Does anyone know of applications/variations of Simulated annealing (or similar search strategies) where the search space is dynamic, i.e. changes over time between iterations ("generations" in genetic algorithms)? Like travelling salesman but the roads keep changing.

Maybe there's no fundamental change required? Just keep searching for better solutions and possibly accept worse ones with some probability, even if the space changes? The problem I'm seeing in my area is defining the neighborhood function when possible paths change over time.