Perhaps someone should write a book, "The Meme Delusion".  Just joking.  This sounds a lot like lateral gene transfer to me.  Could you add some more information, please?

Hi Douglas,

I am using hyper-heuristics to evolve an Memetic algorithm.  These generated Memetic Algorithms  are compared against a hard-coded one.  Now I have been tuning the restart element of these algorithms and I am concentrated to the hard-coded one. I am not convinced I am doing that correctly. 

So I am asking how you would tune the restart of a memetic algorithm. 

Many  thanks

Patricia 

The TSP is known to be globally convex -- see the link.  Since good local optima share many similarities with other good local optima, you in general want to concentrate search around the best solutions (which are all local optima in a memetic algorithm).

If your population stalls (e.g. no improving solutions in a generation), you should not only restart your population, but maybe try to focus on new population members that are as different as possible from the stalled population -- i.e. concentrate search an in a distinctly different part of the search space (because many real-world TSP instances have "structures" in which different clusters of local optima exist).

http://www.crt.umontreal.ca/~pesant/BIRS/sellmann-TSPsolnDistribution.pdf

That is very interesting thank you very much! 

Patricia