Modeling of HVDC in Dynamic State Estimation

Using Unscented KaIman Filter Method

Hesam Khazraj, F. Faria da Silva and Claus Leth Bak
Department of Energy Technology, Aalborg University
Aalborg, Denmark
[email protected], [email protected], [email protected]

Abstract-HVDC transmission is an integral part of various Load flow and state estimation analysis are important tools
power system networks. This article presents an Unscented for deciding the stable operation and control of power system,
KaIman Filter dynamic state estimator algorithm that considers as weIl as future planning of power systems. High voltage
the presence of HVDC links. The AC - DC power f10w analysis,
which is implemented as power f10w sol ver for Dynamic State AC - DC technology has made considerable advances in
Estimation (DSE), creates an updated admittance matrix. First, recent years and engineers are now considering DC multi-
a hybrid AC / DC network model is developed to combine the terminal network as a feasible option. Therefore, the load flow
AC network and DC links. Then a non-linear state estimator and state estimation techniques are being extended to deal with
can solve for hybrid AC/ DC states by applying the unscented such mixed AC - DC systems.
KaIman filter (UKF) algorithm. It is demonstrated that UKF is
easy to implement and accurate in estimation. The dynamic state State estimation is a key Energy Management System
variables of multi-machine power systems, which are generator (EMS) function, responsible for estimating the state of the
rotor speed and rotor angle, are estimated to study transient power system. Power system is a quasi-static system and
behavior of the power system network.Finally, a dynamic state hence changes slowly with time [5]. State estimation plays an
estimation model is built for a 14 bus power system network to
evaluate the proposed algorithm for hybrid AC/ DC networks.
important role in monitoring, control and stability analysis of
electric power systems. An efficient, timely and accurate state
Index Terms-HVDC Transmission, Power System Transients, estimation is aprerequisite for a reliable operation of modern
Power System Stability, State Estimation power grids. Traditional state estimators cannot capture the
system dynamics very weIl due to the slow updating rate
I. INTRODUCTION of SCADA systems. After Wide-Area Measurement Systems
Demand for reliable supply of electricity is growing, thus (WAMS) wh ich is based on Phasor Measurement Unit (PMU)
many requirements have to be taken into account such as was invented, the way of state estimation process has been
system reliability. One possible solution might be the incorpo- revolutionized. These lead to the development of Dynamic
ration of controllable power components in the system. One State Estimation (DSE) techniques, which enables the dynamic
such component is a High Voltage Direct Current (HVDC) view of power systems in the control center.
link, which is capable of transmitting large amount of power, Extended Kalman Filter (EKF) and Unscented Kalman
while improving stability of system. HVDC has been used Filter (UKF) are two common methods of DSE. In [6] [7],
in power system since 1954 and it can be used for long Extended KaIman Filter dynamic state estimation are used to
distance power transmission and/or for connecting two AC maintain the elegant and efficient recursive update form of the
power systems with different frequencies [1], [2] such as Kalman filter, but this method suffers from serious limitations
50 Hz and 60 Hz. This improves the stability and economy due to linearization and Jacobian matrix calculation. The EKF
of each grid, by allowing exchange of power between non- is probably the most widely used estimation algorithm for
synchronous networks. nonlinear systems. However, more than 35 years of experience
HVDC technologies have undergone considerable advance- in the estimation community has shown that it is difficult
ments in recent years. Utilities can improve transmission line to implement, sophisticated to tune, and only reliable for
efficiencies by controlling the power flow through their sys- systems that are almost linear on the time scale of the updates
tems. One new attractive technology that enables the control [8]. In [9], UKF is used for dynamic state estimation and
of power flow in the system is HVDC transmission. The trend compared study between the UKF and EKF, demonstrated
of power system is towards supplying electric power of high er on three representative test power systems. In [10], it is
quality by improving the system security [3]. The power flow implement UKF for non linear estimation and found that UKF
analysis is very valuable for setting the proper protection consistently achieves a beuer level of accuracy than the EKF
devices to insure the security of system. This analysis is an at a comparable level of complexity.
important tool for deciding stable operating conditions, control The objectives of this paper is to develop the HVDC model
of power system and the future planning of power systems [4]. in AC - DC power flow for UKF dynamic state estimation.
978-1-4673-8463-6/16/$31.00 © 2016 IEEE FinaIly, transient stability of the system will be studied and re-

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on August 16,2024 at 11:16:52 UTC from IEEE Xplore. Restrictions apply.
 liability assessments for HVDC systems in DSE are presented.
J y,
This paper presents an DSE algorithm that takes DC links

