Modeler’s Corner
Tips for Modeling
Multiple Scenarios or Alternatives
Rick Kossik
Principal
GoldSim Technology Group
rkossik@goldsim.com
Many GoldSim models are intended to compare various alternatives or
options, or to run multiple scenarios for a particular system. What is
the best way to model multiple scenarios or options in a GoldSim model?
There are four ways to approach this:
- Run all of the scenarios/options simultaneously in a single
model.
- Run each scenario/option separately using a scenario/option
flag.
- Run all of the scenarios/options in series as multiple
realizations.
- Run all of the scenarios/options in series as multiple realizations
using SubModels
Each of these four approaches (and their pros and cons) are discussed
below.
1. Run All of The Scenarios/Options Simultaneously in a Single Model
In this approach, you build all scenarios into a single model.
Typically, this means having a localized Container for each
scenario/option. By using cloned Containers for the primary parts of the
algorithm in each Container, you can ensure that when you edit it, it
only needs to be edited in one place.
NOTE: To learn more about cloning Containers, open the GoldSim Help
system, go to the Index tab, and search for "Clones" and select
the subtopic "Containers".
To ensure that each scenario/option uses the same random number seeds
for Stochastic variables, all common Stochastics should be modeled as
global parameters (at the top level of the model)
Pros: All results are in a single model. Simulation can be
probabilistic.
Cons: This approach works well for simple models, but breaks down
if your model is very large and complex.
2. Run Each Scenario/Option Separately Using a Scenario/Option Flag
In this approach, you build a single model that simulates one
scenario/option at a time. The scenario/option being simulated is
determined by a single Data element (a flag). If statements (and/or
Selectors) are then set up that are functions of the flag (and switch the
logic appropriately for each option).
NOTE: Starting with Version 9.5, the scenario flag could also be a
Global Property (defined in the Globals tab of the Simulation Settings).
In order to view all the option results simultaneously, it is
necessary to export the results for each run (e.g., to a spreadsheet).
Pros: Can handle large, complex models. Simulation can be
probabilistic.
Cons: Results for different scenarios/options must be exported to
be viewed simultaneously.
3. Run All of the Scenarios/Options in Series as Multiple
Realizations
In this approach, you build a single model that simulates one
scenario/option at a time (just as in Approach #2). Furthermore, the
scenario/option being simulated is determined by a single element (a
flag). If statements (and/or Selectors) are then set up that are
functions of the flag (and switch the logic appropriately for each
option).
The difference with Approach #2 is that that flag is not a constant
Data element, but is instead an Expression defined as
"Realization". The model is run in Monte Carlo mode, with the
number of realizations equal to the number of scenarios/options. Of
course, this means that true Monte Carlo (probabilistic) simulations of
the scenarios/options cannot be carried out.
Pros: Can handle large, complex models. All results are in a
single model.
Cons: Simulation can not be probabilistic.
4. Run All of the Scenarios/Options in Series as Multiple
Realizations Using SubModels
This Approach is identical to Approach #3, but uses GoldSim's SubModel
feature (introduced on Version 9.5).
NOTE: To learn more about SubModels, open the GoldSim Help system, go
to the Contents tab, and under "Advanced Modeling Concepts", go
to the topic "Using SubModels to Embed Models Within Models".
In this approach, the "inner model" is the scenario/option
(which can be run as a Monte Carlo simulation). The "outer
model" is run in Monte Carlo mode, with the number of realizations
equal to the number of scenarios/options. The flag is the primary element
in the "outer model" that is sent to the "inner
model".
It is important to note that the outputs of a SubModel are statistics
(e.g., a mean or percentile). They cannot be time histories. As a result,
this approach only works if the scenarios/options can be compared in
terms of statistics (e.g., mean peak water level, 95th percentile of
cumulative cost), rather than by directly viewing and comparing time
histories.
Pros: Can handle large, complex models. All results are in a
single model. Simulation can be probabilistic.
Cons: Cannot view time histories of individual scenario results.
Suggestions?
Do you have any ideas to share for simulating different
scenarios/options, or suggestions for future Modeler's Corner articles?
If so, I'd love to hear from you. Please contact me directly at rkossik@goldsim.com.
|
