Informative performance metrics for dynamic optimisation problems

Existing metrics for dynamic optimisation are designed primarily to rate an algorithm's overall performance. These metrics show whether one algorithm is better than another, but do not indicate any specific aspects of the performance. In this paper we split the offline error metric into two component parts. We propose a new metric to measure convergence speed, and show how this, when combined with a population diversity metric, correlates strongly with the overall performance. We then use these metrics to analyse several optimisation algorithms, yielding new insight into both the test function and how the algorithms' characteristics can be improved.