The Smart Grid: Distributed optimization & control challenges - - PowerPoint PPT Presentation

Kameshwar Poolla UC Berkeley June 24, 2011 HYCON2 – Trento, June 2011

The Smart Grid:

Distributed optimization & control challenges

Objectives

Cover essential background on power systems Tell you about what is changing

• Renewables
• Distribution automation
• Markets
• Communication and sensing

Discuss some opportunities

• Distributed optimization
• Distributed control

Describe our vision of Grid2050

HYCON2 – Trento, June 2011

• A. The Legacy Grid
• 1. Components
• 2. Complications
• 3. Fact & Figures
• 4. Basic Power Systems Engineering
• 5. Basic Power Systems Economics

HYCON2 – Trento, June 2011

Power System Components

Generation system: power source

ideally with fixed voltage and frequency

Load or demand: consumes power

ideally constant and resistive

Transmission system: transmits power

ideally as a perfect conductor

Distribution system: local reticulation of power Control equipment: many functions

coordinate supply with load, regulate voltage and frequency, protection, handle component failures

HYCON2 – Trento, June 2011

A simple power system

generator transmission load

HYCON2 – Trento, June 2011

Complications

Loads vary a lot, not purely resistive

not known in advance day-ahead forecast accuracy 2-4%

Generators have constraints: ramping, capacity Transmission lines are reactive, have capacity

constraints

Everything is connected in a complex network of

heterogeneous elements

Must be robust to component failure Must deliver power economically through markets

HYCON2 – Trento, June 2011

Load variability

Aggregate loads in CA on a hot day in 1999 Peak is 2-6pm [system can be very stressed] Variation is 50% of peak load

HYCON2 – Trento, June 2011

Generation, T&D

HYCON2 – Trento, June 2011

Generators

Limited capacity Contain control systems to regulate

frequency and voltage

Many different types with different costs, cold-start

times, ramp rates, inertia

Coal, gas, nuclear, hydro, wind, solar, Other points

• Stability issues are important for thermal generators
• Start-up times require unit commitment in advance
• Require to be taken off-line for maintenance/repair

HYCON2 – Trento, June 2011

Transmission

Operate at high voltages to reduce resistive line loss

[765, 500, 345, 230 kV]

Mainly 3 phase AC HV DC for long lines and undersea cables T & D losses in 2007 Moving electricity across 1000 Km is cheap

0.005 – 0.02 $/kW hr [not counting capital cost] Key fact: aging transmission infrastructure is causing many problems in the US

HYCON2 – Trento, June 2011

HYCON2 – Trento, June 2011

It’s a Complex Network

Distribution network

• has tree structure
• Serving millions of customers
• Changing topology

Transmission network

• has loops, node degree < 5
• Enables routing of power from multiple generators
• 4000-12000 buses

Network aspect can cause problems

• Cascading failures
• Islanding
• Stability issues

HYCON2 – Trento, June 2011

Operations

HYCON2 – Trento, June 2011

Too complex, so broken into control areas Each area handled by a Balancing Authority/System

Operator

Inter-area power transfer for control area imbalances Balancing functions of System Operator

• Ex ante:

Economic Dispatch Schedule generation to meet forecast demand

• Real time: Monitoring

For system failures, power quality, line congestion

• Ex post:

Reserve management Schedule generation to handle unexpected events

North America Interconnections

HYCON2 – Trento, June 2011

Geographic segmentation for easier management of electricity

Electricity Markets

Intimately connect to power systems engineering Electricity is different from bananas Bilateral Contracts 65% of power is sold in long-term bilateral contracts Bulk-power markets Multiple time-scales: Day-Ahead, Hour-Ahead, Real-Time Markets are partially regulated Price caps, subsidies, extra-market mechanisms Ancillary Services Markets for other stuff Reserves, Inertia,

HYCON2 – Trento, June 2011

Economic Dispatch

HYCON2 – Trento, June 2011

Scheduling generation to meet forecast demand Done by SO in advance through electricity markets

1.

Forecast the demand

2.

Get bid stacks from generators

3.

