Multiple Ant Colonies Presented by Brandon Borkholder Based on MACS-VRPTW: A Multiple Ant Colony System for Vehicle Routing Problems with Time Windows
Problem Route a fleet of delivery vehicles Minimize cost and time Primary variable is cost/tours Most realistic as vehicles are variable cost
Solution Define two Ant Colony Systems (ACS) One for each parameter
Solution, cont. Cooperate by pheromone updating Pheromone levels indicate how desirable an arc is Opposing influences: exploration and exploitation Ants decide where to go (probabilistically) based on maximal pheromone levels As an ant passes over an arc, it decreases the pheromone
Implementation Procedure MACS-VRPTW() Repeat v ← #active_vehicles(ψ gb ) Activate ACS-VEI(v - 1) Activate ACS-TIME(v) While ACS-VEI and ACS-TIME are active Wait an improved solution ψ from ACS-VEI or ACS-TIME ψ gb ← ψ if #active_vehicles(ψ gb ) < v then kill ACS-TIME and ACS-VEI End While until a stopping criterion is met
Implementation, cont. Find an initial solution for given cost Improve on time until we can improve on cost Then improve on time with that parameter of cost
Implementation, cont. The ants for time might not find a complete solution Import the other edges in a greedy best-first manner
Alternative Methods Exact methods Tabu and large neighborhood searches Works well for constricted time windows Solution space is small
Results
Advantages/Disadvantages Very fast < 30 minutes for ~100 nodes All the advantages of swarm Adaptable, etc. Disadvantages Is a complex system with many concepts Really not optimizing both parameters simultaneously
Questions