Parallel optimization based on artificial bee colony algorithm

This paper aims to tackle the shortcomings of the standard artificial bee colony algorithm (ABC) such as slow convergence, long solving time and being easy to fall into local optima. We study the state transformation formula and propose a parallelized ABC algorithm with Message Passing Interface (MPI). We use the traveling salesman problem (TSP) as the case study. Our experiments show that the parallel ABC algorithm has an advantage in speed over the standard algorithm w.r.t. iterations and convergence speed.

I. Introduction

Combinatorial optimization, an important branch in the field of optimization, uses mathematical methods to find the optimal discrete solution. This kind of problem cannot be solved by conventional methods with increasing scale and time complexity. In order to solve this problem, people have been inspired by the study of animal groups and invented the swarm intelligence algorithm.[1] Swarm intelligence refers that individuals with simple intelligence, which through mutual cooperation and organizations to show the characteristics of the behavior of group intelligence. In 2005, Karaboga successfully applied the principle of honey bee harvesting to function optimization, and put forward the artificial bee colony algorithm (ABC)[2]. Artificial bee colony algorithm solves the problem of solution space of combinatorial optimization problem[3], but for larger problems, the time in finding the solution is long and the convergence is slow.