A Self-Adaptive Heuristic Algorithm for Combinatorial Optimization Problems

Berna Dengiz,Cigdem Alabas-Uslu

doi:10.1080/18756891.2014.966992

Abstract

AbstractThis paper introduces a new self-tuning mechanism to the local search heuristic for solving of combinatorial optimization problems. Parameter tuning of heuristics makes them difficult to apply, as parameter tuning itself is an optimization problem. For this purpose, a modified local search algorithm free from parameter tuning, called Self-Adaptive Local Search (SALS), is proposed for obtaining qualified solutions to combinatorial problems within reasonable amount of computer times. SALS is applied to several combinatorial optimization problems, namely, classical vehicle routing, permutation flow-shop scheduling, quadratic assignment, and topological design of networks. It is observed that self-adaptive structure of SALS provides implementation simplicity and flexibility to the considered combinatorial optimization problems. Detailed computational studies confirm the performance of SALS on the suit of test problems for each considered problem type especially in terms of solution quality.

Highlights

Due to the practical and the theoretical importance of combinatorial optimization problems, interest in research to develop exact and heuristic algorithms has been evolved consistently
We aim to show that SelfAdaptive Local Search (SALS) is able to generate very good solutions to combinatorial optimization problems without any tuning effort by applying it to problems selected from different application areas, namely, the classical vehicle routing (VRP), permutation flow shop scheduling (PFSP), quadratic assignment (QAP), and topological design of computer networks (TDP), problems
The performance of SALS is compared with the metaheuristics listed in Table 20 on the QAP by SkorinKapov[14]

Summary

Introduction

Due to the practical and the theoretical importance of combinatorial optimization problems, interest in research to develop exact and heuristic algorithms has been evolved consistently. The run time of exact algorithms often increases exponentially with the instance size and only small or moderate-sized problems can be solved. The best combination of parameter values is a crucial task This task is generally called parameter optimization, parameter tuning or parameter setting. A careful selection of the best parameter set requires either a deep knowledge of the problem structure or a lengthy trial-and-error process. Tuning a set of parameters to achieve robust and high performance of the metaheuristic is a tedious and time consuming process. Adenso-Diaz and Laguna[3] state that about 10% of the total time dedicated to designing and testing of a new heuristic is spent for development, and the remaining 90% is consumed by fine-tuning of parameters. The most of them still are influenced by tediousness of parameter optimization

Objectives

Results

Conclusion