Cross-border transfer business processing method and system based on dynamic port knowledge graph

Abstract

The application provides a cross-border transfer business processing method and system based on a dynamic port knowledge graph, comprising the following steps: S1, obtaining structured and unstructured information from multiple sources of data and constructing a dynamic port knowledge graph containing several entity types and relationship types; S2, receiving task information of goods to be transferred and performing standardized processing to generate a goods task feature vector and a task constraint vector; S3, under the constraint of the dynamic port knowledge graph, running a multi-agent collaborative decision engine to search for multiple candidate transfer paths that meet the access conditions according to the goods task feature vector and the task constraint vector, and sequentially performing customs clearance mode matching, declaration element matching, risk matrix matching, and multi-objective comprehensive scoring processing to obtain a candidate scheme ranking result, so as to guide a user to process cross-border transfer business. The beneficial effect is that the application can effectively improve the automation level, decision accuracy, and risk resistance of cross-border transfer business.

Description

Technical Field

[0001] This invention relates to the technical field of cross-border business processing, and more specifically, to a method and system for cross-border transit business processing based on a dynamic port knowledge graph. Background Technology

[0002] With the continued growth of cross-border e-commerce, international supply chain collaboration, and third-country transit business, cross-border goods often need to undergo continuous declaration and transshipment processing under multiple ports, modes of transport, and regulatory systems overseas. Especially in scenarios such as rail transit, road transit, air freight, and bonded warehouse transshipment, the same batch of goods may face significantly different access conditions, licensing requirements, labeling rules, inspection intensity, and time limits in different countries and ports. How to efficiently and accurately generate cross-border transit declaration schemes that match the actual regulatory requirements of each port has become a pressing technical challenge in the field of cross-border logistics and customs services.

[0003] In existing technologies, relevant solutions can be broadly categorized into the following types: The first type is a scheme that classifies HS codes based on textual information such as product name, purpose, and material. This type of technology typically uses the product description as the main input and outputs a predicted tariff code through classification models or knowledge graph matching. However, this type of scheme lacks the capability to integrate the entry rules, licensing requirements, rules of origin restrictions, sensitive category regulations, and unforeseen risk events of different ports into a machine-processable structure, resulting in a disconnect between the classification results and the actual regulatory requirements of the target port.

[0004] The second category is solutions based on logistics nodes and transportation costs for cross-border route planning. These solutions focus on transportation scheduling or route optimization, but rarely consider legal and regulatory factors such as customs supervision rules, compliance of declaration elements, and differences in port policies. The lack of linkage between route selection and subsequent declaration compliance requirements easily leads to the predicament of "optimal cost but inadequate permits and licenses."

[0005] The third type is schemes that score risk at a single point based on historical inspection records or declaration behaviors. These schemes typically output a risk score only for a single declaration behavior, lacking structured collaboration with declaration scheme generation, compliance verification, and replanning during the implementation period. Risk prediction results are often displayed independently and fail to serve as input for scheme selection and dynamic adjustment.

[0006] The fourth category is decision-making support solutions based on knowledge graphs. Most existing knowledge graph technologies remain at the level of knowledge retrieval, display of regulatory relationships, or rule association analysis. Knowledge graphs are only used as information query tools, rather than as the underlying decision-making foundation for constraint propagation, solution generation, and risk correction.

[0007] In summary, while existing solutions have some value in specific tasks, they still have the following shortcomings in the complex scenario of cross-border transit, which involves multiple ports, rules, and time constraints: First, dynamic port regulatory conditions are difficult to model uniformly and participate in automated decision-making. The access rules, licensing requirements, inspection intensity, and policy events of different ports are scattered across heterogeneous data sources, lacking a unified structured expression and continuous update mechanism, which makes it impossible for the decision-making process to perceive changes in port regulatory status in real time.

[0008] Secondly, route selection, customs clearance mode, declaration element classification, and risk assessment are fragmented. The existing system treats these aspects as independent modules, lacking the ability to optimize them collaboratively within a unified decision-making framework. Furthermore, the lack of data feedback and constraint propagation mechanisms between modules makes it difficult to promptly identify and resolve decision-making conflicts.

[0009] Third, the system lacks the ability to respond to and adapt to policy changes and emergencies in real time. During the transit of goods, if there are changes in port policies, a sudden increase in inspection intensity, or updates to license rules, the existing system can usually only issue passive alarms and cannot automatically identify the scope of impact or replan the unexecuted links locally or globally.

[0010] Therefore, there is a need for an integrated technical solution that can dynamically and structurally represent the regulatory rules of multiple ports and drive the automatic generation of declaration schemes, risk prediction, and dynamic reconstruction, so as to improve the automation level, decision-making accuracy, and risk resistance of cross-border transit business. Summary of the Invention

[0011] The technical problem this invention aims to solve is how to effectively improve the automation level, decision-making accuracy, and risk resistance of cross-border transit business. To overcome the shortcomings of the existing technology (or related technology), this invention provides a cross-border transit business processing method and system based on dynamic port knowledge graph.

