4404 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO.

5, MAY 2019

Fast Solving Method Based on Linearized Equations

of Branch Power Flow for Coordinated
Charging of EVs (EVCC)
Jian Zhang, Mingjian Cui , Senior Member, IEEE, Bing Li , Hualiang Fang, and Yigang He , Member, IEEE

Abstract—Reported researches on smart charging methods have NOMENCLATURE

the disadvantages of low calculation efficiency or have not simulta-
neously taken the three-phase imbalance, voltage, and power flow
constraints into account. It is an important topic to improve the A. Sets:
computational speed to meet the online rolling optimization re-
quirement for EVCC problems. In this paper, the branch power N Buses of distribution system excluding the root
flow equations of balanced and unbalanced distribution system node.
are derived. The linearization methods for the nonlinear terms of ε Line segments of distribution system.
the branch power flow equations are proposed. Two stages linear
programming (LP) is introduced for EVCC to minimize the total
Hk Child nodes of node k.
charging costs of the holders where three-phase imbalance, charg- B. Constants:
ing demand, voltage, and power flow constraints have been taken
into account. Via ignoring the nonlinear terms of the branch power rik Resistance of line segment (i, k) for balanced dis-
flow equations, the first stage LP is formulated to calculate the es- tribution system.
timated branch power and node voltages as the initial points for xik Reactance of line segment (i, k) for balanced dis-
linearizing the nonlinear terms of branch power flow equations.
The second stage LP is formulated to calculate the optimal charg- tribution system.
ing power using the linearized branch power flow equations. Four zik Impedance of line segment (i, k) for balanced dis-
case studies show that the proposed method without the compro- tribution system.
mise of precision is significantly faster than state-of-the-art works r ik Resistance matrix of line segment (i, k) for unbal-
with respect to the computational speed. anced distribution system, a 3 × 3 matrix.
Index Terms—Branch flow, distribution system, coordinated xik Reactance matrix of line segment (i, k) for unbal-
charging, electric vehicles (EVs), linear programming. anced distribution system, a 3 × 3 matrix.
z ik Impedance matrix of line segment (i, k) for unbal-
anced distribution system, a 3 × 3 matrix.
pdk0 Active power of constant power load at node k for
balanced distribution system.
d
qk0 Reactive power of constant power load at node k
for balanced distribution system.
Manuscript received October 5, 2018; revised January 10, 2019 and March
6, 2019; accepted March 7, 2019. Date of publication March 12, 2019; date pdkz0 Active power of constant impedance load at node
of current version May 28, 2019. This work was supported in part by the Na- k for balanced distribution system when voltage
tional Natural Science Foundation of China under Grant 51577046, in part by magnitude is 1.0 p.u.
the State Key Program of National Natural Science Foundation of China under d
Grant 51637004, in part by the National Key Research and Development Plan qkz0 Reactive power of constant impedance load at node
“Important Scientific Instruments and Equipment Development” under Grant k for balanced distribution system when voltage
2016YFF0102200, and in part by Equipment Research Project in advance un- magnitude is 1.0 p.u.
der Grant 41402040301. The review of this paper was coordinated by Prof. S.
Manshadi. (Corresponding author: Mingjian Cui.) pdk0 Active power of constant power load at node k for
J. Zhang and B. Li are with the School of Electrical Engineering and unbalanced distribution system, a 3 × 1 vector.
Automation, Hefei University of Technology, Hefei 230009, China (e-mail:, q dk0 Reactive power of constant power load at node k
[email protected]; [email protected]).
M. Cui is with the Department of Electrical and Computer Engineering, for unbalanced distribution system, a 3 × 1 vector.
Southern Methodist University, Dallas, TX 75275 USA (e-mail:, mingjian. pdkz0 Active power of constant impedance load at node k
[email protected]). for unbalanced distribution system, a 3 × 1 vector.
H. Fang is with the School of Electrical Engineering, Wuhan University,
Wuhan 430072, China (e-mail:,[email protected]). q dkz0 Reactive power of constant impedance load at node
Y. He is with the School of Electrical Engineering and Automation, Hefei k for unbalanced distribution system, a 3 × 1 vec-
University of Technology, Hefei 230009, China, and also with the School tor.
of Electrical Engineering, Wuhan University, Wuhan 430072, China (e-mail:,
[email protected]). K Total number of EVs with three-phase charging
Digital Object Identifier 10.1109/TVT.2019.2904464 mode.

0018-9545 © 2019 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: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: FAST SOLVING METHOD BASED ON LINEARIZED EQUATIONS OF BRANCH POWER FLOW 4405

M Total number of EVs with single-phase charging Qik Sending end reactive power of line segment (i, k)
mode. for balanced distribution system.
t1 Optimization start time. Skd Apparent power of load at node k for balanced dis-
tmax Optimization end time. tribution system.
β Connecting phase for EVs with single-phase charg- Vk Voltage of node k for balanced distribution system.
ing mode. |Vk | Mode of Vk .
PEVk,max Charging power for the k th EV with three-phase pdk Active power of load at node k for balanced distri-
charging mode. bution system.
PEVm,max Charging power for the mth EV with single-phase qkd Reactive power of load at node k for balanced dis-
charging mode. tribution system.
Δt Time interval of optimization. Uk Voltage magnitude square of node k for balanced
η Charging efficiency. distribution system.
Ekini Initial energy of the k th EV with three-phase charg- cvik (P, Q) Square of voltage loss for line segment (i, k) for
ing mode. balanced distribution system.
ini
Em Initial energy of the mth EV with single-phase cpik (P, Q) Square of active power loss for line segment (i, k)
charging mode. for balanced distribution system.
Ekcap Battery capacity of the k th EV with three-phase cqik (P, Q) Square of reactive power loss for line segment (i, k)
charging mode. for balanced distribution system.
cap
Em Battery capacity of the mth EV with single-phase S ik Sending end apparent power of line segment
charging mode. (i, k) for unbalanced distribution system, a 3 × 1
tks Charging start time for the k th EV with three-phase vector.
charging mode. |S ik | Mode of S ik , a 3 × 1 vector.
tke Charging end time for the k th EV with three-phase P ik Sending end active power of line segment (i, k) for
charging mode. unbalanced distribution system, a 3 × 1 vector.
tms Charging start time for the mth EV with single- Qik Sending end reactive power of line segment (i, k)
phase charging mode. for unbalanced distribution system, a 3 × 1 vector.
tme Charging end time for the mth EV with single-phase Vk Voltage of node k for unbalanced distribution sys-
charging mode. tem, a 3 × 1 vector.
Umin Lower limit for voltage magnitude square. |V k | Mode of V k , a 3 × 1 vector.
Umax Upper limit for voltage magnitude square. Uk Square of voltage magnitude of node k for unbal-
max
Pik,α,t Maximum active power of line segment (i, k) for anced distribution system, a 3 × 1 vector.
phase α in time interval t. pdk Active power of load at node k for unbalanced dis-
max
PT,α,t Maximum active power of distribution transformer tribution system, a 3 × 1 vector.
for phase α in time interval t. q dk Reactive power of load at node k for unbalanced
Pik0 Initial active power of line segment (i, k) for lin- distribution system, a 3 × 1 vector.
earization in balanced distribution system. S lik Power losses across line segment (i, k) for unbal-
Qik0 Initial reactive power of line segment (i, k) for lin- anced distribution system, a 3 × 1 vector.
earization in balanced distribution system. cuik (P, Q) Square of voltage loss for line segment (i, k) for
Sik0 Initial apparent power of line segment (i, k) for lin- unbalanced distribution system, a 3 × 1 vector.
earization in balanced distribution system. cpik (P, Q) Square of active power loss for line segment
Vi0 Initial voltage of node i for linearization in balanced (i, k) for unbalanced distribution system, a 3 × 1
distribution system. vector.
P ik0 Initial active power of line segment (i, k) for lin- cqik (P, Q) Square of reactive power loss for line segment (i, k)
earization in unbalanced distribution system, a 3 × for unbalanced distribution system, a 3 × 1 vector.
1 vector. ρ(t) Power price in time interval t.
Qik0 Initial reactive power of line segment (i, k) for PEVk,a,t Charging power of phase a in time interval t for the
linearization in unbalanced distribution system, a k th EV with three-phase charging mode.
3 × 1 vector. PEVk,b,t Charging power of phase b in time interval t for the
A A constant real number 3 × 3 matrix. k th EV with three-phase charging mode.
B A constant real number 3 × 3 matrix. PEVk,c,t Charging power of phase c in time interval t for the
k th EV with three-phase charging mode.
C. Variables: PEVk,t Charging power of the k th EV with three-phase
Sik Sending end apparent power of line segment (i, k) charging mode in time interval t.
for balanced distribution system. PEVm,β,t Charging power of the mth EV with single-phase
|Sik | Mode of Sik . charging mode in time interval t.
Pik Sending end active power of line segment (i, k) for Un,α,t Voltage magnitude square of node n phase α in time
balanced distribution system. interval t.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 4406 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

