Sei sulla pagina 1di 11

Petrel 2012.

1 AHM Tips
Base Case in AHM
All the variables need
to be run as disabled
to produce a base case
that will have both
input (sampled
variables) and output
variables (HM DOF
calculated).

This then needs to be


supplied to AHM both
to previous cases pane
(to avoid duplicate
sampling) and
initialization pane (to
make sure AHM
improved over base
case OF.
Sensitivity in AHM

Parameters that show low response to


AHM OF in Tornado can be removed
from AHM by disabling these
variables.
Uncertainty in AHM

Uncertainty need to be run with Monte-Carlo


Sampling and Latin-hypercube on.
This will give a better initialization for AHM rather
than default random initialization.
And then this should be put into both Previous
Cases pane (to avoid duplication) and to
Initialization pane ( to disable random initialization)
Restart AHM vs Random Initialization

Random Initialization Restart AHM with Uncertainty


(default behavior) (recommended)
Evolutionary Optimization Reports for
Analysis (OF vs Gen Plot)

This information is available from


Individuals.csv file (but not in Petrel RE)
OF vs. Generation plot gives an idea
how OF is improving and how search
space is changing (narrowing )
Modifying HM DOF

If changes are made to HM DOF all the cases in


the Previous Cases pane/Initialization pane
needs to be recalculated. Best to create a new
DOF HM and use Settings/Options
Add Variables

It is possible to add a new


variable (e.g. $WOC) to
AHM workflow, but some
additional work is required As soon as new variable is
added restart is invalidated

User will need to add column in


case variable spreadsheet
($WOC) and fill in base value for
all the restart cases used in AHM
Save Functionality
This Save Functionality will
Save the changes to Petrel/
Cases pane and update with
newly generated Cases

But cases will not saved in


Previous cases pane in
Optimizer tab
User will have to drop to
restart
Integer problem
Evolution strategy optimizations with
discrete parameters are not support
restartable

For discrete variables in U&O


workflows (i.e. those whose 'Int' check
box is checked) the optimizer generates
floating point numbers which are then
rounded to the nearest integer by the
uncertainty and optimization
framework. The rounded number is
used to run the workflow and recorded
in the resulting case's case
variables. This introduces a difference
between the numbers generated by
the ES and the corresponding case
variable. This difference prevents the
optimizer from recognizing cases it has
previously generated and thus
subsequent optimizations will duplicate
This is a limitation if you like to choose Evolution the cases produced in the first run even
Strategy for tasks as optimizing number of laterals or when no settings have changed and
number of wells (discrete) the previous cases data is available
Case Management with AHM
AHM runs tend to create a large
number of cases and user can do
the following:
It is possible to restart even
though cases are not in .sim
directory, Petrel AHM will continue
as long as cases are in Cases pane
(without results) with case
variable information still available

Potrebbero piacerti anche