[0012] This invention provides a method for processing cross-border transit business based on a dynamic port knowledge graph, comprising the following steps: Step S1: Obtain structured and unstructured information from multi-source data and construct a dynamic port knowledge graph containing several entity types and relationship types; Step S2: Receive the task information of the goods to be transferred, and perform standardized processing on the task information to generate a cargo task feature vector and a task constraint vector; Step S3: Under the constraints of the dynamic port knowledge graph, a multi-agent collaborative decision engine is run to search for multiple candidate transit routes that meet the admission conditions based on the cargo task feature vector and the task constraint vector. Each candidate transit route is then processed sequentially with customs clearance mode matching, declaration element matching, risk matrix matching, and multi-objective comprehensive scoring to obtain a candidate scheme ranking result containing at least one recommended scheme, so as to guide the user in cross-border transit business processing.

[0013] Compared with existing technologies, the cross-border transit business processing method based on dynamic port knowledge graph of the present invention has the following advantages: This invention constructs a dynamic port knowledge graph to unify multi-source data into machine-executable structured constraints. This dynamic port knowledge graph serves as the common constraint base for recommending solutions, ensuring that candidate transit routes are constrained by real regulatory rules from the generation stage. This avoids the problem of classification results and route selection being disconnected from actual port requirements. Furthermore, through the pipelined processing of a multi-agent collaborative decision-making engine, path search, pattern matching, declaration element matching, and risk matrix matching are performed sequentially. This ensures that each decision-making stage is executed within the same regulatory semantic space and converges uniformly through multi-objective comprehensive scoring, effectively overcoming the decision-making conflicts caused by independent serial computation of modules in traditional solutions. Simultaneously, the entire method is driven by standardized task vectors, graph constraints, and quantitative scoring models, reducing the involvement of manual rule pre-setting and subjective judgment. The candidate solution generation process has complete scoring records and ranking criteria, facilitating user traceability and understanding of the recommendation logic. This effectively improves the automation level, decision-making accuracy, and risk resistance of cross-border transit operations.

[0014] In one possible implementation, the entity types in step S1 include port entities, country and region entities, mode of transport entities, customs clearance mode entities, HS code entities, regulatory condition entities, license entities, risk event entities, and cargo category entities; the relationship types include applicable, prohibited, required submission, associated risk, permitted transit, applicable mode of transport, applicable time window, and impact on subsequent nodes, and each entity type or relationship type is bound to attribute fields including effective time, expiration time, source identifier, and credibility score.

[0015] Compared with existing technologies, the above-mentioned technical solution can comprehensively cover the entire semantic association of "goods-port-rules-risk" in cross-border transit scenarios by limiting core entity types such as ports, regulatory conditions, licenses, risk events, and cargo categories, as well as relationship types such as applicable, prohibited, required submission, and associated risks. This provides a complete data foundation for subsequent path search, pattern matching, and risk prediction.

[0016] In one possible implementation, the task information in step S2 includes cargo name, attribute information, country of origin, place of departure, destination, declared value, logistics measurement information, special attribute identifiers, trade terms, acceptable mode of transport range, and enterprise-side target constraints; the task constraint vector includes maximum acceptable timeliness, maximum acceptable cost, countries or ports of entry that cannot be passed through, certification conditions that must be met, and enterprise-bearable risk thresholds.

[0017] Compared with existing technologies, the above-mentioned technical solution can extract standardized cargo task feature vectors from messy enterprise declaration data by limiting fields such as cargo name, attribute information, country of origin, place of departure, destination, declared value, special attribute identifiers, acceptable mode of transport range, and enterprise-side target constraints, thereby avoiding decision-making biases caused by missing or non-standard input fields.

[0018] In one possible implementation, the multi-agent collaborative decision-making engine includes a port screening and pattern matching agent, a declaration element classification agent, a compliance risk prediction agent, and a comprehensive evaluation agent. Step S3 includes: Step S31: The port screening and pattern matching agent searches for multiple candidate schemes that meet the admission criteria based on the place of departure, the destination, the acceptable mode of transport and the passable rules in the dynamic port knowledge graph. It also matches executable customs clearance mode combinations for the candidate transit paths of each candidate scheme and outputs mode information. Furthermore, it performs hard constraint pruning on the path nodes of each candidate transit path to generate path information. Step S32: Use the declaration element classification agent to extract the regulatory constraints corresponding to each path node from the dynamic port knowledge graph, and input the regulatory constraints and the cargo task feature vector into the final output declaration element scheme package of the classification reasoning model. Step S33: The compliance risk prediction agent is used to model the combination of goods-port-mode-time as nodes, and the temporal transition edges and path adjacency edges between nodes are constructed to obtain the risk sample graph structure, and the risk matrix is ​​output based on the risk sample graph structure. Step S34: The comprehensive evaluation agent performs multi-objective comprehensive scoring on the path information, pattern information, application element package and risk matrix of each candidate solution based on the task constraint vector, and filters to obtain a ranking result of candidate solutions containing at least one recommended solution.

