A neural network based auto-tuned regulator

A novel auto-tuning regulation scheme is presented, based on the ability of the multi-layer perceptron to perform as a pattern recognisor. To facilitate the rapid training of such a complex network a parallel version of the Broyden-Fletcher-Goldfarb-Shanno optimisation based learning algorithm is used. The principle of this auto-tuning technique is demonstrated successfully for a number of linear second-order example plants. Finally this technique proves successful for the auto-tuned regulation of a nonlinear CSTR chemical plant, over the complete operating range.