Overview
Geometric LOD study · drop 5 · 264 cells · atlas inputs release 2.4.0 · timed (R5)
Which thermal zoning of a DOE prototype building is coarse enough to be cheap and fine enough to keep its ideal sensible heating and cooling loads within a stated error of the prototype's full zoning? For each prototype, climate zone, vintage and orientation the study simulated the prototype's own geometry as BEMGen writes it, zoned every room (the reference), and then the same building under each zoning simplification and floor stacker (the test sites) with geometry, atlas programs, envelope and weather fixed. Every test site is scored against its reference with the batch runner's paired metrics: the annual sensible heating and cooling error, the hourly CV(RMSE) and NMBE, the peak errors and the change of conditioned floor area. The recommended zoning of a cell is the coarsest one, by modelled zone count, that meets the threshold below; a cell where no zoning meets it says so. The robust zoning of a prototype, climate zone and vintage is the coarsest that meets the threshold at both orientations; each prototype page opens with it.
Threshold
- Basis
- ASHRAE Guideline 14 hourly limits (CV(RMSE) 30 %, NMBE 10 %), annual error 10 %, conditioned area 1 %; proposed, the owner decides
- Loads judged
- Sensible heating, Sensible cooling
- Annual error
- |error| ≤ 10.0 %
- Hourly CV(RMSE)
- ≤ 30.0 %
- Hourly NMBE
- |NMBE| ≤ 10.0 %
- Conditioned floor area
- |change| ≤ 1.0 %
- Loads not judged
- a load whose reference is below 1.0 kWh/m² a year
- Owner choice
- yes: the default of the report writer; a later drop may carry another
Prototypes
| Prototype | Cells | With a recommended zoning | Robust across orientations (cell pairs) | Most often recommended (cells) | Recommended zones | Reference zones |
|---|---|---|---|---|---|---|
| Medium office | 24 | 24 | 12 of 12 | Perimeter core 4.57 m + Stack (16) | 1 to 60 | 65 |
| Full-service restaurant | 24 | 8 | 4 of 12 | Conditioned merge + Single zone building (8) | 1 to 1 | 2 |
| Strip mall | 24 | 24 | 12 of 12 | Conditioned merge + Single zone building (14) | 1 to 3 | 10 |
| Stand-alone retail | 24 | 22 | 11 of 12 | Conditioned merge + Single zone building (15) | 1 to 5 | 5 |
| Supermarket | 24 | 17 | 8 of 12 | Conditioned merge + Single zone building (12) | 1 to 5 | 6 |
| Warehouse | 24 | 2 | 1 of 12 | Conditioned merge + Single zone building (2) | 1 to 1 | 3 |
| Primary school | 24 | 20 | 10 of 12 | Conditioned merge + Single zone building (12) | 1 to 14 | 25 |
| Small hotel | 24 | 24 | 12 of 12 | Conditioned merge + Stacked floor zone multiplier (13) | 8 to 43 | 67 |
| Large hotel | 24 | 24 | 12 of 12 | Semantic merge + Stacked floor zone multiplier (13) | 6 to 33 | 42 |
| Hospital | 24 | 24 | 12 of 12 | Conditioned merge + Stacked floor zone multiplier (20) | 24 to 51 | 54 |
| Outpatient | 24 | 24 | 12 of 12 | Conditioned merge + Single zone building (12) | 1 to 104 | 126 |
A zoning meets the threshold in 213 of 264 cells, and at both orientations of a cell pair in 106 of 132 pairs (robust across orientations). Each prototype page lists every cell, the recommended zoning with its errors, every zoning's metrics and the atlas programs the prototype takes. The heat-transfer sub-study varies the interior-wall and floor heat-transfer switches. Method and caveats.
Recommended zonings
| Zoning | Cells |
|---|---|
| Conditioned merge + Single zone building | 82 |
| Semantic merge + Stack | 50 |
| Conditioned merge + Stacked floor zone multiplier | 40 |
| No simplification + Single zone building | 35 |
| Perimeter core 3 m + Single zone building | 35 |
| Perimeter core 4.57 m + Single zone building | 35 |
| Perimeter core 4.57 m + Stack | 28 |
| Conditioned merge + Stack | 27 |
| Perimeter core 3 m + Stack | 24 |
| Semantic merge + Single zone building | 20 |
| Perimeter core 4.57 m + Stacked floor zone multiplier | 16 |
| Perimeter core 3 m + Stacked floor zone multiplier | 14 |
| Semantic merge + Stacked floor zone multiplier | 13 |
| Single zone per floor + Stack | 1 |
Counts over every cell and tie of the main matrix.
