SEEM 94 Calibration to
Single Family RBSA Data
Analysis and proposed actions
RTF SEEM Calibration Subcommittee
May 7, 2013
Goals for today’s Subcommittee Meeting
• Review the following presentation in detail
– Consensus on next steps
• Are there any needed changes to the analysis?
• Is there subcommittee consensus that the RTF should
make a decision stating SEEM94 is calibrated?
– Receive suggestions from the subcommittee for
improvements in the presentation
• Does it adequately tell the full story?
• Is it the appropriate tool present to the RTF (assuming
previously covered sections will be skimmed over)?
2
SEEM 94 Calibration to
Single Family RBSA Data
Analysis and proposed actions
Regional Technical Forum
May 21, 2013
Overview
• Background
– Purpose
– History
• Methodology
– Data
– Regression
– Calibration
• Discussion
• Proposal
Overview - 4
Overview
• Background
– Purpose
– History
• Methodology
– Data
– Regression
– Calibration
• Discussion
• Proposal
Overview - 5
Purpose of Calibration
6
Align SEEM with Measured Energy Use
• The SEEM model is used to estimate energy savings for
most space-heating-affected residential UES measures
using the “calibrated engineering” estimation
procedure (see section 2.3.3 of guidelines)
–
–
–
–
–
Heat Pumps and Central AC (ASHP, GSHP, DHP)
Weatherization
New Homes
Duct Sealing
Space Conditioning Interaction Factor
• Goal: Ensure SEEM94’s results are grounded in
measured space heating energy use of single family
homes. Use RBSA as source of measured data.
Background - Purpose
7
RTF Savings Guidelines
2.3.3.2. Model Calibration
In most cases, calibrated engineering procedures will involve at least one stage of modeling in
which baseline and efficient case energy consumption are estimated for the measure-affected
end use. For example, the heating load for single-family homes is estimated as part of the
derivation of UES for ductless heat pump conversion. A simulation model is used to derive the
heating end use for typical homes in different climate zones. Ideally, the model would be
calibrated to measured heating end use for a sample of homes. If end use data are not
available, the model should at least be calibrated to metered total use for the sample.
Calibration should also be performed for samples that have adopted the measure, i.e., the
efficient case. For measures that affect new buildings the calibration may be limited to the
efficient case or to comparable buildings of recent vintage.
Background - Purpose
8
Overview
• Background
– Purpose
– History
• Methodology
– Data
– Regression
– Calibration
• Discussion
• Proposal
Overview - 9
RTF Decision History
Date
RTF Decision
Summary
Housing Type
T-stat Results
Nov-2009
SEEM 92 model is
calibrated.
Single Family
HP & Gas FAF
70°F Day ; 64°F Night
Electric FAF and Zonal
66°F Day & Night
Apr-2011
SEEM 93 model is
calibrated.
(implicit decision)
Single Family with
GSHP
70°F Day ; 64°F Night
Dec-2011
Use updated
SEEM94 model
Single Family,
Manufactured
Home
Dec-2011
Sep-2012
SEEM 94 model is
calibrated
SEEM 94 model is
calibrated
n/a
Manufactured
Home
69.4°F Day
61.6°F Night
Multifamily
Walk-up and Corridor
68°F Day& Night
Townhouses
66°F Day & Night
Data Sources Used in Calibration
1. Res New Const. Billing Analysis (RLW 2007)
2. SGC Metered Data
3. NEEA Heat Pump Study (2005)
Note: Very limited representation of Zones 2 & 3
1. Missoula GSHP Study (1996)
Ecotope updated SEEM code to model the physics of the
house infiltration, rather than rely on a constant stipulated
infiltration rate input in previous versions of SEEM.
1. NEEM 2006
2. NEEA Heat Pump Study (2005)
3. MAP 1995
4. RCDP (manufactured homes)
1. Multifamily MCS (SBW 1994)
2. MF Wx Impact Evaluation for PSE (SBW 2011)
3. New Multifamly Building Analysis (Ecotope 2009)
4. ARRA Verification for King County (Ecotope 2010)
Background - History
10
RTF Decision History (Continued)
For “model is calibrated” decisions…
Calibration Methodology:
1. Use available house and operation characteristics data
from billing/metering studies to develop inputs to SEEM
runs;
2. Adjust SEEM thermostat setting input to achieve a good
match (on average) between SEEM output (annual
heating energy use) and billing/metering study results.
