Intelligent Engineering Systems through Artificial Neural Networks, Volume 20
31 Gaussian Random Evaporation in Ant Colony Optimization
Download citation file:
- Ris (Zotero)
- Reference Manager
In ant colony optimization methods, the central pheromone trails data structure τ can be thought of as representing the collective wisdom of the population. In each iteration, after each ant generates its candidate solution based on τ and η (a problem-dependent heuristic function), the τ structure is then updated to integrate the experience gained from the solutions constructed into the collective wisdom of the group. First, each entry of the τ data structure is reduced by a fraction ρ, where ρ is an evaporation rate constant, and then some entries of τ are reinforced based on the solutions constructed. We propose a variation in which instead of ρ being constant, each entry of τ is reduced by a Gaussian random number N(ρm, ρs). We evaluate our proposed variation in the context of MAXMIN Ant System and a 1,173-city instance of the traveling salesman problem, and find that our variation improves solution quality.