Select generation to minimize cost Constraints Existing bilateral contracts Transmission line limits Power system security It is a centralized optimization problem Output: schedule of committed generation and prices

Power System Security

Must deliver power even when a component fails

• Worry about 1 failure
• Called [N-1 contingency]

SO buys reserve capacity

• to handle single largest failure
• Margins are 7% in CA, 11% in Texas

Feasibility of power flow places constraints on

economic dispatch

HYCON2 – Trento, June 2011

Robustness & Reserves

Power system security

• Contingency Reserves

Demand forecast errors

• Operating Reserves
• When load rises too sharply to schedule other resources

Keep power quality [frequency 60 ± 0.25 Hz]

• Regulation reserves

Management of reserves is complex, centralized

HYCON2 – Trento, June 2011

Facts & Figures

HYCON2 – Trento, June 2011

US Generation Sources

HYCON2 – Trento, June 2011

Generation Sources in California

Oregon is 57% Hydro, Washington State is 70% Hydro

!"#$%"&'(") *+, !$-&."% /0, 1.2.3"4&.'

5)6&"%7'38297' :.6#;.%<"&=

/>, ?@9%6 //, A6"&'"29'68& >, B#;.% /, C6$%-.D'EC'FGH7'F&.-#%8-' I63.%'H22$"&'>JJ+'K C#"#.' L"#"'M"4&.)

HYCON2 – Trento, June 2011

Generation Sources in Illinois

Coal 55.1% Nuclear 43.6% Renewable 0.0% Petroleum 0.6% Hydroeletric 0.0% Gas 0.6%

HYCON2 – Trento, June 2011

HYCON2 – Trento, June 2011

Demand side facts

Power will be defined carefully later

Installed U.S. generation capacity GW [about 3 kW per person] Average load of UCB about 25 MW Average load of California about 34 GW

Annual U.S. electric energy consumption

MWh per person [on average each American uses 1.5 kW of power]

HYCON2 – Trento, June 2011

Retail Electricity Prices

HYCON2 – Trento, June 2011

AC Power

HYCON2 – Trento, June 2011

60 Hz in US, 50 Hz in US Phasors Complex power absorbed by a load Complex power delivered by a source Conservation of complex power

! " ! "

Power Flow on a Line

HYCON2 – Trento, June 2011

Real power flow controlled by phase difference Reactive power flow controlled by voltage difference

Reactive power

HYCON2 – Trento, June 2011

What is it?

• Power that is borrowed and returned each cycle
• Not consumed in net

If reactive power is supplied remotely, we have

Reactive power compensation

HYCON2 – Trento, June 2011

Best to supply reactive power locally Need to adjust compensation as load varies Can be done at load bus by

• Adjustable shunt capacitance
• Load-tap-changing transformer
• Power electronics devices

Three Phase

HYCON2 – Trento, June 2011

3 conductors instead of 6 for same power transfer Sum of powers across phases is constant

Balanced 3-phase Operation

HYCON2 – Trento, June 2011

Symmetrical loads Symmetrical generation 120o phase shift from one phase to next Allows analysis one-phase-at-a-time Steady-state frequency = 60 Hz Quasi-steady-state frequency 60 Hz Changes slowly, so can still use phasors for analysis Time-varying magnitudes and angles

Generator Models

HYCON2 – Trento, June 2011

Per-phase Includes voltage regulation loop More complex models for transient analysis

[Swing equation]

Still more complex models to analyze field/armature

effects, frequency swings, rotor vibrations, etc ! "

Transmission line Models

HYCON2 – Trento, June 2011

Per-phase Medium to long lines: -model Short lines (< 150 miles) Line loadability Short lines: thermal limits Long lines: stability limits Decent power system model

sparse circuit driven by variable sources serving uncertain loads

Power Quality

HYCON2 – Trento, June 2011

Voltage 1 ± 0.02 per unit Frequency 60 ± 0.25 Hz

Frequency oscillations are an early indicator of system stability Voltage collapse is the result Modest voltage excursions result from poor reactive power support Result is inefficiency

Dispatch

HYCON2 – Trento, June 2011

Selecting generation to meet forecasted demand Centralized

• System operator communicates with generators
• Schedules setpoints

Day Ahead

• Forecast load ± 3%
• Schedule generation in DA markets

Hour Ahead

• Better load forecast ± 0.5 %
• HA is an adjustment market

Real-time Balancing

• Real time is actually 5-10 minute ahead
• Reissue set-points for select generators

