what to do with all that 2-way end-device...
TRANSCRIPT
© 2015 Eaton. All rights reserved..
What to Do with All That 2-Way End-Device Data
Joseph E. Childs – Sr. Program Manager
Ryan Brager – DR Product Manager
Yigang Wang, Ph.D. - Corporate Research & Technology (CRT)
51st Annual MINNESOTA POWER SYSTEMS CONFERENCE
2 © 2015 Eaton. All rights reserved..
What to Do with All That 2-Way End-Device Data
• This session reviews the value assessment of
utilizing two-way data, and addresses some of the
challenges of managing the abundance of data
available. Discussion will occur about optimizing
the data received through AMI, load management,
and volt-var control applications.
First Minneapolis Presentation: Wood, M.C., J. E. Childs, W. E. Marlatt and D. Renne.
1979 Assessing the Air Pollution "Carrying Capacity" of Northern
Puget Sound: An Applications of TAPAS. Conference Proceedings, 14th Conference on Agriculture and Forest Meteorology &
Fourth Conference of Biometeorology, Minneapolis, Minnesota.
3 © 2015 Eaton. All rights reserved..
Agenda
• The Problems
• Renewable Integration
• Energy Markets
• Value Assessment
• Data Analysis Refresher
• Case Studies
Engineering …. Is the art of doing that
well with one dollar, which any bungler
can to with two after a fashion.
4 © 2015 Eaton. All rights reserved..
PV Market and Policy Drivers
• Decreasing by 30%/year • Driven by off-shore manufacturing (China)
• Initially driven by mandates and subsidies
Note: Data for solar only http://www.seco.cpa.state.tx.us/publications/renewenergy/
2020: 15%
2025: 25%
2020: 33%
2025: 25% 2020:
30%
2030: 40%
2015: 29%
ME - 2020: 30%
VT - 2025: 24.8%
NH - 2017: 20%
MA - 2020: 22.1%
RI - 2020: 16%
CT - 2020: 27%
NJ - 2021: 20.4%
DE - 2026: 25%
MD - 2022: 20%
2015: 15%
2015: 10%
2015: 10%
2020: 20% 2025:
20%
2025: 15% 2020:
20%
2015: 15%
2021: 15%
2021: 18%
105 MW
2015: 5880 MW
< 10%
10% - 20%
> 20%
Legislated Renewable Standards Policies and Goals IREC North Carolina Energy Center
“inflection point” • Economic / social benefits prevail (< $0.10/kWh) • Renewable penetration exceeds 5% (point of
stability problems)
PV Price & Penetration Texas Renewable Energy Resource Assessment
• Penetration of PV at inflection point
• Other intermittent sources and loads continue to drive complexity in distribution grid
5 © 2015 Eaton. All rights reserved..
Distribution System with PV – Power System Models
IEEE 34-Bus Test Feeder with PV integration
Voltage rise
introduced by
reverse power
flow from PV
integration Vo
lta
ge
% o
f n
om
inal
Vo
lta
ge
% o
f n
om
inal
6 © 2015 Eaton. All rights reserved..
Intermittency PV Power Output Real vs. Idealized
Intermittency is one major challenge of renewables
Data source: NREL
PV Power Output “Ideal” Insulation
7 © 2015 Eaton. All rights reserved..
Energy Business Landscape driving Data Analysis
• Regulations
• FERC Orders face challenges (745)
• EPA 111d
• States still have authority on many issues
• Electrical Markets (5, 15, 60 minute) • ISO Markets
• Different market flavors
• Different retail and wholesale settlement
• IOUs and G&Ts required to participate (Moslty)
• Coops and Munis exempt from Competition, but affected
• RTO Markets
• Balancing Authorities
• Inability to Monetize Value
• Electrical Operations • Mismatch between Markets and Operations
• Do More with Less
Too many cooks spoil the stew or Design by committee or …
8 © 2015 Eaton. All rights reserved..
Markets – ISO/RTO DR Program Comparison
IRC Database
Programs (52)
Attributes (55)
• ISO/RTO Product
• Product / Service Features
• Deployment
• Market Participant Roles
• Communications
• Event Timing
• Telemetry
• After The Fact Metering
• Performance Evaluation Filters:
• Resource Type
• Aggregation Allowed http://www.isorto.org
Data from 2013 updated in 2014
GIWH Markets
Market Capacity Energy Reserve Regulation
AESO ● ● ●
CAISO ● ●
ERCOT ● ● ●
IESO ● ●
ISO-NE ● ● ●
MISO ● ● ●
NYISO ● ● ●
PJM ● ● ● ●
SPP ● ● ●
ISO Count 5 9 8 4
Programs 12 18 11 4
9 © 2015 Eaton. All rights reserved..
