Vous êtes sur la page 1sur 11

Electric Power Systems Research 143 (2017) 522532

Contents lists available at ScienceDirect

Electric Power Systems Research


journal homepage: www.elsevier.com/locate/epsr

Myopic real-time decentralized charging management of plug-in


hybrid electric vehicles
R. Jalilzadeh Hamidi , H. Livani
EBME Department, University of Nevada, Reno, 1664 North Virginia Street, Reno, NV 89503, USA

a r t i c l e i n f o a b s t r a c t

Article history: This paper proposes a decentralized control algorithm for charging management of Plug-in Hybrid Elec-
Received 3 August 2016 tric Vehicles (PHEVs) in distribution networks. The objectives of the proposed control algorithm are to
Received in revised form 1 November 2016 mitigate PHEV integration challenges (e.g., over-currents and under-voltages in distribution networks)
Accepted 3 November 2016
and to reduce the charging costs of PHEVs. The proposed algorithm adjusts the charging rates of PHEV
chargers utilizing distributed cooperative control to prevent the network constraints (i.e., voltage and
Keywords:
current limits) from being violated. It also determines the operating modes of the chargers (i.e., charg-
Distributed cooperative control
ing, discharging, or idle) using a decision making algorithm to increase the State Of Charges (SOCs) and
Smart micro-grid (SMG)
Plug-in hybrid electric vehicle (PHEV)
decrease the charging costs only based on the current conditions of the distribution network. The pro-
Vehicle-to-grid (V2G) posed algorithm is evaluated on a modied IEEE 37-Node Test Feeder and the simulation studies are
carried out using OpenDSS and MATLAB. The advantages and disadvantages of the proposed algorithm
are discussed and compared with the exiting methods according to the simulation results.
2016 Elsevier B.V. All rights reserved.

1. Introduction ered, and the technical constraints of the feeder can be taken into
account as well [10,11], 3) distribution voltage level in that PHEVs
Unmanaged connection of PHEVs will probably result in sev- are mostly studied as a eet connected to a medium voltage node of
eral issues such as over-currents and under-voltages in distribution a distribution system, and technical constraints can be considered
networks [14]. Proliferation of bidirectional chargers along with [3,1214], and 4) transmission voltage level in which the transmis-
advanced charging management algorithms provide a potential sion system operator is able to predict the power market and obtain
to address the challenges raised by PHEV connection. The charg- optimum charging costs.
ing management methods in the literature are categorized based Based on communication requirements, charging management
on different aspects such as (i) charging management objectives, methods are divided into two categories as: 1) centralized meth-
(ii) coordination scales, and (iii) communication requirements. ods in that a control center optimizes and schedules PHEV charging
Charging management methods are categorized into three groups based on the received data from PHEVs and the power network (e.g.,
based on their objectives as [5]: 1) technical studies, which con- State Of Charges (SOCs) and voltage measurements from power
sider the satisfaction of electrical constraints (e.g., prevention of network), 2) decentralized methods in which PHEV charging is
over-currents and under-voltages) [5,6], 2) economical studies that managed without a control center and the intelligence is dispersed
provide the minimum charging costs to PHEV owners [7], and 3) throughout the network [5,15,16].
techno-economical studies that simultaneously address both tech- Centralized methods provide more optimal solutions for PHEV
nical and economical aspects [3,5]. charging management compared to decentralized methods. How-
Based on the coordination scales, charging management meth- ever, they require a complex communication network and high
ods are divided into four categories as [5]: 1) vehicle level in that computational capability [5,1517]. On the other hand, decen-
the individual benets of only one PHEV are considered, however, tralized methods provide scalability and self-organizability with
the technical constraints are often neglected [79], 2) low voltage low communications and computational requirements [5,1518].
(LV) level in which the PHEVs connected to an LV feeder are consid- Moreover, PHEV owners hesitate to share their User State Informa-
tion (USI) with third parties (e.g., utility operator) [19]. Therefore,
decentralized methods appear to be more practical for PHEV charg-
ing management [5].
Corresponding author.
E-mail address: reza.j.hamidi@gmail.com (R. Jalilzadeh Hamidi).

