CN-122020718-A - Cross-border trade data processing method and system based on intelligent port

CN122020718ACN 122020718 ACN122020718 ACN 122020718ACN-122020718-A

Abstract

The invention relates to the field of intelligent data processing, and particularly discloses a cross-border trade data processing method and system based on intelligent ports, wherein the method comprises the steps of processing original data in a local trusted execution environment of each port node, generating privacy protection feature vectors, and generating verifiable certificates for verifying calculation compliance by using zero knowledge proof; the method comprises the steps of verifying certificates through a distributed consensus network, adopting secure multiparty calculation to fuse feature vectors of all nodes to generate cross-domain joint features, inputting the features into a decentralised intelligent network to carry out asynchronous negotiation, outputting a collaborative risk research and judgment report, finally optimizing a local model under differential privacy protection according to report feedback by all nodes, and enabling the system to comprise four corresponding modules. The invention realizes the trusted collaboration and the deep risk insight under the premise that the original data does not go out of the domain, and ensures the continuous autonomous evolution of the system.

Inventors

ZHANG ZONGWANG
WANG FENG
MEI TAO
ZHOU JIAN
Zi Zhengyuan

Assignees

山东省电子口岸有限公司

Dates

Publication Date: 20260512
Application Date: 20260204

Claims (10)

