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/TPWRS.2016.2638923, IEEE
Transactions on Power Systems
A State-Independent Linear Power Flow Model
with Accurate Estimation of Voltage Magnitude
Jingwei Yang, Student Member, IEEE, Ning Zhang, Member, IEEE, Chongqing Kang, Fellow, IEEE
and Qing Xia, Senior Member, IEEE
Abstract—Linearized power flow models are of great interest in
power system studies such as contingency analyses and reliability
assessments, especially for large-scale systems. One of the most
popular models–the classical DC power flow model–is widely
used and praised for its state-independence, robustness and
computational efficiency. Despite its advantages, however, the DC
power flow model fails to consider reactive power or bus voltage
magnitude. This paper closes this gap by proposing a decoupled
linearized power flow (DLPF) model with respect to voltage
magnitude and phase angle. The model is state-independent
but is distinguished by its high accuracy in voltage magnitude.
Moreover, the paper presents an in-depth analysis of the DLPF
model with the purpose of accelerating its computation speed,
leading to the fast DLPF (FDLPF) model. The approximation
that is applied to obtain the FDLPF model from the DLPF
model is justified by a theoretical derivation and numerical tests.
The proposed methods are provably accurate and robust for
several cases, including radial distribution systems, meshed large-
scale transmission systems and ill-conditioned systems. Finally,
expressions for sensitivity with regard to MW flow and bus
voltage are provided as a potential application.
Index Terms—Linear power flow model, decoupled load flow,
optimal power flow, voltage magnitude, reactive power, compu-
tational efficiency
N OMENCLATURE
The main notation used in this paper is provided below;
other symbols are defined as required.
n Number of power system buses
m Number of power system lines
R Set of V θ buses (slack buses)
S Set of P V buses
L Set of P Q buses
Y = G + jB Admittance matrix of the power system
Y 0 = G0 + jB 0 Admittance matrix without shunt ele-
ments
Yij = Gij + jBij Element in the ith row and jth column
of Y
Yij0 = G0ij + jBij
0
Element in the ith row and jth column
of Y 0
V ,θ Magnitudes and phase angles of bus
voltages
This work was supported in part by National Key Research and De-
velopment Program of China (No. 2016YFB0900102), National Science
Fund Major International(Regional) Joint Research Project of China (No.
51620105007) and Tsinghua University Initiative Scientific Research Program
(No. 20151080418).
J. Yang, N. Zhang, C. Kang and Q. Xia are with the Department of
Electrical Engineering, Tsinghua University, Beijing 100084, China. (E-mail:
cqkang@tsinghua.edu.cn)
0885-8950 (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.
1
Vi , θi Voltage magnitude and phase angle of
bus i
P,Q Bus injected active and reactive powers
Pi , Qi Injected active and reactive power at
bus i
Pij Lossless MW flow from bus i to bus j
zij = rij + jxij Impedance of line (i, j)
yij = gij + jbij Admittance of line (i, j)
yii = gii + jbii Shunt admittance at bus i
I. I NTRODUCTION
P OWER flow analysis is the backbone of power system
studies and is widely applied in daily operations as well
as short-term and long-term planning. Direct calculations of
power flow are of great significance for contingency analysis
[1], reliability assessment [2] and probabilistic load flow anal-
ysis [3], if they can be performed accurately and efficiently.
Although conventional AC power flow calculations yield ac-
curate results, their nonlinearity, difficulty in convergence and
low computational efficiency limit their application in the
power system industry, especially for large-scale systems [4].
Therefore, a fast and linear power flow model would be of
great interest, provided it could offer reasonable accuracy and
robustness for all types of grids. Such a linear model could
also be beneficial for solving optimization problems, such as
locational marginal pricing (LMP) [5] and security-constrained
unit commitment (SCUC) [6], by allowing these problems to
be transformed into linear programming problems.
The DC power flow model is one of the most widely
used linear power flow models for power systems. Because
it is a linear, non-iterative model with reasonable accuracy
in terms of MW flow, it has considerable analytical and
computational appeal compared with the AC power flow model
[7]. The classical DC power flow model is derived based on
the assumptions of a lossless MW flow and a flat bus voltage
profile. More general versions of the DC power flow model
have been proposed, including hot- and cold-start models, for
broader applications. Hot-start models correct the bus power
injections according to base points obtained from initial AC
power flow solutions in various ways, such as net loss dispersal
and base-point matching [7]. However, in situations such
as DC-model-based SCUC [6], financial transmission rights
allocations [8] and power system planning, no reliable AC
base points are available. Therefore, it is necessary to use
cold-start models. As alternatives to the classical DC power
flow model, some cold-start models modify the bus injections
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/TPWRS.2016.2638923, IEEE
Transactions on Power Systems
2

