5200 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 53, NO.

6, NOVEMBER/DECEMBER 2017

Power Flow Analysis Considering Automatic

Generation Control for Multi-Area Interconnection
Power Networks
Xiaoming Dong , Member, IEEE, Hua Sun, Chengfu Wang , Member, IEEE, Zhihao Yun, Yiming Wang,
Penghui Zhao, Yuanyuan Ding, and Yong Wang

Abstract—In addition to maintaining the power balance, the I. INTRODUCTION

major tasks of multi-area automatic generation control (AGC) in-
N OPEN question regarding the development of China is
volve regulating the power exchange among subareas. This study
presents a new power flow model allowing for multi-area AGC that
achieves cooperation among the participating generators, which
A how to address the shortage of electric power in develop-
ing areas [1] and make the best use of the new energy sources
are deployed separately in the interconnected subarea networks. [2], which are widespread in low-population areas. There is also
The formulations of the power exchange among the subareas are
derived from the active power flow equations of the tie-lines. The
a central concern about promoting the global energy intercon-
node parameters representing the level and allocation of the un- nection (GEI). As a result, GEI-style power networks that are
balanced power are introduced. Then, the proposed power flow capable of transferring power from sources to load centers in a
model is established by integrating the formulations of the node secure, economical, and effective manner must be planned and
power injection and power exchange. Meanwhile, the unknown established. China has a plan to transmit electric power from
variables in the integrated model can be synchronously adjusted
by the Newton iteration method, which depends on the integrated
west to east [3] and establish a transmission network framework
Jacobian matrix. Furthermore, being improved based on the in- based on the ultrahigh voltage alternating current [4] and ultra-
tegrated formulations, some related algorithms, including but not high voltage direct current [5]. Once completed, the transmis-
limited to sensitivity analysis, are expected to produce more precise sion system will cover multiple areas through the interconnected
solutions. The case studies of the IEEE 5 bus test system, the IEEE network architectures and ultimately achieve a wider range of
RTS-1996 test system, and the IEEE 300 bus test system illustrate
the convergence and efficiency of the proposed power flow method.
transfer and interoperability of energy.
Moreover, the case studies also demonstrate that the integrated For such a project, analysis and calculation methods that are
formulations based sensitivity analysis can generate more reliable relevant to the GEI, such as the power flow method, are par-
results. ticularly important. Newton–Raphson algorithm based power
Index Terms—Power system analysis computing, power system flow method centers on obtaining the system state by solving
interconnection, power system control, power transmission, power the nonlinear power flow equations representing the active and
flow. reactive power balance. Moreover, the relationships between the
system state variables and control variables, which are expressed
Manuscript received July 24, 2016; revised November 1, 2016, January 25, by the equations, are valuable for system operation. As an opti-
2017, and June 26, 2017; accepted August 1, 2017. Date of publication August mal linear approximation of the relationship among the physical
10, 2017; date of current version November 20, 2017. Paper 2016-PSEC-
0696.R3, presented at 2016 IEEE Industry Applications Society Annual Meet- variables, the Jacobian matrix, established in the iterative pro-
ing, Portland, OR, USA, Oct. 2–6, and approved for publication in the IEEE cess, is particularly applicable for system stability analysis [6],
TRANSACTIONS ON INDUSTRY APPLICATIONS by the Power System Engineering [7]. However, the traditional power flow method [8]–[10] fails
Committee of the IEEE Industry Applications Society. This work was sup-
ported in part by the National Natural Science Foundation of China under Grant to consider power balancing control and power exchange con-
51607107 and Grant 51477091, in part by the Fundamental Research Funds of trol between interconnected power networks. Accordingly, the
Shandong University under Grant 2016TB001 and Grant 2015GN001, and in conventional power flow calculation is incapable of meeting the
part by the Science and Technology Project of the State Grid Corporation of
China under Grant SGSDDK00KJJS1600061. (Corresponding author: Xiaom- requirements of state analysis for large-scale multi-area power
ing Dong.) systems. Therefore, this study proposes an improved alternat-
X. Dong, C. Wang, Z. Yun, Y. Wang, P. Zhao, Y. Ding, and Y. Wang are ing current power flow method for analyzing the system state,
with the Key Laboratory of Power System Intelligent Dispatch and Control of
the Ministry of Education, School of Electrical Engineering, Shandong Univer- facilitating the interconnection and coordination of subarea
sity, Jinan 250061, China (e-mail: [email protected]; wangcf@sdu. networks. For a multi-area power system, the major tasks of
edu.cn; [email protected]; [email protected]; [email protected]; the automatic generation control (AGC) system are to support
[email protected]; [email protected]).
H. Sun is with the Department of Electrical Automation, Shandong the internal unbalanced power for each subarea and maintain the
Labor Vocational and Technology College, Jinan 250022, China (e-mail: power exchanges between subareas. Once given the base values
[email protected]). and participation factors of the AGC units, the AGC system
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. can adjust the frequency quality with certain power exchange
Digital Object Identifier 10.1109/TIA.2017.2738618 between multiple areas. The outputs of the AGC generators are

0093-9994 © 2017 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.

