Logistics industry digital financial risk analysis method based on internet of things access

Abstract

The application discloses a logistics industry digital financial risk analysis method based on Internet of Things access, relates to the technical field of logistics supply chain financial risk analysis, and comprises the following steps: acquiring the track coordinates of a truck terminal, the access records of a warehouse access control and the load change of an intelligent shelf as initial device data, and extracting the action time corresponding to the bottom physical access behavior from the initial device data to obtain first time sequence data arranged according to the action time; the logistics industry digital financial risk analysis method based on Internet of Things access can reduce the verification difficulty caused by the disconnection between the account records and the actual goods flow state in the existing logistics supply chain financial business, improve the accuracy of goods flow authenticity verification, risk node positioning and abnormal responsibility tracing, and help improve the reliability and penetrating supervision capability of the logistics industry digital financial risk analysis.

Description

Technical Field

[0001] This invention relates to the field of logistics supply chain finance risk analysis technology, specifically to a method for analyzing industrial digital finance risks in the logistics industry based on Internet of Things access. Background Technology

[0002] In industrial digital finance within the logistics sector, the actual flow of goods is a crucial basis for financial institutions to conduct credit granting, loan disbursement, post-loan management, and for regulatory authorities to conduct risk audits. Current logistics supply chain data management methods typically rely on manual data entry by business personnel, electronic document circulation, or post-event aggregation by centralized business systems. These accounting records primarily reflect business declaration results and fail to directly reflect the actual physical handover process of goods during transportation, warehousing, storage, and outbound processes. Therefore, in financing transactions, discrepancies easily arise between accounting documents and the actual flow of goods, increasing financial risks such as duplicate financing, false warehouse receipts, and document forgery.

[0003] In real-world logistics scenarios, IoT devices such as truck terminals, warehouse access control systems, and smart shelves generate underlying physical access data, including trajectory coordinates, access records, and load changes. However, this data is typically scattered across different device nodes or business systems, with inconsistent data formats, collection frequencies, and time bases, making it difficult to effectively align the timelines of actions across different business processes. This is especially true when goods handover involves multiple physical events such as changes in transport location, warehouse entry / exit records, and shelf weight changes; existing systems struggle to organize fragmented device data into a complete temporal sequence reflecting the true flow of goods. Furthermore, current supply chain finance risk control methods primarily rely on document review, manual verification, or partial data comparison, underutilizing the physical access data generated by IoT devices. When auditors encounter an abnormal waybill or financing contract, while they can access accounting records or system-generated inbound / outbound vouchers, they often struggle to reconstruct the actual physical flow path of the goods and determine whether the accounting records match the actual physical handover actions, thus affecting risk identification and accountability. Summary of the Invention

[0004] The purpose of this invention is to provide a method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access, thereby solving the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for analyzing industrial digital financial risks in the logistics industry based on IoT access, comprising: S1, acquiring the trajectory coordinates of truck terminals, access control records of warehouse access control, and load-bearing changes of smart shelves as initial equipment data, and extracting the action time corresponding to the underlying physical access behavior from the initial equipment data to obtain first time-series data arranged according to action time; S2, constructing an IoT access spatiotemporal graph for the first time-series data, and extracting cross-node collaborative features to obtain second time-series data containing physical handover features and spatial location features; S3, encrypting and solidifying the second time-series data to generate a digital certificate with a unique identifier and tamper-proof; S4, extracting the timestamp and spatial coordinates encapsulated within the digital certificate to determine the time alignment between different IoT device nodes. S5. If the time alignment relationship meets the preset threshold, then the fragmented underlying physical access behaviors are spliced ​​together according to the time alignment relationship to construct an interaction sequence diagram reflecting the actual flow status of goods; S6. Extract the flow path of goods in the physical world from the interaction sequence diagram, perform anomaly detection on the flow path, and determine whether the flow path deviates from the pre-established baseline path; S7. If the flow path deviates from the pre-established baseline path, generate an anomaly mark for the waybill, and reconstruct the real scene containing risk nodes according to the anomaly mark and the interaction sequence diagram; S8. Obtain the accounting records corresponding to the risk nodes in the real scene, and determine whether the accounting records match the physical handover actions in the real scene; S9. If the accounting records do not match the physical handover actions in the real scene, output a document forgery alarm for use in the digital financial risk analysis of the logistics industry.

[0006] Preferably, step S1 includes acquiring the trajectory coordinates, access control records, and load changes of the truck terminal as initial device data; extracting spatial displacement features, identity verification identifiers, and load fluctuation values ​​from the initial device data to obtain a multidimensional physical state set; if the multidimensional physical state set undergoes a sudden change, obtaining the underlying physical access behavior; extracting the behavior trigger nodes corresponding to the underlying physical access behavior to determine the action time; and sorting the underlying physical access behavior according to the action time to obtain the first time-series data arranged according to the action time.

[0007] Preferably, step S2 includes extracting the access trajectory and node distance of the first time-series data to construct an IoT access spatiotemporal graph; processing the adjacency matrix and state sequence of the IoT access spatiotemporal graph using a causal graph attention network to obtain causal weights; extracting cross-node collaborative features by weighting and summing the state sequences using causal weights, and determining the aggregation vector corresponding to the cross-node collaborative features; and performing spatiotemporal mapping on the aggregation vector to obtain second time-series data containing physical handover features and spatial location features.

[0008] Preferably, step S3 includes acquiring second time-series data containing physical handover features and spatial location features; performing feature segmentation processing on the second time-series data to obtain time-series data blocks; using a hash algorithm to extract a digest value from the time-series data blocks to obtain a data digest value; determining whether the length of the data digest value meets a preset threshold; if the length of the data digest value meets the preset threshold, then performing encryption and solidification processing based on the data digest value to determine the solidified ciphertext; and adding a timestamp and node identifier to the solidified ciphertext to perform credential signing, generating a digital credential with a unique identifier and that cannot be tampered with.

[0009] Preferably, step S4 includes obtaining a digital certificate, extracting the original timestamp and original spatial coordinates of the digital certificate; calculating an initial alignment deviation based on the original timestamp and original spatial coordinates to obtain a target alignment deviation; calculating a time offset based on the target alignment deviation; generating a target compensation value if the time offset is greater than a preset offset threshold; correcting the original timestamp with the target compensation value to obtain a target timestamp; and determining the time alignment relationship between IoT device nodes based on the target timestamp.

[0010] Preferably, step S5 includes obtaining underlying physical access records, extracting access timestamps from the underlying physical access records to generate discrete access segments; matching the discrete access segments to obtain time alignment features; determining whether the time alignment features meet a preset threshold; if the time alignment features meet the preset threshold, then performing fusion processing on the discrete access segments to obtain a behavior splicing sequence; constructing a temporal correlation matrix based on the behavior splicing sequence; and constructing an interaction temporal sequence diagram reflecting the actual flow status of goods based on the temporal correlation matrix.

[0011] Preferably, step S6 includes acquiring an interaction sequence diagram, extracting flow nodes from the interaction sequence diagram to construct a flow path; processing the flow path using the isolated forest algorithm to obtain a path feature vector, and calculating an anomaly score based on the path feature vector; if the anomaly score is greater than a preset threshold, extracting the deviation degree from the flow path; and comparing the deviation degree with a pre-established benchmark path to determine whether the flow path deviates from the pre-established benchmark path.

[0012] Preferably, step S7 includes parsing spatial coordinates and dwell time from waybills that deviate from the baseline path to calculate the trajectory deviation angle; if the trajectory deviation angle is greater than a preset deviation angle threshold, generating an anomaly marker for the waybill; extracting node states from the interaction sequence diagram based on the anomaly marker, performing feature mapping on the node states to obtain risk nodes and constructing a scene topology; and performing spatial mapping on the interaction sequence diagram based on the scene topology to reconstruct a real scene containing risk nodes.

[0013] Preferably, step S8 includes obtaining the video stream corresponding to the risk node, extracting the action trajectory of the physical handover action from the video stream, retrieving the accounting records based on the action trajectory, the accounting records containing record details, obtaining the difference features if the flow characteristics of the record details are inconsistent with the action trajectory, and performing classification mapping based on the difference features to determine whether the accounting records match the physical handover action.

[0014] Preferably, step S9 includes obtaining the receipt time and transport vehicle identifier from the accounting records; extracting spatial coordinates based on the transport vehicle identifier to obtain a first spatiotemporal dataset; the first spatiotemporal dataset is used to retrieve and identify video frames to obtain a second spatiotemporal dataset containing physical handover actions; comparing the cargo weight characteristics in the second spatiotemporal dataset with the cargo weight in the accounting records to determine the weight difference value; if the weight difference value is greater than a preset difference threshold, extracting the time difference value between the receipt time and the time of the physical handover action; if the time difference value exceeds a preset time threshold, determining that the accounting records do not match the physical handover actions in the real scenario, and outputting a document forgery alarm for industrial digital financial risk analysis in the logistics industry.

[0015] As can be seen from the above technical solution, the present invention has the following beneficial effects: This IoT-based digital financial risk analysis method for the logistics industry acquires initial equipment data such as truck terminal trajectory coordinates, warehouse access control records, and smart shelf load changes. It transforms the dispersed underlying physical access behaviors generated during the logistics process into first-series data arranged by action time. Furthermore, it constructs an IoT access spatiotemporal graph to extract cross-node collaborative features, enabling continuous and analyzable temporal correlations in the physical handover behaviors of goods transportation, warehousing, and storage. Simultaneously, it encrypts and solidifies the second-series data using a hash algorithm, generating uniquely identified and tamper-proof digital credentials, thus improving the efficiency of IoT access. This invention addresses the credibility and tamper resistance of data. Furthermore, it determines the time alignment between different IoT device nodes based on the timestamps and spatial coordinates in digital credentials, and uses this to piece together fragmented underlying physical access behaviors to construct an interaction time sequence diagram reflecting the true flow of goods. This allows for the reconstruction of the actual flow path of goods in the physical world. Based on this, anomaly detection determines whether the flow path deviates from the baseline path, and by combining anomaly markers with the interaction time sequence diagram, it reconstructs a real-world scenario containing risk nodes. This enables financial institutions or regulatory authorities to further determine whether the accounting records match the actual physical handover actions, and to output a document forgery alert when there is a mismatch. Therefore, this invention reduces the verification difficulties caused by the disconnect between accounting records and the actual flow of goods in existing logistics supply chain finance businesses, improves the accuracy of goods flow authenticity verification, risk node location, and anomaly responsibility tracing, and helps enhance the reliability of industrial digital finance risk analysis and the ability for penetrating supervision in the logistics industry. Attached Figure Description

[0016] Figure 1 This is a signal transmission diagram of the present invention; Figure 2 This is a structural block diagram of the local terminal of an exemplary electronic device of the present invention; Figure 3 This is a structural block diagram of the network terminal of an exemplary electronic device of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1: As Figure 1 As shown, this invention provides a technical solution: a method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access, characterized by comprising: S1. Obtain the trajectory coordinates of the truck terminal, the access control records of the warehouse access control, and the load-bearing changes of the smart shelf as initial equipment data, and extract the action time corresponding to the underlying physical access behavior from the initial equipment data to obtain the first time-series data arranged according to the action time. S2. For the first time series data, construct an IoT access spatiotemporal graph and use a causal graph attention network to extract cross-node collaborative features to obtain second time series data containing physical handover features and spatial location features. S3. Encrypt and solidify the second time-series data using a hash algorithm to generate a digital certificate with a unique identifier that cannot be tampered with. S4. Based on the digital certificate, extract the timestamp and spatial coordinates encapsulated within the digital certificate to determine the time alignment relationship between different IoT device nodes. S5. If the time alignment relationship meets the preset threshold, then the fragmented underlying physical access behaviors are spliced ​​together according to the time alignment relationship to construct an interaction sequence diagram that reflects the actual flow status of goods. S6. Extract the flow path of goods in the physical world from the interaction sequence diagram, and use the isolated forest algorithm to detect anomalies in the flow path to determine whether the flow path deviates from the pre-established baseline path. S7. If the transfer path deviates from the pre-established baseline path, an anomaly mark is generated for the waybill, and the real scenario containing risk nodes is reconstructed based on the anomaly mark and the interaction sequence diagram. S8. Obtain the accounting records corresponding to the risk nodes in the real scenario, and determine whether the accounting records match the physical handover actions in the real scenario; S9. If the accounting records do not match the physical handover actions in the real scenario, a document fraud alarm will be output for use in the digital financial risk analysis of the logistics industry.

[0019] In this embodiment, the method collects IoT device data in the logistics scenario through truck terminals, warehouse access control, and smart shelves, and converts vehicle transportation behavior, warehouse passage behavior, and shelf load-bearing change behavior into first time-series data arranged in chronological order.

[0020] Specifically, the truck terminal data collection fields include vehicle identification, longitude, latitude, collection time, and waybill identification; the warehouse access control data collection fields include access control device identification, access subject identification, access direction, access time, and warehouse identification; the smart shelf data collection fields include shelf identification, location identification, load-bearing value, load-bearing change time, and goods identification; the system uses the device data collection time as the action time and arranges the above fields in chronological order according to the action time to generate the first time-series data, so that the underlying physical access behaviors such as vehicle arrival, access control passage, goods being put on the shelf, goods being taken off the shelf, and vehicle departure form a unified time chain.

[0021] Furthermore, after obtaining the first time-series data, the system uses truck terminals, warehouse access control, smart shelves, and waybill goods as nodes in the IoT access spatiotemporal graph, and uses time connection relationships, spatial connection relationships, and business handover relationships as edges in the IoT access spatiotemporal graph.

[0022] Among them, the temporal connection relationship is determined by the time of adjacent actions, the spatial connection relationship is determined by the trajectory coordinates, warehouse coordinates and cargo location coordinates, and the business handover relationship is determined by the waybill identifier, cargo identifier and warehouse identifier. When the vehicle position recorded by the truck terminal enters within a 50-meter radius of the warehouse coordinates, the warehouse access control generates a passage record within 10 minutes after the vehicle enters, and the smart shelf generates a load change corresponding to the waybill cargo within 15 minutes after the access control passes, the system establishes a handover edge between the corresponding nodes.

[0023] Thus, the first time-series data is transformed into an IoT access spatiotemporal diagram that simultaneously includes time sequence, spatial location, and handover actions.

[0024] In this embodiment, the system uses a causal graph attention network to perform cross-node collaborative feature extraction on the spatiotemporal graph of IoT access. The causal graph attention network sets the causal direction according to the action time sequence, so that the previous action node passes features to the next action node, and the next action node does not pass features in the opposite direction to the previous action node.

