A process simulator interface for multiobjective optimization of chemical processes

• Published in: Computers & chemical engineering Vol. 109; pp. 119 - 137
• Main Authors: Muñoz López, Carlos André, Telen, Dries, Nimmegeers, Philippe, Cabianca, Lorenzo, Logist, Filip, Van Impe, Jan
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 04.01.2018
• Subjects: Aspen Plus; Gradient-based optimization; Multiobjective optimization; Process optimization; Aspen Plus; Process optimization; Multiobjective optimization; Gradient-based optimization
• ISSN: 0098-1354; 1873-4375
• DOI: 10.1016/j.compchemeng.2017.09.014

•A novel interface for gradient-based multiobjective optimization of processes.•Multiple relevant tradeoffs in the Pareto front can be evaluated for complex (bio)chemical processes modeled in Aspen Plus.•The structure of an equation set object is exploited to transfer gradient information.•Higher computational performance, better constraint satisfaction and complete Pareto fronts are produced. The (bio)chemical process industry is under an increasing pressure due to smaller margins and increasing societal and legislative demands for a sustainable future. In this context model-based optimization contributes to the solution because it serves to improve the processes’ performance. Furthermore, multiobjective optimization techniques provide the decision maker with a deeper insight in the tradeoffs when choosing an operating condition. However, an accurate process model is needed to apply these techniques efficiently. In this paper, a novel interface is developed between state-of-the-art gradient-based optimization techniques and the widely used process simulator Aspen Plus. Furthermore, specific challenges and solutions for overcoming the gap between process simulators and optimization tools are highlighted. The resulting interface allows gradient-based techniques to be exploited for optimization of complex industrial processes modeled in the advanced Aspen Plus environment. The interface ensures constraints satisfaction, and a higher computational performance than gradient free methods.