[0019] Compared with existing technologies, the above-mentioned technical solution can form a complete decision chain from candidate route generation to multi-objective comprehensive scoring and ranking by sequentially linking port screening and pattern matching intelligent agents, declaration element classification intelligent agents, compliance risk prediction intelligent agents, and comprehensive evaluation intelligent agents. The output of each intelligent agent serves as the input of the next intelligent agent, ensuring the continuity of the decision-making logic. The declaration element classification intelligent agent does not output a single HS code for the goods themselves, but outputs a declaration element scheme package bound to the regulatory environment of each candidate transit route. This allows for clear identification and comparison of the differences in declaration requirements for the same batch of goods under different port routes, avoiding classification mismatch. At the same time, the comprehensive evaluation intelligent agent performs multi-objective scoring based on data from four dimensions: comprehensive route information, pattern information, declaration element scheme packages, and risk matrix. This ensures that the final recommended solution achieves a balance between cost, timeliness, compliance, and risk control.

[0020] In one possible implementation, the hard constraints in step S31 include at least whether the cargo category is allowed to transit, whether the current port is open for the clearance mode, whether there is an absolute lack of necessary licenses, whether there are embargoes or strong restrictions, and whether the time window is met.

[0021] Compared with existing technologies, the above-mentioned technical solution can quickly eliminate infeasible solutions during the route screening stage by combining judgments on hard constraints such as whether the cargo category allows transit, whether the port is open for the customs clearance mode, whether the licenses are absolutely missing, whether there are any embargo events, and whether the time window is met. This narrows down the candidate space for subsequent detailed evaluation and improves the overall decision-making efficiency.

[0022] In one possible implementation, the declaration element scheme package output in step S32 includes at least a set of HS candidate codes and the confidence level, declaration element list, missing field list, conflict field list with the target port regulatory conditions, and element completeness score corresponding to each HS candidate code. The classification reasoning model outputs the declaration element scheme package bound to the regulatory environment of each candidate transit path for different candidate transit paths.

[0023] Compared with existing technologies, the above technical solution can provide risk prediction agents and comprehensive evaluation agents with sufficient structured evaluation data by outputting HS candidate code sets and confidence levels, declaration element lists, missing field lists, conflict field lists and element completeness scores, thus avoiding risk omissions or scoring distortions due to insufficient information granularity.

[0024] In one possible implementation, in step S33, the risk matrix is ​​formed according to each node and the risk category corresponding to each node. The risk categories include HS classification questioning risk, missing declaration elements risk, missing license or certification risk, origin restriction risk, price anomaly risk, and policy change transmission risk. Each node includes cargo attribute characteristics, classification and placement characteristics, element completeness characteristics, current port supervision intensity characteristics, historical inspection event characteristics, certificate requirement satisfaction characteristics, and policy update time difference characteristics.

[0025] Compared with existing technologies, the above-mentioned technical solution can unify the risks of HS classification questioning, missing declaration elements, missing licenses or certifications, origin restriction, price anomaly, and policy change transmission into a risk matrix, thereby expanding risk assessment from a single-dimensional inspection probability to a multi-dimensional compliance risk profile and improving the comprehensiveness of risk identification.

[0026] In one possible implementation, after performing step S3, the method further includes: During the actual transit of goods, the system continuously receives update events from the dynamic port knowledge graph and dynamically monitors the recommended schemes selected by the user. When a triggering event is detected, the system freezes path nodes that have been declared or irreversibly executed during the execution of the recommended scheme, identifies the impact domain of the triggering event to determine the subsequent node subgraphs that need to be recalculated, and performs local or global replanning on the subsequent node subgraphs to output explanations of the differences between the old and new schemes, explanations of risk changes, and operational suggestions.

[0027] Compared with existing technologies, the above technical solution can continuously receive update events from the dynamic port knowledge graph and automatically perform impact domain identification and replanning when a triggering event is detected. This enables the system to perform dynamic protection throughout its entire lifecycle, significantly improving the continuity and risk resistance of cross-border transit business. Furthermore, by freezing executed path nodes and replanning only subsequent node subgraphs, the system avoids having to start over due to sudden triggering events, reducing the cost and time loss of replanning.

[0028] In one possible implementation, the triggering events include subsequent changes in port access status, changes in licensing rules, a significant increase in inspection intensity, the inclusion of sensitive HS codes in the key supervision list, changes in the feasibility of transportation methods, and recall events, prohibited or restricted transportation events, or security events related to cargo attributes.

[0029] Compared with existing technologies, the above-mentioned technical solution can accurately identify events that truly affect the feasibility of the solution by limiting the types of triggering events such as changes in port access status, changes in license rules, increased inspection intensity, inclusion of sensitive HS codes in the key supervision list, changes in transportation mode feasibility and recall events related to cargo attributes, prohibited and restricted transportation events, and security events. This avoids frequent unnecessary replanning due to irrelevant information disturbances.

[0030] This invention provides a cross-border transit business processing system based on a dynamic port knowledge graph, which applies the aforementioned cross-border transit business processing method based on a dynamic port knowledge graph, including: The data acquisition module is used to collect policy data, logistics data, declaration data, and risk event data, and perform standardized cleaning to obtain multi-source data; The knowledge graph construction module, connected to the data acquisition module, is used to obtain structured and unstructured information from the multi-source data and construct a dynamic port knowledge graph containing several entity types and relationship types. The task processing module is used to receive task information of goods to be transferred and to standardize the task information to generate a cargo task feature vector and a task constraint vector. The multi-agent collaborative decision-making engine, connected to the graph construction module and the task processing module respectively, is used to search for multiple candidate transit routes that meet the admission conditions under the constraints of the dynamic port knowledge graph, based on the cargo task feature vector and the task constraint vector. It then performs customs clearance mode matching, declaration element matching, risk matrix matching, and multi-objective comprehensive scoring on each candidate transit route to obtain a ranking result of candidate schemes containing at least one recommended scheme, so as to guide users in cross-border transit business processing.

