A Framework for Analyzing the Impact of Data

Integrity/Quality on Electricity Market Operations

Dae-Hyun Choi

(Advisor: Prof. Le Xie)

Department of Electrical and Computer EngineeringTexas A & M University

February 4, 2014

1 / 51

1 Introduction

2 Research Goals

3 Part I: Data Attack on Look-Ahead Dispatch

4 Part II: Sensitivity Analysis of LMP to Data Corruption

5 Conclusions

2 / 51

Smart Grid: A Cyber-Physical System

Image Source: http://www.sgiclearinghouse.org/sites/default/files/images/nist/Slide1-1.png

3 / 51

Advanced Grid Sensors Improve Smart Grid Operations

Transmission

Operator

Distribution

Operator

Distributed

Resources

Industrial

Customer

Commercial

Customer

Residential

Customer

Microgrid

Substation

Other

Substation

Other

Substation

Energy

Storage

Communication

s

Smart Switching Devices Sensor Advanced Computing

Image Source: EPRI

4 / 51

Advanced Grid Sensors Improve Smart Grid Operations

Transmission

Operator

Distribution

Operator

Distributed

Resources

Industrial

Customer

Commercial

Customer

Residential

Customer

Microgrid

Substation

Other

Substation

Other

Substation

Energy

Storage

Communication

s

Smart Switching Devices Sensor Advanced Computing

Image Source: EPRI

5 / 51

Data Quality in Future Grid

3.5

3.0

2.5

2.0

1.5

1.0

0.5

0.0

1980 1990 2000 2010 2020 2030

Exab

yte

syear

(1 e

xab

yte

=250 m

illi

on

DV

Ds)

Distribution

automation

Workforce

management

Substation

automation

Transmission

Sensors

Building energy

management

Distribution

management

AMI

deployment

Smart in-home

devices

Image Source:EPRI

Big data explosion!I Big data analytics

Data quality analysisI Pre-processing for big data

analytic

Transmission sensorsI SCADA/PMU data

6 / 51

Data Integrity in SCADA System

Stuxnet Worm, 2010

Nuclear power plant attacked via SCADA systems

A Silent Attack,

but Not a Subtle One

Stuxnet: Computer worm

opens new era of warfare

Stuxnet: Malware more

complex, targeted and

dangerous than ever

* SCADA (Supervisory Control And Data Acquisition)

SmartGrid Update Report*

“Even small changes in the data could affect the stability of the grid andeven jeopardize human safety”

SCADA Weak Cybersecurity+Data Integrity Violation ⇒ Grid Malfunction

* http://www.smartgridupdate.com/dataforutilities/pdf/DataManagementWhitePaper.pdf 7 / 51

Motivation

EMS (Energy

Management

System)

MMS (Market

Management

System)

MTU (Master

Terminal

Unit)

RTU (Remote

Terminal

Unit) SCADA

Network

Control Center SCADA system

Noise

Adversary

Data quality/integrity violation ⇒ blackouts & financial losses

8 / 51

Background: Power System State Estimation

= ( , ) = ( , )

G G

= ( , )

Three-bus system

G G

Base case power flow solution

= +

= ( , , , , )

= ( , , , , , )

= ( , , , , )

minimize =

s.t = ( )

Weighted least squares method

= ( , , , , , )

State estimation solution

Measurement model

9 / 51

Background: Electricity Market Operations

… Hour(j) Day(k-1) Day(k)

… …

5-min(i) Time

Time

(5-min)

: Forecasted load for static dispatch : Forecasted load for look-ahead dispatch

Present

Ex-post

real-time

price

Ex-ante

real-time

price

Static dispatch

Look-ahead dispatch

Day-ahead

Real-time economic dispatch

Min ( )

s.t. Balance constraint of supply and demand

Operational system constraints

Base Case Power

Flow Solution

State Estimation

10 / 51

Problem Statement

Normal Condition

Physical Layer

(Power Network)

Economic

Dispatch

Measurement Layer

(Sensor Network)

Control/Computation Layer

(Control Center)

State

Estimation

SCADA Network

11 / 51

Problem Statement

Normal Condition

Physical Layer

(Power Network)

Economic

Dispatch

Measurement Layer

(Sensor Network)

Control/Computation Layer

(Control Center)

State

Estimation

SCADA Network

After Data Corruption

Physical Layer

(Power Network)

Economic

Dispatch

Bad/Malicious Data

Control/Computation Layer

(Control Center)

State

Estimation

SCADA Network

1 What are the impacts of data integrity/quality on real-time market prices,namely locational marginal price (LMP), via state estimation?

