Controller tuning and nonlinear compensation based on multilayer neural networks

This paper presents a controller tuning method for nonlinear systems by using the so-called virtual reference feedback tuning method and nonlinear compensation based on multilayer neural networks. To demonstrate the effectiveness of the presented method, simulation is carried out for systems with nonlinearities such as hysteresis, cubic functions, and dead zones. Moreover, architectures and computational efficiency of neural networks are investigated through a widely-used deep learning framework.