Search

CN-121998739-A - Intelligent tracking management method and system for cross-border business based on informatization

CN121998739ACN 121998739 ACN121998739 ACN 121998739ACN-121998739-A

Abstract

The invention belongs to the technical field of electronic commerce informatization, in particular to an intelligent tracking management method and system for cross-border business based on informatization, wherein the method comprises the following steps of responding to business anomaly detection, dispatching independent participants of data ownership; the method comprises the steps of constructing a federal causal map across participants through cooperation of a federal causal discovery module, dynamically configuring a business digital twin model based on causal relations in the map to generate a causal constraint enhanced digital twin instance, coding map structural features in a deep reinforcement learning agent, taking the enhanced digital twin instance as an environment, and operating the agent to output an intervention action sequence. The method solves the problems of inter-border electronic commerce data island, decision delay and lack of intelligent support, realizes privacy protection collaborative analysis of the inter-border data, active prediction and interpretable intelligent decision based on a causal mechanism, and improves the overall toughness and decision efficiency of a supply chain.

Inventors

  • Leng Xiangliang

Assignees

  • 杭州恒健供应链科技有限公司

Dates

Publication Date
20260508
Application Date
20260127

Claims (10)

  1. 1. The intelligent tracking management method for the cross-border business based on informatization is characterized by comprising the following steps: scheduling at least two data-hosting independent parties in response to detection of a target traffic anomaly; Through a federal causal discovery module deployed between the participants, cooperatively constructing a federal causal map representing causal dependency relationships among service entities crossing the participants; Based on the causal relation in the federal causal map, carrying out dynamic parameter configuration on a pre-constructed business digital twin model to generate a digital twin instance with enhanced causal constraint; In a deep reinforcement learning agent, encoding structural features in the federal causal map as part of a state vector, and taking the causal constraint enhanced digital twin instance as a training and deduction environment; and outputting an intervention action sequence aiming at the target business abnormality by operating the deep reinforcement learning intelligent agent.
  2. 2. The method of claim 1, wherein the federal cause and effect discovery module performs the following operations: Each participant generates a weighted local causal subgraph by utilizing a differentiable causal structure learning model based on local data; Safely exchanging the structural gradient of the local causal subgraph among all the participants through homomorphic encryption channels; And on a coordination server, the structural gradients are aggregated, and a global federal causal map meeting the property of the directed acyclic graph is obtained through optimization solution with constraint.
  3. 3. The method of claim 2, wherein the differentiable causal structure learning model is a NOTEARS algorithm based on a neural network or a variant thereof, and wherein the constrained optimization solution uses an augmented lagrangian approach, the constraint being a directed acyclic graph condition.
  4. 4. The method according to claim 1, wherein the dynamic parameter configuration of the business digital twin model based on federal causal map is specifically: Identifying a causal path in the federal causal map, wherein the confidence of the causal path exceeds a preset threshold; Mapping each of the causal paths into a logical rule or probability transfer function of a combination direction; The logic rules or probability transfer functions are injected as part of a core state machine when initializing the business digital twin model.
  5. 5. The method of claim 4, wherein the mapping process employs a rule engine in combination with a neural network to generate deterministic logic rules using the rule engine for deterministic causal relationships that are well-structured and to generate new probability transfer functions using a trained graph neural network for complex causal relationships that have uncertainty.
  6. 6. The method according to claim 1, wherein the encoding of structural features in the federal causal map as state vectors, in particular: performing one-time reverse breadth-first search with a target abnormal node as a starting point on the federal causal map, and extracting all causal adjacent nodes and edges in K hops; feature aggregation is carried out on the causal adjacent nodes and edges by using a graph annotation force network, so that a graph embedding vector with fixed dimension is generated; And splicing the map embedding vector with the current macroscopic state vector of the digital twin instance to be used as the complete state observation of the deep reinforcement learning intelligent agent.
  7. 7. The method of claim 1, wherein the training of the deep reinforcement learning agent employs a causal action mask mechanism: at each decision moment, calculating a subset of actions allowed to be performed according to the federal causal map and the current state of the digital twin instance; Masking actions outside the subset of actions with a negative reward, and guiding the deep reinforcement learning agent to be used only in a preset causally reasonable action space.
  8. 8. The method of claim 7, wherein the method of calculating the subset of actions allowed to be performed is: Constructing a temporary intervention diagram, and simulating the influence of an execution candidate action on a key state variable in the digital twin instance; Judging whether the influence violates strong causal constraint or business safety boundary in the federal causal map by using a lightweight causal effect prediction model; candidate actions that do not violate constraints and boundaries are included in the subset of actions that are allowed to be performed.
  9. 9. The method of claim 1, further comprising a policy evaluation step based on a counter fact interpretation model: Starting a counter fact interpretation model coupled with the causal constraint enhanced digital twin instance while outputting the intervention action sequence; the counterfactual interpretation model generates a simplest comparison intervention scheme which is different from the recommended sequence by minimizing the intervention cost; And quantifying the expected difference of the comparison intervention scheme and the recommended sequence on the key business indexes by simulating and executing the comparison intervention scheme and the recommended sequence, and attaching the difference as a reference basis of decision reliability to an output result.
  10. 10. An informationized cross-border business intelligence tracking management system for implementing the method of any one of claims 1 to 9, the system comprising: the federal cause and effect discovery module is configured with a privacy calculation middleware and is used for cooperatively constructing the federal cause and effect map; the causal rule injection engine is connected with the federal causal discovery module and is used for dynamically configuring the business digital twin model; the deep reinforcement learning subsystem integrating the state encoder and the action mask module is respectively connected with the outputs of the federal causal discovery module and the causal rule injection engine and is used for generating the intervention action sequence; And the strategy interpretation and evaluation module is connected with the deep reinforcement learning subsystem and is used for executing strategy steps corresponding to the intervention action sequence.

