摘要：AbstractThe key idea of the fourth industrial revolution is to use the huge amount of data from the increased process digitalisation in order to make better decisions at all levels: from the design and control, to operation and management. However, advanced decision support systems usually rely on good plant models. Despite the increased popularity of machine learning, in the process industry many of these approaches may fail in building reliable prediction models: that is, models whose output can be trusted even out of the region where actual data was collected. This paper illustrates how to get a reliable grey-box model of a chemical plant for optimisation purposes via sum-of-squares (SOS) constrained regression, a method that guarantees full enforcement of physical features on the identified model, no matter the quality and quantity of the collected data. The approach is used here to identify a reliable model for the reaction kinetics in a hybrid CSTR, a pilot plant where the chemical reactions are emulated over a harmless fluid.
