Neural Network Controllers in Chemical Technologies

Abstract

Chemical technologies benefit greatly from optimization-based control strategies. This paper addressed the problem of substituting the optimization based controller with a neural network (NN). The NN-based controller offers several advantages, first it can be derived in an analytical form and second, it can make the closed-loop implementation tunable. It is not possible to incorporate these aspects into optimizationbased controller easily.The contribution is also to address the problem of the quality of the neural net, that approximates the control law. We show, which activation functions and structure yield the best approximation.