Zakutsky 2020

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Simulation of thyristor-controlled reactor using

susceptance-to-firing angle conversion with


regression models
Vadim I. Zakutsky Qiuwei Wu Magomed G. Gadzhiev
Department of Power Electrical Systems Centre for Electric Power and Energy Department of Power Electrical Systems
Moscow Power Engineering Institute Department of Electrical Engineering Moscow Power Engineering Institute
Moscow, Russia Technical University of Denmark Moscow, Russia
vadim.zakutsky@ieee.org Kongens Lyngby, Denmark gmg.hhh@gmail.com
qw@elektro.dtu.dk

Abstract—This paper describes an approach to simulation have controlled reactors installed including transformer based
of thyristor-controlled reactor (TCR) with automatic voltage reactors, magnetic saturation controllable reactors and TCRs.
regulator (AVR). Two approaches to design susceptance-to-firing Shunt connected controlled reactor is the most common type
angle (B-to-α) conversion block were considered. First one is
look-up table approach, second one is based on fitting inverse of FACTS devices in the UES [6].
relationship between firing angle of thyristors and susceptance The control of the thyristor-controlled reactor in the SVC
by using regression models. Despite that look-up table is common type var generator needs to provide four basic functions:
way to obtain firing angle from input susceptance by searching synchronous timing, reactive current (or susceptance) to firing
for the appropriate angle from a table with pre-calculated values, angle (b-to-α) conversion, computation of the required funda-
the regression model based approach shows better perfomance
in terms of increasing speed of calculations and simplicity of mental reactor current and thyristor firing pulse generation
modelling. Regression models might be used in order to decrease [5].
delays in control system of TCR. Different regression models According to [1]–[3], [5], [7], [10], [14], the most common
are derived and compared. A numerical comparison with the way of modelling b-to-α conversion block is by implementing
classical look-up table approach is given.
look-up table. But this approach leads to increasing simulation
Index Terms—FACTS, thyristor-conrolled reactor, TCR, SVC,
AVR, simulation, susceptance-to-firing angle conversion, power time because it requires an iterative process to obtain the firing
electronics, thyristors, regression angle from the susceptance value at each sample time. Also
this may increase delays in control system of TCR in real time
I. I NTRODUCTION operation. Regression analysis is one of the most widely used
It’s well known that in the mid 60’s traditional energy and simple methods to determine the functional dependency
systems by implementing various types of reactive power between variables. Inverse regression model based relationship
compensators began to transform into the Flexible AC Trans- α = f (B) is suggested to be used instead of look-up table to
mission Systems (FACTS) - alternating current transmission obtain the firing angle from input susceptance.
systems incorporating power electronic-based and other static The way of detailed modelling both the reactor itself and
controllers to enhance controllability and increase power trans- the automatic voltage regulator (AVR) with b-to-α conversion
fer capability. FACTS devices can be classified in three groups block is considered in this paper. Two ways of modelling b-
by implemented type of controller: series connected, shunt to-α conversion block are compared:
connected and combined shunt and series connected con- • By using digital look-up table;
trollers. A shunt connected thyristor-controlled reactor (TCR) • By implementing regression models which fit the inverse
is one of the most important building blocks of thyristor- relationship between firing angle and susceptance of
based static var compensators (SVCs). Although it can be used TCR.
alone, it is more often employed in conjunction with fixed
Matlab/Simulink was used for design and simulation.
or thyristor-switched capacitors to provide rapid, continuous
control of reactive power over the entire selected lagging-to- II. TCR MODEL
leading range. By using power thyristors, the static compen-
sators present many advantages from several points of view, A. Thyristor-controlled reactor
such as high reaction speed, insignificant contribution to the A TCR branch contains reactors which are phase angle
short-circuit power, and low maintenance [1]–[5]. controlled by thyristor valves. Three single phase branches
As of 2016, according to the statistical rewiev in the Unified are connected in delta (Fig. 1) to reduce the generation of
Energy System (UES) of Russia, 32 substations 220-500 kV triplen harmonics in symmetrical operation or multiples of

l-))) 

Authorized licensed use limited to: San Francisco State Univ. Downloaded on November 12,2020 at 15:10:18 UTC from IEEE Xplore. Restrictions apply.
third harmonics from percolating into the transmission lines C. Synchronization block
[2], [4]. Synchronization block is intended to synchronize pulses
from Pulse generator with the voltage phase in the POC. In
order to fire thyristors accurately, it is essential to look for the
time crossing zero of synchronous signal. Common methods to
get zeros are by sampling and calculating zero-crossing instant
by linear interpolation [9].
For the power flow simulation studies the utilization of
simple phase synchronization block is suggested. Fig. 3 shows
the scheme of this block. There are two sinusoidal signals as
a inputs:
Fig. 1: Delta connected TCR [2] • Measured sinusoidal voltage magnitude at the MV side
of POC.
• Sine wave with 0° phase shift.