Pik,α,t Power of the line segment (i, k) for phase α in time In [11], [12], a real-time smart load management strategy is
interval t. proposed for the coordinated charging of EVs by using the sen-
PT,α,t Power of the distribution transformer for phase α sitivity analysis technique. However, the control variables are
in time interval t. the charging locations rather than the charging power of EVs. It
x A 3 × 1 vector. is still challenging to ensure that the EVs can be fully charged.
y A 3 × 1 vector. As the coordinated charging of EVs is a large scale optimization
problem, many techniques are proposed to improve the com-
putational speed. In [13], a linear constrained convex quadratic
I. INTRODUCTION programming is formulated to iteratively correct nodal voltages
ORLDWIDE energy sectors face critical challenges using the power flow calculation. The objective function is to
W with regard to the security of power supply, environ-
mental impacts, and energy costs. Energy investments are trend-
minimize the power losses, while the constraints on voltage
magnitudes and thermal loadings of lines/transformers are ig-
ing towards innovations to improve both the energy efficiency nored. However, if the objective function is sensitive to nodal
and the environmental friendliness. Compared with traditional voltages, such as minimizing the total charging costs, the method
vehicles, electric vehicles (EVs) present more significant ben- developed in [13] cannot be applicable.
efits due to the capability of the non-reliance on oil, reducing In [14], [15], with inequality constraints on nodal voltage and
harmful gas emissions, and lowering fluctuations of renewable thermal loadings of transformers and lines linearized, a LP for
sources. Currently, many countries have accelerated the devel- the coordinated charging of EVs is proposed to maximize the
opment of distributed generators (DGs) and EVs. Consequently, total charging energy. However, the deviation of linearization
some hot research topics come to the fore, including the impact of this method is relatively large. Moreover, this method cannot
of DGs and EVs on the power system [1]–[3], the optimal op- be applicable to the nonlinear objective function, which is not
eration of distribution networks [4], and the active distribution linearly related to the charging power of EVs, such as the mini-
network technology [5]–[7]. However, the uncoordinated charg- mization of total power supply. In [16], based on Cartesian coor-
ing of massive EVs could significantly increase network losses, dinate power flow equations, a mixed integer LP of coordinated
overload distribution transformers or lines, reduce the energy charging of EVs is proposed to maximize the revenue of power
efficiency, and lower system voltages. Whereas smart charging corporations with linearized constraints. However, many aux-
of EVs can significantly improve both economy and reliability iliary variables and constraints are used to linearize inequality
benefits of the distribution system. Generally, researches on the constraints on both the voltage and current, which can signifi-
coordinated charging of EVs can be divided into the distributed cantly increase the complexity of the developed model. Last but
and centralized methods. The distributed coordinated charging not the least, the charging location rather than the charging power
mainly uses the fuzzy mathematics theory [8], sensitivity anal- is optimized. The formulated mixed integer LP is much more dif-
ysis [9], and iterative method [10]. The centralized coordinated ficult to solve than the LP. In [17], a quadratic programming is
charging generally utilizes the sensitivity analysis [11], [12] and proposed to optimize the charging and discharging power of EVs
the optimization techniques [13]–[20]. considering the time-of-use power price and battery degradation
When the objective function is to minimize the total charg- costs. However, the electricity price is proportional to charging
ing costs of holders, distributed EV charging scheduling cannot power and other conventional load. In [18], a coordination strat-
be applied, because the voltage magnitude and branch power egy for optimal charging of EVs is developed by considering the
constraints cannot be taken into account. For example, the elec- congestion of the distribution system. In [19], a quadratic pro-
tricity price is low in peak wind or solar power time and high gramming is formulated to minimize the power losses with load
in peak load time. If a large number of EVs are scattered in dif- balancing. In [20], load factor, load variance, and network losses
ferent nodes of distribution network, such as EVs in residential are demonstrated to be equivalent under certain conditions. As
distribution network, it is difficult to take voltage magnitude and an outcome, minimizing network losses can be transformed to
branch power constraints into account if distributed charging is minimizing the load factor or load variance, which can reduce
used to tracking the low electricity price. As a result, safe and the computational complexity. However, the constraints on nodal
economic operation of distribution network cannot be guaran- voltages or thermal loadings of transformers and lines are not
teed. Therefore, centralized coordinated charging is preferable. considered in the aforementioned models. When there are mas-
However, centralized coordinated charging is a large-scale non- sive EVs connected to the distribution system, the constraints
linear optimization problem. It is very difficult to solve because on nodal voltages and/or thermal loadings of transformers and
of high dimension of optimization variables and large number of lines can be really a factor that limits the charging power of EVs.
constraints. With the popularization of EVs and the progress of Though neglecting the constraints on nodal voltages and/or ther-
battery technology, a large number of EVs will adopt fast charg- mal loadings of transformers and lines may significantly improve
ing mode. As a result, optimization time interval must be greatly the computational speed, it may also make the solution to the
reduced, and the dimension of optimization variables, number charging power of EVs unfeasible. When the objective function
of constraints will increase dramatically. How to improve the is to minimize the total charging costs, the constraints on volt-
computational speed to meet the online rolling optimization re- ages magnitudes and/or thermal loadings of transformers/lines
quirement is an important topic worthy of study. That is, the can be a factor that limits the charging power of EVs. As a result,
computational time is very important in this problem. the aforementioned four methods cannot be applicable.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: FAST SOLVING METHOD BASED ON LINEARIZED EQUATIONS OF BRANCH POWER FLOW 4407

