Ant colonies for traveling salesman problem BioSystems 1997 Present Sherry Y.T.Chen
Auther Marco Dorigo IRIDIA Université Libre de Bruxelles Belgium Luca Maria Gambardella IDSIA ,Department of Electronics and Informatics of Politecnico di Milano 2006/05/22 OPLab, Dept. of IM, NTU 2
Outline Introduction TSP problem ACO (artificial ant) Simulation & Results Conclusion & Future work Reference 2006/05/22 OPLab, Dept. of IM, NTU 3
Outline Introduction TSP problem ACO (artificial ant) Simulation & Results Conclusion Future work 2006/05/22 OPLab, Dept. of IM, NTU 4
Introduction Ants and positive feedback (Dorigo 1992) Pheromone trail deposited on TSP graph Assumption:TSP graph is completely connected 2006/05/22 OPLab, Dept. of IM, NTU 5
Outline Introduction TSP problem ACO (artificial ant) Simulation & Results Conclusion & Future work Reference 2006/05/22 OPLab, Dept. of IM, NTU 6
TSP problem What is TSP problem? All cities were visited once returns to the starting city cheapest round-trip 2006/05/22 OPLab, Dept. of IM, NTU 7
TSP problem Algorithms (I) The Greedy Method Divide-&-Conquer Enumerating Branch & Bound Dynamic Programming Approximation 2006/05/22 OPLab, Dept. of IM, NTU 8
TSP problem Algorithms (II) Simulated annealing (SA) Neural nets (NNs) Annealing-genetic algorithm (AG) Neural nets (NNs) Elastic net (EN) Self organizing map (SOM) Evolutionary programming (EP) Genetic algorithm (GA) 2006/05/22 OPLab, Dept. of IM, NTU 9
Outline Introduction TSP problem ACO (artificial ant) Simulation & Results Conclusion & Future work Reference 2006/05/22 OPLab, Dept. of IM, NTU 10
ACO (artificial ant) Real ants Real ants seems have some memory Real ants are completely blind Real ants live in an discrete environment 2006/05/22 OPLab, Dept. of IM, NTU 11
ACO (artificial ant) Example for real ants 2006/05/22 OPLab, Dept. of IM, NTU 12
ACO (artificial ant) Example for artificial ants t=0.5 t= 1 2006/05/22 OPLab, Dept. of IM, NTU 13
ACO (artificial ant) From real to artificial (i) the preference for paths with a high pheromone level (ii) the higher rate of growth of the amount of pheromone on shorter paths (iii) the trail mediated communication among ants. 2006/05/22 OPLab, Dept. of IM, NTU 14
ACO (artificial ant) :Euclidean distance between i and j :the number of ants in town i at time t :total number of ants. 2006/05/22 OPLab, Dept. of IM, NTU 15
ACO (artificial ant) :intensity of trail on edge (i,j) (1) (2) (3) 2006/05/22 OPLab, Dept. of IM, NTU 16
ACO (artificial ant) transition probability from town i to town j for the k-th ant (4) 2006/05/22 OPLab, Dept. of IM, NTU 17
ACO-Algorithm Initialize, set value Loop and updating NC reached? 2006/05/22 OPLab, Dept. of IM, NTU 18
ACO-Algorithm 1. Initialize: Set t:=0 Set NC:=0 For every edge (i,j) set an initial value τij(t)=c for trail intensity and Δτij= 0 Place the m ants on the n nodes 2. Set s:=1 For k:=1 to m do Place the starting town of the k-th ant in tabuk(s) 2006/05/22 OPLab, Dept. of IM, NTU 19
ACO-Algorithm 3. Repeat until tabu list is full Set s:=s+1 For k:=1 to m do Choose the town j to move to, with probability pkij (t) given by equation (4) Move the k-th ant to the town j Insert town j in tabuk(s) 2006/05/22 OPLab, Dept. of IM, NTU 20
ACO-Algorithm 4. For k:=1 to m do Move the k-th ant from tabuk(n) to tabuk(1) Compute the length Lk of the tour described by the k-th ant τij(t+n)=ρ×τij(t)+ Δτij Update the shortest tour found 2006/05/22 OPLab, Dept. of IM, NTU 21
ACO-Algorithm 5. If (NC < NCMAX) and (not stagnation behavior) then Empty all tabu lists Goto step 2 else Print shortest tour Stop 2006/05/22 OPLab, Dept. of IM, NTU 22
ACO (artificial ant) ACS TSP 2006/05/22 OPLab, Dept. of IM, NTU 23
Outline Introduction TSP problem ACO (artificial ant) Simulation & Results Conclusion & Future work Reference 2006/05/22 OPLab, Dept. of IM, NTU 24
Simulation & Results Compared with other optimization methods 2006/05/22 OPLab, Dept. of IM, NTU 25
Simulation & Results Compared with TSPLIB TSPLIB (maintained by G. Reinelt): http://www.iwr.uniheidelberg.de/iwr/comopt/soft/TSPLIB95/TSPLIB.html 2006/05/22 OPLab, Dept. of IM, NTU 26
Simulation & Results Compared with different candidate lists 2006/05/22 OPLab, Dept. of IM, NTU 27
Simulation & Results Communication determines a synergistic C No-C 2006/05/22 OPLab, Dept. of IM, NTU 28
Outline Introduction TSP problem ACO (artificial ant) Simulation & Results Conclusion & Future work Reference 2006/05/22 OPLab, Dept. of IM, NTU 29
Conclusion & Future work ACO is appropriate to TSP problem Improvement Local optimization Number of ants Specialized ants, tighter reinforcement 2006/05/22 OPLab, Dept. of IM, NTU 30
Reference Ant System_Optimization by a colony of cooperating agents, M Dorigo, LM Gambardella - Evolutionary Computation, IEEE Transactions on, 1997 http://www.neotech-web.com/technology_03.html http://en.wikipedia.org/ http://uk.geocities.com/markcsinclair/aco.html 螞蟻演算法在即時戰略遊戲上的應用-以美式足球為例, 尹邦嚴 2006/05/22 OPLab, Dept. of IM, NTU 31
Q&A Thanks for your listening 2006/05/22 OPLab, Dept. of IM, NTU 32
Are arcs limited the solution? Only ACS + greedy ? 2018/11/13 OPLab, Dept. of IM, NTU
TSP problem Elastic net (EN) 2006/05/22 OPLab, Dept. of IM, NTU 33