1. The intelligent port-based cross-border trade data processing method is characterized by comprising the following steps of: processing the original data in a local trusted environment of each node to generate a privacy protection feature vector and a corresponding verifiable credential; Verifying the verifiable certificate in a distributed network, and fusing privacy protection feature vectors of all nodes to generate cross-domain joint features; inputting the cross-domain joint characteristics into a decentralised intelligent agent network for asynchronous negotiation, and outputting a collaborative risk research and judgment report; Each node generates and applies an optimization gradient locally based on the collaborative risk study report to update a feature extraction model.
2. The method for processing cross-border trade data based on intelligent port according to claim 1, wherein the steps of processing raw data in a local trusted environment of each node to generate a privacy protection feature vector and a corresponding verifiable credential comprise: Analyzing the original cross-border trade data in a trusted execution environment, extracting structured fields, mapping each field value to a preset standard term dictionary, and generating a standardized structured data record; Inputting the structured data record into a preset feature extraction model, and mapping the input into a numerical vector serving as a privacy protection feature vector by the model through a multi-layer perception network and an attention mechanism; Verifiable credentials for verifying their computational integrity and compliance are generated for the privacy preserving feature vectors based on a zero knowledge proof method.
3. The intelligent port based cross-border trade data processing method according to claim 2, wherein generating verifiable credentials for verifying its computational integrity and compliance for the privacy preserving feature vector based on a zero knowledge proof method, specifically comprises: constructing an arithmetic circuit according to the feature extraction model, wherein the public input of the arithmetic circuit is the cryptographic promise of the privacy protection feature vector, and the secret input is the structured data record and model intermediate variable; In the trusted execution environment, using secret variables in the structured data record and model calculation process as witness, running a proof generation algorithm of the zero knowledge proof method, and outputting a zero knowledge proof for proving that an arithmetic circuit is correctly satisfied; and packaging the zero knowledge proof, the cryptographic commitment of the privacy protection feature vector and the public parameters for verification together, generating the verifiable certificate, and binding and outputting the verifiable certificate and the corresponding privacy protection feature vector.
4. The method for processing cross-border trade data based on intelligent port as claimed in claim 3, wherein verifying the verifiable certificate in a distributed network and fusing privacy protection feature vectors of each node to generate cross-domain joint features comprises: Each node broadcasts the privacy protection feature vector and the verifiable credentials bound by the privacy protection feature vector to a distributed consensus network as a transaction through a data trusted delivery protocol; after receiving a plurality of related transactions, the cooperative nodes in the distributed consensus network perform parallel verification on all verifiable certificates based on a preset zero knowledge proof verification contract, and mark corresponding privacy protection feature vectors as valid data to be fused only after all certificates pass verification; The cooperative nodes input all privacy protection feature vectors marked as effective, a preset safe multiparty calculation fusion function is input, and a cross-domain joint feature fusing all node features is output.
5. The method for processing cross-border trade data based on intelligent port according to claim 4, wherein the cooperative node inputs all privacy protection feature vectors marked as valid, a preset secure multiparty computation fusion function, and outputs a cross-border joint feature fusing all node features, specifically comprising: each node splits the effective privacy protection feature vector into encrypted shares through a secret sharing algorithm and distributes the encrypted shares to all cooperative nodes; all cooperating nodes are locally based on the set of shares held Executing a secure multiparty computing fusion function, and specifically adopting weighted average fusion: Node setting Is given by the weight of Weighted average joint feature The method comprises the following steps: Wherein: For the privacy-preserving feature vector, Is the number of cooperative nodes; Node Computing local results on shares : Wherein: A plaintext share for local storage; Finally each node Generating a result share of cross-domain federated features Satisfies the following conditions ; The nodes summarize the result shares through the security protocol, and jointly recover and output the complete plaintext cross-domain joint characteristics.
6. The intelligent port based cross-border trade data processing method according to claim 5, wherein the cross-domain joint feature is input to a decentralized intelligent agent network for asynchronous negotiation, and a collaborative risk study report is output, specifically comprising: inputting the cross-domain combined features into a pre-trained decentralised risk research and judgment agent network, wherein a plurality of special agents in the network respectively correspond to different risk research and judgment dimensions, and each agent performs reasoning based on the cross-domain combined features to generate primary risk research and judgment sub-results of each dimension; Each intelligent agent exchanges respective preliminary risk research judging sub-results based on a preset rule through a communication channel built in the network and carries out multiple rounds of negotiations to form a risk judging set after negotiations; and automatically generating a collaborative risk study report according to the negotiated risk decision set, wherein the collaborative risk study report comprises cross-gateway area associated risk labels, risk conduction path speculation and collaborative treatment priority suggestions.
7. The method for processing cross-border trade data based on intelligent port according to claim 6, wherein each agent exchanges respective preliminary risk development sub-results based on a predetermined rule through a communication channel built in a network and performs multiple rounds of negotiations to form a risk development set after negotiations, specifically comprising: Each intelligent agent broadcasts the preliminary risk studying and judging sub-result and the corresponding confidence coefficient through a communication channel; based on a preset consensus rule, each agent iteratively adjusts own judgment through a Bayesian updating or consensus algorithm according to the received sub-results and confidence levels of other agents in each round of negotiation; after multiple rounds of interaction, when the judgment of existence, type and association relation of all the agents on risks reaches a preset consensus threshold, locking and finally judging and aggregating to form a risk judgment set after negotiation.
8. The intelligent port based cross-border trade data processing method according to claim 7, wherein each node locally generates and applies an optimization gradient based on the collaborative risk development report to update a feature extraction model, comprising: each intelligent port node subscribes to the distributed consensus network and acquires a collaborative risk research report containing self node identification and a subsequent disposal verification result; in a local trusted execution environment, based on a differential privacy technology, carrying out security association analysis on a treatment verification result in a report and a privacy protection feature vector generated by self history to generate an optimization gradient aiming at a local feature extraction model; and updating the parameters of the locally stored feature extraction model by a secure aggregation algorithm in the local trusted execution environment by using the optimization gradient.
9. The intelligent port based cross-border trade data processing method according to claim 8, wherein the generating an optimized gradient for the local feature extraction model specifically comprises: pairing the historically stored privacy protection feature vectors with the disposal verification results in the local trusted execution environment to form a group of encryption training sample pairs for model optimization; setting the parameters of the local feature extraction model as Calculating predictions using forward propagation Wherein: For the feature extraction model's mapping function, In order to output the layer weights, The function is activated for Sigmoid, For the risk prediction probability, The privacy protection feature vector; the loss function adopts binary cross entropy: Wherein: As a function of the loss, To dispose of the validation tag; Computing raw gradients ; Calculating L2 norms of gradients ; Applying clipping thresholds Obtaining gradient after cutting ; Injection of Gaussian noise ; Wherein: For the purpose of a privacy budget, Is the gradient after noise injection; A multi-element Gaussian noise with the mean value of zero; And performing norm clipping on the initial noise gradient, performing standardization processing on the clipped gradient, and finally outputting an optimized gradient which meets the privacy protection requirement and has stable numerical value.
10. A cross-border trade data processing system based on an intelligent port, which is used for implementing the cross-border trade data processing method based on an intelligent port, comprising: the local trusted characteristic and credential generation module is configured to process locally stored original cross-border trade data in a trusted execution environment in a data main authority domain of any intelligent port node participating in collaborative analysis; The cross-domain trusted feature fusion module is configured to issue privacy protection feature vectors generated by each node and verifiable credentials attached to the privacy protection feature vectors to a distributed consensus network through a data trusted delivery protocol; in the distributed consensus network, after verifying that all received verifiable credentials are valid, fusing privacy protection feature vectors of a plurality of sources to generate a global cross-domain joint feature; The intelligent risk-decentralizing judging module is configured to input the cross-domain joint characteristics into a pre-trained intelligent risk-decentralizing judging agent network, asynchronously reasoning and negotiating by a plurality of special agents in the intelligent agent network based on the cross-domain joint characteristics, and outputting a collaborative risk judging report; The privacy security feedback optimization module is configured to enable each intelligent port node to acquire a cooperative risk research report related to the intelligent port node from the distributed consensus network, conduct security association analysis on treatment verification results in the report and privacy protection feature vectors generated by the privacy protection feature vectors based on a self history in a local trusted execution environment based on a differential privacy technology to generate an optimization gradient aiming at a local feature extraction model, and update the local model by utilizing the optimization gradient.

