Accepted:
December 2004
| |
/vol12/msid54/ | © Copyright 2008 | |||
| Volume 12 | Received: Accepted: |
November 2004 December 2004 |
|||
|
Particle swarm optimization with combined mutation and hill climbing
Ian Grundy, Andrew Stacey |
|
| Abstract | |
| Particle swarm optimization is an established, effective technique for global optimization, but one which is prone to delivering sub-optimal results when faced with multimodal functions, especially when the local optima are found in regions where the objective function varies rapidly. The failure of the swarm to venture into these regions is not just of theoretical interest; it has been observed in practical settings, e.g. in the training of neural networks. In this paper we propose a simple modification to the particle swarm algorithm in which, at each generation, a small number of particles are mutated and are allowed to hill climb. The mutation has the effect of randomly bouncing particles towards other parts of the search space, while the effect of hill-climbing is to greatly increase the effective size of the "target" region of interest around the global optimum. We provide results for a selection of well-known test functions, and demonstrate that our modification improves the ability of the swarm to find the global optimum. In most cases this is achieved without a prohibitive increase in the number of function evaluations. | |
| Full Text |
|
|
Multimedia Links (none) Reference Links (none) Citation Reference Ian Grundy, Andrew Stacey 2008, Particle swarm optimization with combined mutation and hill climbing, Complexity International, Volume 12, Paper ID: msid54, URL: http://www.complexity.org.au/vol12/msid54/ |
Get viewers for PS & PDF   |
