Oil refining may be regarded as a complex hierarchical system based on large-scale processes differing in inertia and energy requirements and occurring in series and in parallel. The challenges of effectively controlling such processes are illustrated in the present work, and the role of neural network models in the automated control of oil refining is considered. An iterative dynamic model is proposed for the development of classifications in a prototype-based control system. The following hypothesis is verified experimentally: that maximum reliability of associative prediction by a Hopfield neural network may be achieved by adjusting the size of the input word that encodes the values of the observable process parameters.
S.Yu. Tyryshkin (Wed,) studied this question.