Digital twins, AI and autonomous optical networks

Modern optical transport networks must ensure error-free transmission under dynamic conditions and heterogeneous ROADM-based infrastructures. Maintaining high quality of transmission (QoT) is challenging due to interdependent launch power settings across optical multiplex sections, where naïve adjustments may induce transient signal-to-noise ratio (SNR) degradation. Increasing system complexity further motivates the integration of physical-layer modeling with intelligent automation. This thesis presents proof-of-concept contributions toward autonomous optical network control and management by combining digital twin (DT) technology with AI-based approaches. First, DT-enabled multi-step lookahead strategies are proposed for autonomous power equalization. With monitoring data, the DT predicts QoT before applying configuration changes. By evaluating sequences of control actions with fixed and adaptive step sizes, DT-enabled closed-loop control mitigates QoT degradation during intermediate steps and avoids suboptimal convergence, while DT-assisted parallel configuration improves efficiency. Experiments on a commercial ROADM based optical network testbed demonstrate fast and safe network reconfiguration without intermediate SNR margin degradation. Second, an AI-agent-based framework is developed for intent-driven control and management. A control agent translates high-level intents into APIs calls for automated configuration, while a management agent leverages retrieval-augmented generation to analyze alarms and support root-cause analysis. Beyond task automation, the proposed agent architecture illustrates how AI agents can coordinate decision-making, integrate domain knowledge, and support the evolution toward autonomous optical network operation. Overall, this work demonstrates the feasibility of combining DT-based closed loop control with AI-agent-based automation, providing a foundation for future autonomous optical networks.