Preferences and their application in evolutionary multiobjective optimization

Cvetkovic, D.; Parmee, Ian

Abstract

The paper describes a new preference method and its use in multiobjective optimization. These preferences are developed with a goal to reduce the cognitive overload associated with the relative importance of a certain criterion within a multiobjective design environment involving large numbers of objectives. Their successful integration with several genetic-algorithm-based design search and optimization techniques (weighted sums, weighted Pareto, weighted coevolutionary methods, and weighted scenarios) are described and theoretical results relating to complexity and sensitivity of the algorithm are presented and discussed. Its usefulness has been demonstrated in a real-world project of conceptual airframe design (a joint project with British Aerospace Systems).

• Journal Article Type: Article
• Publication Date: Feb 1, 2002
• Journal: IEEE Transactions on Evolutionary Computation
• Print ISSN: 1089-778X
• Publisher: Institute of Electrical and Electronics Engineers
• Peer Reviewed: Not Peer Reviewed
• Volume: 6
• Issue: 1
• Pages: 42-57
• DOI: https://doi.org/10.1109/4235.985691
• Keywords: algorithm design and analysis, decision making, delta modulation, design engineering, design optimization, helium, mathematical model, operations research, optimization methods, pareto optimization
• Public URL: https://uwe-repository.worktribe.com/output/1079290
• Publisher URL: http://dx.doi.org/10.1109/4235.985691
• Additional Information: Additional Information : Presents an objective preference system where designer sets preferences linguistically (eg 'A more important than B') and its integration with a multi-objective framework within an Interactive Evolutionary Design System (IEDS). IEDS development involved several researchers collaborating with BAE Systems (EPSRC Grant GR/L/31654). IGR outcome was 'Tending Towards International Excellence'. This laid a foundation for current user-centric intelligent systems research that is now at the core of ACDDM Lab activity. 2006 impact factor for the journal is 3.770. 3rd amongst 85 "CS Artificial Intelligence" journals; 2nd amongst 75 "CS Theory & Methods" journals and 9th amongst all 365 CS journals.