Note: The data sources used were free of (or mostly free of)
supplemental fuel usage (wood, propane, oil, etc.)
• Collection of reliable electric and gas usage data for space
heat consumption is relatively easy compared to other
fuels.
Background - History
11
SF Calibration to RBSA - Recent History
Date
Forum
Topic
Proposal to adopt calibration:
Heating
High °F
(day)
Heating
Low °F
(night)
64
64
Gas FAF
68.6
63.9
Heat Pump
69.6
65.4
Heating
System Type
1/23/13
Full RTF
Electric Zonal
Electric FAF
3/20/13
Subcommittee
Status update and check in.
5/7/13
Subcommittee
Review staff’s proposal in
detail. Decide whether to
recommend RTF adoption.
Outcome
Links
Send staff back to
assess calibration
needs related to
climate and measure
parameters; and
engage
subcommittee.
Presentation
Minutes
Presentation
Minutes
Today
Presentation
Minutes
History - 12
Overview
• Background
– Purpose
– History
• Methodology
– Data
– Regression
– Calibration
• Discussion
• Proposal
Overview - 13
Methodology – Overview
14
Two Sources of Heating Energy Estimates
RBSA-PRISM. Estimates of annual “space heating use” for each
house determine by using PRISM
– PRISM is a “change-point” regression model that uses billing data to
estimate temperature-sensitive use
– PRISM analysis based on monthly billing data (at least 2years)
SEEM. Estimated annual space heating energy use for each
house based on SEEM engineering model
– RBSA individual home characteristics (e.g., thermal envelop,
heating system type, duct tightness) used as model inputs;
– Initial model runs use thermostat set to 68°F day & night
• SEEM is a one-zone model, so t-stat setting input represents the average
setting for the entire house
• Actual t-stat settings are not well documented (occupant reported
settings are unreliable, especially for “zonal systems”)
• Thermostat setting will be used (step 2, below) as the “calibration knob”
Methodology - Overview
15
Step 1 (Regression)
Use regression techniques to identify building
characteristics that drive systematic differences
between SEEM(68°F) and PRISM space heating energy
use estimates.
Methodology - Overview
16
Step 2 (Calibration)
Use regression results to determine thermostat setpoint that will align (i.e., “calibrate”) SEEM with PRISM
annual space heating use
– Calibration based on comparing average of all SEEM (68)
annual estimates to average of all PRISM annual estimates
– Calibration is based on building characteristics identified in
regression.
– SEEM run for each house at varying “day-time” thermostat
settings, with “night-time” thermostat settings based on
occupant reported setbacks
Methodology - Overview
17
Overview
• Background
– Purpose
– History
• Methodology
– Data
– Regression
– Calibration
• Discussion
• Proposal
Overview - 18
Data Sources
Data Source used in this calibration:
Underlying database* for the Single Family Residential
Building Stock Assessment (2012)
– RBSA study’s database offers recent billing analyses results
and detailed house characteristics on 1404 single family
houses in the Region.
– RBSA data allows inputs for SEEM runs to be well defined
for individual homes.
* Using a pre-release version of the database for this analysis .
Methodology - Data
19
Key Model Input
Parameters
RBSA Data Availability
UA
Available for each house.
Weather
Zip code (available for each house) linked to nearest TMY3
weather station.
Gas Heating Efficiency
Available for some houses; used average for remaining houses.
HP Operation & Efficiency
Not readily available. Used ARI control & 7.9 HSPF.
Duct System Leakage and
Surface Area
Available for some houses; used average for remaining houses
with ducts.
Duct System Insulation and
Location
Available for each house.
Infiltration
Available for some houses; used a floor area-scaled average (by
foundation type) for remaining houses
Mechanical Ventilation
Not available. Assumed 2 hours /day at 50 cfm.
Non-Lighting Internal Gains
Not available. See next slide for details.
Lighting Internal Gains
LPD available for each house; assumed 1.5 hours/day.
T-stat Setting
Available based on interviews, but used this as the “calibration
knob”.
Methodology - Data
20
Detail: Non-Lighting Internal Gains
• Equation:
𝐼𝑛𝑡𝑒𝑟𝑛𝑎𝑙 𝐺𝑎𝑖𝑛𝑠
𝐵𝑡𝑢
ℎ𝑟
= 805 + 367 × 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑃𝑒𝑜𝑝𝑙𝑒
• Based loosely on Building America Benchmark*
– Used the original equation and values (averaged) to determine average
internal gains for RBSA homes.
