The effect of computer network errors on genetic algorithms

Abstract

Genetic algorithms are widely used in engineering, to solve nonlinear, multi-target optimization problems with multiple variables (e.g. optimization of geometry of flow domains, parameters of control systems). The parallelization of software using genetic algorithms is very important because in a typical practical problem they need huge computational power. Fortunately it is easy to implement a master-slave style parallelization. Our goal was to investigate the effect of random errors that can occur in a cluster of workstations on the efficiency of the genetic algorithm.