Robust Dynamic State Estimation for PMU Data Quality and ...

29
Robust Dynamic State Estimation for PMU Data Quality and Security 1 IEEE PES GM 2019 Panel Dynamic State Estimation for Power System Monitoring, Protection and Control-- Paving the Way for A More Resilient Grid Assistant Professor Department of Electrical and Computer Engineering Mississippi State University Starkville, MS, 39762, USA Junbo Zhao, Ph.D. Acknowledgement-Professor Lamine Mili

Transcript of Robust Dynamic State Estimation for PMU Data Quality and ...

Page 1: Robust Dynamic State Estimation for PMU Data Quality and ...

Robust Dynamic State Estimation for PMU Data Quality and Security

1

IEEE PES GM 2019 Panel

Dynamic State Estimation for Power System Monitoring, Protection and Control--Paving the Way for A More Resilient Grid

Assistant ProfessorDepartment of Electrical and Computer Engineering

Mississippi State UniversityStarkville, MS, 39762, USA

Junbo Zhao, Ph.D.

Acknowledgement-Professor Lamine Mili

Page 2: Robust Dynamic State Estimation for PMU Data Quality and ...

Outline

Introduction of Power System DSE

Motivations and Implementations

Robust DSE Needs

Proposed Robust Unscented Kalman Filter

Bad Data and Cyber Attack Detection

Robust Filtering

Results and Discussions

Conclusions

2

2

Page 3: Robust Dynamic State Estimation for PMU Data Quality and ...

Source: Solar in US-NREL

Integration of intermittent DERs and flexible loads adds more uncertainties and dynamics to the system operating states;Steady-state assumption may be violated.

Source: EV in US-IHS Automotive

Dynamic State Estimation- Motivations3

Page 4: Robust Dynamic State Estimation for PMU Data Quality and ...

Synchrophasor measurements at 30-60 samples/s, provide the opportunity to capture power system dynamics.

Source: www.naspi.org

Dynamic State Estimation- Motivations4

Page 5: Robust Dynamic State Estimation for PMU Data Quality and ...

Dynamic State Estimator

Exciter

Governor

Synchronous Generators with PMU

Measurements

P

V

refV

MP

fdE

refP

'

dE '

qEAdvanced System

Control and Protection

For advanced protection and control of power systems.

Q

Dynamic State Estimation- Motivations5

Page 6: Robust Dynamic State Estimation for PMU Data Quality and ...

Discrete-time state-space model of the power system DAE:

๐”ผ ๐’˜๐‘˜๐’˜๐‘˜๐‘‡ = ๐‘ธ๐‘˜

๐”ผ ๐’—๐‘˜๐’—๐‘˜๐‘‡ = ๐‘น๐‘˜

๐’™๐‘˜ = ๐’‡ ๐’™๐‘˜โˆ’1, ๐’š๐‘˜โˆ’1, ๐’–๐‘˜ +๐’˜๐‘˜

๐’›๐‘˜ = ๐’‰ ๐’™๐‘˜ , ๐’š๐‘˜ , ๐’–๐‘˜ + ๐’—๐‘˜

subject to various types of constraints

(1)

Problem Formulation-DSE

6

Page 7: Robust Dynamic State Estimation for PMU Data Quality and ...

โ€ข Implementation using Kalman Filter Framework

Implementation of DSE

โ€œState Predictionโ€Dynamic Simulation

โ€œState correctionโ€

Prediction step Filtering step

๐’™๐‘˜ = ๐’‡ ๐’™๐‘˜โˆ’1, ๐’š๐‘˜โˆ’1 +๐’˜๐‘˜โˆ’1๐’›๐‘˜ = ๐’‰ ๐’™๐‘˜, ๐’š๐‘˜ + ๐’—๐‘˜

๐’›๐‘˜

๐’™๐‘˜โˆ’1 ๐’™๐‘˜

Extended Kalman filter-poor performance for strong nonlinear system

Unscented Kalman filter and its variants, Ensemble Kalman filter-sensitive to non-Gaussian noise

Particle filter and its variants-sensitive to bad data

7

Page 8: Robust Dynamic State Estimation for PMU Data Quality and ...

