefficients based on the five-flow relational graph and the five-flow residual ledger, perform relational convolution, and form the waybill propagation residual state; The delayed verification module is used to write the relationship edges with the trusted credential status of not returned to the delayed verification queue, and update the five-flow relationship graph and the five-flow residual ledger after the third-party trusted data source returns the data; The supplementary evidence write-back module is used to generate supplementary evidence nodes and supplementary evidence relationship edges after the appeal supplementary evidence data is approved, and to update the five-flow relationship graph and the five-flow residual ledger; The results output module is used to output the freight business verification results based on the updated waybill propagation residual status.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a network freight business verification scenario. In this scenario, freight business data sources are dispersed. Waybill basic data, contract data, trajectory data, fund flow data, and invoice data are uploaded by different business links. The return speed of third-party trusted data sources is inconsistent, and some trajectory data, bank flow data, and invoice verification data are returned with delays. Traditional verification methods mainly rely on single-field comparison and manual review, which easily leads to problems such as the inability to link and verify contract amount, payment amount, and invoice amount; the inability to close the verification between trajectory field and loading / unloading location field; the long-term suspension of abnormal records when third-party trusted data sources have not returned; and the need for manual re-judgment after appeal and supplementary evidence data are received.

[0031] In this embodiment, a batch of anonymized freight business data is selected as the verification object, covering waybill basic data, contract data, trajectory data, fund flow data, and document data. Each freight business data entry is processed according to the waybill number, forming a waybill verification record. After the waybill verification record enters the five-flow graphing process, it is mapped to information flow nodes, contract flow nodes, transportation flow nodes, fund flow nodes, and document flow nodes. Relationship edges are established between the five types of nodes according to the subject field, time field, amount field, trajectory field, and document number field, and a rule number is bound to each relationship edge. After the relationship edges are established, the five-flow relationship graph can express the verification relationships of subject consistency, time sequence, amount closure, trajectory matching, and document correspondence under the same waybill number.

[0032] During the verification process, the relationship edges in the five-flow relationship graph read the verification rules according to the rule numbers and retrieve the fields returned by the third-party trusted data source. For data that has already been returned, such as vehicle qualification, personnel qualification, trajectory, bank statements, and invoice verification, it is directly compared with the fields of the nodes at both ends of the relationship edge to generate subject residuals, time residuals, amount residuals, trajectory residuals, and document residuals. For data that has not yet been returned by the third-party trusted data source, a "not returned" flag is registered, and the trusted document status is registered as "not returned." All residual items are written to the five-flow residual ledger, which records the waybill number, relationship edge number, residual type, residual value, data source, trusted document status, deductible document flag, and closure flag.

[0033] After the five-flow relationship graph and five-flow residual ledger are input into the improved R-GCN model, the improved R-GCN model selects the relationship convolution weight matrix according to the relationship edge type and generates relationship edge gating coefficients based on the residual value, data source, and trusted document status. These relationship edge gating coefficients are loaded into the residual message propagation process of adjacent nodes, enabling the subject residual, time residual, amount residual, trajectory residual, and document residual to propagate weighted along the corresponding relationship edges. The propagated residuals are aggregated by waybill number to form relationship propagation residuals. Residual items with a trusted document status of "returned" and matching deductible document markers enter the deduction process; residual items with a trusted document status of "not returned" or mismatched deductible document markers retain their residual values ​​and enter delayed verification or appeal for supplementary documentation processing.

[0034] After the trusted third-party data source returns data, the delayed verification queue locates the delayed verification record according to the waybill number and relation edge number, adds the returned fields to the corresponding node and relation edge, rereads the verification rules, and replaces the residual values ​​in the five-flow residual ledger. After the appeal and supplementary verification data is approved, a supplementary verification node and a supplementary verification relation edge are generated. The supplementary verification node is connected to the start and end nodes of the abnormal relation edge, and the supplementary verification fields participate in the re-comparison. After the five-flow relation graph and the five-flow residual ledger are updated, the affected waybill subgraph is extracted. Only the updated relation edge, the supplementary verification relation edge, the adjacent first-order node, and the adjacent first-order relation edge are re-executed with R-GCN relation convolution. All waybill data is no longer re-verified. After verification, waybills with all closed closure markers output normal verification results. Waybills with unclosed residual items output abnormal waybill records, abnormal relation edges, residual types, and verification basis.