~
into consideration. The network model utilized builds on [l1]
and extends to inc1ude HVDC links. The reminder of artic1e
is organized as folIows: Seetion 11, AC - DC Power ftow V
N:l
caIculation and HVDC Model in AC and DC power system
components are described . Seetion III, UKF dynamic state
estimation extended to inc1ude c1assic HVDC links. This UKF-
DSE is validated through simulations in transient stability test
systems in Seetion IV. Seetion V presents conc1usions and
Fig. 1. AC model of network
future work is discussed.
11. AC-DC POWER FLOW CALCULATION AND B. Modeling of ACIDC Network
AC-DC POWER SYSTEM MODEL The authors of [4], [13]-[15], present dynamic modeling of
A. AC-DC POWER FLOW AC - DC systems. Dynamic modeling of AC - DC systems
This seetion explains the formulation of the AC, DC and has been carried out and several power ftow approaches for
hybrid AC / DC power ftow models. steady state AC / DC systems have been proposed.
1) AC power fiow: AC power ftow equations will be Figure 1 shows the i - th line is represented by aseries
formulated with following variables at each bus: admittance Yi and shunt admittance YiO in per unit. An ideal
transformer is represented by the off-nominal tap N : 1 in the
• angle
case of phase shifting transformers, where N is a complex
• Voltage magnitude
number.
• Active power injection
The current 11 and 12 given are by (7) and (8)
• Reactive power injection
At each bus , two variables are known and the remaining two
are obtained by solving nonlinear power ftow equations. The
- - ( 1) -
h = - Yi V2 + Yi + "2Yio VI (7)

complex power can be derived as:

(8)
Sj = P;V + jQ"/ = VjI; = Vj [LYjtVi]* (1)
t= 1

Where P;V and Q"/ are a specified active and reactive power
magnitude at the PV Bus i.
At generator bus

Pj = Vj L Vi [G jt cos(B j - Bt ) + Bjt sin(B j - Bt )] (2) N, : I

Rectifier In ve rter

(3)

At load bus
Fig. 2. DC link model from [15]
Pj = Vj L Vi [G jt cos(B j - Bt ) + Bjt sin(B j - Bt )] (4)
Figure 2 shows a simplified DC link model, where the
Vj and Vi are the voltage magnitude of the two end buses of a
subscript r refers to the rectifier side of the DC link and i
line. Yjt = G jt + j B jt is an element of the system admittance
refers to the inverter. YAC in matrix form result is taken from
matrix.
[15] shows in (9)
Qj = Vj L Vi [G jt sin(B j - Bt ) - B jt cos(B j - Bt )] (5)
Where Pj , Qj is a real-reactive power injection at bus j, and
Bj , Bt are the voltage angles at bus j, t, respectively. ,
[ Yi+ !Y-iO
- N- 2 -
_ 1li.
N
v
Yi
_ J1i...
N*
+ 2Yio
1
1,
(9)

2) DC power fiow: In [12], some assumptions are used for YA C

solving DC power ftow, but in order to limit error within 5% Then matrix form result of YDC shows in (10)
on average, X / R should be greater than 5 and the standard
voltage deviation should be less than 0.0l.
• Voltage magnitudes on all buses are 1p.u.
IVr l cos 0:
0 - 1 0 _ lX lVi i cos <5
[n ~ [T
• sin(B j - Bt ) ~ Bj - Bt , cos(B j - Bt ) ~ 1 ~ r
KNi 0 - 1 - :;rXi
1
V rdc
• Reactive power injections on all buses are ignored 1 1
0 - 1 Vidc
Based on the above assumptions, the active power ftow is: R dc R dc
I dc
Y OC
(6) (10)

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on August 16,2024 at 11:16:52 UTC from IEEE Xplore. Restrictions apply.
 Where K = 0.995* 3~. The AC network system is equivalent
to AC voltage sources Vr and Vi at the AC buses of converter
Measurement
stations, whereas the AC network of the HVDC is equivalent to
injection current sources Ir and I i at the interface AC buses R
and I. In other words, the load is linear and the bus admittance
matrix of the studied AC/ DC system has been reduced to
generator internal buses with converter ac buses remained.
The corresponding reduced bus admittance matrix equation
Prediction
takes the form of (11). The subscript G is used to denote the
Equations
internal generator nodes, and the subscript U is used for all
the remaining nodes.