by estimating loss as a percentage of net load, whereas others The approximation is justified by a theoretical derivation
apply an empirical fixed voltage magnitude. Several tests have and numerical tests.
shown, however, that there is no large difference in MW-flow (3) To provide linear expressions for sensitivity computations
error between the classical DC power flow model and other regarding MW flow and bus voltage magnitude. The
cold-start models [7]. sensitivity is much greater than a small-signal sensitivity.
Although cold-start DC power flow models are widely used Numerical results prove that the model can also be applied
in many fields, it is worth noting that they can neither consider in cases of large signal perturbations.
reactive power nor shed light on bus voltage magnitude. In The remainder of the paper is organized as follows. Sec-
fact, the lack of a voltage magnitude consideration leads to tion II formulates the DLPF model based on the initial AC
many problems in system planning and operation when cold- power flow equation and derives the expressions for voltage
start DC power flow models are applied. For instance, it magnitude, phase angle and branch MW flow. In section III,
is a common practice to simply consider the MW flows in the FDLPF model is formulated by making a subtle change to
reliability evaluations. The bus voltage quality is inevitably the DLPF model, which is justified by a theoretical derivation
ignored, and moreover, ZIP loads are modeled at nominal and and numerical examples. The specific steps for utilizing the
typical voltages [2]. In terms of electricity markets, only active FDLPF model are presented in order in section IV. In section
generation is priced in markets such as PJM. The reactive V, several cases, including radial distribution systems, meshed
power offered by generators is paid for merely as an ancillary large-scale transmission systems and ill-conditioned systems,
service [9]. Probabilistic load flow calculations, which are an are analyzed to demonstrate the accuracy and robustness of
increasingly important tool in power system analysis, typically the DLPF/FDLPF approach and the computational efficiency
face the dilemma that DC power flow models are incapable of the FDLPF model. Large-signal sensitivity expressions are
of accounting for voltage magnitude, whereas AC power flow provided in section VI as an application of the FDLPF model.
models that are linearized around the operating point lead to Finally, conclusions are drawn and further discussions are
errors in scenarios with large signal perturbations [10]. presented.
However, it is not easy to consider bus voltage magnitudes
in a decoupled linear flow model, because they are sensi-
tive to both active and reactive power injections. In [11], a II. F ORMULATION AND D ERIVATION OF THE DLPF
linear model was proposed for radial distribution systems.
However, this model is dependent on the radial topology of A. Decoupling of Voltage Magnitude and Phase Angle
the distribution system and cannot be extended to meshed For a power system with n buses, the polar power-voltage
transmission systems. The authors of [12] have achieved a AC power flow model is well known:
great advancement in formulating a linear load equation that
n n
considers reactive power. This power flow model reflects both X X
active and reactive power balance with respect to the square of Pi = Gij Vi Vj cos θij + Bij Vi Vj sin θij (1)
j=1 j=1
the voltage magnitude, V 2 , and the modified phase angle, V 2 θ.
n n
Nevertheless, the voltage magnitude and phase angle are not X X
Qi = − Bij Vi Vj cos θij + Gij Vi Vj sin θij (2)
completely decoupled, giving rise to a quadratic programming
j=1 j=1
problem if this model is applied for optimization.
i = 1, 2, · · · , n
This paper makes a further step toward a V − θ decoupled
linear power flow model, which is distinguished by its accu- For most scenarios in real power systems, the bus voltage
racy in voltage magnitude and robustness for various types magnitudes are approximately 1 p.u. The absolute value of
of systems. Moreover, we direct considerable attention to im- phase angle differences across lines rarely exceed 30◦ [12] and
proving its computational efficiency for potential application to most of them are within 10◦ [13]. Moreover, the admittance
large-scale systems or simulations that require tens of millions matrix of the power system has a special structure, in which the
of repeated load flow calculations. The specific contributions diagonal elements are the sum of the non-diagonal elements
of this paper are threefold: in the corresponding rows, with the contribution of shunt
(1) To propose a decoupled linearized power flow (DLPF) elements as well:
model with respect to the voltage magnitude V and the 
− yij if j 6= i
phase angle θ. The model is distinguished by its accuracy


n
in voltage magnitude and robustness for application to Yij = X (3)
 yii +
 yik if j = i
different types of systems. The error in MW flow is k=1,k6=i
comparable to that of cold-start DC power flow models,
which is much smaller in practice than is predicted, as Note that in this paper, the “shunt elements” yii denote
mentioned in [7]. not only the contribution of shunt capacitors but also the
(2) To design a fast DLPF (FDLPF) model by developing line-charging susceptance and the equivalent admittance of
a fast approximation of the matrices used in the DLPF transformers and phase shifters (Part C of this section provides
model, which is as fast as classical DC power flow the details of the consideration of transformers and phase
calculations while still as accurate as DLPF calculations. shifters).

0885-8950 (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/TPWRS.2016.2638923, IEEE
Transactions on Power Systems
3

Considering these facts, the following linear approximations Approx 1

2 Approx 2
are applied to (1) to decouple the voltage magnitudes and the Approx 3
No approximation
phase angles:

Branch MW flow (p.u.)

n
X n
X 1
Pi = Gij Vi Vj cos θij + Bij Vi Vj sin θij
j=1 j=1
n 0
X
= gii Vi2 + (gij Vi (Vi − Vj cos θij ) − bij Vi Vj sin θij )
j=1,j6=i
-1
Xn n
X
≈ gii Vi + gij (Vi − Vj ) − bij (θi − θj )
j=1,j6=i j=1,j6=i -2
-10 -5 0 5 10
 
n n
X X Phase angle difference (degree)
= Vi gij + (−gij ) Vj 
(a) Case I: r/x = 1/4 with r = 0.02 p.u. and x = 0.08 p.u.
j=1 j=1,j6=i
 
n
X n
X Approx 1
Approx 2
− θi bij + (−bij ) θj  (4) 2 Approx 3
No approximation
j=1,j6=i j=1,j6=i

Branch MW flow (p.u.)

n
X n
X
0
= Gij Vj − Bij θj
1
j=1 j=1

Similarly, equation (2) for reactive power can be approximated

as follows: 0
Xn n
X
Qi = − Bij Vj − Gij θj (5)
j=1 j=1
-1
Note that G0ij
≈ Gij is assumed in (5) because the shunt con- -10 -5 0 5 10
Phase angle difference (degree)
ductance is negligible compared with the shunt susceptance.
(b) Case II: r/x = 4/1 with r = 0.08 p.u. and x = 0.02 p.u.
The approximations made in (4) and (5) are not the same as
those made in other linear power flow models. Consider the Fig. 1. An illustration of the accuracy of different approximations with Vi =
1.05 p.u. and Vj = 0.98 p.u.
following three alternative approximations in the expression
of branch MW flow, corresponding to the DC power flow, the
approximation made by [12], and the proposed approximation, Note that Vi is decomposed into (1 + ∆Vi ), where ∆Vi is
respectively: typically one order of magnitude smaller than Vi . In the second
• Approx.1: gij Vi (Vi − Vj cos θij ) ≈ 0 step of (6), ∆Vi2 and ∆Vi ∆Vj can be neglected without
2 2


