Diversity rate of change measurement for particle swarm optimisers

Abstract

The diversity of a particle swarm can reflect the swarm's explorative/exploitative behaviour at a given time step. This paper pro- poses a diversity rate of change measure to quantify the rate at which particle swarms decrease their diversity over time. The proposed measure is based on a two-piecewise linear approximation of diversity measurements sampled at regular time steps. The proposed measure is the slope of the first of the two lines. It is shown that, when comparing the measure among different algorithms, the measure reflects the differences in the behaviour of algorithms in terms of their exploration-exploitation trade-o . The measure can potentially be used to characterise and classify different algorithms based on algorithm behaviour.