The Chylla-Haase problem: a neural network controller

Presents the solution to an industrial challenge problem using neural networks. The control problem statement is authored by Chylla and Haase (1993). The objective is to maintain reaction temperature of a semibatch polymerization reactor at a set point inspite of a variety of conditions ranging from raw material impurity, product variety, repeated batching and varying ambient temperature. Although the reactor configuration and control requirements appear simple, the realities of the problem make it a challenging problem to tackle. The controller design methodology is based on process inverse dynamics modelling. The learning database for the controller training is generated in an open loop fashion and training of the network is carried out off-line by considering the future plant outputs as the reference set points. The learning procedure is not goal directed in that an explicit goal is not defined. This paper follows the design of the controller