Authorized licensed use limited to: VEL TECH MULTI TECH Dr RR Dr SR ENGG. COLLEGE. Downloaded on July 28,2023 at 03:54:11 UTC from IEEE Xplore. Restrictions apply.
 DONG et al.: POWER FLOW ANALYSIS CONSIDERING AGC FOR MULTI-AREA INTERCONNECTION POWER NETWORKS 5201

dependent on the system frequency deviation and power devia- model. The power transferred by high voltage direct current
tion on the interregional tie-lines [11]. As such, the power flow lines can be treated as power injection at the conventional PV
formulations must be improved for a large interconnected power or PQ bus [23]. Furthermore, for a more precise expression,
grid by considering the operation of the multi-area AGC. the static unified power flow controller formulation presented in
Because multi-area AGC is ignored in the conventional power [10] could be included in the power flow formulations to estab-
flow formulations, in which the unbalanced power must be ad- lish an integrated power flow model, which is not the primary
dressed by a sole slack bus, an error has been introduced in the focus of this paper.
results of the system status. Therefore, improved approaches
for the power flow calculation of the interconnected power grid II. DESCRIPTION OF CONVENTIONAL POWER FLOW
have been proposed. In [12], a new method is proposed to split FORMULATIONS
the two areas into two systems, the sender and receiver, and
the two subsystems are calculated separately by alternatively With the assumption that only one slack bus is involved in
iterating to obtain the final result. Yu et al. [13]–[16] solved the the conventional power flow model, the formulations involving
distributed power flow formulations of a multi-area power grid the active power balance equations for all of the buses except
using the asynchronous iterative method, in which the power for the slack bus and reactive power balance equations for all
exchange and the output of AGC generators are adjusted sepa- PQ buses are given as
rately from the key iterative process. In [17], by classifying the 
operating units into the slack mode and the control mode, the fi = Qi − hQ i (V, θ) = 0 (i = 1, 2, . . . , m)
(1)
unbalanced power calculated using the sensitivity method [18] fm +j = Pj − hPj (V, θ) = 0 (j = 1, 2, . . . , n − 1)
is distributed to the different categories of units. These stud-
ies illustrate that the power flow calculation of interconnected where P and Q represent the active power and reactive power
networks is a pressing issue. injected into the network, respectively; f represents the active
However, current approaches for calculating the power flow and reactive power balance equation; and subscripts n and m
of interconnected power networks are dependent on the asyn- are the total number of buses and the number of PQ buses,
chronous iteration, in which the variables cannot be calculated respectively. The unknown vector V is an m-dimensional vector
and updated synchronously because of the lack of integrated that represents the voltage magnitudes of all PQ buses
formulations of Jacobian matrix in each iteration. For certain  
analyses of power flow-based problems, such as static voltage V = V1 V2 · · · Vm . (2)
stability [19], [20], the synchronous, or alternatively integrated,
The unknown vector θ is an n − 1-dimensional vector that
Newton-iteration formulations must be established for the cal-
represents the voltage phase angle of all PV and PQ buses
culation. For example, in each predictor–corrector step of a
continuous power flow [21] calculation, the integrated Jacobian  
θ = θ1 θ2 · · · θn −1 . (3)
matrix must be established to confirm the limit-induced bifur-
cation point or saddle node bifurcation point. Moreover, some hP (V, θ) and hQ (V, θ) are expressed as follows:
typical algorithms for static voltage stability, such as modal
analysis [6] and the V–Q sensitivity method [7], are found dur- 
n

ing the analysis of the integrated Jacobian matrix. Therefore, hPi (V, θ) = Vi Vj (Gij cos θij + Bij sin θij ) (4)
j ∈i
the integrated power flow formulations and integrated Jacobian
matrix must be generated for the static voltage stability anal- 
n

ysis of multi-area interconnected grids [22]. As a result, this hQ

i (V, θ) = Vi Vj (Gij sin θij − Bij cos θij ) (5)
study proposed an integrated power flow model that considers j ∈i
the multi-area AGC operation by introducing the tie-line power
where Gij and Bij denote the mutual impedance between buses i
control model and multi-area AGC distribution model.
and j, respectively, and θij = θi − θj is the phase angle between
The remainder of this paper is organized as follows. Section II
complex bus voltages Vi ∠ θi and Vj ∠ θj .
presents the model of conventional power flow formulations. In
Section III, the formulations of multi-area AGC, which consist
of the subarea unbalanced power operation and the interface III. POWER FLOW MODEL CONSIDERING MULTI-AREA AGC
power flow control among subareas, are derived. Using the anal- The improvement approach presented below is used to inte-
yses in Section III, the integrated formulation and iteration of grate the operation of multi-area AGC into the formulations of
the new power flow model is proposed in Section IV. Section V power flow.
presents the case studies based on the IEEE 5 bus test system,
IEEE RTS-1996 test system, and IEEE 300 bus test system to
demonstrate the performance of the proposed formulations. In A. Considering the Subarea Unbalanced Power Operation
Section VI, the main results are summarized and the key con- It is assumed that the system is divided into d areas and
clusions are drawn. that each area possesses an individual AGC system. The AGC
The proposed algorithm only accounts for the multi-area alter- participating factor of each bus corresponding to different ar-
nating current grid and ignores the direct current transmission eas is defined by α (α ≥ 0). Thus, the factor matrix could be

Authorized licensed use limited to: VEL TECH MULTI TECH Dr RR Dr SR ENGG. COLLEGE. Downloaded on July 28,2023 at 03:54:11 UTC from IEEE Xplore. Restrictions apply.
 5202 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 53, NO. 6, NOVEMBER/DECEMBER 2017

obtained as
⎡ ⎤
a11 a12 ··· a1d
⎢ a21 a22 ··· a2d ⎥
⎢ ⎥
A = ⎢. .. .. .. ⎥. (6)
⎣ .. . . . ⎦
an 1 an 2 ··· an d Fig. 1. Schematic diagram of interconnected networks.
aik represents any element in matrix A in the ith row and kth
column, corresponding to the ith bus and kth area, respectively. In (11), the number of unknown variables (m + n + d − 1)
If the ith bus is not in the kth area, then aik = 0; if the ith bus is greater than the number of equations m + n, thus requiring
is in the kth area but does not participate in the AGC operation, d − 1 equations to be added for solvability.
then aik = 0; if the ith bus is in the kth area and participates
in the AGC operation, then aik = 0. Elements in each column B. Considering the Interface Power Flow Control Among
satisfy the following relationship: Subareas

