LMI-Based Design of a Robust Model Predictive Controller for a Class of Recurrent Neural Networks With Guaranteed Properties

This work proposes a novel robust nonlinear model predictive control (NMPC) algorithm for systems described by a generic class of recurrent neural networks. The algorithm enables tracking of constant setpoints in the presence of input and output constraints. The terminal set and cost are defined based on linear matrix inequalities to ensure convergence and recursive feasibility in presence of process disturbances. Simulation results on a quadruple tank nonlinear process demonstrate the effectiveness of the proposed control approach.