On-line optimization of batch processes using neural networks coupled with optimal state feedback

In most practical applications in batch processing, the optimal operating strategy consists of nonsingular regions where the manipulated input hits a bound (upper or lower) and singular regions where the manipulated input has a value between the upper and lower bound. In this paper, we demonstrate the use of artificial neural networks to identify online the switching times between singular and nonsingular regions. This information on switching times will be coupled with the optimal state feedback laws developed in Palanki et al. (1993) in the singular region to provide the complete solution to end-point optimization of batch reactors. As an illustrative example of this online methodology, we will consider the yield optimization of alcohol in a semi-batch reactor.