1. Concept of a Wide Area Defense System for the Power Grid
Chen-Ching Liu
University College Dublin
National University of Ireland, Dublin
Seminar at NTUA and IEEE PES Greece, June 2009

2. Catastrophic Power Outages

3. Western Electricity Coordinating Council (WECC) system - Aug 10th, 1996 Blackout
Initiating events
System becomes unstable
Blackout
15:48 p.m. Keeler-Allston 500-kV line contacts a tree due to inadequate right-of-way maintenance. Additionally the Pearl-Keeler line is forced out of service due to the Keeler 500/230-Kv transformer being OOS.
Mild .224 Hz oscillations were seen throughout the system and began to appear on of the PDCI.
The WECC broke into 4 asynchronous islands with heavy loss of load.
7.5 million people lost power.
Shunt capacitor banks were switched in to raise the voltage but the oscillations were not being damped.
With the loss of these 2 lines, 5 lines are now out of service, removing hundreds of MVAR.
15:48:51 p.m. Oscillations on the POI reached 1000MW and 60-kV peak-to-peak. WA MT ND
Lines throughout the system begin to experience overloads as well as low voltage conditions. Additional lines trip due to sagging. OR
PDCI Remedial Action Schemes (RAS) began to actuate. Shunt and series capacitors were inserted. ID SD WY NV NE CA UT CO
15:47:40-15:48:57 p.m. Generators at the McNary power house supplying 494 MVAR trip. The system begins to experience “mild oscillations”. AZ NM

4. Eastern Interconnection – August 14th, 2003 Blackout
Initiating events
System becomes unstable
Blackout
1:07 p.m. FE turns off their state estimator for troubleshooting.
4:05:57 p.m. The loss of 138-kV lines overloads the Sammis-Star line.
265 power plants tripped off line and 50 million people are without power.
1:31 p.m. Eastlake 5 generation unit trips and shuts down.
2:02 p.m. Stuart-Atlanta 345-kV line tips off due to contact with a tree.
4:08:59 p.m. Galion-Ohio and Central-Muskinghun 345-kV lines trip on Zone 3 causing major power swings through New York and Ontario and into Michigan.
2:14 p.m. FE’s control room lost alarm functions followed by a number of the EMS consoles.
Low voltage/ high load conditions and system disturbances propagate through the system tripping transmission lines and generators. ND ME MN VT MI NH SD WI NY MA
2:54 p.m. The primary and secondary alarms servers failed. CT RI NE IA PA IL OH NJ IN MD DE KS WV MO VA
3:05:41-3:57:35 p.m kV lines trip due to contact with trees. This overloads the underlying 138-kV system and depressed voltages.
4:13 p.m. most of the North East and parts of Canada blacked out. There are only a few islands which remain operating. KY NC OK TN AR SC MS AL GA LA FL
3:39:17-4:08:59 p.m kV lines trip due to overloading.

5. Hydro-Québec Blackout- April 18th, 1988

6. Blackout Propagation (without defense systems)
Complete System Collapse
Triggering Event

7. Occurrences and Extent of Blackouts in North America
1000
Number of Blackouts
100 10
100
102
104
106
108
Number of customers affected

8. Strategic Power Infrastructure Defense (SPID)
Design self-healing strategies and adaptive reconfiguration schemes
To achieve autonomous, adaptive, and preventive remedial control actions
To provide adaptive/intelligent protection
To minimize the impact of power system vulnerability

9. SPID System Fast and on-line power & comm. system assessment
Hidden failure monitoring
Adaptive: load shedding, generation rejection, islanding, protection

10. Multi-Agent System for SPID
Knowledge/Decision exchange
Protection Agents
Generation
Agents
Fault
Isolation Agents
Frequency
Stability
Model
Update
Command
Interpretation
Planning
Agent
Restoration
Hidden
Failure
Monitoring Agents
Reconfiguration
Vulnerability
Assessment
Power System
Controls
Inhibition Signal
Plans/Decisions
Event
Identification
Triggering Events
Event/Alarm
Filtering
Events/Alarms
Inputs
Update Model
Check
Consistency
Comm.
DELIBERATIVE LAYER
COORDINATION LAYER
REACTIVE LAYER

11. Multi-Agent System for SPID
Subsumption Architecture (Brooks) for Coordination
Agents in the higher layer can block the control actions of agents in lower layers
Load Shedding
Agent
Global View/Goal(s) R
Under Frequency Relay
Local View/Goal(s)
Tripping Signal
Inhibition Signal

12. combined into sequences
Cascaded Events
Some Basic Patterns
Line tripping due to overloading
Generator tripping due to over-excitation
Line tripping due to loss of synchronism
Generator tripping due to abnormal voltage
and frequency system condition
Under-frequency/voltage load shedding
Identifying the basic patterns of cascaded events and explore how these patterns can be
combined into sequences

13. Cascaded Zone 3 Operations
Zone 3 Relay Operations Contributed to Causes of Blackouts.
Heavy Loaded Line
Low Voltage
High Current
Zone 3
Relay
Operation(s)
Lower Impedance Seen by Relay
Loss of Transmission Lines
Other Heavy Loaded Lines
Catastrophic Outage

