Multi-objective optimisation of chemical processes via improved genetic algorithms: A novel trade-off and termination criterion

Abstract

A sustainable chemical process operation often requires optimality with respect to multiple conflicting objectives as economic, societal and environmental aspects need to be addressed. Multi-objective optimisation aims at solving such problems. Asingle optimal solution for all objectives does not exist as one cannot improve with respect to one objective without worsening with respect to one of the other objectives. The result of a multi-objective optimisation algorithm is hence a Pareto set comprising equally optimal trade-off solutions. In this contribution an improved version of NSGA-II, one of the current state-of-the-art algorithms in evolutionary multi-objective optimisation, is presented and applied to the optimisation of an industrially relevant case. The proposed novel algorithm overcomes, amongst others, one of the major shortcomings of the currently used evolutionary multi-objective optimisation algorithms: the inability to distinguish between solutions based on their trade-off and distribution. This results in a Pareto front cluttered with irrelevant solutions. The performance of the improved algorithm has been evaluated for the multi-objective optimisation of a methane tri-reforming process for methanol production. Three objectives have been studied: (i) minimisation of the total energy demand, (ii) maximisation of the carbon efficiency and (iii) an economic profit function. For this INPROP, the recently developed interface between Matlab and Aspen Plus for multi-objective optimisation of chemical processes, is extended to perform multi-objective optimisation using the proposed method. Up till now, INPROP only exploited scalarisation-based multi-objective optimisation methods. Scalarisation methods convert the multi-objective optimisation problem (MOOP) in a sequence of single objective optimisation problems, which are solved separately using deterministic gradient-based methods. Evolutionary algorithms, like NSGA-II, are vector-based and tackle the MOOP in its entirety. The results obtained with the improved evolutionary algorithm are compared with earlier presented scalarisation-based multi-objective process optimisation results.