|Start Page Number:||615|
|End Page Number:||633|
|Publication Date:||May 2017|
|Journal:||International Transactions in Operational Research|
|Authors:||Hansen Pierre, Mladenovic Nenad, Hanafi Sad, Mjirda Anis, Todosijevic Raca|
|Keywords:||heuristics, programming: travelling salesman, combinatorial optimization, performance|
In a single local search algorithm, several neighborhood structures are usually explored. The simplest way is to define a single neighborhood as the union of all predefined neighborhood structures; the other possibility is to make an order (or sequence) of the predefined neighborhoods, and to use them in the first improvement or the best improvement fashion, following that order. In this work, first we classify possible variants of sequential use of neighborhoods and then, empirically analyze them in solving the classical traveling salesman problem (TSP). We explore the most commonly used TSP neighborhood structures, such as 2‐opt and insertion neighborhoods. In our empirical study, we tested 76 different such heuristics on 15,200 random test instances. Several interesting observations are derived. In addition, the two best of 76 heuristics (used as local searches within a variable neighborhood search) are tested on 23 test instances taken from the TSP library (TSPLIB). It appears that the union of neighborhoods does not perform well.