Dynamic Paradigm for Grid Operations - Department of … · Dynamic Paradigm for Grid Operations ... Transfer limit of a critical path, ... – Dynamic state estimation based on Advanced

Dynamic Paradigm for Grid Operations

Henry Huang

Pacific Northwest National Laboratory

Collaborating Organizations: Binghamton University, Powertech Labs

June 17, 2014

1

Presentation outline

• Project Purpose: Capturing Dynamics

• Significance and Impact

• Technical Approach – A Dynamic Paradigm for Operation

• Technical Accomplishments

– Dynamic State Estimation

– Look-Ahead Dynamic Simulation

– Dynamic Contingency Analysis

• Transient Stability

• Voltage Stability

• Synergistic Work

• Summary

• Contacts

2

Capturing dynamics: Enable grid operations from reactive to predictive

3

Today’s paradigm:Slow data

Static modeling

State Estimation

Contingency AnalysisSCADA

DynamicState

Estimation

Look‐ahead Dynamic Simulation

Dynamic Contingency Analysis

Graphical Contingency Analysis

Future paradigm:High‐speed dataDynamic modeling

Phasor

Predictive

Reactive

Data cycle

Time1/30 sec 2/30 sec 3/30 sec

Dynam

ic States

Look-ahead dynamic simulation

1 min

Dynamic contingency analysis

Dynamics state estimation

5 10 15 200

500

1000

1500

2000

2500

Time, hour

Tra

nsfe

r lim

it of

a c

ritic

al p

ath,

MW Real-time Path Rating

Offline path rating, current practice

25.74% more energy transferusing real-time path rating

5 10 15 200

500

1000

1500

2000

2500

Time, hour

Tra

nsfe

r lim

it of

a c

ritic

al p

ath,




Real‐Time Path Rating

Other Applications

Projects and teams

• Dynamic Paradigm for Grid Operations– Henry Huang, Ning Zhou (Binghamton University), Steve Elbert,

Shuai Lu, Da Meng, Shaobu Wang, Ruisheng Diao

• Look-Ahead Dynamic Simulation– Shuangshuang Jin, Ruisheng Diao, Di Wu, Yousu Chen

• Dynamic Contingency Analysis– Yousu Chen, Mark Rice, Shuangshuang Jin, Kurt Glaesemann

• Non-Iterative Voltage Stability Analysis– Yuri Makarov, Bharat Vyakaranam, Da Meng, Pavel Etingov,

Tony Nguyen, Di Wu, Zhangshuan (Jason) Hou, Shaobu Wang,Steve Elbert, Laurie Miller

4

See project details on posters…

2007 2,008 2,009 2,010 2,011 2,0120.01

0.011

0.012

0.013

0.014

0.015

0.016

0.017

0.018

0.019

0.02

Year

Roo

t Mea

n S

quar

e E

rror

of F

requ

ency

Sig

nal,

Hz

0.005

0.01

0.015

Mea

n of

Abs

olut

e F

requ

ency

Dev

iatio

n, H

z

Trend of frequency variance

5

Freq

uen

cy

60 Hz

• Traditional state estimation does not accurately capture the system’sdynamic status

– During emergency situations, frequency is significantly off nominal 60Hz

– During normal operation, frequency tends to deviate more often from 60Hz

Challenges in future power grid operations

“Grid evolution meets information revolution” • Grid Evolution –

stochastic & dynamic– Generation: intermittent

renewable energy, distributed generation

– Demand: smart loads, plug-inhybrids

– Other: storage, new marketdesign/incentives

• Information Revolution –data rich but information scarce– Large number of phasor

measurement units, smart meters, and intelligent devices

– Requirements of cybersecurity

Transmission Grid

6

Dynamic paradigm for grid operation

• National Driver: clean and efficient power grid as well as being affordable,reliable, and secure dynamic and fast operation

• Technical Approach: combine model prediction and measurementobservations to determine where we are, where we are going, and what-ifs

– Fuse models and data with nonlinearity, discontinuity, model deficiency, and datasparsity

– Develop Advanced Kalman Filter and HPC codes to estimate states and models

– Solve a large number of ODE systems to predict future states and alternativestates

7

Data collection cycle


Dynam

ic States


1 min


Dynamic state estimation

Better Reliability

Clean Energy

Integration

Better Asset

Utilization

Enabling factor ─

Phasor measurement

• Time-synchronized, high-speed measurement at 30 samples persecond, able to capture the majority of grid dynamics

8

NASPI: www.naspi.org

Enabling factor –

Computing technology

• CMOS technology hitting its limits due to power dissipation for theincreased clock speed (4-6 GHz)