[0025] Furthermore, for each handover edge, the system calculates attention weights based on node time difference, node spatial distance, waybill identifier consistency, and cargo identifier consistency. After processing by a causal graph attention network, the system outputs second time-series data. The second time-series data includes physical handover features and spatial location features. The physical handover features include vehicle arrival, access control passage, load increase, load decrease, vehicle departure, and handover completion status. The spatial location features include vehicle latitude and longitude, warehouse coordinates, cargo location coordinates, and distance between nodes.

[0026] Furthermore, after obtaining the second time-series data, the system encrypts and solidifies the second time-series data using a hash algorithm.

[0027] Specifically, the system serializes the waybill identifier, device node identifier, action time, spatial coordinates, physical handover characteristics, spatial location characteristics, and data generation time according to a fixed field order to form a string to be processed, and calculates a hash digest using the SHA-256 algorithm or SM3 algorithm. The system encapsulates the hash digest, timestamp, spatial coordinates, device node identifier, and waybill identifier into a digital certificate. A one-to-one correspondence is formed between the digital certificate and the second time-series data. When any field in the second time-series data is modified, the recalculated hash digest is inconsistent with the hash digest in the digital certificate. Therefore, the system can confirm whether the second time-series data has been tampered with by comparing the hash digests.

[0028] Furthermore, the system determines the time alignment relationship between different IoT device nodes based on the timestamp and spatial coordinates encapsulated in the digital credential.

[0029] Specifically, for truck terminal nodes, warehouse access control nodes, and smart shelf nodes under the same waybill identifier, the system calculates the first time difference between the truck arrival time and the access control time, the second time difference between the access control time and the shelf load change time, the first spatial distance between the truck arrival coordinates and the warehouse coordinates, and the second spatial distance between the shelf location coordinates and the warehouse coordinates. When the first time difference is no more than 10 minutes, the second time difference is no more than 15 minutes, the first spatial distance is no more than 50 meters, and the second spatial distance is within the same warehouse coordinate range, the system determines that the time alignment relationship meets the preset threshold.

[0030] Furthermore, when the time alignment relationship meets a preset threshold, the system splices together the fragmented underlying physical access behaviors according to the time alignment relationship.

[0031] In the inbound scenario, the system connects vehicle arrival, access control entry, shelf load increase, and vehicle departure into an inbound handover link; in the outbound scenario, the system connects shelf load decrease, access control release, vehicle loading, and vehicle departure into an outbound handover link. Each handover link records waybill identifier, cargo identifier, equipment node identifier, action time, spatial coordinates, and action type. The system uses the above handover links as edges and truck terminals, warehouse access control, smart shelves, warehouses, and cargo as nodes to construct an interaction sequence diagram.

[0032] This interactive sequence diagram is used to express the actual flow status of goods from the origin node, transit node, warehousing node to the delivery node.

[0033] In this embodiment, the system extracts the flow path of goods in the physical world from the interaction sequence diagram.

[0034] Specifically, the circulation path includes the originating warehouse, transport vehicles, transit warehouses, inbound shelves, outbound shelves, delivery vehicles, and signing nodes, and includes the action time, spatial coordinates, dwell time, node sequence, and load change value corresponding to each node. The system converts the circulation path into a feature vector and inputs it into the Isolation Forest algorithm for anomaly detection. The Isolation Forest algorithm calculates anomaly scores based on the path length of the feature vector in a randomly partitioned tree. When the anomaly score is greater than a preset anomaly threshold of 0.65, the system determines that the circulation path deviates from a pre-established baseline path. The baseline path is generated from normal historical waybill paths under the same route, the same warehouse, the same carrier, and the same cargo category, and includes the baseline node sequence, baseline transport interval, baseline inbound node, and baseline outbound node.

[0035] Furthermore, when the transfer path deviates from the pre-established baseline path, the system generates an anomaly flag for the waybill.

[0036] Specifically, the anomaly markers include waybill identifier, anomaly node, anomaly action time, anomaly spatial coordinates, anomaly type, and associated device node. The system reconstructs the real scenario containing risk nodes based on the anomaly markers and the interaction sequence diagram. If the accounting record shows that the goods have been put into storage, but there is no access control entry record or shelf load increase record in the interaction sequence diagram, the system will mark the corresponding storage node as a risk node. If the accounting record shows that the goods have been put out of storage, but there is no shelf load decrease record or vehicle departure record in the interaction sequence diagram, the system will mark the corresponding outbound node as a risk node. If the accounting record shows that the goods are transported according to the agreed route, but the truck terminal trajectory coordinates do not pass through the designated transfer node in the baseline path, the system will mark the missing transfer node as a risk node.

[0037] Furthermore, the system acquires the accounting records corresponding to risk nodes in the real-world scenario. These accounting records include waybill records, inbound slips, outbound slips, signed receipts, inventory ledgers, accounts receivable vouchers, and financing application materials. The system compares the time, location, goods identification, goods weight, handover entity, and handover status in the accounting records with the action time, spatial coordinates, load change values, equipment node identification, and physical handover actions in the interaction sequence diagram. When any key field in the accounting record is inconsistent with the corresponding field in the interaction sequence diagram, the system determines that the accounting record does not match the physical handover actions in the real-world scenario and outputs a document fraud alarm.

[0038] The document fraud alert includes risk waybill, risk node, accounting record field, IoT device field, and mismatch reason, and is used for digital financial risk analysis in the logistics industry.

[0039] Example 2: S1 includes acquiring the trajectory coordinates, access control records, and load changes of the truck terminal as initial equipment data; extracting spatial displacement features, identity verification identifiers, and load fluctuation values ​​from the initial equipment data to obtain a multi-dimensional physical state set; if the multi-dimensional physical state set undergoes a sudden change, the underlying physical access behavior is obtained; extracting the behavior trigger nodes corresponding to the underlying physical access behavior to determine the action time; sorting the underlying physical access behavior according to the action time to obtain the first time-series data arranged according to the action time.

[0040] In this embodiment, after receiving the data uploaded by the truck terminal, warehouse access control and smart shelf, the executing entity first uses the waybill identifier as the main associated field to classify the trajectory coordinates, access control records and load changes generated under the same logistics task into the same processing batch.

[0041] In this embodiment, the executing entity is the logistics industry digital financial risk analysis platform. This platform can be deployed on a local server, cloud server, or edge computing server to uniformly receive and process IoT device data, video data, and accounting record data in order to complete the logistics industry digital financial risk analysis.

[0042] Specifically, the data uploaded by the truck terminal includes vehicle identification, waybill identification, collection time, longitude, latitude, and positioning status; the data uploaded by the warehouse access control system includes access control device identification, warehouse identification, access subject identification, identity verification result, direction of passage, and time of passage; the data uploaded by the smart shelf includes shelf identification, location identification, cargo identification, collection time, and load-bearing value. For data whose trajectory coordinates, access control records, and load-bearing changes already contain the same waybill identification, the executing entity directly establishes an association based on the waybill identification; for data lacking a waybill identification but containing vehicle identification or cargo identification, the executing entity supplements the association based on the binding relationship between vehicle identification and waybill identification, and the binding relationship between cargo identification and waybill identification, and only retains the access control records and load-bearing records generated within 30 minutes before the truck enters the warehouse within a 50-meter radius of the coordinates to 60 minutes after entering, so that the initial equipment data corresponds to a defined logistics handover process.

[0043] Furthermore, when the executing entity extracts spatial displacement features from the initial equipment data, it uses the two trajectory coordinates continuously uploaded by the truck terminal as the calculation object.

[0044] Specifically, the executing entity first reads the longitude, latitude, and acquisition time of the previous trajectory coordinate, and then reads the longitude, latitude, and acquisition time of the next trajectory coordinate. Next, it converts the longitude and latitude of the two trajectory coordinates from degrees to radians. For distance calculation, the executing entity first calculates half the difference between the two latitudes and obtains the square of the sine of that value. Then, it calculates the cosine of each of the two latitudes and half the difference between the two longitudes, obtaining the square of the sine of that value. Afterward, it multiplies the square of the sine corresponding to the latitude difference with the square of the cosine of the two latitudes and the square of the sine corresponding to the longitude difference, and adds the results to obtain the trajectory angle parameter. Finally, it takes the square root of the trajectory angle parameter, takes the arcsine value, and multiplies it by twice the Earth's radius, 6,371,000 meters, to obtain the vehicle's displacement distance between the two consecutive trajectory coordinates. The executing entity subtracts the acquisition time of the previous trajectory coordinate from the acquisition time of the next trajectory coordinate to obtain the displacement time. Finally, it divides the displacement distance by the displacement time to obtain the vehicle's displacement speed. The executing entity also calculates the spatial distance between the next trajectory coordinates and the warehouse coordinates using the same distance calculation method; when the spatial distance is no more than 50 meters, it is recorded as the vehicle entering the warehouse area, and when the spatial distance is greater than 50 meters, it is recorded as the vehicle being outside the warehouse area.

[0045] The resulting spatial displacement features include vehicle identification, waybill identification, displacement distance, displacement speed, vehicle position relative to the warehouse, subsequent trajectory coordinates, and the time of acquisition of the subsequent trajectory coordinates.

[0046] Furthermore, when the executing entity extracts the identity verification identifier from the access control record, it reads the access entity identifier, identity verification result, access direction, and access time.

[0047] Specifically, when the authentication result is successful, the executing entity records the authentication status as successful; when the authentication result is unsuccessful, the executing entity records the authentication status as unsuccessful; when the passage direction is "entering", the executing entity records the passage direction as "inbound"; when the passage direction is "leaving", the executing entity records the passage direction as "outbound".

[0048] Subsequently, the executing entity compares the access entity identifier with the vehicle identifier, driver identifier, and warehouse worker identifier bound to the current waybill one by one; if the access entity identifier matches any of the bound identifiers, it is recorded as entity match; if the access entity identifier does not match any of the above bound identifiers, it is recorded as entity mismatch; the identity verification identifier consists of the access control device identifier, warehouse identifier, access entity identifier, identity verification status, passage direction, entity match result, and passage time, and is used to indicate the identity verification behavior that actually occurred at the warehouse access control point.

[0049] In this embodiment, when the executing entity extracts the load fluctuation value from the load change, it uses two consecutive load values ​​under the same shelf identifier and the same storage location identifier as the calculation object.

[0050] The execution entity reads the previous load-bearing value and its collection time, and then reads the next load-bearing value and its collection time. It then subtracts the previous load-bearing value from the next load-bearing value to obtain the load change. A load change greater than 0 is recorded as a load increase; a load change less than 0 is recorded as a load decrease; and a load change equal to 0 is recorded as no change. The execution entity subtracts the previous load-bearing collection time from the next load-bearing collection time to obtain the time taken for the load change. Subsequently, it reads the cargo weight registered in the waybill and calculates the difference between the absolute value of the load change and the registered cargo weight. If the registered cargo weight does not exceed 100 kg and the difference does not exceed 2 kg, the load change is determined to be consistent with the registered cargo weight; if the registered cargo weight exceeds 100 kg and the difference does not exceed 5 kg, the load change is determined to be consistent with the registered cargo weight; otherwise, the load change is determined to be inconsistent with the registered cargo weight.

[0051] Specifically, the load fluctuation value consists of shelf identification, location identification, cargo identification, load change amount, load change direction, load change time, weight difference, weight consistency result, and the time of the last load collection.

[0052] Furthermore, the executing entity will merge the extracted spatial displacement features, identity verification identifiers, and load fluctuation values ​​according to waybill identifiers, warehouse identifiers, and cargo identifiers to obtain a multidimensional physical state set.

[0053] Specifically, each status record in this set retains the device source, device identifier, service identifier, status content, and status time.

[0054] The status information corresponding to the vehicle trajectory includes displacement distance, displacement speed, and the vehicle's position relative to the warehouse; the status information corresponding to the access control record includes identity verification status, passage direction, and subject matching result; and the status information corresponding to load change includes load change amount, load change direction, and weight consistency result. After these status records are entered into the same set according to their time of occurrence, they can represent the sequence of occurrence of vehicle position changes, access control verification actions, and cargo load changes within the same logistics handover process.

[0055] Furthermore, when the executing entity determines whether a sudden change has occurred in the multidimensional physical state set, it makes judgments on vehicle trajectory, access control records, and load changes respectively. For vehicle trajectory, if the distance between the previous trajectory coordinate and the warehouse coordinate is greater than 50 meters, the distance between the next trajectory coordinate and the warehouse coordinate is not greater than 50 meters, and the time interval between the collection of the two trajectory coordinates is not more than 5 minutes, then a sudden change in vehicle arrival is determined, and a vehicle arrival behavior is generated. If the distance between the previous trajectory coordinate and the warehouse coordinate is not greater than 50 meters, the distance between the next trajectory coordinate and the warehouse coordinate is greater than 50 meters, and the time interval between the collection of the two trajectory coordinates is not more than 5 minutes, then a sudden change in vehicle departure is determined, and a vehicle departure behavior is generated. For access control records, if the identity verification status is passed, the subject matching result is subject matching, and the passage direction is the inbound direction, then an access control entry behavior is generated. If the identity verification status is passed, the subject matching result is subject matching, and the passage direction is the outbound direction, then an access control departure behavior is generated.

[0056] Specifically, for load changes, if the load change direction is load increase, the weight consistency result is consistent, and the load change time does not exceed 3 minutes, then a goods put-on action is generated; if the load change direction is load decrease, the weight consistency result is consistent, and the load change time does not exceed 3 minutes, then a goods take-off action is generated.

[0057] Furthermore, after obtaining the underlying physical access behavior, the executing entity extracts the behavior trigger node corresponding to the behavior and determines the action time based on the behavior trigger node.

[0058] Specifically, vehicle arrival and departure are triggered by the trajectory coordinates of the truck terminal, with the trigger node being the truck terminal that generates the next trajectory coordinate, and the action time being the acquisition time of the next trajectory coordinate. Access control entry and exit are triggered by the access records of the warehouse access control system, with the trigger node being the warehouse access control system that generates the access record, and the action time being the access time in the access control record. Goods shelving and goods de-shelving are triggered by changes in the load capacity of the smart shelf, with the trigger node being the smart shelf that generates the next load capacity value, and the action time being the acquisition time of the next load capacity value. The executing entity converts the above action times into Beijing time and writes them into the underlying physical access behavior record in the format of year, month, day, hour, minute, and second.

[0059] In this embodiment, the executing entity sorts the underlying physical access behaviors according to the action time to obtain the first time-series data arranged according to the action time.

[0060] Specifically, during sorting, the executing entities first arrange the actions in ascending order of time. When two underlying physical access behaviors have the same action time, they are arranged in the order of vehicle arrival, access control entry, goods shelving, goods de-shelving, access control exit, and vehicle departure. When both action time and behavior type are the same, they are arranged according to the character encoding order of the device identifier. After sorting, the executing entity writes a continuously increasing time sequence number for each underlying physical access behavior. Each record in the first time-series data includes a time sequence number, waybill identifier, goods identifier, warehouse identifier, behavior type, behavior trigger node, action time, spatial coordinates, authentication identifier, and load fluctuation value.

