You are on page 1of 39

Taylor-Fourier PMU

in RTLab environment
G. Frigo, Y. Zuo, M. Paolone
Research scenario
Before their deployment on the field, Phasor Measurement Units (PMUs) have
to be characterized in terms of estimation accuracy and reporting latency w.r.t.
IEEE Std C37.118.1 requirements and expected worst-case conditions.

y Experimental testing hard to control or even unfeasible outside lab

y Real Time Simulator easy test a wide range of operating conditions PMUs’ reliability
during critical scenarios

Î PMU RTS models are a promising solution as they allow for reproducing
large-scale applications involving multiple PMUs.
PMU simulation approaches
y Option 1: Hardware in the Loop

Advantages Limitations

• Allow testing real • $$$ (>10k$ per


Option 1 PMU devices device)
(real • Typically not • Analog outputs
devices) computationally (number/accuracy)
intensive (low TS) • Cabling
PMU simulation approaches
y Option 2: Simulated PMU

Advantages Limitations

• Allow testing real • $$$ (>10k$ per


Option 1 PMU devices device)
(real • Typically not • Analog outputs
devices) computationally (number/accuracy)
intensive (low TS) • Cabling
• Extreme cost
• Complexity of the
Option 2 reduction
PMU model vs.
(simulated • Not limited by
computational
devices) available AOs and
power of the RTS
PMUs
Static vs dynamic phasors
Static phasor Dynamic phasor
y stationary signal parameters y accounts for parameters’ time-variations

y ROCOF defined as finite-difference of y ROCOF is a state variable, i.e. the 2nd order
consecutive frequency estimates time-derivative of phase

Accuracy comparison
‫ = ݐ ݔ‬cos ʹߨ݂‫ ݐ‬+ ʹߨ‫ ݐ‬ଶ + 0.05 ‫ ڄ‬cos Ͷߨ‫ ݐ‬െ ߨ
ROCOF t = 2 െ 0.4ߨ ‫ ڄ‬cos Ͷߨ‫ ݐ‬െ ߨ

y dynamic phasor does not introduce delay

y dynamic phasor guarantees more accuracy


PMU signal model
Time-varying power signal affected by narrow- & wide-band disturbances:

‫ ڄ ࡭ = ݐ ݔ‬1 + ߝ஺ ‫ݐ‬ ‫ ڄ‬cos ʹߨࢌ‫ ݐ‬+ ࣐ + ߝఝ ‫ݐ‬ + ߟ(‫ )ݐ‬+ ߩ(‫)ݐ‬

y ‫ܣ‬, ݂, ߮ Æ fundamental tone parameters


y ߝ஺ , ߝఝ Æ amplitude/phase modulations
y ߟ Æ spurious narrow-band tones
y ߩ Æ measurement noise (AWGN)

Î Typically, PMU RTS models consider a window of ࡺ࢝ samples and compute its Discrete Fourier
Transform (DFT) ࢄ(ࢌ) in order to extract ‫ܣ‬, ݂, ߮.
Dynamic phasor via CS-TFM
Compressive Sensing Taylor-Fourier Multifrequency (CS-TFM) model:

y identification of signal main components


CS-based routine to get a super-resolution effect in the DFT domain

y 2nd-order Taylor-Fourier decomposition


higher order TF coefficients account for time-varying parameters

y extraction of fundamental dynamic phasor


estimated three-term phasor ‫݌‬ො = ‫݌‬ො଴ , ‫݌‬ොଵ , ‫݌‬ොଶ that accounts for:
o dynamic synchrophasor Æ ‫ܣ‬መ଴ ‫ ڄ‬exp ͳ݆ ‫߮ ڄ‬ො ଴
o fundamental frequency Æ ݂௡௢௠ + ߮ො ଵ Τʹߨ
o instantaneous ROCOF Æ [Ա ‫݌‬ොଶ exp െͳ݆ ‫߮ ڄ‬ො ଴ െ ‫ܣ‬መଵ ‫߮ ڄ‬ො ଵ Τ(ʹߨ ‫ܣ ڄ‬መ଴ )
P-class vs M-class
P-class M-class

60 ms, fundamental + 4 harmonics

100 ms, fundamental + 4 harmonics


+ inter-harmonics
Adopted RTS features
Opal-RT eMEGAsim PowerGrid Real-Time Digital Simulator:

y industrial PC with 12 cores


runs the RTS model with an integration time-step of 100 Ɋs
GPS antenna provides UTC-timestamp and PPS time-base

y Dolphin DXE410 PCI Express Expansion Chassis


is
manages the blocks’ communication

y Spartan-3 FPGA board


locks the internal time-base to a more
deterministic clock
Simulated PMU model

Synchronization module

Synchrophasor estimation algorithm


• Modulated-Sliding DFT (MS-DFT)
• Compressive Sensing Taylor-Fourier
CS-
Multifrequency Model (CS-TFM)
TFM