Settlement - ERCOT C&I Baseline
• Middle 8 of 10 Preceding Like Days
• Filter out Non Similar Days
• Take Last 10 Days
• Remove Highest and Lowest
• Average Remaining
• Matching Day Pair
• Today Pair (Yesterday & Today)
• Historic Pairs for 1 Year
• Sum Square Difference Interval Data
• Select 10 Lowest SSD
• Average Intervals
• Statistical Regression Model
• kW(d,h,int) = F(Weather(d), Calenday(d),
Daylight(d))
• Coefficients Calculated Offline
• Uses “Lookup” Factors
Final Baseline
Event Day Adjustment
– Event Day Interval Data
– Baseline Interval Data
– Factor = Baseline Sum /
Event Day Sum
– Final Baseline(t) = Factor*
Unadjusted Baselline(t)
10 © 2015 Eaton. All rights reserved..
Agenda
• The Problems
• Value Assessment
• AMI
• DR
• CVR
• Statistical Analysis Refresh
• Data Analysis
“The Graduate” 1967
Mr. McGuire: I just want to
say one word to you. Just
one word.
Benjamin: Yes, sir.
Mr. McGuire: Are you
listening?
Benjamin: Yes, I am.
Mr. McGuire: Analytics
…
11 © 2015 Eaton. All rights reserved..
AMI Business Case Value Assessment
ROI Analysis (NPV) Baseline Analysis
10 15 20
Total Cost of Ownership (1,468,000) (1,752,000) (1,985,000)
Total Benefits 1,992,000 2,762,000 3,394,000
Present Value Annualized ROI 3.6% 3.8% 3.5%
Total Benefits/ Cost of Ownership Ratio 1.36 1.58 1.71
NPV of Investment 524,000 1,010,000 1,409,000
Cash Flow Positive (Years) 2
Project Breakeven (Years) 6
65% of AMI Value comes using Interval Data (Customer Service, Operations etc.)
12 © 2015 Eaton. All rights reserved..
Demand Response Value Assessment
Supply Value Analysis
10
14,918,634$
45,218,539$
30,299,905$
20.3%
3.03
3Project Breakeven (Years)
Analysis Period
Total System Costs (NPV)
Total Value from System (NPV)
DR Program Benefit (NPV)
Annual Return on Investment
Value/ Cost Ratio
DR Benefit
Equivalent
Supply Side DR System
4,905,297$ 19,823,931$ 14,918,634$
162$ 829$ 668$
3%
1.33
ROI (Annual)
DR / Supply Side Ratio
Peaking Plant Comparison
Present Value Costs
Cost per kw
Data Value: Fast DR – Market Value 2 to 5 times capacity and energy markets
Data Cost Reductions: DR System Maintenance & Marketing & Enrollment
13 © 2015 Eaton. All rights reserved..
CVR: Peak Power Price Model
Savings Voltage Reduction during Peak Hours
Percent/Voltage Reduction (Avg.) per Feeder 3.0% Actual Volts Reduced per Feeder 3.75 CVR Factor 0.7 Actual kW Yield (%) 2.10% kW Reduction from Peak 8,400 Expected Daily Hours of CVR On-Peak 6 Expected Number of Days of Operation per week 5 Annual Hours of CVR 1560 Estimated Yield (reduction) Per Year (kWH) 13,104,000 Peak Generation Cost per KWH $0.08 Peak Demand Charges $126,000.00 Annual CVR Savings Potential $1,174,320
Revenue Loss Peak Distribution Charge (per KWH) to Consumer $0.17 Annual Revenue Loss $2,227,680
Eaton Volt/Var Management Value Assessment Calculator
Data: 5 to 15 second per phase voltage along feeder
14 © 2015 Eaton. All rights reserved..
Agenda
• The Problems
• Value Assessment
• Data Analysis Refresh
• References and Methods
• Fundamental Concepts
• Data Samples
• Data Analysis
15 © 2015 Eaton. All rights reserved..
References and Key Concepts
• Reference Material
• The Signal and the Noise, by Nate Silver. 2012. “Why
So Many Predictions Fail – but Some Don’t”
• Analysis & Forecasting
• Regression vs.
• Time Series vs.
• Non-Parametric and Distribution-Free Methods
• Reporting
• Net Present Value
• Baseline
16 © 2015 Eaton. All rights reserved..