In [21], the influence of charging on the heating and life span

of distribution transformer is analyzed. A non-linear model is
constructed. However, none of the references [1]–[20] have con-
structed a non-linear model for transformer heating. Instead, a
simple linear inequality with simple branch power or current is
formulated. In [22], stochastic analysis is used to analyze the im-
pact of charging randomness on distribution network. At present,
for EVCC problem, the rolling optimization is generally used to
take into account the uncertainty of EVs and load forecasting. Fig. 1. Branch flow of distribution systems.
To cope with the inefficiency of calculation, this paper de-
rives the branch flow equations of balanced and unbalanced dis- III. NLP MODEL FOR THE EVCC PROBLEM
tribution system. Moreover, the three-phase imbalance, voltage
A. Branch Flow Equations in Balanced Distribution Systems
constraints, and power flow constraints are also considered. The
coordinated charging model of EVs is established to minimize The distribution system with symmetrical parameters is re-
total charging costs of holders. The contributions of this paper ferred as balanced distribution system while that with three-
are as follows. 1) We propose a method to linearize the non- phase conductors not transposed or with large load differences
linear terms in branch power flow equations of balanced and among three-phase is referred as unbalanced distribution system.
unbalanced distribution system and apply it to solve the EVCC Balanced distribution system can be represented by single-phase
problem. As a result, the computing time is greatly reduced. model while unbalanced distribution system must be represented
2) The conventional load in branch power flow equations in- by a three-phase model.
cludes constant impedance and constant power load. 3) We have To improve the computational speed, models of the three-
proposed how to compute the initial point of linearization. 4) phase balance and unbalanced distribution systems are con-
The capabilities of the proposed method-fast calculation speed structed by using branch flow equations. Given an N + 1 bus
and high accuracy are verified by four simulation cases and com- distribution system with a tree topology, i.e., a radial net-
pared with recent similar work. work without loops between branches, the root bus is denoted
The organization of this paper is as follows. Branch flow equa- by N + 1 and the remaining N buses of the system are de-
tions of balanced and unbalanced distribution systems are intro- noted by the set N = {1, 2, . . . , N }. The edge-set that rep-
duced in Section II. The coordinated charging model of EVs is resents the set of distribution line segments (including con-
formulated in Section III-C. A fast solving method is described ductors for single-, two-, and three-phase circuits) is denoted
in Section V. Both the accuracy and computational efficiency of by ε ⊆ {N ∪ {N + 1}} × {N ∪ {N + 1}} with (i, k) ∈ ε, if
the developed method are discussed in Section VI. Section VII there is a distribution line segment between bus i and bus k (bus
concludes the paper. i closer to the feeder). Note that all the edges are directed so
that we can get (i, k) ∈ ε ⇒ (k, i) ∈ / ε. The Π type equivalent
circuit is used to represent the line segments. The impedance of
II. EVCC PROBLEM the line segment (i, k) is given by zik = rik + jxik . The demand
The EVCC problem is to determine an EV battery charging at bus k considering the shunt capacitance of the line is denoted
schedule so that distribution system operates with optimal cost by sdk = pdk + jqkd . Considering the circuit shown in Fig. 1, both
and satisfies operational constraints. In this paper, the following the voltage drop and power flow equations are formulated by
descriptions are assumed [16]. using the notation and orientation.
1) The EV batteries must be charged in a given period of Let Hk = {j|(k, j) ∈ ε} be the set of buses downstream of
time, which is divided into several time intervals. bus k. The total power flow Sik ∈ C transferred through the
2) The energy required by each battery is known at the be- sending end of distribution line segment (i, k) is given by:
ginning of the time period.  |Sik |2
3) The EVs have communication devices that allow the dis- Sik = Skj + Skd +zik (1)
tribution system operator to control the charging power of |Vi |2
j∈Hk
the batteries. That control can be carried out in each time
where the line power flow Sik is always relative to the sending
interval of the time period.
end voltage, Vi = |Vi |∠θi , of the distribution line segment. The
Operational constraints, such as voltage magnitude limits,
voltage of bus k based on the upstream bus i is given by:
power generation limits, and maximum circuit power must be
satisfied. The optimal charging schedule defines the charging Pik − jQik
Vk = Vi − zik (2)
power of each EV battery in each time interval. The estimated Vi∗
arrival and departure times for the EVs are considered using pa-
where the root node voltage is fixed with VN +1 = V s = |V s |∠0
rameters tks and tke , respectively. These parameters, as well as
for a constant |V s |. The dependence on the phase angles in (2)
the initial charge state of a battery (Ekini ), can be obtained using
is removed by taking the product of each side of (2) with its
estimation techniques applied to EVs, such as those in [23]–[25].
conjugate. Thus, the branch flow equations for an N + 1 bus
The mathematical model considers these parameters, allowing
network is given by:
the EV to be charged only during the time interval between ar-
rival and departure. VN +1 = V s = |V s |∠0 (3a)
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 4408 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

|Sik |2 of (6) with its conjugate vector, the voltage equation in (5b) can
|Vk |2 = |Vi |2 − 2 (rik Pik + xik Qik ) + |zik |2 2
(3b) be updated as:
|Vi |
 |Sik |2 |V k |2 = |V i |2 − 2 (r̃ ik Pik + x̃ik Qik )
Pik = Pkj + pdk + rik (3c)
|Vi |2 + cuik (P , Q) , k ∈ N (7)
j∈Hk
 T
 |Sik |2 a = 1 e−j2π/3 ej2π/3 (8)
Qik = Qkj + qkd + xik (3d)
|Vi | 2    
j∈Hk r̃ ik = Re aaH r ik + Im aaH xik (9)
The power demand is usually a mix of constant power and    
x̃ik = Re aaH xik − Im aaH r ik (10)
constant impedance load. Thus, it can be expressed as:
cuik (P , Q) = [z ik (S ∗ik ∅V ∗i )] [z ∗ik (S ik ∅V i )] (11)
pdk = pdk0 + pdkz0 |Vk |2 (4a)
where the symbol of denotes the element-wise multiplication.
qkd = qk0
d
+ qkz0
d
|Vk |2 (4b)
We assume that the three-phase voltages magnitudes of each
According to (3)–(4) and given Ui = |Vi |2 , the branch flow node are equal in order to obtain a constant equivalent resistance
equations of the three-phase balanced distribution system can matrix r̃ ik and reactance matrix x̃ik , thus simplifying the volt-
be simplified as: age equation. This hypothesis is only used to derive Eq. (7) and
is not used for other purposes. Eq. (7) shows that when the three-
UN +1 = |V s |2 = U s (5a)
phase power of each branch is unequal, the three-phase voltages
Uk = Ui − 2(rik Pik + xik Qik )+ cvik (P, Q), k ∈ N (5b) magnitudes of each node are unequal as well. Thus, Eq. (7) sim-
 ulates the three-phase unbalanced distribution network. Eq. (7)
Pik = Pkj + pdk + cpik (P, Q), k ∈ N (5c) has high accuracy, because the unbalance of three-phase voltage
j∈Hk
of each node in the actual distribution network is very small. The

Qik = Qkj + qkd + cqik (P, Q), k ∈ N (5d) voltage imbalance limit in the distribution system is that the neg-
j∈Hk ative sequence voltage divided by the positive sequence voltage
must be below 2% which is required by the National Electri-
pdk = pdk0 + pdkz0 Uk , k ∈ N (5e) cal Manufacturers Association (NEMA). In the actual distribu-
qkd = d
qk0 + d
qkz0 Uk , k∈N (5f) tion network, the imbalance of three-phase voltage is very small
while the imbalance of three-phase power may be large.
where cvik (P, Q) = |zik |2 |Sik | /|Vi | , 2 2
cpik (P, Q) = rik |Sik |2 Branch flow equations in (5c) and (5d) can be updated as:
/|Vi |2 , and cqik (P, Q) = xik |Sik |2 /|Vi |2 . 
P ik = P kj + pdk + cpik (P , Q), k ∈ N (12)
B. Branch Flow in Unbalanced Distribution Systems j∈Hk

In the actual distribution system, the overhead lines are usually Qik = Qkj + q dk + cqik (P , Q), k ∈ N (13)
not transposed. Thus, the off diagonal elements of the line mutual j∈Hk
impedance matrix are not equal any more. Moreover, the three-
phase loads connected to each node are usually not equal. As where cpik (P , Q) = Re{(S ik ∅V i ) (V i − V k )}, cqik (P , Q)
a result, the three-phase parameters of the distribution system = Im{(S ik ∅V i ) (V i − V k )}
are asymmetrical. For each line segment (i, k) ∈ ε, the voltage Power demand equations in (5e) and (5f) can be updated as:
equation is given by: pdk = pdk0 + pdkz0 U k , k ∈ N (14)
∗
V k = V i − z ik [(P ik − jQik ) ∅V i] (6) q dk = q dk0 + q dkz0 U k , k∈N (15)
where z ik ∈ C , V k = [Vka , Vkb , Vkc ] , V i = [Via , Vib ,
3×3 T
The three-phase voltage of the root node in (5a) is updated as:
Vic ]T , P ik = [Pika , Pikb , Pikc ]T , and Qik = [Qik a , Qik b ,
Qik c ]T . The symbol of ∅ denotes the element-wise division. U N +1 = |V s | |V s | (16)
Unlike the per-phase equivalent case, both sides of (6) can-
not remove the dependence on phase angles by multiplying the C. Model for EVCC
complex conjugate. This is due to the fact that there is a cou-
The objective function for the coordinated charging model of
pling between phases that arises from the cross-product of the
EVs is to minimize the total charging costs of holders, given by:
three-phase equation for the phase voltage and line current. To
t
K M

address this problem, it can be observed that voltage magnitudes max  
between phases are similar, i.e., |Via | ≈ |Vib | ≈ |Vic | [26] and J = min ρ(t) PEVk,t + PEVm,β,t Δt (17)
the unbalance on each phase are not that severe. Thus, voltage t=t1 k=1 m=1
magnitudes are assumed to be approximately equal. The unbal- Constraints on the charging power of each EV with the three-
ance of the three-phase angle α is relatively small (typically phase charging mode are formulated by:
within 1◦ ∼ 3◦ ). Thus, we ignore α and assume that the three-
phase voltage at each node is equal. By multiplying both sides 0 ≤ PEVk,t ≤ PEVk,max (18)
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: FAST SOLVING METHOD BASED ON LINEARIZED EQUATIONS OF BRANCH POWER FLOW 4409

