I have made use of an evolutionary algorithm that performs randomized search. How many iterations are needed to ensure that I will find a very good solution close to the global optimum?
you first fix 1000 iterations and see whether the results get converged. if not so, increase the number until you get the global best solution and the results get converged.
i think it helps.
Getting optimal solution is convergence of results. If the deviations are very negligible there iteration can be stopped. In any optimization process like GA, MOGA, NSGA, NN and PSO etc .. the results are obtained on convergence, not by fixing iterations. Also for each run you will get global optimal solutions for varying iterations.
You need to decide what change in your optimum value is negligble to the value you are optimizing. You can then monitor the difference between successive iterations and once the difference between values falls below you pre-decided cut off value you can terminate the optimisation.
There is no strict or clear rule for determining the total number of generations to get an optimal or near-optimal solution. Not only number of generations, but many parameters (or algorithm's characteristic components) affect on the quality of the final solution. Fixing values of all, but total generation numbers, means you have to start with a very large number (considering all related works) and investigate where global convergence and/or premature convergence appears. If global convergence happens then you can decide how long generations could be. On the other hand, if premature convergence occurs, it means that you need further investigations and modifications for other algorithm's parameters.
It depends on the problem you want to solve. You must try some efficient algorithms. The most efficient algorithm (reported in the publications) maybe be inefficient for your problem. You should try several algorithm for various parameter settings (including the iteration number).
- Badges
- Science topic