This chapter presents an overview of the design and analysis of computational experiments with optimization algorithms. It covers classic designs and their corresponding (meta)models; namely, Resolution-III designs including fractional factorial two-level designs for first-order polynomial models, Resolution-IV and Resolution-V designs for two-factor interactions, and designs including central composite designs for second-degree polynomials. It also reviews factor screening in experiments with very many factors, focusing on the sequential bifurcation method. Furthermore, it reviews Kriging models and their designs. Finally, it discusses experiments aimed at the optimization of the parameters of a given optimization algorithm, allowing multiple random experimental outputs. This optimization may use either generalized response surface methodology or Kriging combined with mathematical programming; the discussion also covers Taguchian robust optimization.
|Title of host publication||Empirical Methods for the Analysis of Optimization Algorithms|
|Editors||T. Bartz-Beielstein, M. Chiarandini, L. Paquete, M. Preuss|
|Place of Publication||Heidelberg|
|Publication status||Published - 2010|
Kleijnen, J. P. C. (2010). Design and analysis of computational experiments: Overview. In T. Bartz-Beielstein, M. Chiarandini, L. Paquete, & M. Preuss (Eds.), Empirical Methods for the Analysis of Optimization Algorithms (pp. 51-72). Springer Verlag.