PEVk,t The linearizing process is closely associated with the charging

PEVk,a,t = PEVk,b,t = PEVk,c,t = (19)
3 scheduling model.
The constraint on the charging power of each EV with the The advantage of LP is that it can be solved quickly by using
single-phase charging mode is formulated by: sophisticated solver. A mixed-integer LP is formulated in [16].
However, polar coordinate power flow equations are adopted.
0 ≤ PEVm,β,t ≤ PEVm,max (20) While in our paper, branch power flow equations are utilized.
Constraints on the power demand of each EV with the three- The variable number is much less than that in [16]. Further, non-
and single-phase charging modes are given by: linear inequalities are linearized by introducing new auxiliary
variables in [16]. As a result, variables have sharply increased.
tke
 While in our paper, we linearize the non-linear terms in branch
η PEVk,t Δt = Ekcap − Ekini (21) power flow equations by using Taylor expansion and the variable
t=tks
number keeps constant.
tme

η PEVm,β,t Δt = Em
cap
− Em
ini
(22) A. Linearization of Branch Flow Equations in the Balanced
t=tms
Distribution System
The rates of charging power constraints are considered in In (5a)–(5f), only cvik (P, Q), cpik (P, Q), and cqik (P, Q) are
Eqs. (18)–(20), while the SOC constraints are indirectly con- nonlinear terms. To fast solve the developed model, it should be
sidered in Eqs. (21) and (22). The discharging mode (Vehicle to linearized. Let:
Grid) can be added to this work as well. To do this, it is only
needed to modify Eqs. (18)–(22). However, since the discharg- |Sik |2 2
Pik + Q2ik
hik (P, Q) = = (26)
ing mode can reduce the battery life span, it is not considered in |Vi |2 |Vi |2
this paper.
The constraint on the nodal voltage of the distribution system cvik (P, Q) = |zik |2 hik (P, Q) (27a)
is given by: cpik (P, Q) = rik hik (P, Q) (27b)
Umin ≤ Un,α,t ≤ Umax (23) cqik (P, Q) = xik hik (P, Q) (27c)
The constraint on the thermal loading of each line is given by: hik (P, Q) can be linearized as:
max
0 ≤ Pik,α,t ≤ Pik,α,t (24)
2Pik0 Pik + 2Qik0 Qik − |Sik0 |2
hik (P, Q) ≈ (28)
The constraint on the thermal loading of each transformer is |Vi0 |2
given by:
max B. Linearization of Branch Flow Equations in the Unbalanced
0 ≤ PT,α,t ≤ PT,α,t (25)
Distribution System
For the balanced distribution system, the objective function is
formulated in (17). Equality constraints are formulated in (5a)– 1) Linearization of the Nonlinear Term in Voltage Equations:
(5f), (19), and (21). Inequality constraints are formulated in (18) Let āi = [1, ej2π/3 e−j2π/3 ]T ∅|V i |, then the nonlinear term of
and (23)–(25). For the unbalanced distribution system, the ob- the voltage equation can be expressed as:
jective function is formulated in (17). Equality constraints are cuik (P , Q) = [z ik (S ∗ik ∅V ∗i )] [z ∗ik (S ik ∅V i )]
formulated in (7), (12)–(16), (19), and (21)–(22). Inequality con- (29)
straints are formulated in (18), (20) and (23)–(25). ≈ [z ik (S ∗ik ā∗i )] [z ∗ik (S ik āi )]
The charging power of EVs is constrained by the voltage A new branch impedance matrix is defined as:
magnitude and branch power. For different distribution network
models, the function relationship between the voltage magni- z̄ ik = z ik diag (ā∗i ) = r̄ ik + jx̄ik (30)
tude, branch power, and charging power is different, which leads
to different objective function values, i.e., total charging costs of Then, (29) can be rewritten as:
holders. If the distribution network is balanced and all EVs are
cuik (P , Q) = (r̄ ik P ik ) (r̄ ik P ik ) + (x̄ik Qik ) (x̄ik Qik )
charged with the three–phase mode, the balanced distribution
network model can be used. Otherwise, the unbalanced distri- + (x̄ik P ik ) (x̄ik P ik ) + (r̄ ik Qik ) (r̄ ik Qik )
bution network model must be adopted.
+ 2(r̄ ik P ik ) (x̄ik Qik ) − 2(x̄ik P ik )
IV. LP MODEL FOR THE EVCC PROBLEM (r̄ ik Qik ) (31)
In the formulated NLP EVCC problem, only the model of dis- The linearization of cuik (P , Q) is defined as:
tribution network is nonlinear, while other parts are linear. We
propose a method to linearize the model of balanced and unbal- cuik (P , Q) ≈ upik (P ik0 , Qik0 ) (P ik − P ik0 )
anced distribution network. However, for the charge scheduling + uqik (P ik0 , Qik0 ) (Qik − Qik0 ) (32)
problem, the initial point of linearizing is unknown in advance.
We propose a method to calculate the initial point of linearizing. + cuik (P ik0 , Qik0 )
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 4410 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

According to (31), the partial derivatives are given by: Partial derivative terms in (44) and (45) are respectively given
by:
∂cuik
upik = = hxx (r̄ ik , P ik ) + hxx (x̄ik , P ik ) ∂cpik
∂P ik f pik = = h̃xx (r̂ ik , P ik ) + h̃xy (x̂ik , P ik , Qik )
(33) ∂P ik (46)
+ 2hxy (r̄ ik , x̄ik , P ik , Qik )
− h̃yx (x̂ik , Qik , P ik )
− 2hxy (x̄ik , r̄ ik , P ik , Qik )
∂cpik
∂cuik f qik = = h̃xx (r̂ ik , Qik ) − h̃xy (x̂ik , Qik , P ik )
uqik = = hxx (r̄ ik , Qik ) + hxx (x̄ik , Qik ) ∂Qik (47)
∂Qik
+ h̃xy (x̂ik , P ik , Qik )
(34)
+ 2hxy (x̄ik , r̄ ik , Qik , P ik )
∂cqik
g pik = = h̃xx (x̂ik , P ik ) − h̃xy (r̂ ik , P ik , Qik )
− 2hxy (r̄ ik , x̄ik , Qik , P ik ) ∂P ik (48)
where functions hxx and hxy are respectively given by: + h̃yx (r̂ ik , Qik , P ik )
∂cqik
hxx (A, x) = 2diag (Ax) A (35) g qik = = h̃xx (x̂ik , Qik ) + h̃xy (r̂ ik , Qik , P ik )
∂Qik (49)
hxy (A, B, x, y) = diag (By) A (36) − h̃yx (r̂ ik , P ik , Qik )
2) Linearization of the Nonlinear Term in Power Equations: where functions h̃xx , h̃xy , and h̃yx are respectively given by:
The power loss across line segment (i, k) is given by:
h̃xx (A, x) = diag (Ax) + diag (x) A (50)
S lik = (S ik ∅V i ) [z ik (S ∗ik ∅V ∗i )] (37)
h̃xy (A, x, y) = diag (Ay) (51)
Another branch impedance matrix is redefined as: h̃yx (A, y, x) = diag (y) A (52)

