Inversion control of nonlinear systems with neural network modelling

Abstract

A method of controlling certain types of nonlinear dynamical systems whose dynamics can be modelled by a multilayer neural network is proposed. The control algorithm assumes that the plant equations are not known but the dimension of the system is known, The control input is derived by inversion of a forward neural network via the Newton Raphson method. During inversion of the multilayer neural network some optimal control senses are resolved. To suppress the control error due to the modelling error of the forward neural network, the inversion controller with a conventional feedback controller is proposed, which provides a better performance than a pure inversion controller. The proposed algorithm shows various advantages, and computer experiments on a bioreactor prove the effectiveness of this algorithm.