[0061] Through the above processing, trajectory coordinates, access control records, and load changes are converted into underlying physical access behavior data with clear triggering conditions, clear action times, and clear arrangement order.

[0062] Example 3: S2 includes extracting the access trajectory and node distance of the first time series data to construct an IoT access spatiotemporal graph; using a causal graph attention network to process the adjacency matrix and state sequence of the IoT access spatiotemporal graph to obtain causal weights; extracting cross-node collaborative features by weighting and summing the state sequences through causal weights, and determining the aggregation vector corresponding to the cross-node collaborative features; performing spatiotemporal mapping on the aggregation vector to obtain second time series data containing physical handover features and spatial location features.

[0063] In this embodiment, after obtaining the first time-series data, the executing entity first reads the waybill identifier, cargo identifier, behavior type, behavior trigger node, action time, spatial coordinates, identity verification identifier, and load fluctuation value from each time-series record.

[0064] Specifically, for records under the same waybill identifier, the executing entity determines the access trajectory according to the chronological order of the actions. The process of generating the access trajectory is as follows: the truck terminal corresponding to the vehicle arrival action is taken as the starting access node, the warehouse access control corresponding to the access control entry action is taken as the next access node, and the smart shelf corresponding to the goods shelving action is taken as the inbound access node. In the outbound stage, the smart shelf corresponding to the goods unshelving action is taken as the outbound access node, the warehouse access control corresponding to the access control departure action is taken as the release access node, and the truck terminal corresponding to the vehicle departure action is taken as the ending access node. When there is an action time succession relationship between two adjacent access nodes, the executing entity establishes an access edge between the two access nodes.

[0065] Furthermore, when the executing entity extracts the node distance, it processes it according to the scenario in which the node is located.

[0066] Specifically, for the distance between the vehicle trajectory node and the warehouse node, the executing entity reads the longitude and latitude from the vehicle trajectory coordinates and the longitude and latitude from the warehouse coordinates. After converting the longitude and latitude values ​​from angles to radians, it first calculates half of the difference between the two latitudes and obtains the square of its sine, then calculates half of the difference between the two longitudes and obtains the square of its sine. Subsequently, it multiplies the cosine value of each of the two latitudes by the square of the sine corresponding to the longitude difference, and adds it to the square of the sine corresponding to the latitude difference. Then, it takes the square root of the sum, takes the arcsine value, and multiplies it by twice the Earth's radius of 6,371,000 meters to obtain the spatial distance between the vehicle trajectory node and the warehouse node.

[0067] Specifically, for the distance between the warehouse access control node and the smart shelf node, the executing entity reads the horizontal and vertical coordinates of the access control in the warehouse plane coordinate system, and reads the horizontal and vertical coordinates of the smart shelf location in the warehouse plane coordinate system; calculates the difference between the two horizontal coordinates and the difference between the two vertical coordinates respectively, multiplies the two differences by themselves, adds them together, and then takes the square root of the sum to obtain the distance between the nodes inside the warehouse; in terms of action time, the executing entity subtracts the action time of the previous access node from the action time of the next access node to obtain the node time interval.

[0068] Furthermore, after the access trajectory and node distance are determined, the executing entity constructs an IoT access spatiotemporal map.

[0069] Specifically, the nodes in this spatiotemporal graph include truck terminal nodes, warehouse access control nodes, smart shelf nodes, warehouse nodes, and cargo nodes. Node attributes include device identifier, business identifier, action time, spatial coordinates, authentication identifier, and load fluctuation value. The edges in this spatiotemporal graph are formed by the access relationships between adjacent accessing nodes. Edge attributes include the identifier of the previous node, the identifier of the next node, the spatial distance between nodes, the time interval between nodes, and the behavior acceptance type. The executing entity generates an adjacency matrix according to the node arrangement order. The rows and columns of the adjacency matrix correspond to the nodes in the spatiotemporal graph. When there is an access edge between two nodes, a 1 is written in the corresponding position; when there is no access edge between two nodes, a 0 is written in the corresponding position.

[0070] Because logistics actions have a temporal sequence, the executing entity only writes connection values ​​in the direction from the earliest to the latest action time, and no longer writes connection values ​​in the reverse direction.

[0071] Furthermore, the executing entity simultaneously forms a state sequence based on the first time-series data. Each access node corresponds to one state record in the state sequence.

[0072] Specifically, the behavior types in the status record are converted into fixed behavior numbers in the order of vehicle arrival, access control entry, goods shelving, goods de-shelving, access control exit, and vehicle departure; the action time is converted into the time interval relative to the first action record of the current waybill; the spatial coordinates are retained as vehicle latitude and longitude, warehouse latitude and longitude, or internal warehouse plane coordinates; the identity verification identifier is retained as verification passed, verification failed, subject matched, or subject mismatched; the load fluctuation value is retained as the increase in load, decrease in load, time taken for load change, and weight consistency result.

[0073] The aforementioned state records are arranged in chronological order according to the first time series data, and are used as input data for the causal graph attention network.

[0074] Furthermore, when processing the adjacency matrix and state sequence, the causal graph attention network first determines whether there is a transitive relationship between nodes based on the adjacency matrix, and then determines the transitive direction based on the action time.

[0075] Specifically, if the action time of the previous access node is earlier than the action time of the next access node, and the corresponding position in the adjacency matrix is ​​1, the state record of the previous access node is included as a candidate input for the next access node; if the action time of the previous access node is not earlier than the action time of the next access node, or the corresponding position in the adjacency matrix is ​​0, the node does not participate in the causal weight calculation.

[0076] Through this process, vehicle arrival can affect access control entry, access control entry can affect goods placement, goods removal can affect access control departure, and access control departure can affect vehicle departure. Reverse actions are not involved in the transmission.

[0077] Specifically, the causal weight is determined by the behavior succession relationship, time interval, spatial distance, and business identifier consistency. The executing entity first determines whether the behavior of two adjacent access nodes conforms to the preset succession order. When the vehicle arrives and then the access control enters, the access control enters and then the goods are put on the shelf, the goods are taken off the shelf and then the access control leaves, and the access control leaves and then the vehicle leaves, the succession relationship is recorded as valid. If the order does not conform, the succession relationship is recorded as invalid. The executing entity then determines whether the node time interval and node spatial distance meet the corresponding thresholds.

[0078] The spatiotemporal conditions are considered valid if the time interval between vehicle arrival and access control is no more than 10 minutes, and the distance between the vehicle trajectory node and the warehouse node is no more than 50 meters; if the time interval between access control entry and goods placement is no more than 15 minutes, and the internal warehouse distance between the access control node and the shelf node is no more than 300 meters; if the time interval between goods removal from shelf and exit from access control is no more than 15 minutes, and the internal warehouse distance between the shelf node and the access control node is no more than 300 meters; and if the time interval between exit from access control and vehicle departure is no more than 10 minutes, and the distance between the vehicle trajectory node and the warehouse node is no more than 50 meters. The executing entity also compares the waybill identifier and goods identifier in two adjacent access nodes; if both are consistent, the business identifier consistency is considered valid. When the connection relationship, spatiotemporal conditions, and business identifier consistency are all valid, the access edge participates in causal weight generation; if any one of these is invalid, the weight of the access edge is reset to 0.

[0079] As a preferred implementation, for access edges participating in weight generation, the executing entity converts the valid results of the acceptance relationship, the valid results of the spatiotemporal conditions, and the valid results of the consistency of the business identifier into candidate weight values.

[0080] During the transformation, a valid connection relationship is recorded as 1, a valid spatiotemporal condition is recorded as 1, and a valid business identifier consistency is recorded as 1. The three values ​​are added together to obtain the candidate weight value of the access edge. For the same result node, the executing entity summarizes all candidate weight values ​​pointing to the result node and divides the candidate weight value of each access edge by the summary value to obtain the causal weight corresponding to the access edge. If there is only one valid access edge pointing to the result node, the causal weight of the access edge is 1; if there is no valid access edge pointing to the result node, the result node retains its own state record and does not perform cross-node state aggregation.

[0081] Furthermore, after obtaining the causal weights, the executing entity performs a weighted summation of the state sequence using the causal weights to extract cross-node collaborative features.

[0082] Specifically, the executing entity takes the result node as the processing object, reads the state record of the cause node pointing to that result node, and multiplies the time interval, spatial distance, identity verification status, subject matching result, load change, and weight consistency result in the cause node state record by their corresponding causal weights. Then, the products of the same field are summed to obtain the cross-node collaborative feature corresponding to that result node. For the access control entry node, this cross-node collaborative feature reflects the connection status between vehicle arrival and access control verification; for the goods shelving node, this cross-node collaborative feature reflects the connection status between access control entry and load increase; for the access control exit node, this cross-node collaborative feature reflects the connection status between goods removal and access control release; for the vehicle departure node, this cross-node collaborative feature reflects the connection status between access control exit and vehicle departure.

[0083] Specifically, the aggregation vector is obtained by merging the state of the resulting node itself with the cross-node collaborative features.

[0084] The executing entity takes the behavior type, action time, spatial coordinates, identity verification identifier or load fluctuation value of the result node as the first part, and the time connection result, spatial connection result, identity connection result and load connection result formed by weighting the cause node as the second part, and writes them into the same vector record in a fixed field order; this vector record is the aggregation vector corresponding to the cross-node collaborative features. Each aggregation vector corresponds to a result node, and retains the location of the result node in the access trajectory, the source device and the waybill to which it belongs.

[0085] Furthermore, when the executing entity performs spatiotemporal mapping on the aggregated vector, it writes the action time, time interval, and node order in the aggregated vector into the time dimension, writes the vehicle latitude and longitude, warehouse coordinates, internal planar coordinates of the warehouse, and node distance into the spatial dimension, and writes the identity verification status, entity matching result, load change direction, load change amount, and weight consistency result into the physical handover dimension.

[0086] Specifically, the aggregated vector formed by a vehicle arriving and pointing to the access control gate is mapped to the vehicle-to-warehouse handover feature; the aggregated vector formed by a access control gate entering and pointing to goods being shelved is mapped to the inbound handover feature; the aggregated vector formed by goods being unshelved and pointing to leaving the access control gate is mapped to the outbound handover feature; and the aggregated vector formed by leaving the access control gate and pointing to vehicle leaving is mapped to the vehicle-to-warehouse handover feature. After the above processing, the executing entity obtains second time-series data containing physical handover features and spatial location features. The records in this second time-series data include waybill identifier, goods identifier, node identifier, action time, physical handover features, spatial location features, node time interval, node spatial distance, and causal weight, which are used for subsequent encryption and solidification processing.

[0087] Example 4: S3 includes acquiring second time-series data containing physical handover features and spatial location features; performing feature segmentation processing on the second time-series data to obtain time-series data blocks; using a hash algorithm to extract a digest value from the time-series data blocks to obtain a data digest value; determining whether the length of the data digest value meets a preset threshold; if the length of the data digest value meets the preset threshold, then performing encryption and solidification processing based on the data digest value to determine the solidified ciphertext; adding a timestamp and node identifier to the solidified ciphertext to perform credential signing, generating a digital credential with a unique identifier and that cannot be tampered with.

[0088] In this embodiment, after the execution entity completes the generation of the second time-series data, it reads the time-series records under the same waybill identifier.

[0089] Specifically, each second time-series data item includes at least a waybill identifier, cargo identifier, node identifier, action time, physical handover characteristics, spatial location characteristics, node time interval, node spatial distance, and causal weight. The executing entity first collects data according to the waybill identifier, and then arranges it in order of action time from first to last; when the action times are the same, they are arranged according to the character encoding order of the node identifier.

[0090] Furthermore, after the executing entity obtains the second time-series data containing physical handover features and spatial location features, it performs feature block processing on it.

[0091] During the segmentation process, the executing entity divides each time-series record into a business identifier block, a time feature block, a spatial feature block, and a physical handover block according to the purpose of each field. The business identifier block contains the waybill identifier, cargo identifier, and node identifier; the time feature block contains the action time, node time interval, and time sequence number; the spatial feature block contains the vehicle's latitude and longitude, warehouse coordinates, shelf location coordinates, internal warehouse planar coordinates, and node spatial distance; and the physical handover block contains the vehicle arrival handover characteristics, inbound handover characteristics, outbound handover characteristics, vehicle departure handover characteristics, and causal weights.

[0092] Each time-series data block is accompanied by a waybill identifier, a node identifier, and a block number. The block number is assigned the values ​​1, 2, 3, and 4 in the order of business identifier block, time feature block, spatial feature block, and physical handover block.

[0093] In this embodiment, before extracting the summary, the executing entity performs normalization processing on the time-series data block. The normalization processing includes field completion, field sorting, time unification, coordinate unification, and numerical unification.

[0094] Among them, field completion means that all data blocks of the same type are written with field names and values ​​according to the preset field directory; fields without values ​​are written with empty placeholders; field sorting means that fields are arranged in a fixed order in the field directory; time unification means that both action time and generation time are converted to Beijing time; coordinate unification means that longitude and latitude are retained to 6 decimal places, and the internal planar coordinates of the warehouse are retained to 2 decimal places; numerical unification means that the spatial distance of nodes is retained to 2 decimal places, and the causal weight is retained to 6 decimal places. The normalized field content is encoded in UTF-8 and concatenated into a character sequence to be summarized in the order of field name, equal sign, field value, and separator.

[0095] Furthermore, when the executing entity uses a hash algorithm to extract the digest of the time-series data block, it takes the character sequence to be digested as the hash input. The hash algorithm can be either SHA-256 or SM3.

[0096] When using the SHA-256 algorithm, the executing entity performs padding, grouping, compression, and hexadecimal conversion on the character sequence to be digested, outputting a 256-bit digest result. When using the SM3 algorithm, the executing entity performs padding, message expansion, iterative compression, and hexadecimal conversion on the character sequence to be digested according to the processing flow of the national cryptographic hash algorithm, also outputting a 256-bit digest result. Since one hexadecimal character corresponds to four bits of binary data, the 256-bit digest result has 64 characters after being converted into a hexadecimal string, and these 64 characters are the data digest value.

[0097] Furthermore, when the executing entity determines whether the length of the data digest value meets the preset threshold, it first reads the hash algorithm identifier used by the current time-series data block, and then determines the digest length threshold based on the algorithm identifier. The digest length threshold is determined by the fixed output bit width of the hash algorithm.

