Parallel Gene Transfer Operations for the Bacterial Evolutionary Algorithm

Abstract

Bacterial evolutionary algorithm (BEA) is a special evolutionary method, originally developed to optimize fuzzy system parameters, but it is also useful in many design and engineering problems. In the latter case, applications kill most of their time with the evaluation of the objective function. One possible way to speed up BEA is the use of many CPUs, and evaluate the objective functions of different individuals on them simultaneously. This can be realized e.g. with a cluster of workstations or with an SMP/SMC system. Unfortunately, the BEA in its original form is not suitable for parallelization, because one of the two main operations, the gene transfer is sequential. In this article we propose three new alternatives of BEA and present the results of test calculations with 5 test problems, on 1 to 16 processors to find a good parallel version of BEA.