+- jeplus.org forums (http://jeplus.org/mybb)
+-- Forum: Building simulation tools (http://jeplus.org/mybb/forumdisplay.php?fid=1)
+--- Forum: jEPlus+EA (http://jeplus.org/mybb/forumdisplay.php?fid=3)
+--- Thread: Defining population size and max. generations in jEPlus+EA (/showthread.php?tid=94)
Defining population size and max. generations in jEPlus+EA - Susanneb - 03-24-2015
Hi all,
I am not an expert on EA but an architect and have troubles to define the population size in jEPlus+EA. As I understood more design variables requires larger population size. But you wrote the choice is also influenced by the number of processor cores you have for running these simulations.
Running the same model for different climate, I found that the total number of "Best solutions" can be very different.
Is there any formula how to calculate the population size? My model has now 14 variables and 3 objectives. I was also wondering how the order of the objectives influences the generation of solutions?
Warm regards,
Susanne
RE: Defining population size and max. generations in jEPlus+EA - navid - 03-25-2015
Hi my dear Susanne
Your question is very common. Anyone who asks me this question, I have the best answer for her/him.NO,There is no formula for calculating Population size.However, there are several tricks in optimization problems for number of crossover,number of mutation and population size.for instance,Some optimization experts believe that choosing 4 to 6 times the number of decision variables is Appropriate for Population size.But In my opinion,based on my experience,the number of Population size, Mutation, generations and crossover selections for both single and multi-objective functions are completely random. Basically, they will depend on the type of your optimization problem. For example, for an optimization problem case(A),it may be optimized perfect just with Population size( = 10) , Mutation(=0.05) ,generations(=10) and crossover(=0.7) ,while for another optimization problem case(B), it may be optimized perfect with Population size( = 200) , Mutation(=0.2) ,generations(=100) and crossover(=0.6). The most important point is that you have to run your optimization problem for different selections and find the best selection .it is very common.
Based on my experience in the optimization, Choosing among Population size( = 20 to 50) , Mutation(=0.05 to 0.2) ,generations(=10 to 100) and crossover(=0.55 to 0.8) work very well.
If you still have any questions about optimization, I am very happy to ask me your questions.
Cheers
Navid Delgarm