Closed joseph-robertson closed 1 year ago
I tried running datapoints 1 through 5 for the national project:
qoi_report.qoi_peak_magnitude_use_kw | report_simulation_output.peak_electricity_summer_total_w | report_simulation_output.peak_electricity_winter_total_w |
---|---|---|
5 | 4560.3 | 5002.4 |
11.86 | 6528.1 | 11862.3 |
2.62 | 2577.1 | 2620.1 |
2.73 | 1996.3 | 2126 |
6.56 | 6556.1 | 5315 |
Results look pretty good for datapoints 1, 2, 3, and 5.
Datapoint 4 appears to have the opposite behavior described in this issue. Winter is limited to Dec/Jan/Feb and Summer is limited to Jun/Jul/Aug, so this seems reasonable/possible to me.
@lixiliu Are you able to provide more detail here so I can reproduce the issue you are seeing?
@joseph-robertson Yes, I am seeing the discrepancy in EUSS Run Round 1 found on AWS: s3://euss/final-2018/euss_res_final_2018_550k_20220901/baseline/
Attached is a downselected version containing only building_id, total electricity, qoi_peak, and the seasonal peaks.
I am seeing qoi_peak < seasonal peak 529469 times (out of 548916). I also noticed that the some of the seasonal peaks = 0 when total electricity is non-zero (e.g., building_id 6, 19) which makes me think that there's a bug with seasonal peaks.
euss_amy2018_results_up00_peaks.csv
Let me know if I should extract run files for some of these problematic buildings for debugging
The algorithm for summer/winter peak electricity changed several months ago (see here). Do we know if @lixiliu's runs are using the newer algorithm? With the old algorithm, I could see how you could get this result, but not with the newer algorithm.
Sorry, I wasn't very clear there. I could see how the older algorithm could give the result of "seasonal peaks = 0 when total electricity is non-zero", not how it could give the result of qoi_peak_magnitude_use_kw < max(peak_electricity_summer_total_w, peak_electricity_winter_total_w).
@lixiliu I can confirm that qoi vs seasonal peaks are consistent when the simulation timestep is hourly. However when the simulation timestep is subhourly (e.g., 15-minutely), I see the discrepancy similar to that reflected in your csv file posted above. Basically it comes down to reporting peaks across hourly timesteps in QOIReport even if the simulation timestep is subhourly (whereas OS-HPXML reports peaks across subhourly timesteps). The following items help to illustrate what I'm talking about:
Suggest changing name of QOI report value to something like "peak_hour..."
"qoi_report.qoi_peak_magnitude_use_kw" appears to be smaller than the max of: