Control of nonlinear dynamical systems modeled by recurrent neural networks

Abstract

An integrated methodology was presented, for the modeling and controller design of nonlinear dynamical systems. The methodology, which comprises three steps (model the nonlinear plant using an RNN (recurrent neural network); exact-linearize the nonlinear RNN; design a linear controller for the exact-linearized model, and implement it on the real plant), was tested on a CSTR (continuous stirred tank reactor) and shown to perform better than a linear, optimally-tuned controller. A number of theoretical issues remain to be investigated, most notably robust stability and performance