n For multi-area connected networks, several AGC systems
aik = 1. (7) cooperate with each other to enable control of the power ex-
i=1 changes. An example of an interconnected power grid between
The conventional power flow model consists of a sole slack two areas is shown in Fig. 1.
bus, known as the V θ bus with constant V and θ (θ = 0), The power measurement terminals could be located at either
which compensates the system’s unbalanced real power. How- side of the tie-lines, which are denoted as “M” in Fig. 1. The
ever, in actual power system operation, the unbalanced power is measuring data from the lth tie-lines are shared to both the
supported by a group of generators, which are distributed and sending system S1 and the receiving system S2 . For a certain
participate in the AGC operation according to the participat- mode of operation, AGC units in each subarea would take charge
ing factor shown in (6). Because the AGC system in each area of the exchange power as being a settled value, which can be
operates independently, the unbalanced power factors for the expressed as
network consisting of d areas are defined by 
gk = ϕl Pl − ΔPk = 0 (k = 1, 2, . . . , d) (13)
 T
μ = μ1 μ2 · · · μd . (8) l∈L k

with
If μd > 0, then the output of the AGC units in the dth area
increases to adjust the unbalanced power, and vice versa. Thus, Pl = Vi2 Gl − Vi Vj (Gl cos θij + Bl sin θij ) (14)
the following relations can be obtained in terms of the vector μ 
and matrix A: 1 (i ∈ Ωk , j ∈
/ Ωk )
ϕl = (15)
  T −1 (i ∈ / Ωk , j ∈ Ωk )
Aμ = β1 β2 · · · βn
 (9) where Pl represents the active power transferred by the lth tie-
βi = dj=1 αij μj
line; subscripts i and j represent the two sides of the tie-line;
where the parameter βi represents the adjustment power injected ϕl represents the direction of the power transmission for the lth
into the network at the ith bus. Including the adjustment power tie-line; Lk is the collection of the tie-lines connected to the
βi in the equations allows (1) to be restated as kth area; Gl and Bl represent the real and imaginary parts of
 impedance for the lth tie-line; Ωk is the collection of the buses
fi = Qi − hQi (V, θ) = 0 (i = 1, 2, . . . , m) located in the kth area; ΔPk is the value of the exchange power
fm +j = Pj − hPj (V, θ) + βj = 0 (j = 1, 2, . . . , n) injected into the kth area, ΔPk < 0 indicates feeding power
(10) into the kth area, and ΔPk < 0 indicates outputting power into
the kth area. For a multi-area power grid with d subareas, the
which consists of the real power balance equation of the slack power exchange of all subareas can be given by the following
bus fm +n , such that the number of equations in (10) is one relation:
more than that in (1). As a result, the unbalanced power would 
 d  d
be adjusted by a group of buses that participate in AGC instead Pl = ΔPk = 0. (16)
of the conventional sole “slack bus,” which still provides a ref- k =1 l∈L k k =1
erence for the voltage phase angle (θ = 0). Thus, (10) can be
expressed in the compact form as Because the total amount is equal to 0, the formulation for
any d − 1 subareas could be sufficient to describe the power
f (X) = 0 (11) exchange relationship of the entire power grid.
where X represents the unknown vector of all of the variables
IV. INTEGRATED FORMULATION AND ITERATION
in (11)
  Taking any d − 1 of the d equations in (13), the first d − 1
X = [V, θ, μ] = x1 x2 · · · x(m +n +d−1) . (12) equations, for example, the relations of power exchange, can be

Authorized licensed use limited to: VEL TECH MULTI TECH Dr RR Dr SR ENGG. COLLEGE. Downloaded on July 28,2023 at 03:54:11 UTC from IEEE Xplore. Restrictions apply.
 DONG et al.: POWER FLOW ANALYSIS CONSIDERING AGC FOR MULTI-AREA INTERCONNECTION POWER NETWORKS 5203

described by ∂f /∂μ is expressed as

  ⎡ ∂ β1 ⎤
gk = l∈Ω k ϕl Pl − ΔPk = 0 ∂ β1
··· ∂ β1
g (X) = . (17) ⎢ ∂ β2
∂ μ1 ∂ μ2 ∂ μd
⎥
(k = 1, 2, . . . , d − 1) ⎢ ∂ β2
··· ∂ β2⎥
∂f ⎢ ∂ μ1 ∂ μ2 ∂ μd⎥
=⎢ ⎥
.. ⎥ . (25)
Thus, the compact formulation (18), in which the number ∂μ ⎢ .
⎢ ..
.. ..
of variables and the number of equations both equal m + n, ⎣ . . . ⎥
⎦
represents the integrated power flow model for an interconnected ∂ βn ∂ βn
··· ∂ βn
∂ μ1 ∂ μ2 ∂ μd
power grid
 The elements of the matrix ∂f /∂μ are given by the following
f (X) = 0
F (X) = . (18) relations:
g (X) = 0 ∂βb
= αbk (k = 1, 2, . . . , d; b = 1, 2, . . . , n) . (26)
The Newton-iteration scheme in the following equation is ∂μk
established to solve the nonlinear, algebraic, equations in (18): Thus, (25) can be abbreviated as follows:
  