• Original equation also includes Number of Bedroom and Finished Floor Area terms
– Set Number of Bedrooms and Finished Floor Area terms to zero and adjusted
Number of People term to achieve same average internal gains for RBSA
homes.
• Building America Benchmark based on
– “The appliance loads were derived by NREL from EnergyGuide labels, a
Navigant analysis of typical models available on the market that meet current
NAECA appliance standards, and several other studies. ”
– “The general relationship between appliance loads, number of bedrooms, and
house size, was derived empirically from the 2001 RECS. ”
*Hendron, Robert. "Building America Research Benchmark Definition, Updated
December 20, 2007." NREL/USDOE EERE. January 2008. NREL/TP-550-42662
Methodology - Data
21
Realistic SEEM Simulations Not Feasible/Possible for All
Homes in RBSA
Issue
Count
More than one foundation type
331
25% > Ceiling Area to Floor Area > 200%, or Missing Ceiling U-value
36
Footprint Area to Floor Area < 20%
36
30% > Wall Area to Floor Area > 200%, or Missing Wall U-value
24
Missing Floor U-value for Crawlspace Foundation
5
Window Area = 0
3
Window u-value = 0
3
• Resulting House Count: 1011
– These issues overlap on some houses, so the sum of
the counts cannot be subtracted from 1404 to get
1011.
Methodology - Data
22
Data Filters Excluded Some RBSA Homes
Variable
Filter
Value(s)
Notes
SEEM Run
Valid
SEEM run must be valid (> 0 kWh/yr).
4
Billing Energy Use
> 1,500
kWh/yr
Intends to screen out partially used or
unused houses
38
eRsq and gRsq
= 0 or
≥0.45
Screens out houses with poor billing analysis
results (0.45 per David Baylon)
398
Non-natural-gas &
non-electric Fuel Use
0
Screens out houses with wood, oil, propane,
etc. consumption because billing analysis not
performed.
352
Primary Heating
System
eZonal,
eFAF,
gFAF, HP
Removes gas boilers, wood stoves, etc.
216
Secondary Heating
Electric
Removes wood stoves, propane heaters, etc.
System Fuel
or Gas
• Gas Billing converted to kWh/year using reported AFUE
• Resulting House Count: 293
• (The counts for each item overlap here, too)
Methodology - Data
Count
(filtered out)
274
23
Additional Data Filter for PRISM Excluded
Additional Homes
Exclude any home that had an out-of-range PRISM T-balance
for one or more components.
– The PRISM analysis restricted balance point temperatures to be
between 48 and 70 ⁰F.
• T-balance below 48⁰ is plausible.
• T-balance above 70⁰ is not physically plausible. We filter these out
since such values are evidence of a poor PRISM fit.
• Our 293 sites’ T-bal values include…
– 10 that defaulted to 70⁰ when PRISM’s initial fit exceeded the
max. (Excluded from analysis.)
– 16 that defaulted to 48⁰ when PRISM’s initial fit was below the
minimum. (Kept)
– 267 whose PRISM balance points were within the acceptable
range. (Kept)
• This leaves 283 sites for the present analysis.
Methodology - Data
24
Final
Data
Set
SEEM values calculated with t-stat = 68°F (constant)
Seem (68) Heating Energy (kWh)
Methodology - Data
25
Overview
• Background
– Purpose
– History
• Methodology
– Data
– Regression
– Calibration
• Discussion
• Proposal
Overview - 26
Methodology - Regression
27
Regression Overview (1)
• Analysis
Identify and quantify any systematic patterns (trends) in
the differences between SEEM(68°F) and PRISM savings
estimates ( ∆ kWh = SEEM(68°F) kWh ‒ PRISM kWh.
• Systematic means “explained by known variables.”
(Example: SEEM(68°F) kWh tends to exceed PRISM
kWh in cooler climates.)
• Tacit assumption: PRISM estimates roughly unbiased.
• Definitions
• “PRISM kWH” = Heating energy use via billing analysis; from
RBSA SF dataset.