ẑ ik = r̂ ik + jx̂ik = z ik āi āH

i (38) V. FAST SOLVING METHOD
According to (38), by separating the real part from the imag- The accuracy of linearization is closely associated with the
inary one, it can be obtained: initial point. However, it is challenging to know the initial point
    since it is not the actual operation point. Toward this end, the non-
r̂ ik = Re āi āH
i r ik − Im āi āH
i xik (39) linear terms of the voltage cuik (P , Q), active power cpik (P , Q),
    and reactive power cqik (P , Q) are first ignored. The simplified
x̂ik = Re āi āHi xik + Im āi āHi r ik (40)
LP for coordinated charging of EVs is formulated. Then, the
For the sake of simplicity, (37) can be rewritten as: sophisticated LP solver is used to solve this simplified LP. The
output is taken as the initial point of linearization. Whereafter,
S lik = (P ik + jQik ) [ẑ ik (P ik − jQik )] (41) the approximated LP for coordinated charging of EVs is formu-
lated. Finally, the sophisticated LP solver is used to solve this
According to (39)–(41), by separating the active power from approximated LP once again and output the optimal charging
the reactive power, it can be obtained: power of EVs. Losses of the voltage, active power, and reac-
tive power are much less than the corresponding linear terms
cpik (P , Q) = real S lik = P ik (r̂ ik P ik + x̂ik Qik )
in branch flow equations. In addition, the linearization is per-
+ Qik (r̂ ik Qik − x̂ik P ik ) (42) formed at the initial point that is the result of the simplified
linear model. Thus, the deviation is relatively small. That is,
cqik (P , Q) = imag S lik = P ik (x̂ik P ik − r̂ ik Qik ) the accuracy of the proposed linearization strategy can be guar-
+ Qik (r̂ ik P ik + x̂ik Qik ) (43) anteed. The computational speed of the proposed method can
also be guaranteed since both formulated models belong to the
The linearization of cpik (P , Q) and cqik (P , Q) are respec- LP problem. A schematic diagram of the proposed fast solving
tively defined as: method is shown in Fig. 2.
cpik (P , Q) ≈ f pik (P ik0 , Qik0 ) (P ik − P ik0 ) VI. CASE STUDIES
+ f qik (P ik0 , Qik0 ) (Qik − Qik0 ) (44) In the simulation cases, noon or night are chosen as the charg-
+ cpik (P ik0 , Qik0 ) ing periods since these time periods are in coincidence with the
charging habits of most EV holders.
cqik (P , Q) ≈ g pik (P ik0 , Qik0 ) (P ik − P ik0 )
A. Case 1
+ g qik (P ik0 , Qik0 ) (Qik − Qik0 ) (45)
1) Simulation Conditions: Fig. 3 shows the IEEE 33-node
+ cqik (P ik0 , Qik0 ) medium voltage (MV) distribution system to test the capability

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: FAST SOLVING METHOD BASED ON LINEARIZED EQUATIONS OF BRANCH POWER FLOW 4411

TABLE I
LINE IMPEDANCE OF IEEE 33-NODE DISTRIBUTION NETWORK

Fig. 2. Schematic of the proposed fast solving method.

Fig. 3. IEEE 33-node distribution system.

of the proposed method. The impedance matrix of transmission

lines is shown in Table I. In this system, node 33 is taken as
the slack node and its voltage is kept to be 1.00 p.u.. The rest
of nodes are taken as PQ nodes. The base conventional load at
each node is shown in Table II. The constant power load model is
deployed. There are four parking lots of EVs connected at nodes
17, 21, 24, and 32, respectively. There are 40 EVs in each parking
lot. The single-phase base√ power and voltage are chosen to be
1000/3 kVA and 12.66/ 3 kV, respectively. The back forward
sweep method is used to calculate the power flow.
Other simulation conditions are set as follows:
1) All EV owners are willing to participate in the coordi-
nated charging. The charging power of each EV is fully
controllable. The charging time period is between 12:00 ∼
14:00.
2) The conventional load at each node is equal to the base
load between 12:00 ∼ 13:00 and that of 80% base load
between 13:00 ∼ 14:00. The power factor is 0.95.
3) The power prices in the time range of 12:00 ∼ 13:00 and
13:00 ∼ 14:00 are 0.8 and 0.4 Yuan/kWh, respectively.
4) All EVs adopt the three-phase charging mode. 9) The upper and lower voltage limits are 1.0 and 0.9 p.u.,
5) The charging demand of each EV is 10 kWh. respectively.
6) Minimal and maximal charging power of each EV are 0 2) Simulation Results: All the programs are written with
and 10 kW, respectively. MATLAB. The LP is solved by using the library function lin-
7) The charging efficiency is set to be 1.0. prog. The precise model is solved by the primal dual interior
8) The optimization time interval is 1 hour. point method [27]. The CPU of the computer is Intel (R) Core

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 4412 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

TABLE II TABLE IV
LOAD OF IEEE 33-NODE DISTRIBUTION SYSTEM OPTIMAL VOLTAGES FOR DIFFERENT METHODS IN CASE 1

As can be seen in Table III and Table IV, the results obtained
TABLE III by the simplified and approximate LP are relatively close. Thus,
OPTIMAL CHARGING POWER FOR DIFFERENT METHODS IN CASE 1 the result is of high precision via the initial point obtained by
the simplified LP for linearizing the nonlinear terms of branch
flow equations. Moreover, the optimization results obtained by
the approximate LP are very close to those of the primal dual
interior point method. This observation demonstrates that the
proposed method has a high precision. However, the computa-
tional speed of the proposed method is about 40 times higher
than that of the primal dual interior point method. As can be
seen from the results of power flow calculation in Table IV, the
optimal charging power obtained by the simplified LP may re-
sult in voltages dropping out of the lower limit (see the bold
(TM) i3-4510. The main frequency of the CPU is 3.5 GHz with
font). However, for the approximate LP, voltage results of the
32G RAM.
power flow calculation are very close to those of the optimiza-
Some results are shown in Table III and Table IV, where
tion results. Furthermore, all of the voltages are within the rated
f.0 and f.1 represent the optimization results during 12:00 ∼
range.
13:00 and 13:00 ∼ 14:00 solved by the simplified LP, respec-
tively. PM.0 and PM.1 represent the optimization results during
B. Case 2
12:00 ∼ 13:00 and 13:00 ∼ 14:00 solved by the approximate LP,
respectively. PD.0 and PD.1 represent the optimization results 1) Simulation Conditions: In this case, all the EVs are as-
during 12:00 ∼ 13:00 and 13:00 ∼ 14:00 solved by the precise sumed to adopt the single-phase charging mode to testify the
nonlinear model, respectively. Pf.0 and Pf.1 represent voltages capabilities of the proposed method. The simulation platform is
of power flow calculations during 12:00 ∼ 13:00 and 13:00 ∼ the same as that of case 2 in [28]. Time-of-use electricity prices
14:00, by substituting the optimal charging power of EVs using are set as 0.8, 0.4 Yuan/kWh during 12:00 ∼ 13:00 and 13:00 ∼
the simplified LP into the precise power flow equations, respec- 14:00, respectively. The conventional household loads at each
tively. PF.0 and PF.1 represent voltages of power flow calcula- node of each phase are 4.5 and 3.6 kW during 12:00 ∼ 13:00
tions during 12:00 ∼ 13:00 and 13:00 ∼ 14:00, by substituting and 13:00 ∼ 14:00, respectively. The charging demand of each
the optimal charging power of EVs using the approximate LP EV is 5 kWh. The maximum charging power is 4 kW. Other
into the precise power flow equations, respectively. simulation conditions are set as same as those of case 2 in [28].

