Classical vs. neural network control: an industrial application

A comparison between a conventional PI control vs. a multilayered neural network (MNN) scheme over an industrial process is presented. The process is a distillation column which has a conventional temperature feedback control loop. Also taken under consideration is the first order dynamic of a thermowell with which the temperature of the third stage of the column is measured. The control action takes place 5 minutes after startup of the process. Two kinds of networks were used: one with three layers and another with four layers, all trained under the back-propagation learning algorithm.>