fast,&scalable&demand0shaping&with& colorpower · 2019. 12. 23. · the demand...

Fast, Scalable Demand-‐Shaping with ColorPower Jacob Beal

January, 2013

ColorPower in a Nutshell:

Time

Pow

er (M

W)

10am noon 2pm 4pm0

50

100

150

200

Energy service =ers…

… allow scalable distributed demand management… … bringing disrup=ve

change to the energy industry.

Tradi0onal U0lity

Customer

Demand Aggregator

Ver0cal Resale U0lity

Retail Services

Talk Outline

•  Energy Flexibility Tiers •  The ColorPower algorithm •  Implications and Impact

The Demand Challenge

•  Future retail utilities must be able to effectively manage demand fulfillment at the margin –  Defer consumption –  Pull forward consumption –  Cancel uneconomic consumption (in customers’ best

interest!)

•  Problem: inadequate coordination between the grid and end user devices

Economists vs. Customers

Microeconomics View •  Customers can be

modeled as rational marginal demand functions for a commodity

•  Customers can be modeled as virtual power plants

•  Customers need to be sent price signals to modify their behavior

Customer View •  I do not have a marginal

demand for power, I want reliable service

•  I am not a virtual power plant

•  I don’t want price volatility risk or to do laundry at midnight

Retail Power Is a Service

•  Not a hot concert ticket •  Not a basket of commodity electrons

•  Customers prefer to buy power as a service, not a commodity –  Just like many other service industries

The service sector has largely abandoned conges5on pricing as a way to manage demand peaking

Paradigms for Peak Demand Control

Industry Process / Technology

Various (e.g. manufacturing) “First Come First Serve” / Backlog Queue

Hospitals “Most Urgent First” / Wai0ng Room

Road Transporta0on “Alterna0ng Access” / Traffic Light

General Digital Communica0ons “Best Effort Transport” / TCP/IP

Cellular Automa0c Protocol / CDMA

Cable Networking Automa0c Protocol / DOCSIS

DSL Automa0c Protocol / ATM

Electric Power -‐ Today Free-‐for-‐All / Circuit Breaker

Electric Power – Future Consensus Real-‐0me Auc0on / Smart Grid

HOW MUCH IS THIS POWER WORTH TO YOU NOW?

HOW ABOUT NOW?

HOW ABOUT NOW?

SORRY YOU WERE JUST OUTBID BY R1CH_P0WRHAWG

YOUR POWER HAS MOVED TO A PLACE WHERE IT IS MORE APPRECIATED

[POWER OFF]

THE GOOD NEWS IS YOU HAVE NOW SAVED 26 OPEN SODA CANS WORTH OF CO2 EMISSIONS

Technology As Auction Enabler

FrigidTrader 3000 •  Aggressive risk/reward tradeoff! •  Detects and avoids hedge-‐fund price-‐manipula0on strategies! •  Free trading strategy upgrades! •  Icemaker upgrade available

Energy efficiency or clever trading?

WE’LL PAY YOU TO TAKE THESE PEANUTS AWAY

PEANUTS ARE VERY EXPENSIVE RIGHT NOW

The Smart Grid

The Smart Grid

Price Signal Smart Appliance/EV Stampedes

GOOD FOR SPECULATORS BAD FOR CONGESTION CONTROL

CATASTROPHIC FOR SYSTEM RELIABILITY

WIDE-‐SCALE PRICE VOLATILITY:

Retail Price Volatility: Be Careful What You Wish For

Core problem: price is overloaded

•  Too many stakeholders, not enough leverage: –  Customer preferences –  Customer impact –  Procurement of supply –  Operations reliability –  Regulatory policy –  Shareholders and financial traders

Alternative: less information

ColorPower: Tiered Energy Priority

Humans: Always In Control Privacy Respected

ColorPower Appliance Priori=es 1.  Obey Your Humans’ Preferences 2.  Donate Flexibility to Power Grid

Smart Grid: Coordinates Orderly Power

Access For Flexible Appliances & Machines Invisible to Humans

Flexible

Emergency

Not Flexible

✔ ✔

✔ ✔

“Shed 12 MW of Flexible”