Analysis & Modelling Fundamentals
• Representative Sample Data
• Independent measurements of dependent and predictor variables
• Non-biased sample set
• Measurement errors are random (or distribution is known)
• Sample size is adequate
• Model errors (variance is stable)
• Range of predictors is adequate
• Continuous Over Prediction Range
• Model Assumptions
• Weak Exogeneity. Predictor variables are fixed and error free
• Linearity vs. Non-Linear. Predictor relationship can be expressed
as a*X + b*Y where a and b are constants
• Independence. Errors are uncorrelated
• Minimal Multi-Collinearity. Predictors don’t have perfect correlation.
17 © 2015 Eaton. All rights reserved..
Sample Data – Plot of Runtime vs. Temperature Data Error
Outliers
Variance
Sample not Uniform
Sample:
~146,000 data points
50 Thermostat Data Logs
June 1 – September 30
(with some missing data)
Out of Range – Not Continuous
• What are the sources of bias and errors? • Is the relationship linear?
• What is going on with the outliers?
• Is the variance constant with temperature?
• What are the sources of the Variance?
• If I clean up the data what are the side effects?
Asymptotic Bend
18 © 2015 Eaton. All rights reserved..
Agenda
• The Problems
• Value Assessment
• Statistical Analysis Refresh
• Data Analysis
• PacifiCorp – Forecasting and Program Design
19 © 2015 Eaton. All rights reserved..
• Residential and Small Commercial
• ~100,000 program participants
• Single & Multi Family Residential
• Small Commercial
• Operational for last 10 years
•Operating Parameters
• June – August
• 14:00 – 20:00 Control Hour Window
• Non-Holiday Weekdays
• Western Electricity Coordinating Council
•Climate
• High Altitude Desert
• Large Daily Temperature Swings
• Low Humidity During Control Hours
Rocky Mountain Power Cool Keeper AC Program
20 © 2015 Eaton. All rights reserved..
Customer Segmentation
• What are the key customer segments and how
do they differ?
• Are their subpopulations within a segment that
affect my program?
• What is the normal customer usage pattern?
What do we want to understand about our customers?
21 © 2015 Eaton. All rights reserved..
Segment Comparisons – Average Daily Usage
Peak Time (Hour) Peak Value (% Run) KWh (Control Period)
Start End High Average Max Average
Single Family 17 20 81% 68% 17.97 14.76
Multi-Family 18 21 55% 45% 7.58 6.02
Commercial 16 19 60% 50% 12.51 10.21
22 © 2015 Eaton. All rights reserved..
Single Family Runtime – August Data
Peak Time (Hour) Peak Value (% Run) KWh (Control Period)
Start End High Average Max Average
Single Family 17 20 81% 68% 17.97 14.76
23 © 2015 Eaton. All rights reserved..
Multi-Family Runtime - August Data
Peak Time (Hour) Peak Value (% Run) KWh (Control Period)
Start End High Average Max Average
Multi-Family 18 21 55% 45% 7.58 6.02
24 © 2015 Eaton. All rights reserved..
Commercial Runtime - August Data
Peak Time (Hour) Peak Value (% Run) KWh (Control Period)
Start End High Average Max Average
Commercial 16 19 60% 50% 12.51 10.21
25 © 2015 Eaton. All rights reserved..
Questions from stakeholders
• Was the event successful?
• How many? How much?
• Were our customers inconvenienced?
• Are we controlling at the right level?
• What is the snapback?
• How did it effect my peak?
• Is there an energy efficiency effect? How Much?
What do your stakeholders want to know about a DR event?
26 © 2015 Eaton. All rights reserved..
PacifiCorp - 14 August Event Analysis
• Event Statistics
• Started: 14:30 Ended: 17:30
• Run & Shed • Logged in 15:00- 18:00
• 16:00 & 17:00 have full hour shed times
• Analysis Statistics
• Participation – 96,047 • Single Family - 65,764 participated
• Multi-Family – 23,017 participated
• Commercial – 7,266 participated
• Results – 81.4 MWs estimated • Single Family – 67.5
• Multi-Family – 9.6
• Commercial – 4.6
• Recovery
• Segments back to normal run % by 22:00 • Note: Multi-Family Usage increases until morning hours
Devices Control
No
Control % Control
Not
Available
Total 103,383 96,047 2,755 97% 4,581
Single Family 70,783 65,764 1,780 97% 3,239
Multi-Family 24,670 23,017 827 97% 826
Commercial 7,930 7,266 148 98% 516
27 © 2015 Eaton. All rights reserved..
