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.
