You are on page 1of 10

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2709463, IEEE
Transactions on Smart Grid
1

Fast Decoupled State Estimation for Distribution


Networks Considering Branch Ampere
Measurements
Yuntao Ju, Member, IEEE, Wenchuan Wu, Senior Member, Fuchao Ge, Student Member, Kang Ma,
Member, IEEE, Yi Lin and Lin Ye, Senior Member

Z base Complex impedance base


Abstract Fast decoupled state estimation (FDSE) is proposed Pi Active power injected at bus i
for distribution networks, with fast convergence and high
Qi Reactive power injected at bus i
efficiency. Conventionally, branch current magnitude
measurements cannot be incorporated into FDSE models; avg , avg , Heuristic parameters calculated according to
however, in this paper, branch ampere measurements are network parameters and load condition
reformulated as active and reactive branch loss measurements Phase angle vector with n dimension
and directly formulated in the proposed FDSE model. Using the v Voltage magnitude vector with n dimension
complex per unit normalization technique and special chosen x State variables vector
state variables, the performance of this FDSE can be z Measurements vector
guaranteed when it is applied to distribution networks. Active power measurements vector, include
za
Numerical tests on seven different distribution networks show voltage phase angle measurements
that this method outperforms Newton type solutions and is a Reactive power measurements vector, include
zr
promising method for practical application. voltage magnitude measurements
Calculated values responding to active power
ha
Index TermsDistribution network, Fast decoupled state measurements
estimation, Ampere measurements Calculated values responding to reactive power
hr
measurements
NOMENCLATURE a Active power measurements error
Sbase Complex power base r Reactive power measurements error
Ibase Complex current base H Measurement Jacobian
1 Weight matrix
Sbase Magnitude of complex power base R
Constant measurement Jacobian matrix for
Base angle, rad. Ba
base active power measurements
m Number of branches Br
Constant measurement Jacobian matrix for
Xi Reactance of branch i reactive power measurements
Square of branch current magnitude from bus
Ri Resistance of branch i I ij2
i to bus j
l Number of PQ load G Gain matrix
n Number of state-undetermined buses m Measurements value
r Number of slack buses Average value
U base Complex voltage base Standard deviation
U base Magnitude of complex voltage base

I. INTRODUCTION
estimators (SEs) are significant data processors in
This work was supported partly by the National Science Foundation of
China (Grant.51477083) ,the State Grid Fujian electric power company
Economic institute of technology science and technology project (Grant
S TATE
electric power control centers (EPCCs), providing reliable
information for real time markets [1]. Distribution system state
SGFJJY00GHWT1600081) and the State Key Laboratory foundation project
(Grant SKLD16KZ08). estimation (DSE) is indispensable for active distribution
Yuntao Ju and Lin Ye is with China Agricultural University, Beijing, China network modeling and integrated operation with distributed
(e-mail: juyuntao@cau.edu.cn, yelin@cau.edu.cn). energy resources [2].
Wenchuan Wu is with Tsinghua University, Beijing, China (corresponding Due to budget limitations, on many feeders, only branch
author, email: wuwench@mail.tsinghua.edu.cn).
Yi Lin is with Fu Jian Electric Company (email: jyy.lin_yi@fj.sgcc.com.cn) current magnitude measurements (BCMMs) are deployed to

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2709463, IEEE
Transactions on Smart Grid
2

monitor operating conditions. Some investigations have shown In summary, to the best of our knowledge, with the complex
that these real-time BCMMs can significantly improve the pu technique, FDSE can be implemented on distribution
accuracy of load estimation in DSE [3]. However, BCMMs are networks with promising convergence; however, incorporation
usually ignored in transmission SE due to their coupled of BCCMs in an FDSE model remains a problem requiring a
property and mathematical problems [4], such as zero elements solution. In a typical distribution network, branch current
in the measurement Jacobian evaluated at flat start and multiple measurements represent the major portion of their limited
solutions [5]. real-time measurements. In this paper, a novel fast decoupled
In DSE, BCMMs are usually handled indirectly. In [3], a distribution state estimator (FD-DSE) is proposed. The branch
strategy with two stages is proposed. In the first stage, the ampere measurements are equivalently reformulated as the
branch current measurements are disregarded. In the second branch active power and reactive power losses, and the power
stage, the BCMMs are approximately formulated as linear loss measurements can be incorporated into the FD-DSE model
complex measurements, with the phase angle estimated in the in a decoupled manner. Numerical tests show that this FD-DSE
first stage, and the updated SE model is iteratively solved. In [6], has a low computational burden with fast convergence.
BCCMs, and active and reactive power measurements, are all
transformed to linear complex current measurements based on II. BRIEF REVIEW OF FAST DECOUPLED STATE ESTIMATION
estimated phase angle and voltage. Clearly, these solutions will By introducing the complex per unit normalization [21],
introduce additional errors due to measurement transformation. normalized r/x ratio of distribution network becomes smaller, a
BCCMs can be directly formulated in DSE with special fast decoupled algorithm can be realized with robust
selection of state variables. A branch current-based DSE is performance. The main concept underlying complex per unit
proposed to accommodate large-scale BCCMs, where the normalization is explained briefly here.
magnitude and phase angle of the branch current were selected
as the state variables in [7]. Another formulation of DSE adopts A. Complex per unit Normalization
branch current magnitude and power as state variables [8]. First, a complex volt-ampere base is given by:
However, these methods suffer from a heavy computational Sbase Sbase e j , base
(1)
burden for large-scale networks, because repeated factorization
of the gain matrix is involved. where Sbase denotes complex power base, Sbase denotes
Fast decoupled state estimation (FDSE) algorithms are used magnitude of complex power base, base denotes angle base.
widely for transmission networks [9], [10] [11] due to their high
The voltage base is given by:
efficiency. FDSE has become a standard method for SE in
U base U base j 0 (2)
energy management systems (EMSs) [12].
Numerical tests on variable transmission networks indicate where U base denotes complex voltage base, U base denotes
that the gain matrix of SE changes slightly during the iteration magnitude of complex voltage base.
procedure [9]. Under certain assumptions, a constant gain According to (1) and (2), the current and impedance base will
matrix in a decoupled pattern can be obtained. The right hand be:
side (RHS) of normal equations can also be decoupled [10]. 2
U base U2
Ill-conditioned problems of SE occur when there are small Z base *
base e jbase (3)
Sbase Sbase
impedance branches [13], weights that are too large for virtual
*
zero injection measurements, etc. These problems can be Sbase S
I base base e jbase (4)
addressed by normal equations with constraints [14] [15] or U base U base
Hachtels augmented matrix method [16]. A fast decoupled
where I base denotes complex current base, Z base denotes
formulation for SE based on Hachtels sparse tableaux
approach is proposed in [12]. To improve the numerical complex impedance base.
stability of FDSE, an orthogonal row processing algorithm can The determination of the base angle is achieved by:
be used with Givens transformations. [17]. Because the avg avg
base 1 (5)
orthogonal transformation needs to store a dense and high 2 2
dimension transformation matrix, it is computationally where
complex and not highly practical. Decoupled SE can also be m
Xi
carried out in rectangular form via transformation of active and tan 1