[0031] Compared with existing technologies, the cross-border transit business processing system based on dynamic port knowledge graph of the present invention has the following advantages: This invention constructs a complete technical chain for data acquisition, knowledge graph construction, and intelligent decision-making through modular interconnection of a data acquisition module, a knowledge graph construction module, a task processing module, and a multi-agent collaborative decision-making engine. The responsibilities of each module are clear and the interfaces are well-defined, facilitating engineering deployment and iterative optimization. Furthermore, each module is connected with standardized interfaces. The data acquisition module can interface with policy databases and logistics systems from different sources, and the multi-agent collaborative decision-making engine can independently upgrade or replace the agent models within it, giving the system good scalability and adaptability to different business scenarios. Attached Figure Description

[0032] Figure 1 This is a flowchart of the method steps of the present invention; Figure 2This is a flowchart illustrating step S3 of the present invention. Figure 3 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0033] First, those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention. Those skilled in the art can make adjustments as needed to adapt to specific application scenarios.

[0034] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0035] See Figure 1 This invention discloses a method for processing cross-border transit business based on a dynamic port knowledge graph, including the following steps: Step S1: Obtain structured and unstructured information from multi-source data and construct a dynamic port knowledge graph containing several entity types and relationship types; Step S2: Receive the task information of the goods to be transferred, and standardize the task information to generate the goods task feature vector and task constraint vector. Step S3: Under the constraints of the dynamic port knowledge graph, the multi-agent collaborative decision engine is run to search for multiple candidate transit routes that meet the access conditions based on the cargo task feature vector and task constraint vector. Each candidate transit route is then processed sequentially with customs clearance mode matching, declaration element matching, risk matrix matching, and multi-objective comprehensive scoring to obtain a candidate scheme ranking result containing at least one recommended scheme, so as to guide users in cross-border transit business processing.

[0036] In this embodiment of the invention, step S1 includes at least the following multi-source data: customs announcements, regulatory interface data, port rules database, logistics execution records, historical declaration data, and risk event data. Unstructured text data such as customs announcements and regulatory interface data are parsed using an event extraction model in natural language processing to identify event information such as changes in port policies, upgraded inspections, changes in prohibited or restricted transport rules, changes in licensing rules, and tariff adjustments. Structured data such as the port rules database and logistics execution records are directly extracted and mapped into graph entities and relationships using ETL tools.

[0037] In this embodiment of the invention, the event extraction model adopts a sequence labeling architecture based on a pre-trained language model. The input is unstructured text, and the output is structured fields such as event type, event subject, event object, effective time, and scope of influence. The extracted event information is converted into graph operation instructions, including types such as adding entities, adding relationships, deleting relationships, and updating attributes, thereby realizing the continuous evolution of the dynamic port knowledge graph. Preferably, for conflict rules from multiple data sources, conflict resolution is carried out according to a combination strategy of "source credibility, publication time, and accuracy of applicable scope". When a conflict cannot be automatically determined, the corresponding rule is marked as pending review and its constraint weight in automatic decision-making is reduced.

[0038] In this embodiment of the invention, the entity types in step S1 include port entities, country and region entities, mode of transport entities, customs clearance mode entities, HS code entities, regulatory condition entities, certificate entities, risk event entities, and cargo category entities; the relationship types include applicable, prohibited, required submission, associated risk, permitted transit, applicable mode of transport, applicable time window, and impact on subsequent nodes, and each entity type or relationship type is bound to attribute fields including effective time, expiration time, source identifier, and credibility score; preferably, the storage of entity types and relationship types of the dynamic port knowledge graph is implemented using a graph database (such as Neo4j or JanusGraph), which supports multi-hop path query and subgraph matching operations to accelerate candidate path search and constraint propagation processes.

[0039] In this embodiment of the invention, step S2 includes the following standardization processes: product name normalization (mapping synonyms, abbreviations, and full names to standard product names), unit of measurement unification (e.g., unifying boxes, pieces, kilograms, etc. to preset standard units), brand and model field cleaning (removing special characters and unifying uppercase and lowercase formats), usage tag extraction (extracting keywords from usage descriptions and mapping them to a preset tag system), category mapping (mapping goods names to goods category entities in a knowledge graph), and missing field identification (marking required fields not provided in the task information). After the above processing, a goods task feature vector and a task constraint vector are generated. Preferably, the system simultaneously generates a decision preference weight vector based on the enterprise-side target constraints in the task information, which at least represents one or more combinations of cost priority, timeliness priority, risk priority, or multi-objective weighted priority.

