Am uploading this 20 page PDF for safekeeping. It is also worth reading.
-
DeCarolis, Joseph F, Sauleh Siddiqui, A LaRose, J Woollacott, Cara Marcy, Chris Namovicz, J Turnure, K Dyl, A Kahan, J Diefenderfer, N Vincent, B Cultice, and A Heisey (October 2024). A new generation of energy-economy modeling at the US Energy Information Administration — Working Paper 204. Washington DC, USA: US Energy Information Administration (EIA). Public domain status.
-
2024-decarolis-etal-a-new-generation-of-energy-economy-modeling-at-the-us-energy-information-administration-working-paper-204.pdf (626.9 KB)
▢
1 Like
Thanks for sharing. I quickly went through and perceived two main ideas:
- LP optimization is limiting (with quote of @stefan.pfenninger 2014 article), but I didn’t get what it is supposed to be replaced with: I didn’t see MILP mentioned, but HiGHS is cited as solver (MILP/MIQP) and Iptopt (local NLP), along with probable commercial solvers
- Modularity is important, which is being prototyped with Pyomo’s blocks which I have heard about (and seen several discussions on the Julia/JuMP discussion forum, since JuMP doesn’t have an equivaleent), but never practiced
I guess you are referring to this article from @stefan.pfenninger @hawkez99 and James Keirstead:
- Pfenninger, Stefan, Adam Hawkes, and James Keirstead (1 May 2014). “Energy systems modeling for twenty-first century energy challenges”. Renewable and Sustainable Energy Reviews. 33: 74–86. ISSN 1364-0321. doi:10.1016/j.rser.2014.02.003.

yes it is their Renewable and Sustainable Energy Reviews that DeCarolis et al. is citing in the section “2.1. Limitations of linear optimization models”, with the following comment:
While the use of linear optimization represents a convenient methodology, it is challenging to maintain the flexibility to introduce new governing dynamics without compromising the model’s overall mathematical consistency and computational tractability.
Sticking with a single governing dynamic for computational performance or other benefits can end up constraining model development options and precludes more faithful representations of certain real-world dynamics.
However, reading better this section today, I think I got mislead by the section title: the discussion is less about the LP assumption than the “implicit assumption of these [LP] models is that there is an omniscient social planner”. The link with LP is just that minimization the society-wide system is conveniently expressed in an LP.
Still, the discussion about what to do instead is quite general (nonlinear elastic response of tech choices against fuel price changes)