reactive measurement pairs to linear complex current avg
i 1 Ri ,
measurements [18]. FDSE shows bad convergence m
performance for networks with high r/x ratios. Therefore, a new X X
tan 1 i tan 1 i
FDSE is required, maintaining nonzero coupling in the i max
R Ri min
Jacobian off-diagonal matrix block in an iterative procedure to avg ,
2
improve performance under high r/x ratios [19]. l Qi
Several practical solutions have been developed to resolve cos tan 1

the problem in FDSE of high r/x ratios in a distribution network. i 1 Pi
1 ,
A rotation-based method has been presented to handle this l
difficulty [20]. O. L. Tortelli et al. propose a fast decoupled where m denotes the number of branches, X i and Ri are the
algorithm via complex pu normalization [21] for distribution reactance and resistance of branch i , l is the number of PQ
networks.

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2709463, IEEE
Transactions on Smart Grid
3

load, Pi and Qi are active and reactive power injected at bus i, BrT Rr 1 Br v BrT Rr 1
zr hr v, (14)
avg , avg , are heuristic parameters calculated according to Because of its high efficiency and low memory cost, FDSE
network parameters and load conditions. with (13) and (14) has become a standard algorithm in EMSs.
With the above-mentioned complex per unit normalization, C. Difficulties in Accounting for Branch Current Magnitude
the normalized r/x ratio of a distribution becomes smaller; a Measurements in FDSE
fast-decoupled algorithm can be used for distribution networks.
The branch current can be expressed as [23]:
B. Fast Decoupled State Estimation I ij2 gij2 bij2 vi2 v 2j 2vi v j cosij (15)
For an electrical power network, the state variables are
The measurement Jacobian for the square of BCCM is given
composed of phase angle and voltage magnitude v .
by:
The objective of the weighted least squares SE model is:
I ij2
min Ri1 ( zi hi (x)) 2 (6) 2 g ij2 bij2 vi v j cos ij (16)
i m vi
where x {vi ,i } . I ij2
2 g ij2 bij2 vi v j sin ij (17)
This SE problem can be solved using the Newton method, i
and the iterative equation is [1]: For flat start, assume that vi v j 1 , i j , (16) and (17)
x = G 1 H T R 1 z h x (7) are both equal to zero and are useless at flat start. Reference [23]
h x concludes that BCMM cannot be decoupled or used in FDSE.
where G H RH , H = T
.
x Since BCMMs occupy a large portion of real-time
The measurement vector can be divided into active and measurements in distribution networks, they should be
reactive power parts: exploited thoroughly in DSE.
za ha , v a
z (8) III. PROPOSED DISTRIBUTION STATE ESTIMATION MODEL
zr hr , v r
The voltage vector ( vi i ), and real and reactive branch
The measurement Jacobian matrix can be divided into two power flow ( Pij , Qij , Pji , Q ji ) are selected as state variables.
parts and is written as:
Therefore, there are 4m additional state variables compared
ha ha
Haa Har with the conventional SE model. Let z represent the
H = v = (9) measurement vector, and Ri1 the weight of the ith measurement
hr hr Hra Hrr
The measurement equation h( y) relating the measurement
v
vector and the state vector includes:
1
The weight matrix R can also be divided into two parts: 1) Real and reactive power measurements of the branch:
Ra 1 P Pij Pij , Qij Qij Pij
m m
1 ij (18)
R 1
(10)
Rr where represents the measurement error, and the
superscript m denotes measurement quantities.
hr ha
Considering that 0, 0 after complex per 2) Injection power measurements:
v
Pi Pij P , Qi Qij Qi ,cap Q
m m
(19)
0, vi 1 , the
i i
unit normalization, and assuming that ij ji ji

where j i denotes that j is connected to i and Qi ,cap is


gain matrix is expressed by:
ha reactive power injected by capacitors.
Ba (11) 3) Branch current measurements:
I ijm 2 rij Pijm,loss Pij Pji Pij ,loss (20)
hr
Br (12) I m2
ij x Q
ij
m
ij , loss Qij Q ji v b 2
i sh , ij v b
2
j sh , ij Qij ,loss (21)
v m
According to the proposition of fast decoupled power flow where bsh,ij is the 1/2 branch charging susceptance. I denotes ij