Data encapsulation module


• IEEE Std. C37.118 compliant
• UTC-synchronized via GPS
Computational complexity
Given an integration time of 100 ࣆs, the complexity is evaluated in terms of:
y PMUs per RTS core
y Computation time

Stream
CS-TFM
MS-DFT Simulation
(n-1)Tr nTr (n+1)Tr time
PMU performance evaluation
IEEE Std C37.118.1 for Synchrophasor Measurements

Total Vector Error (TVE)


Im
ଶ ଶ
ܺ െ ܺ෠ ෢௥
ܺ௥ െ ܺ െ ܺ௜ െ ܺ෡௜
ܺ ‫ܣ = ݐ‬଴ ‫ ݐ‬ȉ ݁ ௝ ଶగ௙బ ௧ ௧ାఝబ TVE = =
ܺ ܺ௥ ଶ െ ܺ௜ ଶ

ܺ෠ ‫ܣ = ݐ‬መ ଴ ‫ ݐ‬ȉ ݁ ௝ ଶగ௙መబ ௧ ௧ାఝబ


Frequency Error (FE)
FE = ݂ െ ݂መ

In modern distribution networks:


Re TVE < 0.01%, FE ؄ 10 ݉‫ݖܪ‬
Model assessment
IEEE Std C37.118.1 for Synchrophasor Measurements

Test parameters:
y sampling rate equal to 5 kHz
y reporting period equal to 20 ms
y test duration equal to 2 (or 4) s
y AWGN with SNR equal to 80 dB

Î for each test we report the worst-case performance in terms of TVE, FE, RFE
Static and dynamic tests
P-class (60 ms, 5 PMUs per core)

M-class (100 ms, 1 PMU per core)


Response times
P-class

M-class
Real-world scenario
Inter-area oscillation of Continental European system on December 1, 2016
y fundamental frequency: affected by modulations and linear ramps:
modulation frequency 150 mHz
modulation factor 60 mrad
ramp ROCOF [-2.3, +2.5] Hz/s

y need for dynamic phasor estimates;

y high accuracy with both models:


P-class: FE ൑ 200 ߤHz
M-class: FE ൑ 75 ߤHz
Conclusions
y We presented the design, implementation and characterization of two
dynamic phasor extraction models based on CS-TFM algorithm within a RTS.
available at our Github repository: https://github.com/DESL-EPFL/CS-TFM-PMU-Model

y We described the RTS platform and quantified the computational complexity.


P-class Æ 5 PMUs per core, M-class Æ 2 (or 1) PMU per core
computation time negligible if compared with reporting period (tens ߤs vs tens ms)

y Performance evaluation in terms of accuracy and response time in IEEE Std


tests and real-world power system dynamics.
compliant with P- and M-class requirements (apart from P-class Phase step test)
high-accuracy estimates also during inter-area oscillations (FE ൑ 200 ߤrad)
Inertia-less IEEE 39-bus
grid for UFLS applications
G. Frigo, Y. Zuo, M. Paolone
Content
y An inertia-reduced 39-bus power system dynamic model & compare with IEEE
39-bus benchmark grid

y ROCOF-based under frequency load shedding relays

y Real-time Simulation
Inertia-less 39-bus Power Grid
Original IEEE 39-bus benchmark grid Inertia-reduced 39-bus power grid

G1, G5, G8 and G9


replaced by WP1,
WP2, WP3 and WP4
Synchronous Generators
Conventional generation relies on both hydro- and thermal-power plants:

y a six-order state-space model for the synchronous machine;

y a prime mover (i.e., hydro or steam turbine) and governor

y a DC1A excitation system associated with an AVR.

The primary frequency regulator (PFR) has a droop coefficient ܴ௣ = 5%.


In G7, also a secondary frequency regulator (SFR) with time constant 120 s.
Dynamic Loads
EPRI LOADSYN model reproduces the load response to voltage and frequency variations

௄೛ೡ
ܸ ‫ݐ‬
ܲ ‫ܲ = ݐ‬଴ ‫ݐ‬ 1 + ‫ܭ‬௣௙ ݂ ‫ ݐ‬െ ݂଴
ܸ଴

௄೜ೡ
ܸ ‫ݐ‬
ܳ ‫ܳ = ݐ‬଴ ‫ݐ‬ 1 + ‫ܭ‬௤௙ ݂ ‫ ݐ‬െ ݂଴
ܸ଴

where ܲ ‫ ݐ‬and ܳ ‫ ݐ‬are the total load active and reactive power, and the parameters ‫ܭ‬௣௩ , ‫ܭ‬௣௙ , ‫ܭ‬௤௩ and
‫ܭ‬௤௙ are obtained from typical load voltage and frequency profiles in:

y “Load representation for dynamic performance analysis (of power systems)”, IEEE Trans on Pow Sys, vol. 8, no. 2, pp. 472–482, May 1993.
Dynamic Loads

A PLL and a RMS operator measure the bus frequency and voltage feeding the dynamic load model. In
order to get a smoother PLL response, a moving average is added.

