Energy capture enhancement in wind turbine systems using neural prediction and model predictive yaw control

Wind prediction and yaw control strategies in large-scale wind turbine systems are often constrained by inaccurate wind direction estimates, leading to yaw misalignment, reduced aerodynamic efficiency, increased control effort, and lower energy capture. To address these challenges, this study proposes an integrated framework that combines learning-based wind direction prediction with a predictive yaw control. First, a residual multilayer perceptron-based predictor is developed to reduce the higher prediction errors observed in conventional feed-forward neural network-based predictors due to limited accuracy and generalization under wind variations. Comparative analysis demonstrates that the proposed Residual multilayer perceptron achieves significantly lower centered root mean square error than feedforward neural network and two long short-term memory models across multiple datasets. Second, a continuous set model predictive yaw control scheme is formulated to reduce delayed yaw orientation and large transient yaw errors by the conventional yaw control methods due to their slow corrective action and lack of explicit constraint handling during rapid wind changes. Third, the wind direction predictions are integrated with the model predictive yaw control framework and validated on a detailed system-level 4.8 MW benchmark wind turbine model in MATLAB/Simulink. Simulation results show that the proposed integrated framework consistently outperforms proportional-integral and existing MPYC approaches. Finally, energy extraction analysis confirms improved aerodynamic and electrical efficiencies, demonstrating the robustness of the proposed method.