C. Sandrock and P.L. de Vaal (South Africa)
Multi-objective, stochastic, particle swarm, curve fitting, Pareto optimality
An important aspect of stochastic simulation is the devel opment of realistic input scenarios. This work describes a technique for determining the frequencies of transitions be tween input prototypes by fitting historic data. Instead of deciding on a single objective function, multiple curves are fit that are Pareto optimal in terms of a number of objectives using the Multi-objective Particle Swarm Optimisation al gorithm. The objectives are: fit error, number of curves and curvature of the prototypes. For this study, prototypes were chosen that represent first order step responses. The fit prototypes are then interpreted as being a certain type of event. The resulting list of possible event sequences is used to populate an event transition probability matrix with better coverage than any one fit would have given.
Important Links:
Go Back
