ARTIFICIAL INTELLIGENCE IN THE ANALYSIS OF VOLS TRAFFIC TO IDENTIFY ANOMALIES
Modern fiber-optic communication lines are the fundamental basis of digital infrastructure, however, their operation is associated with growing requirements for reliability, bandwidth and protection from physical and cyber threats, which makes traditional threshold and statistical monitoring methods insufficiently effective for timely detection of complex, multifactorial anomalies in conditions of high-speed data transmission and dynamically changing loads. This article examines in detail the use of artificial intelligence technologies, in particular machine learning and deep learning algorithms, for the intelligent analysis of network traffic and optical fiber metrics in order to automate the detection of abnormal conditions, including hidden component degradation, microfractures, unauthorized interference, attacks on the management level and routing anomalies that are difficult to formalize using classical rules. A multi-level architecture of the AI system is proposed, covering aggregation and normalization of telemetry data from OTDR, spectral analyzers and network sensors, extraction of time, frequency and correlation features, training of ensemble and neural network models (autoencoders, recurrent and transformer architectures, isolation methods) and adaptive classification of incidents in real time with support for online learning and additional training on new patterns.