Multilayer neural network controllers for multivariable dynamic systems

This paper presents multilayer neural networks to control a class of multivariable dynamical systems. The controlled system is represented by two neural networks. The first is used for system emulating, that is, input-output mapping, and the second produces the control signals required to drive the plant at hand to a desired state. Different learning architectures are proposed for these purposes. An aircraft control problem is carried out to demonstrate the design procedure. Simulation results show the potential applicability and practical feasibility of the developed schemes.