Description

Intelligent tracking management method and system for cross-border business based on informatization Technical Field The invention relates to the technical field of electronic commerce informatization, in particular to an intelligent tracking management method and system for cross-border business based on informatization. Background With the rapid development of globalization electronic commerce, cross-border electronic commerce has long chain, multiple links and complex participation subject, and relates to a plurality of nodes such as domestic goods collection, international transportation, import and export clearance, overseas warehouse allocation and the like. At present, the following problems exist in a common cross-border business tracking management system: And the data island and the information barriers are that information such as orders, logistics, storage, payment, customs, customer service and the like are scattered in different platforms and systems, the data formats are different, the integration is difficult, and an end-to-end transparent view is difficult to form. Tracking passive and early warning hysteresis, wherein most systems only provide node state inquiry based on logistics list numbers, early warning depends on simple and fixed rules (such as overtime threshold values), and active and predictive risk identification cannot be performed based on multi-source data fusion (such as predicting that a certain batch of goods can clear the delay in a specific port with high probability). Decision support lacks intelligence-when anomalies occur, the system is typically only able to alert of anomalies, but fails to provide root cause diagnostics and optimal handling advice based on historical data and intelligent analysis (how the alternate logistics channel should be started, when customers should be actively contacted, and compensation schemes provided), decision making is highly dependent on human experience. Compliance and privacy challenges-cross-border business involves data compliance requirements of different legal domains, and it is difficult to achieve maximum utilization of data value on the premise of legal compliance so as to improve the toughness of the whole supply chain. The prior art (such as CN118569759B, CN117495229A, CN119398479 a) is improved in some aspects of logistics tracking, data association or anomaly early warning, but fails to systematically solve the above problems, and lacks a closed-loop management framework with active prediction and intelligent decision making capability for deep fusion of multi-source heterogeneous data. Disclosure of Invention The invention provides an informatization-based cross-border electric business intelligent tracking management method and system, and aims to solve the technical problem of how to realize the integration of causal cognition and intelligent decision-making of cross-participants under the condition of scattered data ownership and privacy compliance. In a first aspect, an embodiment of the present invention provides an intelligent tracking management method for cross-border business based on informatization, including: scheduling at least two data-hosting independent parties in response to detection of a target traffic anomaly; Through a federal causal discovery module deployed between the participants, cooperatively constructing a federal causal map representing causal dependency relationships among service entities crossing the participants; Based on the causal relation in the federal causal map, carrying out dynamic parameter configuration on a pre-constructed business digital twin model to generate a digital twin instance with enhanced causal constraint; In a deep reinforcement learning agent, encoding structural features in the federal causal map as part of a state vector, and taking the causal constraint enhanced digital twin instance as a training and deduction environment; and outputting an intervention action sequence aiming at the target business abnormality by operating the deep reinforcement learning intelligent agent. The intelligent tracking management method for the cross-border electronic business based on informatization has the technical effects that the method realizes root cause positioning, active prediction and interpretable intelligent decision sequence generation of the complex business abnormality of the cross-border electronic business on the premise of data privacy compliance by constructing a federal causal map of the cross-participants and enhancing digital twin and guiding reinforcement learning. Further, the federal cause and effect discovery module performs the following operations: Each participant generates a weighted local causal subgraph by utilizing a differentiable causal structure learning model based on local data; Safely exchanging the structural gradient of the local causal subgraph among all the participants through homomorphic encryption channels; And on a coordination server,