Add class set to model outputs by wesleyjcole · Pull Request #217 · ReEDS-Model/ReEDS

wesleyjcole · 2026-09-09T16:37:50Z

Summary

(Note: this is getting merged into a dev branch and not into main)

This is the 5th PR toward #185, and adds the resource class index c to the model outputs and downstream consumers. The class index is still redundant, so each technology maps to only one class.

Two other meaningful changes:

INV_REFURB gained a class index because it was necessary for several output parameters.
csp-ns is no longer hard-coded to upv_5, but is converted dynamically. This revealed a bug where some of the the csp-ns was getting dropped in the input processing (this PR fixes that bug).

Technical details

Implementation notes

INV_REFURB.l was added to the Sw_RemoveSmallNumbers rounding block in 5_varfix.gms. It was previously the only investment variable absent from that block.

The previous approach for dealing with csp-ns was both fragile and inaccurate: it subtracted all csp-ns fromupv_5, so it would silently do nothing if upv_5 did not exist in a run. writecapdat.py now assigns each csp-ns unit to a upv resource class from its capacity factor, using the same get_class_cf_bounds/assign_class helpers used for upv itself, and writes cap_cspns by (r,c,t). Because the csp-ns resource is in the best-class areas, the reassignment draws from the best class first and moves into the remaining classes only as needed. report.gms then takes csp-ns capacity and generation back out of those same classes,
in proportion to the upv capacity present in each. The csp-ns assignment is also added to the error_check parameter.

Additional changes

The excess error check calculation is added to compare_cases.py so that it will display properly if it has a non-zero value.

Issues resolved

Partially addresses #185.

Known incompatibilities

The PR breaks R2X, so fixes for that will be needed before this dev branch is merged into main. I've added that information to #185 to track it.

Validation, testing, and comparison report(s)

I verified that before making the csp-ns fix, the outputs were all the same to within rounding noise. A national USA_defaults case returned bascially identical outputs:

The csp-ns changes model outputs because some of the csp-ns generation was missing from the model. The change is pretty small because the missing csp-ns generation was small:

Compare report after csp-ns fix: results-Main,CSP-nsFix.pptx.

Checklist for author

Details to double-check

Charge code provided to reviewers
Included comparison reports for appropriate test cases
~~[ ] Documentation updated if necessary~~
Code formatting standardized
Reusable functions used where possible instead of copy/pasted code

General information to guide review

Zero impact on results of default case
No large data file(s) added/modified
No substantive impact on runtime for full-US reference case
No substantive impact on folder size for full-US reference case
No change to process flow (runreeds.py, reeds/core/solve/solve.py)
No change to code organization
No change to package requirements (environment.yml or Project.toml)

Did you use LLM tools (chatbot or copilot) in the preparation of this PR? If so, describe how

Yes. I used Claude Opus 5 to implement the changes in this PR and to do many of the output comparisons. I also used it to diagnose and fix the csp-ns bug found during this PR.

Tag points of contact here if you would like additional review of the relevant parts of the model

…to use undegradated capacity.

cavraam

No major concerns in this one, especially given results. A few suggestions to either collapse the 'c' index in reporting or potentially rename some of the output parameters to reflect the specific outputs. No strong feelings either way!

A couple of sanity checks included, prompts to double-check a couple of seemingly trickier code, but the respective parts should be fine.

PR looks good to go!

cavraam · 2026-09-09T19:08:31Z

    # Concatenate existing and inv_rsc
    df_inv = pd.concat([df_bin_exist, df_inv_rsc], sort=False, ignore_index=True)
-    # Split tech from class
-    df_inv[["tech_cat", "class"]] = df_inv["tech"].str.rsplit("_", n=1, expand=True)


got it, because 'class' is already part of the data upstream, right?

That's right.

cavraam · 2026-09-09T19:21:34Z

(Not related to this PR) I am also getting a suggested edit on line 1610: 'rcby' seems to be defined in line 1647... I wonder if it needs to be renamed to 'rcy' defined earlier, or move some code around?

reeds_to_rev.py isn't working right now, but this is a quick fix, so I can add it in here.

cavraam · 2026-09-09T19:25:20Z

Looks great programmatically. Conceptually, it might make sense to include the class in cap_energy_ivrt and/or cap_firm and cap_firm_iter?