• “SEEM(68°F) kWh” = Heating energy use via SEEM runs using
house-specific characteristics data from the RBSA SF dataset
with thermostat set to 68°F
Methodology – Regression
28
Regression Overview (2)
• Problem is multivariate…
– A single underlying trend (example: ∆ increasing with
heating use) may appear in multiple guises (∆
increasing with HDD, or with U-value, or with building
heat loss)
• Approach is multiple regression…
– Compare PRISM kWh with SEEM kWh when SEEM is
run with a constant T-stat setting (68°F day, 68°F
night.)
– Y-variable is the percent difference between SEEM
kWh and PRISM kWh (when SEEM uses T-Stat=68°F).
– X-variables are physical characteristics known through
RBSA. (Specifying the x-variables is a large part of the
work.)
Methodology – Regression
29
Setting up the Regression (1)
Primary interest is in differences between SEEM(68)
kWh and PRISM kWh—the Y-variable must capture
these differences.
– Heteroskedasticity. The SEEM(68) /PRISM differences
generally increase in magnitude in proportion to SEEM(68)
kWh (or PRISM kWh). (See earlier graph.)
– Measurement error (random noise). As estimates of
heating kWh, SEEM(68) and PRISM both have substantial
standard errors.
Methodology – Regression
30
Setting up the Regression (2)
𝑦
= Percent difference between SEEM and PRISM
=
SEEM kWh − PRISM kWh
?? kWh
• Note choice of signs: 𝑦 > 0 means SEEM > PRISM.
• What goes in the denominator? “??” = “Actual kWh”
would be ideal.
– Using SEEM kWh or PRISM kWh would skew y-values.
(Next slide.)
– Log-transforms (closely related) not quite right either.
– Instead, divide by midpoint: “??” = (SEEM + PRISM)/2.
(Two slides down.)
Methodology – Regression
31
Dividing by PRISM kWh magnifies differences where PRISM’s random error
happens to be negative (since these values get artificially small denominators).
This biases the percent differences upwards.
Methodology – Regression
32
Upward bias (mostly) goes away when we divide by the value halfway between
SEEM(68°F) and PRISM.
Methodology – Regression
33
Upward bias (mostly) goes away when we divide by the value halfway between
SEEM(68°F) and PRISM.
Methodology – Regression
34
Upward bias (mostly) goes away when we divide by the value halfway between
SEEM(68°F) and PRISM.
Methodology – Regression
35
Upward bias (mostly) goes away when we divide by the value halfway between
SEEM(68°F) and PRISM.
Methodology – Regression
36
Building the Regression Model
• Goal is to identify variables that lead to
systematic differences between SEEM(68°F) and
PRISM.
– “Lead to” is only seen in rough trends (think:
correlation).
• Looking to capture unknown effects – not a physical
model.
• Model development is iterative.
– A variable may be weakly correlated with raw y-values
but strongly correlated with y’s that have been
adjusted to account for some other variable’s
influence.
Methodology – Regression
37
Important Limitations
• Avoiding Colinearity - When a potential x-variable closely
tracks some combination of variables that are already
included.
– Example – Including both heat loss rate and vintage
– This redundancy leads to unstable model fits.
– Threshold for “tracks too closely” gets low when the usual suspects
are around:
High noise / faint signal / small sample.
• Pursuing Parsimony. General principle: Don’t over-fit the data
(by including too many explanatory variables).
• Incomplete data variables. Some variables (e.g., duct
tightness and infiltration) aren’t known for very many houses.
Methodology – Regression
Methodology (Regression)
- 38
Prominent x-variable candidates
• Characteristics that likely influence differences
between SEEM(68°F) and PRISM estimates of
use
– Thermal efficiency drivers (U-values, duct
tightness, infiltration, …)
– Heating system type
– Climate (i.e., HDDs)
• Following graphs illustrate “influence” of
several variables (separately) on percent
difference between SEEM(68°F) and PRISM
Methodology – Regression
39
SEEM 68 −PRISM
Midpoint
SEEM 68 −PRISM
Midpoint
Methodology – Regression
40
SEEM 68 −PRISM
Midpoint
SEEM 68 −PRISM
Midpoint
Methodology – Regression
41
Methodology – Regression
42
Insulation Variables
The big surfaces: Wall, ceiling, and floor.
• Express in terms of heat loss (U-values,
weighted by surface area as appropriate)
• We separate out Floor U because of different
foundation types.
– One variable accounts for ceiling and wall heat
loss.
– Another variable accounts for floor heat loss in
crawlspace homes.
