1Suryanarayana Doolla, IIT Bombay
INTRODUCTION TO DEMAND RESPONSE
8.12.2016
PROF. SURYANARAYANA DOOLLA
DEPARTMENT OF ENERGY SCIENCE AND ENGINEERING
2Suryanarayana Doolla, IIT Bombay
Reduction of utility load primarily during periods of peak demand
Reduction of utility loads, more or less equally, during all or most hours of the day
Increase of utility loads, more or less equally, during all or most hours of the day
Improvement of system load factor by building load in off-peak periods
Options to alter customer energy consumption on an as-needed basis
Reduction of loads during periods of peak demand with building load in off-peak time
Load Shape Objectives
3Suryanarayana Doolla, IIT Bombay
• Great achievement in generation and transmission but
still
– Concerns on reliability of power due to outages
– Cost of power to utility becomes very high for a short time due to
• Transmission Constraints
• Generator failure
• Demand surge due to special events such as festivals or state
elections
• Sudden heat waves etc.
• Besides capacity additions are cost intensive as well as
have longer gestation periods.
• Short-time peak management is a challenge for the
Discoms.
Demand Response – What is the Need
Demand response is a solution which can solve these issues plus help in integrating renewables especially wind
4Suryanarayana Doolla, IIT Bombay
DSM & Demand Response
• DSM consists of two major components: Demand Response (DR)[part of load management] and Energy Efficiency (EE), which is alsoconsidered as conservation
• Definition of Demand response : Changes in electric usage byend‐use customers from their normal consumption patterns inresponse to changes in the price of electricity over time, or toincentive payments designed to induce lower electricity use at timeof high wholesale market prices or when system reliability isjeopardized
(Definition as per FERC– USA)
5Suryanarayana Doolla, IIT Bombay
Demand Response
o Electricity Consumers can
o Voluntarily trim their electricity use during peak period
o Choose to use electricity at a less congested time
o Electricity Utilities provide
o Incentives to reduce consumption during peak periods
(residential, commercial, industrial etc)
o Initiating Demand Response:
o Manual
o Automatic
o Semi-Automatic
5
6Suryanarayana Doolla, IIT Bombay
DR is a very complex option
Technology Intensive
Regulatory Intervention
Close Monitoring
Confidence
Understanding Demand Response
7Suryanarayana Doolla, IIT Bombay
Types of Demand Response
DR Programs
Price Responsive DR
Time of Use
Real time
Pricing
Critical Power Pricing
Incentive Based DR/DispatchbleDR
Reliability DR
DLC I / CAncillary market
Emergency DR
Economic DR
Demand bidding/Bu
y Back
8Suryanarayana Doolla, IIT Bombay
A promising way to manage flexible demand is to communicate dynamic prices to the consumers and incentivize them to change their consumption amount or to shift load demand in time.
Various types of Demand Response Programs•Price based DR Program
─Time of the day tariff, Real time pricing, Critical peak pricing
• Incentive based DR Mechanisms
─Direct Load Control, Interruptible/Curtailable DR, Demand Bidding, Emergency DR services, Capacity Bidding and Ancillary DR services
Understanding Demand Response
9Suryanarayana Doolla, IIT Bombay
Framework for Technology Assessment of DR
• Identify the stakeholders involved and their need-cum-objective in the involvement.
─Three broad divisions – Utilities, Load Aggregators and Customer
─Demand response programs require a certain degree of technological readiness from the perspective of each of the stakeholders
• Technology enable DR Program for a target load category
─ The technological requirement identified can be categorized as hardware component, software component and the communication technology that enables exchange of information across various entities involved.
─ Technological Prerequisite for smooth conduct of a residential DR Program (Cycle explained)
Components of Demand Response
10Suryanarayana Doolla, IIT Bombay
Utility
• Triggers a DR event through various price based and incentive based mechanisms
• Payment to Customers
Customer
Load aggregator
• Small consumers may not qualify
• Penalty associated
Regulator
Stakeholders in Demand Response
11Suryanarayana Doolla, IIT Bombay
Infrastructure/Architecture for Demand Response
12Suryanarayana Doolla, IIT Bombay
Technology Options – DR in Residential/Small
Consumers
13Suryanarayana Doolla, IIT Bombay
Architecture for Residential DR Program
14Suryanarayana Doolla, IIT Bombay
Network requirements for DR Program
Latency for DR programs represents the time interval between trigger and response. This varies between 500ms to several minutes as per the type of DR programs.
Ancillary and emergency DR programs usually have low latency requirements.
