Authors: Wei Zhan; Chao Guo; Leiping Xiong
Addresses: School of Computer Science, Yangtze University, Jingzhou, Hubei, China ' School of Computer Science, Yangtze University, Jingzhou, Hubei, China ' School of Computer Science, Yangtze University, Jingzhou, Hubei, China
Abstract: Manifold learning algorithm can find out the low-dimensional smooth manifold embedded in high-dimensional data. So, in this paper, the manifold learning algorithm is introduced into multiobjective optimisation algorithm for multiobjective optimisation problems (MOPs), and a manifold-learning-based multiobjective evolutionary algorithm (ML-MOEA) is proposed to overcome deficiency of the traditional evolutionary multi-objective optimisation algorithms (EMOAs) and model-based multi-objective optimisation algorithms (MOEAs) for reducing dimension of data and mining manifold in the decision space of MOPs, build accurate model, guide algorithm evolution and accelerate convergence. The steps of ML-MOEA is as follows: 1) randomly initialisation; 2) modelling via manifold learning algorithm; 3) extend and reproduction; 4) elite selection; 5) halt or go to step 2. Based on the framework of ML-MOEA, a ML-MOEA via self-organising maps (ML-MOEA/SOM) and a ML-MOEA via SOM locally linear embedding (ML-MOEA/LLE) is proposed, and comparison experiment of algorithm performance is done.
Keywords: manifold learning; multiobjective optimisation; evolutionary algorithms; model-based EDs; estimation of distribution algorithms; modelling; locally linear embedding; LLE.
International Journal of Computing Science and Mathematics, 2015 Vol.6 No.3, pp.287 - 296
Available online: 28 May 2015 *Full-text access for editors Access for subscribers Purchase this article Comment on this article