A Model-Free Control Strategy Based on Artificial Neural Networks for PeWEC

This paper introduces a model-free control strategy aimed at maximizing the power absorbed by a Pendulum Wave Energy Converter (PeWEC). This control strategy is based on the development of a metamodel and on the optimization of the control action through it. The metamodel is built only from the collected data by linking the applied control action with an artificial neural network, which in this case is represented by a damping coefficient, and the faced sea-state parameters with the average absorbed power experienced with that configuration. To manage properly the choice between actions aimed at developing a sufficiently precise metamodel (exploration) and at maximizing the absorbed power (exploitation) a greediness function has been designed and adopted. The developed strategy has been tested by simulating the working conditions of PeWEC for 4 weeks, adopting 14 different sea states, each one with significant height, energy period and probability of occurrence typical of the Mediterranean Sea. Finally, the influence on the learning process of the time length adopted for the applied control action has been analyzed.