A Machine Learning Approach for Model-free Control of PeWEC

Among the devices aimed to the conversion of the wave energy source, PeWEC (Pendulum Wave Energy Converter) stands out for his ability to reliably absorb power by exploiting the relative rotation between its hull pitching and the oscillation of an internal pendulum. The conversion process is performed by the PTO (Power Take-Off), a generator aimed of damping the pendulum motion. By regulating the PTO torque, PeWEC can adapt to different sea states and maximize the absorbed power. However, the control adopted for the definition of this action follows a model-based approach, affected by uncertainties and unable to conform to system changes. To solve these issues, this work presents a model-free control strategy that is based on learning a metamodel from real data and optimizing its action through it.