• Moore’s Law still applies – multiple-core processors (parallelcomputers)

9

– PCs: Two-, four-, andeight-cores

– HPCs: 100s~1,000s cores

– Forthcoming high-endparallel systems expectedto offer million cores

• Parallel computingbecomes a fundamentaltechnique

Advanced Kalman Filter for dynamic state estimation

10HPC Platform

x(k‐1)

z(k)

x(k)x’(k) “Correction”Measurement Eq’s

z = h(x, α) +

R

“Prediction”Dynamic Simulation

dx/dt = f(x, α)

Q

Standard KF

Correction Cycle ~1/30 second

Sensor Placement

Sequential Estimation

Measurement Selection

NonlinearityModel deficiency DiscontinuityData sparsityLarge dimension

Adaptive Tuning

P

dα/dt = g(x, α)

,α’(k) ,α(k),α(k‐1)

Prediction Cycle milliseconds

0 5 10 15 200.5

1

1.5

States tracking (Basic EnKF)

Re

lativ

e R

oto

r A

ng

le (

rad

)

0 5 10 15 20

0

5

10

15

x 10-3

Sp

ee

d D

evi

(p

u)

TrueMeanMean+/-3*Std100 MC Ests

0 5 10 15 20

0.85

0.9

0.95

Eq

' (p

u)

Time (sec)0 5 10 15 20

0.5

0.55

0.6

0.65

Ed

' (p

u)

Time (sec)

Performance evaluation – estimation accuracy

• Excellent tracking with realistic evaluation conditions– 3% measurement noise; 40 ms measurement cycle (phasor measurement)

– 5 ms interpolation cycle; modeling errors considered; unknown inputs;unknown initial states

11

Blue/green tracking red is the goal

Filtering technology assessment

12

Extended

Kalman Filter

Unscented

Kalman Filter

Ensemble

Kalman FilterParticle Filter

Accuracy

The 2nd best

with 0%

diverged

33% divergedThe best with

0% diverged

20% diverged

(PF 2000)

Efficacy of

interpolationHigh High Low High

Number of samples

neededNone Small Medium Large

Sensitivity to missing

dataLow Low Low Low

Sensitivity to outliers Low Low Medium High

Computation time

(non‐parallel)Shortest

Same order

as EKF

longer than

EKF

Same order

as EnKF

Computational performance – scalability

• Current codes scale to ~1,000 cores

• Current computational performance meets the real-time requirement forregional systems

• Performance is expected to be real-time soon for interconnection-scalesystems

13

0.0001

0.001

0.01

0.1

1

10

100

1000

10000

0.003 0.030 0.300 3.000 30.000

TeraFLOPS

Seconds

2017?

30 ms

Oct 2012

June 2014

Oct 2012

June 2014

Computational time per estimation step

Look‐Ahead Dynamic Simulation

• Performance evaluation with the WECC system

• 13x faster compared to a commercial tool with sequential computing

Time distribution for WECC system(Simulation length: 30s with 0.005s time step)

# of threads

Algebraic Equation

Integration Total

1 13.26 218.87 232.14

2 8.99 111.97 120.95

4 6.46 57.49 63.96

8 5.07 29.58 34.65

16 4.58 10.64 15.22

32 4.14 6.41 10.55

64 4.16 4.88 9.04

14


• Solving a large set of Differential Algebraic Equations

• Computational challenge is load balancing – dynamic load balancingvs. static load balancing

15

0

2000

4000

6000

8000

10000

12000

0 2000 4000 6000 8000 10000

Speedup

Number of cores

0 50 100 150 2000

10

20

30

40

50

number of cores

spee

dup

Speedup vs. number of cores

Commercial Tool on HPCvia Virtual Machine

Native HPC Implementation

Voltage Stability Analysis

16

• Objectives:– Generate voltage stability boundary (VSB) accurately and quickly

– Use parallel computing to further enhance speed

– Provide connectivity to commercial tools such as PowerWorld

• A combination of methods toexplore the VSB:– Continuation power flow (CPF)

– New PNNL method based on theX-ray theorem (XR)

– Orbiting method (OM) - a directVSB tracing procedure

– High-order numerical methods(HO)

Voltage Stability Analysis –

Computational performance

• Minimizing the CPF runs results in great improvement incomputational performance

– 5.6x speedup for WECC-sized system in this example

• Parallel computing version is being implemented and expected tofurther improve the performance

17

1Boundary Points2Average Time per Run (s)

Current Operating Point

Boundary Points

Use cases of dynamic paradigm

• Real-time predictive gridoperation with calibrated modeland parameters for fasterresponse and control

• Effective management of large-scale integration of smart gridtechnologies such as renewablegeneration, demand response,electric vehicles, and distributedgeneration

• Better asset utilization tomaximize power transfercapabilities and defertransmission expansion

18

Phasor measurement

Dynamic state estimation

Base case

analysis

Contingencycase

analysis

Violations?