Robust across orientations: 106 of 132 cell pairs
| Robust zoning | Cell pairs |
|---|---|
| Conditioned merge + Single zone building | 40 |
| Semantic merge + Stack | 28 |
| Conditioned merge + Stack | 12 |
| Perimeter core 3 m + Stack | 12 |
| Semantic merge + Stacked floor zone multiplier | 7 |
| Conditioned merge + Stacked floor zone multiplier | 5 |
| Perimeter core 4.57 m + Stack | 2 |
Counts over the cell pairs (prototype × climate zone × vintage) where one zoning meets the threshold at both orientations; the robust zoning of a pair is the coarsest that does. The rule and its limits.
Cost
- Timed
- 244 sites of 12 cells (12 test cases of the serial pass), 9 prototypes; not timed: Large hotel, Medium office
- Protocol
- serial timing pass: --parallel 1 --timing-repeats 3 --timing-test-cases 12 (the first 12 test cases by the SHA-256 of their IDs)
- Statistic
- a site's cost is the median of its repeats; its ratio is that median over its reference's; a prototype's cost is the median of its timed test sites' ratios without the duplicate twins (144 dropped), and a zoning's and the overall cost are the medians of the prototype medians, each timed prototype one vote (D-222)
- Overall
- processor time 0.479, wall clock 0.495 and peak committed memory 0.981 of the reference's: the median of the 9 prototype medians (88 distinct test sites)
- Gate
- processor-time CV per site below 5.0 %: 242 of 244 sites pass (median 1.7 %, largest 7.0 %); not met
- Repeat check
- not run for this drop: a check between sessions needs a second session on an otherwise idle machine (D-265)
Costs are the serial timing pass (R5): one EnergyPlus at a time, three repeats, on an otherwise idle machine. They are reported as medians, since the gate was not met (2 of 244 sites miss it) and no check between sessions was run for this drop; the medians by prototype and by zoning below are the figures to quote, and the per-cell ratios on the prototype pages show the spread. Cost protocol.
By prototype
| Prototype | Test sites | CPU ratio | Wall ratio | Memory ratio | CPU median s | CPU CV median % |
|---|---|---|---|---|---|---|
| Medium office | not timed | |||||
| Full-service restaurant | 3 | 1.580 | 1.608 | 1.010 | 2.0 | 1.3 |
| Strip mall | 4 | 0.479 | 0.495 | 0.981 | 1.6 | 1.1 |
| Stand-alone retail | 6 | 0.494 | 0.507 | 0.986 | 1.3 | 1.9 |
| Supermarket | 12 | 0.468 | 0.473 | 0.980 | 1.4 | 2.1 |
| Warehouse | 6 | 0.495 | 0.522 | 0.990 | 1.6 | 2.5 |
| Primary school | 7 | 0.230 | 0.229 | 0.921 | 2.0 | 2.3 |
| Small hotel | 32 | 0.141 | 0.142 | 0.825 | 3.7 | 1.1 |
| Large hotel | not timed | |||||
| Hospital | 10 | 0.875 | 0.873 | 0.997 | 18.3 | 0.6 |
| Outpatient | 8 | 0.064 | 0.066 | 0.752 | 3.2 | 0.8 |
By zoning
| Zoning | Test sites | Prototypes | CPU ratio | Wall ratio | Memory ratio | CPU median s | CPU CV median % |
|---|---|---|---|---|---|---|---|
| Semantic merge + Single zone per floor type | 1 | 1 | 0.064 | 0.064 | 0.753 | 3.1 | 0.3 |
| Semantic merge + Single zone building | 1 | 1 | 0.064 | 0.067 | 0.751 | 3.2 | 1.7 |
| Conditioned merge + Single zone per floor type | 3 | 2 | 0.072 | 0.074 | 0.752 | 2.6 | 1.8 |
| Single zone per floor + Stacked floor zone multiplier | 2 | 1 | 0.079 | 0.083 | 0.809 | 2.1 | 2.2 |
| Perimeter core 4.57 m + Single zone per floor type | 2 | 1 | 0.081 | 0.084 | 0.811 | 2.1 | 1.7 |
| Perimeter core 3 m + Single zone per floor type | 2 | 1 | 0.081 | 0.081 | 0.810 | 2.1 | 1.2 |
| Conditioned merge + Stacked floor zone multiplier | 5 | 3 | 0.141 | 0.141 | 0.826 | 3.7 | 2.0 |
| Perimeter core 4.57 m + Stacked floor zone multiplier | 2 | 1 | 0.235 | 0.240 | 0.846 | 6.2 | 1.9 |
| Perimeter core 3 m + Stacked floor zone multiplier | 2 | 1 | 0.245 | 0.247 | 0.851 | 6.5 | 0.8 |
| Conditioned merge + Single zone building | 10 | 8 | 0.266 | 0.276 | 0.973 | 0.9 | 2.1 |
| Perimeter core 4.57 m + Single zone building | 7 | 5 | 0.332 | 0.338 | 0.975 | 1.1 | 2.4 |
| Perimeter core 3 m + Single zone building | 7 | 5 | 0.337 | 0.335 | 0.974 | 1.1 | 1.8 |
| Conditioned merge + Stack | 4 | 2 | 0.339 | 0.341 | 0.939 | 7.5 | 0.6 |
