m&v 2.0: a user’s guide · what is new about m&v 2.0? what isn’t new? m&v 2.0 tools...
TRANSCRIPT
1
M&V 2.0: A User’s Guide
West Coast Energy Management
Congress
June 8, 2017
David Jump, Ph.D., P.E.
kW Engineering
Agenda
What is M&V 2.0?
Process Steps & Tools
Non-Routine Events
Benefits and Risks of M&V 2.0
Uses & Case Studies
What is New About M&V 2.0?
What isn’t New? M&V 2.0 tools are built upon savings estimation
techniques that have been used for decades Whole-building and submeter-based pre/post (Option C)
Retrofit Isolation (Option B)
What’s new is: Degree of automation in data acquisition, and model creation
Granularity and volume of data can improve quality of result
Potential for continuous feedback
Integration of M&V capability with other analyses for operational efficiency
Software as a service offerings for owners, managers, program administrators
Option C: Whole Facility
Data Sources:
• Utility bills
• Local weather stations
• Occupancy Schedules
• Production Rates
𝑘𝑊ℎ𝑠𝑎𝑣𝑒 = 𝑘𝑊ℎ𝑏𝑎𝑠𝑒 𝑇𝑝𝑜𝑠𝑡 − 𝑘𝑊ℎ𝑝𝑜𝑠𝑡(𝑇𝑝𝑜𝑠𝑡)
M&V 1.0 – Monthly Data
Linear regressions
12 months/data points per year
Less Accuracy
12 mo. monitoring duration
M&V 2.0 - lnterval Data
Advanced analytics
8760 hourly/365 daily points per
year
More Accuracy
Shorter monitoring duration: 3 to 6
months
Applicable to subsystems (Option B)
M&V 2.0• High frequency data & advanced analytics
• Rapidly process data and visualize key information
• Uses: Building Audits / Commissioning / M&V /
Performance Tracking
Whole Building
(Option C)
Boiler
VSD
Building Subsystem
(Option B)
Advanced Analytics
Familiar Linear OLS Regression
More Advanced ASHRAE RP1050 Change-Point Models
LBNL Temperature and Time-of-Week Model
Exotic Neural Networks
Nearest Neighbor
Machine Learning
Much More..
Comparison
Linear OLS Model
Temperature only
Advanced Model
Temperature and
Time-of-Week
Project Level M&V Tools
Public DomainASHRAE RP1050 Change Point Models Energy Explorer
http://academic.udayton.edu/kissock Energy Charting and Metrics Tool
(Excel add-in)http://www.sbwconsulting.com/ecam/
LBNL TTOW Model Universal Translator, v3
www.utonline.org
Various R, Python M&V Code
Transparent/Repeatable
Proprietary BuildingIQ FirstFuel Gridium More!
Validation with test data sets and protocols (LBNL)
https://eta.lbl.gov/publications/accuracy-automated-measurement
Process
Metering Concern: Meter Accuracy
Bias Measurement Error - eliminate
Random Measurement Error – reduced as more data used
Energy Source Type Typical Accuracy Common Mfgrs
Electric Solid state ± 0.2% of reading
Square D
Eaton
Natural Gas Positive displacement ± 1 - 2% of reading
Dresser
American
CHW/HHW
Temperature sensors: solid state
Flow meter: turbine, electromagnetic,
ultrasonic, or vortex
Temp sensors: ± 0.15°F from 32-200 °F
Flow meter: ± 0.2% to ± 2.0% per flow meter
Calculator accuracy: within ± 0.05%
Onicon
Flexim
Steam
Flow: Vortex shedding
Temperature: RTD Mass flow: ± 2% of mass flow calculation
Rosemount
Yokogawa
Mfgr’s product test results, installed meter calibration reports,
submitted with the documentation for all meters.
Coverage Factor
Good models have maximum range of energy and temperature values
Rule: Don’t extrapolate 10% beyond max or min baseline temperatures
Coverage factor determines how much data to collect prior to project install
Develop and Assess Models - I
Goodness-of-Fit and Accuracy Metrics
Baseline Models
NMBE (bias error) < 0.5%
CV(RMSE) (random error) < 25%
R2 (independent variables check) > 0.7
Linear Model, CV = 25% TTOW Model, CV = 11%
Develop and Assess Models - II Assess Uncertainty
F
mnn
nCV
tE
E
msave
msave
2/1
''
.
