Darwin Designer GA parameters

Question

There are various GA parameters in Darwin Designer. I want to know the use of all of them and how will they affect the solution...

Wayne Hartell · Accepted Answer

The GA parameters are used to control the underlying genetic algorithm used in the optimization process. 
 
The help topic entitled "Advanced Darwin Designer Tips" contains some useful information on the nature of these parameters. I recommend that you take some time to review that topic in detail. 
 
In general though, the most common parameters for a user to change, would be simply population size and random number seed. 
 
Random Seed: 
All other variables (and input) being the same, using the same random number seed produces the same optimized result(s). Is thus follows that it is necessary to change the random number seed in order to test the sensitivity of the random starting population to the solutions of the problem being optimized. Or, in other words, if changing the random number seed has little effect on results, then you can be fairly confident that the optimization is converging and additional runs will likely yield negligible benefit. If changing the random number seed results in very different results (in terms of fitness) then it may be that the problem needs to be run longer (the stopping criteria relaxed) or that the optimization has a configuration problem of some kind. 
 
Population Size: 
With GA optimization having a stochastic nature it may be beneficial to vary population size up or down. Down will result in faster runs, but with less population diversity and potentially less optimized results, but it's a useful way to debug runs more quickly when initially configuring your optimization problem. Up will increase the initial population and genetic diversity so that in theory there is a higher chance, that through having more random starting points, one of those starting points will be very beneficial to the overall optimization. The down side is that the optimization run will take longer to execute and converge. 
 
Other GA Parameters: 
Generally most users will not want, nor need to change these variables as they have already been set to good defaults. You would only really experiment with these values for two reasons. 
 
1. You are being paid for you work by the amount of savings you generate in your design and thus running many optimization runs with slightly different GA parameter values may result in a better optimized result and more money to you or your company. 
 
2. You are an academic and/or researcher who is interested in the effect of the GA parameters in a specific problem domain and you completely understand the underlying algorithms. 
 
There is no blanket statement that I am able to make about the effect of changing any of these variables (excluding random seed) as it will likely be quite unique to the specific model and optimization configuration that has been set up by the user. 
 
My advice for using Darwin Designer is simply to consider the Designer input as an extension of the model in that it defines the limits of the potential design. The pipe sizes, the costs, the hydraulic constraints, but at the same time when you do this, realize that the optimization process will tend to exploit any technicalities (think loop holes) that are incorporated into the problem definition (the way the optimization has been configured). Thus most, if not all users will benefit a great deal more from spending more time considering how they will embody the optimization problem within Designer (sometimes it is necessary to make compromises or other qualitative assessments) in order for Designer to produce valid and usable results, rather than spending time on the potential effect of the change in a GA parameter would have. 
 
I personally worked as a GA design consultant for several years and am happy to answer any specific additional questions you may have. 
 
Regards, 
Wayne.