Fitness Landscape Based Parameter Estimation for Robust Taboo Search

Metaheuristic optimization algorithms are general optimization strategies suited to solve a range of real-world relevant optimization problems. Many metaheuristics expose parameters that allow to tune the e ort that these algorithms are allowed to make and also the strategy and search behavior [1]. Adjusting these parameters allows to increase the algorithms' performances with respect to different problemand problem instance characteristics. The di culty in exposing parameters of metaheuristics is that these parameters need to be set and should be adjusted for good performance. Also, choosing a reasonable default value is not an easy task for algorithm developers. The purpose of this work is, on the one hand to explore the e ect of parameter settings and provide more suited default values, and on the other hand to introduce a new method to use tness landscape analysis (FLA) for the prediction of algorithm parameterization. An overview of tness landscape analysis is given in [6]. The scope of such a study can in general be extended to any algorithm and problem combination, but this paper is restricted to the robust taboo search (RTS) when applied to problem instances of the quadratic assignment problem library (QAPLIB). With a similar intention [2] used tness landscape analysis to choose the best algorithm to solve instances out of the BBOB benchmark set. In [7] a study was made that predicted the hardness of a problem instance using FLA ngerprints.