[0098] When using either the SHA-256 or SM3 algorithm, the output bit length is 256 bits. Converting this to hexadecimal character length, 256 divided by 4 yields 64. Therefore, the preset threshold is set to 64 hexadecimal characters. The executing entity performs character counting on the data digest value and verifies whether the character content consists only of 0 to 9 and a to f. If the number of characters in the data digest value is equal to 64 and the character content conforms to the hexadecimal format, the length of the data digest value is determined to meet the preset threshold. If the data digest value is empty, the number of characters is not equal to 64, or non-hexadecimal characters are present, it is determined that the preset threshold is not met, and the time-series data block is marked as a digest aberration block and does not proceed to encryption and solidification processing.

[0099] Furthermore, when the length of the data digest value meets the preset threshold, the executing entity performs encryption and solidification processing based on the data digest value. The executing entity combines the data digest value, waybill identifier, cargo identifier, node identifier, block sequence number, digest generation time, and hash algorithm identifier into solidified plaintext according to a fixed field order.

[0100] After the plaintext is generated, the executing entity reads the key identifier and the encryption algorithm identifier. In the implementation using the SM4 algorithm, the executing entity uses a 128-bit business key to encrypt the plaintext in blocks of 128 bits each, with any blocks shorter than 128 bits padded according to a pre-defined method. In the implementation using the RSA algorithm, the executing entity uses the credential recipient's public key to encrypt the plaintext or the symmetric key of the plaintext. After the encryption operation is completed, the ciphertext is obtained. There is a definite correspondence between the ciphertext and the data digest value. During subsequent verification, the data digest value can be obtained by decrypting the ciphertext and then compared with the recalculated digest result.

[0101] As a preferred embodiment, the executing entity adds a timestamp and node identifier to the solidified encrypted text to perform credential signing.

[0102] Specifically, the timestamp is returned by a trusted time service and includes the date, time, time source identifier, and timestamp sequence number. The node identifier is the IoT device node identifier or data processing node identifier that generated the second time-series data. The executing entity concatenates the encrypted text, timestamp, node identifier, waybill identifier, cargo identifier, block sequence number, and key identifier in a fixed order to form the data to be signed. Subsequently, the executing entity calls the signing private key to perform a signing operation on the data to be signed, generating a signature value. The signing private key is stored in the cryptographic module or key management server and is not output along with the digital certificate. During signature verification, the verifier uses the public key corresponding to the signing private key to verify the signature.

[0103] When the verification is successful, it means that the encrypted text, timestamp, node identifier and their binding relationship have not been altered.

[0104] In this embodiment, the digital certificate consists of a certificate number, a waybill identifier, a cargo identifier, a node identifier, a block sequence number, a data digest value, a solidified ciphertext, a timestamp, a signature value, a hash algorithm identifier, an encryption algorithm identifier, and a signature certificate identifier. The certificate number is generated by recalculating the digest based on the waybill identifier, node identifier, block sequence number, and data digest value. Since the waybill identifier, node identifier, block sequence number, and data digest value all participate in the generation of the certificate number, each time-series data block corresponds to a unique certificate number.

[0105] If any field is modified, the recalculated data digest value, the decryption result of the solidified ciphertext, or the signature verification result will be inconsistent with the content stored in the digital certificate.

[0106] Example 5: S4 includes obtaining a digital certificate, extracting the original timestamp and original spatial coordinates of the digital certificate; calculating the initial alignment deviation based on the original timestamp and original spatial coordinates to obtain the target alignment deviation; calculating the time offset based on the target alignment deviation; if the time offset is greater than a preset offset threshold, generating a target compensation value; correcting the original timestamp with the target compensation value to obtain the target timestamp, and determining the time alignment relationship between IoT device nodes based on the target timestamp.

[0107] In this embodiment, after obtaining the digital certificate, the executing entity first reads the certificate number, waybill identifier, cargo identifier, node identifier, signature value, and solidified ciphertext. It then verifies the signature value using the public key corresponding to the signature private key. After successful verification, the executing entity decrypts the solidified ciphertext to obtain the data digest value corresponding to the digital certificate. Subsequently, it rereads the field content corresponding to the certificate number in the second time-series data, recalculates the digest value according to the field order used when generating the digital certificate, and compares the recalculated digest value with the decrypted data digest value. If they match, the executing entity confirms that the digital certificate has not been altered and extracts the original timestamp and original spatial coordinates from the encapsulation field of the digital certificate.

[0108] Among them, the original timestamp is the action time written when the IoT device node generates a physical handover behavior, the original spatial coordinates are the device location corresponding to the action time, the original spatial coordinates of the truck terminal node are the vehicle's longitude and latitude, the original spatial coordinates of the warehouse access control node are the longitude and latitude of the warehouse entrance where the access control is located or the warehouse plane coordinates, and the original spatial coordinates of the smart shelf node are the horizontal and vertical coordinates of the shelf location in the warehouse plane coordinate system.

[0109] In this embodiment, the executing entity groups digital vouchers belonging to the same cargo transfer event into the same aligned batch based on the waybill identifier and cargo identifier.

[0110] Specifically, the nodes within the alignment batch are divided into inbound node group and outbound node group according to the order of business actions. The inbound node group includes truck terminal node, warehouse access control node, and smart shelf node, corresponding to vehicle arrival, access control entry, and goods being put on the shelf; the outbound node group includes smart shelf node, warehouse access control node, and truck terminal node, corresponding to goods being taken off the shelf, access control exit, and vehicle departure. The executing entity simultaneously uses the original timestamp and original spatial coordinates to calculate the initial alignment deviation, so as to avoid the handover actions being incorrectly split due to differences in equipment clocks.

[0111] Furthermore, when calculating the initial alignment deviation, the executing entity first determines the adjacent node pairs. In the inbound stage, the adjacent node pairs are the truck terminal node and the warehouse access control node, and the warehouse access control node and the smart shelf node; in the outbound stage, the adjacent node pairs are the smart shelf node and the warehouse access control node, and the warehouse access control node and the truck terminal node.

[0112] For each pair of adjacent nodes, the executing entity subtracts the original timestamp of the previous node from the original timestamp of the later node to obtain the time difference with the direction of precedence. During the calculation, both original timestamps are first converted into the number of seconds from the same reference time, and then the number of seconds of the later node is subtracted from the number of seconds of the previous node. When the result is greater than 0, it means that the record of the later node is later than the record of the previous node; when the result is less than 0, it means that the record of the later node is earlier than the record of the previous node; when the result is equal to 0, it means that the records of the two nodes occurred in the same second.

[0113] Specifically, spatial deviations are calculated according to node coordinate types. The distance between the truck terminal node and the warehouse access control node is calculated using latitude and longitude. The executing entity first converts the vehicle longitude, vehicle latitude, access control longitude, and access control latitude into radians, and then obtains half of the difference between the two latitudes and half of the difference between the two longitudes. Subsequently, the square of the sine of half the latitude difference and the square of the sine of half the longitude difference are calculated, and the cosine of the two latitudes is multiplied by the square of the sine corresponding to the longitude difference. Then, the product is added to the square of the sine corresponding to the latitude difference, the square root of the sum is taken, and the arcsine value is taken. Finally, it is multiplied by twice the Earth's radius of 6,371,000 meters to obtain the spatial distance between the vehicle and the access control.

[0114] The warehouse access control node and the smart shelf node are calculated using the warehouse plane distance method. The execution subject reads the horizontal coordinates of the access control, the vertical coordinates of the access control, the horizontal coordinates of the shelf, and the vertical coordinates of the shelf, calculates the difference between the horizontal coordinates and the difference between the vertical coordinates respectively, multiplies the two differences by themselves and adds them together, and then takes the square root of the sum to obtain the internal warehouse distance between the access control and the shelf. The initial alignment deviation is composed of the time difference and spatial distance between adjacent node pairs.

[0115] As a preferred embodiment, the target alignment deviation is determined by the initial alignment deviation after spatial validity screening.

[0116] Specifically, the spatial threshold between the truck terminal node and the warehouse access control node is 50 meters. This threshold is determined by adding a 30-meter warehouse access control entrance recognition radius, a 15-meter truck terminal positioning error, and a 5-meter buffer distance at the warehouse entrance / exit. The spatial threshold between the warehouse access control node and the smart shelf node is 300 meters. This threshold is determined by adding a 250-meter distance from the warehouse access control to the farthest shelf aisle, a 40-meter forklift or manual handling detour distance, and a 10-meter location recognition error. The executing entity compares the calculated spatial distances with the corresponding spatial thresholds. When the spatial distance between the truck terminal node and the warehouse access control node does not exceed 50 meters, the time difference between the node pair is retained. When the internal warehouse distance between the warehouse access control node and the smart shelf node does not exceed 300 meters, the time difference between the node pair is retained. Node pairs exceeding the corresponding spatial thresholds are not included in the time alignment calculation and are recorded as spatially invalid node pairs. The retained time difference is the target alignment deviation.

[0117] Furthermore, the executing entity calculates the time offset based on the target alignment deviation.

[0118] In specific processing, the executing entity reads the directional time difference from the target alignment deviation and obtains its absolute number of seconds as the time offset. The directional time difference is used to determine the compensation direction, and the absolute number of seconds is used to determine whether the compensation condition has been met. For example, if the original timestamp of the warehouse access control node corresponds to 3600 seconds and the original timestamp of the truck terminal node corresponds to 3540 seconds, then the directional time difference is 60 seconds and the time offset is 60 seconds. If the original timestamp of the smart shelf node corresponds to 4210 seconds and the original timestamp of the warehouse access control node corresponds to 4230 seconds, then the directional time difference is -20 seconds and the time offset is 20 seconds.

[0119] Specifically, the offset threshold is determined according to the node device's acquisition cycle, data writing delay, and timing error, and is pre-written into the node parameter table.

[0120] The offset threshold between the truck terminal node and the warehouse access control node is obtained by adding the truck terminal data acquisition cycle, the warehouse access control data writing delay, and the device timing error. The truck terminal data acquisition cycle is taken as 30 seconds, the warehouse access control data writing delay is taken as 5 seconds, and the device timing error is taken as 5 seconds. The sum of these three is 40 seconds. Therefore, the offset threshold between the truck terminal node and the warehouse access control node is 40 seconds. The offset threshold between the warehouse access control node and the smart shelf node is obtained by adding the warehouse access control data writing delay, the smart shelf data acquisition cycle, and the device timing error. The warehouse access control data writing delay is taken as 5 seconds, the smart shelf data acquisition cycle is taken as 10 seconds, and the device timing error is taken as 5 seconds. The sum of these three is 20 seconds. Therefore, the offset threshold between the warehouse access control node and the smart shelf node is 20 seconds.

[0121] Among them, the device timing error comes from the timing gateway log; the device acquisition cycle comes from the device network access configuration; the data write delay comes from the write record from the device to the data receiving end. When the above parameters are updated, the execution subject uses the record in the parameter table that overwrites the current action time as the calculation basis.

[0122] Furthermore, when the time offset is greater than the offset threshold of the corresponding node pair, the execution entity generates the target compensation value.

[0123] Specifically, the target compensation value includes the compensation object, compensation direction, and compensation seconds. The compensation object is the next node in an adjacent node pair; the compensation direction is determined by the directional time difference. When the directional time difference is greater than 0, it indicates that the next node record is later than the previous node record, and the executing entity performs a forward correction on the original timestamp of the next node; when the directional time difference is less than 0, it indicates that the next node record is earlier than the previous node record, and the executing entity performs a backward correction on the original timestamp of the next node. The compensation seconds are obtained by subtracting the offset threshold from the time offset.

[0124] For example, the time offset between the truck terminal node and the warehouse access control node is 65 seconds, the offset threshold is 40 seconds, and the compensation time is 25 seconds; the time offset between the warehouse access control node and the smart shelf node is 32 seconds, the offset threshold is 20 seconds, and the compensation time is 12 seconds. When the time offset is not greater than the offset threshold, the executing entity does not generate a target compensation value and directly uses the original timestamp as the target timestamp.

[0125] As a preferred embodiment, the executing entity corrects the original timestamp by the target compensation value to obtain the target timestamp.

[0126] In the forward correction process, the executing entity deducts the compensation seconds from the original timestamp of the compensation object; in the backward correction process, the executing entity adds the compensation seconds to the original timestamp of the compensation object. After the correction is completed, the executing entity writes the target timestamp into the alignment result table. The alignment result table includes the voucher number, node identifier, original timestamp, target timestamp, target compensation value, compensation direction, and offset threshold. This alignment result table is associated with the digital voucher through the voucher number and does not rewrite the signature field and solidified ciphertext in the digital voucher body.

[0127] In this embodiment, the executing entity determines the time alignment relationship between IoT device nodes based on the target timestamp.

[0128] Specifically, during the inbound phase, if the vehicle arrives at the corresponding target timestamp no later than the access control entry timestamp, the access control entry timestamp no later than the goods being shelved, and the time interval between the vehicle's arrival and access control entry does not exceed 10 minutes, and the time interval between access control entry and goods being shelved does not exceed 15 minutes, then an inbound time alignment relationship is established between the truck terminal node, the warehouse access control node, and the smart shelf node. During the outbound phase, if the target timestamp for goods being unshelved is no later than the access control exit timestamp, the access control exit timestamp no later than the vehicle's exit timestamp, and the time interval between goods being unshelved and access control exit does not exceed 15 minutes, and the time interval between access control exit and vehicle exit does not exceed 10 minutes, then an outbound time alignment relationship is established between the smart shelf node, the warehouse access control node, and the truck terminal node.

[0129] The aforementioned time interval thresholds are determined by warehouse loading and unloading operation parameters. Specifically, the 10 minutes from vehicle arrival to access control entry is obtained by adding 5 minutes for vehicle queuing confirmation, 2 minutes for access control identity verification, and 3 minutes for vehicle entry buffer time. The 15 minutes from access control entry to goods being shelved is obtained by adding 5 minutes for inbound travel time, 5 minutes for unloading preparation time, and 5 minutes for shelf weighing and recording time. The 15 minutes from goods being unshelved to exiting access control is obtained by adding 5 minutes for picking and transferring time, 5 minutes for outbound verification time, and 5 minutes for access control release preparation time. The 10 minutes from exiting access control to vehicle departure is obtained by adding 5 minutes for loading confirmation time, 2 minutes for exit verification time, and 3 minutes for vehicle departure buffer time.

[0130] Therefore, after deviation calculation, threshold judgment and compensation correction, the original timestamp and original spatial coordinates in the digital certificate form a time alignment relationship used to splice the underlying physical access behavior.

[0131] Example 6: S5 includes obtaining the underlying physical access records, extracting access timestamps from the underlying physical access records to generate discrete access segments; matching the discrete access segments to obtain time alignment features; determining whether the time alignment features meet a preset threshold; if the time alignment features meet the preset threshold, then performing fusion processing on the discrete access segments to obtain a behavior splicing sequence; constructing a temporal correlation matrix based on the behavior splicing sequence; and constructing an interaction time sequence diagram reflecting the actual flow status of goods based on the temporal correlation matrix.