2 What are analytical tools for quantifying such impacts?

11 / 51

Previous and Missing Work

I. Data Integrity Attack on Physical and Economical Grid Operations

Attack modeling & analysis based on continuous data manipulation:[1, Liu et al., 2009], [2, Kosut et al., 2010], [3, Kim et al., 2011]

Attack modeling & analysis based on discrete data manipulation:[4, Kim et al., 2013]

Data attack on static economic dispatch: [5, Xie et al., 2011]

Data attack on look-ahead economic dispatch: ?

II. LMP Sensitivity Analysis Subject to Power System Condition

Impact of physical system conditions (e.g., load variations) on LMPsensitivity: [6, Conejo et al., 2005], [7, Li et al., 2007]

Impact of sensor data quality on LMP sensitivity: ?

12 / 51

Research Goals

I A Market Participant’s Perspective

Part I: Data Integrity Attack on Look-Ahead Economic Dispatch

Ramp-induced data (RID) attack [Choi, Xie, TSG2013]

Undetectable and profitable RID attack strategy

Economic impact of RID attack

I A System Operator’s Perspective

Part II: Sensitivity Analysis of LMP to Data Corruption

Impact of continuous data quality on real-time LMP[Choi, Xie, TPS2014]

Impact of discrete data quality on real-time LMP[Choi, Xie, SmartGridComm2013]

13 / 51

Part I: Malicious Ramp-Induced Data (RID) Attack

I A Market Participant’s Perspective

RID Attack on Look-Ahead Dispatch in Real-Time Market

1 Attack Modeling- Generation capacity withholding- Covert change of generators’ inter-temporal ramp constraints

2 Performance Evaluation- Undetectability- Profitability

14 / 51

State Estimation Model

Power

Network

Sensor

Data

State

Estimation

Economic

Dispatch

z

I Measurement Model ⇒ z = Sx+ e

z: measurements vector, e ∼ N (0,R)

S =

[IHd

]: system factor matrix

x: (nodal power injection) states vector

I Weighted Least Squares Estimate

x(z) = (STR−1S)−1STR−1z = Bz

I Bad Data Detection (Chi-squares test)

J(x(z)) = rTR−1rH1

≷H0

ηχ

where r = z− Sx(z)

15 / 51

Economic Dispatch Model

Power

Network

Sensor

Data

State

Estimation

Economic

Dispatch

Pgi(z)

I Look-Ahead Dispatch Model

minPgi

[k]

K∑

k=1

∑

i∈G

Ci(Pgi [k ])

s.t.

∑

i∈G

Pgi [k ] =

N∑

n=1

Dn[k ] ∀k = 1, . . . ,K

|Pgi [k ]−Pgi [k−1]|≤Ri∆T ∀k = 1, . . . ,K

Pmingi

≤Pgi [k ] ≤ Pmaxgi

∀k = 1, . . . ,K

Fminl ≤Fl [k ] ≤ Fmax

l ∀k = 1, . . . ,K

∀l = 1, . . . , L

Attack Target: Pgi [0] is updated with Pgi (z) at every dispatch interval!

16 / 51

Data Attack Model

Power

Network

Bad/Malicious

Data

State

Estimation

Economic

Dispatch

Pgi(za)

za

(za)

P*gi(za)

I Attack Measurement Model

⇒ za = Sx+ e+ a

za: corrupted measurement vector

a: injected attack vector

I A Domino Effect of Data Attack

za ⇒ Pgi (za) ⇒ λ(za)

17 / 51

Two Main Features of RID Attack: (1) Undetectability

I After data attack, we have

New estimator: x(za) = Bza = x(z) + Ba

New residual:

||r′||2 = ||r + (I− SB)a||2 ≤ ||r||2︸︷︷︸Without attack

+ ||(I− SB)a||2︸ ︷︷ ︸With attack

I For undetectability, the attacker’s goal is to

Construct a such that the contribution of ||(I− SB)a||2 still makesthe following healthy detection condition hold true:

||r′||2 < η

18 / 51

Two Main Features of RID Attack: (2) Profitability

Without Attack

P*gi[1]

Ri

Pgi,min

Pgi,max

P*gi[0]= Pgi[0]

