A method for federated learning optical module colony life and early failure warning

By using a federated learning framework and a TA-LSTM model, the problems of capturing the progressive degradation trajectory of optical modules and mining common patterns among groups are solved. This enables early warning of optical module lifespan and cross-client collaborative analysis of faults, improving prediction accuracy and location efficiency.

CN122137463AActive Publication Date: 2026-06-02CHENGDU GUANGCHUANGLIAN CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHENGDU GUANGCHUANGLIAN CO LTD
Filing Date
2026-05-06
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

Existing technologies struggle to capture the gradual degradation trajectory of optical module performance parameters, making it difficult to identify latent and gradual failures in their early stages. There is a lack of exploration into common degradation patterns among different individual groups, and data from each workstation is isolated, lacking collaborative analysis and early warning mechanisms, resulting in low fault location efficiency and a high rate of missed detections.

Method used

Using a federated learning framework, the optical power and bias current of the optical module are collected through the degradation trajectory feature extraction module to construct a degradation feature vector. The feature vector is then processed using a TA-LSTM model. Gradient aggregation and early warning are performed on the client and cloud to establish a dynamic fluctuation benchmark and achieve cross-client collaborative analysis and early warning.

Benefits of technology

Without exposing the raw data, it accurately captures the gradual degradation trajectory and abrupt failure precursors of optical module performance parameters, significantly improving the accuracy and real-time performance of individual optical module lifetime prediction, reducing the false alarm rate and missed detection rate, and realizing collaborative early warning across clients.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122137463A_ABST
    Figure CN122137463A_ABST
Patent Text Reader

Abstract

This invention relates to the field of optical communication technology and discloses a federated learning method for predicting the collective lifetime and early failure of optical modules. It addresses the challenges of capturing individual degradation trajectories, uncovering common patterns within the group, and the lack of collaborative early warning due to isolated workstation data in optical module lifetime prediction. The invention extracts optical power degradation rate, bias current rise rate, and correlation features from the client side, constructs a degradation feature vector, and inputs it into a TA-LSTM model. This enables accurate capture of gradual degradation and abrupt precursors in individual optical modules. In the cloud, federated learning aggregates gradients from each client side to uncover common degradation patterns within the group. Gradient variance regularization and adaptive weights enhance the robustness of the global model. Dynamic hierarchical early warning thresholds are set based on the mean and standard deviation of the group degradation features, effectively distinguishing between occasional fluctuations and genuine failure precursors. This significantly reduces the false alarm and false false alarm rates, achieving the dual goals of cross-client collaborative early warning and privacy protection.
Need to check novelty before this filing date? Find Prior Art