[0132] In this embodiment, after the executing entity obtains the underlying physical access records, it first collects them according to the waybill identifier and the cargo identifier, so that the access records generated by the same cargo in the same logistics flow process enter the same processing batch.

[0133] Specifically, the underlying physical access records include vehicle arrival and departure records generated by the truck terminal, access control entry and departure records generated by the warehouse access control system, and goods placement and removal records generated by the smart shelf. Each underlying physical access record includes at least the device node identifier, behavior type, access timestamp, spatial coordinates, identity verification result, load change direction, and load change value. The executing entity first removes records that fail digital credential verification, and then arranges them in order of access timestamp from first to last. If the access timestamps are the same, they are arranged in the business sequence of vehicle arrival, access control entry, goods placement, goods removal, access control departure, and vehicle departure.

[0134] Furthermore, the executing entity extracts access timestamps from the underlying physical access records and generates discrete access segments. For vehicle arrival records, the access timestamp is the collection time corresponding to the first time the truck terminal enters the warehouse within a 50-meter radius of the warehouse coordinates; for access control entry records, the access timestamp is the passage time when the warehouse access control verification is successful and the direction of passage is entering; for goods shelving records, the access timestamp is the collection time after the intelligent shelf load capacity is increased; for goods de-shelving records, the access timestamp is the collection time after the intelligent shelf load capacity is decreased; for access control exit records, the access timestamp is the passage time when the warehouse access control verification is successful and the direction of passage is leaving; for vehicle departure records, the access timestamp is the collection time corresponding to the first time the truck terminal leaves the warehouse within a 50-meter radius of the warehouse coordinates.

[0135] Each discrete access segment is written with a segment number, waybill identifier, cargo identifier, device node identifier, behavior type, access timestamp, spatial coordinates, and trigger field, so that the data originally distributed across different devices is converted into time-aligned segment data.

[0136] Furthermore, after generating discrete access fragments, the executing entity matches the fragments according to the logistics handover sequence.

[0137] In the warehousing stage, starting with the vehicle arrival segment, the system searches for access control entry segments under the same waybill identifier and the same cargo identifier, and then uses this access control entry segment as the preceding segment to search for cargo shelving segments. In the outbound stage, starting with the cargo de-shelving segment, the system searches for access control exit segments under the same waybill identifier and the same cargo identifier, and then uses this access control exit segment as the preceding segment to search for vehicle departure segments. For each candidate segment pair formed, the executing entity converts the access timestamp of the subsequent segment into seconds and the access timestamp of the preceding segment into seconds, and subtracts the corresponding seconds of the preceding segment from the corresponding seconds of the subsequent segment to obtain the time interval between adjacent segments. When the time interval is less than 0, it indicates that the candidate segment pair does not conform to the order of actions, and the executing entity does not consider the candidate segment pair as a valid matching object.

[0138] In this embodiment, the time alignment feature consists of the time interval between adjacent segments, spatial distance, behavior succession result, and service identifier consistency result. The spatial distance is calculated according to the device type corresponding to the segment.

[0139] Specifically, between the vehicle arrival segment and the access control entry segment, the executing entity reads the vehicle's longitude and latitude, the access control's longitude and latitude, converts the longitude and latitude into radians, processes the latitude and longitude differences separately, and then converts them into metric distances between the vehicle and the access control based on the Earth's radius of 6,371,000 meters. Between the access control entry segment and the goods shelving segment, the executing entity reads the horizontal and vertical coordinates of the access control in the warehouse's planar coordinate system, and reads the horizontal and vertical coordinates of the shelf locations in the same coordinate system. First, it calculates the difference between the two horizontal coordinates and the difference between the two vertical coordinates, then multiplies the two differences by themselves and adds them together. Finally, it takes the square root of the sum to obtain the internal distance of the warehouse.

[0140] Among them, the spatial distances between the goods unloading segment and the access control exit segment, and between the access control exit segment and the vehicle departure segment are calculated separately according to the same coordinate type. The behavior acceptance result is judged according to the segment type. When the vehicle arrives and accepts the access control entry, the access control enters and accepts the goods loading, the goods are unloaded and accept the access control exit, or the access control exits and accepts the vehicle departure, it is recorded as a valid acceptance. Other combinations are recorded as invalid acceptance. The consistency result of the business identifier is obtained by comparing the waybill identifier and the goods identifier. When both identifiers are consistent, it is recorded as a consistent business.

[0141] In this embodiment, when the executing entity determines whether the time alignment feature meets the preset threshold, it calls the corresponding time threshold and spatial threshold according to the segment continuation type. The time threshold from the arrival of the vehicle to the access control entry segment is 10 minutes, which is determined by adding the vehicle queuing confirmation time of 5 minutes, the access control identity verification time of 2 minutes, and the vehicle entry buffer time of 3 minutes. The spatial threshold is 50 meters, which is determined by adding the access control entrance recognition radius of 30 meters, the truck terminal positioning error of 15 meters, and the warehouse entrance and exit road buffer distance of 5 meters.

[0142] Specifically, the time threshold from the access control entry segment to the goods shelving segment is 15 minutes, which is determined by adding 5 minutes for inbound travel time, 5 minutes for unloading preparation time, and 5 minutes for shelving weighing and writing time; its spatial threshold is 300 meters, which is determined by adding 250 meters for the distance from the access control to the farthest shelving aisle, 40 meters for handling detour distance, and 10 meters for cargo location identification error; the time threshold from the goods unshelving segment to the access control exit segment is 15 minutes, which is determined by adding 5 minutes for picking and transferring time, 5 minutes for outbound verification time, and 5 minutes for access control release preparation time; its spatial threshold is also 300 meters; the time threshold from the access control exit segment to the vehicle exit segment is 10 minutes, which is determined by adding 5 minutes for loading confirmation time, 2 minutes for exit verification time, and 3 minutes for vehicle departure buffer time; its spatial threshold is 50 meters; when adjacent segments simultaneously meet the conditions that the time interval does not exceed the corresponding time threshold, the spatial distance does not exceed the corresponding spatial threshold, the behavior is valid, and the business is consistent, the executing entity determines that the time alignment feature meets the preset threshold.

[0143] Furthermore, when the time alignment feature meets a preset threshold, the executing entity performs fusion processing on the discrete access fragments to obtain a behavior splicing sequence.

[0144] The fusion process is based on the original records of the fragments, without rewriting the original access timestamps and original spatial coordinates, but instead writes the succession relationship between the fragments; the inbound behavior splicing sequence is composed of the vehicle arrival fragment, the access control entry fragment, and the goods shelving fragment in sequence; the outbound behavior splicing sequence is composed of the goods unshelving fragment, the access control exit fragment, and the vehicle departure fragment in sequence.

[0145] If a preceding segment corresponds to more than two candidate following segments, the executing entity first compares the time intervals of the candidate segment pairs and selects the candidate following segment with the smallest time interval value; if the time intervals are the same, then the spatial distances are compared and the candidate following segment with the smallest spatial distance value is selected; if both the time interval and spatial distance are the same, the candidate following segment with the device node identifier character encoding order first is selected; each behavior splicing sequence is written with the link number, segment sequence number, waybill identifier, cargo identifier, behavior type, device node identifier, access timestamp, spatial coordinates, time interval, and spatial distance.

[0146] As a preferred implementation, the executing entity constructs a temporal correlation matrix based on the behavior splicing sequence.

[0147] Specifically, the rows and columns of the matrix correspond to segment nodes in the row concatenation sequence. Segment nodes are arranged according to link number and segment sequence number. When the preceding and following segments belong to the same row concatenation sequence, and the temporal alignment feature between the two segments meets the corresponding threshold, the executing entity writes 1 at the intersection of the row containing the preceding segment and the column containing the following segment in the matrix, indicating a temporal correlation. If there is no correlation or the threshold is not met, 0 is written at the corresponding position. Because logistics access behavior has a definite temporal direction, the matrix only writes correlation values ​​at positions where the preceding segment points to the following segment, and not at reverse positions.

[0148] The executing entity also adds time interval, spatial distance, behavior inheritance type and link number to the matrix position with a value of 1, so that the matrix not only expresses whether there is a connection between segments, but also expresses the spatiotemporal conditions on which the connection is based.

[0149] Specifically, the executing entity constructs an interaction sequence diagram based on the temporal correlation matrix.

[0150] In the interaction sequence diagram, the nodes are formed by fragment nodes in the behavior splicing sequence. The node attributes include waybill identifier, cargo identifier, equipment node identifier, behavior type, access timestamp, spatial coordinates, authentication result, and load fluctuation value. The directed edges in the interaction sequence diagram are formed by the positions with a value of 1 in the temporal association matrix. The edge attributes include link number, time interval, spatial distance, and behavior acceptance type. The inbound link is represented in the interaction sequence diagram as the truck terminal node pointing to the warehouse access control node, and then the warehouse access control node pointing to the smart shelf node; the outbound link is represented as the smart shelf node pointing to the warehouse access control node, and then the warehouse access control node pointing to the truck terminal node.

[0151] Example 7: S6 includes obtaining an interaction sequence diagram, extracting flow nodes from the interaction sequence diagram to construct a flow path; processing the flow path using the isolated forest algorithm to obtain a path feature vector, and calculating an anomaly score based on the path feature vector; if the anomaly score is greater than a preset threshold, extracting the deviation degree from the flow path; comparing the deviation degree with a pre-established baseline path to determine whether the flow path deviates from the pre-established baseline path.

[0152] In this embodiment, after obtaining the interaction sequence diagram, the executing entity reads the node data and edge data within it. The node data includes waybill identifier, cargo identifier, device node identifier, behavior type, access timestamp, spatial coordinates, authentication result, and load fluctuation value; the edge data includes preceding node identifier, following node identifier, time interval, spatial distance, and behavior acceptance type.

[0153] The execution entity uses waybill identifiers and cargo identifiers as search criteria to extract flow nodes belonging to the same cargo from the interaction sequence graph. The order of nodes is determined according to the direction of directed edges: in the warehousing stage, the node order is vehicle arrival node, access control entry node, and cargo shelving node; in the outbound stage, the node order is cargo de-shelving node, access control exit node, and vehicle departure node. If a preceding node has more than two directed edges pointing to a subsequent node, the execution entity first compares the access timestamps of each directed edge and selects the directed edge whose subsequent node time is later than the preceding node time. If there are more than two directed edges satisfying this condition, the directed edge with the smallest spatial distance is selected; if the spatial distances are the same, the directed edge with the earlier device node identifier character encoding is selected. After this processing, a flow path with a defined node order is obtained.

[0154] Furthermore, the executing entity converts the flow path into a path feature vector that the isolated forest algorithm can process.

[0155] Specifically, the executing entity reads each node field in the flow path and converts them into numerical fields. For example, the vehicle arrival node is converted to vehicle arrival time, vehicle arrival coordinates, and the distance between the vehicle and the warehouse entrance; the access control entry node is converted to access control entry time, authentication result, and subject matching result; the goods shelving node is converted to shelving time, shelf location coordinates, load increase value, and weight consistency result; the goods de-shelving node is converted to de-shelving time, load decrease value, and weight consistency result; the access control departure node is converted to access control departure time, authentication result, and direction of travel; and the vehicle departure node is converted to vehicle departure time, vehicle departure coordinates, and the distance between the vehicle and the warehouse entrance. A value of 1 is written when authentication is successful, subject matches, and weight matches; a value of 0 is written when authentication fails, subject does not match, or weight does not match.

[0156] In this embodiment, the time interval between adjacent flow nodes is calculated by the access timestamp. The execution entity first converts the access timestamp of the preceding node into the number of seconds calculated from the same reference time, then converts the access timestamp of the following node into the number of seconds under the same reference, and then subtracts the number of seconds corresponding to the preceding node from the number of seconds corresponding to the following node to obtain the time interval.

[0157] The spatial distance between adjacent circulation nodes is calculated according to coordinate type. The distance between vehicle nodes and warehouse entrance nodes is calculated using latitude and longitude. The executing entity converts longitude and latitude into radians, calculates the trigonometric function results corresponding to the latitude and longitude differences, and then converts them into metric distances using the Earth's radius of 6,371,000 meters. The distance between nodes inside the warehouse is calculated using planar distance. The executing entity reads the horizontal and vertical coordinates of two nodes, calculates the difference between the horizontal and vertical coordinates, multiplies each difference by itself, adds them together, and then takes the square root of the sum to obtain the distance inside the warehouse.

[0158] Furthermore, the path feature vector consists of node order, time interval between adjacent nodes, spatial distance between adjacent nodes, load change value, identity verification result, subject matching result, and weight consistency result.

[0159] Specifically, the execution entity writes the above values ​​in a fixed field order, consistent with the field order used during the training phase of the isolated forest model. For missing node fields, the execution entity does not use null values ​​in the calculation, but instead writes a preset missing value of -1 and simultaneously writes 1 in the missing value field; when the field exists, the missing value field is written with 0. Thus, the flow path is converted into a path feature vector with a defined number and order of fields.

[0160] Specifically, the execution entity uses the isolated forest algorithm to process the path feature vector. The isolated forest model is trained from historical waybill path samples. The training samples include historical normal waybill paths that have not generated anomaly markers and waybill paths that have been manually reviewed as abnormal. Each path in the training samples is generated according to the field order of the current path feature vector.

[0161] During the training of each isolated tree, the execution entity extracts a fixed number of path feature vectors from the training samples, randomly selects one feature field, and generates a split value between the minimum and maximum values ​​of that feature field. Samples with values ​​less than the split value enter the left child node, and samples with values ​​not less than the split value enter the right child node. This splitting process continues until a sample is isolated or a preset tree depth is reached. After the current path feature vector is input into the isolated forest model, the execution entity records the number of node layers it traverses when isolated in each isolated tree, then adds up the node layers corresponding to all isolated trees and divides by the number of isolated trees to obtain the average number of isolation layers.

[0162] In this embodiment, the anomaly score is calculated based on the average number of isolation layers.

[0163] Specifically, the execution entity first reads the number of samples used for each isolated tree during training and obtains the theoretical average isolation layer based on this number of samples. Then, it calculates the ratio between the average isolation layer of the current path and the theoretical average isolation layer to obtain the standardized isolation result. Then, according to the anomaly score conversion method of the isolated forest algorithm, the standardized isolation result is converted into an anomaly score between 0 and 1. The shorter the average isolation layer, the earlier the feature vector of the path is isolated in the random segmentation process, and the closer the converted anomaly score is to 1. This anomaly score is used to represent the degree of difference between the current flow path and the normal flow path in the training samples.