: Ri -

(Shortage Power)

Ri -

With Attack

P*gi[1]

Ri

Pgi,min

Pgi,max

Pgi,a[0]

P*gi[0]

-CM(a)

P*gi,a[1]

:(Excess Power)

Capacity withholding condition: −CM(a) > Ri∆T −∆L

I increasing LMP and profit

19 / 51

Attack Strategy: Compute the Attack Vector a

maxa∈span(A)

δ

s.t.

||(I − SB)a||2 ≤ ε ⇒ Undetectable Condition

αCM(a) + βCB(a) ≤ ∆L− Ri∆T − δ ⇒ Profitable Condition

δ > 0

where

CM(a) = E [Pgi,a [0]− P∗gi[0]] = Bia

CB(a) =∑

j∈G cM

E [Pgj,a [0] + Rj∆T − Pmaxgj

[0]] =∑

j∈G cM

[Bja+ Rj∆T ]

I α = 1, β = 0: Marginal unit attack (Case I)I α = 0, β = 1: Binding unit attack (Case II)

I α = 1, β = 1: Coordinated attack (Case III)20 / 51

Simulation Setup

G

: Compromised Injection Sensor

G

G

G

G

1

2

5

3

4

78

6

12

13

11 10

9

14

Figure : IEEE 14-Bus System.

Unit Type Pmin Pmax Ramp Rate Marginal Cost

Coal(1) 0MW 200MW 10MW/5min 30$/MWhWind(2) 0MW 300MW 150MW/5min 20$/MWh

Nuclear(3) 0MW 300MW 8MW/5min 40$/MWhCoal(6) 50MW 250MW 15 MW/5min 55$/MWhOil(8) 60MW 150MW 60 MW/5min 60$/MWh

Table : Generator Parameters.

21 / 51

Attack Performance

Case Static (PE(3)%) Look-ahead (PE(3)%) J (ηχ=37.6)

I 131.9 148.9 28.2

II 101.2 102.6 35.5

III 108.9 113.8 31.5

I Case I: P3 injection sensor compromisedI Case II: P1 injection sensor compromisedI Case III: P1, P3 injection sensors compromised

Observation 1

Attack profitability (PE(3) > 100%)

Attack undetectability (J < ηχ=37.6)

22 / 51

Ramp-Induced Data Attack Increases LMPs

0 50 100 150 200 250 3000

5

10

15

20

25

30

Time interval (5 min)

LMP

incr

ease

($\

MW

h)

Figure : LMP Increase of Look-ahead Dispatch with Case 1 Attack.

23 / 51

Attack Relative Magnitude vs Attack Performance

Attack Relative Magnitude (ARM)(

||a||∞||z||∞

%)

0.25 0.5 0.75 1Static (PE(3)) 111.8 120.8 126.4 126.9

Look-ahead (PE(3)) 112.2 125.8 127.6 137.7J 21.1 25.4 29.2 33.1

Observation 2

Increasing ARM ⇒ increasing attack profit at the expense ofincreasing J

24 / 51

Ramp Rate & Data Accuracy vs Attack Profit

Ramp Rate (MW/5min) Variance (σ2)8 10 12 14 0.0005 0.005 0.05 0.5

Static (PE(3)) 131.9 119.7 106.4 100.5 123.2 129.1 130.3 136.9Look-ahead (PE(3)) 148.9 123.5 108.5 103.1 143.5 144.75 146.1 152.8

Observation 3

A slower ramp rate unit targeted ⇒ increasing attack profit

Observation 4

A less accurate sensor compromised ⇒ increasing attack profit

25 / 51

Part I: Remarks

Main Contributions1 Problem formulation of a novel ramp-induced data attack

I covert generation capacity withholding

2 An optimization-based undetectable/profitable attack strategy

3 Economic impacts on real-time electricity market operations

26 / 51

Part II: Sensitivity Analysis of LMP to Data Corruption

I A System Operator’s Perspective

SCADA

Telemetry

Topology

Processor

Observability

Analysis

State

Estimation SCED

Bad Data

Processing

Topology Error

Processing

EMS MMS SCADA

Impact flow of continuous data Impact flow of discrete data

Data corruption

(a)

(b)

