Dynamic floating function: A novel test problem generator for non-stationary environments

Dynamic optimization is one of the important research area in the intelligence computation field. For a decade, various dynamic benchmark test functions have been put forward. Generally speaking, though these functions help to improve a lot on the dynamic algorithms design, the fact that the whole landscape might affects the algorithm's performance rather than that of the way it changes is ignored. For example, the reason of sticking to the local optima may be not for the dynamic change, but for the complex landscape. In this paper, a novel dynamic test problem generator, named dynamic floating function is proposed. It inherits the advantages of other dynamic benchmarks, as well as includes the complexity of landscape after the change occurs. And the dynamic can be changed just by adjusting some simple variables of the basic function and floating function. Several typical test environments are given and a comparative study of a genetic algorithm and two particle swarm optimization algorithms is done.