A recurrent neuronal approach for the nonlinear discrete time output regulation
Jorge M. O. Henriques, Paulo Sousa Gil, Antonio Carlos Dourado, Bernardino Castillo–Toledo, André Titli · 2002
The combination of a recurrent neural network with the output regulation control theory is proposed so that a robust controller for general nonlinear discrete time systems is obtained. It is intended with this approach to profit from the identification capabilities of neural networks with the stability properties of the output regulation theory. Given the universal approximation properties, a recurrent neural network is applied for modelling nonlinear systems. Learning is implemented online, based on input-output data, ensuring that the learning error converges to zero. Due to the combination of the adaptive neural learning procedure aspect and the output regulator technique the proposed control scheme behaves with strong robustness with respect to unknown dynamics and nonlinear characteristics. To solve the regulator equations a new iterative procedure is presented. The proposed algorithm, based on the recurrent neural network ensures the convergence of the regulator equations. Experimental results collected from a laboratory heating system confirm the viability and effectiveness of the proposed methodology.