I don't think it's necessary in cap_energy_ivrt because there are no class-relevant technologies for energy in the initial scope. It would make sense if PSH gets a class at some point.

cap_firm, though, should get a class designation. We'll want to know exactly the capacity contribution of a technology. That will require making m_cc_mar adopt a class, which also needs to happen at some point. I've been focusing on stress periods, but I'll note this for follow-on work once I have a working stress-period formulation with collapsed classes. The output rename would also go along with that.

cavraam · 2026-09-09T19:26:42Z

        'RPS': {'title': 'Non-RecMap\nRECS\n[GWh]', 'scale':1e-3},
        'OpRes': {'title': 'Non-valgen\nopres\n[MWh]', 'scale':1},
        'm_rsc_dat': {'title': 'Supply curve\ntweaks [GW]', 'scale':1e-3},
+        'cspns': {'title': 'Unreassigned\ncsp-ns [MW]', 'scale':1},


could this tag be 'unassigned'?

It is capacity that was not re-assigned to csp-ns, but that's likely not to be intuitive. I'll update to unassigned.

cavraam · 2026-09-09T19:31:39Z

                if items_column == 'tech':
                    df['tech'] = df['tech'].str.lower().map(lambda x: items_map.get(x, x))
-                    df = df.groupby(['tech', 'r', 'year']).sum().reset_index()
+                    valcols = df.select_dtypes('number').columns.difference(['year'])


a little less familiar with this part of ReEDS, so commenting just as a point to double-check... post-processing passed the checks, so we should be good

I took another look and decided that summing over the resource would also help here. Reviewing these and reproducing outputs for ancillary scripts like this will be a key part of testing after the i set is collapsed.

cavraam · 2026-09-09T19:51:15Z

 cap_exist_i(i)$valcap_i_filt(i) = sum{(r,v), cap_exist(i,v,r) } ;

-cap_ivrt(i,v,r,t)$([not (upv(i) or wind(i))]$valcap(i,v,r,t)$trange(t)) = CAP.l(i,v,r,t) ;
+cap_ivrt(i,c,v,r,t)$[i_c(i,c)$(not (upv(i) or wind(i)))$valcap(i,v,r,t)$trange(t)] = CAP.l(i,v,r,t) ;


nitpick: do we actually need to track the class? I wonder if it would make sense to just sum over 'c'... otherwise, do we want to potentially rename this to cap_icvrt? I understand that it might be a little bit of a pain to update downstream in reeds.py etc...

I think we want to track the class. That way this parameter holds the equivalent information as it does now. Otherwise you wouldn't have the information about which wind/pv/etc. class is getting built.

And I agree with the rename and that it will be a bit of a pain. I'm noting that for a follow-up change as well.

cavraam · 2026-09-09T20:02:33Z

+cap_init(i,v,r)$(distpv(i)$valcap_ivr(i,v,r)) = sum{(c,t)$[i_c(i,c)$tfirst(t)], cap_ivrt(i,c,v,r,t) } ;
+inv_ivrt(i,c,v,r,t)$[i_c(i,c)$valcap(i,v,r,t)$trange(t)] =
+    [INV.l(i,v,r,t)$(not rsc_i(i))
+     + sum{rscbin$m_rscfeas(r,i,c,rscbin), INV_RSC.l(i,c,v,r,rscbin,t) }$rsc_i(i)


this line prompted me to double-check if $i_c(i,c) is accounted for in m_rscfeas in reeds/core/setup/b_inputs.gms. I think the edits in b_inputs.gms below are optional, given that rscfeas is based on rsc_dat (which only has the relevant i-c pairs), but flagging just in case:

Lines 1443 -- 1453 (related to 'rscfeas'): Do we need sum{ii$i_c(ii,c)$...], ...}?

Line 1751, 1753: Do we need sum{ii$[i_c(ii,c)$...], ...}?

Lines 5258 --5260: I see a few sum{ii, ...}, would we need sum{ii**$i_c(ii,c)**, ...}?

I checked these, and I think they are ok. There may be some redundancy (e.g., some $ conditionals can be removed without impacting the calculation), but I'd rather not clean those up until the classes are collapsed becaused then it will be easier to see if removing a conditional actually does change things.

cavraam · 2026-09-09T20:03:48Z


-ret_ivrt(i,v,r,t)$([trange(t)$(not tfirst(t))$newv(v)]$valcap(i,v,r,t)) = sum{tt$tprev(t,tt), cap_ivrt(i,v,r,tt)} - cap_ivrt(i,v,r,t) + inv_ivrt(i,v,r,t) ;
-ret_ivrt(i,v,r,t)$([abs(ret_ivrt(i,v,r,t) < 1e-6)]$valcap(i,v,r,t)) = 0 ;
+ret_ivrt(i,c,v,r,t)$[i_c(i,c)$trange(t)$(not tfirst(t))$newv(v)$valcap(i,v,r,t)] = sum{tt$tprev(t,tt), cap_ivrt(i,c,v,r,tt)} - cap_ivrt(i,c,v,r,t) + inv_ivrt(i,c,v,r,t) ;


similar nitpick to 'cap_ivrt' here: Do we need to explicitly track 'c' in the outputs? If so, would it make sense to rename into ret_icvrt here or later on?