[0040] In this embodiment of the invention, the task information in step S2 includes the goods name, attribute information, country of origin, place of departure, destination, declared value, logistics measurement information, special attribute identifiers, trade terms, acceptable mode of transport range, and enterprise-side target constraints; the task constraint vector includes the maximum acceptable timeliness, maximum acceptable cost, countries or ports of entry that cannot be traversed, certification conditions that must be met, and the enterprise's acceptable risk threshold; specifically, the attribute information includes at least one or more of the following: brand, model, material, purpose, composition, or functional attributes; the logistics measurement information includes at least the number of pieces, weight, and volume; the special attribute identifiers include at least one or more of the following: whether it contains batteries, battery type, whether it contains wireless modules, whether it has magnetic components, whether it has dangerous goods attributes, or whether there are special certification requirements.

[0041] See Figure 2 In this embodiment of the invention, the multi-agent collaborative decision-making engine includes a port screening and pattern matching agent, a declaration element classification agent, a compliance risk prediction agent, and a comprehensive evaluation agent. Step S3 includes: Step S31: Using the port screening and pattern matching agent, search for multiple candidate solutions that meet the admission criteria based on the place of departure, destination, acceptable mode of transport and the passable rules in the dynamic port knowledge graph. Match executable customs clearance mode combinations for the candidate transit paths of each candidate solution and output mode information. Perform hard constraint pruning on the path nodes of each candidate transit path to generate path information. Step S32: Use the declaration element classification intelligent agent to extract the regulatory constraints corresponding to each path node from the dynamic port knowledge graph, and input the regulatory constraints and cargo task feature vector together into the final output declaration element scheme package of the classification reasoning model. Step S33: Use the compliance risk prediction agent to model the combination of goods-port-mode-time as nodes, construct the temporal transfer edges and path adjacency edges between nodes to obtain the risk sample graph structure, and output the risk matrix based on the risk sample graph structure. Step S34: The comprehensive evaluation agent performs multi-objective comprehensive scoring on the path information, mode information, application element package and risk matrix of each candidate solution based on the task constraint vector, and selects the candidate solution ranking result containing at least one recommended solution.

[0042] In this embodiment of the invention, in step S31, the port screening and pattern matching agent first performs a path search in the dynamic port knowledge graph, with the origin as the starting point and the destination as the ending point. The search algorithm adopts bidirectional breadth-first search or A* algorithm to reduce the complexity of the search space. During the search process, hard constraint pruning is performed on each path node. The hard constraints include at least whether the cargo category allows transit, whether the current port is open for the clearance mode, whether there is an absolute lack of necessary certificates, whether there are embargo or strong restriction events, and whether the time window is met.

[0043] In this embodiment of the invention, for candidate transit routes pruned through hard constraints, the port screening and pattern matching agent further matches executable customs clearance mode combinations for each path node, including but not limited to: domestic export transit, transit country transit declaration, bonded transit, and import clearance followed by transshipment, etc., and then outputs the candidate path-mode set as shown below: ; in, Represents the candidate path-pattern set. Indicates the first The ports corresponding to each path node Indicates the first Customs clearance mode matching each path node. Indicates the first The estimated transit time for each path node. This indicates the total number of path nodes.

[0044] In this embodiment of the invention, the hard constraints in step S31 include at least whether the cargo category is allowed to transit, whether the current port is open for the customs clearance mode, whether there is an absolute lack of necessary licenses, whether there are embargoes or strong restrictions, and whether the time limit window is met. The method for determining whether the time limit window is met is as follows: calculate the time to reach each path node based on the planned shipment date and the estimated processing time of each path node. If the time exceeds the effective time range of the time window entity associated with the port entity in the dynamic port knowledge graph, or exceeds the maximum acceptable time limit in the task constraint vector, then the hard constraint is not met.

[0045] In this embodiment of the invention, in step S32, the declaration element classification agent extracts the regulatory constraints corresponding to each path node from the dynamic port knowledge graph for each candidate path-pattern set. The regulatory constraints include at least: applicable HS code range, mandatory declaration fields, label requirements, certificate requirements, sensitive attribute identifiers, and historical dispute rules. The above regulatory constraints are concatenated with the cargo task feature vector to form the input feature vector of the classification reasoning model. The classification reasoning model adopts a multi-classification model or graph neural network model based on the Transformer architecture. The classification reasoning model outputs: HS code candidate set and the confidence of each candidate code, declaration element list, missing field list and conflicting field list of target port regulatory conditions, and element completeness score.

[0046] In this embodiment of the invention, the intelligent agent classifying declaration elements does not only output a single HS code, but outputs a declaration element scheme package bound to a specific candidate transit route, so that the differences in declaration requirements under different ports can be explicitly expressed. For example, the same batch of goods to be transferred needs to provide a wireless module frequency band description under candidate transit route A, while it needs to provide a battery UN38.3 test report under candidate transit route B. The declaration element scheme packages for different candidate transit routes list their respective supplementary material requirements.

[0047] In this embodiment of the invention, the declaration element scheme package output in step S32 includes at least a set of HS candidate codes and the confidence level, declaration element list, missing field list, conflict field list of the target port regulatory conditions, and element completeness score corresponding to each HS candidate code. The classification reasoning model outputs declaration element scheme packages bound to the regulatory environment of each candidate transit path for different candidate transit paths. The element completeness score is calculated based on the declaration field completeness rate, field consistency verification results, and the degree of matching with the regulatory condition entity requirements in the knowledge graph. Each item in the missing field list is marked with its corresponding regulatory condition source and supplementation suggestion.

