In the quest to harness the power of ocean waves, a team of researchers from the University of Exeter, including Yi Zhan, Iván Martínez-Estévez, Min Luo, Alejandro J. C. Crespo, and Abbas Khayyer, has made significant strides in improving the efficiency of wave energy converters. Their work, published in the journal Renewable Energy, focuses on enhancing the performance of point absorber wave energy converters, a technology that has shown great promise in capturing wave energy.
The team tackled the challenge of optimizing the power take-off (PTO) system, which is crucial for converting wave motion into electrical energy. The complexity lies in the nonlinear interactions between the waves and the device’s motion. To address this, the researchers developed an advanced control model that combines Smoothed Particle Hydrodynamics (SPH) with multi-agent deep reinforcement learning. This innovative framework allows for real-time communication between high-fidelity SPH simulations and intelligent control agents. These agents learn coordinated policies to maximize energy capture, receiving instantaneous hydrodynamic states from the SPH environment and outputting continuous damping actions. The reward system is based on power absorption, incentivizing the agents to optimize their actions.
The researchers employed the Multi-Agent Soft Actor Critic algorithm within a centralised-training and decentralised-execution scheme. This approach ensures stable learning in continuous, multi-body systems. The entire platform was implemented in a unified GPU-accelerated C++ environment, enabling long-horizon training and large-scale three-dimensional simulations. The effectiveness of the learned control policy was validated through various benchmark cases under regular and irregular wave conditions. Compared to constant PTO damping, the learned control policy increased overall energy capture by 23.8% and 21.5%, respectively. These results highlight the strong potential of intelligent control for improving the performance of wave energy converter arrays.
The developed three-dimensional GPU-accelerated multi-agent platform is not only a significant advancement in computational hydrodynamics but also extendable to other fluid-structure interaction engineering problems that require real-time, multi-body coordinated control. This research paves the way for more efficient and intelligent wave energy converters, bringing us closer to harnessing the full potential of ocean wave energy.
Source: Zhan, Y., Martínez-Estévez, I., Luo, M., Crespo, A. J. C., & Khayyer, A. (2023). Coupling Smoothed Particle Hydrodynamics with Multi-Agent Deep Reinforcement Learning for Cooperative Control of Point Absorbers. Renewable Energy, 205, 1186-1202.
This article is based on research available at arXiv.