[0164] Specifically, the preset threshold is determined by the verification samples. The executing entity selects 1,000 historical normal waybill paths and 200 verified abnormal waybill paths as verification samples, and uses the trained isolated forest model to calculate the abnormality scores of the 1,200 verification samples respectively.

[0165] Subsequently, the executing entity starts from 0.50 and takes values ​​in increments of 0.01 until it reaches 0.90, forming 41 candidate thresholds. For each candidate threshold, the executing entity counts the number of normal waybill paths that are judged as abnormal, and the number of abnormal waybill paths that have been reviewed that are judged as normal.

[0166] Specifically, the selection criteria for candidate thresholds are as follows: the number of normal waybills being judged as abnormal does not exceed 50, and the number of abnormal waybills being judged as normal after review is the minimum among the candidate thresholds that meet the aforementioned conditions; if there are more than two candidate thresholds that meet the above conditions and have the same number of missed judgments, then the candidate threshold with the smallest value is selected. According to the above rules, the preset threshold in this embodiment is determined to be 0.65, and it is written into the threshold parameter table along with the route type, warehouse type and cargo category.

[0167] Furthermore, when the abnormal score of the current flow path is greater than 0.65, the executing entity extracts the deviation degree from the flow path.

[0168] Specifically, the degree of deviation includes the degree of deviation in node sequence, time, space, and load. The degree of deviation in node sequence is obtained by comparing the node sequence of the current flow path with that of the reference path. When the current flow path is missing one necessary node from the reference path, the deviation count increases by 1. When the current flow path has one node outside the reference path, the deviation count increases by 1. When the order of two adjacent nodes in the current flow path is inconsistent with that of the reference path, the deviation count increases by 1. The degree of time deviation is obtained by comparing the time interval between adjacent nodes. The executing entity subtracts the standard time interval of the corresponding node segment in the reference path from the time interval between adjacent nodes in the current flow path, and takes the absolute number of seconds of the difference as the time deviation value of that node segment.

[0169] The spatial deviation is obtained by comparing node coordinates; vehicle nodes are calculated based on latitude and longitude distances, while warehouse internal nodes are calculated based on planar distances, and the resulting distances are used as the spatial deviation values. The load deviation is obtained by comparing the load change value with the registered cargo weight; the executing entity subtracts the registered cargo weight from the absolute value of the load change value and takes the absolute value of the difference as the load deviation value.

[0170] As a preferred implementation, the baseline route is pre-generated from historical normal waybills under the same route, the same warehouse, the same carrier, and the same cargo category.

[0171] When establishing the baseline path, the executing entity first filters completed historical waybills that have not generated anomaly markers. Then, from each historical waybill, it extracts the vehicle arrival node, access control entry node, goods shelving node, goods de-shelving node, access control exit node, and vehicle departure node, forming a node order according to the access timestamp. For each node segment, the executing entity adds the corresponding time intervals from the historical normal waybills and divides by the number of historical waybills involved in the calculation to obtain the standard time interval. For each node location, the executing entity adds the longitude, latitude, warehouse internal horizontal coordinates, and warehouse internal vertical coordinates from the historical normal waybills and divides each by the number of historical waybills involved in the calculation to obtain the standard node coordinates. The baseline path consists of the standard node order, standard time intervals, standard node coordinates, and registered cargo weight.

[0172] Furthermore, the executing entity compares the degree of deviation with the baseline path and determines whether the flow path deviates from the baseline path.

[0173] Specifically, a node deviation is determined when the node sequence deviation count is not zero. A time deviation is also determined when the time deviation value of any node segment exceeds the corresponding time tolerance. The time tolerance is determined according to the work composition: the time tolerance from vehicle arrival to access control entry is 3 minutes, determined by adding 1 minute for vehicle queuing confirmation time, 1 minute for access control identity verification floating time, and 1 minute for gate entry buffer time; the time tolerance from access control entry to goods being shelved is 5 minutes, determined by adding 2 minutes for warehouse driving floating time, 2 minutes for unloading preparation floating time, and 1 minute for shelf weighing and recording floating time; the time tolerance from goods being unshelved to leaving access control is 5 minutes, determined by adding 2 minutes for picking and transferring floating time, 2 minutes for outbound verification floating time, and 1 minute for release preparation floating time; the time tolerance from leaving access control to vehicle departure is 3 minutes, determined by adding 1 minute for loading confirmation floating time, 1 minute for gate exit verification floating time, and 1 minute for vehicle departure buffer time.

[0174] Furthermore, if the spatial deviation value of any flow node exceeds the corresponding spatial tolerance, it is determined that there is a spatial deviation.

[0175] Specifically, the spatial tolerance for vehicle nodes is 50 meters, determined by adding a 30-meter access control radius, a 15-meter truck terminal positioning error, and a 5-meter road buffer distance. The spatial tolerance for internal warehouse nodes is 300 meters, determined by adding a 250-meter distance from the access control to the furthest shelf aisle, a 40-meter handling detour distance, and a 10-meter location identification error. A load deviation is considered to exist when it exceeds the corresponding weight tolerance. The weight tolerance is 2 kg for registered goods weighing no more than 100 kg and 5 kg for registered goods weighing more than 100 kg. If any of the node deviation, time deviation, spatial deviation, or load deviation occurs, the executing entity determines that the flow path deviates from the pre-established baseline path; if none of the above deviations occur, the flow path is determined not to deviate from the pre-established baseline path.

[0176] Example 8: S7 includes parsing spatial coordinates and dwell time from waybills that deviate from the baseline path to calculate the trajectory deviation angle; if the trajectory deviation angle is greater than a preset deviation angle threshold, an anomaly marker is generated for the waybill; based on the anomaly marker, node states are extracted from the interaction sequence diagram, feature mapping is performed on the node states to obtain risk nodes and construct a scene topology; based on the scene topology, the interaction sequence diagram is spatially mapped to reconstruct a real scene containing risk nodes.

[0177] In this embodiment, after receiving a waybill that has been determined to deviate from the baseline path, the executing entity uses the waybill identifier and cargo identifier as search criteria to read the corresponding flow nodes, directed edges between nodes, spatial coordinates, access timestamps, dwell records, and behavior types from the interaction sequence diagram. Flow nodes include vehicle arrival nodes, access control entry nodes, cargo shelving nodes, cargo de-shelving nodes, access control exit nodes, and vehicle departure nodes. The executing entity arranges these nodes according to the direction of the directed edges, so that vehicle arrival, access control entry, cargo shelving, cargo de-shelving, access control exit, and vehicle departure form the same waybill flow chain.

[0178] Furthermore, when the executing entity analyzes spatial coordinates, it first converts coordinates from different sources to the same metric coordinate system.

[0179] Specifically, for the latitude and longitude coordinates uploaded by the truck terminal, the executing entity uses the coordinates of the starting warehouse of the baseline path as the coordinate origin, converts the longitude difference of the vehicle node into east-west distance, and the latitude difference into north-south distance.

[0180] During the conversion, the executing entity first converts the longitude and latitude values ​​from degrees to radians, and then uses the Earth's radius of 6,371,000 meters to calculate the ground distance corresponding to the longitude and latitude differences. For warehouse access control nodes and smart shelf nodes, the executing entity reads the horizontal and vertical coordinates in the warehouse plane coordinate system. When the access control node only records longitude and latitude coordinates, the executing entity converts them to metric coordinates with the warehouse entrance as the origin, according to the conversion method of truck terminal coordinates. After the coordinate conversion, vehicle nodes, access control nodes, and shelf nodes are all represented by horizontal and vertical coordinates.

[0181] Furthermore, when the executing entity analyzes the dwell time, it determines the start and end times according to the node type. The dwell time of the vehicle within the warehouse is determined by the access timestamp of the truck terminal first entering the warehouse within a 50-meter radius and the access timestamp of the truck terminal first leaving the area. The executing entity converts both timestamps into seconds and subtracts them to obtain the vehicle dwell time.

[0182] The warehouse access control dwell time is calculated by subtracting the entry time from the exit time, converting both to seconds. Similarly, the smart shelf dwell time is calculated by subtracting the goods' removal time from the shelf, using the goods' placement time as the start time and removal time as the end time. If a node lacks an end time, the executing entity uses the access timestamp of the next node in the flow as the end time. If a node lacks a start time, the executing entity records this node as a time-missing node, and this node is not included in the trajectory deviation calculation.

[0183] Specifically, the trajectory deflection angle is calculated using three consecutive flow nodes with spatial coordinates. The executing entity first takes the previous node, the current node, and the next node, calculates the lateral and longitudinal displacements from the previous node to the current node, and then calculates the lateral and longitudinal displacements from the current node to the next node.

[0184] Subsequently, the executing entity multiplies the previous lateral displacement with the next lateral displacement, and the previous longitudinal displacement with the next longitudinal displacement, and adds the two products to obtain the directional product result. The executing entity calculates the length of the previous displacement and the length of the next displacement respectively. The calculation process is as follows: multiply the lateral displacement by itself, multiply the longitudinal displacement by itself, add the two multiplication results, and then take the square root of the sum. The executing entity divides the directional product result by the result of multiplying the length of the previous displacement by the length of the next displacement to obtain the cosine value of the included angle. Then, it performs an inverse cosine operation on the included angle cosine value and converts it into an angle value. This angle value is the trajectory deflection angle of the current node. When the length of the previous displacement is equal to 0 or the length of the next displacement is equal to 0, the executing entity records the node as a repeated coordinate node and stops calculating the trajectory deflection angle of the node.

[0185] In this embodiment, the dwell time is used to correct the trajectory deviation angle. The execution subject reads the actual dwell time of the current node and the reference dwell time of the same node in the reference path.

[0186] Specifically, the baseline dwell time is calculated from historical normal waybills under the same route, warehouse, carrier, and cargo category. The executing entity adds up the dwell times of the same node in the historical normal waybills used for calculation, and then divides by the number of historical normal waybills to obtain the baseline dwell time for that node. The executing entity subtracts the baseline dwell time from the actual dwell time and takes the absolute value of the difference to obtain the dwell deviation. When the dwell deviation does not exceed 5 minutes, the trajectory angle is not corrected; when the dwell deviation exceeds 5 minutes but does not exceed 15 minutes, the executing entity increases the trajectory angle by 5 degrees; when the dwell deviation exceeds 15 minutes, the executing entity increases the trajectory angle by 10 degrees. The corrected angle is used as the trajectory angle for anomaly detection.

[0187] Furthermore, the deflection angle threshold is determined according to the scenario to which the node belongs. The deflection angle threshold for trunk transportation nodes is 45 degrees, which is obtained by adding the allowable turning angle of the baseline path (20 degrees), the vehicle positioning error compensation angle (10 degrees), and the road turning margin (15 degrees). The deflection angle threshold for warehouse park nodes is 60 degrees, which is obtained by adding the allowable turning angle within the park (30 degrees), the vehicle positioning error compensation angle (10 degrees), and the detour margin of the loading and unloading area (20 degrees). The deflection angle threshold for nodes inside the warehouse is 70 degrees, which is obtained by adding the allowable turning angle of the warehouse passage (40 degrees), the cargo location coordinate error compensation angle (10 degrees), and the turning margin of the handling passage (20 degrees).

[0188] The executing entity calls the corresponding deflection threshold based on whether the current node belongs to a trunk transportation node, a warehouse park node, or a warehouse internal node.

[0189] Furthermore, when the trajectory deviation angle of any transfer node is greater than the corresponding deviation angle threshold, the executing entity generates an anomaly mark for that waybill.

[0190] Specifically, the anomaly marker is written into the waybill identifier, cargo identifier, anomaly node identifier, anomaly node behavior type, trajectory deviation angle, deviation angle threshold, dwell deviation, anomaly occurrence time, and anomaly spatial coordinates. When there are more than two anomaly nodes in the same waybill, the executing entity writes the anomaly detail number under the same waybill anomaly marker and associates each anomaly detail with the node identifier and edge identifier in the interaction sequence diagram.

[0191] In this embodiment, the execution entity extracts the node state from the interaction sequence diagram based on the exception marker.

[0192] Specifically, for an abnormal node, the executing entity reads the node's behavior type, access timestamp, spatial coordinates, authentication result, subject matching result, load fluctuation value, weight consistency result, number of inbound edges, number of outbound edges, and time interval between adjacent nodes; at the same time, it reads the status of one preceding node and one following node of the abnormal node; if the preceding node or the following node does not exist, the executing entity writes a missing flag at the corresponding position. Through this process, the upstream and downstream status of the abnormal node in the waybill flow link is preserved.

[0193] As a preferred embodiment, the executing entity performs feature mapping on node states to identify risky nodes.

[0194] Specifically, authentication success is mapped to 0, and authentication failure is mapped to 1; subject matching is mapped to 0, and subject mismatch is mapped to 1; weight consistency is mapped to 0, and weight inconsistency is mapped to 1; node existence is mapped to 0, and node absence is mapped to 1; trajectory deviation angle not greater than the deviation angle threshold is mapped to 0, and trajectory deviation angle greater than the deviation angle threshold is mapped to 1; dwell deviation not exceeding 15 minutes is mapped to 0, and dwell deviation exceeding 15 minutes is mapped to 1. The executing entity adds the above mapping values ​​to obtain the node risk count; when the node risk count is not less than 2, the executing entity identifies the node as a risk node; when the node risk count is less than 2, the node only participates in scenario reconstruction as an associated node.

[0195] Furthermore, after identifying the risk nodes, the executing entity constructs the scene topology. The scene topology is centered on the risk nodes and includes the risk nodes, their predecessor nodes, successor nodes, and directed edges directly connected to these nodes.

[0196] Specifically, the topology node attributes include device node identifier, behavior type, target timestamp, spatial coordinates, and node risk count; the topology edge attributes include preceding node identifier, following node identifier, time interval, spatial distance, and behavior connection type; when the risk node is a vehicle node, the scene topology retains the vehicle trajectory point, warehouse entrance location, and access control node; when the risk node is an access control node, the scene topology retains the vehicle node, access control node, and shelf node; when the risk node is a shelf node, the scene topology retains the access control node, shelf node, and load change record.

[0197] Specifically, the executing entity performs spatial mapping of the interaction sequence diagram based on the scene topology in order to reconstruct a real scene containing risk nodes.

[0198] In the spatial mapping process, the executing entity converts the vehicle's latitude and longitude into metric coordinates with the warehouse entrance as the origin. It then writes the access control coordinates and shelf location coordinates into the same warehouse plane coordinate system and connects nodes according to the directed edges in the scene topology. Vehicle nodes are mapped to vehicle location points, warehouse access control nodes are mapped to warehouse entrance or exit location points, smart shelf nodes are mapped to shelf location points, and goods nodes are mapped to shelf occupancy locations or vehicle loading locations. The executing entity arranges nodes according to the target timestamp and generates flow direction lines between adjacent nodes, thereby forming a realistic scene that can represent changes in vehicle location, access control status, and shelf load-bearing capacity.