B. Control system structure


There are two types of control system structure for SVC 2 ˜ Vmeas ˜ Sin(Zt  T ) Zt  T T
PLL block
proposed by IEEE: the IEEE Basic Model 1 and the IEEE MV side of POC 
Basic model 2 [2], [8]. This schemes can also be implementing a ˜ Sin(Zt ) Zt
PLL block
in a independent TCR’s control system structure (Fig. 2).

Fig. 3: Synchronization block scheme


, PHDV
. 6/ .,
9UHI  9H V  %UHI %WRĮ D
&RQYHUVLRQ
3XOVH
*HQHUDWRU
PLL block in Simulink is dedicated to calculate the value of
9PHDV 
.5  EORFN 
the frequency and phase angle of the input sine signal. After
  V75
PLL blocks output values ωt and ωt+θ compares with each
+9VLGHRI32&

9PHDV
09VLGHRI32&
6\QFKURQL]DWLRQ
%ORFN
T other by subtraction to obtain the phase shift θ of the voltage
at the MV side of POC. After that θ compares with a firing
Fig. 2: The IEEE Basic Model 2 control system of TCR angle α (Fig. 3) in order to synchronize pulses, genereted by
Pulse generator, with the voltage phase in the POC.
The IEEE Basic model 2 corresponds to the integrator with D. Susceptance-to-firing angle conversion block
the current droop format. This scheme works by converting
This block is intended to convert an output value of a
voltage error Ve - difference between reference voltage Vre f ,
reference susceptance from the AVR to firing angle of thyris-
measured value Vmeas in the point of connection (POC) of
tors. Firing angle is varying between 90 and 180 degrees.
TCR and a current droop KSL - to the reference value of
When α = 180°, the reactor becomes non-conducting. When
susceptance of TCR Bre f and after that to firing (delay) angle
α = 90°, the reactor becomes fully conducting [5].
α of thyristors. This approach is called phase control [2].
Dependence of reference susceptance (B, p.u.) from firing
A proportional gain K p is employed to increase the speed angle (α, rad) and can be expressed as [2], [3]:
of response.
The AVR can be equivalently expressed as: 2π − 2α + sin 2α
B = f (α) = (3)
  π
KI 1 + sTQ
G(s) = · (1) The output current of TCR can be calculated by using
s 1 + sTp equation [2]:
where:  
VL 2α 1
Kp IL (α) = · 1− − · sin 2α (4)
TQ = Tp + (2) ωL π π
KI
Obviously, the dependence (3) is nonlinear. In order to
The voltage regulator transforms to the simple propor- derive the inverse relationship between firing angle and sus-
tional–integral (PI) type if Tp is kept at zero. ceptance several approaches can be implemented [1], [5]:
Pulse generator is intended to convert firing angle to pulses • An analog function generator producing in each half-
in order to control power output of TCR by gating two cycle a scaled electrical signal that represents the B(I)
oppositely poled thyristors which conduct on alternate half- versus α relationship;
cycles of the supply frequency [3]. • A digital look-up table for the normalized current (suscep-
The basics of modelling b-to-α conversion and synchroniza- tance) versus a function which is read at regular intervals
tion blocks are described below. (e.g., at each degree) starting from α = 90° until the



