Top Banner
Security of Smart Grids: A Cyber‐Physical Perspec:ve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon 1 TexPoint fonts used in EMF. CyLab Silicon Valley Briefing March 25, 2011
52

Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Feb 07, 2020

Download

Documents

dariahiddleston
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Security of Smart Grids: A Cyber‐Physical Perspec:ve �

Bruno Sinopoli Assistant Professor 

Department of ECE Carnegie Mellon �

1

TexPoint fonts used in EMF.

CyLab Silicon Valley Briefing March 25, 2011

Page 2: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

The smart grid 

Page 3: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

From a smart grid to a smarter grid

•  Integration of

3

Page 4: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Is it a worthwhile effort?

•  Pros –  Efficiency –  Safety –  Green –  Competitiveness

•  Cons –  Cost –  Complexity –  Vulnerability

4

Page 5: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

What are Cyber‐Physical Systems? 

Computing

Control Communication Cyber Physical

Page 6: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

6

Cyber vs Cyber‐Physical Security 

•  Key goals of informa;on security: –  Confiden;ality: aAacker cannot read data packets. –  Integrity: aAacker cannot modify data packets. –  Availability: data packets are available for es;ma;on and control purpose. 

–  Etc.. •  Key goal of CPS security: 

–  Guaranteeing reliable system opera;on 

•  Cyber security is a tool not a goal 

Page 7: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Goal/Scope of the attack in CPS •  Disrupt operations, e.g. destabilize the

system (e.g. Stuxnet) •  Reduce system’s performance •  Financial gain •  Context

– Cyber warfare – Commercial advantage – Criminal intent

7

Page 8: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Types of CPS Attacks/ Remediation

•  Attacks –  Cyber range of attacks –  Physical Attacks –  Insider attacks

•  Remediation –  Detection/isolation –  Guarantee continuity of operation –  Graceful degradation –  Service restoration

8

Page 9: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Today’s talk: provide some insights via case studies

•  System definition – Focus on control systems

•  Attacks on sensors – Analysis of Sensor Replay attacks – Analysis of Integrity attacks on sensors

•  Examples •  Conclusion

9

Page 10: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

System model�

•  We model the underlying physical system as a linear ;me‐invariant system: 

•  Sensors are used to monitor the system: 

•  Each element in      represents the reading of a certain sensor at ;me    . �

10

xk+1 = Axk + wk

yk = Cxk + vk

yk

k

Page 11: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Illustra:ve Example�

•  We consider a vehicle moving along the - axis.

•  Two sensors are used to measure position and velocity respectively.

11

x

xk+1 = xk + wk,1,

xk+1 = xk + xk + wk,2

yk,1 = xk + vk,1,

yk,2 = xk + vk,2.

x

Page 12: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

12

Kalman Filter and LQG controller 

Page 13: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Failure Detector �

•  A failure detector is used to detect abnormality in the system, which triggers an alarm based on the following condi;on: 

where 

and the func;on     is con;nuous. 

13

gk > threshold

gk = g(yk, xk, . . . , yk!T , xk!T ),

g

Page 14: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Failure Detector �

•  For example,       for a chi‐square detector takes the following form: 

where  

and        is the covariance of      . 

14

gk

zk = yk ! CAxk!1,

P zk

gk = zTk P!1zk

Page 15: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

15

Replay AJack Model (Allerton conf. ‘09) 

•  The aAacker can –  Record and modify the sensors’ readings  –  Inject malicious control input 

•  Replay AAack –  Record sufficient number of       without adding control inputs. 

–  Inject malicious control input to the system and replay the previous  .    . We denote the replayed measurements to be       . 

•  When replay begins, there is no informa;on from the systems to the controller. As a result, the controller cannot guarantee any close‐loop control performance. The only chance is to detect the replay. 

yk

yk

yk

y!k

Page 16: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Our Abstract

16

Page 17: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

16 months later…

17

Page 18: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

18

System Diagram 

Page 19: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

19

Simula:on •  Suppose the aAacker records from ;me –T and replay 

begins at ;me 0. 

•  For some systems, the Chi2 detector cannot dis;nguish system under replay and system without replay. 

