Incorporation of human decision making preference into evolutionary multi-objective optimization

Abstract

Human preference in multi-objective decision making contains uncertainties and anomalies that are to be taken into account in a formal model of preference. The uniqueness of the evolutionary computation approach renders the direct adoption of modelling and implementation techniques developed for classical optimization approaches unsuitable. This thesis documents research effort into the articulation and incorporation of preference information into EMOO. Models of preference formulated in terms of the importance ranking of objectives, an imprecisely specified reference vector, and objective trade off and their implementations in MOEAs are reviewed. Three preference incorporation schemes encompassing the representation, elicitation and implementation of preference information are also proposed. The approaches are designed for easy adoption into major state-of-the-art MOEAs. The first guides the population of solutions towards an imprecisely specified goal vector. The second directs the search to regions of optimum trade-off in the Pareto front. The third incorporates importance ranking of objective functions into MOEAs.