Authorized licensed use limited to: San Francisco State Univ. Downloaded on November 12,2020 at 15:10:18 UTC from IEEE Xplore. Restrictions apply.
requested value is found, at which instant a firing pulse its coefficient values β = [β1 , β2 , ..., βn ] in terms of accuracy
is initiated. of fitting.
• Using a microprocessor and compute the delay angle 4 nonlinear regression models were considered: polynomial,
corresponding to the required current (susceptance). The exponential, Gaussian and Fourier (Eq. 7-10, accordingly). βk ,
actual firing instant is then determined simply by a timing γk , σk denote coefficient values. Each regression model was
circuit (e.g., a counter) "measuring" a from the peak of investigated with respect to different number of terms n.
the voltage. n
In this paper 2 approaches are compared: f (x) = ∑ βk · xk + β0 (7)
• A digital look-up table approach implementation; k=1
• Derivation of an inverse dependence between firing angle n
and susceptance α = f (B) by using regression analysis. f (x) = ∑ βk · exp(γk · x) (8)
k=1
III. A PPROACHES FOR SUSCEPTANCE - TO - FIRING ANGLE n
CONVERSION BLOCK DESIGN f (x) = ∑ βk · exp(−((x − γk )/σk )2 ) (9)
k=1
A. look-up table approach
The most common and actionable way to design look-up n
table is by using dimensional interpolation. The table is [B, α] f (x) = ∑ [βk · cos(k · ω · x) + γk · sin(k · ω · x)] + β0 (10)
k=1
and the aim is to look up the Bre f in B = [B1 , B2 , ..., B91 ],
and, based upon its location, return value α linear interpolated In order to estimate which model represents inverse re-
within the elements of α = [α1 , α2 , ..., α91 ] at each sample lationship α = f (B) with the appropriate accuracy, various
time: regression metrics can be used. In this paper we use 3 metrics:
Root Mean Squared Error (RMSE), Sum of Squared Estimate
(αi+1 − αi ) · (Bre f − Bi ) of Errors (SSE) and Coefficient of Determination (R2 or R
α = αi + (5)
Bi+1 − Bi squared).
It can be done by using Matlab command interp1. RMSE is one of the most commonly used scale-dependent
measures are based on the absolute error or squared errors
B. Regression models [12]. RMSE can be calculated as a square root of the averaged
Regression analysis is one of the most widely used statistical squared difference between actual and predicted value:
tools because it provides simple methods for establishing 
∑mi=1 (yi − f (xi , β ))
2
a functional relationship among variables. It has extensive RMSE = (11)
applications in many subject areas. There are several types m
of regression analysis: linear, multiple linear, and nonlinear. SSE was defined previously as a sum of a squared residuals
Regression analysis includes 4 main steps: model speci- (Eq. 6). R squared is a statistical metric of fitting that indicates
fication, choice of fitting method, model fitting and model how much variation of a dependent variable is explained by
validation. the independent variable in a regression model. R squared can
The form of the model should be specified initially based be calculated by Eq. 12:
on knowledge or objective and/or subjective judgments. After
SSE
the model has been defined, the next step is parameter esti- R2 = 1 − (12)
i=1 (yi − y)
∑m 2
mation or model fitting. The most commonly used method of
estimation is called the least squares method [11]. where
Mathematical formulation of regression analysis in fol-
lowing: suppose, that there is a sample of m pairs (xi , yi ). 1 m 2
Regression model is defined as a f (x, β ). It is necessary to
y= ∑ yi
m i=1
(13)

find such coefficient values β to minimize the Sum of Squared In order to esrtimate coefficient values Curve Fitting App
Estimate of Errors (SSE) - sum of squared differences between was used [13]. The results are presented in Table I.
actual and predicted values or residuals (yi − f (xi , β )) (Eq. 6): According to the results presented in Table I, it is observed
m that generally fitting accuracy can be improved by increasing
SSE = ∑ (yi − f (xi , β ))2 (6) number of terms n. But at the same time complexity of
i=1 modelling is also increasing. The Gaussian model with n = 4
As was mentioned above, variations of the least squares shows appropriate fitting accuracy with RMSE= 0.8133° and
method (such as linear and nonlinear) are usually used to find simplicity of modelling. The coefficients of Gaussian regres-
coefficient values. sion model are listed in Table II.
In this paper pre-calculated susceptance values and corre- Fig. 4 shows selected regression model and residuals.
sponding firing angles are represented as a 91 pairs (xi , yi ) and In order to estimate the elapsed time, both approaches
the aim is to find appropriate regression model f (xi , β )) and were written in M-code. Then 10 iterative calculations of



Authorized licensed use limited to: San Francisco State Univ. Downloaded on November 12,2020 at 15:10:18 UTC from IEEE Xplore. Restrictions apply.
TABLE I: The regression metrics data

Regression model n RMSE SSE R2

Polynomial 2 5.527 2688 0.9572


3 4.012 1401 0.9777
4 3.169 863.8 0.9862
5 2.625 585.9 0.9907
6 2.247 424 0.9932
7 1.968 321.4 0.9949
8 1.754 252.3 0.996
9 1.585 203.5 0.9968

Exponential 1 7.051 4424 0.9295


2 1.708 253.8 0.996

Gaussian 2 1.751 260.5 0.9959


3 1.868 286.2 0.9954
4 0.8133 52.25 0.9992
5 1.663 210.3 0.9967
6 1.222 108 0.9983
7 1.442 145.6 0.9977
8 0.674 30.44 0.9995