Available Communication Options for DR
Network attributes Demand Response
Latency 500ms to several minutes
Bandwidth 14 to 100 Kbps per node
Minimum reliability 99 %
Security High
Power Back up Not required
15Suryanarayana Doolla, IIT Bombay
Coverage range and data rate specifications for various networks used in DR
Available Communication Options for DR
16Suryanarayana Doolla, IIT Bombay
Technology Review for Wireless (ZigBee, Wi-Fi, WiMaX, RF), Wired (PLC, Ethernet, (DSL) Optical fibre), Cellular (GSM, GPRS, 3G/4G (LTE))
Comprehensive comparison of various communication technologies (Wireless, Wired and Cellular) for DR
• Data Transfer
• Cost
• Power Consumption
• Existing Support Structure
• Security
• Application
Available Communication Options for DR
17Suryanarayana Doolla, IIT Bombay
18Suryanarayana Doolla, IIT Bombay
DR Programs – Level of Interaction with Grid, Control
and Telemetry Speed requirements
SpiningReserve (Fast DR)
Real-time DR
Day ahead DR (Slow)
Daily peak load management
ToU Tariff
Daily EE measures
Increased interaction with the Grid
Increased level of control and speed
19Suryanarayana Doolla, IIT Bombay
Open Automated Demand Response (OpenADR) protocols provide an open (nonproprietary), standardized and secure way for Utilities and Independent system operators (ISOs) to communicate DR signals with each other and with their customers using a common language over any existing IP-based communications network, such as the Internet.
OpenADR 1.0 standards were designed for simpler devices such as residential thermostats. Many features have been added in new versions of OpenADR 2.0 such as handling the flow of information back from the buildings.
Open ADR Standards
20Suryanarayana Doolla, IIT Bombay
This reporting feature has been made more robust and can handle real-time results from end users.
Real-time energy consumption changes in building loads can be confirmed with same ease as verification of a home or small business thermostat or load-control device having received the DR signal and if they have responded as per the contract.
Pacific Gas & Electric (PG&E), San Diego Gas & Electric and Southern California Edison (SCE) have added and recommended OpenADR 2.0 certified products to implement their dispatch of emergency and price DR resources.
Open ADR Standards
21Suryanarayana Doolla, IIT Bombay
Case Study – TPDCL - Delhi
o Customers enrolled: 144/173 (connected load > 100 kW)
o Total event period: 6 Months
o Event interval: 30 min to 60 min
o No. of Events: 17
o Achieved about 8% of demand reduction in the pilot project
o 5/10 with morning adjustment is used as baseline
21
22Suryanarayana Doolla, IIT Bombay
Tasks of DRAS
o Task 1
oCollects information from Load forecasting tool, SCADA, and MDMS (Meter Data Management Systems)
o Task 2
oSupply-Demand is determined for present and next block.
o Task 3
oBased on the deficit/surplus situation, commands are sent to the consumers.
22
Source: TPDDL Delhi
23Suryanarayana Doolla, IIT Bombay
DR Deployment – Utility Consumer
23
Source: TPDDL Delhi
24Suryanarayana Doolla, IIT Bombay
Sub-Systems Connectivity – Auto DR
24
Source: TPDDL Delhi
25Suryanarayana Doolla, IIT Bombay
Measurement and Verification
for DR
26Suryanarayana Doolla, IIT Bombay
Sequence of Events in DSM Cycle
Load Research & Market Potential
Assessment
Strategic Planning
Program DesignProgram
Implementation
Program Evaluation & Measurement
27Suryanarayana Doolla, IIT Bombay
Aim of M&V - evaluating the actual amount of load that is curtailed during an event.
Requirement - Estimate of the load that the customer is going to consume on the event day more precisely during the event period.
Baseline- Predicted amount of the electricity that would have been used up by a customer in the absence of a DR event.
Consumers - Compensated financially for reducing electricity use•Difference between the customer baseline (CBL) and the actual metered usage.
Baseline is challenging aspect as it should represent the exact usage of electricity by the customers in the absence of load curtailment during DR event
Importance of Baseline
28Suryanarayana Doolla, IIT Bombay
Components of Baseline
Data Selection
Estimation
Adjustment
29Suryanarayana Doolla, IIT Bombay
Baseline window is identified
• Precedes a DR event period over which electricity usage data is collected for the purpose of establishing a baseline.
Some of the instances of baseline windows include:
• Last 10 non-holiday weekdays;
• 10 most recent program-eligible non-event days;
• 10 most recent program-eligible days beginning 2 days before the event;
• Last 45 calendar days
• Previous year
Data Selection
30Suryanarayana Doolla, IIT Bombay
All the above baseline windows exclude some of the days from the from the baseline window so as to neutralize the effect of certain exceptionaldays/events on the estimation of baseline.