1 a branch current measurement, and subscript loss denotes


[22], Ba is formulated with , where xb is the branch branch loss. Since the branch charging current for a distribution
xb
network is very small, these two approximate equations have
reactance and Br is formulated with the imaginary part of the only trivial errors.
branch admittance. When BCCMs are the only ones to make the branch
Finally, the FDSE is formulated with two incremental observable, the SE solution is not unique [23]( chapter 9). It
equations. First, the phase angles are corrected using: means BCCMs cannot be critical measurements to make the
branch observable, i.e. BCCMs are redundant measurements to
BaT Ra 1 Ba BaT Ra 1 za ha v, (13)
make SE more accurate.
Then, the voltage magnitudes are corrected using: 4) Bus voltage measurements:

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2709463, IEEE
Transactions on Smart Grid
4

vi vi Ui represent additional real and reactive branch power flow as


m
(22)
additional state variables. Vmag represents voltage magnitude
5) Virtual measurements:
Because 4m additional state variables have been introduced measurements.
into state estimation, 4m extra constraints should be The corresponding elements of matrix H are described as:
supplemented, as virtual measurements, to insure the Pi Qi Pij Qij Pij ,loss Pij ,loss
1 , 1 , 1 , 1 , 1 , 1 ,
observability of the system. Pij Qij Pij Qij Pij Pji
For a branch, the branch active and reactive power
constraints can be formulated as virtual measurements and Qloss Q fij , p ,v
1 , loss 1 1,
expressed as: Q ji Qij Pij
rij vi v j cos i j +xij v j sin i j fij , p ,v
f p , v =Pij -vi 2
r 2
xij
=0 (23) 2vi g v j g cos ij v j b sin ij ,
ij vi

xij vi v j cos i j +rij v j sin i j fij , p ,v
f q , v Qij -vi =0 (24) vi v j g sin ij b cos ij ,
rij2 xij2 i
Similar constraints can be derived for Pji and Q ji flow from fij , p ,v
vi g cos ij b sin ij ,
the other side of the branch. Effects of charging susceptance are v j
neglected in (23) and (24).
fij , p ,v
For shunt capacitors, the branch power flow constraints are vi v j g sin ij b cos ij ,
formulated as: j
Qi ,cap -vi2bi ,cap =0 (25) fij , q ,v fij , q ,v
where bi , cap is the susceptance of capacitors. 1, 2vi b yc v j g sin ij b cos ij ,
Qij vi
The whole measurement Jacobian is expressed as H , where
fij , q ,v
v Pij Pji Qij Q ji vi v j g cos ij b sin ij ,
i
Pinj Pinj
fij , q ,v
0 vi g sin ij b cos ij ,
0 0 0
Pij Pji v j
Qinj Qinj
0 fij , q ,v Vi
vi v j g cos ij b cos ij ,
0 0 0
Qij Q ji 1 . Here g and b
j vi
Pflow Pflow denote branch admittance, yc denotes shunt susceptance.
0 0
Pij Pji
0 0
It shows that no elements are faced with zero problem under
Q flow Q fllow the assumption ij 0 . So, the proposed measurement model
0 0 0 0
Qij Q ji can cope with branch currents model in a decoupled manner.
Ploss Ploss FDSE depends on the PQ decoupled properties of the
0 0 0 0 measurement equations. Branch power measurements (18),
H Pij Pji injection power measurements (19) and branch loss

0 Qloss Qloss measurements (20)(21)(25) all have PQ decoupled properties;
0 0 0
Qij Q ji
however, the virtual branch constraints (23)(24) are involved in
both v and . It has been observed that v has little impact on
f p ,v f p ,v f p ,v f p ,v active power and that has little impact on reactive power for
0 0
v Pij Pji a distribution network with high R / X in a complex per unit
f f q ,v f q ,v f q ,v system [21], and that sin i j i j , cos i j 1 ,
q ,v 0 0
v Qij Q ji and v j 1 . Therefore, according to the proposition from fast
decoupled power flow, the elements of the measurement
Vmag ,
0 0 0 0 0 Jacobian for the virtual branch constraints of (23)(24) can be
v
formulated as:
Pij i j
where Pflow , Q flow represent active and reactive power flow (26)
vi xij
measurements vectors, Ploss , Qloss represent active and reactive Qij xij vi v j
(27)
branch power loss measurements vectors, Pinj , Qinj represent vi r 2
ij xij2
active and reactive power injection measurements vectors,
f p,v , f q,v represent virtual measurements provided by branch
constraints in equation (23) and equation (24), Pij , Pji , Qij , Q ji

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2709463, IEEE
Transactions on Smart Grid
5

For FD-DSE, the measurement jacobian matrix is expressed The usage of complex pu normalization in a state estimation
in equation (9), where Har =0, H ra =0, and the other two problem includes three steps:
(1)The key for complex pu normalization method is to
matrixes are estimate the base angle base , according to equation (5), this
Pij Pji value is mainly determined by network parameters and load
factor. In distribution network, load factor can be estimated
Pinj Pinj
0 according to load type.
Pij Pji (2)Calculate complex volt-ampere base, impedance base and
Pflow Pflow current base according to the estimated base angle base .
0
Pij Pji (3)All measurements and network parameters are
H aa normalized according to the base values using complex pu
0 Ploss Ploss normalization.
Pij Pji
Initialized with
f p , v f p , v f p , v vi i 10
Pij 0
Pij Pji
Qij 0
v Qij Q ji
Qinj Qinj P BaT Ra 1 Ba

