Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
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