Some of the common exclusion rules include:
• Excluding days with DR events.
• Excluding days with outages.
• Excluding days with extreme weather.
• Excluding days with the highest or lowest loads.
Data Selection
31Suryanarayana Doolla, IIT Bombay
Baseline Estimation Methodology -
Example
Select 10 most recent non-event weekday Event days are non-denied settlement days
Exclude the following day-types: National and State holidays on which industry is closed
Weekend Days (Saturdays and Sundays)
Strikes or any other special event
Demand Response event days
Replace excluded days with next valid day
Final Weekday CBL Basis Window should contain 10 days
Actual Load data for these 10 days to be used to calculate CBL of consumer
Sunday Monday Tuesday Wednesday Thursday Friday Saturday
1 (D-10) 2 (D-9) 3 (D-8) 4 (D-7) 5 6
7 8 9 (D-6) 10 11 (D-5) 12 13
14 15 (D-4) 16 (D-3) 17 18 19 20
21 22 (D-2) 23 24 25 26 (D-1) 27
28 29 30
Previous Event Days Base Line Days Current Event Day Weekends
32Suryanarayana Doolla, IIT Bombay
Data Selection – Exception Days
50006000700080009000
10000110001200013000
00:0
0
00:5
6
01:5
2
02:4
8
03:4
4
04:4
0
05:3
6
06:3
2
07:2
8
08:2
4
09:2
0
10:1
6
11:1
2
12:0
8
13:0
4
14:0
0
14:5
6
15:5
2
16:4
8
17:4
4
18:4
0
19:3
6
20:3
2
21:2
8
22:2
4
23:2
0
MW
Hours
Exception Day : 14th Jan 2015 ( Kite festival in Gujrat, Makar Sankranti)
GEB Dmd 13-01-15 GEB Dmd 14-01-15 GEB Dmd 15-01-15
26000
28000
30000
32000
34000
36000
38000
40000
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
MW
Hours
WR - Load Curve Holi Day 06.03.2015
4TH MARCH 15 5TH MARCH 15
6TH MARCH 15 7TH MARCH 15
3200034000360003800040000420004400046000
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
MW
Hours
Diwali Effect on WR Demand met: 23rd Oct 2014
20-10-2014 21-10-2014 22-10-2014
23-10-2014 24-10-2014 25-10-2014
260002800030000320003400036000380004000042000
00:0
0
00:5
6
01:5
2
02:4
8
03:4
4
04:4
0
05:3
6
06:3
2
07:2
8
08:2
4
09:2
0
10:1
6
11:1
2
12:0
8
13:0
4
14:0
0
14:5
6
15:5
2
16:4
8
17:4
4
18:4
0
19:3
6
20:3
2
21:2
8
22:2
4
23:2
0
MW
Hours
WR Demand on 25th, 26th and 27th Jan 2015
WRDMD 25-01-15 WRDMD 26-01-15 WRDMD 27-01-15
33Suryanarayana Doolla, IIT Bombay
Methods adapted
• Averaging method calculates for each hour/half-hour of the day, the average of the load at that hour over the included days.
• Regression method calculates load using linear regression model from the included days on weather and other variables, usually with separate regression coefficients by hour of the day.
• Maximum value method takes the maximum of the loads in the included period.
• Rolling average method uses the updated unadjusted baseline for an operating day which is equal to 0.9 times the prior unadjusted baseline plus 0.1 times the most recent included day.
Estimation
34Suryanarayana Doolla, IIT Bombay
In averaging or “X of Y” or “X in Y” method, the baseline for each interval of curtailment day is calculated as the simple average, across all the days chosen by the data selection criteria. The most common “X in Y” baselines include:
10 in 10 - Out of 10 selected days, highest 10 day’s data is taken and the baseline is calculated.
7 in 10 - Out of 10 selected days, highest 7 day’s data is taken and the baseline is calculated.
5 in 10 - Out of 10 selected days, highest 5 day’s data is taken and the baseline is calculated.
3 in 10 - Out of 10 selected days, highest 3 day’s data is taken and the baseline is calculated.
