Driving dynamic multi-objective optimizations constrained by decision-makers' preferences

Abstract

Dynamic multi-objective optimization problems (DMOOPs) are an interesting and a relatively complex class of optimization problems where elements of the problems, such as objective functions and/or constraints, change with time. These problems are characterized with at least two objective functions in con ict with one another. Sometimes, human decision-makers seek to in uence ways (by restricting the search to a specific region of the Pareto-optimal Front (POF)) in which algorithms that optimize these problems behave by incorporating personal preferences into the optimization process. This dissertation proposes approaches that enable decision-makers to in uence the optimization process with their preferences. The decision-makers' imparted preferences force a reformulation of the optimization problems as constrained problems, where the constraints are defined in the objective space. Consequently, the constrained problems are then solved using variations of constraint handling techniques, such as penalization of infeasible solutions and the restriction of the search to the feasible region. The proposed algorithmic approaches' performance are compared using standard performance measures for dynamic multi-objective optimization (DMOO) and newly proposed measures. The proposed measures estimate how well an algorithm is able to find solutions in the objective space that best re ect the decision-maker's preferences and the paretooptimality goal of DMOO. This dissertation also proposes a new di erential evolution algorithm, called dynamic di erential evolution vector-evaluated non-dominated sorting (2DEVENS). 2DEVENS combines elements of the dynamic non-dominated sort genetic algorithm version II (DNSGA-II) and the dynamic vector-evaluated particle swarm optimization (DVEPSO) algorithm to drive the search for solutions. The proposed 2DEVENS algorithm compared favorably with other nature-inspired algorithms that were used in the studies carried out for this dissertation. The proposed approaches used in incorporating decision-makers' preferences in the optimization process also demonstrated good results.