Self-Identification of Demand Flexibility

Leaf Bukon: On = Flexible

Green = Price Sensi0ve Yellow = Reliability Responsive Red = Opt Out

Cloud Somware With More Op0ons

Turn Off

Resource Tier 500

Resource Tier 499

Resource Tier 1

Resource Tier 5

Resource Tier -‐5

Resource Tier -‐1

Resource Tier -‐499

Resource Tier -‐500

Priority Tiers More Demand

Less Demand

No Feedback

Pool Pump

Run Now

Charge

Discharge Bakery

Generator On

PHEV

Stay under 6kW

Facility Management

-‐3 degrees

Thermostat

-‐6 degrees

Solar Export Power

From Flexibility to Priority Tiers

Default Tier Assignment Via Color/Device Type

Tiered Aggregations

HVACs – Green

Service Levels •  Max Curtailment: 60% •  Min Cycle Power Timeslice: 10 min

•  Max Cycle Down0me: 20 min

•  Max Ramp Rate: 200kW/min

Pool Pumps -‐ Green

HVACs -‐ Green

Misc. Flex -‐ Green

Pool Pumps -‐ Yellow

HVACs -‐ Yellow

Misc. Flex -‐ Yellow

Bakery Storage

Night Run Dishwashers

Emergency Ra0oning

Sheddable Load Resources

Dispatchable Load Resources

Emergency Dispatchable Load

No Inconvenience

Emergency Storage

Emergency Load Dumping

Major Inconvenience

Major Inconvenience

Outline

•  Energy Flexibility Tiers •  The ColorPower algorithm

–  Control Architecture –  ColorPower Algorithm –  Validation in Simulation

•  Implications and Impact

[Beal, Berliner, & Hunter, SASO’12; Ranade & Beal, SASO’10]

•  Groups and Individual Devices Act Randomly, but Precisely in Aggregate

•  Feedback loop recruits resources until demand target satisfied

Demand Mgmt. Server

ADAPTIVE FEEDBACK LOOP

Groups and Individual Devices

Local Probabilis0c Coopera0on Calcula0on

SYSTEM STATE: NEED -‐12 MW TOTAL DEMAND RELIEF

DEMAND REPORT: ON: 203 MW OFF: 342 MW SHED: 127 MW NOSHED: 432 MW

Demand Shaping Target

Basic ColorPower Architecture

ColorPower State Transitions

•  (E)nabled vs. (D)isabled •  (R)efractory vs. (F)lexible

!"#!$#

%"# %$#

&'(#&')#

*+&')#

*+&'(#

,!$-./01,!23#

,%$-./01,%23#

Formal Control Problem

!"#"$%&#'()")*'(

+*'"&*,,$&

-."/#0121.3*44/5/#

/13*'/5/#

/1

11111111111113*44/6/#

/13*'/6/#1

7770

8."0

!"#$%&'()*+#

!"#"$%&#'()")*'(

!"#"$%&#'()")*'(

!"#"$%&#'()")*'(

9."0121.:;<:5/1:=>:

5/17771011

(."/#0121.:;<:5/#/1:=>:

5/#/1:=<:

5/#/1:;>:

5/#/

1111:;<:6/#/1:=>:

6/#/1?01111111

@ @

+*'"&*,,$&+*'"&*,,$&+*'"&*,,$&

A&*#B-#("

!"#"$;(")C#"*&

A&*#B-#("

D88&$8#"$For each ColorPower client, set pon , poff for each device group, such that the total enabled power in s(t) tracks g(t)

Outline


–  Control Architecture –  ColorPower Algorithm –  Validation in Simulation


ColorPower 2.0 Algorithm: Intuition

•  Flexibility accumulates as Refractory devices finish their timeouts.

•  Allocate flexibility budget in order of importance: 1.  Goal tracking & hard priority constraints 2.  Soft priority 3.  Cycling devices to ensure fairness 4.  Maintaining flexibility reserves for future needs

Constraints

•  Goal tracking: shape power demand

•  Color priority: respect user preferences

•  Fairness: no devices are favored

•  Cycling: don’t keep the same devices off

ColorPower 2.0 Algorithm: Equations

Alloca=on 1: Goal Tracking

Alloca=on 2: Color Priority

Alloca=on 3: Cycling

Compu=ng pon and poff

Upward is converse

Downward shim:

Correc0on Goal:

Upward is converse

Downward shim:

Boundary color b:

and similar for other states

Reserve frac0on f:

when enabled domainates, else converse

Analysis

•  Convergence: –  Sufficient Flexibility: –  Insufficient Flexibility:

•  Quiescence: (conservative)

•  Ramp Tolerance:

!"#

$!!"#

%&'()*

+#,-

(*.-#!"#$%&"#'

!# /0%123/0412#05'67-8#/2.-'

/ 0%129:/ 0

%123/ 0

4129;

<#

/0%12#

Quadra0c:

Linear:

Total:

Priority constraint:

Outline


–  Control Architecture –  ColorPower Algorithm –  Analysis & Simulation


Control Example: Hot Summer Day

Time

Pow

er (M

W)

10am noon 2pm 4pm0

50

100

150

200

Flex

Emergency

Inflexible

Control Example: Emergency Response

Time (seconds)