Page 20: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

20

Detec:on of Replay AJack �

•  Manipula;ng equa;ons: 

•  If          converges to 0 very fast, then there is no way to dis;nguish the compromised system and healthy system. Ak

Page 21: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

21

Counter Measure 

•  Replay is feasible because the op;mal es;mator and controller are determinis;c 

•  If we add random control input to the system: –  If the system responds to this input, then there is no replay 

–  If not, then there is a replay –  Random control inputs act like ;me stamps –  Cost: The controller is not op;mal any more 

Page 22: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

22

Counter Measure 

•  Let control input to be 

 where        is the op;mal control input,       is an i.i.d. Gaussian random control input with zero mean and covariance of     . can be seen as an authen;ca;on signal  

•  The increase in control cost is given by 

uk = u!k + !uk,

u!k !uk

Q

trace!(U + BT SB)Q

"

!uk

Page 23: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

23

Counter Measure 

•  Innova;on with random input: 

Page 24: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

24

New System Diagram

Page 25: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

25

Simula:on Result 

•  One dimensional system, single sensor: 

•  Parameters: –  R = 0.1, Q = 1 –  W = U =1 

–  Detector window size 5, false alarm rate 5% 

xk+1 = xk + uk + wk,

yk = xk + vk.

Page 26: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

26 Detec;on Rate of Different Random Signal Strength 

10 11 12 13 14 15 16 17 18 19 200

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Time(k)

De

tec

tio

n R

ate

Page 27: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Chemical Plant (A + C → D)

Objectives: Maintain production rate by controlling valves Minimize operating cost (function of purge loss of A and C)

Restrictions:

Page 28: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Regular vs. Secure controller

Time for detection = 25 ms

Page 29: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Integrity AJack strategy�

•  The aAacker has full knowledge of the system’s model. 

•  The aAacker can change the readings of a subset of sensors. 

•  The goals of the aAacker are: –  To affect the system’s opera;ons;  –  Not being detected. 

29

y!k = Cx!

k + vk + !yak

Page 30: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Ques:ons�

•  Can the aAacker successfully destabilize the system?�

•  If not what is the extent of the perturba;on that the aAacker can inflict to the system? 

30

Page 31: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Integrity AJack Model 

•  An aAack sequence      is defined as an infinite sequence of the aAacker’s input  

•  The innova;on is defined as 

•  An aAack sequence     is call              feasible if the following condi;ons hold from ;me 0 to ;me T: 

31

Yya0 , ya

1 , . . .

zk = yk ! CAxk, z!k = y!

k ! CAx!k

Y (T, !)

12(z!k ! zk)TP"1(z!k ! zk)T " !

Page 32: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Reachable Set 

•  Define es;ma;on error as: 

•  Define the bias introduced by the aAacker as: 

•  The     reachable region is defined as: 

•  The reachable region is defined as: 

32

k

!ek = e!k ! ek.

Rk = {x ! Rn : x = !ek(Y), and Y is (k, 1) feasible}.

R =!!

k=1

Rk.

ek = xk ! xk, e!k = x!

k ! x!k

Page 33: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Which sensors should I aJack/protect?�

•  To check the resilience of control system, one can find all the unstable eigenvector of A and compute Cv. 

•  If Cv is sparse, then the aAacker only need to compromise a few sensors to launch an aAack along the direc;on v. 

•  To improve the resilience, the defender could add redundant sensors to measure every unstable mode. 

33

The reachable region R is unbounded if and only if A has an unstable eigen-value and the corresponding eigenvector v satisfies:

1. Cv ! span(!).

2. v is reachable for dynamic system "ek+1 = (A"KCA)"ek "K!yak+1.

Page 34: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Resilient systems allow only a finite reachable set �

•  In general compu;ng the reachable set is very hard, since the number of inequali;es needed to describe the set quickly explodes.  

•  As a result, we use ellipsoids to approximate the reachable region. 

34

Page 35: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Illustra:ve Example�

•  We consider a car moving along the - axis.

•  Two sensors are used to measure position and velocity respectively.

•  We assume that . 35

x

Q = R = I2

xk+1 = xk + wk,1,

xk+1 = xk + xk + wk,2

yk,1 = xk + vk,1,

yk,2 = xk + vk,2.