Primary Generation Control

HYCON2 – Trento, June 2011

Also called Load-frequency control To account for minute imbalances between supply

and demand

Completely decentralized How it works

• Frequency drop:

add generation

• Frequency rise:

decrease generation Details

• Done by adjusting governor in thermal generation
• Time scale 0.1 seconds

Contingency Operations

HYCON2 – Trento, June 2011

When something goes wrong

• Load ramps up/down unexpectedly
• Line trips, Generator fails
• Centralized

Beyond capacity of primary AGC Call on reserves: Secondary AGC

• Spinning reserves: can come on line in 5-10 minutes for 30

minutes

• Non-spinning reserves: can come on line in 10-15 minutes for

30 minutes Tertiary AGC

• Reschedule contracted generation

Kameshwar Poolla UC Berkeley June 24, 2011

The Smart Grid:

Distributed optimization & control challenges

• B. The New Frontier
• 1. Drivers for Change
• 2. New Components
• 3. New Problems
• 4. The Smart Grid

Drivers for Change

Global climate change

• Reduce Carbon emissions
• Sequestration: Carbon capture and storage

Energy Security

• Reduce dependence on “unreliable” resources
• Wage war on small countries with oil & gas reserves

These concerns drive

• Renewable generation [ex: wind/solar]
• Efficiency [ex: buildings]
• Increased use of electric vehicles

No Silver Bullet

Source: EPRI

The Grid works now but

New renewable generation Utility-scale wind, Rooftop solar, Thermal solar, PV farms New hardware Power electronics, materials, electricity storage New communication & sensing New loads: EVs, demand response New business practices Intraday markets, aggregators New Problems

Renewable Generation

Utility scale: Wind farms, Thermal solar, PV farms Distribution side: Rooftop solar Utility scale wind

• US leads the world with 35 GW capacity
• Growth rate in China is 100% annually
• 38GW installed in 2009 worldwide

Solar

• China is the world leader
• Costs dropping but heavily subsidized [ex: Germany]
• Bulk-power parity target of 1$ installed per watt

Wind Energy Facts

2008 penetration levels [by energy] 3.7% in EU 2% in the USA 19% in Denmark 11% in Spain, Portugal Aggressive future goals [consumption, not capacity] 20% from renewable sources by 2020 in EU 2-14% from wind by 2020 in EU 30% from renewable sources by 2020 in Denmark “20% Wind Energy by 2030” – US DOE technical feasibility report

Variability

Dealing with variability is the single biggest problem in

utility-scale renewable integration

• Intermittent: Large fluctuations within a day
• Uncertain: power output is random, not known in advance
• Uncontrollable: power output cannot be regulated

Two Consequences

• Under-utilization of T&D resources

Transmission must be sized for peak renewable generation

• Increased reserves to absorb variability

Expensive, defeats the carbon benefits The obstacle: Coordination among and conflicting objectives of power producers, system operators, and regulatory agencies

!"#$%&'()*%+,*",-"."/0

1'2*.0%("#$%)#)*30% 4*'$25/"'#% 6'*$"5%*)3"'# 7)-*2,*0%8999 :;%<%'=%"#;/,..)$% 5,4,5"/0

Source: P. Norgard et al, Fluctuations and predictability of windand hydropower, Risø National Laboratory, Denmark, June 2004

!"#$%&'())*

Dealing with Variability

• 1. Curtail wind
• Makes renewable generation unprofitable
• Spilled wind has value
• 2. Back up with operating reserves
• Costly, who should pay?
• Defeats carbon benefit
• 3. Programmable loads
• Ex: Evs and PHEVs, demand response
• Requires comm infrastructure

New Hardware

Power electronics

• FACTS devices
• STATVARs for reactive power compensation
• Smart Inverters
• Switches for two-way power flow on distribution end

Materials

• For transmission line, longer life transformers

Storage

• Flywheels, CAES, NaS batteries

New Sensing

AMI [advanced metering infrastructure]