[0048] In this embodiment of the invention, in step S33, the compliance risk prediction agent constructs a graph-structured risk sample for each candidate path-mode-classification scheme. Each node represents a combination state of "goods-port-mode-time". The node features include at least: goods attribute features (such as category, whether it contains batteries, whether it contains wireless modules, etc.), classification confidence features (HS code confidence distribution), element completeness features (completeness rate of declared elements), current port supervision intensity features (such as inspection rate in the past 30 days, average release time), historical inspection event features (historical inspection frequency and rejection rate of the same type of goods at the port), certificate requirement satisfaction features (completeness of required certificates), and policy update time difference features (interval between relevant rules and the current time). Two types of edges are constructed between nodes: temporal transition edges, which connect successive path nodes in the same candidate transit path, representing the temporal dependency of goods flowing along the path; and path adjacency edges, which connect path nodes in different candidate transit paths corresponding to the same physical port or the same transportation stage, for cross-path transmission analysis of risk events.

[0049] In this embodiment of the invention, based on the above graph structure, a graph attention network or a graph convolutional network is used for node-level risk prediction, and the output node-level risk matrix is ​​shown below: ; in, Represents a node-level risk matrix. Represents a path node. The risk category is indicated, and the risk category includes at least: HS classification questioning risk, missing declaration elements risk, missing license or certification risk, origin restriction risk, price abnormality risk, and policy change transmission risk. Preferably, the policy change transmission risk is calculated as follows: when the regulatory rules of a certain path node in the dynamic port knowledge graph are updated, the risk value is propagated along the relationship edge of "affecting subsequent nodes" and the path adjacent edge. The farther the propagation distance and the stronger the path dependence, the higher the transmission risk.

[0050] In this embodiment of the invention, in step S33, a risk matrix is ​​formed based on each path node and the risk category corresponding to each path node. The risk categories include HS classification questioning risk, missing declaration element risk, missing license or certification risk, origin restriction risk, price anomaly risk, and policy change transmission risk. Each node includes cargo attribute characteristics, classification confidence characteristics, element completeness characteristics, current port supervision intensity characteristics, historical inspection event characteristics, certificate requirement satisfaction characteristics, and policy update time difference characteristics. Preferably, before inputting the node characteristics into the risk prediction model, they are standardized (e.g., Z-score standardization or Min-Max normalization) to eliminate the impact of differences in the dimensions between different characteristics on the model prediction results.

[0051] In this embodiment of the invention, in step S34, the comprehensive evaluation agent receives the path information (including the path node sequence and estimated time), mode information (customs clearance mode of each node), declaration element scheme package (including HS code candidates, declaration element list, element completeness score) and risk matrix (various risk values ​​of each node) of each candidate scheme, and performs multi-objective comprehensive scoring according to the enterprise-side target constraints; preferably, the multi-objective comprehensive scoring considers at least the following dimensions: estimated total cost (summed based on the transportation cost of each node, customs clearance service fee, possible demurrage fee, etc.), estimated total timeliness (summed based on the transportation time of each node, customs clearance processing time, cut-off time, etc.), comprehensive risk value (weighted aggregation of each risk category in the risk matrix), classification confidence (the highest confidence or average confidence of HS code candidates), and material completeness score (element completeness score in the declaration element scheme package).

[0052] In this embodiment of the invention, the scoring model may adopt a weighted summation model: ; in, This indicates a multi-objective comprehensive score. This represents the scores for each positive dimension. Represents various risk values, and This represents the weighting coefficient, which is determined based on the decision preferences in the enterprise's objective constraints, or by using the TOPSIS ranking model based on the distance to the ideal solution. The distance between each candidate solution and the positive and negative ideal solutions is calculated to generate the final ranking result of the candidate solutions. Each recommended solution includes at least: recommended transit port sequence, recommended customs clearance mode for each node, corresponding HS code and declaration element list, required license / label / supplementary material list, risk pretreatment suggestions, and alternative solutions.

[0053] In this embodiment of the invention, after performing step S3, the method further includes: During the actual transit of goods, the system continuously receives update events from the dynamic port knowledge graph and performs dynamic monitoring on the recommended schemes selected by the user. When a trigger event is detected, the system freezes the path nodes that have been declared or have been irreversibly executed during the execution of the recommended scheme, identifies the impact domain of the trigger event to determine the subsequent node subgraphs that need to be recalculated, and performs local or global replanning on the subsequent node subgraphs to output explanations of the differences between the old and new schemes, explanations of risk changes, and operational suggestions. Specifically, dynamic monitoring is achieved by listening to the update event stream of the dynamic port knowledge graph. When the attribute of any entity type or relation type in the dynamic port knowledge graph changes (such as the port access status changing from "permitted" to "restricted", the certificate requirement changing from "recommended" to "mandatory", the inspection intensity level being increased, etc.), or when a new risk event entity is added, the system determines it as a potential triggering event.

