Comparison of surrogate models for wastewater process synthesis

Abstract In this work, the optimization based method for wastewater process synthesis is studied by the application of more-rigorous process models for wastewater treatment processes units. A cooling-tower and a steam stripper are approximated by surrogate models for their utilization in a known refinery wastewater treatment synthesis problem proposed by Huang et al. (1999). The data for obtaining the surrogate models are generated from a semi-phenomenological algebraic cooling-tower model built in MatLab 7.8 and from a phenomenological steam stripper model simulated in a commercial simulator (Aspen Plus version 7.2). Two types of surrogate models are compared: a polynomial black box model and a neural-network model. The refinery wastewater problem was formulated as an MINPL and solved with different starting points. Different minima were obtained with both the models. The results showed that both the approaches are computationally tractable with a similar accuracy, allowing their use in the design of more realistic water network synthesis.