,
12126.1
U < 10%
CV(RMSE) 0.05 0.1 0.15 0.175 0.2 0.25
% savings
2% 11% 22% 33% 39% 44% 56%
4% 6% 11% 17% 19% 22% 28%
6% 4% 7% 11% 13% 15% 19%
8% 3% 6% 8% 10% 11% 14%
10% 2% 4% 7% 8% 9% 11%
12% 2% 4% 6% 6% 7% 9%
14% 2% 3% 5% 6% 6% 8%
16% 1% 3% 4% 5% 6% 7%
18% 1% 2% 4% 4% 5% 6%
20% 1% 2% 3% 4% 4% 6%
Uncertainty (90% CI)
Table shows acceptable CV for target savings F, if:∆𝐸𝑠𝑎𝑣𝑒
𝐸𝑠𝑎𝑣𝑒< 10% (90% CI)
Monitor Savings
Non-Routine Events
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
4/1
3/2
006
4/1
7/2
006
4/2
1/2
006
4/2
5/2
006
4/2
9/2
006
5/3
/2006
5/7
/2006
5/1
1/2
006
5/1
5/2
006
5/1
9/2
006
5/2
3/2
006
5/2
7/2
006
5/3
1/2
006
6/4
/2006
6/8
/2006
6/1
2/2
006
6/1
6/2
006
Date
Da
ily
kW
h
0
10
20
30
40
50
60
70
80
De
g.
F
AHU-3 Supply Fans Avg. Daily Temp.
SF-1 Fails. The other fans
ramp up to meet setpoint.
SF-1 repaired. Fans
return to normal
operation
Prefilters and
bags changed
Non-Routine Adjustments Process
Identify the NRE (visualize data or owner report)
Determine if NRE Impact is Material (if not, stop)
Assess
Temporary or Permanent?
Constant or Variable Load?
Added or Removed Load?
Quantify Impact
Engineering calcs + assumptions (low quality/cost)
Engineering calcs + logged data (med-high quality/cost)
Analysis of before/after NRE using metered data (high
quality/low cost)
Adjust Savings Estimate
Non-Routine Adjustment
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
4/1
3/2
006
4/1
7/2
006
4/2
1/2
006
4/2
5/2
006
4/2
9/2
006
5/3
/2006
5/7
/2006
5/1
1/2
006
5/1
5/2
006
5/1
9/2
006
5/2
3/2
006
5/2
7/2
006
5/3
1/2
006
6/4
/2006
6/8
/2006
6/1
2/2
006
6/1
6/2
006
Date
Da
ily
kW
h
0
10
20
30
40
50
60
70
80
De
g.
F
AHU-3 Supply Fans Avg. Daily Temp.
Difference ~ 500 kWh
Avoided Energy Use (actual savings)
-40.0
-20.0
0.0
20.0
40.0
60.0
80.0
100.0
150
250
350
450
550
650
750
850
950
De
g F
kW
h
` ,
Baseline Period Post Install Period
Source: Universal Translator v3
Normalized Savings
Reduces risk of
extreme weather years
Case Study - Pay For Performance
1 2 3
•Cumulative savings -
continuous tracking &
feedback
0
50
100
150
200
250
Th
erm
s/d
ay
Actual
Expected
Energy ‘Diagnostics’1. Software identifies high natural
gas usage, especially on weekends.
Estimated Savings: 65,000 kWh
(Annual) 14,000 therms
$20,000
2. Trend review uncovers AC
units running continuously over
the weekend.
0
20
40
60
80
100
120
140
160
Th
erm
s/d
ay
Actual
3. AC units
rescheduled
M&V Documentation
M&V Plan
Describe Model
Why chosen?
Mathematical form
Independent variables
Baseline Period
Coverage factor
Goodness-of-fit statistics
Uncertainty Assessment
Calculations
How often & how savings are reported
Non-routine adjustments
More!
Best Applications – Project Level M&V
‘Predictable’ buildings, systems
Weather sensitive, regularly scheduled
Multiple and interactive ECMs
Affecting multiple building systems (HVAC, lighting, etc.)
Deep savings projects
Savings are “above the noise”
Data useful for other purposes
Anomaly detection, Performance drift
What are the Potential Benefits of M&V
2.0? What is the Value Proposition?
Increase visibility, quickly obtain ongoing and interim
results feedback
Increase savings and enhance customer experience?
Improve transparency and trustworthiness of EE savings?
Automate parts of the process that computers do well,
streamline data acquisition and processing
Reduce time and cost to quantify savings?
Maintain/improve accuracy in savings?
Increase throughput, number of projects going through the
pipeline?
Risks and Issues
Sub Meter Calibration Requirements & Frequency
Complex Analysis Methods
Not simple OLS anymore!
Unpredictable buildings
Prescreening may be required
Non-Routine Events
Added building loads, major occupancy shifts
Must remove impacts from savings estimations
Data accessibility and security (not covered)
Predict/Forecast
The Good
The Bad
The Ugly
Good buildings:
Predictable operation
Bad buildings
Requires intervention?
Ugly buildings
Cannot predict future use