Hello,
Can someone please tell how to do "sensitivity analysis"
for a lookup function using a Hearne's method ?
Hearne' method
Re: Hearne' method
I assume you mean this?
https://onlinelibrary.wiley.com/doi/abs ... 09.00058.x
The paper is paywalled, but I'm going to speculate that in general it would be tricky to implement. You'd have to have a library of functions with desired properties that meet all the reality checks needed for them to make sense in context.
A simple method for doing sensitivity analysis on a lookup is easy, if the lookup is properly normalized to begin with, like:
price = reference price * effect of inventory lookup( inventory / reference inventory )
With this construction, you can already shift and scale the effect using the reference parameters. Then you could add additional distortions of the effect, like:
price = reference price * effect of inventory lookup( inventory / reference inventory ) ^ sensitivity scale
(Bear in mind that this has infinite slope at 0 if scale < 1.)
Or, you could mix two different shapes:
price = reference price * (weight to A*effect of inventory lookup A( inventory / reference inventory )
+ (1-weight to A)*effect of inventory lookup B( inventory / reference inventory ))
https://onlinelibrary.wiley.com/doi/abs ... 09.00058.x
The paper is paywalled, but I'm going to speculate that in general it would be tricky to implement. You'd have to have a library of functions with desired properties that meet all the reality checks needed for them to make sense in context.
A simple method for doing sensitivity analysis on a lookup is easy, if the lookup is properly normalized to begin with, like:
price = reference price * effect of inventory lookup( inventory / reference inventory )
With this construction, you can already shift and scale the effect using the reference parameters. Then you could add additional distortions of the effect, like:
price = reference price * effect of inventory lookup( inventory / reference inventory ) ^ sensitivity scale
(Bear in mind that this has infinite slope at 0 if scale < 1.)
Or, you could mix two different shapes:
price = reference price * (weight to A*effect of inventory lookup A( inventory / reference inventory )
+ (1-weight to A)*effect of inventory lookup B( inventory / reference inventory ))
/*
Advice to posters (it really helps us to help you)
http://www.ventanasystems.co.uk/forum/v ... f=2&t=4391
Blog: http://blog.metasd.com
Model library: http://models.metasd.com
Bookmarks: http://delicious.com/tomfid/SystemDynamics
*/
Advice to posters (it really helps us to help you)
http://www.ventanasystems.co.uk/forum/v ... f=2&t=4391
Blog: http://blog.metasd.com
Model library: http://models.metasd.com
Bookmarks: http://delicious.com/tomfid/SystemDynamics
*/
Re: Hearne' method
Hi Tom,
Yes the link goes to the Hearne's method.
and thanks for other useful description.
Yes the link goes to the Hearne's method.
and thanks for other useful description.
Re: Hearne' method
Is it possible to use this method in an electricity market. I have a large-scale model which takes hours for one run, so I even cannot use Monte Carlo for sensitivity analysis. That's why I was wondering do you have any experience for using this method on models related to energy systems?tomfid wrote: ↑Mon Apr 15, 2019 1:49 pm I assume you mean this?
https://onlinelibrary.wiley.com/doi/abs ... 09.00058.x
Re: Hearne' method
It seems impractical in a large model - just too many parameters to explore.
/*
Advice to posters (it really helps us to help you)
http://www.ventanasystems.co.uk/forum/v ... f=2&t=4391
Blog: http://blog.metasd.com
Model library: http://models.metasd.com
Bookmarks: http://delicious.com/tomfid/SystemDynamics
*/
Advice to posters (it really helps us to help you)
http://www.ventanasystems.co.uk/forum/v ... f=2&t=4391
Blog: http://blog.metasd.com
Model library: http://models.metasd.com
Bookmarks: http://delicious.com/tomfid/SystemDynamics
*/
Re: Hearne' method
Although he has not addressed what large model means from Author's point of view, he has mentioned in the paper that "The application of
this simpler method before resorting to the triangular functions might yield computational savings, which would be important for very large models. pp. 13"
Re: Hearne' method
This is an old thread, but it seems to still be applicable.
This Paper https://onlinelibrary.wiley.com/doi/abs ... 2/sdr.1518
Was extremely useful in understanding and applying the Hearne' method.
This Paper https://onlinelibrary.wiley.com/doi/abs ... 2/sdr.1518
Was extremely useful in understanding and applying the Hearne' method.