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: FAST SOLVING METHOD BASED ON LINEARIZED EQUATIONS OF BRANCH POWER FLOW 4413

TABLE V TABLE VII

OPTIMAL CHARGING POWER FOR DIFFERENT METHODS IN CASE 2 OPTIMAL CHARGING POWER FOR DIFFERENT METHODS WITH SINGLE- AND
THREE-PHASE CHARGING COEXISTING

TABLE VI
OPTIMAL VOLTAGE FOR DIFFERENT METHODS IN CASE 2

Fig. 4. Computational efficiency with different methods.

power rather than the single-phase charging power for EV with

the three-phase charging mode as the optimization variable. The
charging power of each phase is one-third of this variable. That
is, with some simple mathematical techniques, the number of
optimization variables, equality and inequality constraints can
2) Simulation Results: All the programs are written with be the same as those in case 2. The imbalance of the distri-
MATLAB. The simplified and approximate LP is solved using bution system is reduced compared with that of case 2 due to
library function linprog. The precise model is solved by the the existence of three-phase charging mode. Thus, the numeri-
primal dual interior point method [27]. The computer configura- cal stability of the program is improved and the computational
tion is the same as that in case 1. Simulation results of different speed is slightly higher than that of case 2. The voltage results
optimization algorithms are shown in Table V and Table VI. using different optimization algorithms are similar with those
It can be seen that results of the proposed method are in good in Table VI. Based on simulation results, we can draw the same
agreement with those of the primal dual interior point method. conclusions as case 2.
However, the computational speed is significantly superior to The computational time of aforementioned three cases with
the primal dual interior point method. different optimization methods are shown in Fig. 4. As can be
seen, the calculation efficiency of the proposed method is signif-
C. Case 3 icantly better than that of the primal dual interior point method
and slightly worse than that of the simplified LP. However, the
1) Simulation Conditions: In this case, the coexistence of precision is very close to the primal dual interior point method.
the single- and three-phase charging modes for EVs in the dis-
tribution system to test the capabilities of the proposed method.
Simulation conditions are the same as those in case 2 except D. Case 4
that EVs connected to nodes 3, 4, 5, 8, 9, and 10 adopt the three- 1) Simulation Conditions: A 354-node distribution system is
phase charging mode and the maximum charging power of each used to test the capabilities of the proposed method. As shown
EV is 12 kW. in Fig. 5, the MV distribution system is a 34-node distribution
2) Simulation Results: The optimal charging power of EVs at system and the rated phase to ground voltage is 13.8 kV. There
different nodes with different algorithms and the computational are 16 low voltage (LV) distribution systems in the simulation
time of the program are shown in Table VII. As can be seen, platform. The topology of each one is shown in Fig. 6. The rated
the computational efficiency of the proposed method is slightly phase-to-ground voltage is 220 V. In Fig. 6, the capacity of the
better than that of case 2. This is because we choose the total transformer is 250 kVA and the terminal node ‘xx’ is connected

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 4414 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

Fig. 5. MV distribution system.

Fig. 7. Total load at different time periods.

charging load obtained by the simplified and approximate LP

are denoted as blue, orange, and yellow boxes, respectively, as
Fig. 6. LV distribution system.
shown in Fig. 7. As can be seen, all of EVs are only charged
during 4:00 ∼ 6:00, when the electricity prices are relatively
low. However, the total charging power difference between the
to the MV distribution system at nodes 11, 13, 14, 17, 18, 20, 22, simplified and approximate LP is significant. Moreover, since
24–26, and 28–33, respectively. The rated currents of the cables the lowest electricity price and conventional load level occur
in the MV and LV distribution networks are 200 A and 368 A, during 5:00 ∼ 6:00, the total charging load is the maximal during
respectively. There is a conventional household load connected 5:00 ∼ 6:00. The electricity price during 4:00 ∼ 5:00 is higher
to each node in the LV distribution system and each household than that during 6:00 ∼ 7:00. The total charging load during
has one EV connected. Node 34 is taken as the slack node and 4:00 ∼ 5:00 is much lower than that during 6:00 ∼ 7:00. This is
its voltage is kept to be 1.1 p.u.. The rest of nodes are taken as because most of the charging loads are supplied during 5:00 ∼
PQ nodes. The single-phase base power and voltages are chosen 7:00 by the optimization programming.
to be 1000 kVA, 13.8 kV and 0.22 kV, respectively. The back Let the cable connected with the root node of the distribution
forward sweep method is used to calculate the power flow. system and that connected with the LV side of the transformer
Other simulation conditions are set as follows: be called the MV and LV main cables, respectively. The load-
1) All EV owners are willing to participate in the coordinated ing of the MV and LV main cables and transformer for phase
charging and charging power of each EV is fully control- A are shown in Fig. 8. Clearly, the loadings of transformer and
lable. The charging time period is between 18:00 ∼ 8:00. cables are low when the electricity price is high. Because EVs
2) The conventional house hold load at each node is set to are not charged when the electricity price is high. Otherwise, the
be the same. loadings of transformer and cables are high when the electricity
3) At 18:00 ∼ 19:00, the conventional household load price is low. Because EVs are charged with high power when the
connected to phases A, B, C are 0.8666, 0.8000, and electricity price is low. During 5:00 ∼ 6:00, since the electricity
0.7334 kW, respectively. Power factor is set to be 0.95. price is minimal, the total charging power of EV is maximal. As a
The conventional household load model is set to be 60% result, the loadings of transformer and main cable are the largest
constant power load plus 40% constant impedance load. during 5:00 ∼ 6:00. However, under both circumstances, the
4) All EV adopt three-phase charging mode. loadings of transformer and cables are not more than 90%. The
5) Charging demand of each EV is 15 kWh. capacity of the distribution system equipments is more than suf-
6) Minimal and maximal charging power of each EV is 0 ficient to accommodate the charging load. Moreover, the branch
and 10 kW, respectively. flows obtained by the simplified and approximate LP are very
7) Charging efficiency is set to be 1.0. close. Thus, it is reasonable to take the results obtained by the
8) The optimization time interval is 1 hour. simplified LP as the initial point for linearizing the nonlinear
9) The upper and lower voltage limits are 1.1 and 0.9 p.u., terms of the branch flow equations.
respectively. Both minimum voltages of each time period obtained by the
2) Simulation Results: We utilize MATLAB to call the cplex simplified and approximate LP are shown in Fig. 9a. Compared
LP library function cplexlp for the optimization calculation. with Fig. 7, both the minimum voltages are high when the to-
The configuration of computer is the same as that in case 1. tal load is low. Otherwise, the minimum voltages are low when
During the optimization period, the total conventional load and the total load is high. The minimum voltage is always within

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: FAST SOLVING METHOD BASED ON LINEARIZED EQUATIONS OF BRANCH POWER FLOW 4415

Fig. 9. Minimum voltage comparison at each time period.

charging power obtained by the simplified LP is not able to meet

the node voltage constraint. That is, the optimal charging power
obtained by simplified LP may be unfeasible. As the voltage
constraint has been loosened, too many EVs are charged with
high power during the lowest electricity price time periods dur-
ing 5:00 ∼ 6:00 and 6:00 ∼ 7:00 by simplified LP. However,
based on the optimal charging power obtained by the approxi-
mate LP, the minimum voltages during 5:00 ∼ 6:00 and 6:00 ∼
7:00 are 0.8963 p.u. and 0.8975 p.u., respectively. This result
can approximately meet the node voltage constraint. Overall,
this observation can demonstrate that the proposed linearization
strategy has a high precision.
The optimal charging power of EVs that are nearest and far-
thest from the root node is shown in Fig. 10. When the electric-
ity price is low during 4:00 ∼ 7:00, the charging power is close
to its maximum value. When the electricity price is high during
18:00 ∼ 3:00 and 7:00 ∼ 8:00, the charging power is zero. Since
EVs nearest from the root node are charged with the maximum
power at the lowest price during 5:00 ∼ 7:00, those farthest
Fig. 8. Loadings of electric equipments on phase A. from the root node can be charged at the maximum power dur-
ing 4:00 ∼ 5:00 when the conventional load level is the lowest
the rated range during the optimization time periods. The mini- rather than 5:00 ∼ 7:00. Thus, voltages cannot drop out of the
mum voltages of power flow calculation results using the opti- lower limit. The charging power of EV is determined firstly by
mal charging power obtained by the simplified and approximate electricity price and then by charging location. Although the
LP are shown in Fig. 9b. Since the voltage and power losses electricity is the lowest during 5:00 ∼ 6:00, the charging power
are ignored in the simplified LP, the minimum voltage of the of the EV farthest from the root node is zero. This is because too
power flow calculation drops to 0.8651 p.u. and 0.8659 p.u. dur- many EVs in front of it have been charged with high power. In a
ing 5:00 ∼ 6:00 and 6:00 ∼ 7:00, respectively. Thus, the optimal conclusion, in order not to exceed the voltage magnitude limit,

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 4416 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

the branch flow equation. The optimization variables do not con-

tain the branch current and the phase angle of the node voltage.
However, the model in [16] adopts the current type power flow
equation in the Cartesian coordinate system. Many variables and
constraints are introduced to linearize the branch current con-
straint. The number of variables in [16] is several times of the
proposed model. The number of branch current and node volt-
age constraints in [16] is more than 10 times and 5 times of
the proposed method. This results in a significant increase of
the computational time. In addition, discrete variables are also
introduced in [16] and makes the model non-convex. Also, this
can significantly increase the total computational time.

