One of the biggest drawbacks with things like ACO (ant colony optimization) are actually more in dynamic environments rather than static ones like these. Pheromone trails tend to reinforce the information from the old topology rather than the new topology and you need to add more wrinkles to the algorithm to better adapt to a changing environment.

When it comes to static problems like TSP, the translation of the problem has a lot of influence on how likely the heuristic is to fall into local optima. For instance in the case of ACO you have pheromone trails that are intended to dissipate faster as the distance between food increases. When translating something like TSP you have a decision to make: you can go literal and use the physical distance between edges on a graph (causing closer nodes to be more heavily weighted). Or you could use the total distance of the circuit (causing edges that occur in better solutions to be more heavily weighted). Or you can blend them both -- which is what it looks like the author did.

These algorithms are heuristics and are not intended to necessarily give you the "best" answer, but to get you a "pretty good" answer on messy problems that don't neatly fit into a problem that has been heavily researched like TSP.

not really, depends on the evaporation rate of the pheromone. If you increase the evaporation co-efficient it's more suited to a dynamic environment

It's much more nuanced than that though. Increasing the evaporation rate also makes exploitation much harder, not to mention that the positive feedback cycle on previously very good edges can take a long time to dissipate.

ACO for dynamic environments generally require much more attention to how the problem is applied to the environment or the generation of a more hybrid approach.

Can you provide a source for that? Would like to read that paper.