Estimation – Averaging Method
35Suryanarayana Doolla, IIT Bombay
Sample – Utilities in US
ISOBaseline
typeAverage of Out of
CAISO,
MISO10 in 10
10 most recent
weekdays
10 most recent
weekdays
ERCOT Mid 8 of 10
10 most recent
weekdays,
dropping highest
and lowest kWh
days
10 most recent
weekdays
NYISO 5-of-105 highest kWh
days
10 most recent
weekdays
PJM 4-of-5 4 highest kWh
days
5 most recent
weekdays
ISO Baseline
type
Average of Out Of
CAISO,
MISO
10-in-10 4 most recent
weekend days
4 most recent
weekend days
ERCOT Mid 8-of-10 10 most recent
weekend days,
dropping highest
and lowest kWh
days
10 most
recent
weekend days
NYISO 5-of-10 2 highest kWh
weekend days
3 most recent
weekend days
PJM 4-of-5 2 highest kWh
weekend days
3 most recent
weekend days
Weekdays
Weekends
ERCOT’s (Electric Reliability Council of Texas)
NYISO’s (New York ISO)
CAISO (California ISO)
MISO (Midcontinent ISO)
Jaipur Vidyut Vitran Nigam Limited
(JVVNL), for its DR project which was from
July 2013 March 2014, used “5 in 10”
method for estimation of baseline.
36Suryanarayana Doolla, IIT Bombay
Regression model consisting of daily energy equation
• Include consumer’s daily consumption (kWh) as dependent variable
• 24 hourly energy fraction equations.
The explanatory variables in model include calendar variables
• Day of the week, holidays
• Weather variables
• Daylight variable
─ Daylight saving time
─ Time of sunrise and sunset
Estimation - Regression Method
37Suryanarayana Doolla, IIT Bombay
The average method is not weather sensitive and does not depends on occupancy level, due to which it requires adjustments in baseline
Adjustment factor is evaluated to align the baseline with observed conditions of the event day.
• Temperature, humidity; calendar data, Sunrise/Sunset time and event day operating conditions.
Adjustment factors are either additive or scalar (multiplicative).
• Additive - Shifts the curve up or down by a constant amount
• Scalar - Scales the shape by a constant amount
Adjustment
38Suryanarayana Doolla, IIT Bombay
TPDDL - Auto-DR project
•5 out of 10 baseline model with morning adjustment baseline.
In “5 in 10” method, the highest 5 days load out of previous 10 days was averaged excluding the weekend, holidays.
• Morning adjustment baseline, additive factor was used.
5 in 10 with morning adjustment was found to be more close to the actual load curve.
Adjustment – Indian Experience
39Suryanarayana Doolla, IIT Bombay
Energy consumption data of various consumer categories in Mumbai was used to evaluate various unadjusted baselines through averaging method.
These baselines were plotted along with one of the actual days from the data sample which was not used for baseline estimation in order to investigate the performance of these baselines in predicting the load on a DR event day.
Comparison of various Baseline
Techniques- Unadjusted; Mumbai Data
40Suryanarayana Doolla, IIT Bombay
Residential Colony
0
200
400
600
800
1000
1200
1400
1600
Lo
ad
(k
Wh
)
Hours
Load Curve for Residential Colony
(Sep - Oct 2012)
ACTUAL 10 IN 10 7 IN 10 5 IN 10 3 IN 10
41Suryanarayana Doolla, IIT Bombay
Hospital
100
150
200
250
300
350
400
450
Lo
ad
(k
Wh
)
Hours
Load Curve for Hospital (Sep- Oct 2012)
ACTUAL 10 IN 10 7 IN 10 5 IN 10 3 IN 10
42Suryanarayana Doolla, IIT Bombay
Hotel
600
800
1000
1200
1400
1600
kW
h
Load Curve for Hotel - A (Sep - Oct 2012)
Actual 10 in 10 7 in 10 5 in 10 3 in 10
43Suryanarayana Doolla, IIT Bombay
Similar analysis carried out for other sectors• IT Sector
• Commercial establishments
• Apartments
• Manufacturing Setups
• Shopping Centre
The analysis reveals that there cannot be one baseline that is a best fit for all customers. Although, “10 in 10” baseline shows best values for accuracy, bias and variability metrics for most of the customer categories, it needs further analysis before declaring these values conclusive as the effects of temperature and other factors have not been taken into account
Analysis for other sectors
44Suryanarayana Doolla, IIT Bombay
Barriers• Cost (varies depending on extent of EM&V)
• Available Personnel
• Access to energy data
• Understanding and organizing energy data: kWh vs. kW Demand;
• Rate structures;
• Meters and corresponding equipment/facilities
EM&V approaches/methods range from simple and direct to complex and indirect• Sometimes combined
• More complex methods generally require more detailed data and higher cost
Summary
45Suryanarayana Doolla, IIT Bombay
Reports are available in http://dsm-india.org
Further Reading
46Suryanarayana Doolla, IIT Bombay
Thank You!