Experimental Validation and Analysis of Deep Reinforcement Learning Control for Wave Energy Converters

Zou, S.; Subramanian, A.; Bosma, B.; Lou, J.; Beringer, C.; Robertson, B.; Lomonaco, P. (2025). Experimental Validation and Analysis of Deep Reinforcement Learning Control for Wave Energy Converters [Presentation]. Presented at OREC/UMERC 2025 Conference, Corvallis, United States of America.

@conference{Zou-2025-24631,
author = {Zou, S and Subramanian, A and Bosma, B and Lou, J and Beringer, C and Robertson, B and Lomonaco, P},
title = {Experimental Validation and Analysis of Deep Reinforcement Learning Control for Wave Energy Converters},
year = {2025},
month = {aug},
series = {OREC/UMERC 2025 Conference},
pages = {17},
address = {Corvallis, United States of America},
url = {http://umerc2025.exordo.com/programme/presentation/86},
keywords = {Wave, Lab Data, Control},
}

Export Citation to BibTex

TY - CONF
TI - Experimental Validation and Analysis of Deep Reinforcement Learning Control for Wave Energy Converters
AU - Zou, S
AU - Subramanian, A
AU - Bosma, B
AU - Lou, J
AU - Beringer, C
AU - Robertson, B
AU - Lomonaco, P
T2 - OREC/UMERC 2025 Conference
C1 - Corvallis, United States of America
AB - This paper presents the real-time implementation and experimental validation of a Deep Reinforcement Learning (DRL) control strategy for a Wave Energy Converter (WEC). Wave energy technology is still underdeveloped. One important challenge is the inefficient Power Take-Off (PTO) control technologies. To address this challenge, the research team has developed a DRL control which shows promising performance in wave power production and outperforms existing model-based controls on paper. However, the practical performance of this control remains unknown. Therefore, the team at Michigan Technological University (MTU) collaborated with Oregon State University (OSU) in testing the control performance in the wave flume. The Laboratory Upgrade Point Absorber (LUPA) WEC is utilized for control demonstration. Varied wave tank tests have been conducted, including both regular and irregular wave tests. The testing results suggest: (1) the real-time implementation of the DRL control is relatively straightforward with a very small computational cost; (2) the robustness of the control is a challenging problem in practice, which can be addressed by incorporating significant randomness in training; (3) the control shows promising performance in practice. Specifically, the power produced in regular waves is close to the practical maximum, and the power produced in irregular waves outperforms an existing control.
DA - 2025/08//
PY - 2025
SP - 17
UR - http://umerc2025.exordo.com/programme/presentation/86
LA - English
KW - Wave
KW - Lab Data
KW - Control
ER -

Export Citation to RIS

Abstract

This paper presents the real-time implementation and experimental validation of a Deep Reinforcement Learning (DRL) control strategy for a Wave Energy Converter (WEC). Wave energy technology is still underdeveloped. One important challenge is the inefficient Power Take-Off (PTO) control technologies. To address this challenge, the research team has developed a DRL control which shows promising performance in wave power production and outperforms existing model-based controls on paper. However, the practical performance of this control remains unknown. Therefore, the team at Michigan Technological University (MTU) collaborated with Oregon State University (OSU) in testing the control performance in the wave flume. The Laboratory Upgrade Point Absorber (LUPA) WEC is utilized for control demonstration. Varied wave tank tests have been conducted, including both regular and irregular wave tests. The testing results suggest: (1) the real-time implementation of the DRL control is relatively straightforward with a very small computational cost; (2) the robustness of the control is a challenging problem in practice, which can be addressed by incorporating significant randomness in training; (3) the control shows promising performance in practice. Specifically, the power produced in regular waves is close to the practical maximum, and the power produced in irregular waves outperforms an existing control.