Takeover Time in Evolutionary Dynamic Optimization: From theory to practice

Making theoretical has been a hard task for researchers in the field of Evolutionary Dynamic Optimization (EDO), as only a few approaches have appeared in recent years. In EDO, problems change over time, requiring from the solver, an Evolutionary Algorithm (EA), to continuously adapt to new conditions. Mathematical tools such as the takeover time models, extensively used to characterize and compare EAs in static problems, become much more difficult to understand when the problem changes over time. A preliminary takeover time model have been recently introduced for tournament selection and diversity-generating approaches. In this article, we propose a new enhanced model that takes into account important scenarios that were not initially considered. We use predictive modeling to describe the EAs performance and statistical analysis to validate our equations. Finally, we show how these theoretical models can be used to build novel techniques in EDO.