BaT Ra 1 za ha v, , Pij , Qij
0 Iteration Pij

Qij Q ji
Qv
Q flow Q flow v
BrT Rr 1 Br BrT Rr 1
0 Iteration Qij
zr hr v, , Pij ,Qij

Qij Q ji
.
Qloss Qloss
H rr 0 No
Qij Q ji v, , Pij , Qij

f q , v f q , v f q , v
Yes
v Qij Q ji
V Stop
mag 0 0
v
Figure 1 Flowchart of the whole fast decoupled distribution state estimator
(FD-DSE)
The corresponding elements of matrix H aa and H rr are Discussion on its observability: For a distribution network
described as: with m branches, 4m additional state variables are involved in
Qi Qij Q ji ,loss f V this SE model, and branch constraints (23)(24) can provide a
1 1 1 ij , q ,v 1 i 1
Qij Qij Q ji Qij vi total of 4m extra equations in the same time. Therefore, the
proposed SE model has the same observability as a
Pi P P P
1 , ij 1 , ij ,loss 1 , ij ,loss 1 . conventional SE. Because the dimensions of Ba and Br being
Pij Pij Pij Pji
larger than those of the conventional SE is not very important
The Jacobian matrix entries keep constant during the iteration. for sparse matrices [14], the proposed FDSE has promising
fij , p ,v 1 computational performance for practical applications.
Every nonlinear entries are approximated with ,
i xij
IV. BAD DATA ANALYSIS
fij , q ,v xij
, according to fast decoupled power flow Chi-squares test is used for detecting bad data [23].
vi rij2 xij2 Normalized residuals are usually used for bad data
solution proposed in[22]. Other elements are either 1 or 0. identification [32]. Hypothesis testing identification (HTI) can
be used for identifying multiple errors [33]. In proposed
The element of the measurement Jacobian for the shunt FD-DSE, HTI is used to identify bad data and the analysis
capacitor is: procedure is described as followings:
Qi ,cap (1)suspect measurement set s is selected according to
bi ,cap (28)
vi normalized residuals r N and calculate es S ss1 z z ,where
By adopting the above measurement Jacobian formulation
Sss represents residual sensitivity matrix corresponding to
scheme, a fast decoupled distribution SE (FD-DSE) can be
formulated, in which the two Jacobian matrixes for the P- and suspect measurement set. z denotes estimated measurements
Q-V iterations are both constant. The calculation procedure for value and z denotes measurements value.
the FD-DSE is similar to the FDSE introduced in (13) and (14)
[23]. The whole FD-DSE flow chart is depicted in Fig. 1.

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2709463, IEEE
Transactions on Smart Grid
6

esi i N Tii 1 P2m P21 a ,1 (29)


(2)Calculate N , where

1 i i Tii Q Q21 r ,1
m
2 (30)
2
The equations for the virtual measurements are expressed as:
T Sss1 , i2 variance, denotes the probability of making an
v1r v1 v1 cos 1 2 +v1xv2 sin 1 2
error in rejection of valid measurements, N b ,(for example, 0 P12 (31)
b 2.32 for 0.01 ).
2
r2
x
v2r v2 v2 cos 2 1 +v2 xv1 sin 2 1
(3)If 0 N N , i i Tii N , where 0 P21 (32)

1 i
2

1 max
2

1 i
2
r 2
x2

N 3.0 .If v1x v1 v2 cos 1 2 +v1rv2 sin 1 2


0 Q12 (33)
1 max
2
r 2
x2
N 0 , i i Tii N
.If N N , v2 x v1 v1 cos 2 1 +v2rv1 sin 2 1
1 i 1 i max 1 i 1 max 0 Q21 (34)
r x2
2 2 2 2 2

i i Tii N
.
1 i max The power loss measurements are given by:
2 m
Ploss P12 P21 (35)
(4)Taken as suspect measurements if esi i .
Q m
loss Q12 Q21 (36)
(5)Repeat steps 1-4 until all measurements that are suspected
Since all the measurements satisfy PQ decoupled
in the previous iteration are all selected again at step 4. Bad data
characteristics, matrix Ba is given by:
identification algorithm can be accelerated using algorithm in
[35]. 2 P12 P21
The detail of HTI based bad data identification can refer to P2m 0 0 1
0m 1 x 1 0
[23] and [32].
(37)
Ba m
V. SIMPLE ILLUSTRATIVE SAMPLE 0 1 x 0 1
m
To explain the procedure of the proposed method more Ploss 0 1 1
clearly, the FD-DSE is implemented on a simple two-bus Similar to the active power part, Br is formulated as:
sample network. v2 Q12 Q21
U1 1 0 r jx U 2 U 2 2
Q2m 0 0 1
P12 jQ12 P21 jQ21 0m
x r x 1 0
2 2
(38)
Br m
0 x r x 0 1
2 2

