Inventing an Energy Internet Concepts, Architectures and Protocols for

Smart Energy Utilization

Lefteri H. TsoukalasPurdue University

Fermi National Accelerator Laboratory Colloquium, April 29, 2009

Research Sponsored by Grants from EPRI/DOD, NSF, DOD, DOE

Consortium for the Intelligent Management of the Electric Power Grid (CIMEG)Purdue University, The University of Tennessee, Fisk University, Exelon, TVA, ANL

http://helios.ecn.purdue.edu/~cimeg

Outline

• Global Energy Realities

• Conservation

• Smart Energy

• Energy Internet

• Examples/Applications

• Summary

Global Energy Realities

• World demand for energy: approximately 210 million barrels of oil equivalent (boe) per day (7.5 boe ~ 1 mtoe - metric ton of oil equivalent).

• World oil demand: ~ 85 million barrels of oil per day (MM bpd)

• Aggregate world supply: ~85+ million barrels of oil per day (MM bpd)

• International markets allocate resources by price– Need ~5% excess capacity for a stable market– Markets have difficulties directing capital towards

infrastructural investments– Are we witnessing the beginning of a series of oil-

induced market crises?

Oil Consumption Per Capita

USA: ~25 barrels/year per capita

Japan/S. Korea: ~15 barrels/year per capita

China (2003): ~1.7 barrels/year per capita

India (2003): ~0.7 barrels/year per capita

Global Energy Growth

• Nearly 6.6 billion people with modern needs• Growth in energy demand still has significant

margins for growth– 12% of the world uses 54% of all energy– 33% of the world still has no access to modern energy– The other 45% uses 1/4 of the energy consumed by the

remaining 12% • Energy use by world’s richest 12%

– U.S.: 65 boe energy per person– Japan: 32 boe energy per person– U.K.: 30 boe energy per person– Germany: 32 boe energy per person

Peak (Long Plateau) in Global Oil Production?

Peak 2005(?) 85 mm bpd

By 2020: 50 mm bpd

Bakhtiari, S. A-M. World Oil Production Capacity Model Suggests Output Peak by 2006-07 , Oil and Gas Journal (OGJ), May 2004

~70% of Remaining Reserves

EnergyCrisis

EnergyCrisis Climate

Change

ClimateChange

Future: Utopia – Dystopia?

Resource Constraints

Resource Constraints

FinancialCrisis

FinancialCrisis

First: Do More with Less

• Conservation may be our greatest new energy discovery in the near future

• Smart energy can facilitate further convergence of IT, power, and, transportation infrastructures

• Smart energy can facilitate integrated utilization of new energy carriers – H2, Alcohols, Biofuels

• Can harvest energy usually wasted• Ambient energy MEMS

Electric Power Grid

• Secure and reliable energy delivery becomes a pressing challenge– Increasing demands with higher quality of service– Declining resources

• The North American electric power grid is operating under narrower safety margins– More potential for blackouts/brownouts– Efficient and effective management strategy needed

• Current energy delivery infrastructure is a super complex system (more evolved than designed)– Lack of accurate and manageable models– Unpredictable and unstable dynamics

• Can we build an inherently stable energy network?

US Electric GridUS Electric Grid• Evolved, not-designed• Developed in the first half of the 20th century without a clear

awareness and analysis of the system-wide implications of its evolution

Energy + Intelligent Systems

• Smart Energy = Energy + Intelligent Systems• Smart energy extends throughout the electricity

value chain– Smart Generation– Smart Grid– Smart Loads (End Use)

Smart Energy Distribution Management Systems

• Advanced Metering Infrastructure

• Supervisory Control and Data Acquisition (SCADA)

• Capacitor Bank Control

Source: HP’s, Smart Energy Distribution Management Systems, 2005

HP’s Utility Integration Hub for real-time service integration

Singapore:Smart Energy Vending

Smart Grids

• Smart grids is an advanced concept– Detect and correct incipient problems at avery early stage– Receive and respond to a broader range of information– Possess rapid recovery capability– Adapt to changes and reconfiguring accordingly– Build-in reliability and security from design– Provide operators advanced visualization aids

• Most of these features can be found in the Information Internet– The Internet is also a super complex system– It is remarkably stable

• Can we find an Internet-type network for energy systems?

An Energy Internet

• Benefits of an Energy Internet– Reliability

• Self-configuration and self-healing– Flexibility and efficiency

• Customers can choose the service package that fits their budget and preferences

• Service providers can create more profits through real-time interactions with customers

• Marketers or brokers can collect more information to plan more user-oriented marketing strategies

• The regulation agency can operate to its maximal capacity by focusing effectively on regulating issues

– Transparency• All energy users are stakeholders in an Energy Internet

Storage and Buffer

Buffered Network

Energy Storage• Similar protocols could be developed for the power

grid in order to – Resolve conflicts due to the competition over

resources (peak demand), and– Identify and contain problems locally

IF electricity can be stored in the grid(!)