| No simplification + Single zone building | 5 | 4 | 0.593 | 0.602 | 0.985 | 1.9 | 1.4 |
| Perimeter core 3 m + Stack | 9 | 7 | 0.629 | 0.631 | 0.989 | 2.0 | 2.3 |
| Perimeter core 4.57 m + Stack | 9 | 7 | 0.632 | 0.627 | 0.992 | 2.0 | 1.7 |
| Semantic merge + Stack | 7 | 5 | 0.673 | 0.676 | 0.980 | 16.2 | 0.6 |
| Semantic merge + Stacked floor zone multiplier | 5 | 3 | 0.785 | 0.799 | 1.043 | 18.3 | 0.8 |
| No simplification + Stacked floor zone multiplier | 5 | 3 | 0.906 | 0.903 | 0.968 | 26.0 | 0.5 |
The report writer's medians: by prototype over its timed test sites, by zoning over the prototype medians (the prototypes column counts them): the ratios of processor time, wall clock and peak committed memory to the reference, the median processor time in seconds and the median per-site CV. Peak committed memory hardly changes because EnergyPlus's own floor dominates; the saving in time follows the zone count only for large models, since a fixed start-up dominates the small single-storey prototypes. A ratio above 1 (Full-service restaurant) means the prototype's distinct timed zonings cost more than its reference at the median: some of them have more zones than the reference (the prototype page lists each zoning's zone count).
Status
Data drop 5 of the geometric LOD study: drop 4's results on atlas release 2.4.0 inputs (every stage at BEMGen 35e8a132), now timed by R5, the serial timing pass run on an otherwise idle machine (BEMGen stage LOD2-R5, D-265), and without the processor-time ratio of the parallel stage runs (D-266). Schema 1.2.0 as before. The recommendations are drop 4's unchanged (R3: 213 of 264 cells; R4: 18 of 22); the files differ from drop 4's only by the cost and timing members R5 adds and the cpu_time_ratio member D-266 removes. R5 timed 244 sites of 12 test cases (12 references and 232 test sites, 732 runs) on 2026-11-02 from about 04:20 to 05:11 local time without interruption, the machine's total processor load 5.6 % on average and at most 9.5 %. The gate (processor-time CV below 5 % over the three repeats) is not met by 2 of 244 sites, both 0.9 s supermarket 3C runs, so the costs are medians (D-202); the late-repeat slowdown of the earlier timed drop did not recur. No repeat check between sessions was run for this drop. The medium office and the large hotel are not among the timed test cases.
Nothing is pending.
- Study report (R6):
research/lod-study/report/2026-11-02-lod-study-report.mdin BEMGen - Methodology:
research/lod-study/report/2026-11-02-lod-study-methodology.mdin BEMGen - Run plan:
docs/plans/2026-10-29-lod-study-run-plan.mdin BEMGen - Results:
research/lod-study/results/in BEMGen - Previous drop (drop 4, not timed):
research/lod-study/results/in BEMGen - Archived earlier drop (drop 3, atlas release 2.0.0 inputs):
research/lod-study/archive/2026-10-29-atlas-2-0-0/in BEMGen
| Stage | Suite | Run | BEMGen | Gate |
|---|---|---|---|---|
| R3 main matrix | lod-main-20261101-225850 | run_20261101-230342 | 1.10.0+35e8a132f785a5329f4fd6015fce82a655e634ab | complete: 5280 of 5280 simulated sites succeeded; 1056 refused by the generator by documented limits (listed with their reason) |
| R4 heat-transfer sub-study (IWHT/FHT) | lod-ht-20261102-001857 | run_20261102-002039 | 1.10.0+35e8a132f785a5329f4fd6015fce82a655e634ab | complete: 1760 of 1760 simulated sites succeeded; 352 refused by the generator by documented limits (listed with their reason) |
| R5 serial timing pass | lod-timing-20261102-034653 | run_20261102-034718 | 1.10.0+35e8a132f785a5329f4fd6015fce82a655e634ab | CPU-time CV < 5 %: 242 of 244 sites, not met; reported as medians (D-202) |
| R6 study report, methodology and consolidated results | – | – | 1.10.0+99c0b0adcd4449bedbe0f7a1b86b0db75e2d3ce5 | written by scripts/lod-study/study_tables.py --write from report.py of D-266 and compared with the archived drop; tables verified by --check (D-264, D-265, D-266) |
Original data CC BY 4.0 (ADR 0048, ADR 0050); attribution: Archetype Atlas, Energy-Atlas, LOD recommendations drop 5.