m

Qloss 0 1 1
P2 jQ2
Then, according to (13) and (14), this SE model can be
Figure 2 Two-bus illustrative sample system
solved in a decoupled manner. The FD-DSE for this simple
The original impedance, and active and reactive power under
illustrative sample network converged after three iterations.
a conventional per unit system, are r +jx = 0.0967 + 0.0397j p.u.
It can be seen that Ba and Br have zero pivot in
and P2 + jQ2 = 0.0042 + 0.0026j. According to (5), the base
angle for complex per unit normalization can be obtained as lower-upper triangular (LU) decomposition, which can be
base 1.3581 rad. Then, by implementing complex unit accounted for easily with an MC64 pivoting algorithm [24] to
ensure the numerical stability of the FD-DSE.
normalization on this simple two-bus system, measurements
can be generated with original power flow adding errors, which
satisfy a Gaussian distribution. The measurement parameters
VI. TEST CASES
for the Gaussian distribution are given in Table I.
The FD-DSE was programmed in C++ and tested on several
TABLE I practical or hypothetical radial and meshed distribution
PARAMETERS FOR DISTRIBUTIONS OF MEASUREMENTS networks [25]. Six cases on radial or meshed networks, with all
Measurement
branches deployed with BCCMs and all loads equipped with
P2 -1.6548e-3 1.656e-9
active and reactive power measurements, were designed for
Q2 4.6542e-3 4.655e-9 testing. The average values for all measurements were the
Square BCCM of 0.17997 1.7997e-7 original power flow values. The measurement errors satisfy a
branch 1-2 Gaussian distribution, with standard deviations given by
Active power loss -0.003310 3.31e-9 106 z pf 1012 (39)
of branch 1-2
Reactive power 0.009311 9.31e-9 where z pf denotes the power flow result for this measurement.
loss of branch 1-2
(1) Case 1: 29-bus test feeder, with 1 reference bus and 1 tie
branch [26];
Then, the active power injection measurements are given by:
(2) Case 2: 32-bus test feeder, with 1 reference bus and 5 tie

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2709463, IEEE
Transactions on Smart Grid
7

branches [27]; PERFORMANCE OF THE TWO DIFFERENT SE METHODS IMPLEMENTED ON A


RADIAL NETWORK
(3) Case 3: 83-bus test feeder, with 11 reference buses and 13
Maximum
tie branches [28]; FD-DSE Newton DSE
voltage deviation
(4) Case 4: 135-bus test feeder, with 8 reference buses and 21 Case
CPU CPU between FD-DSE
Iter. Iter.
tie branches [25]; time(s) time(s) and Newton DSE
(5) Case 5: 201-bus test feeder, with 3 reference buses and 15 1 7 0.0088 5 0.0338 6.25 106
tie branches [25]; 2 6 0.0093 5 0.0319 3.24 106
(6) Case 6: 873-bus test feeder, with 7 reference buses and 27 3 7 0.0279 5 0.1123 3.79 106
tie branches [25];
(7) Case 7: 10476-bus test feeder, with 84 reference buses
4 8 0.0481 5 0.2182 5.71106
and 260 tie branches [25]; 5 8 0.0774 5 0.3535 5.66 106
The programs were compiled using g++ -o0 g3 Wall c 6 6 0.2680 5 0.9618 6.62 106
MMD MP -MF on a Fedora 25 Linux operating system and 7 7 3.3766 5 13.7085 9.49 106
executed on a notebook PC (ThinkPad X1Carbon; 2.3-GHz
CPU). The functions and operations on dense and sparse
matrices were developed based on the Armadillo [29], Eigen
TABLE III
[30] and SuperLab [31] open source codes. PERFORMANCE OF THE TWO DIFFERENT SE METHODS
IMPLEMENTED ON A MESHED NETWORK
A. Performance Comparison with Newton Base SE Maximum
FD-DSE Newton DSE
From numerical test cases 1-7, with only deploying critical Case
voltage deviation
measurements, the results show that proposed method and CPU CPU between FD-DSE
Iter. Iter.
time(s) time(s) and Newton DSE
traditional SE can both obtain SE results. These tests verify that
the proposed SE model has the same observability as a 1 7 0.0151 5 0.0654 3.76 106
conventional SE.
2 5 0.0180 4 0.0617 5.89 106
The test results of the FD-DSE and Newton-DSE are listed in
Table II and Table III, with an average value of 100 Monte 3 6 0.0355 5 0.1358 7.99 106
Carlo simulations and where Iter. denotes the number of
iterations. Table II lists the results for radial networks, while 4 7 0.0648 5 0.2843 4.11106
Table III lists the results for meshed networks. The FD-DSE
5 8 0.0924 5 0.4470 5.22 106
and Newton DSE produce very similar SE results, since exact
measurement constraints are used to calculate the power 6 7 0.3422 5 1.4683 4.19 106
deviation over iterations. The FD-DSE gain matrix was
enlarged by introducing additional variables ( Pij , Qij , Pji , Q ji ). 7 8 4.2965 5 20.4278 1.19 106
However, this matrix is very sparse, thus maintaining the
efficiency of the FD-DSE. From Tables II and III, although the B. Impact from Complex per unit Normalization
Newton-DSE uses fewer iterations, it requires more CPU time. Complex per unit normalization is critical for implementing
This is because the gain matrix has to be reformulated and FD-DSE on a distribution network. From the results in Tables
factorized at every iteration in the Newton-DSE, but is only IV and V, complex per unit normalization improves the
needed during initialization for the FD-DSE. Since the branch convergence of FD-DSE. Without complex per unit
current measurements are formulated as linear loss normalization, FD-DSE cannot converge within a limited
measurement functions, the results also show that FD-DSE number of iterations. From the results for Case 5 given in
does not need more iterations when these measurements are Tables IV and V under conventional p.u., after closing the tie
involved. From the results of Case 7, the proposed FD-DSE is lines in the 201-bus test feeder, the FD-DSE failed to converge
more efficient than the Newton DSE for a large-scale system. in more than 20 iterations. The impedance parameters for tie
Since the efficiency of a program also depends heavily on its lines are all identical, at 2.98257 10-9 + 2.98257 10-9j p.u.
coding techniques, the CPU times for these two methods listed These are small impedance branches compared to other
in Table II are only for comparison under the same branches with about 0.0010.01 p.u. For FD-DSE under
implementation conditions. From the results in Tables II and III, conventional p.u., small impedance branches have a greater
it can also be seen that the meshed network has little impact on impact on convergence, while for FD-DSE under complex p.u.,
the convergence of the proposed method. From the results for small impedance branches have a smaller impact on
Cases 2 and 3 in Table III, the number of iterations for meshed convergence. It can be concluded that complex p.u. improves
networks is less than that for radial networks. the numerical stability of FD-DSE.
TABLE II
TABLE IV
PERFORMANCE OF THE TWO DIFFERENT SE METHODS
IMPLEMENTED ON A RADIAL NETWORK
FD-DSE with FD-DSE with
case complex p.u. conventional p.u.
base Iter. Iter.

