Heterogeneous Ant Colony Optimisation Methods and their Application to the Travelling Salesman and PCB Drilling Problems

Abstract

Ant Colony Optimization (ACO) is an optimization algorithm that is inspired by the foraging behaviour of real ants in locating and transporting food source to their nest. It is designed as a population-based metaheuristic and have been successfully implemented on various NP-hard problems such as the well-known Traveling Salesman Problem (TSP), Vehicle Routing Problem (VRP) and many more. However, majority of the studies in ACO focused on homogeneous artificial ants although animal behaviour researchers suggest that real ants exhibit heterogeneous behaviour thus improving the overall efficiency of the ant colonies. Equally important is that most, if not all, optimization algorithms require proper parameter tuning to achieve optimal performance. However, it is well-known that parameters are problem-dependant as different problems or even different instances have different optimal parameter settings. Parameter tuning through the testing of parameter combinations is a computationally expensive procedure that is infeasible on large-scale real-world problems. One method to mitigate this is to introduce heterogeneity by initializing the artificial agents with individual parameters rather than colony level parameters. This allows the algorithm to either actively or passively discover good parameter settings during the search. The approach undertaken in this study is to randomly initialize the ants from both uniform and Gaussian distribution respectively within a predefined range of values. The approach taken in this study is one of biological plausibility for ants with similar roles, but differing behavioural traits, which are being drawn from a mathematical distribution. This study also introduces an adaptive approach to the heterogeneous ant colony population that evolves the alpha and beta controlling parameters for ACO to locate near-optimal solutions. The adaptive approach is able to modify the exploitation and exploration characteristics of the algorithm during the search to reflect the dynamic nature of search. An empirical analysis of the proposed algorithm tested on a range of Travelling Salesman Problem (TSP) instances shows that the approach has better algorithmic performance when compared against state-of-the-art algorithms from the literature.