(a): Power Flow Estimate (b): Network Topology Estimate

Part II-A: impact of undetectable error in (a) on LMP

Part II-B: impact of undetectable error in (b) on LMP

27 / 51

Research Goal

I Develop analysis tools to study the impact of data quality on LMP

G

G

G

G

G

1

2

5

3

4

78

6

12

13

11 10

9

14

: Injection sensor : Flow sensor

: Voltage magnitude sensor

$

$ : LMP change

$ $

$

$

$

$ $

$

$

$

$

$

$

(a) Continuous data corruption (Part II-A)

G

G

G

G

G

1

2

5

3

4

78

6

12

13

11 10

9

14

: “On” circuit breaker

$

$ : LMP change

$ $

$

$

$

$ $

$

$

$

$

$

$

: “Off” circuit breaker

: Line exclusion

(b) Discrete data corruption (Part II-B)

28 / 51

Part II-A: LMP Sensitivity to Continuous Data Corruption

G G

= ( ) +

( ) =

State Estimation

Ex-ante

Pricing

Ex-post

Pricing

( ) ( )

( ( ))

{ ,

( ( ))}

Power System

SCED Ex-ante Dispatch

I Composite function of the Ex-ante andEx-post LMP vectors:

LMP = π(x(z))

I Proposed sensitivity matrix:

Λ =∂π

∂z=

∂π

∂x︸︷︷︸ΛA

∂x

∂z︸︷︷︸ΛB

ΛA : Sensitivity matrix of LMPs to state estimates (Economic Impact)

ΛB : Sensitivity matrix of state estimates to sensor data (Cyber Impact)

⇒ A unified closed-form LMP sensitivity matrix Λ

29 / 51

Continuous Data Corruption Manipulates LMP

minPgi

Nb∑

i=1

Ci (Pgi )

s.t.

λ(za) :

Nb∑

i=1

Pgi =

Nb∑

i=1

Ldi

τ (za) : Pmingi

(za) ≤ Pgi ≤ Pmaxgi

(za) ∀i = 1, . . . ,Nb

µ(za) : Fminl ≤

Nb∑

i=1

Sli(Pgi − Ldi ) ≤ Fmaxl ∀l = 1, . . . ,Nl

I Domino effect: za ⇒{Pmingi

(za), Pmaxgi

(za)}

⇒ {λ(za),µ(za)} ⇒ LMP(za)

LMP(za) = λ(za)1Nb− ST [µmax(za)− µmin(za)]

30 / 51

Derivation of ΛA: KKT Condition Perturbation Approach

I KKT equations

(i)∂Ci (Pgi

)

∂Pgi

− λ +

Bg∑

j=1

τjAji +

Bf∑

l=1

µlSli = 0

∀i = 1, . . . ,Nb

(ii)

Nb∑

i=1

Pgi=

Nb∑

i=1

Ldi

(iii)

Nb∑

i=1

AjiPgi= Cj ∀j = 1, . . . ,Bg

(iv)

Nb∑

i=1

Sli [Pgi− Ldi

] = Dl ∀l = 1, . . . ,Bf .

I Perturbed KKT equations

(i)∂

∂Pgi

(∂Ci (Pgi

)

∂Pgi

)

︸ ︷︷ ︸

Mi

dPgi− dλ +

Bg∑

j=1

Ajidτj

+

Bf∑

l=1

Sli dµl = 0 ∀i = 1, . . . ,Nb

(ii)

Nb∑

i=1

dPgi=

Nb∑

i=1

dLdi

(iii)

Nb∑

i=1

Aji dPgi= dCj ∀j = 1, . . . , Bg

(iv)

Nb∑

i=1

SlidPgi=

Nb∑

i=1

SlidLdi∀l = 1, . . . ,Bf .

For example,

(ii)∑Nb

i=1 Pgi =∑Nb

i=1 Ldi =⇒∑Nb

i=1 dPgi =∑Nb

i=1 dLdi

31 / 51

Derivation of ΛA: KKT Condition Perturbation Approach

I Perturbed KKT equations in matrix form

M −1NbΥ

1TNb0 0

ΥT 0 0

︸ ︷︷ ︸Ξ

dPg

dλdτ s

dµs

=

[U1

T U2T

]︸ ︷︷ ︸

Φ

[dLd

dCs

]