• Not so important in my view
• Ex: residential metering

PMUs [Phase measurement units]

• Very important in my view
• Direct measurement of voltage phase at a bus
• Also called synchro-phasors
• Uses GPS clocks
• Accurate, 60 Hz sampling
• IEEE C37.118-2005 standard

!""#$%%&&&'#!()*+,+"-.)'/*.%#!()0+/*1/0#")%#!()*+2(-324(5'!".6

PMU Applications

Wide-area monitoring, protection, and control Early indicators of problems Ex: Florida disturbance of 2003 Replace state estimation by direct state measurement

PMU Deployment

!"##$%&'()*++,)-./0)1%0&2''$34 567$8&$3)&9)$68$$3):++,)-./0);()*+<= =>)?@A2;(&$0)9B)32&2)89''$8&$3)C2@'()B#9D)<++)-./0 EF)<)G$#2;(&$)7$#).9%&HI4 !+,,-./01'23*'4567'#.7/800-9

2:

New Communications

NASPI-net for PMU data

• 1 Terrabyte per month
• Secure
• Publish/subscribe model

Other Multi-layer architectures under development Issues

• Mode: Wireless, cable, power-line
• Protocols
• NIST interoperability standards
• Security
• Ownership

New Loads

Electric vehicles, PHEVs

• These are energy consumers

Deferrable loads

• Commercial refrigeration
• Some manufacturing

Redefining QoS

• Not power-on-demand
• Net energy service in a time window

Demand Response

What is it? [partly] Controllable demand FERC Order 719 elevated DR to generator status Examples

• Industrial:, commercial refrigeration, dimmable lighting in

smart buildings,

• Residential: price-responsive AC, smart appliances,

PHEV/EVs Applications

• Peak-shaving during severe system congestion
• Deferring loads to times when there is a surplus of generation
• Another balancing resource for “generation control”
• Absorbing variability from renewable generation

DR Benefits & Business Models

Benefits

• Reduce wholesale prices during peak periods
• Reduce price volatility
• Reduce capital expenditures in peak generation, transmission,

distribution

• Increase grid reliability

Business models

• Top-down: price signals
• Bottom-up: aggregation [ex: Enernoc]
• I don’t buy residential DR in the US
• The European consumer is different
• Lottery mechanisms for residential DR [Prabhakar@Stanford]

New Business Models

Markets

• Additional intraday markets
• New Ancillary Services Markets
• Inertia markets

Energy Aggregators ex: EnerNoc

• ~$275M in revenue in 2010
• Aggregator or Curtailment Service Provider
• Commercial refrigeration
• ENOC pays for right to defer loads
• Aggregates this and sells to SO as an ancillary service
• Could benefit from distributed decision-making

New Problems

Renewable integration Management of distributed resources Millions of control/optimization loops Requires intelligent architecture

Smart Grids

7*8+/0$9:'9;'9<=>!"10+?9 @'7'9AB:?9C8>8)"9DEEF

Smart Grid Application Areas

$;+,<-='>?5'@-7-8,<A

• C. Opportunities
• 1. Renewable Integration
• 2. Feeder Automation
• 3. Grid2050

Renewable Integration Today

Focus on wind >)#?,*@%A 7*'?%="B)$%4*"5)%='*%("#$%)#)*30%C ?,*@)/%4*"5)%

D%;2-;"$0%

E)*?,#0%C F)#)(,-.)%G#)*30%H'2*5)%:5/I%?2;/%-20%,..%

'==)*)$%("#$%)#)*30%4*'$25/"'#%,/%="B)$%4*"5);

JK%A ,%#)(%=))$C"#%/,*"==%'=%L9M8NO@!P%='*%;?,..%QRST!U%

("#$%4*'V)5/;%

W:%A &XF&%4*'3*,? H2-;"$");Y%=))$C"#%/,*"==;Y%)B/*,C?,*@)/%?)5P,#";?; !"..%#'/%;5,.)%,/%$))4%4)#)/*,/"'#%.)Z).;%5,..)$%='*%"#%F&H

