Adaptive RIS Optimization for Secure Underwater Optical Communications

Underwater Optical Wireless Communication (UOWC) systems are subject to significant degradation due to dynamic salinity gradients, temperature variations, and turbulence-induced noise, which distort light propagation and compromise data integrity. To mitigate these challenges, we propose an adaptive framework integrating: Long Short-Term Memory (LSTM)-based environmental prediction, Deep Reinforcement Learning (DRL)-based control, and a Non-dominated Sorting Genetic Algorithm II (NSGA-II) multi-objective optimization for real-time reconfiguration of Reconfigurable Intelligent Surface (RIS). The LSTM model forecasts underwater conditions with a mean absolute error (MAE) of 0.008 PSU, facilitating proactive system adaptation. Experimental evaluations demonstrate that the system achieves a 41.4% reduction in BER, reaching a minimum of $2.3 \,\, {\times }{10}^{-5}$ compared to the static RIS baseline, a secrecy rate of 8.2 bps/Hz, 22.2% energy savings, and low reconfiguration latency of 47.3 ms, ensuring operation within the channel coherence time. These results set a new performance benchmark for secure and efficient UOWC systems, with applications in underwater surveillance, environmental monitoring, and offshore communication networks.