Comparison of optimisation methods tested on testing functions and discrete event simulation models

This paper deals with the comparison of selected optimisation methods (random search, hill climbing, tabu search, local search, downhill simplex, simulated annealing, differential evolution and evolution strategy) using various evaluation criteria proposed by us which express the success of the optimisation method in different ways (the success of the method in finding the global optimum, the difference between optimum and local extremes, the distances of quartiles of possible solutions, the number of simulation experiments until the optimum was found). Some of the selected optimisation methods (downhill simplex, simulated annealing, differential evolution and evolution strategy) were modified to improve their behaviour to find the global optimum. Optimisation methods were tested on selected testing functions (De Jong's, Rosenbrock's, Michalewicz's, and Ackley's function ‐ the substitution of the simulation model) and three discrete event simulation models (the manufacturing system and logistics, the penalty, and the assembly line). We have modified downhill simplex and differential evolution which are implemented in the simulation optimiser after the initial testing on simulation models.