• Approx.2: gij Vi (Vi − Vj cos θij ) ≈ gij Vi − Vj excessive error because they are two orders of magnitude
• Approx.3: gij Vi (Vi − Vj cos θij ) ≈ gij (Vi − Vj ) smaller than Vi and Vj .
The common assumption is that cos θij ≈ 1. However, the A linear expression for the average branch MW flow Pij
methods dealing with voltage magnitude are not identical. In can also be easily obtained from the first step of (4):
the DC model, the voltage differences across branches are
neglected. However, these are considered in Approx.3, which Pij = gij (Vi − Vj ) − bij (θi − θj ) (7)
is the key approximation step in the derivation of the proposed
Because the line loss is ignored for linearity, we have Pij =
DLPF model. Although [12] takes voltage differences into
−Pji , where Pij denotes the power flow from bus i to bus
account by using Approx.2, the result is not satisfactory. Fig.1
j. It should be noted that equation (7) has to be modified for
illustrates the effects of the three approximations on branch
lines with transformers and phase shifters. The details are also
MW flow in two typical scenarios (a high r/x ratio and a low
provided in Part C.
r/x ratio). It is evident that the approximation proposed in
this paper (Approx. 3) performs the best in both cases.
In fact, a theoretical mechanism exists underlying the pro- B. Matrix Formulation of the DLPF Model
posed approximation:
Equations (4) and (5) represent the basic formulation of the
gij Vi (Vi − Vj cos θij ) ≈ gij Vi (Vi − Vj ) DLPF model. The matrix form is as follows:
 0
= gij (1 + ∆Vi ) (∆Vi − ∆Vj )
   
P B −G θ
=− (8)
≈ gij (∆Vi − ∆Vj ) (6) Q G B V
= gij (1 + ∆Vi − 1 − ∆Vj ) It is evident that both θ and V consist of three sub-vectors,
= gij (Vi − Vj ) corresponding to the V θ, P V , and P Q buses. Without loss of

0885-8950 (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/TPWRS.2016.2638923, IEEE
Transactions on Power Systems
4

generality, we arrange these buses in the following sequence: C. Transformers and Phase Shifters
V θ, P V , P Q. This section describes the changes to the admittance ma-
 T T T T trix and MW flow approximations needed for branches with
θ = θR , θS , θL
T (9) transformers or phase shifters.
V = VR , VST , VLT
 T

The admittance matrix Y can also be partitioned in the same

manner:
YRR YRS YRL
 

Y =  YSR YSS YSL  (10)

YLR YLS YLL Fig. 2. A general digram for a branch with phase shifting transformers.
With the aid of the known sub-vectors θR , VR , and VS , we
transform (8) into the following equation: 1) Contributions to the Admittance Matrix: According to
     the formation of nodal admittance matrix Y [14], the contri-
P̃ H N θ̃ bution of the branch with a phase shifiting transformer (Fig.2,
= (11)
Q̃ M L Ṽ t for tap ratio and θs for phase shifting angle) to the nodal
where admittance matrix is
   0 T
PS BSR −GSR −GSS θR
 
  ∆Y = Cyij C (22)
P̃ 0
=  PL  + BLR −GLR −GLS  VR  (12) 1
Q̃ where C is a n × 1 vector of all zeros except Ci = te−jθs
QL GLR BLR BLS VS
 0 and Cj = −1. C is the conjugate of C.
0

  BSS BSL −GSL ∆Y can be split into two parts:
H N 0 0
= − BLS BLL −GLL  (13)
M L GLS GLL BLL ∆Y = ∆Y1 + ∆Y2 (23)

T T
θ̃ = θST , θL
  where ∆Y1 has four non-zero elements:
(14) yij
Ṽ = VL ∆Y1,ii = −∆Y1,ij =
te−jθs (24)
Performing elementary row operations on both sides of (11) yij
∆Y1,jj = −∆Y1,ji = jθs
leads to the following equations: te
P̃ − N L−1 Q̃ while ∆Y2 is a diagonal matrix with only two non-zero
H − N L−1 M 0 θ̃
    
= (15) elements:
Q̃ M L Ṽ yij yij
∆Y2,ii = 2
− −jθs
and t te (25)
yij
  
H N
  ∆Y2,jj = yij − jθs
P̃ θ̃ te
= (16)
Q̃ − M H −1 P̃ 0 L − M H −1 N Ṽ In fact, ∆Y1 represents the contribution of equivalent line
Combining the first part of (15) and the second part of admittance to Y . ∆Y2 can be regarded as the equivalent shunt
(16) into one equation leads to the decoupling of voltage elements of phase shifting transformers and is considered in
magnitudes and phase angles: yii as stated in Part A.
2) Influence on the Branch MW Flow: For the branch
P̃ − N L−1 Q̃
    
H̃ 0 θ̃ shown in Fig.2, the branch complex power has the following
= (17)
Q̃ − M H −1 P̃ 0 L̃ Ṽ relationship with bus voltage and phase angle:
where #  yij yij  "
− −jθs
" # " #
S ij Vi e−jθi  t 2 te V i e jθi
H̃ = H − N L−1 M (18) = yij

S ji Vj e−jθj − jθs yij Vj ejθj
L̃ = L − M H −1 N (19) te
(26)
According to (17), the expressions for calculating the un- The branch MW flow Pij is the real part of S ij :
known Ṽ and θ̃ with respect to the known bus injections P̃ y

ij yij 
and Q̃ are as follows: Pij = Re Vi e−jθi V i e jθi
− V j e jθj

y t2 te−jθs 
θ̃ = H̃ −1 P̃ − H̃ −1 N L−1 Q̃ (20) ij 2 yij −j(θi −θj −θs )
= Re 2 Vi − Vi Vj e
t t
Ṽ = L̃−1 Q̃ − L̃−1 M H −1 P̃ (21) 
gij Vi
= Vi − Vi Vj cos(θi − θj − θs ) (27)
Note that the decoupling of the voltage magnitudes and phase t t
angles is not achieved by simply ignoring the relatively small bij
conductance matrix G. On the contrary, the coupling between − Vi Vj sin(θi − θj − θs )
t
the active and reactive powers is considered in equations (20)
 
gij Vi bij
and (21). ≈ − Vj − (θi − θj − θs )
t t t

0885-8950 (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/TPWRS.2016.2638923, IEEE
Transactions on Power Systems
5

The approximation made in the last step of (27) is similar

to that in (4) and this provides a linear expression for branch
MW flow with transfromers or phase shifters. Because the line
loss is ignored for linearity, we have Pij = −Pji . It should
be noted that (27) and (7) have the same pattern. Equation (7)
can be seen as a special case of (27) where t = 1 and θs = 0.