[0199] Furthermore, the reconstructed real-world scenario includes waybill identifier, cargo identifier, risk node, risk node spatial coordinates, risk node target timestamp, preceding node, following node, trajectory deviation angle, deviation angle threshold, dwell deviation, and node risk count. This real-world scenario is used to represent the actual flow status of waybills that deviate from the baseline path in physical space.

[0200] Example 9: S8 includes obtaining the video stream corresponding to the risk node, extracting the action trajectory of the physical handover action from the video stream; retrieving the accounting records based on the action trajectory, wherein the accounting records contain record details; if the flow characteristics of the record details are inconsistent with the action trajectory, obtaining the difference characteristics; performing classification mapping based on the difference characteristics, and determining whether the accounting records match the physical handover action.

[0201] In this embodiment, after identifying a risk node, the executing entity reads the node identifier, node type, target timestamp, spatial coordinates, waybill identifier, and cargo identifier corresponding to the risk node, and queries the camera device binding table based on the node identifier.

[0202] Specifically, the camera binding table records the binding relationships between camera equipment and warehouse access control, smart shelves, loading and unloading platforms, and vehicle parking areas. When the risk node is a warehouse access control node, the executing entity retrieves the video stream corresponding to the access control entrance or exit; when the risk node is a smart shelf node, it retrieves the video stream covering the aisle of that location; when the risk node is a truck terminal node, it retrieves the video stream from the loading and unloading platform or vehicle parking area. The video stream capture time is centered on the target timestamp of the risk node, capturing 5 minutes forward and 5 minutes forward to form the video segment to be identified.

[0203] The first 5 minutes are determined by adding 2 minutes of work preparation time, 2 minutes of equipment writing delay, and 1 minute of allowable error for document registration; the last 5 minutes are determined according to the same composition and are used to cover the start and end of the physical handover process.

[0204] Furthermore, when the executing entity extracts the motion trajectory of the video stream, it first splits the video segment into consecutive image frames.

[0205] When the video frame rate is 25 frames per second, the execution subject extracts 1 frame every 5 frames as a detection frame, so that 5 detection images are formed per second; when the video frame rate is 30 frames per second, 1 frame is extracted every 6 frames as a detection frame; at other frame rates, the execution subject divides the actual video frame rate by 5, and takes the integer frame interval corresponding to the obtained value as the frame extraction interval. Each detection frame inherits the original video timestamp and writes the frame sequence number and frame time.

[0206] Subsequently, the executing entity performs target identification on vehicles, personnel, pallets, containers, and forklifts in the detection frame to obtain the target category and the target bounding rectangle.

[0207] Specifically, the circumscribed rectangle includes the pixel coordinates of the top left corner, the pixel coordinates of the bottom right corner, the center point of the rectangle, and the center point of the bottom edge. When performing target association between adjacent detection frames, the executing entity first calculates the pixel distance between the center point of the target rectangle in the previous detection frame and the center point of the candidate target rectangle in the next detection frame.

[0208] The pixel distance calculation process is as follows: First, calculate the difference between the two center points in the horizontal pixel coordinates, and then calculate the difference between the two center points in the vertical pixel coordinates; multiply the horizontal difference by itself, multiply the vertical difference by itself, add the two multiplication results, and then take the square root of the sum to obtain the pixel distance between the two frames. When the target categories of the two consecutive frames are the same and the pixel distance does not exceed 60 pixels, the executing entity will merge the candidate target in the subsequent detection frame into the same target trajectory corresponding to the previous detection frame. The 60 pixels are determined by adding 40 pixels of the image displacement of the manually moved object within 1 second, 10 pixels of detection box jitter compensation, and 10 pixels of video compression error compensation.

[0209] In this embodiment, in order to make the motion trajectory in the video correspond to the actual work location, the executing entity uses the calibration parameters of the camera device to convert the pixel coordinates into warehouse planar coordinates.

[0210] Specifically, the calibration parameters are determined by four fixed calibration points within the image. These four fixed calibration points correspond to four metric coordinates measured on the warehouse floor. The executing entity establishes a perspective transformation relationship based on four sets of pixel coordinates and four sets of ground coordinates, and converts the center points of the base edges of the circumscribed rectangles of vehicles, personnel, pallets, boxes, and forklifts into horizontal and vertical coordinates in the warehouse planar coordinate system. The planar coordinates of each target in the continuous detection frames are arranged according to the frame time to form a motion trajectory. The motion trajectory includes at least the target category, trajectory start point, trajectory end point, traversed area, motion start time, motion end time, and direction of movement.

[0211] Furthermore, the executing entity identifies the physical handover actions based on the action trajectory.

[0212] In the access control scenario, the executing entity reads the position of the access control line segment in the warehouse's planar coordinate system. When the target trajectory crosses the access control line from the outer area of ​​the warehouse and enters the inner area, it is determined as an inbound action; when the target trajectory crosses the access control line from the inner area of ​​the warehouse and enters the outer area, it is determined as an outbound action. In the shelving scenario, the executing entity reads the planar coordinate range of the storage location area. When the trajectory endpoint of the box or pallet falls into the storage location area, and the smart shelf has a record of increased load within the corresponding time period, it is determined as a shelving action; when the trajectory starting point of the box or pallet is located in the storage location area, and the smart shelf has a record of decreased load within the corresponding time period, it is determined as a de-shelving action. In the loading and unloading scenario, when the cargo trajectory moves from the vehicle loading area to the warehouse area, it is determined as an unloading action; when the cargo trajectory moves from the warehouse area to the vehicle loading area, it is determined as a loading action.

[0213] Furthermore, when the executing entity retrieves accounting records based on the action trajectory, it uses the waybill identifier and cargo identifier as the main search fields to extract detailed records from waybill records, inbound slips, outbound slips, signed receipts, inventory ledgers, and financing application materials.

[0214] Specifically, the record details include document number, waybill identifier, cargo identifier, type of action in the books, flow in the books, location in the books, time in the books, quantity of goods, weight of goods, handover entity and vehicle identifier. The executing entity also performs time matching based on the start time and end time of the action. The time matching calculation process is as follows: convert the start time of the action, the end time of the action, the start time of the book and the end time of the book into the number of seconds under the same base time; when the start time of the action is not later than the end time of the book and the end time of the action is not earlier than the start time of the book, it is determined that the action trajectory and the record details have time overlap.

[0215] Furthermore, after obtaining the record details, the executing entity extracts the flow characteristics from the record details and compares them with the action trajectory.

[0216] Specifically, when the book action type is "inbound," the book flow should correspond to the inbound or shelving action obtained from video recognition; when the book action type is "outbound," the book flow should correspond to the outbound or de-shelving action; when the book action type is "loading," the book flow should correspond to the goods moving from the warehouse area to the vehicle loading area; when the book action type is "unloading," the book flow should correspond to the goods moving from the vehicle loading area to the warehouse area. The executing entity will determine the consistency between the direction type corresponding to the book flow and the action direction obtained from the action trajectory recognition; when the two correspond to the same direction, it is recorded as consistent flow; when the two correspond to opposite directions or have no corresponding relationship, it is recorded as inconsistent flow.

[0217] Furthermore, when the flow characteristics of the recorded details are inconsistent with the action trajectory, the executing entity obtains the difference characteristics.

[0218] Specifically, the discrepancies include flow direction discrepancies, time discrepancies, location discrepancies, weight discrepancies, subject discrepancies, and missing actions. Flow direction discrepancies are obtained by comparing the book flow direction with the action direction. Time discrepancies are obtained by comparing the book time with the action time. The executing entity converts the book time and the action start time into seconds, subtracts them, and takes the absolute value of the difference. When this difference exceeds 5 minutes, it is recorded as a time discrepancy. The 5 minutes are determined by adding 2 minutes of video capture pre-capture time, 2 minutes of device write delay, and 1 minute of allowable error for document registration. Location discrepancies are obtained by comparing the book location coordinates with the action trajectory endpoint coordinates. The executing entity calculates the planar distance between the two coordinates. When the planar distance in the access control or loading / unloading area exceeds 50 meters, it is recorded as a location discrepancy. When the planar distance in the shelf area exceeds 10 meters, it is recorded as a location discrepancy. The 50 meters are determined by adding 30 meters of access control entrance recognition radius, 15 meters of positioning error, and 5 meters of road buffer distance. The 10 meters are determined by adding 5 meters of warehouse area width, 3 meters of camera calibration error, and 2 meters of goods placement offset.

[0219] As a preferred embodiment, the weight difference is obtained by comparing the book weight of the goods with the load change value.

[0220] Specifically, the executing entity reads the weight of the goods in the record details and reads the load-bearing change value of the smart shelf within the time period associated with the action trajectory; the absolute value of the load-bearing change value is subtracted from the book weight of the goods, and the absolute value of the difference is taken; when the book weight of the goods does not exceed 100 kg and the difference exceeds 2 kg, it is recorded as a weight difference; when the book weight of the goods exceeds 100 kg and the difference exceeds 5 kg, it is recorded as a weight difference; the subject difference is obtained through comparison with the handover subject. The executing entity compares the personnel identification, vehicle identification or access control subject identification obtained by video recognition with the handover subject in the record details; when the identification is inconsistent, it is recorded as a subject difference.

[0221] Among them, missing actions are obtained by comparing the actions in the books with the actions in the video. When the books record that there are actions of entering, leaving, loading or unloading, but the corresponding action trajectory is not extracted from the video stream, it is recorded as missing actions.

[0222] Specifically, the executing entity is classified and mapped according to the difference characteristics. Flow direction difference is mapped to flow direction conflict class, time difference is mapped to time conflict class, location difference is mapped to location conflict class, weight difference is mapped to quantity weight conflict class, entity difference is mapped to entity conflict class, and action missing is mapped to action non-existent class.

[0223] For each type of difference, the executing entity writes a difference count for that record detail; when there are two types of differences for the same record detail, the difference count is 2, and so on; when the difference count is 0, the executing entity determines that the accounting record matches the physical handover action; when the difference count is not less than 1, the executing entity determines that the accounting record does not match the physical handover action, and outputs the mismatch category, the corresponding document number, the action trajectory number, the risk node identifier, and the difference field.

[0224] Example 10: S9 includes obtaining the receipt time and transport vehicle identifier from the accounting records; extracting spatial coordinates based on the transport vehicle identifier to obtain a first spatiotemporal dataset; using the first spatiotemporal dataset to retrieve and identify video frames to obtain a second spatiotemporal dataset containing physical handover actions; comparing the cargo weight characteristics in the second spatiotemporal dataset with the cargo weight in the accounting records to determine the weight difference value; if the weight difference value is greater than a preset difference threshold, extracting the time difference value between the receipt time and the time of the physical handover action; if the time difference value exceeds a preset time threshold, determining that the accounting records do not match the physical handover actions in the real scenario, and outputting a document forgery alarm for industrial digital financial risk analysis in the logistics industry.

[0225] In this embodiment, when the executing entity enters the document fraud alarm output stage, it first reads the receipt time, transport carrier identifier, waybill identifier, cargo identifier, document number, cargo weight, and receipt location from the accounting records.

[0226] Specifically, the transport carrier identifier is any one of the following: vehicle license plate number, vehicle terminal number, electronic seal number, or transport container number. The executing entity uses the waybill identifier as the main index and the transport carrier identifier as the positioning index to retrieve the spatial coordinate record corresponding to the transport carrier from the truck terminal trajectory table, warehouse access control table, and loading and unloading area equipment table.

[0227] Furthermore, when extracting spatial coordinates, the executing entity converts the receipt time into the number of seconds calculated from the same reference time, and reads the transport vehicle trajectory record from 30 minutes before the receipt time to 30 minutes after the receipt time. This 30 minutes is determined by adding the vehicle arrival buffer time of 10 minutes, the loading and unloading operation time of 15 minutes, and the equipment writing delay of 5 minutes.

[0228] For each trajectory record, the executing entity converts the trajectory acquisition time into seconds at the same reference time, then subtracts the receipt time from the number of seconds corresponding to the trajectory acquisition time, and takes the absolute value of the difference to obtain the time distance of the trajectory record relative to the receipt time. Trajectory records with a time distance of no more than 30 minutes are written into the candidate trajectory set. When there are more than two records in the candidate trajectory set, the record with the smallest time distance value is selected as the receipt-related trajectory record. The longitude, latitude, acquisition time, vehicle status, and positioning status in the receipt-related trajectory record, together with the receipt time, receipt location coordinates, waybill identifier, cargo identifier, and transport carrier identifier, constitute the first spatiotemporal dataset.

[0229] In this embodiment, the first spatiotemporal dataset is used to retrieve video frame images. The executing entity queries the camera equipment coverage area table based on the spatial coordinates in the signed-off trajectory record. The camera equipment coverage area table records the camera identifier, installation location, coverage boundary coordinates, bound work area, and corresponding risk node.

[0230] Specifically, when the spatial coordinates of the transport vehicle fall within the coverage boundary of a certain camera device, the executing entity identifies that camera device as the target camera device. When the same spatial coordinates fall within the coverage boundaries of two or more camera devices, the executing entity calculates the planar distance between the installation position of each camera device and the spatial coordinates of the transport vehicle, and selects the camera device with the smallest distance value. The calculation process of the planar distance is as follows: read the horizontal and vertical coordinates of the installation position of the camera device, read the horizontal and vertical coordinates of the position of the transport vehicle, calculate the difference between the horizontal coordinates and the difference between the vertical coordinates respectively, multiply the two differences by themselves, add them together, and then take the square root of the sum to obtain the distance value.

[0231] Furthermore, once the target camera device is identified, the executing entity captures the video stream centered on the signing time, capturing 5 minutes forward and 5 minutes backward to form a video frame image to be identified.

[0232] Specifically, the 10 minutes is determined by adding 2 minutes of work preparation time, 3 minutes of physical handover action duration, 2 minutes of loading and unloading confirmation time, and 3 minutes of equipment write delay. When the video stream is 25 frames per second, the executing entity extracts 1 frame every 5 frames as an identification frame; when the video stream is 30 frames per second, 1 frame is extracted every 6 frames as an identification frame; at other frame rates, the executing entity divides the actual frame rate by 5 to obtain the frame extraction interval, and generates an identification frame sequence of 5 frames per second according to the frame extraction interval. Each identification frame is written with a frame number, frame time, camera identifier, and corresponding waybill identifier.

[0233] In this embodiment, the executing entity performs target recognition on the recognition frame. The recognition objects include vehicles, personnel, pallets, boxes, forklifts, and cargo labels. For each recognition object, the executing entity records the target category, the target's circumscribed rectangle, the rectangle's center point, the bottom edge's center point, and the detection time.