Agreed - and saving for a follow-up PR.

cavraam · 2026-09-09T20:20:45Z


 gen_new_uncurt(i,r,h,t)$[(vre(i) or storage_hybrid(i)$(not csp(i)))$valcap_irt(i,r,t)] =
-      sum{(v,c)$[valinv(i,v,r,t)$i_c(i,c)], (INV.l(i,v,r,t) + INV_REFURB.l(i,v,r,t)) * m_cf(i,c,v,r,h,t) * hours(h) }
+      sum{(v,c)$[i_c(i,c)$valinv(i,v,r,t)], (INV.l(i,v,r,t) + INV_REFURB.l(i,c,v,r,t)) * m_cf(i,c,v,r,h,t) * hours(h) }


are we potentially double-counting the INV values by having the sum applied to both INV and INV_REFURB instead of the latter only (like above)?

No - these are unique, so we will need both.

cavraam · 2026-09-09T20:23:51Z

-        INV.l(i,v,r,tt) + INV_REFURB.l(i,v,r,tt)$[refurbtech(i)$Sw_Refurb]}) / ilr(i) ;
-
-cap_out(i,r,t)$[valcap_irt(i,r,t)$tmodel_new(t)] = sum{v$valcap(i,v,r,t), cap_ivrt(i,v,r,t) } ;
+cap_out(i,c,r,t)$[i_c(i,c)$valcap_irt(i,r,t)$tmodel_new(t)] = sum{v$valcap(i,v,r,t), cap_ivrt(i,c,v,r,t) } ;


could sum over 'c' as well and not worry about it in reeds2.py?

I tried to keep the same information as was in the original data where feasible. Since class information was already available and can be calculation directly, I maintained class in the output.

wesleyjcole added 24 commits September 3, 2026 11:51

Replace instances of CAP with CAP_CLASS in outputs.

640e164

Add class to INV_REFURB and m_cf_filt.

92e9cc4

Propagate "c" through 6_data_dump.gms

5c1a5e5

Add class to resources input.

a7c08a5

Add class to relevant resource_adequacy parameters.

250e101

Add class to reeds_to_rev.py inputs.

b97a64c

Add class to more outputs

8f3f31e

Add class to cap_ivrt.

65735b2

Add class to Bokeh.

d19892f

Add class to cap_out.

a31ff0d

Remove class splitting from reeds_to_rev.py

701deec

Add class to cap_new_ivrt_refurb.

d0cdb47

Add class to LCOE outputs.

89f7e25

(forgot to include this file with the last commit...).

485daef

Add class to additional capacity by ivrt outputs

2f948ab

Add class to cap_avail and capex_ivrt

1bfc477

Fix issue in Bokeh

005e479

Adjust csp-ns allocation

098e2a4

Merge branch 'wjc/dev/separate_class' into wjc/outputs_class

f5b2076

Update bokeh class assignments.

e5eff11

Add "excess" and and unallocated csp-ns to compare_cases.py

ea86ae1

Move csp-ns reassignment parameters to report_params.csv

550fbb0

Fix csp-ns accounting error.

56e65c2

Update comment for cap_cspns

a17e8bb

wesleyjcole requested a review from cavraam September 9, 2026 16:37

github-actions Bot added data_changes hourlize model_changes postprocessing labels Sep 9, 2026

Add class to merges in bokeh.

3036306

wesleyjcole added 3 commits September 9, 2026 11:28

Select numeric value columns in uncertainty_plots.py

caa4fff

Reorder indices of cap_cspns to be consistent. Also adjust reporting …

9860c76

…to use undegradated capacity.

Add groupby rather than drop class column

d1d3c79

cavraam approved these changes Sep 9, 2026

View reviewed changes

wesleyjcole added 3 commits September 11, 2026 09:12

Change description to unassigned.

7ff1628

Fixes for reeds_to_rev.py

e92fd37

Add additional groupby for uncertainty_plots.py

8a60ebf

wesleyjcole merged commit e8c2c15 into wjc/dev/separate_class Sep 11, 2026
1 check passed

wesleyjcole mentioned this pull request Sep 11, 2026

Separate resource class from technology set i #185

Open

28 tasks

Uh oh!

Conversation

wesleyjcole commented Sep 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Technical details

Implementation notes

Additional changes

Issues resolved

Known incompatibilities

Validation, testing, and comparison report(s)

Checklist for author

Details to double-check

General information to guide review

Did you use LLM tools (chatbot or copilot) in the preparation of this PR? If so, describe how

Tag points of contact here if you would like additional review of the relevant parts of the model

Uh oh!

cavraam left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

wesleyjcole commented Sep 9, 2026 •

edited

Loading