Ant Colony Optimization Artificial Ants as a Computational Intelligence Technique Marco Dorigo, Mauro Birattari, and Thomas St¨utzle Universit´e Libre de Bruxelles, BELGIUM September 2006
Swarm is an approach for problem solving technique that takes in account of social behavior of insects, birds animals etc. The natural behavior of ants helps us in many successful optimization techniques. one is known as ANT colony optimization
What ants do is they produce pheromone on the path they travel from which other family members trace that pheromone to find favorable path.
This article introduces us to ACO and its application. Section I gives us knowledge about natural behavior Section II introduces us to ACO. Section III surveys theoretical results of ACO, and Section IV gives us its application. Section V is about some current hot research topics, and Section VI gives us the overview of other algoritm that is somehow related to the behavior of ants. Section VII concludes the article.
Deneubourg investigated an experiment “double bridge experiment” in which there are two bridges one longer than the other one that connects ants’ nest to their food source. The ant that chooses the shorter bridge are the ones that reaches the nest first that is why the shorter bridge receives pheromone earlier than the other bridge and this fact somehow increases the probability for the further ants of their family member to choose that particular shorter bridge.
A model was proposed for this behavior to find the probability for ants to choose shorter path
Where m1 ants chooses the first bridge and m2 chooses the second bridge , monte carlo simulations shows good result for k=20 and h=2.
Application of ACO
Lots of researchers are applying ACO for N-P hard problems and only few are concern with dynamic and stochastic aspects as well as multiple objectives. In near future how to apply ACO to these problems may emerge as a major research.
Around 15 years from now when this idea was first coined it may seems to be crazy idea. But now many applications that are given above changes are thinking. This approach of using ants’ behavior approximate solution of difficult optimization problems.
Ant Colony Optimization Saad Ghaleb Yaseen Nada M. A.AL-Slamy June 2008
ACO is bio-simulation because of their relative individual simplicity & composite group behavior. This paper takes account of ACO into TSP. It takes a case study from Amman Seaport to derive the shortest route for good transportation inside Amman.
TSP ant system
This above frmula is used to obtain ant routing table for node i and j where Ni is set of all neighbor node
is pheromone trail and is local heuristic value . Where ? and ? are parameters that control the relative weight of pheromone trail and heuristic value
The probability Pij (t) with which at the t-th algorithm iteration an ant k located in city I chooses the city j ? N to move to is given by the following above probabilistic decision rule.
The functioning of an ACO algorithm can be summarized as follows:
Ants move by using stochastic local decision policy based on following parameters, attractiveness & trails. Each ant’s trail contributes to the solution. During construction phase ant modifies the trail value Updated information of pheromones helps future ants’ trails. In ACO, an artificial ant builds a solution by traversing the fully connected construction graph G (C, L), where C is a set of vertices and L is a set of edges. An artificial ant moves through random edge and vertices and deposit pheromone on their trail to contribute towards promising result.
Ant Colony System: A Cooperative Learning Approach to the Traveling Salesman Problem Marco Dorigo , Luca Maria Gambardella
This is how ant chooses their paths. At first some ants choose longer path some shorter the choice is purely random. Approximately ants’ speed is nearly same those whose chooses the shorter path reaches the destination early than those who chooses the longer path. On the shorter path pheromone accumulation is more .
An intuitive explanation of how ACS works, which emerges from the experimental results presented in the preceding sections, is as follows. Once all the ants have generated a tour, the best ant deposits (at the end of iteration t) its pheromone, defining in this way a “preferred tour” for search in the following algorithm iteration t+1. In fact, during iteration t+1 ants will see edges belonging to the best tour as highly desirable and will choose them with high probability. Still, guided exploration (see Eqs. (3) and (1)) together with the fact that local updating “eats” pheromone away (i.e., it diminishes the amount of pheromone on visited edges, making them less desirable for future ants) allowing for the search of new, possibly better tours in the neighborhood5 of the previous best tour. So ACS can be seen as a sort of guided parallel stochastic search in the neighborhood of the best tour.