DECOR: Differential Evolution using Clustering based Objective Reduction for many-objective optimization

•Proposes DECOR with Improved Differential Evolution for Multi-objective Optimization.•Features clustering for fast online objective reduction & new elitism approach.•Offers a selection metric & tackles singleton clusters to avoid early termination.•Tested on DTLZ problems for 10–20 objectiv...

Full description

Saved in:
Bibliographic Details
Published in:Information sciences Vol. 423; pp. 200 - 218
Main Authors: Pal, Monalisa, Saha, Sriparna, Bandyopadhyay, Sanghamitra
Format: Journal Article
Language:English
Published: Elsevier Inc 01.01.2018
Subjects:
Differential evolution
DTLZ test problems
Many-objective optimization
Objective reduction
Pareto-optimality
Differential evolution
Pareto-optimality
DTLZ test problems
Many-objective optimization
Objective reduction
ISSN:0020-0255, 1872-6291
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•Proposes DECOR with Improved Differential Evolution for Multi-objective Optimization.•Features clustering for fast online objective reduction & new elitism approach.•Offers a selection metric & tackles singleton clusters to avoid early termination.•Tested on DTLZ problems for 10–20 objectives & parallel coordinates are visualized.•DECOR is superior concerning convergence & equivalent to others concerning diversity. Challenges like scalability and visualization which make multi-objective optimization algorithms unsuitable for solving many-objective optimization problems, are often handled using objective reduction approaches. This work proposes a novel many-objective optimization algorithm, viz. Differential Evolution using Clustering based Objective Reduction (DECOR). Correlation distance based clustering of objectives from the approximated Pareto-front, followed by elimination of all but the centroid constituent of the most compact cluster (with special care to singleton cluster), yields the reduced objective set. During optimization, the objective set periodically toggles between full and reduced size to ensure both global and local exploration. For finer clustering, number of clusters is eventually increased until it is equal to the remaining number of objectives. DECOR is integrated with an Improved Differential Evolution for Multi-objective Optimization (IDEMO) algorithm which uses a novel elitist selection and ranking strategy to solve many-objective optimization problems. DECOR is applied on some DTLZ problems for 10 and 20 objectives which demonstrates its superior performance in terms of convergence and equivalence in terms of diversity as compared to other state-of-the-art optimization algorithms. The results have also been statistically validated. Source code of DECOR is available at http://decor.droppages.com/index.html.
ISSN:0020-0255
1872-6291
DOI:10.1016/j.ins.2017.09.051