Abstract:
|
The algorithm developed in this work differs strongly from the other Ant algorithms in the literature both in the approach and in new elements introduced. For the way the solutions are built, the algorithm is called Extending Ant algorithm.
The high-level procedure that governs the algorithm is quite classical (see Algorithm 1): in each generation, several feasible solutions are built; then, for the best solutions of the generation, three classical Local Search methods are applied (2-opt and forward and backward Or-exchange); and these solutions are used to update the Trail system indicating the likelihood of two customers being consecutive in the best solution. In addition, the minTrails value (used as threshold to switch to the 2nd phase of the construction procedure) and the probabilities for each route of the best solution to be copied into new solutions (at the beginning of the construction of every solution) are computed. |