Evolutionary Multiobjective Optimization
Evolutionary Multiobjective Optimization
Optimization problems with multiple, often conflicting, objectives arise in a natural fashion in most real-world applications, such as aerodynamic design, financial decision making, and electronic circuit development. An important task in multiobjective optimization is to identify a set of optimal trade-off solutions (called a Pareto set) between the conflicting objectives, which helps gain a better understanding of the problem structure and supports the decision-maker in choosing the best compromise solution for the considered problem.
Evolutionary algorithms are particular suited for approximating the entire Pareto set because they work with a population of solutions rather than a single solution candidate. This enables approximating several members of the Pareto set simultaneously in a single algorithm run.
This Demonstration shows how an evolutionary multiobjective optimization algorithm (NSGA-II) approximates the Pareto set of Kursawe's two-objective optimization problem, which has a nonconvex, disconnected two-dimensional Pareto front and a disconnected three-dimensional Pareto set.
In each generation, all solution candidates are represented as decision vector in decision space (where the evolutionary search takes place) and as objective vector in objective space. Since the quality of a candidate solution is evaluated as a vector rather than a scalar, the concept of Pareto dominance is used to compare the quality of the solutions: The solution is said to dominate another solution if is not worse than in all objectives, and is strictly better than in at least one objective.
x=
T
(,,)
x
1
x
2
x
3
f(x)=
T
((x),(x))
f
1
f
2
(1)
x
(2)
x
(1)
x
(2)
x
(1)
x
(2)
x
In each generation, all solutions that are not dominated by any other are called nondominated solutions. These solutions form the current approximation of the Pareto set and are marked green. In objective space, the image of the nondominated solutions is called the Pareto front approximation.