A multiscale energy polarization attention mechanism and H-shaped network for AI-driven biomimetic oceanic energy farms

The worldwide shift towards renewable energy has generated considerable interest in tapping into the extensive potential of ocean energy resources. Nevertheless, current marine energy technologies face challenges stemming from inefficiencies caused by unpredictable conditions and low conversion rates. In contrast to previous research that has concentrated exclusively on hardware enhancements or utilized isolated machine learning (ML) models without closed-loop control, this study presents a fully integrated hierarchical cyber-physical system. We propose an Artificial Intelligence (AI)-Driven Biomimicry system that combines nature-inspired designs with a sophisticated hierarchical AI control network. This system includes two advanced components, which are (1) a Multiscale Energy Polarization Attention Mechanism (MEPAM) that effectively filters out turbulent noise to emphasize high-energy vortices, and (2) an H-Shaped Neural Network featuring a central bridging layer for explicit cross-domain integration of environmental and operational data, a configuration that is not found in standard single-encoder models. The framework was evaluated in three separate simulation protocols, which were surrogate prediction, closed-loop farm control, and fault prognostics. In the prediction protocol, the dual-branch surrogate achieved an R 2 of 0.98 and a Root Mean Squared Error (RMSE) of 0.04. In the closed-loop protocol, the Multi-Agent Proximal Policy (MAPPO) controller demonstrated a 53.7% increase in annual energy production compared to the traditional Wave Energy Converter (WEC) baseline from 15,100 MWh to 23,200 MWh, utilizing 100-harvester device model and environmental conditions. In the prognostic protocol, the model achieved an impressive F1 score of 0.96. Furthermore, it can predict mechanical failures up to 36 h in advance by analyzing vibrational signatures. Simulation results reveal a 40–60% enhancement in energy extraction efficiency relative to conventional systems, alongside a 35% reduction in maintenance costs. Potential applications include offshore power grids, energy supplies for remote islands, and maritime monitoring systems.