Methodology – Regression
43
Variable for Wall/Ceiling U
Applies to all homes (regardless of foundation
type).
A simple indicator variable:
“Wall/Ceiling Insulation is Poor” if
Wall u-value > 0.25, OR
Ceiling u-value > 0.25, OR
Both u-values > 0.25.
This variable captures the main effect of the
weighted average. (See next slide)
Methodology – Regression
44
Methodology – Regression
45
Variable for Floor U
Particularly interested in crawlspace heat loss
since crawlspace insulation is a common
measure.
Variable definition: “Yes/No” indicator for
uninsulated crawlspace.
Note: Sites with basements, slabs, and insulated
crawlspaces all have Uninsulated Crawl = “No”
Methodology – Regression
46
Do these indicator variables really capture the
insulation effects?
The next two slides compare various u-values’
relationships with
– Unadjusted (raw) percent differences;
– Percent differences that have been adjusted for
the two insulation variables included in the
regression.
Methodology – Regression
47
Methodology – Regression
48
Methodology – Regression
49
Heating System Variable
Four distinct heating systems in the sample:
Electric zonal
Gas FAF
Electric FAF
Heat pump
After controlling for insulation, heating system
effect appears to be captured with just two
groups:
“Electric Resistance” = Electric zonal / Electric FAF
“Gas/HP” = Gas FAF / Heat Pump
Parsimony: two is better than four!
Methodology – Regression
50
Methodology – Regression
51
Methodology – Regression
52
Model 1 fit summary:
(Intercept)
elec. resistance
poor.ins.ceil.wall
uninsulated.crawl
Est.
s.e.
p-value
-0.01
(0.04)
0.80
0.27
(0.05)
0.00
0.42
(0.08)
0.00
0.15
(0.07)
0.04
Adjusted R-square = 0.212
…and with an interaction term for insulation:
Est.
s.e.
p-value
Intercept
-0.18 (0.04)
0.62
elec. resistance
0.27 (0.05)
0.00
poor.ins. ceil.wall
0.49 (0.09)
0.00
uninsulated.crawl
0.21 (0.08)
0.01
poor.ins.c.w*unins.crawl -0.21
(0.16)
0.19
Adjusted R-square = 0.214
Methodology – Regression
No strong
recommendation
either way because
of low p-value, but
proposal is to drop
the interaction term.
53
Climate variable
HDD effect is not very pronounced. Next slide
shows percent differences (adjusted for effects
in the previous regression), versus HDDs
• Standard HDDs with constant (65⁰) base.
• Plot shows (slight) positive correlation
between HDDs and adjusted y-values.
• Group means (x-mean, y-mean) lie very near
the overall trend line.
Methodology – Regression
54
(Black line indicates OLS linear regression fit.)
Methodology – Regression
55
Climate Variable
• A modest linear trend is clear from the plot.
• Could either use indicator variables, or the
actual HDDs values (a single continuous
variable).
– Group means agree with overall linear trend
almost perfectly, so little practical difference.
• We use the continuous variable, x = HDDs.
Methodology – Regression
56
Previous fit:
Estimate
s.e.
p-value
(Intercept)
-0.01
(0.04)
0.80
elec. resistance
0.27
(0.05)
0.00
poor.ins.ceil.wall
0.42
(0.08)
0.00
uninsulated.crawl
0.15
(0.07)
0.04
Adjusted R-square = 0.212
And now with HDDs:
Estimate
s.e.
p-value
Intercept
-0.40
(0.15)
0.01
elec. resistance
0.27
(0.05)
0.00
poor.ins.ceil.wall
0.44
(0.07)
0.00
uninsulated.crawl
0.13
(0.07)
0.07
Base-65 HDDs
7.3e-5 (2.7e-5)
0.01
Adjusted R-square = 0.230
Methodology – Regression
57
Interpreting the HDD Coefficient
Our fitted coefficient for HDDs was 7.3 × 10−5 . What
does this mean in practical terms?
In our sample, the HDDs averages differ by about 1500
HDDS from one climate zone to the next.
Since 7.3 × 10−5 × 1500 = 0.1095, the climate zone
effect corresponds to about an 11% difference.
Methodology – Regression
58
So far, so good…
• The next two slides compare four variables’
relationships with
– Unadjusted (raw) percent differences; and
– Percent differences that have been adjusted for all
four variables included in the regression.
• HDDs and heat source show zero relationship
with adjusted differences.
