Parallel Implementation of Particle Swarm Optimization (PSO) Through Digital Pheromone Sharing

In this paper, a parallelization model for PSO through sharing of digital pheromones between multiple particle swarms to search n-dimensional design spaces is presented. Digital pheromones are models simulating real pheromones produced by insects for communication to indicate a source of food or a nesting location. Particle swarms search the design space with digital pheromones aiding communication within the swarm to improve search efficiency. Digital pheromones have demonstrated the capability of searching design spaces within PSO in the previous work by authors in both single and coarse granular parallel computing environments. Multiple swarms are simultaneously deployed across various processors in the coarse granular scheme and synchronization is carried out only when all swarms achieved convergence. This was done in an effort to reduce processor-to-processor communication and network latencies. With an appropriate parallelization scheme, the benefits of digital pheromones and swarm communication can potentially outweigh the network latencies resulting in improved search efficiency and accuracy. A swarm is deployed in the design space across different processors to explore this idea. Each part of the swarm is made to communicate with each other through an additional processor. Digital pheromones aiding within a swarm, communication between swarms is facilitated through the developed parallelization model. In this paper, the development and implementation of this method together with benchmarking test cases are presented.Copyright © 2008 by ASME