ܲ଴ ‫ ݐ‬, ܳ଴ (‫ )ݐ‬are active and reactive power profiles at the rated frequency and voltage (i.e., 50 Hz and
345 kV). These demand profiles are derived from a monitoring system based on PMUs installed on the
125 kV sub-transmission system of the city of Lausanne, Switzerland.
Wind Power Plants

The wind generator model consists of a DFIG and an averaged back-to-back converter model.

The power output is approximated by multiplying the power output of a detailed model of a single wind
turbine to match the total nominal capacity of the whole wind farm.
Wind Profiles
The wind power profiles are generated at 1 second resolution by re-sampling the measurements
at 1 minute resolution from ERCOT.

y GE Energy, Analysis of wind generation impact on ERCOT ancillary services requirements, 2008.

The re-sampling approach is based on iterated smoothing and differentiating operations that use
the statistical characteristics of the aggregated wind generation profiles presented in:

y K. Coughlin and J. H. Eto, “Analysis of wind power and load data at multiple time scales”, Lawrence Berkeley
National Laboratory (LBNL), Berkeley, CA (United States), Tech. Rep., 2010.
BESS-VSC Model
y The BESS model is a Three-time Constant (TTC) model whose parameters have been
identified using real data from a 720kW/560kWh LiTi BESS available at EPFL- DESL.

y The BESS is integrated into the IEEE 39-bus through an aggregated fully modeled three-
level neutral-point clamped (NPC) converter.
BESS RTS Model
݀‫ݐ ݔ‬
= ‫ܣ‬௖ ܱܵ‫ ݐ ݔ ڄ ܥ‬+ ‫ܤ‬௖ ܱܵ‫)ݐ(ݑ ڄ ܥ‬
݀‫ݐ‬
‫ܥ = )ݐ(ݕ‬௖ ܱܵ‫ ݐ ݔ ڄ ܥ‬+ ‫ܦ‬௖ ܱܵ‫)ݐ(ݑ ڄ ܥ‬

nominal power capacity and


energy capacity
225 MW and 176 MWh

Since the nominal active power of the DESL BESS is way lower than the one to be connected to a
HV transmission grid, its SSM is scaled-up to obtain a 225 MW SSM.
Three-Level NPC Converter
VSC Grid-Feeding Control
The VSC is implemented with a grid-feeding structure embedding outer control loop that decouples the
active and reactive power regulations.

An external frequency regulation control is implemented: droop coefficient ‫ܭ‬௣ି௙ = 20, limit for
maximum/minimum active power to ± 0.95 ‫ݑ݌‬
Simulation Setup
y Software

MATLAB/Simulink and RT-LAB: MATLAB version 8.5.1.281278 (R2015a) Service Pack 1, RT-LAB version
v11.2.2.108 with associated ARTEMIS Blockset version 7.2.2.1206.

y Hardware

OPAL-RT OP5600 installed with the corresponding RT-LAB version, with at least 7 cores available.

y Contingency for this analysis

The tripping of generator G6 that causes about 800 MW of generation loss;

Simulation time step equal to 25 Ɋs.


Simulation Demo
Compare with IEEE 39-bus benchmark grid

The inertia-less 39-bus power system dynamic model is available on Github (DESL-EPFL):
https://github.com/DESL-EPFL/Extension-of-the-IEEE-39-bus-Test-Networkfor-the-Study-of-
Fundamental-Dynamicsof-Modern-Power-System.git
ROCOF-based load shedding and load restoring relays
Recap on the PMUs
The Frequency and ROCOF measurements are from two P-class PMUs:

PMU csTFM
y accounts for parameters’ time-variations
y ROCOF is a state variable, i.e. the 2nd order time-derivative of phase

PMU e-IpDFT
y accounts for stationary signal parameters
y ROCOF is defined as finite-difference of consecutive frequency estimates
Real-time Simulation
y Install PMUs for every load

y Implement the UFLS relays for every load

y Reproduce the same contingency: tripping of G6, 800MW generation loss


Results--Load Shedding
Results--Frequency
Results--ROCOF
THANK YOU

You might also like