Tom Overbye Dept. of Electrical & Computer Engineering University of Illinois at Urbana-Champaign

1

PSERC

Power System Control: Enhancing the Human-System Interface

The Mathematics of August 14th 2003: How Complex?

Tom OverbyeTom Overbye

Dept. of Electrical & Computer EngineeringDept. of Electrical & Computer Engineering

University of Illinois at Urbana-ChampaignUniversity of Illinois at Urbana-Champaign

overbye@ece.uiuc.eduoverbye@ece.uiuc.edu

March 13, 2004

2

PSERC

Humans as the key link

• Some of power system operations is automated– fault detection, under & over-frequency load-

shedding, under voltage load shedding

• But degree of automation is much lower than many people assume

• Humans are very much “in the loop”

• This is particularly apparent during emergency system events

3

PSERC

August 14th 2003

• The August 14th blackout demonstrated how crucial this link can be, and the critical need for an optimized human-system interface

• This talk demonstrates several techniques for enhancing this interface, with the August 14th blackout as a motivating example

• Talk also looks at accuracy of the mathematical models for the initial August 14th events

4

PSERC

Causes of August 14th Blackout

• US-Canada Interim Report determined three groups of causes for the blackout– Inadequate situational awareness by FirstEnergy

(FE)

–FE failed to adequately manage tree growth in its transmission rights-of-way

–Failure of the grid reliability organizations to provide effective diagnostic support (mostly the Midwest Independent System Operator [MISO])

5

PSERC

NERC Reliability Coordinators

6

PSERC

Control Implications of August 14th

• From a control perspective the August 14th event lasted for over an hour– Interim report noted that prior to 15:05 EDT the

system was in a reliable operational state

–From the first event at 15:05:41 until the blackout was complete at 16:13 there were essentially no human-initiated corrective control actions• There was a lot of talk, and some were prepared to act,

but the state of the grid was almost entirely dictated by its physics and automatic controls

• Talk looks at why and how to do better

7

PSERC

Early Power System Control (in 1919)

8

PSERC

Late 1990’s Control Centers (ComED)

9

PSERC

Control Center Trend Towards Overview Displays (ComED Now)

10

PSERC

Overview of Real-time Power System Operations

• Off-line studies used to plan system dispatch

• Real-time data comes to control center via SCADA–SCADA data is displayed to operators

–user entered topology is used to calculate line outage distribution factors (LODFs)

– flowgates values determined for “critical” facilities

– flowgate overloads are curtailed by TLR (transmission load relief)

11

PSERC

Overview of Real-time Power System Operations

• State estimator (SE) uses SCADA data and a system model to calculate the system state (mostly voltages at all system buses)– the key output of SE is a system power flow model

• Power flow model is used in advanced applications, such as contingency analysis (CA), optimal power flow (OPF), and security constrained OPF (SCOPF)–SCOPF calculates bus marginal prices (LMPs)

12

PSERC

State Estimator Algorithm

• Most state estimators use a weighted least-squares approach

1min ( ) ( ) ( )

where is the measurement covariance matrix

Tmeas meas

xJ x z f x R z f x

R• Because the power system is non-linear, the SE requires an iterative

solution–advanced apps can’t run without an SE solution

– topology errors in f can cause non-convergence

13

PSERC

Power Flow Equations

• The steady-state power flow equations, which must be satisfied at each bus i, are

1

1

( cos sin )

( sin cos )

where is the bus admittance matrix

n

i k ik ik ik ik Gi Dik

n

i k ik ik ik ik Gi Dik

V V G B P P

V V G B Q Q

j

G B

• The power flow solves for the bus voltage magnitude and angle vectors, V and

14

PSERC

The DC Power Flow

• The DC power flow makes a number of approximations to greatly simplify the non-linear AC power flow–completely ignores the reactive power flow

–assumes all voltage magnitudes are one per unit (i.e., at their nominal values)

– ignores line resistive losses

– ignores tap dependence of the impedance of LTC and phase shifting transformers

15

PSERC

The DC Power Flow Equation

• With these approximations the power flow is reduced to a linear, state-independent, set of equations

1

where is the vector of bus voltage angles,

is the negative of the imaginary portion of the bus

admittance matrix, and is the vector of real power

injections

θ B P

θ B

P

16

PSERC

Power Transfer Distribution Factors (PTDFs)

• The DC power flow approximation is used extensively by NERC to calculate both PTDFs and LODFs

• PTDFs approximate the incremental impact a power transfer has on the network (i.e., how power flows from the seller to the buyer.