14. Prediction of Zone 3 Relay Tripping Based on On-Line Steady State Security Assessment
Contingency Evaluation Performed On Line Every Several Minutes
Contingency Evaluation …
Case
Relay
Status
Contingency Description 1
N/A
Secure
3 phase fault at bus 1 2
Zone3
Insecure
3 phase fault at bus 2 .. … N
3 phase fault at bus N
Case
Relay
Status
Contingency Description 1
N/A
Secure
3 phase fault at bus 1 2
Zone3
Insecure
3 phase fault at bus 2 .. … N
3 phase fault at bus N
Case
Relay
Status
Contingency Description 1
N/A
Secure
3 phase fault at bus 1 2
Zone3
Insecure
3 phase fault at bus 2 .. … N
3 phase fault at bus N
Case
Relay
Status
Contingency Description 1
N/A
Secure
3 phase fault at bus 1 2
Zone 3
Insecure
3 phase fault at bus 2 .. … N
3 phase fault at bus N
Post-Contingency
Power Flow
FIS
Post-Contingency
Apparent Impedance
Fuzzy Inference System (FIS) Developed
Using Off-Line Time-Domain Simulations
Corrected Post-Contingency
Apparent Impedance

15. Impedances Obtained by Power Flow and Time Domain Simulation
Post-Contingency Impedance Obtained by Power Flow Does Not Coincide with Impedance Obtained by Time-Domain Dynamic Simulations

16. Automatic Development of Fuzzy Rule Base
Wang & Mendel’s algorithm is a “learning” algorithm:
1) One can combine measured information and human linguistic information into a common framework
2) Simple and straightforward one-pass build up procedure
3) There is flexibility in choosing the membership function
Pre-determine number of
membership functions N
Give input and output
data sets
FIS
created
automatically
(Inp1, Inp2, Out) = (10, 1, -2)
(Inp1, Inp2, Out) = (8, 3, -1)
(Inp1, Inp2, Out) = (5, 6, -4)
(Inp1, Inp2, Out) = (2, 8, -5) …
In this example, N is 7

17. Case A
Fault Location
Relay Location

18. Impedance on R-X (Case A)
Z obtained by power flow solution is outside Zone 3 circle.
Case A

19. Load Shedding
Studies have shown that the August 10th 1996 blackout could have been prevented if just 0.4% of the total system load had been dropped for 30 minutes.
According to the Final NERC Report on August 14, 2003, Blackout, at least 1,500 to 2,500 MW of load in Cleveland-Akron area had to be shed, prior to the loss of the 345-kV Sammis-Star line, to prevent the blackout.

20. Automatic Load Shedding
Under Voltage
Under Frequency
Rate of Frequency Decrease
Remedial Action Scheme

21. Adaptive Self-healing: Load Shedding Agent
A control action might fail
Unsupervised adaptive learning method should be deployed
Reinforcement Learning
Autonomous learning method based on interactions with the agent’s environment
If an action is followed by a satisfactory state, the tendency to produce the action is strengthened

22. Load Shedding Options
frequency
Time (multiples of 0.02 sec)

23. Adaptive Self-healing: Load Shedding Agent
179 bus system resembling WSCC system
ETMSP simulation
Remote load shedding scheme based on frequency decline + frequency decline rate
Temporal Difference (TD) method is used for adaptation: Need to find the learning factor for convergence

24. Adaptive Load Shedding Agent
State 1
State 2
State 3
Freq := 59.5
Dec.rate > threshold value
Freq := 58.8
Freq := 58.6
179 bus system

25. Adaptive Load Shedding Agent
Expected normalized system frequency that makes the system stable
Normalized frequency
“The load shedding agent is able to find the proper control action in an adaptive manner based on responses from the power system”
Number of trials

26. Flexible Grid Configuration to Enhance Robustness
Flexible Grid Configuration can play a significant role in defending against catastrophic events.
Power infrastructure must be more intelligent and flexible.
To allow coordinated operation and control measures to absorb the shock and minimize the potential damages caused by radical events .

27. Cascading Events
A Cascading Event Refers to a Series of Tripping Initiated by One or Several Component Failures in a Power System
Here the initial component(s) failure is designated as “shock” to the power infrastructure

28. Simulated Cascading Events (179 Bus System)
Compute Power Flows after Tripping
Six lines are found on limit violation
Trip these lines
Identify New Network Configuration and Solve Power Flows Again
Fifteen lines are found with limit violations
Continue This Simulation Procedure
Finally system collapses: most transmission lines are tripped and most loads are lost

29. 2-Area Partitioning Algorithm (from VLSI)
Spectral 2-way Ratio-Cut Partitioning
Theorem Given an edge-weighted graph G = (V, E), the second smallest eigenvalue λ2 of the graph’s Laplacian matrix Q yields a lower bound on the cost c of the optimal ratio cut partition, with c = e(U,W)/(|U|·|W|) ≥ (λ2/n)
Cut-Size: e(U,W) ≥ (λ2/n) (|U|·|W|)