• Unfortunately, large scale storage of electricity is technologically and economically not feasible

• Solution: a virtual buffer– Electricity can be virtually stored if enough information

is gathered and utilized

Virtual Buffer: Concepts• Assume predictability of demand • Based on demand forecast, the desired amount of electricity is ordered (by an intelligent

agent on behalf of the customer) ahead of time• The supplier receives orders and accepts them only if all constraints are met. Otherwise,

new (higher) price may be issued to discourage customers (devices) from consuming too much electricity

• Price elasticity is used by the supplier to determine the amount of adjustment on price• Once the order is accepted, from a customer's point of view, electricity has been virtually

generated and stored

Time

Period when electricity is virtually stored

Virtual Buffer: Implementation• Information can be used to

achieve the virtual storage of energy

• Two keys for implementation– know electricity demand for

individual customers in advance

– Regulate demand dynamically• Hardware

– An intelligent meter for every customer to handle the planning and ordering automatically

• Algorithms– Demand forecast– Dynamical regulation via price

elasticity

Virtual buffer

Unbuffered Network

An Intelligent

Meter

Example LAG

Customer ID’s

Long- and Short-term Elasticity• Long-term Elasticity

– Average elasticity in within a long period (moths, years)– Usually is an overall index including a large number of

customers– Good for long-term strategic planning– More reliable to estimate

• Short-term Elasticity– Instant elasticity within a very short period (e.g., minutes,

hours)– Can be a local index for a particular customer– Critical for control of the power flow– Difficult to estimate

Managing Short-term Elasticity

• Short-term price elasticity characterizes a particular customer’s nearly instantaneous responsiveness to the change of price

• Short-term elasticity can be estimated from– Historical price-demand data – Psychological models of customer energy behaviors

• The use of intelligent meters is important for– Increasing short-term elasticity => more effective for control– Regulating customers’ behavior => more reliable for prediction

Example Approach

Grid is viewed as polycentric and multilayered system

Customer-driven Grid segmented by

groups of customers (LAGs)

Accurate predictions of nodal demand drive the system

Optimal dispatch of units (storage)

Plug and play tool: TELOS

Transmission

Distribution

Sub-transmission

Residential Customers

Medium-Large Customer

Large CustomersVery Large Customers

To other LAGsTo other LAGs

Customer-SideAgent Space

Generators

Generators

Local Area Grid - LAG

– Defined as a set of power customers

– Power system divided into Local Area Grids each with anticipatory strategies for• Demand-side

management

• Dispatching small units

• Energy storage

• Good neighborly relations

Central Dispatch

Small Generator

Sheddable Load

Data Network

Local Grid 1

Local Grid

Local Grid 2Local Grid 3

TELOS Design Requirements

• TELOS = Transmission-distribution Entities with Learning and On-line Self-healing

• Local Area Grid (LAG)

• Customer-centric

• System Model

• Power System Calculations

• User Interface

• Automated Execution

Examples of Customers in TELOS

0 20 40 60 80 100 1200.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 20 40 60 80 100 120350

400

450

500

550

600

Large Commercial /Industrial (LCI) Customer Hourly Demand (KW-h) for a week

Residential (RSL) Customer Hourly Demand (KW-h) for a week

KWh

KWh

Hourly data starting at 00:00 Monday

Intelligent Power Meter

Pow

er

Info

Prediction Agent

Historical data

Decision Module

Effectors

Database

Actions

LAG Manager

TELOS Simulation

Demand Forecast in TELOS

1/25/2002 2/25 3/25 4/25 5/25/2002

2

4

6

8

10

12

14

Time (30 min interval)

MW

actualprediction

Argonne National Laboratory (ANL)

Dynamic Scheduling via Elasticity

OrderingPowerFlow

PowerPrediction

SecurityCheck

ElasticityModel

PowerFlow

CapacityCheck Scheduling

ElasticityModel

BackupPower

Generation Agent

Customer withIntelligent Meter

Transmission/DistributionAgent

Pric

ing

Info

N

Y

N

Y

N Y

Convergence of IT and Power

• Technological advancements– More information/data is available

• Transmission and delivery system monitoring: SCADA, ….

• Smart grid/smart meter

– More analytical tools are available.• Economics models

• Power system analysis tools

• Can we build around current technologies a more reliable and efficient infrastructure?– Utilize complex systems theories (multi-agents)– Software (information infrastructure) upgrade

Open- vs Closed-Loop System

Generationg(t)

Consumptionc(t)

Storages(t)

Generationg(t+t0)

Consumptionc(t+t0)

Delivery System

Pricing p()

Order f()

Open-Loop System

Closed-Loop System w/ Anticipation

Equilibrium is reached at future state!

