Evolutionary Algorithms for the Optimization of Simulation Models Using PVM

Simulation models usually share some specific characteristics that make the automatic optimization of their input parameters an extremely difficult task. Evolutionary algorithms--- search and optimization methods gleaned from the model of organic evolution--- are applicable to this problem and known to be able to yield good solutions for many difficult practical optimization problems. The paper presents a parallel, steady-state evolutionary algorithm which exploits the available parallel machine configuration in an optimal manner. The algorithm is implemented under PVM and runs in a LAN of SUN SPARC workstations. The basic algorithm is applicable to arbitrary simulation models, and only the individual structure and the genetic operators must be specified for a particular application. As an application example, the problem of optimizing pressurized water reactor core reload designs is briefly discussed and first experimental results are presented