How to disable the random stream control in the Optimization conducted in the Experimenter
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report
Hi all!
I'm learning the FlexSim, especially the Experimenter part. I have created an optimizatio job inside the Experimenter. After setting all the parameters, I run the optimization and get the results. From the results I can know the "optimial" solution and the associated objective value. I know FlexSim uses the OptQuest to perform the optimization, and some heuristic optimization algorithms are used in OptQuest.
So I know that with limited number of iteration, this found solution may not be the true optimal solution. Moreover, the objective value on each explored solution is also noisy (subject to randomness) because many things (such as arrivals, service times) are randomly generated in the model. In the optimization setup, I purposely set a small number of replications.
Now, a natural question is, whether the above optimization result is reliable? Or it is just misled by some random noises? To see that, I want to re-run the optimization job. I hope that in this run, different random stream can be used. If different runs give me close "optimal" solutions and their objective values, then I will say that the optimization result is reliable; otherwise I can not trust the results of any run (because the noises are too large) and the optimizaton parameters should be improved.
However, I find FlexSim always gives me exactly the same results, when I run the optimization jobs for several times. I believe the reason is FlexSim uses the same random stream when I re-run the optimization job. I have already unchecked the "Repeat Ramdon Streams" under the Statistics menu. I'm not sure if there is any way to disable the random stream control in the Optimization conducted in the Experimenter. I believe it is an important issue.
I have attached the toy model. For this model, the optimal solution is obvious. But with limited replications, the estimation of the objective value may be very inaccurate. So I use it to study the above mentioned issue.
Thanks!