Energy Internet: Anticipation and Pricing

Without anticipation

Power Generation Level

0 20 40 60 80 100350

400

450

500

550

600MW

120Hours

0 20 40 60 80 100350

400

450

500

550

600MW

120Hours

Dynamic Generation Consumption

0 20 40 60 80 100350

400

450

500

550

600MW

120Hours

Dynamic Generation Consumption

With anticipation

With anticipation and pricing

0 2 4 6 8 10 12 14 16 18 20 22 2411,000

12,000

13,000

14,000

15,000

16,000

17,000

50

60

70

80

90

100

110

120

130

Peak Demand and Pricing

Source: NE ISO

Prof. G. Gross, UIUC

pri

ce (

$/M

Wh

)

time of day

dem

and

(M

W)

average demand

Research Issues• Multi-agent based methodology• Asynchronous and autonomous system• Demand Side: Smart Meters

– Anticipation– Programmable energy management– Communication and negotiation– Plug-ins

• Supply Side– (distributed) Power system analysis – (distributed) Pricing models– Renewables (solar, PV, geothermal…)– Vehicle-to-grid

Outstanding Issues

• Feasibility– Can system reach equilibrium?

• Efficiency– How much electricity can be conserved (negawatts)?

• Stability– How does the system react in unexpected events?

• Scalability– What if the size of the system increases?

• Cost– How much investment is needed for upgrading current

infrastructure and evolving towards energy internet?

Summary• An Internet-like energy network, an Energy Internet,

representing the smart convergence of Power and IT, is a technically plausible next step

• Intelligent Systems can provide virtual energy storage (via anticipation)

• The Energy Internet may positively shape a sustainable future through more transparent energy relations

• All sources of primary energy will be needed to produce the most easily available energy we have, grid electricity

• Nuclear power has a special role as an important source of emissions-free electricity with some sustainability features

• Important nuclear physics advancements needed to enable global standards for future nuclear fuel cycles

Extra Slides

Load Identification and Wavelets

– Different types of load show a characteristic behavior on the change of wavelet coefficients with respect to scale

– The type of load is possible to be identified with a neural-wavelet approach

Wavelet Decomposition

20 40 60 80 100 120 140 160

-101

Ori

gin

al

20 40 60 80 100 120 140 160

-1-0.5

00.5

s

20 40 60 80 100 120 140 160-0.5

0

0.5

d5

20 40 60 80 100 120 140 160

-0.5

0

0.5

d4

20 40 60 80 100 120 140 160-0.5

0

0.5

d3

20 40 60 80 100 120 140 160-0.5

0

0.5

d2

20 40 60 80 100 120 140 160

-0.2

0

0.2

d1

20 40 60 80 100 120 140 160

0

2

4

Ori

gin

al

20 40 60 80 100 120 140 160

-0.20

0.20.40.6

s

20 40 60 80 100 120 140 160

-0.50

0.5

d5

20 40 60 80 100 120 140 160

-0.50

0.5

d4

20 40 60 80 100 120 140 160-1

0

1

d3

20 40 60 80 100 120 140 160

-1

0

1

d2

20 40 60 80 100 120 140 160

-101

d1

LCI RLS

Load Identification

LCI RLS

0 50 100 150 200 250 300 350-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

0 50 100 150 200 250 300 350-0.5

0

0.5

1

Structure of LAG

LAGMANAGER

Producer #1

Producer #2

Producer #N

Consumer #1

Consumer #2

Consumer #M

To neighboring LAG

LAG Manager

• Collects current demand and supply data• Checks for grid stability based on predictions

of individual customer demand and available pathways

• Makes decisions on when to dispatch local units and/or manage load based on demand/supply and contracts between customers and producers/providers

• Sends decisions to individual agents

Interconnecting LAGs

• Reliable network connections with adequate redundancy

• Each LAG connecting to at least two neighboring LAGs

• Neighboring LAG Manager should be able to take over in case of local fault

LAG Agent Hierarchy

Substation

Transformers

Feeders

Customers

Agent Functionalities

• Customer Agent: contains a neurofuzzy predictor to predict future demand

• Feeder Agent: sums up predicted demands of all customers connected to the feeder, performs internal overload check

• Transformer Agent: sums up predictions of all the feeders

TELOS Implementation

• Distribution Agency: Propagates Agents Through the Network

• Intelligent Meters Contract for Power in a Central Database

Intelligent Power MeterPow

er

Valu

e

Prediction Agent

Historical data

Decision Module

Effectors

Database

Actions

LAG Manager

Anticipatory Control of Small Units

The objective at time k is to find an open loop set of constrained control actions u(k) to drive the plant outputs y(k) along a desired trajectory with anticipated disturbances v(k).

Controller Plant

Anticipated Disturbances

Current System Trajectory Estimation

Noise

ky * ky

k

kukv

Load Schedule

Forecasting

Nk

k

k

N

u

u

u

U

1

Logging Optimal Control Patterns• Iterate to find the optimal control response

• Log the system trajectory at time 0

• Log the predicted error at time 1

• Log the optimized control action at time 0

Control Action

Logging Optimal Control Patterns• Train a neural

network to learn the optimal, tuned, anticipatory control response

• Example: If the error will be negative and the current system trajectory is level, then decrease fuel flow rate.

Control Action

Anticipatory Control of Small Units

Smoothness of Control - Time Rate of Change of Fuel Flow

Conventional Control

Anticipatory Control

• 1/25th of Control Effort• Reliability and Maintenance Benefits• Energy Savings