Real-time optimization of wind farm energy capture for below rated wind speed is critical for reducing the levelized cost of energy (LCOE). Performance of model based control and optimization techniques can be significantly limited by the difficulty in obtaining accurate turbine and farm models in field operation, as well as the prohibitive cost for accurate wind measurements. The Nested-Loop Extremum Seeking Control (NLESC), recently proposed as a model free method has demonstrated its great potential in wind farm energy capture optimization. However, a major limitation of previous work is the slow convergence, for which a primary cause is the low dither frequencies used by upwind turbines, primarily due to wake propagation delay through the turbine array. In this study, NLESC is enhanced with the predictor based delay compensation proposed by Oliveira and Krstic [1], which allows the use of higher dither frequencies for upwind turbines. The convergence speed can thus be improved, increasing the energy capture consequently. Simulation study is performed for a cascaded three-turbine array using the SimWindFarm platform. Simulation results show the improved energy capture of the wind turbine array under smooth and turbulent wind conditions, even up to 10% turbulence intensity. The impact of the proposed optimization methods on the fatigue loads of wind turbine structures is also evaluated.
A Distributed Reinforcement Learning Yaw Control Approach for Wind Farm Energy Capture Maximization*