Fig. 4: Gaussian model (a) and its residuals (b)


Fourier 1 5.559 2688 0.9572
2 3.188 863.8 0.9862
3 2.261 424.4 0.9932
4 1.768 253.2 0.996
5 1.468 168.5 0.9973
6 1.262 122.6 0.998
7 1.104 91.48 0.9985
8 0.9915 71.76 0.9989

TABLE II: The Gaussian model coefficients

Coefficient Value

β1 , β2 , β3 , β4 5.299 · 1014 , 55.94, 5.31 · 1015 , -0.2926

γ1 , γ2 , γ3 , γ4 -0.3681, -0.2739, -136.6, 0.579

(a) (b)
σ1 , σ2 , σ3 , σ4 0.06588, 0.261, 24.43, 0.1191
Fig. 5: Elapsed time for iteration of B-to-α conversion block
based on Gaussian regression model (a) and look-up table (b)

firing angles were conducted for each approach by using as


IV. S IMULATION
the input vector pre-calculated values of susceptance β =
[90°, 91°, ..., 180°] at each calculation (A total of 910 iterations A. Case system data
were provided for each method). The Matlab timer-function
tic-toc was applied for timing at each iteration. The research was carried out on the single machine infinite
The basic information of hardware and software: Processor bus (SMIB) test system with nominal voltage and frequency
AMD A10-9600P @ 2.4-3.3 GHz, RAM 16 GB, 64-bit level as 220 kV and 50 Hz respectively. The TCR and load
operating system of Windows 10, Matlab version R2019a. The Sload are connected between two double-circuit power lines
elapsed time is visualized by using box plots (Fig. 5). 220 kV, 150 km by using coupling 220/20 kV transformer.
According to the box plot at Fig. 5(a), it should be em- (Fig. 6). The case system data is presented in Table III
phasized that iteration time of Gaussian regression model lies Matlab/Simulink was used for the power flow simulation. The
between 0.034 and 0.088 ms, excepting 68 outlier values. Generator is represented as a Programmable voltage source
Fig. 5(b) shows distribution of elapsed times for look-up table (PV type) and Infinity bus as Three-phase source (swing type).
as following: 100% values lies between 0.108 and 0.40 ms, Simulink was used for the simulation. B-to-α conversion
excepting 34 outliers. That means that generally Gaussian blocks were written in M-code and represented as a Matlab
regression model performance is 3-5 times faster. Function blocks.



Authorized licensed use limited to: San Francisco State Univ. Downloaded on November 12,2020 at 15:10:18 UTC from IEEE Xplore. Restrictions apply.
TABLE III: Case system data

Element Data

Generator P = 300 MW, Q = 0 − 60 Mvar;

TCR Snom = 100 Mvar;

Power line r0 = 0.0960 Ohm/km,


x0 = 0.429 Ohm/km,
b0 = 2.645 · 10−6 S/km;

Load Sload = 200 + 50 j MVA;

Coupling transformer 220/20 kV Snom = 100 MVA, Y /Δ − 11,


Ino−load = 0.5%, Uimpedance = 11%;

220 kV, 150 km 220 kV, 150 km

Fig. 7: The TCR model in Simulink


Point of
'ĞŶĞƌĂƚŽƌ /ŶĨŝŶŝƚĞƵƐ
connection
(POC)
Sload

T 220/20 kV
Vmeas

Pulses
Control Fig. 8: The TCR control system in Simulink

System

Thyristor 2 Thyristor 1

Gaussian regression model leads to decreasing simulation time


Inductor
more than 8 times.

Fig. 6: SMIB test system with TCR.

Fig. 7 and 8 represents the TCR model and its control


system in Simulink.

B. Scenario
During the time period 1 second the Generator changes the
voltage setpoint Vgen,i = [1.0, 1.03, 1.0, 1.06], p.u. at the time
moments Ti = [0, 0.1, 0.4, 0.7], s, accordingly. Sample time is
0.0001 s. The TCR is set to keep voltage in POC as close as
possible to Vre f = 222 kV .

