Ocean waves carry more energy per unit area than wind or solar, and wave energy has significant global potential to help meet clean-energy demand. Yet wave energy converters (WECs) remain far from mainstream commercial deployment, largely because it is difficult and costly to control them well enough to extract energy efficiently and safely.
A key obstacle is that optimal control of a WEC is inherently noncausal. To extract the most energy, a controller needs to know what the waves will do a few seconds into the future, in addition to their current state. Existing advanced control methods, such as model predictive control, can use this future information, but they require an accurate mathematical model of the device's hydrodynamics. Such models are especially hard to obtain for emerging soft-body WEC concepts, including dielectric elastomer generators and dielectric fluid generators, whose flexible structures deform and interact with water in complex, nonlinear ways.
To avoid this modelling difficulty, a research team from University College London (UCL), the University of Southampton and the University of Nottingham used reinforcement learning (RL), an approach in which a controller learns directly from experience with the environment. In a new study published in Ocean, the team combined proximal policy optimization (PPO), a widely used and stable RL algorithm for continuous control, with a noncausal control scheme driven by short-term wave predictions.
"Wave energy converters, especially the new generation of soft, flexible devices, are notoriously difficult to model precisely. Our framework lets the controller learn how to respond directly from the device's own behaviour and from short-term wave forecasts, so it doesn't need that precise model at all," said Yao Zhang, corresponding author of the study and a researcher in the Department of Mechanical Engineering at UCL.
The team trained a PPO agent to adjust the control gains of a point-absorber WEC, a common benchmark device, in real time, based on the buoy's motion and forecasts of the incoming waves one, three or five steps ahead. They tested the trained controller against a conventional damping-based, causal controller, using 200 seconds of real, irregular wave-height data gathered off the coast of Cornwall, UK, that had not been used during training.
The results showed a clear benefit from giving the controller a longer look-ahead at the waves. With a five-step wave prediction horizon, the PPO controller generated up to 13.9% more energy than the conventional baseline controller, while keeping the device's motion and actuator force within their safety limits throughout. The longer-horizon controllers also finished training in fewer episodes, suggesting that access to wave forecasts can improve the efficiency of the learning process as well as the final control performance.
The researchers also tested how the controller coped with imperfect information, deliberately adding noise to the wave forecasts and introducing mismatches between the training and test environments. The trained agent continued to generate energy and operate stably even under these disturbed conditions, indicating that the approach is robust to the kind of prediction errors and model uncertainty that are unavoidable in real ocean deployments.
"This is, to our knowledge, the first time this kind of policy-gradient reinforcement learning has been combined with wave prediction for wave energy control. It's a promising, model-free method for controlling the next generation of hard-to-model wave energy devices," Zhang said.
The team next plans to test the controller on a hardware-in-the-loop dSPACE real-time platform to evaluate its computational and timing performance under realistic hardware constraints, and to extend the framework beyond mechanical power to account for the electrical power conversion stage of a full wave energy system.
D OI Link:
https://doi.org/10.26599/OCEAN.2026.9470015
About O cean
O cean is an international peer-reviewed journal that offers open access and serves as a multidisciplinary platform for the state-of-the-art research and practice in the domains of ocean science, technology, and engineering. The journal is dedicated to publishing articles, reviews and perspectives in these areas, with the goal of promptly disseminating and promoting theoretical, numerical, site-based, and experimental advancements in the context of global sustainability.
Ocean
Proximal policy optimization–based noncausal control for wave energy conversion systems
18-Jun-2026