1

where only has non-zero entries at the

buyer/seller buses

θ B P

P

17

PSERC

PTDF Visualization of a Power Transaction from Wisconsin to TVA

18

PSERC

Line Outage Distribution Factors (LODFs)

• LODFs are used to approximate the change in the flow on one line caused by the outage of a second line– typically they are only used to determine the change in the MW flow

–LODFs are used extensively in real-time operations

–LODFs are state-independent but do dependent on the assumed network topology

,l l k kP LODF P

19

PSERC

Flowgates

• The real-time loading of the power grid is accessed via “flowgates”

• A flowgate “flow” is the real power flow on one or more transmission element for either base case conditions or a single contingency–contingent flows are determined using LODFs

• Flowgates are used as proxies for other types of limits, such as voltage or stability limits

• Flowgates are calculated using a spreadsheet

20

PSERC

Flows in Northeast Ohio at 15:00 EDT on August 14th 2003

21

PSERC

Northeast Ohio 138 kV Voltage Contour: 15:00 EDT

22

PSERC

Flowgate 2265

• Flowgate 2265 monitors the flow on FE’s Star-Juniper 345 kV line for contingent loss of the Hanna-Juniper 345 Line–normally the LODF for this flowgate is 0.361

– flowgate has a limit of 1080 MW

–at 15:05 EDT the flow as 517 MW on Star-Juniper, 1004 MW on Hanna-Juniper, giving a flowgate value of 520+0.361*1007=884 (82%)

–Chamberlin-Harding 345 opened at 15:05; FE and MISO all missed seeing this

23

PSERC

Flowgate #2265

• At 15:10 EDT (after loss of Chamberlin-Harding 345) #2265 an incorrect value because its LODF was not automatically updated. –Value should be 633+0.463*1174=1176 (109%)

–Value was 633 + 0.361*1174=1057 (98%)

• At 15:32 the flowgate’s contingent line opened, causing the flowgate to again show the correct value, about 107%

24

PSERC

Flows in Northeast Ohio at 15:33 EDT on August 14th 2003

25

PSERC

Northeast Ohio 138 kV Voltage Contour: 15:33 EDT

26

PSERC

Flows in Northeast Ohio at 15:46 EDT on August 14th 2003

27

PSERC

Northeast Ohio 138 kV Voltage Contour: 15:46 EDT

28

PSERC

Flows in Northeast Ohio at 16:05 EDT on August 14th 2003

29

PSERC

Northeast Ohio 138 kV Voltage Contour: 16:05 EDT

30

PSERC

Are DC LODFs Accurate?August 14th Crash Test

• Here are some results from August 14th Time Contingency Element LODF MW (pred) MW (act)

15:05 Chamberlin-Harding 345

Hanna-Juniper 345

0.362 179 176

15:32 Hanna-Juniper 345 Star-Juniper 345

0.465 545 527

15:46 CantonCentral-Cloverdale 138

Sammis-Star 345

0.164 48 54

15:46 same Cloverdale-Star138

0.234 68 64

16:06 Sammis-Star 345Star-Urban 138W.Canton-Dale 138

Star-Juniper345

numerous 517 676

16:06 same Ashtabula-Perry 345

numerous 319 408

31

PSERC

The Results are Actually Quite Good!

• The initial LODF values were accurate to within a few percent

• Even after more than a dozen contingencies, with many voltages well below 0.9 pu, the purely DC LODF analysis was giving fairly good (with 25%) results

32

PSERC

System was Well Behaved

• Until the cascade began at about 16:10 the system was actually quite well behaved mathematically

• How the flow redistributed through the system could have been well predicted by essentially linear means

• Of course, once the cascade started (after more than a dozen contingencies) the dynamics got to be quite complex

33

PSERC

What was missing on August 14th?

• The key missing ingredient on August 14th was a high level view of the system

• Even though SCADA measurements were available, FE, MISO, PJM and AEP did not have a good view of what was happening on the grid, particularly outside of their areas of control/oversight

• Next few slides show some techniques for providing this view

34

PSERC

System with Dynamic Sized Pie Charts used to Indicate Loading

35

PSERC

Contouring

• Contours can be effective for showing large amounts of spatial data–weather maps showing temperatures and weather

radar images provide good examples

–potential power system applications• bus voltage magnitudes and LMPs

• percent loading and PTDFs on transmission lines

• flowgate values

–personally, I think discrete contours are best

36

PSERC

Continuous Contour of Bus LMPs

37

PSERC

Discrete Contour of Bus LMPs

38

PSERC

Interactive 3D Visualization

• Starting point is to re-map traditional one-line into 3D–builds upon the traditional 2D one-line, familiar

to power system users

–existing one-lines can be extended into 3D to highlight relationships between variables

–existing 2D one-lines were redrawn using a 3D visualization language, OpenGL

–easy navigation and interaction very important

39

PSERC

3D View of Generation Sources in Midwest

40

PSERC

Visualization of Contingency Analysis Results

• Contingency analysis results can be presented in a 2D matrix format (contingencies versus violated elements)–but such an approach loses the geographic

information for both the contingencies and the violated elements

• We are working on 3D approaches to supplement traditional 2D displays

41

PSERC

Single Device Contingencies: Contingency to Violated Elements

42

PSERC

Single Device Contingencies: Violated Element to Contingencies

43

PSERC

Conclusion

• Lack of situational awareness was a key cause of the August 14th blackout; this greatly hindered emergency control

• A lack of emergency control requires more constrained operation with increased system cost

• Automatic control, such as price feedback, could certainly help

• Better visualization technology is needed