III. D ERIVATION AND J USTIFICATION OF THE FDLPF

A fast computation speed is one of the key advantages
of DC power flow models. The classical DC power flow
model achieves this speed because the admittance matrix is
Fig. 3. A 2-bus system.
formulated using 1/xij directly from the system information.
For the DLPF model proposed in section II, its solution
requires the factorization of four matrices: H, L, H̃, and It can be seen from (33) and (34) that the updated elements
L̃. Although H and L are sparse matrices, H̃ and L̃ may be are exactly identical to the diagonal elements of H̃ 0 and L̃0 .
non-sparse matrices, for which factorization is computationally This surprising feature can be extended to more general cases,
intractable. This section, however, shows that H̃ and L̃ can be as demonstrated next.
approximated by the sparse matrices H̃ 0 and L̃0 , which have
the same structure as H and L:

1 B. Theoretical Derivation

 − if j 6= i

 xij For a general power system, H̃ and L̃ are too complicated
0
H̃ij = Xn
1 (28) to explicitly derive. However, for a radial system or a constant-
if j = i

r/x-ratio system without P V buses, it can be proven that L̃0

xik


k=1,k6=i
is identical to L̃, whereas H̃ 0 is a good approximation of H̃.
1


 − if j 6= i Let ySL = gSL + jbSL denote the primitive admittance

 xij matrix of a system in which the V θ buses are grounded and
0
L̃ij = n (29) all shunt susceptances are neglected. C represents the node-
X 1
if j = i

to-branch incidence matrix. The shunt susceptances are repre-


 x ik
k=1 sented by the (n − 1) × (n − 1) diagonal matrix diag (b0 ).
Expressions (20) and (21) for bus voltage magnitude and phase For a meshed system with a constant r/x ratio, let us
angle are thus transformed into assume that for any branch l, rl /xl = α. This leads to
θ̃ = H̃ 0−1 P̃ − H̃ 0−1 N L−1 Q̃ (30) rl −xl
gl = − = −αbl
Ṽ = L̃0−1 Q̃ − L̃0−1 M H −1 P̃ (31) xl rl2 + x2l
(35)
r2
 
The approximation always holds for a radial system or −xl 1
1 + α2 bl = 1 + l2


=−
a meshed system with a constant r/x ratio, as justified by xl rl2 + x2l xl
the theoretical derivation presented in this section. For more and thus
complex cases, the approximation error is mild, as can be
demonstrated by a numerical example. H̃ = H − N L−1 M
= −CbSL C T − CgSL C T ·


A. An Illustrative Example
−1
CbSL C T + diag (b0 ) CgSL C T


Let us begin with a 2-bus system (Fig.3) for illustration. In
≈ −CbSL C T − α2 CbSL C T ·


this case, equation (13) becomes (36)

−1
−b12 CbSL C T CbSL C T
   

H N g12
= (32)
−g12 −b12 − b22 = − 1 + α2 CbSL C T


M L
and H̃ and L̃ can be calculated as follows: = H̃ 0
H̃ = H − N L−1 M L̃ = L − M H −1 N
−1
= −b12 − g12 (b12 + b22 ) g12 = −CbSL C T − diag (b0 ) −

−1
≈ −b12 − g12 b−1 CgSL C T CbSL C T CgSL C T



12 g12 (33) (37)
1 = − 1 + α2 CbSL C T − diag (b0 )


=
x12 = L̃0
L̃ = L − M H −1 N
= −b12 − b22 − g12 b−1 For a radial system (usually a distribution system), C is a
12 g12 (34)
1 1 non-singular matrix. It can also be easily found that L̃0 = L̃
= + and H̃ 0 ≈ H̃, in the same manner as for meshed systems.
x12 x22

0885-8950 (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/TPWRS.2016.2638923, IEEE
Transactions on Power Systems
6

Computational Complexity: All matrices are created directly

from the system topology.
• Step 2 Modify bus power injections

Build the modified bus injection vectors P̃ and Q̃ according

to (12). Calculate L−1 Q̃ and H −1 P̃ first, followed by P̃ −
N L−1 Q̃ and Q̃ − M H −1 P̃ .
Computational Complexity: This step requires the factoriza-
tion of two sparse matrices, H and L.
Fig. 4. Grainger & Stevenson 4-bus system. • Step 3 Calculate bus voltage magnitudes, phase angles
and MW flows
C. A Numerical Example The phase angles and voltage magnitudes can be rapidly
calculated using expressions (30) and (31). The branch MW
The approximations of H̃ and L̃ also hold for more flows can be calculated using expression (7).
general meshed systems. Let us demonstrate this claim using Computational Complexity: This step requires the factoriza-
a numerical example of a Grainger & Stevenson 4-bus system tion of two sparse matrices, H̃ 0 and L̃0 .
[14], which is shown in Fig.4. The total computational complexity is very low for the
Using equation (13), the matrices H, N , M , and L are as following three reasons:
follows:
  (1) All of the matrices are created directly from the system
40.97 −25.85 −15.12 −5.17 −3.02 topology. Because all of the matrices have the same
   −25.85 44.93 0 8.99 0  structure as that of the admittance matrix, they can be
H N  
=  −15.12
 0 40.97 0 8.19  updated rapidly after system reconfiguration.
M L  5.17 −8.99 0 44.84 0 

(2) All of the matrices are sparse and easily factorized.
3.02 0 −8.19 0 40.86 (3) All of the multiplication is performed between a matrix
and a vector.
The matrices H̃ and L̃ are derived from equation (18) and
(19): In the next section, it is shown that the FDLPF model is as
fast as the classical DC power flow model.
41.79 −26.88 −15.72
 