http://dx.doi.org/10.1016/j.epsr.2016.11.002
0378-7796/ 2016 Elsevier B.V. All rights reserved.
R. Jalilzadeh Hamidi, H. Livani / Electric Power Systems Research 143 (2017) 522532 523

In Refs. [19,20], a decentralized, non-cooperative, game theory- 2. Problem formulation


based method is proposed for minimizing charging costs, however,
network technical constraints are not considered. In Ref. [21], a 2.1. Technical considerations
decentralized method based on the Lagrangian decomposition is
proposed for scheduling PHEV charging pattern and compensating In this section, the charger features, battery capacities, and
reactive power demand, however, this method requires a coordi- charging rates are considered as follows,
nator to nd the nal solution. In Refs. [22,23], each PHEV sends
its demand to an aggregator. The aggregator calculates an approx- a) Among the existing charger types, the chargers with higher
imate optimal solution using convex relaxation method and sends charging rates (e.g., Level-3) cannot be readily used at ordinary
the result to PHEVs. Then, each PHEV decides between charging and households [3337]. Therefore, we limit our studies to level-1
being idle. In Ref. [24], the utility operator determines the num- and -2 chargers with the following specications,
ber of PHEVs to be charged and informs the PHEVs via one-way
communications. Then, each PHEV individually decides between Level-1 (Mode-1) chargers are designed for plugging a PHEV into
charging and being idle. In Ref. [25], a real-time decentralized an ordinary 120-V outlet. The charging rate of a level-1 charger
algorithm is proposed to manage the contributions of local solar is mostly up to 1.4 [kW], [3336].
panels for PHEV charging based on the load forecast. In Ref. [26], a Level-2 (Mode-2, DC Fast) chargers require a 240-V voltage
decentralized control algorithm based on the Alternating Direction source. The maximum charging rate of a level-2 charger is often
Method of Multipliers is proposed that an aggregator along with limited to 6.6 [kW] by regulations [3337].
PHEVs iteratively nd the global optimum. However, transferring
a large amount of data requires a capable and strong communica-
b) The battery capacity of a commercial PHEV is normally in the
tion infrastructure. In Ref. [27], a decentralized myopic algorithm
range of 340 [kWh], [3,38,39].
is proposed based on the current system conditions. Each vehi-
c) The charging rate of a battery affects the battery lifetime. There-
cle operates in either charging or idle mode mainly according to
fore, several methods have been proposed for application of
the received signals from an aggregator and its SOC to reduce the
inconsistent charging rates, [30,31,4042]. Thus, PHEVs can be
adverse effects of PHEVs on load variation. Based on the proposed
charged with a varying charging rate [43].
technology in Ref. [28], an experimental testbed is developed in
Ref. [29] where voltage and current measurements are sent to an
aggregator using power line carriers. If the measurements exceed 2.2. Control system objectives
predened thresholds, the local aggregator stops some of the PHEVs
from charging. In Ref. [30], a cooperative control-based method The PHEVs, chargers, and meters connected to a specic phase
is proposed to minimize voltage deviations and balance the gen- in a section of a distribution system are considered as a group,
eration and demand of distributed generations and PHEVs in DC as shown in Fig. 1. When a large number of PHEVs are charging
grids. simultaneously, the sending ends of MV feeders are vulnerable to
In this paper, we propose a decentralized, myopic, techno- become overloaded. Therefore, these locations are called current
economical, charging management method for PHEVs connected to critical points. Under the same situation, some of the receiving ends
distribution grids. The proposed method is distributed as each PHEV of LV feeders are prone to experience an under-voltage. Thus, these
receives only local and neighboring data [31,32]. The coordination locations are called voltage critical points. It is assumed that the
scale of the proposed method includes both low- and medium- current and voltage critical points are identied and equipped with
voltage levels. The method is self-organizing and largely scalable, current or voltage measurement devices by the utility. Considering,
and provides plug-and-play ability for smart grids with low cost the above-mentioned assumptions and requirements, the control
sparse communication infrastructure. The proposed method also objectives of the proposed method are:
retains USI more private compared to centralized methods. The
main contributions of this paper are: 1) the charging rates of PHEVs Voltage objective: Referring to Fig. 1, the proposed control
are regulated to prevent the occurrence of over-currents and under- method modies the charging rates in each group such that the
voltages using the distributed cooperative control. 2) The proposed voltages at the corresponding voltage critical points stay greater
charging management algorithm selects the operating modes of the than the minimum acceptable value. This is written as,
chargers to reduce the individual PHEV charging costs.
In the proposed method, all PHEV chargers are equipped with a Pci Pcj : V min Vm , i =
/ j, (i, j) N, m M (1)
local controller receiving local and neighboring data, a time-varying
electricity price from the utility, and a control signal based on the where Pci and Pcj are the charging rates of the ith and jth chargers
power grid measurements for preventing the electrical constraints in a group, N is the set consisting of all the chargers in the group,
from being violated. The control signal is available to some of the Vm [pu] is the voltage of the mth voltage critical point, M is the set
local controller. The PHEV controller then selects either charging, consisting of all the voltage critical points in the group, Vmin [pu] is
discharging, or idle operating mode. During the charging mode, the the minimum acceptable voltage, the sign indicates that the left
local controller adjusts the charging rates utilizing the distributed side approaches the right side, and the sign : denotes such that.
cooperative control to retain the voltages and currents in the nor-
mal condition. Current objective: Referring to Fig. 1, the proposed control
The paper is organized as follows: Section 2 presents problem method modies the charging rates in a group such that the cur-
formulation, including technical considerations and control objec- rent at the corresponding current critical point stays less than the
tives. In Section 3, the design of the proposed control method is maximum acceptable value. This is written as,
described. In Section 4, the case study and results are provided, dis-
cussed, and compared with other methods. In Section 5, conclusion Pci Pcj : I I max , i =
/ j, (i, j) N (2)
is presented. Appendices A, B, and C provide mathematical justi-
cations, the technical parameters of the test case, and the details of where I [pu] is the measured current at the current critical point,
the communication networks, respectively. and Imax [pu] indicates the maximum allowable current at the cur-
rent critical point.
524 R. Jalilzadeh Hamidi, H. Livani / Electric Power Systems Research 143 (2017) 522532