[0054] In this embodiment of the invention, identifying the influence domain of a triggering event specifically includes: taking the port entity or regulatory condition entity associated with the triggering event as the center, performing forward propagation along the "affecting subsequent nodes" relationship edges and path adjacency edges in the dynamic port knowledge graph to determine the range of path nodes affected by the triggering event; then calculating the intersection of the path nodes in the selected recommended scheme with the influence domain; if the intersection only contains subsequent path nodes that have not yet been executed, then only the subsequent node subgraph formed by these path nodes is locally replanned; if the intersection contains executed nodes or the core constraints whose influence scope covers the entire scheme, then global replanning is performed.

[0055] In this embodiment of the invention, the specific execution method of local replanning is as follows: freeze the path nodes that have been declared or irreversibly executed and their satisfied constraints; take the current actual path node as the new starting point and the original destination as the end point; re-execute the search, matching, prediction and scoring process of steps S31 to S34 in the dynamic port knowledge graph; but only calculate the subsequent node subgraphs that have not been executed; during the replanning process, retain reusable declaration elements and satisfied certificate results to reduce redundant calculations; if local replanning cannot find a feasible solution that satisfies all hard constraints and task constraint vectors, it is automatically upgraded to global replanning, that is, re-execute the complete solution search and evaluation process from the original departure point.

[0056] In this embodiment of the invention, the output of the differences between the old and new schemes includes at least: the changed path nodes, the changed customs clearance mode, and the changed declaration elements; the risk change description includes at least: the change in the comprehensive risk value and the change direction of each risk category; the operation suggestions include at least: a list of materials that need to be supplemented, the names of the certificates that need to be recertified, and the suggested operation time window.

[0057] In this embodiment of the invention, the triggering events include subsequent changes in port access status, changes in licensing rules, a significant increase in inspection intensity, the inclusion of sensitive HS codes in the key supervision list, changes in the feasibility of transportation methods, and recall events, prohibited or restricted transportation events, or security events related to cargo attributes. Preferably, the criteria for judging a significant increase in inspection intensity are: the "supervision intensity" attribute value of the port entity in the dynamic port knowledge graph increases by more than a preset threshold (e.g., 20%) within a preset time window (e.g., 7 days or 30 days), or the port adds a "supplementary review" relationship edge related to cargo category. The criteria for judging a sensitive HS code being included in the key supervision list are: the HS code entity in the dynamic port knowledge graph adds a "key supervision" label or is associated with a new risk event entity.

[0058] See Figure 3The present invention also discloses a cross-border transit business processing system based on a dynamic port knowledge graph, which applies the above-mentioned cross-border transit business processing method based on a dynamic port knowledge graph, including: The data acquisition module is used to collect policy data, logistics data, declaration data, and risk event data, and perform standardized cleaning to obtain multi-source data; The knowledge graph construction module, connected to the data acquisition module, is used to obtain structured and unstructured information from multi-source data and construct a dynamic port knowledge graph containing several entity types and relationship types. The task processing module is used to receive task information of goods to be transferred and to standardize the task information to generate a cargo task feature vector and a task constraint vector. The multi-agent collaborative decision-making engine connects the knowledge graph construction module and the task processing module respectively. Under the constraints of the dynamic port knowledge graph, it searches for multiple candidate transit routes that meet the access conditions based on the cargo task feature vector and task constraint vector. Then, it performs customs clearance mode matching, declaration element matching, risk matrix matching, and multi-objective comprehensive scoring on each candidate transit route to obtain a ranking result of candidate schemes containing at least one recommended scheme, so as to guide users in cross-border transit business processing.

[0059] In the description of this invention, the references to "one embodiment," "some embodiments," "in this embodiment," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0060] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for processing cross-border transit business based on a dynamic port knowledge graph, characterized in that, Includes the following steps: Step S1: Obtain structured and unstructured information from multi-source data and construct a dynamic port knowledge graph containing several entity types and relationship types; Step S2: Receive the task information of the goods to be transferred, and perform standardized processing on the task information to generate a cargo task feature vector and a task constraint vector; Step S3: Under the constraints of the dynamic port knowledge graph, a multi-agent collaborative decision engine is run to search for multiple candidate transit routes that meet the admission conditions based on the cargo task feature vector and the task constraint vector. Each candidate transit route is then processed sequentially with customs clearance mode matching, declaration element matching, risk matrix matching, and multi-objective comprehensive scoring to obtain a candidate scheme ranking result containing at least one recommended scheme, so as to guide the user in cross-border transit business processing.

2. The method for processing cross-border transit business based on dynamic port knowledge graph according to claim 1, characterized in that, The entity types in step S1 include port entities, country and region entities, mode of transport entities, customs clearance mode entities, HS code entities, regulatory condition entities, license entities, risk event entities, and cargo category entities; the relationship types include applicable, prohibited, required submission, associated risk, permitted transit, applicable mode of transport, applicable time window, and impact on subsequent nodes, and each entity type or relationship type is bound to attribute fields including effective time, expiration time, source identifier, and credibility score.

