Simulation optimisation using a genetic algorithm

Abstract The paper presents an attempt to apply genetic algorithms (GAs) to the problem of optimising an existing simulation model. A simple real-coded GA is presented and used to change the simulation model parameters. With each new parameter set proposed, a simulation run is performed. From the statistics gathered by running the simulation, a goal function is constructed that measures the quality of these parameters. Because of its nature and the stochastic and unpredictable behaviour of the complex simulation model, the goal function used leads to a highly non-linear, noisy and mixed (discrete and continuous) programming optimisation problem. A GA successfully works on it, and as a result gives a parameter set that measures substantially better than the nominal one. This demonstrates the capability of GAs to solve hard inverse problems even in the area of complex simulation model optimisation.