TY - GEN
T1 - MOEA/D-AMS
T2 - 2011 IEEE Congress of Evolutionary Computation, CEC 2011
AU - Chiang, Tsung Che
AU - Lai, Yung Pin
PY - 2011
Y1 - 2011
N2 - In this paper we propose a multiobjective evolutionary algorithm based on MOEA/D [1] for solving multiobjective optimization problems. MOEA/D decomposes a multiobjective optimization problem into many single-objective subproblems. The objective of each subproblem is a weighted aggregation of the original objectives. Using evenly distributed weight vectors on subproblems, solutions to subproblems form a set of well-spread approximated Pareto optimal solutions to the original problem. In MOEA/D, each individual in the population represents the current best solution to one subproblem. Mating selection is carried out in a uniform and static manner. Each individual/subproblem is selected/solved once at each generation, and the mating pool of each individual is determined and fixed based on the distance between weight vectors on the objective space. We propose an adaptive mating selection mechanism for MOEA/D. It classifies subproblems into solved ones and unsolved ones and selects only individuals of unsolved subproblems. Besides, it dynamically adjusts the mating pools of individuals according to their distance on the decision space. The proposed algorithm, MOEA/D-AMS, is compared with two versions of MOEA/D using nine continuous functions. The experimental results confirm the benefits of the adaptive mating selection mechanism.
AB - In this paper we propose a multiobjective evolutionary algorithm based on MOEA/D [1] for solving multiobjective optimization problems. MOEA/D decomposes a multiobjective optimization problem into many single-objective subproblems. The objective of each subproblem is a weighted aggregation of the original objectives. Using evenly distributed weight vectors on subproblems, solutions to subproblems form a set of well-spread approximated Pareto optimal solutions to the original problem. In MOEA/D, each individual in the population represents the current best solution to one subproblem. Mating selection is carried out in a uniform and static manner. Each individual/subproblem is selected/solved once at each generation, and the mating pool of each individual is determined and fixed based on the distance between weight vectors on the objective space. We propose an adaptive mating selection mechanism for MOEA/D. It classifies subproblems into solved ones and unsolved ones and selects only individuals of unsolved subproblems. Besides, it dynamically adjusts the mating pools of individuals according to their distance on the decision space. The proposed algorithm, MOEA/D-AMS, is compared with two versions of MOEA/D using nine continuous functions. The experimental results confirm the benefits of the adaptive mating selection mechanism.
KW - evolutionary algorithm
KW - mating pool
KW - mating selection
KW - multiobjective optimization
KW - scalarization
KW - subproblem
UR - http://www.scopus.com/inward/record.url?scp=80051992344&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=80051992344&partnerID=8YFLogxK
U2 - 10.1109/CEC.2011.5949789
DO - 10.1109/CEC.2011.5949789
M3 - Conference contribution
AN - SCOPUS:80051992344
SN - 9781424478347
T3 - 2011 IEEE Congress of Evolutionary Computation, CEC 2011
SP - 1473
EP - 1480
BT - 2011 IEEE Congress of Evolutionary Computation, CEC 2011
Y2 - 5 June 2011 through 8 June 2011
ER -