Electric Power Components and Systems · 2023 · 12 citations · 25 references
ComputingMathematicsMachine LearningOptimizationRenewable Energy SystemsAdaptive ControlDeep LearningOptimal ControlReinforcement LearningWind Energy SystemsWind Turbine TechnologyControl System OptimizationControl System ReliabilityControl System RobustnessDeep Reinforcement LearningFarm ResiliencePolicy GradientsResilienceTurbine Control Systems
Today’s Wind Turbines (WT)s are loaded with high-tech devices that help enhance energy harvest while minimizing stress on the structure. Yet, modern WT control systems have competing goals (stability and dependability) that need the use of cutting-edge control approaches capable of addressing multi-objective difficulties. This method proposes adaptive control solutions for Wind Turbines (WTs) using deep reinforcement learning (DRL) to optimize electricity production. The innovation of this study lies in the adaptation of the optimum policy of a training RL agent to derive the probabilistic feature of WT speed, which addresses the multi-objective difficulties of stability and dependability in modern WT control systems. Additionally, a reward regularization A module has been created to calculate the normalized power outputs of wind turbines (WTs) in different yaw configurations and varying environmental conditions, providing high resilience and flexibility to the proposed control system. The combination of this module with the policy gradient’s deep determinism (DDPG) method using multiple agents helps determine the ideal yaw settings for each WT in the farm. The suggested approach improves upon the standard DDPG implementation in MATLAB, contributing to the advancement of WT control strategies for efficient power production in complex networks.
25