Feedback-control operators for improved Pareto-set description: Application to a polymer extrusion process. (February 2015)

• Record Type: Journal Article
• Authors: Carrano, Eduardo G.; Gouveia Coelho, Dayanne; Gaspar-Cunha, António; Wanner, Elizabeth F.; Takahashi, Ricardo H.C.
• Is Part Of: Engineering applications of artificial intelligence. Volume 38(2015:Feb.)
• Page Start: 147
• Page End: 167
• Publication Date: 2015-02
• Subjects: Evolutionary computation -- Multiobjective optimization -- Genetic algorithms -- Polymer extrusion -- Local search; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2014.10.016
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: This paper presents a new class of operators for multiobjective evolutionary algorithms that are inspired on feedback-control techniques. The proposed operators, the archive-set reduction and the surface-filling crossover, have the purpose of enhancing the quality of the description of the Pareto-set in multiobjective optimization problems. They act on the Pareto-estimate sample set, performing operations that eliminate archive points in the most crowded regions, and generate new points in the less populated regions, leading to a dynamic equilibrium that tends to generate a uniform sampling of the efficient solution set. The internal parameters of those operators are coordinated by feedback-control inspired techniques, which ensure that the desired equilibrium is attained. Numerical experiments in some benchmark problems and in a real problem of optimization of a single screw extrusion system for polymer processing show that the proposed methodology is able to generate more detailed descriptions of Pareto-optimal fronts than the ones produced by usual algorithms.