Stochastic Programming

[João Gorenstein Dedecca]

Tom, Jonas and David,

Have you discussed implementing stochastic programming into PyPSA? This could be a very useful feature which may not require major overhauls.

In my view currently it is not an actual stochastic model because all snapshots are linked by the storage and store variables. However, if snapshots had a variable to indicate their temporal linkage, one could apply storage and store constraints by temporal groups, with the snapshots being then actual probabilities instead of only occurrence length weights. With standing losses perhaps one would need to separate probabilities and weights.

Best,

João

[Tom Brown]

Hi João,

We have thought about SP a little bit, but haven't needed it yet for our
own projects. You can use the weights as probabilities, as suggested here

http://pypsa.org/doc/optimal_power_flow.html#stochastic-optimisation

but you're right this isn't proper SP if you have storage, since the OPF
uses the weights as a measure of the passing of time.

Would you suggest something like a series map for the snapshots into
different groupings e.g.

Snapshot Group
1a a
2a a
3a a
1b b
2b b
3b b

and then linking each group using the weights? I guess one would only
need to modify the code by breaking the inter-linkage between each
group/making each group cyclic.

I must say I'm a little fuzzy on the different uses of SP - it can be
used for investment decisions and e.g. operational optimisation with
e.g. discrete decision variables "should I run my lignite plant
tomorrow", then there's the whole recourse stuff...do you know any other
free software with a simple implementation?

Best,

Tom

> --
> You received this message because you are subscribed to the Google
> Groups "pypsa" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to pypsa+un...@googlegroups.com
> <mailto:pypsa+un...@googlegroups.com>.
> To post to this group, send email to py...@googlegroups.com
> <mailto:py...@googlegroups.com>.
> To view this discussion on the web, visit
> https://groups.google.com/d/msgid/pypsa/CAPpM__FLDdYMRp6sRPED5Av7RW2dthfpYbnVrGA0VQh90KgRFA%40mail.gmail.com
> <https://groups.google.com/d/msgid/pypsa/CAPpM__FLDdYMRp6sRPED5Av7RW2dthfpYbnVrGA0VQh90KgRFA%40mail.gmail.com?utm_medium=email&utm_source=footer>.
> For more options, visit https://groups.google.com/d/optout.

--
Tom Brown
Postdoctoral Researcher
Frankfurt Institute for Advanced Studies
https://fias.uni-frankfurt.de/physics/schramm/complex-renewable-energy-networks/
Personal site: http://www.nworbmot.org
Phone: +49 69 798 47693

[Joao Gorenstein Dedecca - TBM]

Hi Tom,
Exactly, something like your groupings should work. I'm not aware of free implementations of it, and a quick look at http://wiki.openmod-initiative.org/wiki/Open_Models didn't bring anything up, except by using the academic license of PLEXOS.
I will test the grouping approach with PyPSA and see how it works, and eventually how fast it is for multiple stochastic scenarios.
Best!

João

To unsubscribe from this group and stop receiving emails from it, send an email to pypsa+un...@googlegroups.com.
To post to this group, send an email to py...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/pypsa/6c682165-add2-7f5a-1362-c7026c8f610d%40fias.uni-frankfurt.de.

[c.d. mclean]

re free stochastic programming implementations ...

this may be of interest:

   https://groups.google.com/forum/#!searchin/julia-opt/stochastic$20programming|sort:relevance/julia-opt/g5LH9RW587k/_X6ugYhu9kIJ

bear in mind that some Python - Julia interops

are implied ( a la https://github.com/JuliaPy/pyjulia)

best,

cdm