研究目的
Investigating the application of statistical machine learning techniques for localizing single-link failures in transparent optical networks to reduce the Mean-Time-To-Repair (MTTR).
研究成果
The proposed fault localization scheme achieves high accuracy (up to 99% for a network load of 20 Erlangs) without requiring extra monitoring equipment. It effectively reduces the MTTR and human effort for fault localization in transparent optical networks.
研究不足
The approach's accuracy is tested on lightly-loaded networks (7 to 20 Erlangs), and its performance in heavier loads or larger networks is not evaluated. Practical feasibility issues, such as data collection time and model scalability, are noted for future work.
1：Experimental Design and Method Selection:
The study employs a Gaussian Process (GP) classifier trained on historical network failure data to model and predict link failure probabilities. A Graph-Based Correlation (GBC) heuristic is used to limit suspect links.
2：Sample Selection and Data Sources:
Data is generated for an OFDM-based elastic optical network, simulating dynamic connection requests and injecting link failures sequentially.
3：List of Experimental Equipment and Materials:
The network topology includes links with distances ranging from 600 km to 2800 km, simulated under different traffic loads (7, 10, and 20 Erlangs).
4：Experimental Procedures and Operational Workflow:
Link failures are injected into a dynamic Routing and Spectrum Allocation (RSA) system, with network state inspected to identify affected and unaffected paths.
5：Data Analysis Methods:
The GP classifier's performance is evaluated based on its accuracy in identifying failed links, with results compared across different network loads.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
