Self-tuning Yaw Control Strategy of a Horizontal Axis Wind Turbine Based on Machine Learning

The design procedure of a Machine Learning (ML) based yaw control strategy for a Horizontal Axis Wind Turbine (HAWT) is presented in the following chapter. The proposed yaw control strategy is based on the interaction of three different Artificial Intelligence (AI) techniques to design a ML system: Reinforcement Learning (RL), Artificial Neural Networks (ANN) and metaheuristic optimization algorithms. The objective of the designed control strategy is to achieve, after a training stage, a fully autonomous performance of the wind turbine yaw control system for different input wind scenarios while optimizing the electrical power generated by the wind turbine and the mechanical loads due to the yaw rotation. The RL algorithm is known to be able to learn from experience. The training process could be carried out online with real-time data of the operation of the wind turbine or offline, with simulation data. The use of an ANN to store the data of the matrix Q(s, a) related to the RL algorithm eliminates the large scale data management and simplifies the operation of the proposed control system. Finally, the implementation of a metaheuristic optimization algorithm, in this case a Particle Swarm Optimization (PSO) algorithm, allows calculation of the optimal yaw control action that responds to the compromise between the generated power increment and the mechanical loads increase due to the yaw actuation.