F(X(t) ) = J(t) ΔX(t) ∂f A
(19) = (27)
X(t+1) = X(t) + ΔX(t) ∂μ 0
where (t) and (t + 1) represent the steps of the iterations; in where matrix A is the AGC participating factor given in (6) and
particular, t = 0 denotes the initial iteration. The unknown vec- 0 represents a matrix whose elements equal zero.
tor X, which is comprised of the unknown vectors (2), (3), and ∂g/∂V is expressed as
(8), is elaborated and initialized for a “flat start,” as ⎡ ∂g ⎤
⎧ (0)   ∂ V1
1 ∂ g1
∂ V2 ··· ∂ g1
∂ Vm
⎢ ⎥
⎨V = 1 1 · · · 1
⎪ ⎢ ∂ g2 ∂ g2
· · · ∂ g2 ⎥
⎢ ⎥
θ (0) = 0 0 · · · 0 . (20) ∂g ⎢ 1∂ V ∂ V 2 ∂ V m
⎥
⎪
⎩ (0)   = ⎢. . . ⎥ (28)
∂V ⎢ . .. ⎥
μ = 0 0 ··· 0 ⎢ .. .. .. ⎥
⎣ ⎦
The following relation is used as the convergence standard of ∂ g ( d −1) ∂ g ( d −1)
· · ·
∂ g ( d −1)
∂ V1 ∂ V2 ∂ Vm
the Newton iteration process:
   where the elements can be obtained as
σ = F(X(t) )1  
(21) ∂gk  ∂Pl
σ ≤ σm = ϕl (k = 1, 2, . . . , d − 1; p = 1, 2, . . . , m)
∂Vp ∂Vp
l∈Ω k
where σ is the convergence standard parameter, which is updated (29)
in each iteration step, and σm is a small positive number that is
given in advance to guarantee the accuracy of the solution. In with
⎧
addition, σm is set to 10–5 in all the subsequent case studies. ⎪2V G − Vj (Gl cos θij + Bl sin θij ) (p = i)
∂Pl ⎨ i l
In (19), J(t) represents the extended integrated Jacobian ma- = −Vi (Gl cos θij + Bl sin θij ) (p = j) .
trix, which is expressed as follows: ∂Vp ⎪
⎩
 ∂f ∂f ∂f
 0 (p = i, p = j)
∂F ∂V ∂θ ∂μ (30)
J= = ∂g ∂g (22)
∂X
∂V 0 ∂θ ∂g/∂θ in (22) can be expressed as
where ∂f /∂V is expressed as ⎡ ∂ g1 ⎤
⎡ ∂f ⎤ ∂ θ1
∂ g1
∂ θ2 ··· ∂ g1
∂ θ n −1
1 ∂f 1
··· ∂ f1 ⎢ ∂ g2 ∂ g2
··· ∂ g2 ⎥
⎢ ∂ V1 ∂ V2 ∂ Vm
⎥ ∂g ⎢ ⎥
=⎢ ⎥
∂ θ1 ∂ θ2 ∂ θ n −1
⎢ ∂ f2 ⎥ (31)
∂θ ⎢ ⎥
. .. .. ..
⎢ ∂ V1
∂ f2
··· ∂ f2
⎥ ⎣ .. . . . ⎦
∂f ⎢ ∂ V2 ∂ Vm
⎥
= ⎢. .. .. ⎥. (23) ∂ g ( d −1) ∂ g ( d −1)
···
∂ g ( d −1)
∂V ⎢ .. .. ⎥ ∂ θ1 ∂ θ2 ∂ θ n −1
⎢ . . . ⎥
⎣ ⎦ where the element is expressed as
∂ f( m + n ) ∂ f( m + n )
···
∂ f( m + n )
 
∂ V1 ∂ V2 ∂ Vm ∂gk  ∂Pl
= (k = 1, 2, . . . , d − 1; q = 1, 2, . . . , n − 1)
∂f /∂θ is given by ∂θq ∂θq
l∈Ω k
⎡ ∂ f1 ⎤
∂ θ1
∂ f1
∂ θ2 ··· ∂ f1
∂ θ n −1
(32)
⎢ ∂ f2 ⎥
⎢ ∂ f2
··· ∂ f2 ⎥ with
∂f ⎢ ∂ θ1 ∂ θ2 ∂ θ n −1 ⎥ ⎧
=⎢ ⎥. (24) ⎨Vi Vj (Gl sin θij − Bl cos θij )
⎪ (q = i)
∂θ ⎢ .
⎢ ..
..
.
..
.
..
.
⎥
⎥ ∂Pl
⎣ ⎦ = Vi Vj (Bl cos θij − Gl sin θij ) (q = j) . (33)
∂ f( m + n ) ∂ f( m + n ) ∂ f( m + n ) ∂θq ⎪
⎩
∂ θ1 ∂ θ2 ··· ∂ θ n −1 0 (q = i, q = j)

Authorized licensed use limited to: VEL TECH MULTI TECH Dr RR Dr SR ENGG. COLLEGE. Downloaded on July 28,2023 at 03:54:11 UTC from IEEE Xplore. Restrictions apply.
 5204 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 53, NO. 6, NOVEMBER/DECEMBER 2017

TABLE II
STEPS OF THE ITERATION PROCESS

Variables t

1 2 3 4 5

V1 1.000 0.966 0.875 0.862 0.862

V2 1.000 1.105 1.079 1.078 1.078
V3 1.000 1.059 1.038 1.036 1.036
θ1 0.000 –0.469 –0.461 –0.465 –0.465
θ2 0.000 –0.075 –0.070 –0.070 –0.070
θ3 0.000 –0.505 –0.458 –0.456 –0.456
Fig. 2. Diagram of the IEEE 5 bus test system. θ5 0.000 –0.436 –0.384 –0.381 –0.381
μ1 0.000 0.001 0.001 0.001 0.001
TABLE I μ2 0.000 –0.280 –0.016 –0.001 –0.001
INFORMATION FOR AGC σ 3.001 0.396 0.011 2E–4 2E–7

Area AGC Bus & a Measure Point (Bus i–Bus j) Power Exchange (p.u.)

