A method and system for protecting consumer product data security based on federated learning

By combining a dynamic polynomial masking mechanism with lightweight Lagrange aggregation and LSTM generator polynomial order, the security and computational efficiency issues of traditional federated learning in the consumer goods industry are solved. This approach achieves a balance between strong privacy protection and dynamic security strategies, blocks model inversion attacks, and improves data security and computational efficiency.

CN120850346BActive Publication Date: 2026-03-03BEIJING JUZHIXING BIG DATA DEVELOPMENT CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202511075733.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-08-01
Publication Date
2026-03-03
Estimated Expiration
2045-08-01

AI Technical Summary

Technical Problem

Traditional federated learning faces challenges in the consumer goods industry, including model inversion attacks, collusion attacks, and high computational overhead, making it difficult to balance strong privacy protection, dynamic security strategies, and lightweight computing.

Method used

Employing a dynamic polynomial masking mechanism and lightweight Lagrange aggregation, the polynomial order is generated through LSTM. Production enterprises only upload the polynomial share after encryption masking, and the public service platform performs aggregation and decryption to block model inversion attacks. Furthermore, by introducing the polynomial into the LSTM-generated order that balances security and efficiency, replay attacks and collusion attacks are resisted.

Benefits of technology

This approach achieves robust privacy protection, dynamic security strategies, and lightweight computation in federated learning for consumer product data security. It blocks model inversion attacks, reduces the likelihood of encryption schemes being reverse-engineered, and improves data security and computational efficiency for manufacturing enterprises.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120850346B_ABST
    Figure CN120850346B_ABST
Patent Text Reader

Abstract

The present application relates to the field of federated learning data security protection, and more particularly to a method and system for protecting consumer product data security based on federated learning, which comprises: generating a polynomial order by a prediction model based on LSTM architecture from the production state data of consumer products of production enterprises and federated learning gradient data; generating a dynamic polynomial and a local end encryption share of the production enterprise based on the polynomial order and the federated learning gradient data; uploading the local end encryption share of the production enterprise to a public service platform by the local end of the federated learning, aggregating and decrypting the local end encryption share of the production enterprise, generating federated learning gradient data, and using the federated learning gradient data to update the federated learning model. The present application realizes the balanced security and efficiency order generated by the polynomial introduction LSTM, and realizes the federated learning consumer product data security method considering privacy protection, security strategy and lightweight calculation.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Asynchronous federal learning method, device and system and storage medium

    CN116245194A

  • Privacy protection method based on generative adversarial network and gradient smoothing, program, equipment and storage medium

    CN118296650A