Pow

er (M

W)

0 500 1000 1500 2000 2500 30000

2

4

6

8

10Flex

Emergency

Inflexible

Simulation Base Configuration

•  10,000 clients, each controlling a 1KW device •  Coloring: 20% green, 50% yellow, 30% red •  Measurement error: 0.1% •  10 second rounds •  Refractory time: U[400,800] seconds •  Flexible reserve ratio: 1:1 •  Attempted correction/round: 80%

Simulation Studies: Convergence

Predicted

Post step Goal (MW)

Pre

step

Goa

l (M

W)

Predicted step response convergence time

0 2 4 6 8 10

0

1

2

3

4

5

6

7

8

9

10 0

100

200

300

400

500

600

700

800

900

Measured

Post step Goal (MW)

Pre

step

Goa

l (M

W)

Step response convergence time

0 2 4 6 8 10

0

1

2

3

4

5

6

7

8

9

10 0

100

200

300

400

500

600

700

800

900

Excellent agreement

Simulation Studies: Quiescence

Predicted

Post step Goal (MW)

Pre

step

Goa

l (M

W)

Predicted step response quiescence time

0 2 4 6 8 10

0

1

2

3

4

5

6

7

8

9

10 0

500

1000

1500

2000

2500

3000

3500

Measured

Post step Goal (MW)

Pre

step

Goa

l (M

W)

Step response quiescence time

0 2 4 6 8 10

0

1

2

3

4

5

6

7

8

9

10 0

500

1000

1500

2000

2500

3000

3500

Much beOer than conserva5ve es5mate

Simulation Studies: Scaling

102 103 104 105 10610 5

10 4

10 3

10 2

10 1

Number of Devices

Mea

n C

onve

rged

Erro

r

Goal Tracking Case 1Steady State Case 1Case 2Case 3Case 4

More devices = beOer accuracy

Simulation Studies: Heterogeny & Error

Heterogeneous Devices

100 101 102

10 4

10 3

Device Size Ratio

Mea

n C

onve

rged

Erro

r

Case 1 50/50 Pop.Case 1 Different Pop.Case 2Case 3Case 4

Estimation Error

10 5 10 4 10 3 10 2 10 1

10 4

10 3

10 2

10 1

Measurement Error ( )

Mea

n C

onve

rged

Erro

r

Goal Tracking Case 1Steady State Case 1Case 2Case 3Case 4

Robust to error and differences in devices

Outline

•  Energy Flexibility Tiers •  The ColorPower algorithm •  Implications and Impact

Locational x Program Control

ISO U0lity Substa0on Tiered Aggrega0ons

Individual Load

Popula0ons


HVACs -‐ Green



HVACs -‐ Yellow


Bakery Storage


Emergency Ra0oning


Emergency Storage



HVACs -‐ Green



HVACs -‐ Yellow


Bakery Storage


Emergency Ra0oning


Emergency Storage



HVACs -‐ Green



HVACs -‐ Yellow


Bakery Storage


Emergency Ra0oning


Emergency Storage



HVACs -‐ Green



HVACs -‐ Yellow


Bakery Storage


Emergency Ra0oning


Emergency Storage


Loca0onal Programs Par0cipants

Price signals to consumers from markets? ColorPower can send demand signals from consumers to markets.

Demand Signals To Markets

Disruptive Business Models

•  Value-added services for appliances: –  ColorPower = 21st Century EnergyStar? –  Deployment as a side-effect of normal

replacement cycles –  Demand shaping capacity sold directly

or contracted to 3rd party managers

•  Vertical utilities: –  Manufacturer / retailer assumes energy

risk for a small premium –  “Air conditioner with brownout protection

plan” –  “Pool pump with a lifetime energy

supply”

Tradi0onal U0lity

Customer

ColorPower enablers: cheap hardware & networking, ability to bundle at organizaWon boundaries

~$1 cost increment

Demand Aggregator

Ver0cal Resale U0lity

Retail Services

Contributions

•  Energy flexibility tiers allow separation of demand management concerns

•  ColorPower algorithm allows fast, robust, and precise control of thousands to millions of devices.

•  ColorPower is pragmatically deployable, and allows disruptive new energy business models.

Next Steps & Collaboration Opportunities

•  Small-scale deployment (starting shortly) •  Standardization •  Improved integration with other grid systems •  Adaptation of base principles to other grid

control problems?

Acknowledgements

Jeff Berliner Ryan Irwin

39

Steve Florek Vinayak Ranade Kevin Hunter

fast,&scalable&demand0shaping&with& colorpower · 2019. 12. 23. · the demand...

Documents