Fig. 1. The schematic of the communication links and distribution system in a section. PHEVs are randomly connected to different phases, and they share information through
communication links.

Table 1
Fair Charging objective: Following the proposed method, all the PHEV specications.
chargers belonging to a group are fairly responsible to elimi- Charging Discharging
nate the probable over-currents and under-voltages at the critical
ACC (0) [$/kWh] 0.120 ACC (2) [$/kWh] 0.105
points. Therefore, we introduce Fair Charging Rate as follows: SOC (0) [%] 10 SOC (2) [%] 19
pursuing the Fair Charging Rate, the ratio of the charging rates Pc [kW] 1 Pd [kW] 1
to the maximum charging capacities of all chargers are the same Pr [$/kWh] 0.080 Pr [$/kWh] 0.120
after reaching the equilibrium. This is shown as, c [%] 90 d [%] 90
ACC (2) [$/kWh] 0.105 ACC (3.62) [$/kWh] 0.102
SOC (2) [%] 19 SOC (3.62) [%] 10
Pc1 Pc2 Pci
= max = . . . = max , i N (3)
Pcmax
1
Pc2
Pci
rent available charge in the battery of a PHEV (i.e., traveling for a
where Pci and Pcmax
i
are the charging rate and the maximum charging specic distance costs less with a lower ACC). ACC is dened as,
capacity of the ith charger.  t


ACCi (t1 )SOCi (t1 )BCi + Pc ()Pri ()d

i
ACCi (t) =  t1
, for charging
Economical objective: In the proposed method, each PHEV indi- SOCi (t1 )BCi +c
i
t
Pc i ()d
t1
vidually attempts to charge at a lower electricity price and  (5)

ACCi (t1 )SOCi (t1 )BCi
t
Pd i ()Pri ()d
discharge at a higher electricity price to reduce its charging cost.
  t1
This is shown as, ACCi (t) = SOCi (t1 )BCi 1/d
t
Pd ()d
, for discharging
i t1 i

