simulated annealing for one dimensional real values

When solving a problem lige tsp with annealing, each step you change your
solution into another candidate solution which relates to the current
solution. That is easy enough, but what do you do if your solution is just a
real value? it seems that if I just use a random new value from somewhere in
the valid inputvalues, then I will be jumping arround too much. If i use a
value which is close to the current value, then I will perhaps not search
the landskape good enough and will miss the global minimum. How do you
select a new value? Do you perhaps use some normal distribution arround the
current value and let the distribution be more and more narrow as the
temperature drops, or what?


.

Quantcast