Change the way LCOX appraisal tool works

[ahawkes]

Is your feature request related to a problem? Please describe

At the moment the LCOX tool uses VoLL to serve any demand a candidate/existing asset cannot serve. This has the effect of pushing the installed capacity to it's maximum value to serve as much demand as possible (and thus avoid the VoLL). Furthermore, the capacity upper bound the tool uses is not updated between "rounds" (i.e. tranches of demand), so it's possible that very large capacities get chosen on later rounds even though they are not needed.

Describe the solution you'd like

We need a new approach for the LCOX appraisal tool. It should not use the VoLL approach, and it should not overinvest. I think the following could almost work and would appreciate thoughts:

1. Determine capacity of candidate asset such that its maximum output equals the tranche maximum demand.
2. For the tranche demand, determine candidate/asset dispatch in previous MSY commodity price conditions. Assets dispatch when their activity cost (AC - defined in docs already) is less than the commodity shadow price (provided shadow prices where also used in the AC calculation). So we can alter the current asset mini optimisation so that demand must be served, using the asset at cost AC(t), or using a fallback option which costs the shadow price when it is lower than AC or the asset is availability-constrained. This will result in dispatch that respects availability constraints, and only operates when the asset could beat the previous MSY shadow price (or equal it - need to add the epsilon to shadow price - similar to current NPV tool).
3. Calculate LCOX using the resulting asset dispatch profile (and commodity prices as per pricing strategy)
4. Repeat for all assets and choose the one with lowest LCOX.
5. Figure out the demand profile of the next tranche and go back to (1).

This could fail when process parameters change between MSY, in that maybe nothing gets dispatched. In that case one could increase the shadow price (e.g. by 10%) and try again. This is messy though!

Also, for existing assets, what if its capacity is wildly different to tranche size? This can happen in the current version too. We have a lot of moving parts here - tranche sizes, asset capacities, divisible assets - a bit mind bending to keep track of them all within one thought...

Maybe we could forget about the tranches, and just rely on user-chosen asset capacities. So the user has to choose the asset size considered, where all the demand is the target, with the shadow-priced fallback option? That would eliminate the tranche size choice issue, and perhaps we could also re-consider the divisible assets thing. At the expense of creating a headache for the user, who has to enter a lot of data!

Thoughts? @tsmbland @alexdewar

[ahawkes]

Was mulling this, and it occured to me that (I think) the dispatch determination step here is the same as for the NPV tool? Essentially we're assuming dispatch if the system would have wanted the asset in the previous MSY, in both NPV and LCOX tools. If this is correct it may have implications for the structure of the code, to avoid repetition.

[alexdewar]

I'm tagging this as a question, along with #1201, because whoever takes this issue will need to have a think about it before tackling it properly.

[ahawkes]

Just to reiterate, I think we should always calculate dispatch assuming profit-maximisation in previous MSY conditions. This - I think - is the approach we're currently using in the NPV appraisal tool. So we might be able to just segregate that bit of code and always use it to figure out assumed dispatch, regardless of which appraisal tool is being applied?