 
46.63 0
H̃ = −26.88 46.73 0  , L̃ = V. N UMERICAL S TUDIES
0 42.50
−15.72 0 42.61 In this section, several cases are analyzed to demonstrate
0 0 the accuracy and robustness of the DLPF/FDLPF approach
The approximate matrices H̃ and L̃ are defined by (28) and
(29): and the fast computation speed of the FDLPF model [15].
Various types of power systems are studied:
42.60 −26.88 −15.72
 
 
46.63 0 (1) Balanced radial distribution systems, including the IEEE
H̃ 0 = −26.88 46.72 0  , L̃0 =
0 42.50 33-node test feeder [16] and the modified 123-node test
−15.72 0 42.61 feeder [11]. These systems are characterized by a radial
It can be easily seen that L̃0 and L̃ are identical, whereas topology and a high r/x ratio [17]. Because voltage drop
H̃ 0 is extremely close to H̃. is an important phenomenon in distribution systems [18]
[19], these cases are studied to demonstrate the accuracy of
the DLPF/FDLPF approach in terms of voltage magnitude.
IV. S TEPS AND C OMPUTATIONAL C OMPLEXITY OF THE
(2) Meshed transmission systems, including IEEE standard
FDLPF M ODEL
systems, large-scale Polish systems and Pegase systems.
Many equations and demonstrations are presented in the Most of the available cases are tested to demonstrate the
sections above. To avoid confusion, this section summarizes robustness of the DLPF/FDLPF approach. The computa-
and highlights the essential equations of the FDLPF model tion times for large-scale systems are also compared to
to be used for direct power flow calculations or related illustrate the fast speed of the FDLPF model and the
applications. For optimization problems, the proposed DLPF usefulness of the proposed approximation.
model (8) can be used directly, but this topic is outside the (3) Ill-conditioned systems, which can be readily simulated
scope of this paper. by increasing the demand within the IEEE 33-node test
The following steps should be applied for the computation feeder. This study is presented to demonstrate the ability of
of bus voltage magnitudes and phase angles as well as the the DLPF/FDLPF approach to cope with an ill-conditioned
branch MW flows. The computational complexity is explained system on the brink of collapse.
for each step. The analyses were performed in MATLAB with the aid
• Step 1 Create matrices of MATPOWER 5.1 [20]. The simulation platform was a
Create the admittance matrices Y and Y 0 and generate H, ThinkPad workstation with an Intel i7 CPU@2.60 GHz and
N , M , and L at the P Q and P V buses according to (13). 8 GB of RAM. Three important results were obtained for
Create the matrices H̃ 0 and L̃0 using definitions (28) and (29). comparison: bus voltage magnitudes, branch MW flows and

0885-8950 (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/TPWRS.2016.2638923, IEEE
Transactions on Power Systems
7

TABLE I
E RRORS OF D IFFERENT L INEAR P OWER F LOW M ODELS FOR D ISTRIBUTION S YSTEMS

† md−[12] md−[11] md−[12] md−[11] ‡

Test Case εDCP
V
F
εV εV εDLP
V
F
εF
V
DLP F
εDCP
P
F
εP εP εDLP
P
F
εF
P
DLP F

IEEE 33-node 0.052 0.027 0.004 0.004 0.004 0.026 0.847 0.979 0.026 0.026
IEEE 123-node 0.051 0.027 0.004 0.003 0.003 0.008 0.291 0.173 0.008 0.008
† The bus voltage profile is assumed to be flat at 1.0 p.u. in the DCPF model.
‡ The linear expression for branch MW flow is not provided in [11]. Because the phase angle is accurate, according to the results reported in
[11], we assume the same expression as that for the DCPF model, namely, ∀ branches (i, j) , P(i,j) = (θi − θj ) /xij .

1.00
ACPF
DCPF
Voltage Magnitude

0.98 md-[12]
md-[11]
DLPF
0.96 FDLPF

0.94

0.92

0.90
0 5 10 15 20 25 30 35
Bus Number
4
MW Flow

-2
0 5 10 15 20 25 30 35 40
Branch Number

Fig. 5. Voltage magnitudes and branch MW flows for the IEEE 33-node test feeder.

computation times. With the results of the AC power flow magnitudes but in active power flows as well. Note that md-
model as the benchmark, the error for each model (md) was [11] exhibits the same accuracy in bus voltage magnitudes as
calculated as follows: the DLPF/FDLPF model but a poor performance in branch
n
1 X  md MW flows. This is not due to the flaws in the model itself but
εmd Vi − ViAC 

V = (p.u.) (38) our choice to use DCPF branch flow expression, as no linear
n i=1
expression for branch MW flow is provided in [11]. In fact,
m
1 X  md the proposed model is a generalization of md-[11] from radial
εmd Pi − PiAC 

P =
m i=1
To avoid random timing errors, each MATLAB program was
run ten times and the average computation time tmd was
considered.
A. Balanced Radial Distribution Systems
Table I compares the performances of different models on
the IEEE 33- and 123-node test feeders. The solutions were
obtained using the DC power flow (DCPF) model, the method
proposed in [12] (md-[12]), the method proposed in [11] (md-
[11]) and the DLPF/FDLPF approach proposed in this paper.
Fig.5 visualizes the solutions for the 33-node system in terms
of voltage magnitudes and branch MW flows. It should be
noted that an average error of 0.05 p.u. in voltage magnitude
is beyond tolerance, given that the operational voltage limit is
typically between 0.95 p.u. and 1.05 p.u.
From Table I and Fig.5, it is obvious that the DLPF
and the FDLPF model perform the best not only in voltage
0885-8950 (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.
(p.u.) (39) distribution systems to meshed networks with P V buses. A
detailed discussion is provided in the Appendix.
B. Meshed Transmission Systems
Table II compares the performances of the different models
for various kinds of transmission systems. The model pro-
posed in [11] is excluded from this comparison because it
was designed only for radial networks. Fig.6 visualizes the
solutions for the IEEE 118-bus system in terms of voltage
magnitude and branch MW flow as an example for illustration.
A comparison of the computation times for the large-scale
system is also presented in Table III.
From Table II and Fig.6, it is evident that the results
of the DLPF and FDLPF models are the most accurate in
terms of voltage magnitude among various cases of large-scale
transmission systems. The accuracy of the DLPF model in
terms of MW flow is also higher than that of the classical DC
power flow model. Although the FDLPF model performs a
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/TPWRS.2016.2638923, IEEE
Transactions on Power Systems
8

