Self-adaptive differential evolution for dynamic environments with fluctuating numbers of optima

Abstract

In this chapter, we introduce the algorithm called: SADynPopDE, a self adaptive multi-population DE-based optimization algorithm, aimed at dynamic optimization problems in which the number of optima in the environment fluctuates over time. We compare the performance of SADynPopDE to those of two algorithms upon which it is based: DynDE and DynPopDE. DynDE extends DE for dynamic environments by utilizing multiple sub-populations which are encouraged to converge to distinct optima by means of exclusion. DynPopDE extends DynDE by: using competitive population evaluation to selectively evolve sub-populations, using a midpoint check during exclusion to determine whether sub-populations are indeed converging to the same optimum, dynamically spawning and removing sub populations, and using a penalty factor to aid the stagnation detection process. The use of self-adaptive control parameters into DynPopDE, allows a more effective algorithm, and to remove the need to fine-tune the DE crossover and scale factors. © Springer-Verlag Berlin Heidelberg 2013.