SEEM Calibration for Manufactured Homes Regional Technical Forum July 15, 2014
Overview Same basic structure as single-family calibration. Phase I: Estimating total heating energy. – Align SEEM with billing data for homes with strong and clear heating energy signatures and no off-grid fuels. Phase II: Estimating electric heating energy in “typical” program homes. – How is electric heating energy affected by the presence of natural gas and off-grid fuels? – What can we say about electric heating energy in homes with weak or unclear heating energy signatures? 2
Phase I: Total heating energy in “well-behaved” homes 3
Phase I general approach 4 Phase I calibration needed because we don’t have perfect knowledge of SEEM inputs. Built around “SEEM (69/64).” This has… – Some inputs based on RBSA data (location, wall insulation, heating equipment…) – Others based on convention (thermostat settings, internal gains…) – “69/64” refers to inside air temperatures (64°F day and 64°F night), but T-stat isn’t the only standardized input. Uses regression to understand differences between SEEM (69/64°F) and billing data (VBDD) heating energy estimates. Regression results provide adjustment factors needed to align SEEM (69/64) with VBDD.
Sneak preview! 5 Adjustment factors look a lot like the SF calibration factors, but… They’re smooth (no abrupt change of slope); The Uo variable has been replaced with a more inclusive heating intensity variable.
More sneak preview! 6 Measure: Attic insulation from R8 to R19. Examples are randomly chosen RBSA sites, not measure prototypes. Home A: Zone 1, Heat pump SEEM.69 kWh Square feet IntensityPhase I factor Phase I kWh Attic R84, ,786 Attic R193, , Home B: Zone 2, Elec. FAF Attic R821, ,928 Attic R1916, ,862 4,6872,065
Phase I groundwork Phase I analysis is restricted to homes whose RBSA entries… – Include building characteristics needed to build SEEM inputs, and – Suggest VBDD reasonably estimates total in-door heating energy 7
Phase I data filters, part 1 8 Some filters not really related to indoor heating energy, don’t really threaten our total indoor heating energy estimates. Some have a lot to do with indoor heating energy. Need to accounts for these. Many “other reason” SEEM failures due to unusually high SLF entries. Staff proposes to included these in baseline SLF averages (but a separate calibration adjustment would be redundant). Filter definitionReason for exclusionBias risk Has large outdoor heating loadBilling data can’t isolate indoor heating energyMedium Has DHPOut of scope (will calibrate MH DHP separately)NA Missing billing dataCan’t generate VBDD estimate for analysisLow Missing SEEM input dataCan’t generate SEEM (69/64) for analysisLow Failed SEEM run, other reasonCan’t generate SEEM (69/64) for analysisHigh Has non-utility heatBilling data can’t isolate indoor heating energyHigh Poor VBDD fit or low VBDD est.Billing estimates not meaningful.High
Phase I data filters, part 2 9 By the numbers…. nFilter description Total number flagged by filter n Number that survive prior filters but get caught by filter n Sample size after first n filters NoneNA 321 1Has large outdoor heating load Has DHP Missing billing data Missing SEEM input data Failed SEEM Run for other reason Has non-utility heat Poor VBDD fit or low VBDD est Phase I sample size:140
Regression background 10 Single-family calibration found that the difference SEEM 69/64 - VBDD tends to be more greater (more positive) in homes with… Poor weatherization (high U-values), Colder climates, Electric resistance heat (instead of gas or heat pump). The SF regression estimated these variables’ effects individually -- a separate coefficient for each.
The Uo trend for manufactured homes… 11 Observations: 1. Looks similar to SF case. 2. Minor variations in data filters lead to a distinct “dip” around Uo=0.12. (partially caused by a handful of homes with 5+ occupants). 3. Smaller MH sample size makes it very hard to separate different variables’ effects (Uo, equipment, climate, other?). Pre-1992 Post-1992
The (SEEM 69/64) heating intensity trend for manufactured homes… 12 Observations: 1. Trend is more clear. 2. Consistent across minor data filter variants. 3. Naturally combines several important variables (Uo, equipment, climate, etc.) 4. Steeper drop at far left due to combined effects of efficient equipment, good weatherization, mild climate.
Wait, what? 13
Capturing the (approximate) heating intensity trend Can’t represent loess smoother (black curve) in simple Excel formula. 2. Splines would work, but aren’t necessary… 3. A cubic polynomial (blue curve) captures the trend very well. 4. SF calibration used a piecewise-linear function for this. That works okay but has caused some headaches.