8

Robust DSE-Motivations

Large measurement bias

Non-Gaussian noise

Data dropout/packet loss/measurement delays

Loss of GPS synchronization

Unknown inputs and noise statistics

Cyber attacks

PMU data quality issues:

Page 9: Robust Dynamic State Estimation for PMU Data Quality and ...

9

Non-Gaussian PMU Noise Distributions

Page 10: Robust Dynamic State Estimation for PMU Data Quality and ...

10

Voltage and current angle errors-roughly Gaussian; Voltage and current magnitude errors-Gaussian mixture; Real and reactive power errors-thick-tail distributions, such as

Laplace and Cauchy.

S. Wang, J. B. Zhao, Z. Huang, R. Diao "Assessing Gaussian Assumption of PMU Measurement

Error Using Field Data," IEEE Trans. on Power Delivery, vol. 33, no. 6, pp. 3233-3236, 2018.

Non-Gaussian PMU Noise Distributions

Page 11: Robust Dynamic State Estimation for PMU Data Quality and ...

11

0 50 100 150 200-0.5

0

0.5

1

1.5

2

Time [seconds]

An

gle

[ra

dia

ns]

UFAM

UFMA

UFPA

UFRGS

UFSC

UNIPAMPA

Bad data

Event

Brazil utility

Occurrence of Outliers (Field Data from Brazil)

Page 12: Robust Dynamic State Estimation for PMU Data Quality and ...

12

Vulnerabilities to Cyber Attacks

Page 13: Robust Dynamic State Estimation for PMU Data Quality and ...

Proposed Robust GM-UKF

13

Page 14: Robust Dynamic State Estimation for PMU Data Quality and ...

Batch-mode Regression FormCombine the measurement equation and the prediction equation to obtain the batch-mode regression form :

Prediction equation:

๐’™๐‘˜ = ๐’™๐‘˜|๐‘˜โˆ’1 + ๐œป๐‘˜|๐‘˜โˆ’1

Measurement equation: Perform statistical linearization of๐’‰(๐’™๐‘˜) around the predicted state ๐’™๐‘˜|๐‘˜โˆ’1:

๐’›๐‘˜ = ๐’‰ ๐’™๐‘˜|๐‘˜โˆ’1 +๐‘ฏ๐‘˜ ๐’™๐‘˜ โˆ’ ๐’™๐‘˜|๐‘˜โˆ’1 + ๐Š๐‘˜ + ๐’†๐‘˜

where ๐‘ฏ๐‘˜ = ๐‘ท๐‘˜|๐‘˜โˆ’1๐‘ฅ๐‘ง ๐‘‡

๐‘ท๐‘˜|๐‘˜โˆ’1๐‘ฅ๐‘ฅ โˆ’1

and ๐’†๐‘˜ is the statistical

linearization error.

14

Predicted state

True value

Prediction error

(2)

(3)

Page 15: Robust Dynamic State Estimation for PMU Data Quality and ...

Batch-mode Regression Form

๐’›๐‘˜ โˆ’ ๐’‰ ๐’™๐‘˜|๐‘˜โˆ’1 +๐‘ฏ๐‘˜ ๐’™๐‘˜|๐‘˜โˆ’1 ๐’™๐‘˜|๐‘˜โˆ’1

= ๐‘ฏ๐‘˜๐‘ฐ๐’™๐‘˜ +๐Š๐‘˜ + ๐’†๐‘˜โˆ’๐œป๐‘˜|๐‘˜โˆ’1

๐’›๐‘˜ = ๐‘ฏ๐‘˜ + ๐’†๐‘˜

โ€ข The covariance matrix of the error ๐’†๐‘˜ is given by

๐”ผ ๐’†๐‘˜ ๐’†๐‘˜๐‘‡ =๐‘น๐‘˜ + ๐‘น๐‘˜ 00 ๐œฎ๐‘˜|๐‘˜โˆ’1

= ๐‘บ๐‘˜๐‘บ๐‘˜๐‘‡ ,

where ๐”ผ ๐’†๐‘˜๐’†๐‘˜๐‘‡ = ๐‘น๐‘˜ and ๐‘บ๐‘˜is obtained from

Cholesky decomposition and used for prewhitening after outlier detection.