I Sensitivity of lagrangian multipliers to estimated capacity limit

dPg

dλdτ s

dµs

= Ξ−1Φ︸ ︷︷ ︸

Λp

[dLd

dCs

]=⇒ Λp =

[ΛLd

ΛCs

]=

∂Pg

∂Ld

∂Pg

∂Cs∂λ∂Ld

∂λ

∂Cs∂τ s

∂Ld

∂τ s

∂Cs∂µs

∂Ld

∂µs

∂Cs

Finally, ΛA is constructed with ∂λ

∂Csand

∂µs

∂Cs

32 / 51

Derivation of ΛB : Iterative State Estimation Equation

I Gauss-Newton iterative equation for state estimation

d xk+1 = [G(xk)]−1HT (xk)R−1

︸ ︷︷ ︸Ψ(xk )

dzk

m[d θ

k+1

dVk+1

]=

[Ψ

θ(xk)

ΨV(xk)

]dzk

⇓

I Sensitivity of linearized real power estimates to sensor data

d zr =

BSPθ

BPθ

BFθ

d θ =

BSPθ

BPθ

BFθ

Ψ

θdz

I Desired sensitivity matrix

ΛB =

BSPθ

BPθ

BFθ

Ψ

θ

33 / 51

Simulation Setup

G

G

G

G

G

1

2

5

3

4

78

6

12

13

11 10

9

14

: Injection sensor : Flow sensor

: Voltage magnitude sensor

Figure : IEEE 14-bus system with a given measurement configuration.

Table : Generator parameters in the IEEE 14-bus system.

Bus Pmingi

(MW) Pmaxgi

(MW) ai ($/MWh) bi ($/(MW)2h)

1 0 332.4 20 0.0432 0 140 20 0.253 0 100 40 0.016 0 100 40 0.018 0 100 40 0.01

34 / 51

Simulation Results

Using a closed-form LMP sensitivity matrix Λ = ΛA · ΛB ,

1 2 3 4 5 6 7 8 9 10 11 12−4

−3

−2

−1

0

1

Real Power Flow Measurement

∂πn/∂

z m

Bus 1Bus 2Bus 3Bus 4Bus 5Bus 10Bus 13The most impacting measurement 8

The most vulnerable bus 3

(a)

1 2 3 4 5 6 7 8 9 10 11 12−0.4

−0.3

−0.2

−0.1

0

0.1

0.2

0.3

0.4

Reactive Power Flow Measurement

∂πn/∂

z m

Bus 1Bus 2Bus 3Bus 4Bus 5Bus 10Bus 13

(b)

35 / 51

Key Observations

1 Sensitivity grouping propertyI Identical positive or negative sensitivity bus group to data corruption

2 Economically sensitive physical and cyber assetsI Buses with LMP highly sensitive to data corruption.I Significantly influential sensors on LMP change.

3 Impact of different types of sensor data on LMPI A more significant impact of real power sensor data on LMP sensitivity

36 / 51

Part II-B: LMP Sensitivity to Network Topology Error

I Two types of topology error

1

closed open

2

1 2

Line Exclusion/

Inclusion Bus Split/

Merging

2’

37 / 51

Part II-B: LMP Sensitivity to Network Topology Error

I Two types of topology error

1

closed open

2

1 2

Line Exclusion/

Inclusion Bus Split/

Merging

2’

I Attack scenario [4, Kim et al., 2013]

1

Injection sensor

2

, ,

1 2

0 0 ,

,

(a) Without attack

(b) With attack (line exclusion)

Flow sensor

Circuit breaker sensor (“Closed=0”)

[Continuous data sensor] [Discrete data sensor]

Circuit breaker sensor (“Open=1”)

37 / 51

Topology Data Attack Manipulates LMP

minpi

∑

i∈G

Ci · pi

s.t.

λ(za) :

Nb∑

n=1

Pgn =

Nb∑

n=1

Ldn

τ (za) : pmini ≤ pi ≤ pmax

i ∀i ∈ G

µ(za) : Fminl ≤

Nb∑

n=1

Hl,n(za)(Pgn − Ldn) ≤ Fmaxl ∀l = 1, . . . ,Nl

I Domino effect: za ⇒ Hl,n(za) ⇒ {λ(za),µ(za)} ⇒ LMP(za)

LMP(za) = λ(za)1Nb− H(za)

T[µmax(za)− µmin(za)]

38 / 51

LMP Sensitivity Analysis to Topology Error

Proposition 1 (A Closed-Form Shadow Price Expression)

The shadow price µl for the congested transmission line l :

µl =∆C (j , i)