[ ~~: ~~~] [ ~~] [ie ] (11)

YHH 'yHU ,YUH and Yuu are H H x H H ,Hu X H H ,HH x H u State
and H u x H u complex matrix respectively.Yuu is admit- Estimation
tance matrix; YH H is a diagonal matrix of inverted generator
impedances; YHH and YUH representing in (12) and (13).
1 Fig. 3. Procedure of Kalman Filter

~"~" 1
j x d,
1
j x d2
(12) (18)

o The execution of KaIman Filter techniques in the power

system dynamic state estimation can compute and estimate
Y {
= G Hand x = H,
..".=.L if z dynamic variables of power systems. KaIman Filter has the
J Xd H (13)
HU xz = 0 otherwise ability of incorporating the noise characteristics into the com-
putations and it can be used to predict the state of a system
IH = (YHH - YHU YuJYuH )Ve = YreducedVe (14)
where there is a lot of input noise. Figure 3 Shows kaIman
The matrix Yreduced is the desired reduced admittance matrix. filter procedure.
III. UNSCENTED KALMAN FILTER FOR POWER SYSTEM A. UKF Algorithm
DYNAMIC STATE ESTIMATION The linear approximations applied to non-linear equations
The states of the system, wh ich represents the power system representing the system dynamics can create low accuracy and
by nonlinear algebraic equations, can be caIculated by the reduced estimation performance. These lead to instability in
dynamic state estimation algorithms. The first step in the DSE the altering due to the reason that high order terms in the
process is to identify the mathematical modeling for the time Taylor series expansion are neglected. The main idea of Un-
behavior of the power system. By using the mathematical scented Transform (UT) is to obtain a set of deterministically
model of the system and collected measurement data, DSE chosen er points which capture the mean and covariance of
predicts the dynamic state vector one step ahead. the original distribution of x , where X is an w - dimensional
A dynamic system can generally be modeled as set of non- random variable with me an q and covariance Pee . Assumption
linear differential equations: of the UT, can be summarized as follow [9], [16]:
dx • Parameter A = a 2 (w + c) - w
dt = f(x,u,w) (15) • 1O- 4 :s; a :s; 1 and c = 3 - w or c = 0
• Square root matrix can be approximated P = AA T ,
In (15), f(o) is the system function, x vector represents the
where A is lower tri angular matrix obtained from the
state variables; the u vector represents the algebraic variables
Cholesky factorization of P.
and w stands for process (system) noise.
Then suppose another random variable y wh ich is related
Xc = Xc- I +f (Xc- I, Uc- I, w c- d 6.t = 9 (Xc- I, Uc - I, Wc- I) to X through a non-linear function as shown at (19)
(16) y = f(x) (19)
c - l is the present instant of time index, c is the next instant of
time index and 6.t is the time step. The measurements at time
the 2w + 1 sigma points can be found as
step c can be represented as a vector of non-linear functions x i = q + h/(w + A))i ,i = I,2, ... ,w (20)
h( 0) in terms of the state variables X and measurement noise
'IjJ as (17) and resulting error between the measured and
x Hw =q- (J(w + A)Pee) i, i = 1,2, ... , w (21)

caIculated values is given by (18) The Unscented KaIman Filter (UKF) can be written as:
(17) (22)

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on August 16,2024 at 11:16:52 UTC from IEEE Xplore. Restrictions apply.
 (23)
Wc rv (0, Qc) (24)

U c rv (0, Qc) (25)

Where noise covariance matrix is Qc. e is an w - dimensional
random variable with me an q, covariance Pee . Y is a second
random variable which is related to e through a non-linear
function. The sigma points (0") are then propagated to calculate
the mean band covariance Hb of b. (J
(w + >. )Pee )i is the ith
row or column of matrix square root of the( w + >')Pee . The
time update equations from c - 1 to c are completed and the
measurement update equations need to be implemented in the
final part of the UKF algorithm. in order to take the process Fig. 4. Proposed 14 bus system [11] with line replaced by HYDC link
noise into account, the apriori error covariance is estimate by
adding Qc- l to the end of equation. Peb is the cross covariance RellciorR R{,lIciorl

between e c and bc . (26), (27) and (28) are shows measurement

update of the state estimate.
K c = Pebp b- 1 (26)

et = e;:- + K c [bc - q (27)

