Suggested GA operators for a TSP problem?

Question

I'm building a genetic algorithm to tackle the traveling salesman problem. Unfortunately, I hit peaks that can sustain for over a thousand generations before mutating out of them and getting better results. What crossover and mutation operators generally do well in this case?

Answer 1

Ordered mutation and ordered cross-over (see this article). Standard mutation and cross-over operations will usually result in invalid solutions (i.e. duplicate and/or missing cities in a route).

There was a similar question recently.

I have a Java applet that implements the TSP using ordered cross-over and mutation, if you are interested in comparing the performance of your implementation.

Answer 2

If your problem is that peaks remain for over one thousand generations, then the problem might not be with the crossover and mutation operators. You might not be introducing or keeping enough variation to your population: I would examine the proportions of crossovers, of mutations, and of survivors from one generation to the next, and possibly raise the proportion of mutations.