1 4&1 2 (2–3) 3.0005

2 5&1 2 (2–1) –3.0005

At the beginning of integrated power flow, an initial point

must be given for the iteration process. Variables represent-
ing power exchange are initialized to 0. Voltage magnitudes
and voltage angles are initialized for a “flat start.” Thus, all of Fig. 3. Schematic of the IEEE RTS-1996 test system.
the variables are synchronously updated with the iterations de-
scribed in (19). Because the relation of the convergence standard Buses i and j represent the buses on the two sides of the tie-
in (21) is confirmed, the numerical solutions of the integrated line and the measurement points are specified to be located at
power flow equations can be obtained. bus i. The results of the Newton iteration calculation process
using the presented methods are presented in Table II. After
V. CASE STUDIES five iterations, the convergence standard parameter σ decreases
continuously to finally become less than σm . All of the variables
In this section, the new power flow model is applied to three in the iteration process show good performance in convergence.
case studies to verify the algorithm.
Case 1: is the IEEE 5 bus test system in which the power B. Case 2
grid is divided into two areas. The simplicity of this case allows
one to explore the detailed iteration process of the proposed In this section, the classical method shown in [15] and the
power flow method. proposed method are applied to calculate the power flow of
Cases 2: and 3 are based on the IEEE RTS-1996 test system the IEEE RTS-1996 test system for comparison. As observed
comprised of three subareas to form the interconnected grid. in Fig. 3, the IEEE RTS-1996 test system comprises of three
Case 2 centers on demonstrating the applicability of the im- subarea grids represented by area 1, area 2, and area 3 separately.
proved algorithm for analyzing complex systems that are char- The detailed information of the interconnection among the buses
acterized by a larger scale and more subareas. In case 3, the is shown in Fig. 4. The partition information and setting values
V–Q sensitivities are calculated according to the integrated Ja- of the AGC are shown in Table III.
cobian matrix derived in this study. Moreover, the values of the The calculation process of the classical method can be sum-
sensitivities are compared with the results of the conventional marized in the following steps.
method. Step 1: Calculate the conventional power flow, where only
In case 4, the calculation results for a large-scale power sys- one sole slack bus is deployed to support the unbal-
tem, IEEE 300 bus test system, with the conventional method anced power and the power exchange is uncontrolled.
and the presented method are deployed. The comparison of the Step 2: Calculate the current exchange power of each sub-
performances of the different methods is elaborated. area. Calculate the deviations between the current
exchange power and target values.
Step 3: If the deviation is sufficiently small, then the solution
A. Case 1 is obtained and the results can be output; else, go to
As shown in Fig. 2, the IEEE 5 bus test system is divided into step 4.
two subareas. Step 4: Adjust the active power output of the generators that
The generator bus voltage magnitudes, load power, and participate in the AGC in accordance with the power
branch parameters are based on [10]. Table I presents the in- exchange deviation and AGC participating factors.
formation of the AGC system. Return to step 1.

Authorized licensed use limited to: VEL TECH MULTI TECH Dr RR Dr SR ENGG. COLLEGE. Downloaded on July 28,2023 at 03:54:11 UTC from IEEE Xplore. Restrictions apply.
 DONG et al.: POWER FLOW ANALYSIS CONSIDERING AGC FOR MULTI-AREA INTERCONNECTION POWER NETWORKS 5205

TABLE IV
PARAMETERS OF THE AGC FOR SIX AREAS

Area AGC Bus & a

1A 101 & 0.3; 102 & 0.3; 107 & 0.4

1B 113 & 0.1; 118 & 0.3; 121 & 0.3; 123 & 0.3
2A 201 & 0.3; 202 & 0.3; 207 & 0.4
2B 213 & 0.1; 218 & 0.3; 221 & 0.3; 223 & 0.3
1A 201 & 0.3; 202 & 0.3; 207 & 0.4
3A 213 & 0.1; 218 & 0.3; 221 & 0.3; 223 & 0.3

16.54 ms. As such, each iteration in the improved method takes

approximately 3.31 ms, which is greater than that in conven-
tional method because of the increase in the dimension of the
equations. However, the total calculation time using the new
method is shorter because of the reduced number of iterations.
Note that, in the improved power flow model, the unbalanced
power factors shown in (8) reflect the operation states of the
AGC under the specified AGC participating factors shown in
(6). The relations in (17) guarantee the fixed value of the inter-
area exchange power.
In order to analyze systems with more subareas, each of the
three areas in the IEEE RTS-1996 test system is divided into
two parts. Then, the diagram of the test system with six areas,
as indicated by 1A, 1B, 2A, 2B, 3A, and 3B in Fig. 4, can be
obtained. The information of AGC sets for six areas is presented
in Table IV.
For six-area test system power flow calculation using the pre-
sented approach, the solution time is approximately 34.95 ms
Fig. 4. Diagram of the interconnection for the test system.
covering five iterations. And it costs about 31.13 ms for 15 iter-
ations since the classical method is applied. It is indicated that
TABLE III the increase in the number of areas and tie lines causes more
PARAMETERS OF THE AGC FOR THREE AREAS burden in solving the equations in the iterative process. The
excessive memory access caused by newly added elements in a
Area AGC Bus & a Power Exchange (p.u.) spare matrix is considered as the major factor that results in this
1 101 & 0.3; 102 & 0.3; 107 & 0.3; 113 & 0.1 –2.0 phenomenon. However, the total consumption of the computa-
2 201 & 0.3; 202 & 0.3; 207 & 0.3; 213 & 0.1 1.0 tion for the presented method still remains in an acceptable level.
3 301 & 0.3; 302 & 0.3; 307 & 0.3; 313 & 0.1 1.0 Note that, because the spare matrix technology is used, the com-
putation burden at different steps differs from each other. The
most time-consuming procedure is in the first iteration where the
Then, the power flow is calculated according to the steps framework of the spare matrix is established and then remains
of the classical method with the programming and testing com- constant in the rest of the iterations.
puter configuration specified as follows. The CPU, memory, and
storage are Intel Core i7-6600U @ 2.60 Hz Dual-Core Quad-
C. Case 3
threading, 16 GB 1867 MHz and Sandisc SSD 512G.
When the classical method is used for the calculation process, When the power flow model in this study is converged, the
the elapsed time of the entire procedure of the power flow cal- Jacobian matrix, shown in (22), can be obtained to calculate the
culation is approximately 18.31 ms, which is the average of the V–Q sensitivities of the load buses according to the following
values from 100 calculations. The total number of iterations of relation:
the power flow calculation is 8, which implies that each iteration  
S = det J−1 (34)
takes approximately 2.29 ms.
Next, the power flow is calculated using the proposed model where the elements of vector S corresponding to the reactive
with the same testing environment shown in Table III. As a re- power balance equations of the specified load buses are the
sult, the unbalanced power in each area can be gradually adjusted desired V–Q sensitivities. Because the operation of multi-area
with good convergence to reach equilibrium in five iterations. AGC is involved, the V–Q sensitivities based on the improved
The continuous decrease in σ with an increase in iteration steps model could be different than the values obtained using the
indicates the numerical convergence of the presented method. conventional method presented in [7]. Therefore, the V–Q sen-
The calculation time of the five iterations is approximately sitivities are calculated using the two methods for comparison.