• Square footage and internal gains relationships
went from weak to weaker (even though they are
not included in the model – that’s good!).
Methodology – Regression
59
Methodology – Regression
60
Methodology – Regression
61
Percent difference versus midpoint has also improved…
(And midpoint isn’t in the model either)
Methodology – Regression
62
And the insulation variables’ plots still look
good…
Methodology – Regression
63
Methodology – Regression
64
Methodology – Regression
65
What else should we consider?
Next slide indicates several variables’ correlation
with adjusted percent differences.
Methodology – Regression
66
Observations
• Duct leakage has the largest apparent
correlation, but this variable is sparsely
populated. (We’ll look at it next.)
• PRISM HDDs have moved up – these had
almost no correlation with unadjusted
differences. (We discuss at the end.)
• RBSA (reported) t-stat values have a slight
negative correlation with % differences.
(This sign makes sense, but we’d expect
more.)
Methodology – Regression
67
Duct Leakage
• Have direct RLF and SLF measurements for 33
homes;
• Also, 87 homes have no ducts (zero leakage);
• Another 38 (excluded from analysis) have
ducts entirely inside of conditioned spaces.
– Some of these spaces are basements designated
“conditioned” simply because they contain ducts.
– 26 of the 38 have “heated basements”
• Not much basis for calibration here…
Methodology – Regression
68
Methodology – Regression
69
Duct Leakage (continued)
• Visually, there’s not much correlation in the range
containing most of the data.
• Numerically…
– Correlation is 53% when only the 33 measured values
are included;
– Drops to 15% when 4 right-most points are omitted;
– Values drop to 31.5% and 5.3% when we include
homes without ducts (zero leakage).
• Weak (and ambiguous) basis for recommending
adjustment specific to duct-leakage.
Methodology – Regression
70
Infiltration
• Have direct infiltration measurements for 95
homes;
• But even less reason for calibration here…
Methodology – Regression
71
Methodology – Regression
72
Infiltration (continued)
• Visually, the relationship is null.
• Numerically…
– Correlation is -11.8% when all points are included;
– Changes to 2.3% when single right-most point is
omitted.
• No basis for adjustment for infiltration once
other variables are included.
Methodology – Regression
73
PRISM Balance Point
• A question was raised at the May 20
subcommittee meeting regarding whether the
regression should take into account the house
balance point determined by PRISM.
Methodology – Regression
74
Methodology – Regression
75
PRISM HDDs
• Definitely a trend, but is it unexpected? Does it require action?
– We know that unobserved variables drive a portion of SEEM-PRISM
deviations that we treat as noise.
– Consider a home where an unobserved variable yields an effective
balance point that is lower than we would expect based only on
observed variables. This home will tend to satisfy both:
• SEEM(68) kWh > PRISM kWh and
• TMY HDD > PRISM HDD
– In other words, the presence of unobserved variables causes a positive
correlation between kWh differences and HDD differences.
• Conclusion: The presence of unobserved variables should yield a
trend like the one seen on the previous slide.
• So the trend is what we would expect.
Methodology – Regression
76
Proposed Final Model
Variable
Estimated
coefficient
Intercept
-0.40
elec. resistance
0.27
poor.ins.ceil.wall
0.44
uninsulated.crawl
0.13
HDDs (Base 65)
7.3e-5
Standard
error
(0.15)
(0.05)
(0.07)
(0.07)
(2.7e-5)
p-value
0.01
0.00
0.00
0.07
0.01
𝑦
= 𝑖𝑛𝑡𝑒𝑟𝑐𝑒𝑝𝑡 + 𝛽𝑒𝑙𝑒𝑐.𝑟𝑒𝑠𝑖𝑠 × 𝐼𝑒𝑙𝑒𝑐.𝑟𝑒𝑠𝑖𝑠 + 𝛽𝑝𝑜𝑜𝑟.𝑖𝑛𝑠.𝑐𝑒𝑖𝑙.𝑤𝑎𝑙𝑙 × 𝐼𝑝𝑜𝑜𝑟.𝑖𝑛𝑠.𝑐𝑒𝑖𝑙.𝑤𝑎𝑙𝑙
+ 𝛽𝑢𝑛𝑖𝑛𝑠.𝑐𝑟𝑎𝑤𝑙 × 𝐼𝑢𝑛𝑖𝑛𝑠.𝑐𝑟𝑎𝑤𝑙 + 𝛽𝐻𝐷𝐷 × 𝐻𝐷𝐷
Adjusted R-square = 0.230
Methodology – Regression
77
Overview
• Background
– Purpose
– History
• Methodology
– Data
– Regression
– Calibration
• Discussion
• Proposal
Overview - 78
Final Step: Interpreting Results
• From the fitted model, we obtain adjustment
factors that apply to SEEM output to align
SEEM with the RBSA-PRISM data.
