UQWorld

Using screening methods to reduce input of UQ analysis

Hello all,
I have a problem with multiple inputs and the evaluations of the main function is also costly.
I want to use Sobol indices with PCE to find the impact of each of the parameters on the output, and I have a question in this regard:
Is there a way so that I can get an estimation of what parameters are the most important ones to use them in my UQ analysis and fix other inputs? I think there should be some screening methods that use some sparse evaluations to give an estimation of the impacts of each of the parameters on the output. I will then run my model to find the Sobol indices for these important inputs rather than all of the inputs.
Can I use Morris method for this aim, or are there any other methods for this?
Thank you in advance.

Hello all,
Any comments on how I can solve the problem I mentioned with uqlab is greatly appreciated.

Hi @Aep93,

Welcome to UQWorld! :smiley:

Yes indeed, Morris screening method is available in UQLab. You can have a look at the Sensitivity User Manual Section 2.1.6 as well as the Method Overview Example of the Sensitivity Analysis module.

Let me know if you have any usage questions!

That said, I’m a bit curious about how costly is your simulation? How many runs that you can comfortably afford for the screening analysis? If the number of model runs is your only concern and if you’re going to use PCE to compute the Sobol’ indices, you might not need that many runs in the first place.

1 Like

Hello @damarginal. Thank you for your answer. Yes, now I think that if I use PCE I may not need the Morris method anymore.