Parameter Tuning Problem in Metaheuristics: A Self-Adaptive Local Search Algorithm for Combinatorial Problems

Abstract

Combinatorial optimization is an important mathematical topic that consists of finding an optimal solution from a finite set of search space. Many problems encountered in real life are defined as combinatorial optimization. There is an increasing interest among researchers to develop heuristic algorithms for combinatorial optimization problems, because enumeration based search is not feasible for them. So, obtaining global optimum solutions for these problems, within a reasonable time, is extremely difficult by exact algorithms. Particularly in recent years, high-level metaheuristics have been developed for combinatorial optimization problems. On the other hand, it is known that metaheuristic algorithms are controlled by a set of parameters. The best parameter set reveals better performance such as solution quality and computer times. The process to find the best parameter set is called parameter optimization or parameter tuning that requires a deep learning of the problem structure or a roughly trial-and-error process. An alternative way for tuning is to control parameters through the running of the algorithms. Those algorithms utilize some feedback from the search and change the parameter values adaptively depending on the knowledge. There is a great deal to develop adaptive algorithms for combinatorial optimization problems to overcome the difficulties of parameter tuning. While a survey is carried out about parameter tuning approaches for metaheuristics, the performance of a new self-adaptive local search (SALS) algorithm is introduced in this chapter, and investigated for the vehicle routing problem considering both single and multi-objectives on a large scale suit of test problems.