TABLE II
E RRORS OF D IFFERENT L INEAR P OWER F LOW M ODELS FOR T RANSMISSION S YSTEMS

F† md−[12] md−[12]
Test Case εDCP
V εV εDLP
V
F
εF
V
DLP F
εDCP
P
F
εP εDLP
P
F
εF
P
DLP F

IEEE 30-bus 0.018 0.011 0.001 0.001 0.004 0.023 0.002 0.007
IEEE 57-bus 0.024 0.012 0.007 0.009 0.015 0.062 0.010 0.017
IEEE 118-bus 0.023 0.003 0.001 0.001 0.035 0.090 0.034 0.042
IEEE 300-bus 0.025 0.016 0.014 0.014 0.105 0.148 0.102 0.105
Polish 2383-bus 0.008 0.005 0.003 0.003 0.026 0.047 0.022 0.038
Polish 3012-bus 0.089 0.005 0.002 0.003 0.022 0.055 0.021 0.034
Pegase 2869-bus 0.034 0.008 0.008 0.008 0.119 0.157 0.115 0.119
Pegase 9241-bus 0.032 0.014 0.007 0.007 0.181 0.208 0.178 0.192
† The bus voltage profile is assumed to be flat at 1.0 p.u. in the DCPF model.

1.06 ACPF
DCPF
1.04
Voltage Magnitude

md-[12]
DLPF
1.02 FDLPF

1.00

0.98

0.96

0.94

0 10 20 30 40 50 60 70 80 90 100 110 120

400 Bus Number

200
MW Flow

-200

-400

-600
0 20 40 60 80 100 120 140 160 180
Branch Number

Fig. 6. Voltage magnitudes and branch MW flows for the IEEE 118-bus system.

little worse than the the DC power flow model with respect A comparison of the computation times for the large-
of MW flow in some cases, the errors are acceptable because scale system is presented in Table III. Comparing it with
the main advantage of FDLPF is its accurate evaluation of Table II suggests that the FDLPF model achieves significantly
bus voltage magnitudes. Moreover, Table III suggests that the improved computational efficiency by sacrificing only limited
FDLPF model achieves significantly improved computational precision in terms of MW flow while maintaining the voltage
efficiency by sacrificing only limited precision in terms of magnitude accuracy of the DLPF model. The FDLPF model is
MW flow while maintaining the voltage magnitude accuracy therefore suitable for applications that require a large number
of the DLPF model. The FDLPF model is therefore suitable of repeated load flow calculations.
for applications that require a large number of repeated load
flow calculations.
C. Ill-conditioned Systems
TABLE III For the first considered case of an ill-conditioned system, the
C OMPUTATIONAL E FFICIENCIES OF D IFFERENT L INEAR P OWER F LOW
M ODELS FOR L ARGE -S CALE S YSTEMS
uniform overload of the system was simulated. The active and
reactive demands on all buses were incrementally increased
Test Case tACP F † tDCP F tmd−[12] tDLP F tF DLP F until the Newton-Raphson method was unable to converge.
Polish 2383 0.24 0.01 1.75 1.60 0.01 Similarly, for the second considered case, we increased the
Polish 3012 0.27 0.02 5.60 9.42 0.02 demand on a single bus (the 30th bus) to simulate a lumped
Pegase 2869 0.30 0.02 1.90 1.71 0.02 overload scenario. Fig.7 and Fig.8 report the errors of the
Pegase 9241 4.03 0.08 43.0 110 0.09 DLPF and FDLPF model in terms of voltage magnitude at
† The unit of computational time is second.
different load levels. A comparison among the results of the
different methods in the two cases shows that the DLPF and

0885-8950 (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/TPWRS.2016.2638923, IEEE
Transactions on Power Systems
9

In the same manner, the distribution shift factor of a branch

0.20
DCPF
md-[12]
MW flow with respect to nodal power injection is as follows:
md-[11]
DLPF
∆P(m,n) = gmn (∆Vm − ∆Vn ) − bmn (∆θm − ∆θn )
Error of Voltage Magnitude

FDLPF
0.15
    
= gmn L̃0−1 0−1
mi − L̃ni + bmn Ñmi − Ñni ∆Qi (42)
0.10     
0−1 0−1
− bmn H̃mi − H̃ni + gmn M̃mi − M̃ni ∆Pi
0.05
0−1
Note that the matrices H̃ji , L̃0−1
ji , Ñji , and M̃ji are constant

0.00
for a system with a fixed topology. They can also be rapidly
1 2 3 4 5 updated after system reconfiguration because H̃ 0 , L̃0 , N , and
Load Scale (Uniform Overload)
M have the same structure as that of the admittance matrix.
Table IV summarizes the sensitivity factors and distribution
Fig. 7. 33-node ill-conditioned system (uniform overload). shift factors for the 4-bus system shown in Fig.4. The changes
in bus voltage, phase angle, and branch MW flow were
0.12
calculated and compared with the results of the AC power
DCPF
md-[12] flow model with P2 and Q2 reduced by 50% (∆P2 = 0.85
md-[11]
0.10 DLPF
FDLPF
p.u. and ∆Q2 = 0.52 p.u.). The results show that the
Error of Voltage Magnitude

0.08
proposed sensitivity factors perform with the highest accuracy
in addressing the bus voltage. The phase angle and MW flow
0.06
variations can also be calculated with reasonable accuracy.
0.04