VII. CONCLUSIONS AND DISCUSSION

In this paper, branch flow equations of balanced and unbal-
anced distribution systems are derived. The model for coordi-
nated charging of EVs is proposed to minimize the total charging
cost of holders. The charging demand, three-phase imbalance
of distribution network, voltage and power flow constraints are
considered. The linearization method is proposed for nonlinear
terms of branch flow equations to develop a fast solving strategy.
Via linearizing nonlinear terms of branch flow equations, the first
stage linear programming (LP) is formulated to calculate the es-
timated branch power and node voltages as initial points. The
second stage LP is formulated to calculate the optimal charging
power based on linearized branch flow equations. Both the high
computational speed and precision of the proposed method are
verified by two case studies. The fast calculation speed and high
precision of the proposed method are verified by four test cases
Fig. 10. Optimal charging power of EVs in the distribution system.
as shown in Tables III–VIII, Fig. 4, and Fig. 9.
In this paper, the simulation time interval is set to 1 hour. Con-
TABLE VIII
sidering the practical application, it should be highlighted that it
COMPARISON WITH THE MODEL IN [16] can be reduced to 15 minutes or even 3 to 5 minutes. Therefore,
forecasting precision can be improved and uncertainties of EVs
and conventional load can be considered by using online rolling
optimization and fast charging EVs can be taken into account so
as to avoid voltage beyond lower limit and branch overloading
caused by fast charging of EVs. Under this circumstance, the ca-
pability of fast computational speed of the proposed method can
be further reflected. This is because the computational speed of
the proposed method is just slightly slower than that of the con-
ventional LP. The proposed method is also applicable to other
objective functions such as minimization of network losses or
charging power of EVs at the end of distribution systems some- total electricity costs of distribution system operator.
times must be charged with low power during low electricity The main contribution of this paper is that we have proposed a
price. fast solution method for EVCC problem. All simulation systems
3) Comparison With Selected Method: As can be seen in are based on actual or IEEE standard distribution networks. The
Fig. 7–Fig. 10, all the EVs are only charged during 4:00 ∼ 7:00 setting of other simulation conditions is also reasonable, which
and the charging power is zero during other time periods. Thus, can fully demonstrate the effectiveness of the proposed method.
for the third and forth steps in Fig. 2, the linearization and opti- The simulation conditions in case 4 are almost the same as those
mization time periods can be reduced to 4:00 ∼ 7:00. Hence, the in reference [16], but the calculation speed is much faster than
computational speed can be further improved. The optimization that in reference [16]. Four simulation cases in this paper show
results of the proposed method are compared with those in [16] that the proposed method has excellent capabilities - high accu-
and shown in Table VIII. As can be seen, the computational racy and fast speed. Even if the access and departure time and
speed of the proposed method is much faster than that of [16]. charging demand of EVs are changed, the proposed method is
This is because the model formulated in this paper is based on still applicable. In future work, more realistic EV user behavior

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: FAST SOLVING METHOD BASED ON LINEARIZED EQUATIONS OF BRANCH POWER FLOW 4417

