Simple recurrent neural network: A neural network structure for control systems

This paper describes: (i) a neural network structure referred as simple rec urrent neural network (SRNN), (ii) a learning algorithm based on the optimi zation layer by layer method for the SRNN and (iii) the SRNN applications f or controlling both a linear and a nonlinear dynamic system. The main objec tives of the SRNN described are: (i) to show the capability for modeling in verse relationships of dynamic systems and (ii) to show that the SRNN needs just a few parameters in order to control dynamic systems. A comparison be tween the SRNN and some other neural network structures, based on the numbe r of parameters and on the number of time steps required for controlling of dynamic systems, is given. (C) 1998 Elsevier Science B.V. All rights reser ved.