15

(4)

(5)

Page 16: Robust Dynamic State Estimation for PMU Data Quality and ...

Detecting Outliers by PS

16

Projection statistics is defined and calculated by

(6)

Page 17: Robust Dynamic State Estimation for PMU Data Quality and ...

Outlier detection using PSWe propose to apply PS on the following matrix

๐’ =๐’›๐‘˜โˆ’1 โˆ’ ๐’‰ ๐’™๐‘˜โˆ’1|๐‘˜โˆ’2 ๐’™๐‘˜โˆ’1|๐‘˜โˆ’2

๐’›๐‘˜ โˆ’ ๐’‰ ๐’™๐‘˜|๐‘˜โˆ’1 ๐’™๐‘˜|๐‘˜โˆ’1

Time instants: ๐‘˜ โˆ’ 1 ๐‘˜

Global redundancy for outlier detection is increased thanks to the batch-mode formulation.

๐’ has two columns to measure temporal correlations. It is found two continuous samples are sufficient to detect outliers effectively.

PS is applied to the innovation vectors and the predicted state vector separately.

17

Innovation vectors

Page 18: Robust Dynamic State Estimation for PMU Data Quality and ...

Probability Distribution of PS

18

Q-Q plots of the sample quantiles of the PS vs. the correspondingquantiles of the chi-square distributions with different degree offreedoms, where the left and the right figures represent Q-Q plots ofPS with Gaussian and Laplace noise, respectively.

Page 19: Robust Dynamic State Estimation for PMU Data Quality and ...

Robust Filtering

19

๐œ›๐‘– = min{1,c22,0.975

๐‘ƒ๐‘†๐‘–

2

} (7)

arg๐‘š๐‘–๐‘› ๐ฝ ๐’™ = ๐‘–=1๐‘š ๐œ›๐‘–2 ๐œŒ ๐‘Ÿ๐‘†๐‘–

New weighting function:

Robust filtering via the GM-estimator that minimizes an objective function given by

(8)

Page 20: Robust Dynamic State Estimation for PMU Data Quality and ...

Robust Filtering

๐๐‘ฑ ๐’™

๐๐’™=

๐’Š=๐Ÿ

๐’Ž

โˆ’๐•๐’Š๐’‚๐’Š๐’”๐ ๐’“๐‘บ๐’Š = ๐ŸŽ

โ€ข The minimization problem is solved via the Iteratively Reweighted Least Squares (IRLS) algorithm:

๐’™๐’Œ|๐’Œ(๐’‹+๐Ÿ)= ๐‘จ๐’Œ๐‘ป๐‘ธ(๐’‹)๐‘จ๐’Œ

โˆ’๐Ÿ๐‘จ๐’Œ๐‘ป๐‘ธ(๐’‹)๐’š๐’Œ

๐‘ธ = diag ๐’’ ๐’“๐‘บ๐’Š and ๐’’ ๐’“๐‘บ๐’Š = ๐ ๐’“๐‘บ๐’Š ๐’“๐‘บ๐’Š.

โ€ข Stopping rule:

๐’™๐’Œ|๐’Œ(๐’‹+๐Ÿ)โˆ’ ๐’™๐’Œ|๐’Œ(๐’‹)< ๐Ÿ๐ŸŽโˆ’๐Ÿ

20

(9)

(10)

(11)

Page 21: Robust Dynamic State Estimation for PMU Data Quality and ...

Robust Covariance Matrix UpdatingTheorem 1: The estimated state by our GM-UKF tends to a Gaussian distribution asymptotically even when the system process and measurement noise follow a non-Gaussian distribution. Furthermore, the estimation error covariance matrix is updated through

๐œฎ๐‘˜|๐‘˜ = 1.0369 ๐‘จ๐‘˜๐‘‡๐‘จ๐‘˜โˆ’1๐‘จ๐‘˜๐‘‡๐‘ธ๐œ›๐‘จ๐‘˜

โˆ’1๐‘จ๐‘˜๐‘‡๐‘จ๐‘˜โˆ’1

where๐‘ธ๐• = diag ๐œ›๐’Š๐Ÿ .

21

๐‘ธ๐œ› determined by PS is used to downweight outliers, yielding robust covariance matrix updating.