ACCi (t2 ) < ACCi (t1 ) , (t2 > t1 , i) if SOCi SOCimin (4) where ACCi (t) [$/kWh] is the ACC of the ith PHEV at time t (t1 indi-
cates the beginning of a charging/discharging period), BCi [kWh]
where ACCi (t) is the Average Cost of Charge (ACC) of the ith PHEV at indicates the battery capacity of the ith PHEV, Pci (t) and Pdi (t) [kW]
any given time t, ACC is dened in 3. 1. PHEV Parameters, SOCimin are charging and discharging rates of the ith charger, Pri (t) [$/kWh]
is the minimum desired SOC of the ith PHEV determined by the is the time-varying electricity price for the ith PHEV at its charging
PHEV owner. location, ci and di are charging and discharging efciencies.

3. Proposed control system design 3.1.1. Numerical example


Assuming that a PHEV with the characteristics provided in
3.1. PHEV parameters Table 1 (BC = 20 [kWh]) starts charging at t1 = 0 and contin-
ues charging for 2 [h]. According to the values in Table 1,
2
A myopic and self-interested method for reducing charging 0.120.120+ 10.08dt
costs is desired as: 1) a PHEV can be charged at different places ACC (2) = 0 2 = 0.105 [$/kWh]. Thus, its ACC
0.120+0.9 1dt
(e.g., in public parking lots during days and at home during nights) 0
reduces from 0.12 to 0.105 [$/kWh]. Assuming that the elec-
with different electricity prices, 2) the connection times of PHEVs
tricity price increases from 0.08 to 0.12 [$/kWh] at t = 2 [h],
are not predictable, 3) electricity price forecasts can be inaccurate
then the PHEV starts discharging (with a pre-set discharging
[19,4446], 4) PHEVs are not informed about the number or the
rate of 1 [kW]) to decrease its ACC. After 1.62 [h], its SOC
states of the other PHEVs in a fully decentralized and distributed
reduces to 10% that equals the SOCmin . Hence, it stops discharg-
charging management framework, and 5) PHEV driving patterns  3.62
0.1050.1920 10.12dt
can be random [27]. Thus, we dene ACC as a new property for ing, and ACC (3.62) = 2
 3.62 = 0.102 [$/kWh].
PHEVs. ACC shows the average cost that has been paid for the cur- 0.1920(1/0.9)
2
1dt
R. Jalilzadeh Hamidi, H. Livani / Electric Power Systems Research 143 (2017) 522532 525

Fig. 2. The schematic of the proposed local controller.

Therefore, the SOC reaches 10% again, however, with a total charg-
ing cost of 0.1 20 0.102 = $0.204 instead of the initial cost of
0.1 20 0.12 = $0.240.

3.2. Controller structure

As the proposed charging method is a distributed method, PHEV


chargers are considered as individual agents. Referring to Fig. 2,
every single charger is equipped with a local controller that receives
the following data: 1) the states of its neighboring chargers through
the communication links, 2) local data (i.e., the maximum charging
capacity of the charger, ACC and SOC of the PHEV), 3) the electricity
price that can be dynamically updated by the utility, and 4) a control Fig. 3. High-level owchart of the function h(.).
signal based on the critical point measurements if it is available to
the local controller.
Eq. (6.2) is the output equation, Pr = [Pr1 , Pr2 , . . ., Pri ]T is the
The state-space representation of the proposed control method
vector of time-varying electricity prices are updated by the util-
for a group (i.e., referring to Fig. 1, the PHEVs, chargers, and critical
ity (i.e., the electricity prices can be different at different nodes
points connected to a specic phase in a section) is
in future smart grids [49,50]),
SOC = [SOC1 , SOC2 , . . ., SOC
i ] is
x = Lx + Bu (6.1) the vector of SOCs, SOC min = SOC1min , SOC2min , . . ., SOCimin is the
 min
 vector of the desired minimum SOCs determined by PHEV own-
P C = h x, Pr, SOC, SOC , ACC, P max (6.2) = [ACC1 , ACC2 , . . ., ACCi ] is the vector of ACCs, P max =
C