3. The method for processing cross-border transit business based on dynamic port knowledge graph according to claim 1, characterized in that, The task information in step S2 includes cargo name, attribute information, country of origin, place of departure, destination, declared value, logistics measurement information, special attribute identifiers, trade terms, acceptable mode of transport range, and enterprise-side target constraints; the task constraint vector includes maximum acceptable timeliness, maximum acceptable cost, countries or ports of entry that cannot be passed through, certification conditions that must be met, and enterprise-bearable risk thresholds.

4. The method for processing cross-border transit business based on dynamic port knowledge graph according to claim 3, characterized in that, The multi-agent collaborative decision-making engine includes a port screening and pattern matching agent, a declaration element classification agent, a compliance risk prediction agent, and a comprehensive evaluation agent. Step S3 includes: Step S31: The port screening and pattern matching agent searches for multiple candidate schemes that meet the admission criteria based on the place of departure, the destination, the acceptable mode of transport and the passable rules in the dynamic port knowledge graph. It also matches executable customs clearance mode combinations for the candidate transit paths of each candidate scheme and outputs mode information. Furthermore, it performs hard constraint pruning on the path nodes of each candidate transit path to generate path information. Step S32: Use the declaration element classification agent to extract the regulatory constraints corresponding to each path node from the dynamic port knowledge graph, and input the regulatory constraints and the cargo task feature vector into the final output declaration element scheme package of the classification reasoning model. Step S33: The compliance risk prediction agent is used to model the combination of goods-port-mode-time as nodes, and the temporal transition edges and path adjacency edges between nodes are constructed to obtain the risk sample graph structure, and the risk matrix is ​​output based on the risk sample graph structure. Step S34: The comprehensive evaluation agent performs multi-objective comprehensive scoring on the path information, pattern information, application element package and risk matrix of each candidate solution based on the task constraint vector, and filters to obtain a ranking result of candidate solutions containing at least one recommended solution.

5. The method for processing cross-border transit business based on dynamic port knowledge graph according to claim 4, characterized in that, The hard constraints in step S31 include at least whether the cargo category allows transit, whether the current port is open for the clearance mode, whether there is an absolute lack of necessary licenses, whether there are any embargoes or strict restrictions, and whether the time window is met.

6. The method for processing cross-border transit business based on dynamic port knowledge graph according to claim 4, characterized in that, The declaration element scheme package output in step S32 includes at least a set of HS candidate codes and the confidence level, declaration element list, missing field list, conflict field list with the target port regulatory conditions, and element completeness score for each HS candidate code. The classification reasoning model outputs the declaration element scheme package that is bound to the regulatory environment of each candidate transit path for different candidate transit paths.

7. The method for processing cross-border transit business based on dynamic port knowledge graph according to claim 4, characterized in that, In step S33, the risk matrix is ​​formed based on each node and the risk category corresponding to each node. The risk categories include HS classification questioning risk, missing declaration elements risk, missing license or certification risk, origin restriction risk, abnormal price risk, and policy change transmission risk. Each node includes cargo attribute characteristics, classification and placement characteristics, element completeness characteristics, current port supervision intensity characteristics, historical inspection event characteristics, certificate requirement satisfaction characteristics, and policy update time difference characteristics.

8. The method for processing cross-border transit business based on dynamic port knowledge graph according to claim 1, characterized in that, After performing step S3, the method further includes: During the actual transit of goods, the system continuously receives update events from the dynamic port knowledge graph and dynamically monitors the recommended schemes selected by the user. When a triggering event is detected, the system freezes path nodes that have been declared or irreversibly executed during the execution of the recommended scheme, identifies the impact domain of the triggering event to determine the subsequent node subgraphs that need to be recalculated, and performs local or global replanning on the subsequent node subgraphs to output explanations of the differences between the old and new schemes, explanations of risk changes, and operational suggestions.

9. The method for processing cross-border transit business based on dynamic port knowledge graph according to claim 8, characterized in that, The triggering events include subsequent changes in port access status, changes in licensing rules, a significant increase in inspection intensity, the inclusion of sensitive HS codes in the key supervision list, changes in the feasibility of transportation methods, and recall events, prohibited or restricted transportation events, and security events related to cargo attributes.

10. A cross-border transit business processing system based on a dynamic port knowledge graph, characterized in that: The cross-border transit business processing method based on dynamic port knowledge graph as described in any one of claims 1-9 includes: The data acquisition module is used to collect policy data, logistics data, declaration data, and risk event data, and perform standardized cleaning to obtain multi-source data; The knowledge graph construction module, connected to the data acquisition module, is used to obtain structured and unstructured information from the multi-source data and construct a dynamic port knowledge graph containing several entity types and relationship types. The task processing module is used to receive task information of goods to be transferred and to standardize the task information to generate a cargo task feature vector and a task constraint vector. The multi-agent collaborative decision-making engine, connected to the graph construction module and the task processing module respectively, is used to search for multiple candidate transit routes that meet the admission conditions under the constraints of the dynamic port knowledge graph, based on the cargo task feature vector and the task constraint vector. It then performs customs clearance mode matching, declaration element matching, risk matrix matching, and multi-objective comprehensive scoring on each candidate transit route to obtain a ranking result of candidate schemes containing at least one recommended scheme, so as to guide users in cross-border transit business processing.

Images