Unnecessary capacity investment

[ahawkes]

Describe the bug

In the attached model, in the 2nd MSY (2021) investment occurs in OILREF and OILRF2. However, as far as I can tell, the additional capacity is not needed. The additional investment corresponds to 1 tranche of capacity for each.

The issue appears to be related to how the activity of the tranches of assets is netted off the total demand profile for each downstream commodity. Somehow, the combined activity of the 3 - 4 tranches of capacity in each asset is less than the potential activity (in final dispatch) of the 3 - 4 tranches combined. As the output commodity is seasonally balanced here, perhaps there's some arbitrariness to where the output falls, leading to some apparent remaining demand after all tranches have been retained in the investment appraisal process? There could be other explanations, but if this is correct, is there a way we can make sure time sliced output is flat in the case of seasonally / annually balanced commodities?

To Reproduce

Run the attached model (on main development branch, as at 09/09/2026)

Expected behavior

No additional capacity investment should occur in these processes.

simple_modified_nostor.zip

[tsmbland]

I haven't looked at your model in detail yet, but I think I've seen this problem before.

I think there's an flaw with the investment appraisal approach where we consider each tranche as if it's its own individual asset, rather than part of a larger asset. In particular, it seems to be a problem for processes with annual/seasonal availability limits. Hard to exactly articulate the problem, but I made an example to demonstrate this here. Hopefully it's clear.

A possible solution might be to appraise the whole asset with each successive tranche added (1 tranche, then 2 tranches, then 3, rather than 1 then 1 then 1), and calculate the marginal LCOX/NPV from each successive tranche. Not sure if that makes sense...

[ahawkes]

Yes that description looks plausible for what's happening here (or something similar). I like the idea of adding successive tranches. In the example you've given, I guess it would still load up the two most profitable time slices first, but then when the third asset is added it would redistribute among them to cover the full load. So that question becomes how to calculate the investment metrics, and how do you compare with other competing processes? I guess both current metric are per unit demand, so should be OK? If the "full" asset still wins you commit it, but if something else wins then you only commit the tranches up to the prior one. It later rounds you revisit the unselected tranches in terms of a "what if this is added onto already selected tranches"?

As usual, I imagine there are several holes in this logic. Thoughts appreciated.