• We use state-of-the-art modeling of WTPS, and optimal control to obtain a data set of optimal control solutions. In particular, we utilize the mathematically rigorous direct optimal control approach (an approach that is based on the calculus of variations principles) to obtain the optimal control solutions. • We provide this optimal control solutions data set to the supervised learning algorithms, which are functional regression algorithms, as a training data set. This in turn allows the supervised learning algorithms to predict the optimal pitch angle control response to different wind speeds. • We present an AI-based framework for the pitch control problem of wind turbine power systems. We believe our work can be very useful in designing online pitch angle controllers, and can be extended to other control methods, such as the torque/tip ratio control of wind turbine systems. In this work, we propose a novel optimal control-based supervised learning framework for pitch control of wind turbine power systems. In this framework, we utilize a recent calculus of variations based optimal control formulation/method, which was developed based on a physically accurate nonlinear differential-algebraic model of wind turbine power systems to construct a supervised learning algorithm. That is, the optimal control solutions are used as the training data set for the supervised learning algorithm. We show the performance of the proposed framework when using both functional linear ridge regression and functional kernel ridge regression methods for the supervised learning algorithm. Multiple wind profiles, including real-world wind data , are used for testing the framework via simulations. Results show that, with minimal training data, and when using functional kernel ridge regression, the proposed framework is able to accurately predict the optimal pitch angle profile corresponding to the different test wind speed profiles, including the real-world wind data, hence enabling an online controller based on the proposed framework.
Abdelfattah et al. (Thu,) studied this question.