A Comparative Study on the Accuracy of Neural Network-Based Fast Fitting Methods for Wind Turbine NMPC

Based on the strong learning capability of neural networks (NN), this paper proposes a fitting technique with fast response characteristics. The study applies the characteristics of neural networks to fitting the output of nonlinear model predictive control (NMPC), aiming to explore solutions to the difficulties in modeling large-scale engineering equipment and dealing with complex nonlinear characteristics, thereby improving the efficiency of equipment operation and maintenance. Conducted in the context of NMPC for maximum wind energy capture of wind turbines, the study verifies the fitting performance of NN by comparing the predictive outputs of NMPC. Meanwhile, through the NN-NMPC strategy, it investigates the impacts of different prediction horizons and the number of training feature values on the fitting accuracy of NN. The study also compares the static data processing capabilities of different network models, providing data-driven support for the selection of optimal models. This study proposes a NN-based fast fitting technique, providing a new method for reducing the computational burden of modeling and improving the prediction accuracy of key nonlinear features.