Simulated Annealing isn't a GA, so there isn't any genetic representation.
Ok, here's a lengthy explanation.
Basically a simulated annealing technique is similar to a hill-climbing searching algorithm, where you always make a move that would improve the situation. For example, always picking a faster model. However, you may reach a point, where there isn't a move that you can make that improves the situation. This could be the most optimised solution, but it can also be what is called a local maxima, which is like the top of the hill, but not the tallest one.
So basically simulated does basically the same thing, except it picks a move at random. If this move makes the situation better, it is imediately accepted. If it isn't, then there is a possibility of it being accept. This possibility lessens with how bad the move makes the situation. Over a period of time the acceptance of these accepting these worse moves is gradually decreases. Much like the temperature in the actually metal annealing process. Hopefully this should escape local maximas and gradually converge to the global maxima.
So basically when i create my random models to select my move from, I'm basically just changing the values in the model, which is why i need a base model and can optimise any model.
|