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2709463, IEEE
Transactions on Smart Grid
8

1 1.524369 7 19 FD-DSE.
The branch current measurement coverage rate is defined
2 0.930988 6 11 as
nm branch
3 0.455284 7 7 (40)
nbranch
4 0.814445 8 7 where nm branch is the number of BCCMs and nbranch is the total
5 0.965050 8 8 number of branches.
To characterize the estimation accuracy, the root mean
6 0.915803 6 10 square (RMS) of the estimation error for load power is defined
as:
7 0.917346 7 10
P Qload Qload
2 2
Pload
se se

RMSload
load
(41)
TABLE V 2nload
PERFORMANCE OF THE TWO DIFFERENT SE METHODS
IMPLEMENTED ON A MESHED NETWORK where nLoad is the number of loads, Pload and Qload are the
FD-DSE with FD-DSE with se
original power flow results and Pload se
and Qload are the state
case complex p.u. conventional p.u.
base estimation results.
Iter. Iter.
As shown in Table VI, when the BCCM coverage rate
1 1.525099 7 18 increases, the estimation accuracy is significantly improved.
Hence, it is necessary to incorporate BCCMs into FD-DSE to
2 0.922711 5 9
obtain better estimation results.
3 0.458971 6 9
TABLE VII
4 0.811444 7 7 RESULTS OF FD-DSE WITH DIFFERENT NUMBERS OF BCCMS
RMSload
5 0.966187 8 > 20 0.1 2.19 102
6 0.860685 7 9 0.3 5.2 103
0.6 2.8 103
7 0.861821 8 13
1 5.1104

C. Comparison with the Newton SE method using constant E. Bad Data Analysis
Jacobian matrix
Additional numerical tests are conducted on a 30-bus
Fast decoupled method is more robust than Newton Jacobian distribution network [26] as illustrated in Figure. 3. The test
method with constant Jacobian under heavy load. Hence, to results are listed in Table VIII. The results show that FD-DSE
verify the performace of the proposed FD-DSE, the active and incorporated with HTI can successfully identify multiple
reactive power of load are increased to 1.3 times. branch current magnitude measurements errors and other types
Table VI shows that Newton DSE with constant Jacobian is of measurements.
less efficient than proposed FD-DSE for a network with heavy
load in all cases. 13 14 15 16 28 29 30
TABLE VI 13 14 15 28 29

PERFORMANCE OF THE FD-DSE METHODS AND NEWTON SE WITH CONSTANT


27
JACOBIAN FOR A DISTRIBUTION NETWORK WITH HEAVY LOAD 1 2 12 3 4 5 6 7 8 9 10 11 12

Newton DSE with 1 2 3 4 5 6 7 8 9 10 11


FD-DSE
Constant Jacobian 16
Case
CPU CPU 17 18 19 20 21
Iter. Iter.
time(s) time(s) 17 18 19 20 21

1 7 0.0079 12 0.0141 Power Flow Branch current magnitude 22


measurements measurements 22
2 7 0.0104 11 0.0173 23
Injection 23
3 7 0.0295 12 0.0978 measurements
Voltage magnitude
measurements
24
27 26
24
4 8 0.0493 14 0.1163 26
25 25

5 8 0.0797 14 0.1539
6 7 0.2798 16 0.6756 Figure 3 30-bus distribution network with various measurements
7 7 3.4986 13 6.7732
TABLE VIII
D. Impact for SE with different BCCM configurations BAD DATA ANALYSIS
Error measurements Identification results
In this section, test cases with a 69-bus distribution network Branch current magnitude measurements Errors are all
with different BCCM configurations are conducted to at branch 1-2 identified
demonstrate how BCCMs improve the estimation accuracy of Branch current magnitude measurements Errors are all

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2709463, IEEE
Transactions on Smart Grid
9