Page 36: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Posi:on sensor is compromised 

36

Page 37: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Simula:on Result: Compromising the Posi:on Sensor�

37

Page 38: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Velocity Sensor is compromised�

38

Page 39: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

An applica:on: Electricity Market pricing�

•  The price of electricity is determined by the state es;ma;on , i.e. genera;on, power flow over transmission and load of the power grid.  

•  If an aAacker was able to compromise some sensors, then it could introduce a bias in the state es;ma;on accordingly. 

•  Eventually, over a finite ;me‐horizon, the aAacker will affect the pricing to his advantage and make a profit. �

39

Page 40: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Day‐Ahead Forward Market and Virtual Bidding�

•  The Regional Transmission Organiza;on (RTO) computes the nodal price based on the predicted load. 

•  The price is published usually 36 hours before actual opera;on. 

•  A market par;cipant could buy/sell virtual power at loca;on j in the day‐ahead market, and is obliged to sell/buy the same amount of power at the same loca;on in the real ;me market.   �

40

Page 41: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Ex‐Post Market (Real Time Market)�

•  A transmission line is posi;vely congested if                       . It is nega;vely congested if                       . 

•  In the real market, the RTO tries to solve the following minimiza;on problem:        �

41

Fl > Fmaxl

Fl < Fminl

minimize!Pgi

I!

i=1

Ci!Pgi

subject toI!

i=1

!Pgi = 0

!Pgmini ! !Pgi ! !Pgmax

i "i = 1, ..., I

!Fl ! 0 "l # cl+

!Fl $ 0 "l # cl!,

Page 42: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Ex‐Post Market (Real‐Time Market)�

•  The Lagrangian of the above minimiza;on problem is defined as �

42

L =I!

i=1

Ci(!Pgi + P g(i))! !I!

i=1

!Pgi

+I!

i=1

µi,max(!Pgi !!Pgmaxi )

+I!

i=1

µi,min(!Pgmini !!Pgi)

+!

l!cl+

"l!Fl +!

l!cl!

#l(!!Fl).

Page 43: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Ex‐Post Market (Real Time Market)�

•  The nodal price at point j is given by 

43

!j = ! +L!

l=1

("l ! #l)$Fl

$Ldj.

Page 44: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Profitability 

•  The nodal loca;onal marginal price (LMP) difference is caused by conges;ons in the transmission line. 

•  Given two node     and     , depending on the power distribu;on matrix,  we could classify the transmission lines into three categories: 

•  If no line in        (      ) is posi;vely(nega;vely) congested, then the price at      will be greater than the price at    . 

j1 j2

L!, L0, L+.

j1L!L+

j2

Page 45: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Profitability 

•  The aAacker first buy/sell at the day ahead market at loca;on       and       ,     units of virtual power, with price                    . Assume that  

•  In the Ex‐post market, sell/buy at the same loca;on, with price             . 

•  Manipula;ng the state es;ma;on to ensure: 

•  The total profit is

!DA1 < !DA

2 .

!1 > !2.

!(!DA

2 ! !DA1 ) + (!1 ! !2)

"" p.

!DA1 , !DA

2

j1 j2 p

!1, !2

Page 46: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Attacker’s strategy

46

Page 47: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Profitability Gaithersburg ($/MWh) Pittsburgh ($/MWh)

Day Ahead Market Buy at 25 Sell at 30 Ex-Post Market without the Attack

Sell at 20 Buy at 26

Ex-Post Market under the Attack

Sell at 24 Buy at 23

•  Without the attack, the attacker could lose 1$/MWh. •  With the attack, the attacker gains 6 $/MWh.

Page 48: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Conclusion 

•  Security of cyber‐physical systems is of paramount importance 

•  Security needs to be integrated with system theory/knowledge 

•  A science of security for CPS systems needs to be developed 

•  Small aAacks that run “under the radar” can have serious consequences 

48

Page 49: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Thank You!�

49

Page 50: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Simulation Result�

50

Page 51: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Simulation Result�

51

Page 52: Security of Smart Grids: A Cyber‐Physical Perspecve · Security of Smart Grids: A Cyber‐Physical Perspecve Bruno Sinopoli Assistant Professor Department of ECE Carnegie Mellon

Simulation Result�

52