∆Hl(i , j)

where

∆C (j , i) = Cj − Ci : Marginal Unit Energy Costs Difference

∆Hl(i , j) = Hl ,i − Hl ,j : Distribution Factors Difference

39 / 51

LMP Sensitivity Analysis to Topology Error (cont’d)

Corollary 2 (A Closed-Form LMP Sensitivity Index to Topology Error)

LMP sensitivity index with respect to the line k status error (k 6= l):

∆LMPkl = ∆C (j , i)vkl

where

∆LMPkl =

[∆LMPk

l ,1, . . . ,∆LMPkl ,Nb

]T

vkl = [vkl ,1, . . . , vkl ,Nb

]T , vkl ,n =Hkl ,n

∆Hkl (i , j)

−Hl ,n

∆Hl(i , j)

I Benefit: less computational time than exhaustive numerical simulations

40 / 51

LMP Sensitivity Analysis to Topology Error (cont’d)

Corollary 3

(a) vkl ,n > 0 ⇒ decreasing LMP at bus n with topology errorI A quick prediction of post-LMP direction by topology error

(b) |vkl ,n| > |vkl ,m| ⇒ LMP sensitivity at bus n is higher than at bus mI A quick comparison of LMP sensitivity magnitude

(c) Increasing ∆C (j , i) ⇒ increasing LMP sensitivity at any busI Guidelines for a bidding strategy of generation company

41 / 51

Simulation Setup

1

2

3

4

5

6

7 8

9

10

14

11

12 13

closed

open

open: line 4-5 exclusion : congested line 5-6

Figure : IEEE 14-bus system includingbus-breaker model.

Table : Generator parameters ofthe IEEE 14-bus system.

Bus Pmin Pmax Marginal Cost

1 0MW 330MW 30$/MWh2 0MW 140MW 20$/MWh3 0MW 100MW 40$/MWh6 0MW 100MW 55$/MWh8 0MW 100MW 60$/MWh

Line 5-6 is congested

Line 4-5 is excluded dueto data corruption

42 / 51

Topology Errors Significantly Change LMPs

1 2 3 4 5 6 7 8 9 10 11 12 13 1420

40

60

80

100

120

140

Bus Location

LMP

($/M

Wh)

Without Topology ChangeWith Topology Change

(a)

1 2 3 4 5 6 7 8 9 10 11 12 13 14−15

−10

−5

0

5

10

15

20

25

30

35

Bus LocationLM

P S

ensi

tivity

($/M

Wh)

SCED ResultAnalytical Result

(b)

43 / 51

LMP Sensitivities Under the Same Line 5-6 Congestion

1 2 3 4 5 6 7 8 9 10 11 12 13 14−4

−3

−2

−1

0

1

2

3

4

5

6

Bus Location

LMP

Sen

sitiv

ity($

/MW

h)

1−2 exclusion

1−5 exclusion

2−4 exclusion

6−12 exclusion

The most sensitivebus 6

The most impactingline/CBs

The highest sensitivity at bus 6 to line 1-2, 1-5 and 2-4 exclusions

Line 1-2 exclusion (blue plot) changes sensitivities the most

44 / 51

Part II: Remarks

Main Contributions1 New analytical frameworks to study real-time LMP sensitivity

with respect to data corruption

2 Derivation of closed-form LMP sensitivity analysis toolsI economically sensitive buses to data corruptionI influential sensors and transmission lines on LMP change

3 Easily integrated with the existing EMS/MMS

45 / 51

Conclusions

I Impact of Data Integrity/Quality on Economic Dispatch

Bad/Malicious

Data

State

Estimation

Economic

Dispatch

?

Part I

I Data Attack on Look-Ahead Dispatch

A market participant’s perspective

Feasible ramp-induced data (RID)attack strategy for:

I UndetectabilityI Profitability

Part II

I LMP Sensitivity to Data Corruption

A system operator’s perspective

Analytical tools for LMP sensitivityquantification with respect to:

I Continuous Data CorruptionI Discrete Data Corruption

46 / 51

The Bigger Picture

Data Quality, Integrity, Privacy-Aware Multi-Scale Decision Making Tool

MMS

EMS

DMS

µGEMS BEMS HEMS

Current

Research - Data quality

- Data integrity

Future

Research - Data quality

- Data integrity

- Data privacy

Sensor

DMS: Distribution Management System, µGEMS: Microgrid Energy Management System

BEMS: Building Energy Management System, HEMS: Home Energy Management System

47 / 51

Multidisciplinary Approach to Future Work

I A Unified System-Wide Monitoring Tool for Multi-Scale SpatialData Quality Analysis

