MOEA/D

Basic information

Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) is a population-based metaheuristic for solving multi-objective optimization problems. The method decomposes a multi-objective problem into a set of scalar subproblems defined by weight vectors and optimizes them collaboratively by means of neighborhood-based offspring generation and replacement.

The implementation added to this library supports both real-valued and binary solution representations. The algorithm uses Tchebyscheff scalarization, neighborhood structures induced by distances between weight vectors, and support classes for offspring generation.

Implementation notes

Implementation of that optimization method is given within the class MoeadOptimizer.

Supporting functionality is implemented in the following modules and classes:

• MoeadVariationSupport

• MoeadVariationSupportReal

• MoeadVariationSupportBinary

• MOEA/D decomposition utilities

• MOEA/D weight utilities

• MOEA/D metrics

References

[ZLi2007]

Zhang, Q.; Li, H. (2007). “MOEA/D: A Multiobjective Evolutionary Algorithm Based on Decomposition”. IEEE Transactions on Evolutionary Computation. 11 (6): 712–731.

[DebAgr1995]

Deb, K.; Agrawal, R. B. (1995). “Simulated Binary Crossover for Continuous Search Space”. Complex Systems. 9 (2): 115–148.

[Deb2001]

Deb, K. (2001). “Multi-Objective Optimization Using Evolutionary Algorithms”. Wiley.