Wind Power Prediction and Pitch Angle Control Based on the Hammerstein System

This paper presents a wind power prediction and pitch angle control using a convolutional neural networks (CNN)-based Hammerstein system. This system consists of a static nonlinear module and a dynamic linear module, where the nonlinear part is represented by a CNN and the linear part by an auto regressive with exogenous input (ARX) model. The two module parameters are identified using a specialized test signal combining Gaussian and random signals. First, correlation analysis is applied to Gaussian signal data to estimate the linear module. Then, the Adam optimization method is used to determine the nonlinear module based on random signal data. A pitch control strategy for wind power systems is developed using this model. By leveraging the distinct linear and nonlinear structures, the nonlinear component’s inverse is derived, enabling a generalized control model with linear characteristics. This reformulates the nonlinear pitch control problem into a linear control framework. Simulation results confirm the effectiveness of the proposed method in identifying the CNN-based Hammerstein model, accurately predicting wind power, and enhancing pitch control performance.