Nonlinear H∞ control for continuous-time recurrent neural networks
Johan A. K. Suykens, Joos P. L. Vandewalle, Bart De Moor · 1997
In this paper we investigate the nonlinear H∞control problem for recurrent neural network models connected to a recurrent neural network controller. Conditions for dissipativity with finite L2-gain are derived and expressed as matrix inequalities, based on a two-hidden layer recurrent neural network in standard plant form. The matrix inequalities are obtained from a storage function of quadratic form or quadratic form plus integral terms. Narendra's dynamic backpropagation procedure for training on a set of specific reference inputs is modified with a dissipativity condition.