[0035] This embodiment compares basic rule verification, ordinary graph verification, and the method of this invention. Basic rule verification performs independent comparisons on various fields without constructing a five-flow relationship graph. Ordinary graph verification constructs a relationship graph but does not introduce the five-flow residual ledger, relationship edge gating coefficients, delay verification queue, or recalculation of the affected waybill subgraph. The method of this invention performs five-flow relationship graph construction, five-flow residual ledger generation, improved R-GCN gating propagation, residual deduction, delay verification, appeal and supplementary documentation write-back, and affected waybill subgraph recalculation on the same anonymized sample. The comparison results are shown in the table below: Table 1: Comparison of Freight Verification Results

[0036] The data in the table shows that, with the same 12,000 anonymized freight business data entries and 186,000 field records, the method of this invention achieves an accuracy rate of 96.7% in identifying main anomalies, higher than the 88.4% of basic rule verification and 92.1% of ordinary map verification; an accuracy rate of 95.2% in identifying time anomalies, higher than the 84.9% of basic rule verification and 90.3% of ordinary map verification; an accuracy rate of 96.1% in identifying amount anomalies, higher than the 86.8% of basic rule verification and 91.6% of ordinary map verification; an accuracy rate of 94.8% in identifying trajectory anomalies, higher than the 82.7% of basic rule verification and 89.5% of ordinary map verification; and an accuracy rate of 95.9% in identifying document anomalies, higher than the 85.6% of basic rule verification and 90.8% of ordinary map verification. The cross-flow anomaly recall rate increased from 79.3% for basic rule verification and 88.9% for ordinary graph verification to 95.4%, indicating that the five-flow relationship graph and the five-flow residual ledger can enhance the ability to correlate anomalies across data sources.

[0037] Regarding the processing of delayed and supplementary data, the delayed data closure rate of the method of this invention reaches 91.2%, higher than the 63.5% of basic rule verification and the 78.6% of ordinary map verification; the automatic closure rate after appeal and supplementary verification reaches 88.6%, higher than the 54.1% of basic rule verification and the 72.4% of ordinary map verification. The false alarm rate of the method of this invention is 3.9%, lower than the 9.8% of basic rule verification and the 6.7% of ordinary map verification; the average verification time per ticket is 2.1 seconds, lower than the 4.8 seconds of basic rule verification and the 3.6 seconds of ordinary map verification; the number of full-volume repeated verifications is 186, lower than the 1280 times of basic rule verification and the 742 times of ordinary map verification; the proportion of manual review is 11.7%, lower than the 31.6% of basic rule verification and the 22.8% of ordinary map verification. The data shows that the improved R-GCN model can improve residual propagation and deduction efficiency, while delayed verification queues, appeal supplementary certificate rewrites, and affected waybill subgraph recalculations can reduce full-volume duplicate verifications and improve the accuracy and traceability of freight business verification results.

[0038] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for verifying freight operations based on big data analysis, characterized in that, Includes the following steps: Receive freight business reported data, collect basic data of waybill, contract data, trajectory data, fund flow data and bill data according to waybill number, and form waybill verification record; The waybill verification record is mapped to five-flow nodes, and relationship edges are established according to the subject, time, amount, trajectory and voucher fields to bind rule numbers, forming a five-flow relationship graph; Read the verification rule according to the rule number, compare the fields of the nodes at both ends of the relationship edge with the fields returned by the third-party trusted data source, and register a no-return flag when no returned field is read. Write the comparison difference and the no-return flag into the five types of residuals to form a five-stream residual ledger. The five-flow relation graph and the five-flow residual ledger are input into the improved R-GCN model. The relation convolution weight matrix is ​​read according to the relation edge type. The relation edge gating coefficients are generated based on the residual value, data source and trusted certificate status. The residual messages of adjacent nodes are weighted and propagated to obtain the relation propagation residual. Propagate residuals by aggregation relationship based on waybill number, deduct residual items whose trusted document status is returned and whose deductible document mark matches, and retain residual items whose trusted document status is not returned or whose deductible document mark does not match, thus forming waybill propagation residual status; Write the relationship edge with the trusted credential status not returned to the delayed verification queue. After the data returned by the third-party trusted data source and the appeal and supplementary evidence data are approved, update the five-flow relationship graph and the five-flow residual ledger, and extract the affected waybill graph. The affected waybill subgraph is re-input into the improved R-GCN model to perform relational convolution, and the freight business verification results are output based on the propagation residual state of the unclosed waybill.

2. The freight business verification method based on big data analysis according to claim 1, characterized in that, The operation of generating the waybill verification record includes: Parse the freight business reporting data and extract the waybill number, reporting stage marker, and field source marker; A collection line is established based on the waybill number, and the collection line is divided into a waybill basic data area, a contract data area, a trajectory data area, a fund flow data area, and a bill data area; The parsed field records are assigned to the corresponding data areas according to the field source markers; The five data areas in the same collection line are merged to form a waybill verification record carrying the waybill number, reporting stage mark, and field source mark.

