Enhancing Reinforcement Learning-Based Control of Wave Energy Converters Using Predictive Wave Modeling (Papers Track)
Vineet Gundecha (Hewlett Packard Enterpise); Arie Paap (Carnegie Clean Energy); mathieu Cocho (Carnegie Clean Energy); Sahand Ghorbanpour (Hewlett Packard Enterprise); Alexandre Pichard (Carnegie Clean Energy); Ashwin Ramesh Babu (Hewlett Packard Enterprise Labs); Soumyendu Sarkar (Hewlett Packard Enterprise)
Abstract
Ocean wave energy is a reliable form of clean, renewable energy that has been under-explored compared to solar and wind. Wave Energy Converters (WEC) are devices that convert wave energy to electricity. To achieve a competitive Levelized Cost of Energy (LCOE), WECs require complex controllers to maximize the absorbed energy. Traditional engineering controllers, like spring-damper, cannot anticipate incoming waves, missing vital information that could lead to higher energy capture. Reinforcement Learning (RL) based controllers can instead optimize for long-term gains by being informed about the future waves. Prior works have utilized incoming wave information, achieving significant gains in energy capture. However, this has only been done via simulated waves (perfect prediction), making them impractical in real-life deployment. In this work, we develop a neural network based model for wave prediction. While prior works use auto-regressive techniques, we predict waves using information available on-device like position, acceleration, etc. We show that replacing the simulated waves with the wave predictor model can still maintain the gain in energy capture achieved by the RL controller in simulations.