A multi-objective adaptive surrogate modelling-based optimization algorithm for constrained hybrid problems

• Published in: Environmental modelling & software : with environment data news Vol. 148; p. 105272
• Main Authors: Sun, Ruochen, Duan, Qingyun, Mao, Xiyezi
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 01.02.2022; Elsevier Science Ltd
• Subjects: Algorithms; computer software; Constrained hybrid problem; Constraint modelling; Continuity (mathematics); Design engineering; Environment models; Environmental management; Environmental modeling; Evaluation; hybrids; MO-ASMO; Multi-objective optimization; Multiple objective analysis; NSGA-II; Optimization; Surrogate model; Systems design; systems engineering; Water distribution; Water distribution systems; Water engineering; MO-ASMO; NSGA-II; Multi-objective optimization; Surrogate model; Constrained hybrid problem
• ISSN: 1364-8152; 1873-6726
• DOI: 10.1016/j.envsoft.2021.105272

Many multi-objective optimization problems in integrated environmental modelling and management involve not only continuous decision variables but also variables like integers and/or discrete variables. Furthermore, the optimization problems are often subject to various constraints. Solving this kind of constrained hybrid problems usually requires a huge number of model evaluations that can be computationally expensive. This study presents an algorithm known as multi-objective adaptive surrogate modelling-based optimization for constrained hybrid problems (MO-ASMOCH). It incorporates several evolutionary operators to handle different types of decision variables and uses a classification surrogate model to deal with model constraints. MO-ASMOCH was evaluated against the widely used NSGA-II method on three engineering design problems and three water distribution system design problems with up to 30 dimensions. The results showed that MO-ASMOCH is able to obtain nondominated solutions of similar quality as that of NSGA-II using much fewer model evaluations. •MO-ASMOCH is proposed to solve multi-objective constrained hybrid optimization problems.•MO-ASMOCH handles various model constraints using a classification surrogate model.•MO-ASMOCH has been evaluated on 6 benchmark problems with referenced Pareto fronts.•MO-ASMOCH is as effective as but is much more efficient than NSGA-II.