Pc+ = Pc- - Kc PbK~ (28)

where K c is the KaIman gain matrix, i t is the state
estimate and Pc+ is the estimation error covariance matrix.
The subscript + indicates the estimate is in an aposteriori
Fig. 5. Proposed HYDC link
estimate and the subscript - denotes the estimate is in an
apriori estimate. The damping power is expressed by the
damping constant, in seconds per electrical radian, times first IV. SIMULATION RESULTS
derivative of the rotor angle wh ich is specified as (29):
Figure 4 and Figure 5 show proposed 14-bus Power Net-
di\ work and HVDC link, respectively. In this section, a case
D
' dt- -- Dw
t ,
(29)
study 14-bus power network is taken from [11] and simulated
B. Estimation Error in PowerFactory (PF). Line 1-5 located between bus NO.1
(30) shows the estimation error for the time instant 9 and bus NO.5 is replaced with one HVDC link. Result of
AC - DC power ftow are taken from Digsilent. Matlab is used
1 2t . . to develop and model AC / DC admittance matrix. HVDC
W = 2t L 1 x~ - X~kJ 1
(30) modeled for power network through power ftow and mixed
i= l admittance matrix. For the transient study of the system, line
where 2t is the number of states (t rotor speed and t rotor 4 - 5 disconnected at t = 2. The dynamic variables generator
angle), x g is the actual state vector calculated at the end of rotor speed w and the rotor angle 0 are estimated by using UKF
transient stability analysis and x g kJ represents the estimated techniques. The estimation process is simulated along with the
state vector as a result of the kaiman transient stability computation procedure on proposed power
filtering process. system. During the simulations, a random Gaussian noise
The estimation error for the time instant 9 is calculated for is assumed with mean equal to zero and standard deviation
the generator rotor speed w variables and the generator rotor equal to 0" = 10- 2 for both system noise and measurement
angle 0 variables by using same formula separately as shown noise. The time step is set as 6,t = 0.002. The mechanical
in (31) and (32): power input Pm i for each generator is constant throughout

Ww = g:ax ~ [~ t I w~ - W~kJ I] (31)

the transient stability procedure. The first order differential
equations representing the second order swing equation for a
synchronous generator in a multi machine power system can

g:ax ~ [~ t lo~ - O~kJ I]

be expressed as follow by using Equation 33 and Equation 34
for describing the electric output power.
Xs = (32)
dO
-dti --wo, (33)
where t is the number of wand 0 in the state vector.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on August 16,2024 at 11:16:52 UTC from IEEE Xplore. Restrictions apply.
 n
M
dWi
- -- Pm,·- EG
t dt i tt
-
2 '
L IE ill Ej l [Bij Sin(Oi - Oj) + Gij COS(Oi - Oj)) , i = 1, 2, 3, .... , n (34)
j = l
j -=J i

- W2(k)

- - - w~~

0.9995

0.999 I

cn- 43.5 :
::J
Cl. 0.9985
..,
Q)
- 04(k_1)
.E: - - - s:(~_1)
1
~ 0.998
oe" - 44
8
0.9975

0.997 - 44.5

0.9965

4 10 12 14 16 18 20 4 6 10 12 14 16 18 20
time (5) time (5)

Fig. 6. W2 estimated by UKF Fig. 8. Estimation of 0 (4- 1) by UKF

high in accuracy. The genera ted measurements from each unit

, were given as input Equation 14 to the corresponding UKF
- 45.4 ,
,
Equation 22. The caIculated states (Blue lines), along with
- 45.6
, their estimated values (red dashed Iines), were plotted for
g> - 45.8 :
, each unit. Estimation accuracy can be seen in figures for every
" - 46 : dynamic state, the plot of estimated values almost coincides
c:
•- ;:- - 46.2 , with those of the caIculated values .
A - 46.4 ' - 03(k_1)
Thus, it is evident that the HVDC link is implemented in
M ,
00 I - - - o;~~_1) dynamic state estimation and decentralized UKF scheme are
- 46.6
generates accurate estimates of all the dynamic states. they
- 46.8 can be reliably used for further control and security decisions
- 47 because wand 0 can been estimated with high accuracies.
- 47.2
V. CONCLUSIONS
6 10 12 14 16 18 20
time (5) The overall purpose of this research was to develop model
of HVDC in dynamic sate estimation and monitor effect of it
Fig.7. Estimation of 0 (3-1) by UKF in transient study of generators. For this purpose, AC - DC
power ftow is implemented into dynamic state estimation.
In order to achieving the research objectives, derivation of
The rotor speed W2 is estimated by using UKF where the proposed AC - DC is completed. The load data from
AC - DC power ftow and Ymatrix are implemented and shown the prefault power ftow are converted to equivalent admit-
on Figure 6. Figure 6 to Figure 9 show that UKF methods are tances. The data for this step are obtained from the result of
capable of capturing the transient responses of the states. k - 1 power ftow study. The power system dynamic state estimation
is the present relative rotor speed and /:).t = 0.002 is the time process, based on KaIman Filtering techniques, is discussed.
step . The relative rotor angle 03 and 04 are estimated by using The dynamic state variables of multi-machine power systems
UKF and are shown in Figure 7 and Figure 8, respectively.The wh ich are generator rotor speed and generator rotor angle
rotor speed W 5 is estimated by using UKF shown on Figure 9. are estimated. UKF was selected as a current method and
The UKF is suitable for this case, because it was recommended implemented for 5 generator of the 14 bus to tracking transient
to apply in non-linear system because its performance is quite on rotor speed and relative rotor angle. The UKF algorithm