and laboratory scale testing will be carried out to validate the [18] J. Hu, S. You, M. Lind, and J. Ostergaard, “Coordinated charging of electric
capabilities of the proposed method. vehicles for congestion prevention in the distribution grid,” IEEE Trans.
Smart Grid, vol. 5, no. 2, pp. 703–711, Mar. 2014.
In future work, the proposed method can also be applicable to [19] S. Weckx and J. Driesen, “Load balancing with EV chargers and PV invert-
other objective functions, such as the minimization of network ers in unbalanced distribution grids,” IEEE Trans. Sustain. Energy, vol. 6,
losses and total electricity cost for distribution system operators. no. 2, pp. 635–643, Apr. 2015.
[20] E. Sortomme, M. M. Hindi, S. J. MacPherson, and S. Venkata, “Coordi-
nated charging of plug-in hybrid electric vehicles to minimize distribu-
ACKNOWLEDGMENT tion system losses,” IEEE Trans. Smart Grid, vol. 2, no. 1, pp. 198–205,
Mar. 2011.
The authors would like to thank the anonymous reviewers for [21] P. Staats, W. Grady, A. Arapostathis, and R. Thallam, “A procedure for
their constructive suggestions to this research. derating a substation transformer in the presence of widespread electric
vehicle battery charging,” IEEE Trans. Power Del., vol. 12, no. 4, pp. 1562–
1568, Oct. 1997.
REFERENCES [22] R.-C. Leou, C.-L. Su, and C.-N. Lu, “Stochastic analyses of electric vehi-
cle charging impacts on distribution network,” IEEE Trans. Power Syst.,
[1] L. Cheng, Y. Chang, and R. Huang, “Mitigating voltage problem in vol. 29, no. 3, pp. 1055–1063, May 2014.
distribution system with distributed solar generation using electric ve- [23] J. D. Melo, E. M. Carreno, and A. Padilha-Feltrin, “Spatial-temporal sim-
hicles,” IEEE Trans. Sustain. Energy, vol. 6, no. 4, pp. 1475–1484, ulation to estimate the load demand of battery electric vehicles charging
Oct. 2015. in small residential areas,” J. Control, Autom. Elect. Syst., vol. 25, no. 4,
[2] Y. P. Agalgaonkar, B. C. Pal, and R. A. Jabr, “Distribution voltage con- pp. 470–480, 2014.
trol considering the impact of PV generation on tap changers and au- [24] K. Qian, C. Zhou, M. Allan, and Y. Yuan, “Modeling of load demand due
tonomous regulators,” IEEE Trans. Power Syst., vol. 29, no. 1, pp. 182–192, to EV battery charging in distribution systems,” IEEE Trans. Power Syst.,
Jan. 2014. vol. 26, no. 2, pp. 802–810, May 2011.
[3] J. Zhang, M. Cui, H. Fang, and Y. He, “Two novel load balancing platforms [25] A. Lojowska et al., “Stochastic modeling of power demand due to EVs
using common DC buses,” IEEE Trans. Sustain. Energy, vol. 9, no. 3, using copula,” IEEE Trans. Power Syst., vol. 27, no. 4, pp. 1960–1968,
pp. 1099–1107, Jul. 2018. Nov. 2012.
[4] J. Zhang, X. Yuan, and Y. Yuan, “A novel genetic algorithm based on all [26] B. A. Robbins and A. D. Domínguez-García, “Optimal reactive power
spanning trees of undirected graph for distribution network reconfigura- dispatch for voltage regulation in unbalanced distribution systems,” IEEE
tion,” J. Modern Power Syst. Clean Energy, vol. 2, no. 2, pp. 143–149, Trans. Power Syst., vol. 31, no. 4, pp. 2903–2913, Jul. 2016.
2014. [27] J. Zhang, Y. He, M. Cui, and Y. Lu, “Primal dual interior point dynamic
[5] W. Zheng, W. Wu, B. Zhang, Z. Li, and Y. Liu, “Fully distributed multi-area programming for coordinated charging of electric vehicles,” J. Modern
economic dispatch method for active distribution networks,” IET Gener. Power Syst. Clean Energy, vol. 5, no. 6, pp. 1004–1015, 2017.
Transmiss. Distrib., vol. 9, no. 12, pp. 1341–1351, Apr. 2015. [28] J. Zhang, M. Cui, H. Fang, and Y. He, “Smart charging of EVs in residential
[6] W. Zheng, W. Wu, B. Zhang, H. Sun, and Y. Liu, “A fully distributed distribution systems using the extended iterative method,” Energies, vol. 9,
reactive power optimization and control method for active distribution no. 12, pp. 985–1005, 2016.
networks,” IEEE Trans. Smart Grid, vol. 7, no. 2, pp. 1021–1033,
Mar. 2016.
[7] S. D. Manshadi, M. E. Khodayar, K. Abdelghany, and H. Üster, “Wireless
charging of electric vehicles in electricity and transportation networks,”
IEEE Trans. Smart Grid, vol. 9, no. 5, pp. 4503–4512, Sep. 2018.
[8] M. Singh, P. Kumar, and I. Kar, “Implementation of vehicle to grid in-
frastructure using fuzzy logic controller,” IEEE Trans. Smart Grid, vol. 3,
no. 1, pp. 565–577, Mar. 2012.
[9] O. Beaude, Y. He, and M. Hennebel, “Introducing decentralized EV charg- Jian Zhang received the B.S., M.S., and Ph.D. de-
ing coordination for the voltage regulation,” in Proc. 4th IEEE/PES Inno- grees in electrical engineering and automation from
vative Smart Grid Technol. Eur., 2013, pp. 1–5. Wuhan University, Wuhan, China, in 2005, 2007, and
[10] L. Gan, U. Topcu, and S. H. Low, “Optimal decentralized protocol for 2011, respectively.
electric vehicle charging,” IEEE Trans. Power Syst., vol. 28, no. 2, pp. 940– He is currently a Lecturer with the Hefei Univer-
951, May 2013. sity of Technology. His research interests include load
[11] S. Deilami, A. S. Masoum, P. S. Moses, and M. A. Masoum, “Real-time modeling, renewable distributed generation, and dis-
coordination of plug-in electric vehicle charging in smart grids to minimize tributed network technology.
power losses and improve voltage profile,” IEEE Trans. Smart Grid, vol. 2,
no. 3, pp. 456–467, Sep. 2011.
[12] A. S. Masoum, S. Deilami, P. Moses, M. Masoum, and A. Abu-Siada,
“Smart load management of plug-in electric vehicles in distribution and
residential networks with charging stations for peak shaving and loss min-
imisation considering voltage regulation,” IET Gener. Transmiss. Distrib.,
vol. 5, no. 8, pp. 877–888, Aug. 2011.
[13] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of charging
plug-in hybrid electric vehicles on a residential distribution grid,” IEEE Mingjian Cui (S’12–M’16–SM’18) received the
Trans. Power Syst., vol. 25, no. 1, pp. 371–380, Feb. 2010. B.S. and Ph.D. degrees in electrical engineering and
[14] P. Richardson, D. Flynn, and A. Keane, “Optimal charging of electric automation from Wuhan University, Wuhan, Hubei,
vehicles in low-voltage distribution systems,” IEEE Trans. Power Syst., China, in 2010 and 2015, respectively.
vol. 27, no. 1, pp. 268–279, Feb. 2012. He is currently a Postdoctoral Research Asso-
[15] P. Richardson, D. Flynn, and A. Keane, “Local versus centralized charging ciate with Southern Methodist University, Dallas, TX,
strategies for electric vehicles in low voltage distribution systems,” IEEE USA. He was also a Visiting Scholar from 2014 to
Trans. Smart Grid, vol. 3, no. 2, pp. 1020–1028, Jun. 2012. 2015 in the Transmission and Grid Integration Group,
[16] J. F. Franco, M. J. Rider, and R. Romero, “A mixed-integer linear pro- the National Renewable Energy Laboratory, Golden,
gramming model for the electric vehicle charging coordination problem CO, USA. He has authored and coauthored more than
in unbalanced electrical distribution systems,” IEEE Trans. Smart Grid, 50 peer-reviewed publications. His research interests
vol. 6, no. 5, pp. 2200–2210, Sep. 2015. include power system operation, wind and solar forecasts, machine learning,
[17] Y. He, B. Venkatesh, and L. Guan, “Optimal scheduling for charging and data analytics, and statistics. He is an Associate Editor for the journal of IET
discharging of electric vehicles,” IEEE Trans. Smart Grid, vol. 3, no. 3, Smart Grid and Journal of Computer Science Research. He is also the Best Re-
pp. 1095–1105, Sep. 2012. viewer of the IEEE TRANSACTIONS ON SMART GRID for 2018.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.
 4418 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

Bing Li received the B.E. degree in automobile en- Yigang He received the M.Sc. degree in electrical en-
gineering from Chongqing Science and Technology gineering from Hunan University, Changsha, China,
University, Chongqing, China, in 1995, and the M.E. in 1992, and the Ph.D. degree in electrical engineer-
and Ph.D. degrees in electrical engineering from Hu- ing from Xi’an Jiaotong University, Xi’an, China, in
nan University, Changsha, China, in 2006 and 2011, 1996.
respectively. In 1990, he joined the College of Electrical and
He has been a Postdoctoral Researcher and a Vis- Information Engineering, Hunan University, and was
iting Scholar with the College of Electrical and Infor- promoted to Associate Professor in 1996 and Profes-
mation Engineering, Hunan University, since 2011. sor in 1999. From 2006 to 2011, he was the Director
He has been an Associate Professor with the School of the Institute of Testing Technology for Circuits and
of Electrical and Automation Engineering, Hefei Uni- Systems, Hunan University. He was a Senior Visiting
versity, Hefei, China, since 2013. His current research interests include ra- Scholar with the University of Hertfordshire, Hatfield, U.K., in 2002. From
dio frequency identification technology, wireless sensor networks, and signal 2011 to 2017, he was the Head of the School of Electrical Engineering and
processing. Automation, Hefei University of Technology. On December 1, 2017, he joined
Wuhan University, China, and is currently the Vice-Head of the School of Elec-
trical Engineering and Automation, Wuhan University. Meanwhile, he was the
Vice-President of the China’s Energy Institute of Science and Technology, the
Vice-President of Anhui Scientists Entrepreneurs Association, and the Direc-
tor of the State Local Joint Engineering Laboratory for renewable energy grid
technology. His teaching and research interests are in the areas of power elec-
tronic circuit theory and its applications, testing and fault diagnosis of analog
and mixed-signal circuits, electrical signal detection, smart grid, satellite com-
munication monitoring, and intelligent signal processing. On the above research
areas, he has presided over a number of state-level projects research, such as
the National Natural Science Foundation of China, the State Key Program of
National Natural Science Foundation of China, the National Key Research and
Development Plan “Important Scientific Instruments and Equipment Develop-
ment,” the National High Technology Research and Development Program of
Hualiang Fang received the B.S., M.S., and Ph.D. China, the Major State Basic Research Development Program of China, etc.
degrees in electrical and electronic engineering from He has authored/coauthored about 300 journal and conference papers and sev-
the Huazhong University of Science and Technology, eral chapters in edited books that have been included more than 1000 times in
Wuhan, China, in 1999, 2003, and 2006, respectively. Science Citation Index of American Institute for Scientific Information in the
He is currently an Associate Professor with the aforementioned areas.
School of Electrical Engineering, Wuhan University. Dr. He has been the General Chair, the Session Chair, and a Committee Mem-
His research interests include self-healing smart grid ber of a number of international academic conferences. He was the recipient of
and power system reliability. a number of national and international awards, prizes, and honors. For example,
he was the winner of the National Outstanding Youth Science fund and the China
National Excellent Science and Technology Worker.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on October 29,2024 at 03:35:08 UTC from IEEE Xplore. Restrictions apply.