A Multiobjective Evolutionary Algorithm based on Decomposition MOEAD
A Multi-objective Evolutionary Algorithm based on Decomposition (MOEA/D) FASLIP Discussion March 29 th, 2017
A multi-objective problem Single-objective sub-problem 1 individual How to decompose? How do sub-problems interact to solve the big problem? How can a single individual solve a sub-problem? …… Single-objective sub-problem 1 individual Ø Each sub-problem is solved by 1 individual Ø The number of individual (population size) is equal to the number of single objective problems.
Decomposition Approaches (1) ØAssume we need to maximize all objectives: ØWeighted Sum Approach: given a weight vector to each sub-problem
Decomposition Approaches (2) ØReference point : ØTchebycheff Approach: minimize the distance between a solution to the reference point.
Decomposition Approaches (3) ØBoundary Intersection Approach: find the intersection point of the most boundary/true Pareto front and a set of lines passing through the reference point
Neighborhood ØFor each sub-problem, its neighbors consists of T sub-problems which have the most similar weight vectors with the current sub-problem, denoted by B(i). ØNote that the neighbors of a sub-problem contains itself. ØThe solutions of neighboring sub-problems should be similar since they have similar fitness functions.
Quick Summary •
Evolutionary process y‘ y Ø Genetic operators: crossover/mutation or DE operators Ø Repair
Flowchart of MOEA/D
- Slides: 9