Contingencylist

Reporting

Real‐timetransferlimits

Real‐timetransferlimits

Advanced computing platform

1

2

3

Real‐time calibrated model and parameters

Synergistic work

• M2ACS Predictive Modeling, DOE-ASCR– Stochastic modeling and data assimilation

• GridPACK™, DOE-OE– Parallel computing library for efficient development and better

compatibility

• Non-Wire Methods for Congestion Management, DOE ARPA-E– Transient and voltage stability analysis

– Asset utilization improvement through real-time path rating

See project details on posters…

19

Summary

• Grid evolution meeting information revolution grid operationstransition from static & slow to dynamic & fast– Enabling technologies are computation advancement and data

development

• A dynamic paradigm is necessary to capture emerging dynamics andunderstand where the grid is, where the grid is going, and where thegrid could end up– Dynamic state estimation based on Advanced Kalman Filter

– Look-ahead dynamic simulation

– Dynamic contingency analysis of transient and voltage stability

• This paradigm is expected to facilitate integration of new generationand load for a more reliable, efficient, and cleaner power grid

20


21


Static modeling

State Estimation


DynamicState

Estimation

Proven math for large‐scale system


13X speedup Sec’s‐Min’s ahead


45X speedup on 64 cores Scalable performanceDays to hours



Phasor

Predictive

Reactive

Data cycle


Dynam

ic States


1 min



5 10 15 200

500

1000

1500

2000

2500

Time, hour

Tra

nsfe

r lim

it of

a c

ritic

al p

ath,




5 10 15 200

500

1000

1500

2000

2500

Time, hour

Tra

nsfe

r lim

it of

a c

ritic

al p

ath,





Other Applications

Contacts

• Dynamic Paradigm for Grid Operations– Henry Huang, 509-372-6781, [email protected]

• Look-Ahead Dynamic Simulation– Shuangshuang Jin, 206-528-3061, [email protected]

• Dynamic Contingency Analysis– Yousu Chen, 206-528-3062, [email protected]

• Non-Iterative Voltage Stability Analysis– Yuri Makarov, 509-372-4194, [email protected]

22

Questions?

23

Backup slides

24


25


Static modeling

State Estimation


DynamicState

Estimation

Proven math for large‐scale system


13X speedup Sec’s‐Min’s ahead


45X speedup on 64 cores Scalable performanceDays to hours



Phasor

Predictive

Reactive

Data cycle


Dynam

ic States


1 min



5 10 15 200

500

1000

1500

2000

2500

Time, hour

Tra

nsfe

r lim

it of

a c

ritic

al p

ath,




5 10 15 200

500

1000

1500

2000

2500

Time, hour

Tra

nsfe

r lim

it of

a c

ritic

al p

ath,





Other Applications

Mitigate intermittency of renewable energy

Source: BPA Fact Sheet, “BPA’s wind power efforts surge forward,” March 2010

500MW

26

Thermal rating

(N‐0 = 10,500 MW)

(N‐1 = 6000 MW)

Stability Rating

(N‐1, N‐2 = 4,800 MW)

WECC/NERC Criteria

Transfer Capacity Example – California Oregon Intertie (COI)

U75 – % of time flow exceeds 75% of OTC (3,600 MW for COI)

U90 ‐ % of time flow exceeds 90% of OTC (4,320 MW for COI)

U(Limit) ‐ % of time flow reaches 100% of OTC (4,800 MW for COI)

WECC

Path Ratings U75, U90 and U(Limit)

% of OTC

75% 90% 100%

Source : Western interconnection 2006 congestion management study

Improve asset utilization

27

Today’s power grid operation paradigm

~1 min

2‐5 mins

• Steady-state model based: static and slow paradigm

28

Computers are more affordable

29

Data Source: http://en.wikipedia.org/wiki/FLOPS

1¢

2005 (8 years later)

Sony VAIO GRT-290Z6 GFLOPS, ~$2,000

2012 (15 year later)

LG Optimus 4X HD P88024 GFLOPS, $600

1997

IBM Deep Blue11.38 GFLOPS, >$100M

Dynamic Paradigm for Grid Operations - Department of … · Dynamic Paradigm for Grid Operations ... Transfer limit of a critical path, ... – Dynamic state estimation based on Advanced

Documents