Segment Comparisons – Event Day
Runtime Analysis 14-Aug-15
Hour
Single
Family
Multi-
Family Commercial
13 56% 27% 43%
14 51% 32% 46%
15 39% 30% 41%
16 42% 25% 31%
17 56% 28% 33%
18 78% 38% 41%
19 71% 52% 53%
20 61% 48% 47%
21 53% 47% 41%
22 45% 44% 39%
23 44% 42% 36%
Event Start 14:30
Event End 17:30
28 © 2015 Eaton. All rights reserved..
Single Family – Event Analysis Baseline Day Load
24 Aug, 27
Jun, 1,2,3
Jul
27 Jun,
1,2,3 Jul
27 Jun, 1,3
Jul
12 (672) 6,972 6,549
13 511 8,322 7,901
14 (2,350) 4,184 4,394
15 23,426 28,728 29,068
16 61,531 65,302 66,358
17 64,702 67,459 69,041
18 37,808 40,404 42,590
19 (14,408) (11,428) (9,917)
20 (7,237) (4,227) (2,176)
21 (2,588) 1,674 3,985
22 (3,431) 718 3,074
23 (4,171) 1,085 3,319
Per Unit Yield
Hour
24 Aug, 27
Jun, 1,2,3
Jul
27 Jun,
1,2,3 Jul
27 Jun, 1,3
Jul
12 (0.01) 0.11 0.10
13 0.01 0.13 0.12
14 (0.04) 0.06 0.07
15 0.36 0.44 0.44
16 0.94 0.99 1.01
17 0.98 1.03 1.05
18 0.57 0.61 0.65
19 (0.22) (0.17) (0.15)
20 (0.11) (0.06) (0.03)
21 (0.04) 0.03 0.06
22 (0.05) 0.01 0.05
23 (0.06) 0.02 0.05
% Error Hour
-16% 12
-8% 13
-4% 14
-3% 15
0% 16
0% 17
0% 18
1% 19
3% 20
6% 21
13% 22
19% 23
29 © 2015 Eaton. All rights reserved..
Single Family – Snapback Scaled to Event Day
Event Snapback Summary 14-Aug-15
MWh
Duration
(Hours)
Event Load Reduction 132.8 4
Event Snap Back (125.6) 7
Difference 7.3
Snapback % 95%
kWh per Home 0.11
Event Snapback - Scaled to Event Day
Event Day
Scaled
Baseline
14-Aug-15 2-Jul-15
Hour
of Day
Event
Hour MWh MWh Yield State
12 1 95.1 94.4 (0.7) ~
13 2 110.2 109.4 (0.8) ~
14 3 129.1 121.1 (8.0)
15 4 117.2 132.4 15.1 -
16 5 89.3 138.3 49.0 -
17 6 96.1 145.0 48.9 -
18 7 128.0 147.8 19.8 -
19 8 179.5 149.4 (30.2) +
20 9 163.1 139.4 (23.6) +
21 10 139.8 122.9 (16.9) +
22 11 120.9 104.6 (16.3) +
23 12 102.7 88.7 (14.0) +
0 13 83.3 70.8 (12.5) +
1 14 66.2 54.2 (12.0) +
2 15 49.7 41.0 (8.7)
3 16 39.5 31.5 (8.0)
4 17 31.2 25.4 (5.8)
5 18 24.3 20.8 (3.5)
Event Load Reduction (-)
Event Load Reduciton (+)
Evant Scale Period (~)
30 © 2015 Eaton. All rights reserved..
Agenda
• Statistical Analysis Refresh
• Value Assessment
• Data Analysis
• Final Statements
31 © 2015 Eaton. All rights reserved..
DR Forecasting Signal or Noise?
32 © 2015 Eaton. All rights reserved..
Cool Keeper Program Segment Comparison
30-Jun Average AC Runtime
Hour Church
Single
Family
Multi-
Family
Small
Comm
12 4.7% 50.9% 32.8% 47.2%
13 4.2% 56.7% 35.3% 51.0%
14 3.7% 62.3% 38.3% 53.2%
15 4.5% 67.2% 40.7% 56.8%
16 2.9% 72.1% 42.6% 59.4%
17 3.2% 75.2% 47.4% 59.7%
18 3.8% 79.2% 49.8% 58.0%
19 5.9% 78.2% 52.6% 55.5%
20 14.0% 74.4% 52.7% 51.4%
21 10.3% 65.9% 50.6% 47.6%
22 8.6% 60.0% 49.4% 43.1%
23 6.0% 52.0% 48.3% 40.1%
Cool Keeper Program Hours
33 © 2015 Eaton. All rights reserved..