• A given site’s adjustment factor depends on
the values of the explanatory variables for
that site.
• Group HDDs by climate zone. Then for each
zone, there are 8 possible configurations of
the three other variables.
• This yields 24 distinct adjustment factors in all.
Methodology – Calibration
79
Calibration Factors
SEEM(68) differs from PRISM by these factors (on average)
100%
80%
60%
40%
20%
0%
No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes
No
Yes
Gas FAF / HP
No
Yes
Elec. Res.
Climate Zone 1
No
Yes
Gas FAF / HP
No
Yes
Elec. Res.
Climate Zone 2
Methodology – Calibration
No
Yes
Gas FAF / HP
No
Yes
Elec. Res.
Climate Zone 3
80
T-Stat Calibration
• Translate percent kWh adjustments into adjustments in daytime tstat setting (from 68 °F).
• No data limitations here: we can directly observe SEEM’s sensitivity
to t-stat settings.
• Method:
1.
Run SEEM for each house at multiple temperature settings in 2 degree increments
–
–
Daytime Settings: … 58, 60, 62, …
Nighttime Setback: Daytime setting - setback
» Setback: Use average difference between reported daytime and nighttime t-stat settings
in RBSA dataset; by heating system type:
Heating System Type
Avg Setback (°F)
Electric FAF
Electric Zonal
Heat Pump
Gas FAF
2.
3.
6.0
4.8
4.3
4.8
Determine relationship of calibration factors to temperature settings for each of
the 24 scenarios.
Interpolate to determine “calibrated” t-stat settings.
(need to add a graph to help explain this)
Methodology – Calibration
81
Calibrated Thermostat Settings
70
65
60
55
50
No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes
No
Yes
Gas FAF / HP
No
Yes
Elec. Res.
Climate Zone 1
No
Yes
Gas FAF / HP
No
Yes
Elec. Res.
Climate Zone 2
Heating System Type
Electric FAF
Electric Zonal
Heat Pump
Gas FAF
No
Yes
Gas FAF / HP
No
Yes
Elec. Res.
Climate Zone 3
Avg Setback (°F)
6.0
4.8
4.3
4.8
Methodology – Calibration
82
Overview
• Background
– Purpose
– History
• Methodology
– Data
– Regression
– Calibration
• Discussion
• Proposal
Overview - 83
Discussion
• Are we done?
• Decision: SEEM94 is “calibrated”; it will give
reliable heating energy consumption results
– for single family houses with the following
characteristics:
• Heating System is one or more of the following: Gas FAF,
Electric FAF, HP, zonal electric (no other heating system
type);
• Occupied/normal houses (PRISM worked);
– if the following inputs are used:
• Calibrated Thermostat Settings (see slide above); and
• Internal Gains:
𝐼𝑛𝑡𝑒𝑟𝑛𝑎𝑙 𝐺𝑎𝑖𝑛𝑠
𝐵𝑡𝑢
ℎ𝑟
= 805 + 367 × 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑃𝑒𝑜𝑝𝑙𝑒
Discussion
84
Next Steps
• If the RTF agrees it’s calibrated, the RTF will be able to use SEEM94
to help estimate energy savings for residential single family
– Heat Pump
• Conversions
• Upgrades
• Commissioning, Controls, and Sizing
– Weatherization
• Insulation
• Windows
• Infiltration reduction
– Duct Sealing
– New Home Efficiency Upgrades
• “Help” is used here because we will still need to deal with “nonelectric benefits” for these measures.
– This topic is out of scope for today’s discussion. The goal today is
simply to determine whether SEEM has been calibrated to provide
reliable results.
Discussion
85
Overview
• Background
– Purpose
– History
• Methodology
– Data
– Regression
– Calibration
• Discussion
• Proposal
Overview - 86
Proposed Motion
“I _______ move that the RTF consider SEEM94
calibrated for single family houses.”
Proposal
87