Reinforcement learning based control for arrays of point absorber wave energy converters

The thesis contains a review of and background material relating to point absorber wave energy converters. These are devices that have small dimensions compared to the incident wavelength, and are axisymmetric in design. In most cases, energy is primarily associated with heave motion. A standard point absorber design was modelled with a PTO that reacted against the seabed. Wave energy is strongly absorbed close to the natural resonant frequency of the point absorber, although outside this resonance wave energy absorption is poor. Performance can be improved using a control system to select different values of damping acting on the point absorber through the PTO. The selection of values can be achieved using reinforcement learning algorithms, such as Q-learning. Performance of the control system was evaluated through simulation of the point absorber. For regular waves, performance could be gauged against an analytical solution, and for irregular waves a value for optimal damping was determined through simulation. Q-learning-initiated control resulted in convergence with analytically and simulation-derived optimal damping values. The power generated in the simulation system using Q-learning was close to the maximum value realisable using passive damping, suggesting that it was an effective control system. Q-learning performance was found to be dependent on unlearnt parameters, termed hyperparameters. It was shown that particle swarm optimisation could be used for the selection of hyperparameters. Point absorbers can be deployed in arrays. Array models were developed to evaluate Q-learning for the control of multiple point absorbers. A characteristic of the array is that elements are hydrodynamically coupled. The application of single-agent and multi-agent Q-learning control was investigated, and both were found to be close in terms of their performance. Proof of principle was established for the single-agent control of four point absorbers.<p></p>

Author dc:creator

• Xuxin Pooley (21041339)

• Renewables
• Reinforcement Learning
• Wave Energy

Statement dc:rights

• All rights reserved
• Open Access after 2027-07-27

Identifier

10779/exe.33078968.v1

OAI identifier oai:identifier

oai:figshare.com:article/33078968