First Approximation of Application of Federated Learning to Wind Turbines

Abstract

This work investigates the application of FederatedLearning techniques to reduce the training time of wind turbine power stabilization controllers on a wind farm. A Reinforcement Learning controller based on Q-learning is implemented and the results of the individual controller are compared with a system of 4 wind turbines using Federated Learning. The simulation results show how this technique significantly improves the convergence time of the controller when compared to control strategies without federated learning. The preliminary results demonstrate how Federated Learning has great potential for improving theeffectiveness of wind turbine controllers while maintaining the privacy and security of their operational data.