Please enable JavaScript.
Coggle requires JavaScript to display documents.
ESA Week12 - Coggle Diagram
ESA Week12
Multi Obj Optimisation
-
-
-
challenges:
- need to converge to global optimal
- find sufficient coverage of set
- even distribution of solutions
-
-
implementation
-
write the nsga function (this is where you define your box constraints in upper/lower bounds)
set num of individuals and generations : rlly by trial and error
note NSGA2 doesnt have a defined seed
-
notes:
-
sizze of circles to match between left and right graph for the activity (to show correspondence between the decision space and obj space)
the more the iterations, the more it will converge --> can print out solution to check if they converge
-
for nsga2: to find the optimal solution you can use utopian point method. if you want to maximise, (the default is minimise), multiply by -1 when defining it
black box optimisation
-
-
feed in decision variables, then see whats the output of the model
use graphing tools / closest to utopian point method to pick one solution of many pareto optimal points on the frontier --> the solution is not necessarily the best one --> it just represents the one that is closest to the utopian point
-
population size = no of individuals = how many individuals you want to test per simulation run... each individual will be represented by all the decision variables --> and you use some sort of selection to retain the best individual while kicking out the poorer ones
homework
when you check for scaling, you can print out the sol data frame (the one with the abs values) and see the min/max there then you can find the term for the min/max to rbind
-
-
Genetic Algorithms
Constraints
If no constraints, then OF = Fitness Function
If constraints exist, then Fitness = OF + Constraints
-
-
-