Intelligent Control of a Wind Turbine based on Reinforcement Learning

Abstract

Advanced controllers of modern wind turbines can help increase energy capture efficiency and reduce structural loading. However, to fulfill the modern wind turbine control demands with contradicting requirements (efficiency and reliability), innovative control methods that can handle multi-objective problems are inevitable. In this paper, we propose adaptive control techniques, which trying to extract the stochastic property of wind speed using a trained reinforcement learning (RL) agent and then apply their obtained optimal policy to the wind turbine adaptive control design. This paper includes the results of tests of the developed torque controller and a multi-input-multi-output (MIMO)-based controller using advanced RL algorithms.