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on August 16,2024 at 11:16:52 UTC from IEEE Xplore. Restrictions apply.
 [12] S. Kim and T. Overbye, "Hybrid power flow analysis: Combination of
ac and dc models," in Power and Energy Conference at lllinois (PECI),
2011 IEEE, Feb 2011, pp. 1-4.
I [13] M. Baradar, M. Ghandhari, D. Van Hertem, and A. Kargarian, "Power
I
flow calculation of hybrid ac/dc power systems," in Power and Energy
0.9995
I Society General Meeting, 2012 IEEE, Jul y 2012, pp. 1-6.
[14] L. Shen, W. Wang, M. Barnes, and J. Milanovic, "Integrated ac/dc model
for power system stability studies," in AC and DC Power Transmission,
I
0.999
11th lET International Conference on, Feb 2015, pp. 1- 7.
::J [15] W. Li and L. Vanfretti, "Inclusion of classic hvdc links in a pmu-based
Q. state estimator," in PES General Meeting - Conference Exposition,
C 2014 IEEE, July 2014, pp. 1- 5.
Sir) 0.9985 [16] D. Simon, Optimal state estimation: Kaiman, H infinity, and nonlinear
approaches. John Wiley & Sons, 2006.

0.998

0.9975

5 10 15 20
time (s)

Fig. 9. W5 estimated by UKF

wh ich is based on unscented transformation is introduced as

a effective method. It is demonstrated that UKF is easy to
implement and accurate in estimation.
REFERENCES
[1] H. Wang and M. Redfern, "Enhancing ac networks with hvdc intercon-
nections;' in Electricity Distribution (CICED), 2010 China International
Conference on, Sept 2010, pp. 1-7.
[2] J. Setreus and L. Bertling, "lntroduction to hvdc technology for reliable
electrical power systems," in Probabilistic Methods Applied to Power
Systems, 2008. PMAPS '08. Proceedings of the 10th International
Conference on, May 2008, pp. 1-8.
[3] K. Basu, "Stability enhancement of power system by controlling hvdc
power flow through the same ac transmission line," in lndustrial Elec-
tronics Applications, 2009. ISIEA 2009. IEEE Symposium on, vo1. 2,
Oct 2009, pp. 663-668.
[4] L. Gengyin, Z. Ming, H. Jie, L. Guangkai , and L. Haifeng, "Power flow
calculation of power systems incorporating vsc-hvdc," in Power System
Technology, 2004. PowerCon 2004. 2004 International Conference on,
vo1. 2, Nov 2004, pp. 1562-1566 Vo1.2.
[5] N. Shivakumar and A. Jain, "A review of power system dynamic
state estimation techniques," in Power System Technology and IEEE
Power India Conference, 2008. POWERCON 2008. Joint International
Conference on, Oct 2008, pp. 1-6.
[6] Z. Huang, K. Schneider, and J. Nieplocha, "Feasibility studies of apply-
ing kaiman filter techniques to power system dynamic state estimation,"
in Power Engineering Conference, 2007. IPEC 2007. International, Dec
2007, pp. 376-382.
[7] H. Tebianian and B. Jeyasurya, "Dynamic state estimation in power
systems: Modeling, and challenges," Electric Power Systems Research,
vo1. 121 , pp. 109-114, 2015.
[8] S. Julier and J. Uhlmann , "Unscented filtering and nonlinear estimation,"
Proceedings of the IEEE, vo1. 92, no. 3, pp. 401-422, Mar 2004.
[9] G. Valverde and Y. Terzija, " Unscented kalman filter for power system
dynamic state estimation," Generation, Transmission Distribution, l ET,
vo1. 5, no. 1, pp. 29-37, Jan 2011.
[JO] E. Wan and R. Van Der Merwe, "The unscented kalman filter for
nonlinear estimation," in Adaptive Systems for Signal Processing, Com-
munications, and Control Symposium 2000. AS-SPCC. The IEEE 2000,
2000, pp. 153- 158.
[11] L. Freris and A. Sasson, " lnvestigation of the load-flow problem,"
Electrical Engineers, Proceedings of the Institution of, vol. 115, no. 10,
pp. 1459- 1470, October 1968.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on August 16,2024 at 11:16:52 UTC from IEEE Xplore. Restrictions apply.