Parallel genetic algorithms for a neurocontrol problem

The major purpose of this work is twofold. One is to understand the capability of genetic algorithms (GA) in artificial neural networks (ANN) design problems; the other is to improve the efficiency and reliability of GA by a coarse-grained parallel processing architecture. A parallel processing architecture is proposed in this paper. Based on the proposed architecture, the ability to allow free exchange of a random number of migration elements between sub-populations of GA, and to allow system expansion without extra coding, is an innovation. Implementation results of an inverse pendulum controller design problem show that, the migration genetic algorithms based on the proposed scheme offer significant improvements in search repeatability and efficiency over the standard GAs.