TABLE IV
0.02 T EST OF THE S ENSITIVITY FACTORS AND D ISTRIBUTION S HIFT FACTORS
D ERIVED FROM THE FDLPF M ODEL
0.00
0 2 4 6 8 10 12
Load Scale (Lumped Overload) FDLPF DCPF Jacobian-based With AC
Perturbation†
SF SF Voltage SF‡ Power Flow
∆θ2 0.0292 0.0314 – 0.0303
Fig. 8. 33-node ill-conditioned system (lumped overload).
∆θ3 0.0085 0.0085 – 0.0090
∆V2 0.0148 – 0.0113 0.0155
∆V3 0 – 0 -0.0002
FDLPF models exhibit a performance not worse than the other
∆P(1,2) -0.6125 -0.6226 – -0.6297
linear power flow models for ill-conditioned systems.
∆P(1,3) -0.2186 -0.2274 – -0.2235
∆P(2,4) 0.2375 0.2274 – 0.2248
∆P(3,4) -0.2186 -0.2274 – -0.2253
VI. A PPLICATION : S ENSITIVITY FACTORS AND
† ∆θ, ∆V , and ∆P are given in units of p.u.
D ISTRIBUTION S HIFT FACTORS ‡ SF is short for sensitivity factor. The Jacobian-based bus voltage
sensitivity is given in [21].
Using the fast and accurate FDLPF model proposed in this
paper, the sensitivity factors for the bus voltage magnitudes
An example of the calculation of the perturbation of the
and phase angles as well as the distribution shift factors for
voltage magnitudes on all P Q buses in the 4-bus system is
the branch MW flows can be easily determined with respect to
provided below:
nodal power injection. Notably, these factors can be applied
0 −0.0043
   
in large-signal cases because the FDLPF model is a state- 0.0214 0 0
L̃0−1 = , M̃ =
independent cold-start model. 0 0.0235 0 0 −0.0047
For simplicity and without loss of generality, we assume ∆V2 = 0.0214 × 0.52 + 0.0043 × 0.85 = 0.0148
that the active and reactive power injections vary at a single
∆V3 = 0 × 0.52 − 0 × 0.85 = 0
bus i and that the perturbations are ∆Pi and ∆Qi . According
to expressions (30), (31) and (7), the corresponding changes in
VII. C ONCLUSION
the voltage magnitude and phase angle at bus j are as follows:
In this paper, we propose a state-independent, V − θ
∆θj = 0−1
H̃ji ∆Pi − Ñji ∆Qi (40) decoupled linearized power flow (DLPF) model which is
0−1
distinguished by high accuracy in voltage magnitude. An in-
∆Vj = L̃ji ∆Qi − M̃ji ∆Pi (41) depth analysis of the matrices used for calculation is presented,
giving rise to a fast version of the DLPF model, namely, the
0−1
where H̃ji , L̃0−1
ji , Ñji , and M̃ji denote the elements in the FDLPF model. The approximation that is applied to obtain
jth row and ith column of the matrices H̃ 0−1 , L̃0−1 , Ñ , the FDLPF model from the DLPF model is justified by a
and M̃ , respectively, where Ñ = H̃ 0−1 N L−1 and M̃ = theoretical derivation and numerical examples. An analysis
L̃0−1 M H −1 . of various cases, including radial distribution systems with

0885-8950 (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/TPWRS.2016.2638923, IEEE
Transactions on Power Systems
10

high r/x ratios, large-scale transmission systems and ill- For the method proposed in this paper, the bus voltage
conditioned systems, proves the accuracy and robustness of magnitudes are calculated using equation (21). H, N , M , L
the two proposed power flow models. In large-scale cases, in equation (21) are represented by
the FDLPF model is proven to be much faster than the 
H N
 
BLL −GLL

DLPF model, thereby demonstrating the effectiveness of the =−
M L GLL BLL
approximation technique. Finally, the sensitivity factors and
distribution shift factors are derived from the FDLPF model and
−1


as a promising application. L̃ = − BLL + GLL BLL GLL
We envision that the proposed models can serve as useful According to (21) and Lemma 1, we have
tools in power system planning and operation. Specifically, the
−1
DLPF model has the potential to be employed in optimization VL =L̃−1 (QL + BLR V0 ) − L̃−1 GLL BLL (PL − GLR V0 )
−1
problem in place of the AC power flow equations, which may =L̃−1 QL − L̃−1 GLL BLL PL
allow nonlinear, non-convex problems to be transformed into −1
+ L̃−1 BLR + GLL BLL


GLR V0
linear programming problems. In addition, the FDLPF model  
−1
is characterized by high computational efficiency and has the =L̃−1 QL − L̃−1 GLL BLL PL + L̃−1 L̃ · 1 V0
potential to be applied in contingency analysis and probabilis- −1
=1V0 + L̃−1 QL − L̃−1 GLL BLL PL
tic load flow problems. By utilizing the FDLPF model, voltage
(A1)
magnitude and reactive power can be considered as in the AC
where V0 , VL are the voltage magnitudes at slack bus and the
power flow model with a computational efficiency that is as
remaining P Q buses.
high as that of the classical DC power flow model.
As to the model in [11]. the expression for bus voltage
magnitudes in the radial network is given as:
A PPENDIX
1

Lemma 1: Let Y denote the admittance matrix of a radial VL = 1V0 + Re ZLL S L (A2)
V0
distribution system, where there are typically no P V buses
where ZLL is the inverse matrix of YLL . SL is the nodel
and the shunt admittances are negligible. Partition Y by the
complex power injection at P Q buses. S L is the complex
type of slack bus and P Q buses as follows:
  conjugate.
YRR YRL With the help of Lemma 2, equation (A2) can be trans-
Y =
YLR YLL formed as follows:
then 1 
−1

VL =1V0 + Re (GLL + jBLL ) (PL − jQL )
YLR = −YLL · 1 V0
1  −1

where 1 is the vector of all ones. =1V0 + −L̃−1 GLL BLL PL + L̃−1 QL
V0
Proof 1: As the shunt admittances are negligible, we have (A3)
From the derivation above, it is easy to verify that the
Y ·1=0 difference between (A1) and (A3) is tiny in radial distribution
then system. If the bus voltage magnitude is 1 p.u. at the slack
YLR + YLL · 1 = 0 bus of the radial network, then the two methods give identical
results.
Therefore,
YLR = −YLL · 1 R EFERENCES
Lemma 2: For a complex matrix C = A + jB in which [1] M. K. Enns, J. J. Quada, and B. Sackett, “Fast linear contingency
analysis,” IEEE Trans. Power App. Syst., vol. PAS-101, no. 4, pp. 783–
the A and B are nonsingular matrices, the inverse matrix of 791, April 1982.
C is [2] R. Billinton and R. N. Allan, “Reliability evaluation of engineering

−1
−1 systems,” 1992.
C −1 = B + AB −1 A AB −1 − j B + AB −1 A [3] Z. Hu and X. Wang, “A probabilistic load flow method considering
branch outages,” IEEE Trans. Power Syst., vol. 21, no. 2, pp. 507–514,
Proof 2: May 2006.
[4] F. Dorfler and F. Bullo, “Novel insights into lossless ac and dc power

−1 flow,” in Power and Energy Society General Meeting, 2013 IEEE, July
C −1 C = B + AB −1 A AB −1 A 2013, pp. 1–5.

