Ant Colony Optimization with Re-Initialization

Abstract

This contribution introduces an Ant Colony Optimization (ACO) algorithm with re-initialization mechanism. The whole search process is broken by re-initialization into shorter semi-independent steps called “macro cycles”. The length of macro cycle depends on pheromone accumulation and can be adjusted by a user parameter. It is shown that re-initialization mechanism prevents ACO algorithm from pheromone saturation and consecutive stagnation. This approach avoids overhead caused by algorithm run with excessive pheromone values where further exploration is hardly possible. The solution offers lower CPU cost of the search process and enables automation of heuristic search especially in changing environments like dynamic networks. The efficiency of proposed method is demonstrated on a path minimization problem on 50 node graph.