30. Area-Partitioning Algorithm
6-Bus System

31. Area-Partitioning Algorithm
Partition {U, W}
Cut Set Size e(U,W)
e(U,W) / (|U|·|W|)
λ2 / n
{(4), ( )}
1.1
1.1 / (1  5) = 0.22
{(4 5), ( )}
1.2
1.2 / (2  4) = 0.15
{(4 5 6), (3 2 1)}
0.3
0.3 / (3  3) =
/ 6 =
{( ), (2 1)}
2.1
2.1 / (4  2) =
{( ), (1)}
1.1 / (5  1) = 0.22
Partition Results of 6-Bus System

32. Flexible Grid Configuration to Absorb the Shock
Solve Power Flows of Area One
All MW loads are supplied, no line flow constraints violations
Solve Power Flows of Area Two
Seven lines on limit violation:
(Bus158-Bus164), (163-8), (64-163), double lines (16-19), and double lines ( )

33. Flexible Grid Configuration to Absorb the Shock
Use “Power Redispatching & Load Shedding” in Area Two
Totally, = MW load are shed
Bus #
Original Load (MW)
Load Shed (MW)
Load Supplied (MW) 8
239
188 51 16
793.4
64.4
729
154
1066 60
1006

34. Split System into Two Areas

35. Flexible Grid Configuration to Absorb the Shock
Shed Load vs. System Total Load
K=1
K=2
K=3 

36. Flexible Grid Configuration to Avoid Cascaded Failures
Step 1 : Compute power flows after initial tripping event(s).
Step 2 : Convert power network to an edge-weighted graph G, weight of each edge is absolute value of real power flow.
Step 3: Multilevel graph partitioning with minimum edge-cut.
Network is separated into k areas to minimize generation / load imbalance.
Graph COARSENED to a smaller number of vertices,
Bisection PARTITIONING of much smaller graph,
UNCOARSENING back toward original graph.
Coarsest graph small, Coarsening can be parallelized, Partitioning efficiency high.

37. Emergency Control with Multilevel Graph Partitioning
Partitioning a 22,000 bus and 32,749 branch system into 2, 3, 4 islands with 0.07s, 0.081s and 0.09s on 2 GHz Pentium CPU and 1GB RAM.
Fast computational speed makes it possible to determine partitioning strategy and identify new network configuration in on-line environment.
An adaptive relaying architecture for controlled islanding
CCUs acquires system data, generates system separation strategy.
SCUs receive system separation commands from CCU and send breaker opening commands to specific auxiliary relays.

38. Flexible Grid Configuration to Avoid Cascaded Failures
Controlled islanding on a 200-bus system:
199 buses, 31 generators, 248 branches.
Sequence of cascading events:
At t=0 s, three transmission lines out-of-service.
At t=60 s, line tripped due to line fault.
At t=120 s, line tripped due to line fault. Generator G70 at bus 70 overloaded.
At t=240 s, generator G70 tripped by over-excitation protection. At t= 260 second, system collapses.
Cut Set
Load-Generation (MW)
Bus , , , , , (1,2)
North: Gen=37862 , Load=37104
South: Gen=24517, Load=23794

39. Flexible Grid Configuration to Avoid Cascaded Failures
Load bus voltages without/with islanding strategy
System islanding initiated at 241s.
Islanding strategy results in balanced generation / load in both islands.
All loads in the system are served. Load shedding scheme not applied.
Islanding strategy successfully prevents the collapse of the system.

40. Conclusion Cascading failures remain a grand challenge
New communication, information and computer technologies enable wide area protection and control
Connectivity also brings cyber vulnerability
“Smart” grid?
“Self-healing” grid?

41. Further Information
J. Li, C. C. Liu, “Power System Reconfiguration Based on Multilevel Graph Partitioning,” IEEE PES Power Tech, Bucharest, Romania, 2009.
K. Yamashita, J. Li, C. C. Liu, P. Zhang, and M. Hafmann, “Learning to Recognize Vulnerable Patterns Due to Undesirable Zone-3 Relay Operations,” IEEJ Trans. Electrical and Electronic Engineering, May 2009, pp
J. Li, K. Yamashita, C. C. Liu, P. Zhang, M. Hoffmann, “Identification of Cascaded Generator Over-Excitation Tripping Events,” PSCC, Glasgow, U.K., 2008.
H. Li, G. Rosenwald, J. Jung, and C. C. Liu, “Strategic Power Infrastructure Defense,” Proceedings of the IEEE, May 2005, pp
J. Jung, C. C. Liu, S. Tanimoto, and V. Vittal, “Adaptation in Load Shedding under Vulnerable Operating Conditions,” IEEE Trans. Power Systems, Nov. 2002, pp
H. You, V. Vittal, J. Jung, C. C. Liu, M. Amin, and R. Adapa, “An Intelligent Adaptive Load Shedding Scheme,” Proc PSCC, Seville, Spain, June 2002.
C. C. Liu, J. Jung, G. Heydt, V. Vittal, and A. Phadke, “Strategic Power Infrastructure Defense (SPID) System: A Conceptual Design,” IEEE Control Systems Magazine, Aug. 2000, pp