Authorized licensed use limited to: VEL TECH MULTI TECH Dr RR Dr SR ENGG. COLLEGE. Downloaded on July 28,2023 at 03:54:11 UTC from IEEE Xplore. Restrictions apply.
 5206 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 53, NO. 6, NOVEMBER/DECEMBER 2017

Fig. 6. Results comparison for the IEEE-1996 3 area test system.

However, as the voltage instability emergencies always accom-

pany with heavy load, an open question is what the errors would
Fig. 5. Distribution of the errors for V–Q sensitivities. be under heavy load cases. In order to address this point, this
study deploys multiple “heavy load” case studies based on the
TABLE V
RESULTS OF THE V–Q SENSITIVITIES FOR TEST SYSTEM
IEEE RTS-1996 three-area test system, in which the active and
WITH THREE AREAS reactive power injection at each bus is simultaneously increased
by a certain percentage. It can be interpreted that the level of
Bus No s1 s2 E power load and generation is changed from a base state to var-
ious heavy levels. Thus, parameter λ is introduced into (10)
110 0.0337 0.0330 2.00% to indicate the load level. Then, the active and reactive power
112 0.0226 0.0222 1.80%
109 0.0339 0.0333 1.71% relations can be expressed as follows:
103 0.0603 0.0593 1.59% 
111 0.0175 0.0173 1.37% fi = λQi − hQ i (V, θ) = 0 (i = 1, 2, . . . , m)
fm +j = λPj − hPj (V, θ) + βj = 0 (j = 1, 2, . . . , n)
(37)
To quantitatively express the error between the two methods,
(35) is defined as where λ = 1.0 represents the base point with no increase in
  power injections and λ > 1.0 represents the heavy load cases.
 2 (s1 − s2 ) 
E=   × 100% (35) The calculation results of ME for different sets of λ are shown
s1 + s2 
in Fig. 6. By observing the dotted line in Fig. 6, the maximum
where s1 and s2 represent the sensitivities obtained by the im- error ME rises continuously with the increase in load level λ.
proved and conventional methods, respectively. Parameter E While λ is equal to 2.0, the value of ME reaches approximately
for each load bus in IEEE RTS-1996 test system is calculated 12%. It implies that the error of V–Q sensitivity analysis using
separately. the conventional method compared to the results of the new
Fig. 5 and Table V illustrate the distribution and detail of method could be significant under the system operation with
the results for the test system with three areas. As shown in heavy load.
Fig. 5, parts of the values are greater than 1%. As observed in Moreover, the red and blue curves in Fig. 6 illustrate the
Table V, E reaches 2.00% at load bus 110, followed by 1.80% performance of iterations corresponding to the conventional
at bus 112, 1.71% at bus 109, etc. Then, the V–Q sensitivities method and new method, respectively. Although the results of
calculated using the presented method differ from the solutions the two methods are consistent with the same flat start and
of the conventional method. In other words, errors are introduced convergence standard, their rates of convergence are quite dif-
in the results of static voltage stability analysis based on the ferent from each other. The new method is shown with more
conventional power flow. In order to evaluate the extent of the reliability of convergence in contrast to the performance of the
errors in different cases, the maximum value of E for all the PQ conventional method. Furthermore, with the maximum allow-
buses is defined by able iterations set to be 500, the conventional method cannot
converge since λ is greater than 2.023. By comparison, the new
ME = Max { Ei | i = 1, 2, . . . , m} . (36)
method is able to converge until λ reaches 2.216. It is reasonable
to explain the high rate and reliable converge of the proposed
Then, the sensitivities for the IEEE RTS-1996 test system with method because all the correction values for the variable devi-
six areas are separately calculated with the presented method and ations including the power exchange through the tie-lines are
the classical method. As a result, ME reaches 1.11%, which obtained by the Newton method which converges quadratically.
indicate that the errors in VQ sensitivity analysis tend to be However, the conventional method, so called asynchronous it-
decreased while higher number of subareas are deployed. erative method, has no such numerical characteristic. Thus, it is
All the above-mentioned results are based on the information less probable that the new method faces convergence problems
of base point corresponding to the general operation condition. while analyzing the system with certain heavy load.

