Extended Experiments with Ant Colony Optimization with Heterogeneous Ants for Large Dynamic Traveling Salesperson Problems

In this work we study the ACS with heterogeneous ants approach to big dynamic problems. When building solutions ACO algorithms rely in two sources of information: static heuristic information about the instance being solved and dynamic trail information acquired during the execution. Conventional ACS always use both sources of informations, ACS with restart clears the trail each time a change occurs and so, immediately after each change, it relies solely on the heuristic information, the heterogeneous ants, or multi-caste, approach, as implemented for this work, have the ability to either use both sources of information or none. We compare the performance of various variants and configurations against both the conventional ACS and conventional ACS with restart, and analize the strengths and weaknesses of each when applied to a set of instances of and dynamic scenarios.