3. The freight business verification method based on big data analysis according to claim 2, characterized in that, The operation of forming the five-flow relationship diagram includes: Read the waybill number, field source marker, and field record from the waybill verification record, and generate information flow node, contract flow node, transportation flow node, fund flow node, and bill flow node according to the field source marker; Extract the subject field, time field, amount field, trajectory field, loading / unloading location field, and voucher number field from the five types of nodes; Under the same waybill number, generate subject relationship edges based on the subject field value and subject correspondence table, generate time relationship edges based on the order of time fields, generate amount relationship edges based on the amount fields of contract flow nodes, fund flow nodes and bill flow nodes, generate trajectory relationship edges based on the trajectory field and loading / unloading location field, and generate voucher relationship edges based on the voucher number field value and voucher correspondence table. Register the waybill number, origin node number, destination node number, relationship edge type, and rule number for each relationship edge to form a five-flow relationship diagram.

4. The freight business verification method based on big data analysis according to claim 3, characterized in that, The operation of forming the five-stream residual ledger includes: Read the edge number, edge type, starting node number, ending node number, and rule number from the five-flow relationship graph; Retrieve the corresponding verification rule according to the rule number, and determine the fields to be compared, the third-party trusted data source, and the deductible voucher marker according to the verification rule; Read the node fields corresponding to the starting node number and the ending node number, and read the fields returned by the third-party trusted data source or the flag that no data was returned; When reading the fields returned by the trusted third-party data source, the node fields are compared with the fields returned by the trusted third-party data source to obtain the comparison differences; When a "not returned" flag is read, the "not returned" flag is used as a comparison difference of the corresponding relation edge, and the trusted credential status is registered as "not returned". When the relationship edge type is a subject relationship edge, the comparison difference will be registered as a subject residual; when the relationship edge type is a time relationship edge, the comparison difference will be registered as a time residual; when the relationship edge type is an amount relationship edge, the comparison difference will be registered as an amount residual; when the relationship edge type is a trajectory relationship edge, the comparison difference will be registered as a trajectory residual; when the relationship edge type is a voucher relationship edge, the comparison difference will be registered as a negotiable instrument residual. For each residual, record the waybill number, relation edge number, residual type, residual value, data source, credible voucher status, deductible voucher mark, and closure mark to form a five-stream residual ledger.

5. The freight business verification method based on big data analysis according to claim 4, characterized in that, The operation of generating relational side gating coefficients includes: For each relation edge in the five-flow relation graph, a corresponding gating processing record is established. The relation edge number, relation edge type, residual type, residual value, data source, and trusted credential status are recorded in the gating processing record. Based on the type of relation edge, select the corresponding relation convolution weight matrix in the improved R-GCN model, and register the matrix number of the relation convolution weight matrix in the gating processing record; Read the corresponding residual interval sequence according to the residual type, compare the residual value with the lower limit and upper limit of the interval in the residual interval sequence in turn, and write the interval level corresponding to the interval to which the residual value belongs as the residual control quantity; Read the corresponding data source level according to the data source, and write the data source level as the source control quantity; Read the corresponding document status level based on the trusted document status, and write the document status level as a document control quantity; The gate control values ​​are read according to the arrangement of the residual control quantity, source control quantity, and voucher control quantity in the gate control value sequence, and the gate control values ​​are determined as the relation edge gate control coefficients. By binding the relation edge gating coefficients with the relation edge numbers and the matrix numbers of the relation convolution weight matrix, a gating relation edge record is formed for weighted propagation in the improved R-GCN model.

6. The freight business verification method based on big data analysis according to claim 5, characterized in that, The operation to form the waybill propagation residual state includes: Extract the relation edges from the five-flow relation graph by waybill number, extract the relation convolution weight matrix and relation edge gating coefficient from the gated relation edge record by relation edge number, and extract the residual type, residual value, reliable voucher status, deductible voucher mark and closure mark from the five-flow residual ledger. Generate adjacent node residual messages based on the starting node number, ending node number, residual type, and residual value of the relation edge; The residual messages of adjacent nodes are weighted and propagated according to the relation convolution weight matrix and relation edge gating coefficients to generate gated residual messages; the gated residual messages are aggregated according to waybill number and residual type to form relation propagation residuals. Perform residual deduction on residual items whose trusted voucher status is returned and whose deductible voucher tag corresponds to the residual type; retain residual values ​​for residual items whose trusted voucher status is not returned or whose deductible voucher tag does not correspond to the residual type. Update the closure marker based on the residual value after deduction, and summarize the residual type, relational propagation residual, deduction result and closure marker by waybill number to form the waybill propagation residual status.