at branch 1-2 and branch 3-4 identified [12] R. R. Nucera and M. L. Gilles, A blocked sparse matrix formulation for
Branch current magnitude measurements Errors are all the solution of equality-constrained state estimation, IEEE Transactions
at branch 3-4 and branch 4-5 identified on Power Systems, vol. 6, no. 1, pp. 214224, Feb. 1991.
Power Flow measurements at branch 1-2 Errors are all [13] A. Monticelli, C. A. F. Murari, and F. F. Wu, A Hybrid State Estimator:
identified Solving Normal Equations By Orthogonal Transformations, IEEE
Power Flow measurements at branch 1-2 Transactions on Power Apparatus and Systems, vol. PAS-104, no. 12, pp.
Errors are all 34603468, Dec. 1985.
and Branch current magnitude
identified [14] L. Holten, A. Gjelsvik, S. Aam, F. F. Wu, and W. H. E. Liu, Comparison
measurements at branch 3-4
of different methods for state estimation, IEEE Transactions on Power
Systems, vol. 3, no. 4, pp. 17981806, Nov. 1988.
VII. CONCLUSION [15] K. A. Clements, G. W. Woodzell, and R. C. Burchett, A new method for
solving equality-constrained power system static-state estimation, IEEE
In this study, branch ampere measurements were formulated
Transactions on Power Systems, vol. 5, no. 4, pp. 12601266, Nov. 1990.
as linear loss measurement functions, and a fast decoupled SE [16] A. Gjelsvik, S. Aam, and L. Holten, Hachtels Augmented Matrix Method
solution was developed for distribution networks based on - A Rapid Method Improving Numerical Stability in Power System Static
complex per unit normalization. In this method, the Jacobian State Estimation, IEEE Transactions on Power Apparatus and Systems,
vol. PAS-104, no. 11, pp. 29872993, Nov. 1985.
matrices are constant and convergence is maintained for
[17] J. W. Wang and V. H. Quintana, A Decoupled Orthogonal Row
distribution networks of different scales. The high efficiency of Processing Algorithm for Power System State Estimation, IEEE
the proposed SE method is suitable for large scale distribution Transactions on Power Apparatus and Systems, vol. PAS-103, no. 8, pp.
networks with ampere measurements. 23372344, Aug. 1984.
[18] F. Zhuang and R. Balasubramanian, A Transformation-Decoupled
The proposed method can be applied in high and medium Estimator for Power System State Estimation, IEEE Transactions on
voltage distribution networks, in which three-phase imbalance Power Apparatus and Systems, vol. PAS-104, no. 7, pp. 17381744, Jul.
is not critical. To apply the method to low voltage distribution 1985.
networks, which are always currently unobservable, this [19] A. Monticelli and A. Garcia, Fast decoupled state estimators, IEEE
Transactions on Power Systems, vol. 5, no. 2, pp. 556564, May 1990.
FD-DSE model should be expanded to a three-phase model [20] L. Roy and T. A. Mohammed, Fast super decoupled state estimator for
with some approximations, which require further investigation. power systems, IEEE Transactions on Power Systems, vol. 12, no. 4, pp.
For unbalanced three-phase distribution systems, three-phase 15971603, Nov. 1997.
fast decoupled pattern can be tried in complex per unit system [21] O. L. Tortelli, E. M. Lourenco, A. V. Garcia, and B. C. Pal, Fast
Decoupled Power Flow to Emerging Distribution Systems via Complex pu
with assumption a b 1200 etc. The detail of three-phase Normalization, IEEE Transactions on Power Systems, vol. 30, no. 3, pp.
fast decoupled model can refer to [34]. 13511358, May 2015.
[22] B. Stott and D. Alsac, Fast decoupled load flow, IEEE Transcations on
PAS, pp. 859867, 1974.
REFERENCES [23] A. Abur and A. G. Exposito, Power System State Estimation:theory and
[1] F. C. Schweppe and J. Wildes, Power System Static-State Estimation, implementation. Boston: Marcel Dekker,Inc, 2004.
Part I: Exact Model, IEEE Transactions on Power Apparatus and Systems, [24] I. S. Duff and J. Koster, The design and use of algorithms for permuting
vol. PAS-89, no. 1, pp. 120125, Jan. 1970. large entries to the diagonal of sparse matrices, Siam Journal on Matrix
[2] A. Primadianto and C. N. Lu, A Review on Distribution System State Analysis & Applications, vol. 20, no. 4, pp. 889901, 1999.
Estimation, IEEE Transactions on Power Systems, vol. PP, no. 99, pp. [25] N. P. Group, REDS: REpository of Distribution Systems.
11, 2016. http://www.dejazzer.com/reds.html, 2016.
[3] M. E. Baran and A. W. Kelley, State estimation for real-time monitoring [26] U. Eminoglu and M. H. Hocaoglu, A new power flow method for radial
of distribution systems, IEEE Transactions on Power Systems, vol. 9, no. distribution systems including voltage dependent load models, Electric
3, pp. 16011609, Aug. 1994. Power Systems Research, vol. 76, no. 13, pp. 106114, 2005.
[4] K. I. Geisler, Ampere Magnitude Line Measurements for Power Systems [27] M. E. Baran and F. F. Wu, Network reconfiguration in distribution
State Estimation, IEEE Transactions on Power Apparatus and Systems, systems for loss reduction and load balancing, IEEE Transactions on
vol. PAS-103, no. 8, pp. 19621969, Aug. 1984. Power Delivery, vol. 4, no. 2, pp. 14011407, Apr. 1989.
[5] A. Abur and A. G. Exposito, Detecting multiple solutions in state [28] C.-T. Su, C.-F. Chang, and J.-P. Chiou, Distribution network
estimation in the presence of current magnitude measurements, IEEE reconfiguration for loss reduction by ant colony search algorithm, Electric
Transactions on Power Systems, vol. 12, no. 1, pp. 370375, Feb. 1997. Power Systems Research, vol. 75, no. 23, pp. 190199, 2005.
[6] C. N. Lu, J. H. Teng, and W. H. E. Liu, Distribution system state [29] C. Sanderson and R. Curtin, Armadillo: a template-based C++ library for
estimation, IEEE Transactions on Power Systems, vol. 10, no. 1, pp. linear algebra, Journal of Open Source Software, vol. 1, p. 26, 2016.
229240, Feb. 1995. [30] G. Guennebaud, B. Jacob, and others, Eigen v3. 2010.
[7] H. Wang and N. N. Schulz, A revised branch current-based distribution [31] X. S. Li, An Overview of SuperLU: Algorithms, Implementation, and
system state estimation algorithm and meter placement impact, IEEE User Interface, toms, vol. 31, no. 3, pp. 302325, Sep. 2005.
Transactions on Power Systems, vol. 19, no. 1, pp. 207213, Feb. 2004. [32] A. Monticellli and A. Garcia, "Reliable Bad Data Processing for Real
[8] W. Wu, Y. Ju, B. Zhang, and H. Sun, A distribution system state estimator Time State Estimation", IEEE Trans, on Power Apparatus and Systems,
accommodating large number of ampere measurements, International Vol.PAS-102, No.3, pp.1126-1139, 1983
Journal of Electrical Power & Energy Systems, vol. 43, no. 1, pp. 839848, [33] L. Mili, Th Van Cutsem and M. Ribbens-Pavella, "Hypothesis Testing
2012. Identification: A New Method for Bad Data Analysis in Power System
[9] A. Garcia, A. Monticelli, and P. Abreu, Fast Decoupled State Estimation State Estimation", IEEE Trans, on Power Apparatus and Systems,
and Bad Data Processing, IEEE Transactions on Power Apparatus and Vol.PAS-103, No.11, pp.3239-3252, 1984
Systems, vol. PAS-98, no. 5, pp. 16451652, 1979. [34] R. K. Portelinha and O. L. Tortelli ,"Three phase fast decoupled power
[10] J. J. Allemong, L. Radu, and A. M. Sasson, A Fast and Reliable State flow for emerging Distribution Systems", 2015 IEEE PES Innovative
Estimation Algorithm for AEPs New Control Center, IEEE Transactions Smart Grid Technologies Latin America (ISGT LATAM), 81-86, 2015.
on Power Apparatus and Systems, vol. PAS-101, no. 4, pp. 933944, Apr. [35] B. M. Zhang and K. L. Lo, A Recursive Measurement Error Estimation
1982. Identification Method for Bad Data Analysis in Power System State
[11] L. S. VanSlyck and J. J. Allemong, Operating experience with the AEP Estimation, IEEE Trans. on Power Systems,Vol. PWRS-6, No. 1, 1991,
state estimator, IEEE Transactions on Power Systems, vol. 3, no. 2, pp. pp. 191-197 0885-8950.
521528, May 1988.