ers,maxACCmax C

Eq. (6.1) is the state equation based on the cooperative control Pc1 , Pc2 , . . ., Pcmax is the vector of the maximum charging

i
T
[47,48], x = [x1 , x2 , . . ., xi ] is the cooperative control state vector capacities, P C = Pc1 , Pc2 , . . ., Pci is the controlled output vector
Pc
(referring to (3), xi = Pcmax
i
, i N). xs are updated by the cooper- and Pc s are charging rates determined by the non-linear function
i h(.). The function h(.) also selects the operating modes of chargers
ative control through an integrator (PI) as follows through a decision making algorithm. The high-level owchart of
the function h(.) is shown in Fig. 3. If the SOC of a connected PHEV
xi = ki xi dt, 0 xi 1, i N (7) is less than SOCmin , it requires to charge regardless of the electric-
ity price since there is no prediction about the electricity price and
where ki is the gain factor for the ith integrator and xs are limited time of connection. However, if the SOC of the PHEV is larger than

0 and 1.
between SOCmin , the charger switches to charging or discharging modes pro-
L = lij is the graph Laplacian matrix of communication net- vided that it is benecial to the PHEV owner in terms of charging
work as, cost. The details of the algorithm are stated in Table 2.
The charging rates are calculated as Pci = xi Pcmax , i N. The
1, j Ni i
utility determines the discharging rates (Pd s) as a common reg-
lij = |Ni |, j = i (8) ulation, considering power grid conditions such as protection

coordination.
0, else

where Ni is the set of the chargers (agents) that directly send their 4. Case study and simulation results
states to the ith charger (the neighbors of the ith charger), |Ni | is
the size of Ni (in-degree of the ith charger) [47,48]. The performance of the proposed charging management algo-
B = [b1 , b2 , . . ., bi ] is the cooperative control input vector, refer- rithm is evaluated on a modied IEEE 37-Node Test Feeder with
ring to Fig. 1, bi = 1 if the ith charger in a group receives the multi-grounded neutral and additional 19 split-phase three-wire
LV feeders, as shown in Fig. 4. The distribution system is simu-
corresponding control signal, or else bi = 0.
min V V min
lated using OpenDSS [51] and the proposed control algorithm is
I max I V1 V min
u = min I max
, , 2 min , . . ., Vm V is the cooper- implemented using MATLAB. The Component Object Model (COM)
V min V V min
ative control input which is the minimum of the relative errors interface enables MATLAB to take control of the execution of
(normalized errors) calculated using the critical point measure- OpenDSS commands, and also provides a platform for exchang-
ments I, V1 , V2 ,. . ., Vm [pu]. ing data between OpenDSS and Matlab [51]. The assumptions and
526 R. Jalilzadeh Hamidi, H. Livani / Electric Power Systems Research 143 (2017) 522532

Fig. 4. The schematic of the test system.

Table 2 Table 3
The algorithm of h(.). Groups of sections and components of the groups.

Section # Group Components

1 Ga F1, F2, I1-1, V1-1

2 Ga F1, F4, I2-1, V2-1


Gb F2, F5, I2-2, V2-2
Gc F3, F6, I2-3, V2-3

3 Ga F1, F4, F7, F10, I3-1, V3-1


Gb F2, F5, F8, F11, I3-2, V3-2
Gc F3, F6, F9, I3-3, V3-3

F, I, and V indicate feeder, current measurement, and voltage measurement, respec-


tively.

