Posted Sun, 21 Feb 2021 19:47:59 GMT by Felipe Gutiérrez

I'm quite new with FEPEST, the thing is that I run the model to estimate calibrated values for Kx, with Ky and Kz tied to this parameter (kx=Ky=Kz). I have also 10 obs points. Only K of only one lithology is being optimized ('Marlstones') spatially, whereas 'Dolomites' is going to have a constant value.. Also, I've added 20 random pilot points in the whole model.

The thing is that, it's been more than 90 hours running, and the model seems to be unstable, since Phi is going upwards after opti-iteration 7. Why is this? is it normal for FEPEST to run this slow? I attach here some images. [b]Also, this is a FEFLOW model with steady-state flux only. [/b] the FEFLOW model runs in like 1:30 min or something like that (or maybe less).

In the tutorial there's an example and the model was ready after almost an hour... am I doing something wrong?

Please any comment is GREATLY appreciated.
Posted Wed, 24 Feb 2021 17:26:55 GMT by Igor Pavlovskii Dalhousie University Post-Doctoral Fellow
I see two unrelated issues
1) each run appears to take 45 min (based on screenshot 94h / 129 runs) instead of 1.5 min mentioned.
The possible cause is that steady state has trouble converging at some combinations of hydraulic conductivities. I would constrain possible range of K values (have you changed the default range?).

2) 20 pilot points mean it takes 20++ model runs to do one iteration. At least for testing I would reduce number of points. Though, here I actually would start with uniform values at different zones to catch other errors before going to reduced number of pilot points

Posted Wed, 24 Feb 2021 18:48:16 GMT by Felipe Gutiérrez
Hi, thanks for your reply.

In fact, the FEFLOW run takes exactly 4 min to finish (steady state, water flow only).

This FePEST run I was asking for, ended after 130 hours of simulation. (sadly the results weren't satisfactory, due to constraints). The model calls aren't related to the time of simulation in FEFLOW, due to the 4 min don't match with the calculation you just made. It's a different thing.

And, indeed... by adding pilot points the time required skyrockets.

Kind regards.

You must be signed in to post in this forum.