C. Results
Fig. 9 shows r.m.s. voltage (a) and current (b) values in
the POC of TCR with a different types of b-to-α conversion
blocks. The control system of TCR with the Gaussian con-
version block demonstrates a good performance with keeping
voltage VPOC as close as it possible to Vre f = 222 kV while
the Generator changes the voltage setpoint. Fig. 9: VPOC (a) and ITCR (b) with a different types of b-to-α
It can be concluded that the voltage magnitude behavior is conversion blocks
almost the same for the presented type of conversion blocks
which demonstrates proper accuracy of regression model
fitting. The average total simulation time is 1:02 and 8:34 V. C ONCLUSION
minutes for B-to-α Gaussian and look-up table based conver- An approach of modelling TCR for a different studies with
sion blocks respectively. That means that implementation of two types of B-to-α conversion blocks was considered in this



Authorized licensed use limited to: San Francisco State Univ. Downloaded on November 12,2020 at 15:10:18 UTC from IEEE Xplore. Restrictions apply.
paper. Gaussian regression model based B-to-α conversion
block shows the best performance by simplicity of modelling,
accuracy of fitting and decreasing a total simulation time
because, unlike look-up table based B-to-α conversion block,
it doesn’t lead to additional iterational process to search
corresponding firing angle.
The derived equation describing the regression Gaussian
model could be implemented in case if it’s crucially important
to decrease delays in control system of TCR and at the same
time ensure proper accuracy.
R EFERENCES
[1] Mircea Eremia, Chen-Ching Liu, Abdel-Aty Edris, "Advanced Solutions
in Power Systems: HVDC, FACTS, and Artificial Intelligence: HVDC,
FACTS, and Artificial Intelligence", John Wiley and Sons Inc., 2016.
[2] R. M. Mathur, R. K. Varma, "Thyristor Based FACTS Controllers for
Electrical Transmission Systems", 1st Edition, John Wiley and Sons Inc.,
2002.
[3] T. J. E. Miller, Ed., "Reactive Power Control in Electric Systems", John
Wiley and Sons, New York, 1982.
[4] H. K. Tyll, F. Schettler, "Historical overview on dynamic reactive power
compensation solutions from the begin of AC power transmission to-
wards present applications", Power Systems Conference and Exposition
2009 PSCE’09. IEEE/PES, 2009.
[5] N.G. Hingorani, L. Gyugyi, "Understanding FACTS: Concepts and
Technology of Flexible AC Transmission Systems", IEEE Press, N.Y.,
2000.
[6] Vadim I. Zakutsky, Magomed G. Gadzhiev, Nikita S. Yermolov, "Analy-
sis of the influence of installed FACTS devices and transformer on-load
tap changers (OLTC) in transmission networks on a total power losses
Unified Energy System (UES) of Russia", 2019 International Youth
Conference on Radio Electronics, Electrical and Power Engineering
(REEPE), 14-15 March 2019.
[7] Chen P., Chen Z., Bak-Jensen B., "Comparison of Steady-State SVC
Models in Load Flow Calculations", 43rd International Universities
Power Engineering Conference, 2008. UPEC 2008, pp. 1-5.
[8] IEEE Special Stability Controls Working Group, "Static Var Compen-
sator Models for Power Flow and Dynamic Performance Simulation",
IEEE Transactions on Power Systems, Vol. 9, No. 1, February 1994, pp.
229–239.
[9] H. Zhang, C. Dai and S. Wu, "Research on single-phase PLL for the
synchronization of thyristor controlled series capacitor", in Proc. 2012
IEEE PES Asia-Pacific Power and Energy Engineering Conference, pp.
1-5.
[10] T. Saravanan, G. Saritha, V. Srinivasan, "Practical Implementation of
Microprocessor Controlled FC-TCR Compensator for Voltage Regula-
tion", Middle-East Journal of Scientific Research 20 (11): pp. 1586-
1589, 2014.
[11] Samprit Chatterjee, Ali S. Hadi, "Regression analysis by example", 4th
edition, John Wiley and Sons Inc., 2006.
[12] Hyndman, Rob J.; Koehler, Anne B., "Another look at measures of
forecast accuracy", International Journal of Forecasting. 22 (4): pp.
679–688, 2006.
[13] se.mathworks.com. (2019). Apps - Curve Fitting Tool-
box for MATLAB/Simulink. [online] Available at:
https://se.mathworks.com/products/curvefitting/apps.html [Accessed
24 Nov. 2019].
[14] Hildo Guillardi Júnior, Eduardo Verri Liberado, José Antenor Pomilio,
Fernando Pinhabel Marafão, "General-compensation-purpose



Authorized licensed use limited to: San Francisco State Univ. Downloaded on November 12,2020 at 15:10:18 UTC from IEEE Xplore. Restrictions apply.

You might also like