−1 [5] F. Li and R. Bo, “DCOPF-based LMP simulation: Algorithm, compar-
+ B + AB −1 A B ison with ACOPF, and sensitivity,” IEEE Trans. Power Syst., vol. 22,

−1 −1
+j B + AB A AB −1 B no. 4, pp. 1475–1485, Nov 2007.
[6] F. Lee, J. Huang, and R. Adapa, “Multi-area unit commitment via
−1
−j B + AB −1 A


A sequential method and a dc power flow network model,” IEEE Trans.
Power Syst., vol. 9, no. 1, pp. 279–287, Feb 1994.
−1 −1
B + AB −1 A



= B + AB A [7] B. Stott, J. Jardim, and O. Alsac, “DC power flow revisited,” IEEE
Trans. Power Syst., vol. 24, no. 3, pp. 1290–1300, Aug 2009.
=I [8] V. Sarkar and S. Khaparde, “A comprehensive assessment of the evolu-
tion of financial transmission rights,” IEEE Trans. Power Syst., vol. 23,
Therefore, C −1 is the inverse matrix of C. no. 4, pp. 1783–1795, Nov 2008.

0885-8950 (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/TPWRS.2016.2638923, IEEE
Transactions on Power Systems
11

[9] PJM Training Materials. [Online]. Available: http://pjm.com/ Jingwei Yang (S’15) received the B.S. degree from
[10] A. Leite da Silva and V. Arienti, “Probabilistic load flow by a multilinear the Electrical Engineering Department of Tsinghua
simulation algorithm,” Generation, Transmission and Distribution, IEE University in China in 2015.
Proceedings C, vol. 137, no. 4, pp. 276–282, Jul 1990. He is currently pursuing Ph.D. degree in Tsinghua
[11] S. Bolognani and S. Zampieri, “On the existence and linear approxima- University. His research interests include optimal
tion of the power flow solution in power distribution networks,” IEEE power flow, renewable energy and multiple energy
Trans. Power Syst., vol. 31, no. 1, pp. 163–172, Jan 2016. system integration.
[12] S. Fatemi, S. Abedi, G. Gharehpetian, S. Hosseinian, and M. Abedi,
“Introducing a novel dc power flow method with reactive power consid-
erations,” IEEE Trans. Power Syst., vol. 30, no. 6, pp. 3012–3023, Nov
2015.
[13] D. Van Hertem, J. Verboomen, K. Purchala, R. Belmans, and W. Kling,
“Usefulness of dc power flow for active power flow analysis with flow
controlling devices,” in The 8th IEE International Conference on AC
and DC Power Transmission, 2006., March 2006, pp. 58–62.
[14] J. J. Grainger and W. D. Stevenson, “Power systems analysis mcgraw-
hill,” 1994.
[15] Source code for the linear power flow models, Github. [Online].
Available: https://github.com/Jingwei-THU/linear-power-flow
[16] M. E. Baran and F. F. Wu, “Network reconfiguration in distribution
systems for loss reduction and load balancing,” IEEE Transactions on Ning Zhang (S’10-M’12) received both a B.S. and
Power Delivery, vol. 4, no. 2, pp. 1401–1407, Apr 1989. Ph.D. from the Electrical Engineering Department
[17] A. Ymeri, L. Dervishi, and A. Qorolli, “Impacts of distributed generation of Tsinghua University in China in 2007 and 2012,
in energy losses and voltage drop in 10 kv line in the distribution sys- respectively.
tem,” in Energy Conference (ENERGYCON), 2014 IEEE International. He is now an Assistant Professor at the same
IEEE, 2014, pp. 1315–1319. university. His research interests include stochastic
[18] J. Huang, S. Ge, J. Han, H. Li, X. Zhou, H. Liu, B. Wang, and Z. Chen, analysis and simulation of renewable energy, power
“A diagnostic method for distribution networks based on power supply system planning and scheduling with renewable en-
safety standards,” Protection and Control of Modern Power Systems, ergy, and multiple energy system integration.
vol. 1, no. 1, pp. 1–8, 2016.
[19] Z. Esau and D. Jayaweera, “Reliability assessment in active distribution
networks with detailed effects of PV systems,” Journal of Modern Power
Systems and Clean Energy, vol. 2, no. 1, pp. 59–68, 2014.
[20] R. Zimmerman, C. Murillo-Sanchez, and R. Thomas, “Matpower:
Steady-state operations, planning, and analysis tools for power systems
research and education,” IEEE Trans. Power Syst., vol. 26, no. 1, pp.
12–19, Feb 2011.
[21] P. S. Seshadri and A. D. Patton, “Bus voltage sensitivity: an instrument
for pricing voltage control service,” in Power Engineering Society
Summer Meeting, 1999. IEEE, vol. 2, 1999, pp. 703–707 vol.2.
Chongqing Kang (M’01-SM’07-F’17) received the
Ph.D. degree from the Department of Electrical
Engineering in Tsinghua University, Beijing, China,
in 1997.
He is currently a Professor in Tsinghua University.
His research interests include power system plan-
ning, power system operation, renewable energy, low
carbon electricity technology and load forecasting.

Qing Xia (M’01-SM’08) received the Ph.D. degree

from the Department of Electrical Engineering in
Tsinghua University, Beijing, China, in 1989.
He is currently a Professor in Tsinghua University.
His research interests include electricity markets,
generation scheduling optimization and power sys-
tem planning.

0885-8950 (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.