An improved multi-objective differential evolution with a termination criterion for optimizing chemical processes

• Published in: Computers & chemical engineering Vol. 56; pp. 155 - 173
• Main Authors: Sharma, Shivom, Rangaiah, Gade Pandu
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 13.09.2013; Elsevier
• Subjects: Alkylation process; Applied sciences; Biological and medical sciences; Chemical engineering; Criteria; Differential evolution; Evolution; Exact sciences and technology; Fermentation process; Fermented food industries; Food industries; Fundamental and applied biological sciences. Psychology; Lists; Mathematical analysis; Mathematical models; Multi-objective optimization; Performance measurement; Reactors; Searching; Termination criterion; Williams-Otto process; Differential evolution; Fermentation process; Termination criterion; Multi-objective optimization; Williams-Otto process; Alkylation process; Space application; Statistical analysis; Chemical engineering; Evolutionary algorithm; Objective analysis; Multiobjective programming; Fermentation; Alkylation; Optimization; Engineering design; Chemical reactor; Metric
• ISSN: 0098-1354; 1873-4375
• DOI: 10.1016/j.compchemeng.2013.05.004

•Variation of five performance metrics with generations is analyzed for MOO problems.•Performance-based termination criterion is proposed for evolutionary algorithms.•Efficacy of MODE is improved by including taboo list and termination criterion.•The proposed termination criterion is tested on constrained MOO problems.•Improved MODE is used to solve three chemical engineering application problems. Application problems have conflicting objectives and constraints, and maximum number of generations is the most common termination criterion in evolutionary algorithms used for solving these applications. This study develops a termination criterion using the non-dominated solutions obtained as the search progresses. For this, several performance metrics are modified, and their variation with generations has been assessed on many test functions. Based on this analysis, it is proposed to terminate the search if the improvement in variance of two selected performance metrics obtained in recent generations is statistically insignificant. Additionally, evaluation of objectives and constraints is computationally expensive in many applications. This study uses taboo list with multi-objective differential evolution to avoid re-visits and for better exploration of search space. Benefits of the termination criterion and taboo list are assessed on constrained benchmark problems. The developed approach is then evaluated on three chemical engineering applications, namely, alkylation, Williams-Otto and fermentation processes.