in normal conditions. The graph Laplacian matrix and the coopera-


tive control input vector are presented in Appendix C. The current
at I1 (the current critical point of Section 1) is limited to 20 [A]
(I max = 20). In the rst study, all four PHEVs are connected to the
system at t = 0.5 [min] and start charging. As shown in Fig. 5(a), the
control signal (u) is the minimum of the relative errors as described
in (6.1). During 0.5 < t < 1 [min], the control signal is equal to the
(1),(2)
Charging and discharging criteria are explained in Appendix A. voltage relative error since it is less than the current relative error.
However, for 1 < t [min], the control signal is equal to the current
relative error. Fig. 5(b) depicts the current critical point measure-
requirements are as follows: 1) Communication links are assumed ment that increases after connection of the PHEVs and reaches
ideal (without delay) [32]. 2) The data exchange rate (so-called its maximum value (20 [A]). The comparison of Fig. 5(a) and (b)
communication frequency) is 5 [Hz] [15]. 3) All LV feeders have shows that the control signal becomes zero when the current mea-
the same features presented in Appendix B. 4) Five level-1 and surement reaches its maximum value. Fig. 5(c) and (d) show Fair
ve level-2 chargers (ci , di = 90%, i N) are available in each Charging Rates and actual charger outputs that start increasing at
LV feeder [27]. In order to reduce the probability that all PHEVs t = 0.5 [min], their nal values are also stated in Table 4. In the sec-
become charged, at least, to SOCmin , a battery capacity of 40 [kW h] ond study, the system is working in steady state with two PHEVs
is considered for the PHEVs. The distribution grid is divided into connected to Chargers 2 and 3. As shown in Table 4 and Fig. 6(c),
three sections with the details given in Table 3. the Fair Charging Rates of both chargers are 1 for t < 0.5 [min]. At
t = 0.5 [min], the other two PHEVs are connected to Chargers 1 and
4.1. Illustrative test case 4. Fig. 6(a) and (b) depict that the control signal becomes zero as the
current at the current critical point reaches its maximum value. Fair
Section 1 in Fig. 4 with four chargers (with the details given in Charging Rates converge to 0.9593, as shown in Fig. 6(c) and stated
Table 4) is considered in the illustrative test case to demonstrate in Table 4. As the system features are identical in both studies, the
the functionality of the cooperative control in retaining the system nal Fair Charging Rates and charger outputs are equal.
R. Jalilzadeh Hamidi, H. Livani / Electric Power Systems Research 143 (2017) 522532 527

Table 4
Specication of the chargers in the illustrative test case.

Studya # Ch.b # Ch.c level Feeder # t < 0.5 [min] 2 < t [min]

Fair charging rates [%] Charger outputs (Pci ) [kW] Fair charging rates [%] Charger outputs (Pci ) [kW]

1 1 1 1 0 0 95.93 1.3430
2 2 1 0 0 95.93 6.3313
3 1 2 0 0 95.93 1.3430
4 2 2 0 0 95.93 6.3313

2 1 1 1 0 0 95.93 1.3430
2 2 1 100 6.6 95.93 6.3313
3 1 2 100 1.4 95.93 1.3430
4 2 2 0 0 95.93 6.3313
a
Study number.
b
Charger number.
c
Charger level.

Fig. 5. (a) The control signal which is the minimum of the relative current and voltage errors. (b) The current at the current critical point. (c) Fair Charging Rates start to
increase at t = 0.5 [min]. As the current critical point reaches its maximum value (20 [A]), they are limited to 0.9593. (d) Charger outputs. Chargers 1 and 3 are level-1. Thus,
Pc1 and Pc3 reach 1.3430 [kW]. Chargers 2 and 4 are level-2. Thus, Pc1 and Pc3 reach 6.3313 [kW].

Fig. 6. (a) The control signal which is the minimum of the relative current and voltage errors. (b) The current at the current critical point. (c) Fair Charging Rates that start
to converge at t = 0.5 [min]. As the current critical point reaches its maximum value (20 [A]), they are limited to 0.9593. (d) Charger outputs. Chargers 1 and 3 are level-1.
Thus, Pc1 and Pc3 reach 1.3430 [kW]. Chargers 2 and 4 are level-2. Thus, Pc1 and Pc3 reach 6.3313 [kW].
528 R. Jalilzadeh Hamidi, H. Livani / Electric Power Systems Research 143 (2017) 522532

Fig. 7. Percentage of the chargers in different operating modes and SOC curves.

Fig. 8. (a) Total PHEV and Non-PHEV load. (b) Real-time electricity price [52].