Regression model fit (v1) VariableCoefficientStd. Errort valuep value (Intercept) (SEEM 69 kWh / sq. ft.) (SEEM 69 kWh / sq. ft.) (SEEM 69 kWh / sq. ft.) Gas heat Heat pump Adjusted R-square: 38% 15 Dependent variable is (SEEM – VBDD)/SEEM. Energy intensity variable gives the regression a chance to take care of climate, heating system, and heat loss all at once. Still had to check to see if the regression treated these variables “fairly”. Some equipment variables still needed to be included. Use caution interpreting these variables’ coefficients.
Regression model fit (v1) 16 Curviness at right end isn’t really data-driven. (It’s caused by the polynomial form of our model, not a pattern in the data.) May be better to force the graph to flatten to the left of the local max (x≈14). See v2.
Regression model fit (v2) 17 Idea is to preserve the shape of the data-driven portion of the v1 polynomial (left part). Method: Define a new variable that equals the v1 polynomial up to the local max, then stays constant to the right. Fit new regression replacing polynomial terms with new variable.
Regression model fit (v2) 18
Regression model fit (v2) 19 Almost identical to v1 to the left of x = Main change is that it’s flattened out to the right. Smallest SEEM.69/sq. ft. values by heat source: HP: 1.8, 1.8, 1.8, 1.8, 2.0 ER: 3.8, 3.9, 4.1, 4.4, 4.6 Gas: 5.4, 6.5, 6.7, 7.3, 7.5
Phase I Adjustment Factors (v2) 20 Note: Only have about 15 points with SEEM/ft 2 < 4. Need to take care around lower x-value range. Not safe to extrapolate beyond observed data.
Phase I Calculation Example 21 Measure: Attic insulation from R8 to R19. Examples are randomly chosen RBSA sites, not measure prototypes. Home A: Zone 1, Heat pump SEEM.69 kWh Square feet IntensityPhase I factor Phase I kWh Attic R84, ,786 Attic R193, , Home B: Zone 2, Elec. FAF Attic R821, ,928 Attic R1916, ,862 4,6872,065
Phase I Decision “I, __________, move that for existing manufactured homes, the RTF approves the Phase I calibration described above (v2).” 22
Phase I Decision for NC? “I, __________, move that for manufactured- home new construction measures, the RTF approve the Phase I calibration described above, but with the following modifications…” 23
Phase II: Electric heating energy in “program-like” homes 24
Phase II general approach 25 Phase I gave us total heating energy estimates for homes with clear VBDD signatures. RTF measure savings needs average electric energy savings for all program homes. Phase II uses regression to find out… – How the presence of non-electric fuels affects electric heating energy, – How heating energy differs in homes with unclear VBDD signatures. Regression focuses on TMY-normalized (VBDD) estimates derived from electric billing data.
Phase II data filters, part 1 26 Sites excluded from Phase-II analysis for three reasons. Don’t have to worry about SEEM input data because Phase II doesn’t use SEEM estimates. Don’t want to remove sites with non-electric heating fuels or weak VBDD signatures since our goal is to estimate those features’ effects. Limit sample to “program-like” homes so we can capture dynamics programs are likely to see. Filter definitionReason for exclusionBias risk Has large outdoor heating loadBilling data can’t isolate indoor heating energyLow Has DHPOut of scope (will calibrate MH DHP separately)NA Missing billing dataCan’t generate VBDD estimate for analysisLow No hard-wired electric heatMinimal program eligibility screenNA
Phase II data filters, part 2 27 By the numbers…. nFilter description Total number flagged by filter n Number that survive prior filters but get caught by filter n Sample size after first n filters NoneNA 321 1Has large outdoor heating load Has DHP Missing billing data No hard-wired electric heat No gas FAF Phase II sample size:201
Phase II regression fit 28 VariableEstimateStd. Errort valuep value Intercept ln(UA × HDD65) Heat pump Gas Ht. (kWh) 4K to 8K Gas Ht. (kWh) Over 8K Wood (kWh) 6K to 12K Wood (kWh) Over 12K Bad VBDD fit Adjusted R-square: 40%
Phase II adjustments, part 1 29 Variable Regression coefficient Adjustment, affected sites Percent of sites affected Net average adjustment Net additive adjustment, Gas Ht. (kWh) 4K to 8K %0.7%-0.2%-0.3% Gas Ht. (kWh) Over 8K %0.5%-0.3%-0.5% Wood (kWh) 6K to 12K %15%-5.0%-4.9% Wood (kWh) Over 12K %11%-5.9%-4.8% Total adjustment due to energy displaced by other fuels:-10.4% Bad VBDD fit %28%-7.8%-6.8% Total adjustment due to energy that doesn’t exist: -17.3%
Phase II Decision “I, __________, move that the RTF adopt the Phase II calibration for existing manufactured homes as described in the previous slides.” 30