Combining First Principles Models and Neural Networks for Generic Model Control

Abstract

Generic Model Control (GMC) is a control algorithm capable of using non-linear process model directly. In GMC, mostly, first-principles models derived from dynamic mass, energy and momentum balances are used. When the process is not perfectly known, the unknown parts of first principles models can be represented by black-box models, e.g. by neural networks. This paper is devoted to the application of such hybrid models in GMC. It is shown that the first principles part of the hybrid model determines the dominant structure of the controller, while the black-box elements are used as state and/or disturbance estimators. The sensitivity approach is used for the identification of the neural network elements of the control-relevant hybrid model. The underlying framework is illustrated by the temperature control of a continuous stirred tank reactor (CSTR) where a neural network is used to model the heat released by an exothermic chemical reaction.