State-of-the-art Multiobjective Evolutionary Algorithms---Pareto Ranking, Density Estimation and Dynamic Population
The major interest of this research work is to investigate multiobjective
optimization problems by evolutionary algorithms. This research classifies
existing Multiobjective Evolutionary Algorithms (MOEAs) and analyzes
several advanced MOEAs based on different design procedures of those
crucial building blocks. Moreover, a Rank-Density based Genetic Algorithm
(RDGA), a Dynamic Multiobjective Evolutionary Algorithm (DMOEA) and a
Dynamic Particle Swarm Evolutionary Algorithm (DPSEA) are designed to
generate a state-of-the-art MOEA, which can efficiently and effectively
approximate a near-complete, near-optimal and uniformly
distributed Pareto front for a given MOP. Findings and
conclusions. From the simulation results of selected quantitative
performance indicators, RDGA is found to be competitive with, or even
superior to, other representative MOEAs in terms of keeping the diversity
of the individuals along the trade-off surface, tending to extend the Pareto
front to new areas, and finding a well-approximated Pareto optimal front.
However, by suing fixed population size, RDGA is found to have difficulties
in dealing with the most challenging issue existing in MOEA design—
the confliction between avoiding and exploiting "genetic drift" phenomenon.
For this reason, DMOEA and DPSEA show their advantages in estimating
optimal population size, resulting in a higher quality Pareto front and
faster convergence speed than RDGA and other representative MOEAs.