Wind in Conventional Markets

How would a single WPP bid in DA markets? DA markets

• accept firm power
• Penalties for contract deviations
• No subsidy

Must curtail some wind

Pricing

Sell price p [clearing price in DA market] Buy price q [clearing price in RT market]

Assume: prices are constant and known in advance

Remarks

• If contract is not honored, IPP must pay deviation penalties q
• Buy price can also capture cost of using local generation

[CCGT]

• Prices capture economic risk which sets energy risk

Problem Formulation

Conventional contract in DA market Objective functions Profit Expected profit Risk-sensitive profit Can also incorporate value of spilled wind

Optimal Contracts

Open Problems

• 1. Many WPPs servicing many distributed loads
• Network problem with line constraints
• What information should be shared?
• 2. Matching variable generation to programmable loads
• Network problem
• Conventional generation must adjust to realize power flow
• 3. Economics of renewable integration
• Reserve sharing & profit allocation
• Distributed coalitional games
• Who should pay for injecting variability?

These are distributed optimization/control problems

Feeder Automation

Voltage support in Distribution System Why?

Induction loads (ex: motors) most efficient at rated voltage Efficiency drop 2-4% per 1-2% deviation in voltage

Actuation

Load-tap-changing transformers Shunt capacitances D-STATVAR PV inverters

Feeder Automation

Constraints

• Actuation is discrete
• Available only at certain buses
• Limit number of changes (for lifetime issues)
• Voltage measured at some buses
• Voltage support desired at buses

Centralized Problem

Open Problems

• 1. Architectural aspects
• Use of local information
• What is the comm overhead needed?
• 2. Monetizing the benefit
• Who pays for voltage support in feeder system?
• How do you audit QoS?

These are distributed optimization/control problems

Grid2050 Vision

Assumptions

• 40% renewable penetration by energy
• Mainly on distribution end

Distributed Resources

• Appliances, EVs, storage, distributed generation
• Networked

Consequences

• Millions of distributed control loops
• Need to manage resources intelligently
• More energy is produced locally, consumed locally
• Legacy grid diminishes in importance, supplies net energy

imbalance

Managing the Complexity

Resource clusters Managed by a resource aggregator (RA) Net energy imbalance of a cluster

• supplied by legacy grid
• adjustable by shifting demand, use of storage
• uncertain because of randomness in load and generation

Intelligence in the periphery Clusters present less variability to bulk power grid

Grid2050 Architecture

Distributed Control Loops

Grid2050 Benefits

1.

CO2 savings

Enables deep renewable penetrations 2.

Energy cost savings

Reduced variability presented to core grid Smaller reserves requirements 3.

Operational benefits

More energy is produced where it is consumed

• Less system stress, lower congestion,

smaller failure probability, greater operational flexibility

• All these are realized by intelligently managing

distributed resources

Distributed Resource Management

Resource aggregator functions

• Aggregate energy supply and demand within a cluster to compute

future net energy imbalance

• Represent cluster to grid operator as a new entity which offers

programmable load as an ancillary service

• If requested, optimize cluster resources at delivery time to deliver

service

• Issues command to device level control loops

Open Problems

1.

Coarse quantification of economic value in our vision

2.

How do we represent uncertain programmable loads?

3.

How can the cluster manager interrogate its resources to calculate net load on a future horizon?

4.

How can programmable loads be sold into AS markets?

5.

How should the RA dynamically update programmable load models based on new information?

6.

Once a certain commitment is made ex ante, how does the RA optimize resources in its control to deliver the promised load schedule at delivery time?

These are distributed optimization/control problems

More Research Topics

• 1. Intelligent Protection
• Relays on lines
• Trip conditions are (I,t) pairs
• Relay trips when current exceeds I for t seconds
• Completely decentralized
• Could be less conservative by setting relay trip condns based
• n remote information
• Result: greater use of transmission capacity
• 2. Wide-area control
• 3. Dynamic reserve procurement
• 4. Optimal operation of distributed storage

What is a good problem?

Power systems is a mature subject So must work on some new Smart Grid element

• Renewables, Electricity Storage, DR
• New market instruments
• PMUs and other sensing

All of these naturally involve distributed

control/optimization

Barriers to entry

• Must get your hand dirty !
• Must go to bed with a power systems domain expert !!