Authors: Musrrat Ali, Millie Pant, Ajith Abraham
Addresses: Department of Paper Technology, Indian Institute of Technology Roorkee, Roorkee-247667, India. ' Department of Paper Technology, Indian Institute of Technology Roorkee, Roorkee-247667, India. ' Machine Intelligence Research Labs (MIR Labs), P.O. Box 2259, Auburn, Washington 98071-2259, USA
Abstract: Differential evolution (DE) is a reliable and versatile function optimiser especially suited for continuous optimisation problems. Practical experience, however, shows that DE easily looses diversity and is susceptible to premature and/or slow convergence. This paper proposes a modified variant of DE algorithm called improved differential evolution (IDE). It works in three phases: decentralisation, evolution and centralisation of the population. Initially, the individuals of the population are partitioned into several groups of subpopulations (decentralisation phase) through a process of shuffling. Each subpopulation is allowed to evolve independently from each other with the help of DE (evolution phase). Periodically, the subpopulations are merged together (centralisation phase) and again new subpopulations are reassigned to different groups. These three phases helps in searching all the potential regions of the search domain effectively, thereby, maintaining the diversity. The promising nature of IDE is demonstrated on a testbed of 16 benchmark problems having box constraints. Comparison of numerical results shows that IDE is either better or at par with other contemporary algorithms.
Keywords: differential evolution; mutation; crossover; decentralisation; centralisation; continuous optimisation; box constraints.
International Journal of Bio-Inspired Computation, 2011 Vol.3 No.1, pp.17 - 30
Published online: 12 Nov 2014 *Full-text access for editors Access for subscribers Purchase this article Comment on this article