Propagation of Interval Values in Simple Processing Networks

Process network optimization problems are prevalent in the processing industries. Streams of differing compositions undergo mixing and splitting steps to attain product streams of desired properties. Constraint propagation of system variable bounds across a process network allows identification of system capabilities prior to optimization, and expedites solution of the optimization itself. In this article, an improved generalized algorithm for interval propagation in simple networks, and an extension of the model structure of the pooling problem beyond that of simple pools and streams, are presented. An application of preprocessing to a branch-and-bound approach to process network NLP global optimization is given. The importance of model structure to tightness of derived bounds is illustrated. This work demonstrates that knowledge of the problem structure may be used to integrate reasoning and optimization, and that specialized structures and specific domain constraints can lead to customized algorithms that are significantly more efficient than general algorithms.