The parametric study on differential evolution must be carried out on a test bed. When the parametric study was initiated in 2004, the initial test bed only included the benchmark electromagnetic inverse scattering problem and two toy functions defined by this author. As such it was seriously inadequate from the point of view of problem features. The features of an application problem may affect the choice of strategy, intrinsic and non-intrinsic control parameters. The relationship between problem features, strategy, intrinsic and non-intrinsic control parameters is of great significance. It is therefore necessary to build a test bed with all representative problem features fully and appropriately addressed.Although many test beds are available, none of them is satisfactory. No effort to benchmark a test bed for unconstrained single-objective optimization has been found with problem features fully and properly addressed. When evaluating an evolutionary algorithm, people usually choose test problems from the available literature. Very often, insufficient attention is given to problem features.
A Survey on Test Problems
SourcesThere are two sources for test problems. Optimization publications including books, journals, conference proceedings, technical reports, degree thesis, and so on, contain numerous test problems. In general, test problems from this source are more trustworthy because of the careful reviewing and proofreading that precede publication. However, using this source is very tedious and time-consuming. Moreover, access to optimization publications is usually limited.