Authorized licensed use limited to: VEL TECH MULTI TECH Dr RR Dr SR ENGG. COLLEGE. Downloaded on July 28,2023 at 03:54:11 UTC from IEEE Xplore. Restrictions apply.
 DONG et al.: POWER FLOW ANALYSIS CONSIDERING AGC FOR MULTI-AREA INTERCONNECTION POWER NETWORKS 5207

TABLE VI study established a new power flow model that allows for multi-
PARAMETERS OF SUBAREAS FOR THE IEEE 300 BUS TEST SYSTEM
area AGC. First, the subarea unbalanced power parameters and
AGC participating factors were introduced into the power flow
Area AGC Bus & a Power Exchange (p.u.) equations to represent the performance of the AGC generators.
1 84 & 0.1; 7002 & 0.1; 7003 & 0.2; 1.0 Next, the analytical expression of the power exchange of the
7012 & 0.1; 7017 & 0.1; 7024 & subarea was derived by analyzing the transmission power of the
0.1; 7039 & 0.1; 7061 & 0.1;
7062& 0.1;
tie-lines. As a result, the integrated power flow formulations that
2 119 & 0.3; 143 & 0.1; 186 & 0.2 –0.4 consist of bus power injection equations and exchange power
187 & 0.2; 7130 & 0.2 equations were calculated using the Newton iteration method.
3 191 & 0.3; 198 & 0.1; 221 & 0.1; –0.6
227 & 0.1; 233 & 0.1; 236 & 0.1;
Moreover, the approach for V–Q sensitivity analysis was pre-
239 & 0.1; 241 & 0.1 sented based on the proposed integrated Jacobian matrix.
The proposed algorithm uses a detailed representation for
multi-area AGC to efficiently generate the power flow results.
The case studies of the IEEE 5 bus test system, IEEE RTS-1996
three-area test system, and IEEE 300 bus test system illustrated
the convergence and effectiveness of the algorithm. The reliable
rate of convergence of presented method is demonstrated by
the case studies with heavy load. The comparisons of the V–Q
sensitivities based on the proposed integrated Jacobian matrix
and conventional approach demonstrated that some errors exist
in multi-area system analysis. Although the range of errors tends
to be decreased with the increasing number of subareas, it could
be amplified to a considerable level for the cases of heavy load.
Since many power flow based algorithms have been pro-
Fig. 7. Iterations of different methods for the IEEE 300-bus test system.
posed and wildly used in analyzing power systems, the pre-
sented model is expected to be further applied to improve the
tools for power system analysis.
D. Case 4
In this case, the IEEE 300 bus test system, which consists of
three subareas and ten tie-lines, is analyzed using the conven-
tional method and the presented method. The detail information REFERENCES
involving the AGC and power exchanges of subareas is pre- [1] X. Yang, Y. Song, G. Wang, and G. Wang, “A comprehensive review
sented in Table VI. on the development of sustainable energy strategy and implementation in
China,” IEEE Trans. Sustain. Energy, vol. 1, no. 2, pp. 57–65, Jul. 2010.
With the parameters according to Table VI, the conventional [2] Z. Zhang et al., “Hubei power grid peak load regulation considering wind
method and the proposed method are both converged at the power transmitted by UHV network,” Elect. Power Autom. Equip., vol. 33,
base point. However, the iterations corresponding to the two no. 4, pp. 6–11, 2013.
[3] A. F. Gubina, X. Xiangyang, and S. Zhengmin, “Analysis and support
methods show different profiles as shown in Fig. 7. As can be policy recommendation of renewable energy sources in western China,”
seen in Fig. 7, the new method successfully converges within in Proc. Int. Conf. Power Syst. Technol., 2006, pp. 1–8.
four iterations which totally cost 248.35 ms. But the iterations [4] T. Zheng, P. L. Chen, Z. Qi, and V. Terzija, “A novel algorithm to avoid
the mal operation of UHV voltage-regulation transformers,” IEEE Trans.
of the conventional method reach 75 with the collapse time Power Del., vol. 29, no. 5, pp. 2146–2153, Oct. 2014.
approaching to 1.36 s. Although timespan for each iteration of [5] C. Zhang, L. Wang, Z. Guan, and F. Zhang, “Pollution flashover per-
the conventional method is 18.15 ms which is much less than formance of full-scale ±800 kV converter station post insulators at high
altitude area,” IEEE Trans. Dielect. Elect. Insul., vol. 20, no. 3, pp. 717–
that of the new method of 62.09 ms, the calculation process of 726, Jun. 2013.
the new method costs much less time due to fewer iterations. [6] B. Gao, G. K. Morison, and P. Kundur, “Voltage stability evaluation using
Moreover, the partial enlarged detail of the iterations of the modal analysis,” IEEE Trans. Power Syst., vol. 7, no. 4, pp. 1529–1542,
Nov. 1992.
conventional method is shown in the gray area. It indicates that [7] P. Kundur, Power System Stability and Control. New York, NY, USA:
although the value of σ is considerably small, a massive amount McGraw-Hill, 1994.
of iterations are still needed to yield an approximation solution [8] Y. S. Zhang and H. D. Chiang, “Fast newton-FGMRES solver for large-
scale power flow study,” IEEE Trans. Power Syst., vol. 25, no. 2, pp. 769–
with the given convergence standard. The root cause of this 776, May 2010.
undesirable phenomenon is that the conventional method lacks [9] B. Zhang and S. Chen, High Power Network Analysis. Beijing, China:
schemes to guarantee efficient adjustment of AGC and exchange Tsinghua Univ. Press, 1996, pp. 144–165.
[10] X. Wang, W. Fang, and Z. Du, Modern Power System Analysis. Beijing,
power among subareas. China: Tsinghua Univ. Press, 2003.
[11] X. Ai, S. Liao, and J. Wen, “An AGC control performance index for
main tie-lines in interconnected power grids,” Autom. Elect. Power Syst.,
VI. CONCLUSION vol. 37, no. 21, pp. 111–117, 2013 (in Chinese).
[12] Y. Yao and N. Hang, “Power flow studies of multi-swing bus based on
The traditional calculation of power flow neglects AGC op- tie-line power equivalence constrains,” Relay, vol. 34, no. 1, pp. 31–33,
eration for multi-area interconnected power systems. Thus, this 2006 (in Chinese).