(12)

Page 22: Robust Dynamic State Estimation for PMU Data Quality and ...

Disturbance: at t=0.5 seconds, transmission line between buses 15 and 16 is removed.

Generator model: two-axis model with IEEE DC1A excitation system and TGOV1 turbine-governor is assumed.

Non-Gaussian noise:

Bimodal Gaussian mixture noise with zero mean, variances of 10โˆ’4 and 10โˆ’3and weights of 0.9 and 0.1, respectively, is added to the voltage magnitudes;

Laplacian noise with zero mean and scale 0.2 is added to the real and reactive power injections.

22

Illustrative Results on IEEE 39-bus system

Page 23: Robust Dynamic State Estimation for PMU Data Quality and ...

Case 1: Non-Gaussian Noiseโ€ขNo outliers;

โ€ขBimodal Gaussian mixture for current and voltage magnitudes;

โ€ขLaplace noises for real and reactive power;

Case 1: Non-Gaussian Noise23

Page 24: Robust Dynamic State Estimation for PMU Data Quality and ...

Case 2: Observation Outliersโ€ขThe real and reactive power measurements of Generator 5 are

corrupted with 20% error from 4s to 6s; Laplace noises for real

and reactive power.

โ€ข State estimates by UKF

are significantly biased;

โ€ข GM-UKF achieves

much higher statistical

efficiency than Huber-

UKF and GM-IEKF.

24

Page 25: Robust Dynamic State Estimation for PMU Data Quality and ...

Case 3: Parameter Errorsโ€ขThe predicted rotor angle of the Generator 5 is incorrect due to

the incorrect parameter of G5 from 4s to 6s; Laplace noises for

real and reactive power.

โ€ขState estimates by UKF

and Huber-UKF are

significantly;

โ€ขGM-UKF achieves much

higher statistical efficiency

than GM-IEKF.

25

Page 26: Robust Dynamic State Estimation for PMU Data Quality and ...

Table II. Average Computing Time at Each PMU Sample (PC with Intel Core i5, 2.50 GHz, 8GB of RAM)

26

Cases EKF UKF GM-IEKF GM-UKF

Case 1 6.24ms 6.28ms 9.64ms 9.52ms

Case 2 6.28ms 6.31ms 9.68ms 9.55ms

Case 3 6.43ms 6.38ms 9.72ms 9.63ms

Case 4 6.45ms 6.40ms 9.71ms 9.62ms

Case 5 6.25ms 6.29ms 9.66ms 9.54ms

Breakdown Point and Computing Efficiency

Handle at least 25% outliers due to cyber attacks, PMU

communication issues or model deficiency;

Suitable for real-time application.

Page 27: Robust Dynamic State Estimation for PMU Data Quality and ...

Conclusions

โ€ข DSE is excepted to be a key tool in future cyber-physical energy system EMS

โ€ข GM-UKF is able to handle non-Gaussian noise, bad data and cyber attacks while achieving good statistical efficiency

โ€ข Numerical and statistical robustness of DSE should be extensively studied

โ€ข Field validation of the developed GM-UKF will be our future work

27

Page 28: Robust Dynamic State Estimation for PMU Data Quality and ...

28

28

โ€ข J. B. Zhao, L. Mili, "A Theoretical Framework of Robust H-infinity Unscented Kalman Filter and Its Application to Power System Dynamic State Estimation," IEEE Trans. Signal Processing, vol. 67, no. 10, pp. 2734-2746, 2019.

โ€ข J. B. Zhao, A. Exposito, M. Netto, L. Mili, A. Abur, V. Terzija, I. Kamwa, B. Pal, A. K. Singh, J. Qi, Z. Huang, A. P. Sakis Meliopoulos, ''Power System Dynamic State Estimation: Motivations, Definitions, Methodologies and Future Work," IEEE Trans. Power Systems, 2019.

โ€ข J. B. Zhao, L. Mili, "A Robust Generalized-Maximum Likelihood Unscented Kalman Filter for Power System Dynamic State Estimation," IEEE Journal of Selected Topics in Signal Processing, vol. 12, no. 4, pp. 578-592, 2018.

References

Page 29: Robust Dynamic State Estimation for PMU Data Quality and ...

29