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2709463, IEEE
Transactions on Smart Grid
10

VIII. BIOGRAPHIES limiter prototype for electrical marine propulsion which was
funded by Rolls-Royce Plc and the Department of Trade &
Yuntao Ju (M13) received his B.Sc in mechanical Industry (DTI) of the United Kingdom.
engineering in 2008 and his Ph.D. degree in electrical He is currently a full professor in electrical power
engineering in 2013, all from Tsinghua University. He engineering at the Department of Electric Power Systems,
was an awardee of excellent graduates from Tsinghua China Agricultural University (CAU), Beijing, P R China.
University in 2008. In November 2013, He has been a His research interests are in the areas of electrical power
visiting scholar to University of Toronto for a year. In system analysis & control, power grid modeling &
simulation, renewable energy generation & system
2015, Dr. Yuntao Ju joined china electric power integration and dispatching as well as wind/solar power
research institute as a research fellow. He is now an forecasting.
associate professor in College of Information and
Electrical Engineering, China Agricultural University
(CAU), Beijing, P R China. His research interests
include hybrid energy system modeling, high speed
dynamic simulation, large scale system parameter
identification, state estimation and uncertainty
optimization..

Wenchuan Wu (SM14) received the B.S., M.S., and Ph.D.


degrees from the Electrical Engineering Department,
Tsinghua University, Beijing, China. He is currently a
Professor in the Department of Electrical Engineering of
Tsinghua University. His research interests include Energy
Management System, active distribution system operation
and control, and EMTP-TSA hybrid real-time simulation.
Prof. Wu is an IET Fellow, associate editor of IET
Generation, Transmission & Distribution, Electric Power
Components and Systems.

Fuchao Ge received the Bachelor Degree of Engineering


in electrical engineering from Linyi University, Linyi, China,
in 2014. Now he is pursuring master degree of China
Agricultural University, his research interests include Power
system state estimation, parameter identification and
electricity market.

Kang Ma is working as a lecturer at University of Bath. His


research focuses on three-phase unbalanced low voltage
networks. He worked as an R&D engineer at China Electric
Power Research Institute (Beijing) from 2011 to 2014,
during which time he developed the first version of the
reliability assessment module for a distribution network
planning platform. This platform has been widely applied to
over 20 provincial grid companies in China. He received his
PhD degree in Electrical Engineering from the University of
Manchester (U.K.) and his B.Eng. degree from Tsinghua
University (China).

Yi Lin is an engineer with Electric Power Research Institute


of Fujian Electric Power Limited Company. His research
interests include distribution network planning and
operation.

Lin Ye (M98,SM04) received his B.Sc. from WuHan


University, P R China in 1992 and his Ph.D. degree in 2000
from the Institute of Electrical Engineering (IEE), Chinese
Academy of Sciences (CAS), all in electrical engineering. He
has been pursuing research at ForschungsZentrum
Karlsruhe(FZK) (*now merged with University of
Karlsruhe(TH) to form Karlsruhe Institute for Technology,
KIT) as a research fellow of Alexander von Humboldt
Stiftung/Foundation (AvH) of Germany from 2000 to 2002.
In 2004, Dr. Lin Ye joined the Interdisciplinary Research
Center (IRC), Department of Engineering/Cavendish
Laboratory, the University of Cambridge, United Kingdom,
as a research fellow. At Cambridge Laboratory, he had been
involved in developing a novel resistive type of fault current

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.