Description

Cross-border trade data processing method and system based on intelligent port Technical Field The invention relates to the technical field of intelligent data processing, in particular to a cross-border trade data processing method and system based on an intelligent port. Background In recent years, intelligent port construction has become a key to improving cross-border trade efficiency and safety. The prior art focuses mainly on data automation and flow optimization inside a single port, such as automatic entry of customs notes by Optical Character Recognition (OCR) technology or tracking of cargo state by using internet of things sensors. However, existing solutions face fundamental bottlenecks when dealing with complex regulatory scenarios requiring multiple port, multiple regulatory agency collaboration. The core contradiction is the conflict between the data master right rigidity requirement and the cross-border collaborative supervision requirement. On one hand, the local storage regulations of data of various countries are increasingly strict, the original trade data cannot flow across the border in principle, and on the other hand, high-risk behaviors such as smuggling, bill washing and the like often have cross-region association characteristics, and collaborative analysis of a data layer is needed to identify hidden risks. The existing main stream technical routes are difficult to properly solve the contradiction: And in the centralized data pool scheme, each port is required to upload data to a central server for analysis, and the cooperation can be realized, but the data main authority principle is directly violated, so that great compliance risks exist, and the implementation is difficult in practice. In the traditional federal learning scheme, although the original data set is avoided, the gradient exchange in the model training process can still implicitly reveal sensitive information, an effective verification mechanism for the compliance of the data preprocessing of the participants is lacked, and the trust basis is weak. The simple data desensitization exchange scheme is that the data value is seriously lost after being processed in a hash mode, a generalization mode and the like, complex associated risk analysis is difficult to support, and the reliability and the consistency of a self-certification desensitization process cannot be ensured. Therefore, the prior art either sacrifices compliance or sacrifices analysis performance, resulting in the intelligent port system being in a data island state for a long period of time, and failing to form a cross-domain supervisory resultant force. What is needed is a new method for co-processing cross-border trade data, which can realize data availability invisible, process credibility auditability and system energy evolution, in legal framework. Disclosure of Invention The invention aims to solve the problem of cross-domain supervision coordination under the constraint of data master rights in the prior art, and provides a cross-border trade data processing method and system based on an intelligent port. In order to achieve the above purpose, the present invention adopts the following technical scheme: a cross-border trade data processing method based on intelligent port comprises the following steps: S1, processing original data in a local trusted environment of each node to generate a privacy protection feature vector and a corresponding verifiable credential; S2, verifying the verifiable certificate in a distributed network, and fusing privacy protection feature vectors of all nodes to generate cross-domain joint features; s3, inputting the cross-domain joint characteristics into a decentralised intelligent agent network for asynchronous negotiation, and outputting a collaborative risk research and judgment report; S4, each node generates and applies an optimization gradient locally based on the collaborative risk research report so as to update a feature extraction model. As a further technical solution of the present invention, the S1 specifically includes: S11, analyzing the original cross-border trade data in a trusted execution environment, extracting structured fields at least comprising commodity description, declaration value, logistics entity and shipper, mapping each field value to a preset standard term dictionary, and generating a standardized structured data record; s12, inputting the structured data record into a preset feature extraction model, and mapping the input into a numerical vector with fixed length and desensitization as a privacy protection feature vector by a multi-layer perception network and an attention mechanism; S13, generating verifiable credentials for verifying the computing integrity and compliance of the privacy protection feature vector based on a zero knowledge proof method. As a further technical solution of the present invention, the S13 specifically includes: s131, constructing an arithmetic circuit ac