Active wave energy extraction by hydrofoil vessels through deep reinforcement learning-based flight control

The hydrofoil vessel is among the few types of ships known to demonstrate negative added resistance in waves. This paper investigates the potential for maximizing this effect through the development of a neural network-based flight control system trained via reinforcement learning to actively extract wave energy. Furthermore, it examines the impact of such a control system on passenger comfort, and whether a feasible trade-off can be made between efficiency and comfort. The results suggest that significant energy savings can be achieved through the proposed methodology, achieving a power reduction of 61.8 % compared to that in calm water, for a simplified model of a hydrofoil vessel operating in 1.5 m regular head waves of 5 s wave period. If optimizing for passenger comfort, a reduction in vertical accelerations of 98% can be achieved, however, at the cost of achieving virtually no power reduction as compared to calm-water operation. By using a reward function with weighted focus on comfort and energy efficiency, it was demonstrated that the presented methodologies can produce a good trade-off between these two performance indicators. The presented cases differ only in terms of the flight control software, demonstrating the profound potential for altering hydrofoil vessel performance through intelligent control. • A method for DRL-based hydrofoil flight control. • Significant negative added resistance in waves, < −60%. • Cancellation of wave-induced motions with no resistance penalty. • Effective balancing of efficiency and comfort. • Discussions of the physics of negative added resistance of hydrofoil vessels.