Combining prior knowledge with data-driven modeling of a batch distillation column including start-up

This paper presents the development of a simple model which describes the product quality and production over time of an experimental batch distillation column, including start-up. The model structure is based on a simple physical framework, which is augmented with fuzzy logic. This provides a way to use prior knowledge about the dynamics, which have a general validity, while additional information about the specific column behavior is derived from measured process data. The model framework is applicable for a wide range of columns operating under a certain control policy. The model framework for the particular column under study makes a priori assumptions about the specific behavior superfluous. In addition, a detailed description of the internal dynamics is not required, which reduces modeling effort. Three different hybrid model structures are compared; the model that uses the available sources of information most effectively can be used to simulate production including part of the start-up by applying constant quality control.