Tools › Building Systems › Energy › Energy Baseline Regression › Example

BPA regression guide: heating degree-day baseline of a classroom building, its net bias, and the avoided energy of a reporting year

ToolEnergy Baseline Regression
SourceBonneville Power Administration (2024). Regression for M&V: Reference Guide. Bonneville Power Administration, Portland, OR, May 2024. Section 6.1, Tables 6-1 to 6-8; Bonneville Power Administration (2024). Verification by Meter-Based Energy Modeling Protocol, Version 3.1. Section 5.5.
LevelProfessional

Problem

Twelve electricity bills of a classroom building heated by heat pumps are given with the days and the heating degree-days of each billing period (24 September 2007 to 22 August 2008). Fit the baseline model of kWh per day against heating degree-days per day as the guide does, with every period weighted equally; report R², CV(RMSE) and the net determination bias of the modeled against the metered total. Then weight each period by its days, as the guide says would have been the best approach, and compare. Finally adjust the baseline to the weather of a reporting year and find the avoided energy use with its uncertainty. The reporting year and the cooling degree-day column of the opening rows are invented for the illustration: the guide gives neither.

Inputs to Enter

  • Weather data: degree-days; model: heating degree-days, E = b1 + b2 HDD; normalization: energy per day, every period weighted equally (the second option; the page opens on the day-weighted fit)
  • Baseline rows (the opening rows; days, kWh, HDD): 30, 6,080, 113; 30, 7,330, 311; 28, 7,470, 463; 28, 10,000, 669; 35, 11,480, 877; 33, 11,420, 782; 30, 9,970, 560; 29, 7,840, 561; 26, 6,800, 265; 32, 5,980, 268; 32, 4,310, 73; 30, 3,330, 57
  • Reporting rows (invented, the opening rows): twelve periods, 365 days, 49,870 kWh metered
  • Confidence 90 %; exact least-squares uncertainty formula with the autocorrelation correction; energy rate 0.11 $/kWh
Open the Energy Baseline Regression tool

Answers

  • Equal weights: kWh/day = 132.17 + 8.8720 HDD/day (book 132.24 + 8.8678); R² 0.8799 (book 0.879); CV(RMSE) 11.69 % (book 11.7 %); standard error 29.64 kWh/day (book 29.69)
  • First period: modeled 165.6 kWh/day with a 90 % confidence half-interval of 24.2 (book 165.5 and 24.3); modeled total 92,330 kWh against 92,010 metered (book 92,331)
  • Net determination bias −0.348 %: the model predicts 0.3 % more than was metered, as the book finds, against the 0.005 % criterion it quotes
  • Day-weighted fit (the page's default): kWh/day = 131.41 + 8.8634 HDD/day; modeled total 92,010 kWh, net bias zero; R² 0.8849, CV(RMSE) 11.58 %
  • Reporting year, day-weighted model: adjusted baseline 90,465 kWh, metered 49,870 kWh, avoided energy use 40,595 kWh (44.9 %)
  • Uncertainty of the avoided energy at 90 % confidence ± 9,334 kWh (23.0 %); fractional savings uncertainty at 68 % confidence 13.3 %, inside the 50 % the BPA protocol requires; cost saving $4,465.42

Notes

Press Submit for the day-weighted results; set Normalization to "Energy per day, every period weighted equally" for the guide's own fit. The guide's numbers are not reproduced to the last digit because it prints whole degree-days and ran its regression on digits it does not print (its Table 6-2 shows 3.7 and 19.4 HDD per day where its own columns give 3.77 and 19.34). Asserted in test_energy_baseline_regression.php, which also checks the change-point search against brute force on 140 seeded cases. The savings figures belong to the invented reporting year and have no printed counterpart.