Smart Meter Solar Power Electric Vehicle Energy Storage

Design of interface between EMS and DMS

Performance index for the impact analysis of distribution data quality

Power system engineering, operations research/optimization

I Smart Grid Cyber Security and Privacy

Data integrity attack modeling and countermeasures

Smart meter data privacy-preserving algorithm

Power system engineering, computer networking, cyber security,statistical signal processing

48 / 51

References

[1] Y. Liu, M. K. Reiter, and P. Ning, “False data injection attacks against state estimationin electric power grids,” Proc. 16th ACM Conf. Comput. Commun. Security, 2009.

[2] O. Kosut, L. Jia, R. Thomas, and L. Tong, “Malicious data attacks on smart grid stateestimation: Attack strategies and countermeasures,” 2010 First International Conference

on Smart Grid Communications, October 2010.

[3] T. T. Kim, and H. V. .Poor, “Strategic Protection Against Data Injection Attacks onPower Grids,” IEEE Trans. Smart Grid, vol. 2, no. 2, pp. 326–333, May 2011.

[4] J. Kim and L. Tong, “On Topology Attack of a Smart Grid: Undetectable Attacks andCounter Measures,” IEEE J. Selected Areas in Communications, July 2013.

[5] L. Xie, Y. Mo, and B. Sinopoli, “Integrity Data Attacks in power market operations,”IEEE Trans. Smart Grid, vol. 2, no. 4, pp. 659–666, December 2011.

[6] A. J. Conejo, E. Castillo, R. Minguez, and F. Milano, “Locational marginal pricesensitivities,” IEEE Trans. Power Syst, vol. 20, no. 4, pp. 2026–2033, November 2005.

[7] F. Li and R. Bo, “DCOPF-based LMP simulation: Algorithm, comparison with ACOPF,and sensitivity,” IEEE Trans. Power Syst, vol. 22, no. 4, pp. 1475–1485, November 2007.

49 / 51

Publications (During Ph.D. Study)

I Journal

[1] D.-H Choi and L. Xie, “Sensitivity Analysis of Real-Time LocationalMarginal Price to SCADA Sensor Data

Corruption,” IEEE Trans. Power Syst (accepted).

[2] D.-H Choi and L. Xie, “Ramp-Induced Data Attacks on Look-ahead Dispatch in Real-time Power Markets,” IEEE

Trans. Smart Grid, vol. 4, no. 3, pp. 1235–1243, September 2013.

[3] L. Xie, D.-H. Choi, S. Kar and H. Vincent Poor “Fully Distributed State Estimation for Wide-Area Monitoring

Systems,” IEEE Trans. Smart Grid, vol. 3, no. 3, pp. 1154–1169, September 2012.

[4] S. Wang, L. Cui, J. Que, D.-H. Choi, X. Jiang, S. Cheng and L. Xie “A Randomized Response Model for Privacy

Preserving Smart Metering,” IEEE Trans. Smart Grid, vol. 3, no. 3, pp. 1154–1169, September 2012.

I Conference Proceedings

[5] D.-H Choi and L. Xie, “Impact Analysis of Locational Marginal Price Subject to Power System Topology Errors,”

2013 Fourth International Conference on Smart Grid Communications, October 2013.

[6] D.-H Choi and L. Xie, “Malicious Ramp-Induced Temporal Data Attack in Power Market with Look-Ahead Dispatch”

2012 Third International Conference on Smart Grid Communications, November 2012 (The Best Paper Award).

[7] D.-H Choi and L. Xie, “Fully Distributed Bad Data Processing for Wide Area State Estimation,” 2011 Second

International Conference on Smart Grid Communications, October 2011.

[8] L. Xie, D.-H. Choi and S. Kar, “Cooperative distributed state estimation: Local observability relaxed,” Proc. IEEE

Power and Energy Society General Meeting, Detroit, 2011.

I Book Chapter

[9] L. Xie, D.-H. Choi, S. Kar and H. Vincent Poor “Bad/malicious data detection in distributed power system state

estimation,” in E. Hossain, Z. Han, and H. V. Poor, editors, Smart Grid Communications and Networking, CambridgeUniversity Press, 2012 (to appear).

50 / 51

Thank You!

51 / 51