AI Model Retraining Threshold for Optical Transmission Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The retraining of AI models in optical transmission networks after an abnormality and its restoration can lead to increased costs due to the need to retrain all affected models, even if only slight changes occur in the network.
Innovation Solution
A controller is implemented with AI models specific to each optical transmission apparatus, which collects and learns from performance monitoring data. The controller determines whether retraining is necessary by comparing post-restoration data to a threshold range derived from average values and allowable errors, thereby reducing unnecessary retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all AI models of optical transmission apparatuses are retrained after abnormality restoration, then the detection accuracy is maintained, but the retraining cost increases significantly
Solution Approach 1:
The patent changes the parameter of retraining frequency from 'all models always retrained' to 'selective retraining based on PM data threshold comparison'. By comparing post-restoration PM data against pre-abnormality baseline data and applying a threshold criterion, the system determines whether retraining is necessary, thus reducing unnecessary retraining operations while maintaining detection accuracy when needed.
2Adaptability or versatility
If AI models are retrained frequently after restoration, then the system adapts to changes, but the operational complexity and time consumption increase
Solution Approach 1:
The patent introduces a threshold parameter for PM data comparison that determines when retraining should occur. This parameter change transforms the retraining process from a frequent, time-consuming operation to a selective process that only occurs when actual significant changes are detected, thereby reducing time loss while preserving necessary adaptability.
3Loss of energy
If selective retraining is performed based on PM data comparison, then the retraining cost is reduced, but the complexity of the determination process increases
Solution Approach 1:
The patent segments the retraining determination process into distinct functional modules: a data acquisition unit that collects PM data, a comparison unit that compares post-restoration data with baseline data, a threshold judgment unit that determines whether retraining is needed, and a retraining execution unit. This segmentation manages complexity by organizing the determination process into manageable, independent components with clear interfaces.
Solution Approach 2:
The patent implements a feedback mechanism where post-restoration PM data is compared against baseline data, and the result feeds into the retraining decision. This closed-loop feedback structure automates the determination process, reducing the need for complex manual evaluation while maintaining systematic control over when retraining occurs.
Data Source
AI summary
A controller: collects PM data of each of a plurality of optical transmission apparatuses provided in an optical transmission network; learns, with a plurality of AI models provided for each of the plurality of optical transmission apparatuses, a variation in time series of the PM data of the optical transmission apparatus; stores an average value and an allowable error of the PM data of the optical transmission apparatus in a database for each of the plurality of optical transmission apparatuses; detects an abnormality within the optical transmission network by using the plurality of AI models; and, in a case where there is an optical transmission apparatus in which the PM data after restoration from the abnormality are outside a threshold range, which is derived from the average value and the allowable error, determines that an AI model of the relevant optical transmission apparatus needs retraining.