High-precision fault management in optical fiber communication systems based on artificial intelligence

Abstract The delivery of high-speed internet services heavily depends on stable and secure data transmission provided by optical fiber infrastructures. Therefore, promptly addressing fiber anomalies – whether deliberate actions like optical tapping or physical issues such as fiber cuts – is essential to ensure service continuity and network robustness. If left unresolved, these disruptions can compromise network stability, cause substantial financial damage, expose sensitive data, and progressively degrade performance. To mitigate these risks, there is a pressing need for intelligent systems capable of autonomously detecting, identifying, and pinpointing faults without relying on manual inspection of OTDR (optical time-domain reflectometry) traces. This study introduces an advanced machine learning-based solution for analyzing OTDR signals during fault recovery. It integrates a semi-supervised anomaly detection ensemble to uncover both previously known and novel faults and employs a multitask BiLSTM architecture enhanced with an attention mechanism to accurately locate and classify the type of fiber issues. The semi-supervised model ensures robust performance even with limited labeled samples. Tests conducted on real OTDR datasets demonstrate high effectiveness, achieving an anomaly detection F1-score of 98.78 % and fault classification accuracy of 98.62 %, along with precise localization. Implementing this approach can greatly enhance the fault tolerance of optical networks, reducing downtime and preserving data security in the face of both attacks and accidental damages.