[0234] In this system, targets in adjacent recognition frames are associated based on their target category and the pixel distance between their center points. The pixel distance is calculated as follows: the difference in horizontal pixels between the center point of the target in the previous frame and the center point of the candidate target in the next frame is calculated, and the difference in vertical pixels is also calculated. The horizontal and vertical differences are multiplied by themselves and then added together. The square root of the sum is then taken to obtain the pixel distance. When the target categories are consistent and the pixel distance does not exceed 60 pixels, the candidate target in the next frame is merged into the same target trajectory. The 60 pixels are determined by adding 40 pixels for the displacement of the manual handling action, 10 pixels for detection box jitter compensation, and 10 pixels for video compression error compensation.

[0235] As a preferred embodiment, in order to make the video recognition results correspond to the real scene, the executing entity converts the pixel coordinates into the planar coordinates of the work area through the camera equipment calibration parameters. The calibration parameters are determined by four fixed calibration points in the video frame and four measured metric coordinates on the ground of the work area.

[0236] Specifically, the executing entity establishes a perspective transformation relationship based on four sets of pixel coordinates and four sets of metric coordinates, and converts the center points of the base edges of the circumscribed rectangles of the cargo box, pallet, forklift, personnel, and vehicle into horizontal and vertical coordinates. After arranging the planar coordinates in the continuous recognition frames according to the frame time, a physical handover action trajectory is formed. The executing entity determines the physical handover action based on the trajectory start point, trajectory end point, traversed area, and target category. When goods move from the vehicle loading area to the warehouse area, it is identified as an unloading action; when goods move from the warehouse area to the vehicle loading area, it is identified as a loading action; when goods enter the storage area and the corresponding load record increases, it is identified as a shelving action; when goods leave the storage area and the corresponding load record decreases, it is identified as a deslaving action. The recognition results, together with the action time, action space coordinates, cargo label recognition results, cargo quantity recognition results, and load change values, constitute the second spatiotemporal dataset.

[0237] Furthermore, the executing entity then compared the cargo weight characteristics in the second spatiotemporal dataset with the cargo weight in the accounting records.

[0238] Specifically, the weight characteristics of the goods are obtained in a predetermined order: if there is a change in the load capacity of the smart shelf within the time period of the action, the absolute value of the change in load capacity is taken; if there is no change in the load capacity of the smart shelf but there is a weighing value on the platform weighbridge, the absolute value of the difference between the current weighing value and the previous weighing value on the platform weighbridge is taken; if neither of the above two conditions exists and the video identifies the goods label and the quantity of goods, the weight of a single item in the goods master data is read, and the weight of a single item is multiplied by the quantity of identified goods to obtain the weight identified by the video; the executing entity subtracts the book weight of the goods from the weight characteristics of the goods and takes the absolute value of the difference to obtain the weight difference value.

[0239] Specifically, the difference threshold is determined based on the book weight of the goods, the error of the weighing equipment, and the packaging error.

[0240] Specifically, when the book weight of the goods does not exceed 100 kg, the difference threshold is 2 kg, which is obtained by adding 1 kg of weighing sensor error and 1 kg of packaging attachment error; when the book weight of the goods exceeds 100 kg but does not exceed 1000 kg, the difference threshold is 5 kg, which is obtained by adding 3 kg of weighing sensor error and 2 kg of pallet packaging error; when the book weight of the goods exceeds 1000 kg, the difference threshold is 10 kg, which is obtained by adding 6 kg of platform weighbridge error, 2 kg of vehicle parking offset error, and 2 kg of packaging attachment error; the executing entity calls the corresponding difference threshold according to the range of the book weight of the goods and compares the weight difference value with the difference threshold.

[0241] Furthermore, when the weight difference value is greater than the corresponding difference threshold, the executing entity extracts the time difference value between the signing time and the physical handover action time. The physical handover action time is determined by the action trajectory in the second spatiotemporal dataset.

[0242] Specifically, for loading and unloading actions, the executing entity takes the frame time when the cargo trajectory first crosses the boundary of the vehicle loading area as the physical handover action time; for shelving and unshelving actions, the executing entity takes the frame time when the cargo trajectory first crosses the boundary of the storage area as the physical handover action time; the executing entity converts both the receipt time and the physical handover action time into seconds at the same reference time, subtracts the physical handover action time from the receipt time in seconds, and takes the absolute value of the difference to obtain the time difference value.

[0243] In this embodiment, the time threshold is determined according to the type of document recorded in the books.

[0244] Specifically, the time threshold for a signed receipt is 5 minutes, calculated by adding the document registration allowance of 2 minutes, the video recognition frame interval error of 1 second, the device write delay of 2 minutes, and the manual confirmation time of 59 seconds; the time threshold for an inbound confirmation is 6 minutes, calculated by adding the inbound verification time of 3 minutes, the video recognition frame interval error of 1 second, the device write delay of 2 minutes, and the manual confirmation time of 59 seconds; the time threshold for an outbound confirmation is 6 minutes, calculated by adding the outbound verification time of 3 minutes, the video recognition frame interval error of 1 second, the device write delay of 2 minutes, and the manual confirmation time of 59 seconds; the executing entity reads the corresponding time threshold according to the document type recorded in the ledger and compares the time difference value with the time threshold.

[0245] Furthermore, when the weight difference value exceeds the difference threshold and the time difference value exceeds the time threshold, the executing entity determines that the book record does not match the physical handover action in the actual scenario; this determination result is written into the risk analysis record, which includes waybill identifier, cargo identifier, transport carrier identifier, document number, book receipt time, physical handover action occurrence time, time difference value, book cargo weight, cargo weight characteristics, weight difference value, difference threshold, time threshold, target camera device identifier, and risk node identifier.

[0246] Furthermore, the implementing entity generates a document fraud alert based on the risk analysis records.

[0247] Specifically, the document forgery alert includes the alert number, waybill identifier, cargo identifier, document number, transport carrier identifier, anomaly type, weight difference value, time difference value, associated video frame number, associated IoT device node identifier, and alert generation time. The anomaly type is recorded as inconsistent weight and inconsistent time; this document forgery alert is included in the logistics industry's digital financial risk analysis results.

[0248] Example 11: Figure 2 This is a structural block diagram of the local terminal of an exemplary electronic device (machine) of the present invention; as shown... Figure 2 As shown, the electronic device of the present invention includes a processor 11, a memory 12, a storage space 13 for storing program code, and program code 14 for executing the method steps according to the present invention. The program code 14 for executing the method steps according to the present invention is used to execute the above-described control logic.

[0249] Figure 3 This is a structural block diagram of the network end of an exemplary electronic device of the present invention; as shown below. Figure 3As shown, the present invention also provides an electronic device (machine), which may include at least one processor 210, at least one memory 230 communicatively connected to the processor, and a communication bus 240 and a communication interface 220 connecting different system components (including the memory 230 and the processor 210). The processor 210, the memory 230 and the communication interface 220 are connected through the communication bus 240 and communicate with each other. The communication interface 220 is used for data interaction with external devices. The memory 230 stores a machine-executable program that can be executed by the processor, and the processor 210 can execute the above-mentioned control logic by calling the machine-executable program.

[0250] Communication bus 240 represents one or more of several bus architectures, including a memory bus or memory controller, peripheral bus, graphics acceleration port, processor, or local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnection (PCI) bus.

[0251] Electronic devices typically include a variety of computer system readable media, which can be any available media that can be accessed by the electronic device, including volatile and non-volatile media, and removable and non-removable media.

[0252] Memory 230 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The electronic device may further include other removable / non-removable, volatile / non-volatile computer system storage media. Memory 230 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the control logic described above.

[0253] A program / utility having a set (at least one) of program modules can be stored in memory 230. Such program modules include, but are not limited to, an operating system, one or more applications, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment.

[0254] Machine-executable programs for performing this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages ​​such as Java, C++, and Python, and may also include specialized engineering languages ​​such as R. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0255] The present invention also discloses a storage medium on which a machine-executable program as described above is stored.

[0256] The aforementioned storage medium may be any combination of one or more computer-readable media. Computer-readable media may be, for example, computer-readable signal media or computer-readable storage media. Computer-readable storage media include, but are not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium may be, for example, any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0257] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0258] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0259] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access, characterized in that, include: S1. Obtain the trajectory coordinates of the truck terminal, the access control records of the warehouse access control, and the load-bearing changes of the smart shelf as initial equipment data, and extract the action time corresponding to the underlying physical access behavior from the initial equipment data to obtain the first time-series data arranged according to the action time. S2. For the first time series data, construct an IoT access spatiotemporal graph and extract cross-node collaboration features to obtain second time series data containing physical handover features and spatial location features. S3. Encrypt and solidify the second time-series data to generate a digital certificate with a unique identifier that cannot be tampered with. S4. Based on the digital certificate, extract the timestamp and spatial coordinates encapsulated within the digital certificate to determine the time alignment relationship between different IoT device nodes. S5. If the time alignment relationship meets the preset threshold, then the fragmented underlying physical access behaviors are spliced ​​together according to the time alignment relationship to construct an interaction sequence diagram that reflects the actual flow status of goods. S6. Extract the flow path of goods in the physical world from the interaction sequence diagram, perform anomaly detection on the flow path, and determine whether the flow path deviates from the pre-established baseline path. S7. If the transfer path deviates from the pre-established baseline path, an anomaly mark is generated for the waybill, and the real scenario containing risk nodes is reconstructed based on the anomaly mark and the interaction sequence diagram. S8. Obtain the accounting records corresponding to the risk nodes in the real scenario, and determine whether the accounting records match the physical handover actions in the real scenario; S9. If the accounting records do not match the physical handover actions in the real scenario, a document fraud alarm will be output for use in the digital financial risk analysis of the logistics industry.

2. The method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access as described in claim 1, characterized in that: S1 includes: The trajectory coordinates, access control records, and load changes of the truck terminal are obtained as initial equipment data; Spatial displacement features, identity verification identifiers, and load fluctuation values ​​are extracted from the initial equipment data to obtain a multidimensional physical state set. If the set of multidimensional physical states undergoes a sudden change, the underlying physical access behavior is obtained; Extract the behavior trigger nodes corresponding to the underlying physical access behavior and determine the action time; The underlying physical access behaviors are sorted according to the action time to obtain the first time-series data arranged by action time.

3. The method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access as described in claim 1, characterized in that: S2 includes: Extract the access trajectory and node distance from the first time-series data to construct an IoT access spatiotemporal graph; A causal graph attention network is used to process the adjacency matrix and state sequence of the spatiotemporal graph of IoT access to obtain causal weights; Cross-node collaborative features are extracted by weighted summation of state sequences using causal weights, and the aggregation vector corresponding to the cross-node collaborative features is determined. Spatiotemporal mapping is performed on the aggregated vector to obtain second time-series data containing physical handover features and spatial location features.

4. The method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access as described in claim 1, characterized in that: S3 includes: Acquire second time-series data containing physical handover features and spatial location features; The second time series data is processed by feature segmentation to obtain time series data blocks; A hash algorithm is used to extract the digest value of the time-series data block; Determine whether the length of the data digest value meets the preset threshold; If the length of the data digest value meets the preset threshold, then the data digest value is encrypted and solidified to determine the solidified ciphertext; Add timestamps and node identifiers to the solidified encrypted text to perform credential signing, generating a digital credential with a unique identifier that cannot be tampered with.

5. The method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access as described in claim 1, characterized in that: S4 includes: Obtain the digital certificate and extract the original timestamp and original spatial coordinates encapsulated in the digital certificate; The initial alignment deviation is calculated based on the original timestamp and the original spatial coordinates, and the target alignment deviation is obtained. Calculate the time offset based on the target alignment deviation; If the time offset is greater than the preset offset threshold, a target compensation value is generated. The target timestamp is obtained by correcting the original timestamp with the target compensation value, and the time alignment relationship between IoT device nodes is determined based on the target timestamp.

6. The method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access according to claim 1, characterized in that: S5 includes: Obtain the underlying physical access records, and extract the access timestamps from the underlying physical access records to generate discrete access fragments; Time-aligned features are obtained by matching discrete access segments. Determine whether the time alignment feature meets the preset threshold; If the time alignment feature meets the preset threshold, the discrete access segments are fused to obtain the behavior splicing sequence. Construct a temporal correlation matrix based on the behavior splicing sequence; An interactive time sequence diagram reflecting the actual flow of goods is constructed based on the time sequence correlation matrix.

7. The method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access as described in claim 1, characterized in that: S6 includes: Obtain the interaction sequence diagram, and extract the flow nodes from the interaction sequence diagram to construct the flow path; The isolated forest algorithm is used to process the flow path to obtain the path feature vector, and the anomaly score is calculated based on the path feature vector. If the abnormal score is greater than the preset threshold, the degree of deviation is extracted from the flow path; The deviation is compared with the pre-established baseline path to determine whether the transfer path deviates from the pre-established baseline path.

8. The method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access according to claim 1, characterized in that: S7 includes: Spatial coordinates and dwell time are analyzed from waybills that deviate from the baseline path to calculate the trajectory deviation angle; If the trajectory deviation angle is greater than the preset deviation angle threshold, an abnormal marker is generated for the waybill. Extract node states from the interaction sequence diagram based on anomaly markers, perform feature mapping on node states to obtain risk nodes, and construct scene topology; Based on the scene topology, the interaction sequence diagram is spatially mapped to reconstruct a real scene containing risk nodes.

9. The method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access according to claim 1, characterized in that: S8 includes: Obtain the video stream corresponding to the risk node, and extract the motion trajectory of the physical handover action from the video stream; The accounting records are retrieved based on the action trajectory, and the accounting records include record details; If the flow characteristics of the recorded details are inconsistent with the action trajectory, then the difference characteristics are obtained; Classification and mapping are performed based on the differences in characteristics to determine whether the accounting records match the physical handover actions.

10. The method for analyzing industrial digital financial risks in the logistics industry based on Internet of Things access according to claim 1, characterized in that: S9 includes: Obtain the receipt time and transport vehicle identifier from the accounting records, extract spatial coordinates based on the transport vehicle identifier to obtain a first spatiotemporal dataset, and use the first spatiotemporal dataset to retrieve and identify video frames to obtain a second spatiotemporal dataset containing physical handover actions. Compare the cargo weight features in the second spatiotemporal dataset with the cargo weight in the accounting records to determine the weight difference value; If the weight difference value is greater than the preset difference threshold, the time difference between the signing time and the physical handover time will be extracted. If the time difference exceeds the preset time threshold, it is determined that the accounting records do not match the physical handover actions in the real scenario, and a document fraud alarm is output for the analysis of industrial digital finance risks in the logistics industry.

Images