Authorized licensed use limited to: VEL TECH MULTI TECH Dr RR Dr SR ENGG. COLLEGE. Downloaded on July 28,2023 at 03:54:11 UTC from IEEE Xplore. Restrictions apply.
 5208 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 53, NO. 6, NOVEMBER/DECEMBER 2017

[13] Y. Yu and Y. Wang, “Asynchronous iterative algorithms of the load flow Zhihao Yun received the Ph.D. degree in electrical
equation in distributed energy management systems (DEMS),” J. Syst. engineering from the School of Electrical Engineer-
Eng., vol. 10, no. 4, pp. 31–38, 1995 (in Chinese). ing, Shandong University, Jinan, China, in 2008.
[14] P. Huang, C. Shen, and Y. Chen, “An asynchronous iterative distributed He is currently an Associate Professor with
algorithm for unified power flow calculation of vertically integrated power Shandong University. His research interests include
systems,” Power Syst. Technol., vol. 32, no. 7, pp. 19–24, 2008 (in Chi- wind power system operation and control.
nese).
[15] T. Gonen, Modern Power System Analysis. Boca Raton, FL, USA: CRC
Press, 2013.
[16] Z. Haibo, Z. Boming, S. Hongbin, and A. Ran, “A new distributed power
flow algorithm between multi-control-centers based on asynchronous it-
eration,” in Proc. Int. Conf. Power Syst. Technol., 2006, pp. 1–7.
[17] P. Ding et al., “Algorithm of power flow for bulk interconnected grid with
transmission interface power constraints,” Proc. CSEE, vol. 30, no. 10,
pp. 8–15, 2010 (in Chinese). Yiming Wang received the B.S. degree in electri-
[18] W. R. Barcelo and W. W. Lemmon, “Standardized sensitivity coefficients cal engineering from the School of Electrical Engi-
for power system networks,” IEEE Trans. Power Syst., vol. 3, no. 4, neering, Fuzhou University, Fuzhou, China, in 2016,
pp. 1591–1599, Nov. 1988. where he is currently working toward the M.S. degree
[19] Q. Han et al., “A practical boundary expression of static voltage stability in electrical engineering.
region in injection space of power systems,” Proc. CSEE, vol. 25, no. 5, His research interests include power system anal-
pp. 8–14, 2005 (in Chinese). ysis and power system operation.
[20] H. Zhong et al., “Preventive control for static voltage stability based on
generation power distribution factor,” Proc. CSEE, vol. 31, no. 34, pp. 8–
14, 2011 (in Chinese).
[21] S. Li and H. Chiang, “Continuation power flow with nonlinear power in-
jection variations: A piecewise linear approximation,” IEEE Trans. Power
Syst., vol. 23, no. 4, pp. 1637–1643, Nov. 2008.
[22] X. Dong, Y. Ding, P. Zhao, C. Wang, Y. Wang, and H. Sun, “Power flow
calculation considering power exchange control for multi-area intercon-
nection power networks,” in Proc. IEEE Ind. Appl. Soc. Meeting, 2016,
pp. 1–7. Doi: 10.1109/IAS.2016.7731971. Penghui Zhao received the B.S. degree in electri-
[23] G. Turan, Modern power system analysis / Second edition[M]. Boca Raton, cal engineering from the School of Electrical Engi-
FL, USA: CRC Press, 2013. neering, Shandong University, Jinan, China, in 2016,
where he is currently working toward the M.S. degree
in electrical engineering.
His research interests include power system sta-
Xiaoming Dong (M’10) received the Ph.D. degree in bility and power system operation.
electrical engineering from the School of Electrical
Engineering, Shandong University, Jinan, China, in
2013.
He is currently a Research Associate Professor
with Shandong University. His research interests in-
clude power system stability, power system control,
and power system operation.

Yuanyuan Ding received the B.S. degree in electri-

cal engineering from the School of Electrical Engi-
neering, Shandong University, Jinan, China, in 2016,
Hua Sun received the M.S. degree in electrical en- where she is currently working toward the M.S. de-
gineering from the School of Electrical Engineering, gree in electrical engineering.
Shandong University, Jinan, China, in 2005. Her research interests include power system sta-
She is currently a Lecturer with the Department of bility and power system operation.
Electrical Automation, Shandong Labour Vocational
and Technology College, Jinan. Her research interests
include power system stability and power system op-
eration.

Chengfu Wang (M’14) received the Ph.D. degree in Yong Wang received the Ph.D. degree in electrical
electrical engineering from the School of Electrical engineering from the School of Electrical Engineer-
Engineering, Shandong University, Jinan, China, in ing, Shandong University, Jinan, China, in 2012.
2012. He is currently an Associate Professor with
He is currently a Research Associate with Shandong University. His research interests include
Shandong University. His research interests include power system reliability, power system operation, and
wind power system operation and planning, and in- dispatch.
tegrated energy system planning and operation.

Authorized licensed use limited to: VEL TECH MULTI TECH Dr RR Dr SR ENGG. COLLEGE. Downloaded on July 28,2023 at 03:54:11 UTC from IEEE Xplore. Restrictions apply.