7. The freight business verification method based on big data analysis according to claim 6, characterized in that, The operation of writing the relation edge with a trusted credential status of "not returned" to the delayed verification queue includes: Extract residual items from the waybill propagation residual status that are marked as unclosed and whose trusted document status is not returned, and read the waybill number, relation edge number, residual type, residual value, data source and deductible document mark corresponding to the residual item; Generate a delay verification record using the waybill number and the relationship edge number. The delay verification record records the residual type, residual value, data source, deductible voucher mark, queuing time, number of retries and queue status. Mark the queue status as pending verification and set the retry count to the initial count; After the trusted third-party data source returns data, locate the delayed verification record according to the data source and relation edge number, and fill the fields returned by the trusted third-party data source into the corresponding node and corresponding relation edge; The verification rules are reread according to the relation edge number. The fields of the nodes at both ends of the updated relation edge are compared with the fields returned by the third-party trusted data source. The residual value corresponding to the same relation edge number in the five-flow residual ledger is replaced. The trusted voucher status is updated to "returned". The closure mark and queue status are updated according to the replaced residual value.

8. The freight business verification method based on big data analysis according to claim 7, characterized in that, The operations of updating the five-flow relationship diagram and the five-flow residual ledger after the appeal and supplementary evidence data are approved include: Extract residual items marked as unclosed from the waybill propagation residual status, and read the waybill number, relation edge number, residual type, residual value and deductible voucher mark corresponding to the residual item; Receive appeal and supplementary certificate data corresponding to the waybill number, and parse the supplementary certificate type, supplementary certificate field and review status in the appeal and supplementary certificate data; When the review status is approved, generate a supplementary certificate node number based on the waybill number and the supplementary certificate type, and register the supplementary certificate field as a supplementary certificate node; Locate the corresponding relation edge in the five-flow relation graph according to the relation edge number, and connect the supplementary proof node to the start node and end node of the corresponding relation edge to form a supplementary proof relation edge; When the type of supplementary voucher is consistent with the deductible voucher mark, update the status of the reliable voucher corresponding to the same relation edge number in the five-flow residual ledger to "returned" and register the type of supplementary voucher as the deductible voucher mark. The verification rules are reread according to the relation edge number. The supplementary verification node, the start node and end node of the corresponding relation edge are compared. The residual value corresponding to the same relation edge number in the five-flow residual ledger is replaced. The closure mark is updated according to the replaced residual value.

9. The freight business verification method based on big data analysis according to claim 8, characterized in that, The operation of extracting the affected waybill subgraph and re-performing relational convolution includes: Read the relation edge numbers whose trusted voucher status is updated to "returned", residual value is replaced, and closure mark is updated from the updated five-flow residual ledger, and generate an updated relation edge list; Locate the waybill number, start node number, and end node number of the corresponding relation edge in the five-flow relation graph according to the updated relation edge list; Using the starting node number and the ending node number as the center, extract the adjacent first-order nodes and adjacent first-order relation edges that are directly connected to the starting node number and the ending node number under the same waybill number. The updated relation edges, the supplementary proof relation edges, the adjacent first-order nodes, and the adjacent first-order relation edges are merged to form the affected waybill graph; Read the gated relationship edge records and the five-flow residual ledger according to the relationship edge number in the affected waybill subgraph, and regenerate the residual messages of adjacent nodes; The residual messages of adjacent nodes are input into the improved R-GCN model to perform relational convolution, update the relational propagation residual, deduction result and closure mark under the corresponding waybill number, and form the updated waybill propagation residual state.

10. A freight business verification system based on big data analysis according to claim 1, executing a freight business verification method based on big data analysis according to any one of claims 1 to 9, characterized in that, Includes the following modules: The data collection module is used to collect freight business reported data by waybill number and form waybill verification records; The five-flow mapping module is used to map waybill verification records into a five-flow relationship diagram; The residual ledger generation module is used to generate five-stream residual ledgers according to the verification rules and the fields returned by the trusted third-party data source; The R-GCN verification module has been improved to generate relational edge gating coefficients based on the five-flow relational graph and the five-flow residual ledger, perform relational convolution, and form the waybill propagation residual state; The delayed verification module is used to write the relationship edges with the trusted credential status of not returned to the delayed verification queue, and update the five-flow relationship graph and the five-flow residual ledger after the third-party trusted data source returns the data; The supplementary evidence write-back module is used to generate supplementary evidence nodes and supplementary evidence relationship edges after the appeal supplementary evidence data is approved, and to update the five-flow relationship graph and the five-flow residual ledger. The results output module is used to output the freight business verification results based on the updated waybill propagation residual status.

Images