Optimization of Simulated System Effectiveness Using Evolutionary Algorithms

Abstract

Using evolutionary algorithms, a search is per formed based on a population where each popula tion member consists of a vector of attribute values and a fitness value. A simulation of a system is run, given a particular set of the member attribute values, producing a fitness value. Fitness measures how well the system achieves its mission objectives. If the fitness has a random component, several runs are made to produce average fitness. The pro cedure is to select the best members from the popu lation based on average fitness and mutate the member attribute values to produce new popula tion members. Since population member attributes can affect process reaction times, wait logic, or decision logic, a search for the best attribute values over 50 to 100 generations can result in optimal fitness. In order to demonstrate the use of evolu tionary algorithms in system optimization, a sim ple inventory system that has a complex fitness surface is considered.