4.2. Results and discussion distributed among the consumers in Sections 1, 2, and 3 (in Fig. 4),
respectively. One thousand Monte Carlo simulations (MCSs) are
The charging management algorithm is studied for the time carried out and the average of the simulation results is presented
period between hour 17 (5 pm) and 9 am with a time step of and discussed below.
10 [min]. The total number of connected PHEVs are 190 (i.e., 10 Fig. 7 shows the percentage of the chargers working in differ-
PHEVs in each LV feeder) and the details of the communication ent operating modes and SOC curves. The Max., Ave., and Min. SOC
links are given in Appendix C and Pdi = 1 [kW] , i N. In order to curves show the maximum, average, and minimum of SOCs. The
study a heavily loaded condition, all PHEVs are assumed to be con- Max. SOC curve is less than 50% at t = 17 [h], indicating that none
nected to the grid at hour 17 with SOCi unif (0.05, 0.35), SOCimin = of the PHEVs has an SOC greater than SOCmin and all the charg-
50%, i P where unif indicates a uniformly distributed random ers are in charging mode regardless of the electricity price. It is
variable between 0.05 and 0.35, and P is the set consisting of all the also noticed that once some of SOCs reach SOCmin (shown with a
PHEVs in the group. Therefore, SOCi < SOCimin , i P and all PHEVs double arrow in Fig. 7), they stop charging as the electricity price
start charging immediately after connection. The ACCs of the PHEVs is high (Fig. 8(b)). If they continue charging, their ACCs rise that
are uniformly distributed as ACCi unif (0.08, 0.16) , i P. All the is not desirable. During 19 : 10 t 22, the Max. SOC curve is
minimum acceptable voltages are 0.95 [pu] and the maximum cur- almost at, indicating that the PHEVs with SOC SOC min are reluc-
rents are: I1max = 45, I2max = 105, and I3max = 150 [A] (shown in Fig. 4). tant to charge. During 22 t 23, the Max. SOC curve decreases
Equal real-time electricity prices based on [52] are assumed for all as the electricity price is high (Fig. 8(b)) and discharging is bene-
the nodes. 15%, 35% and 50% of the total Non-PHEV load are equally cial (i.e., ACC reduces). Therefore, a considerable portion (almost
25%) of PHEVs sell electricity to the grid during this hour. During
hour 23 to 24 the electricity price is low, therefore, discharging is
not benecial to many discharging PHEVs, and they become idle.
A sharp increase in the Max. and Ave. SOC curves occur during
24 t 3 am since the electricity price is low and the idle PHEVs
with SOC > 50% resume charging to reduce their ACCs. The per-
centage of fully charged PHEVs rises after 3 am. Consequently, at
the end of the charging period, almost 50% of PHEVs are idle as
they are fully charged or further charging does not decrease their
ACCs. The average of all SOCs (Ave. SOC curve) increases from 20%
to 79% during the simulation time, and almost 2.5% of SOCs are not
sufciently charged.
Fig. 8(a) shows the total Non-PHEV and PHEV loads during the
simulation period. It is noticed that the proposed control algo-
Fig. 9. ACCs of PHEVs during charging period. rithm prevents the transformer overloading (2.5 [MVA]). During
R. Jalilzadeh Hamidi, H. Livani / Electric Power Systems Research 143 (2017) 522532 529

Fig. 10. Voltage and current measurements at the critical points.

Fig. 11. The proposed method neglecting the economical objective. (a) Total PHEV and Non-PHEV load. (b) The percentage of the chargers in different operating modes and
SOC curves.

Fig. 12. The total system load. (a) 30% PHEV penetration. (b) 50% PHEV penetration [27].

18 : 40 t 23, an increasing number of chargers switch to dis- number of PHEVs become fully charged and the total PHEV load
charging or idle mode as the electricity price increases (Fig. 8(b)) decreases. As the charging pattern of PHEVs responds to the elec-
and the total PHEV load decreases. However, during hour 23 and tricity price, which is lower in the load valley, it is observed that
3 am, the electricity price is low, then more PHEVs start charging the largest PHEV load is placed in the valley. However, the valley is
and the total PHEV load rises. For the time after 3 am, a growing partially lled since the chargers select their operating modes only
530 R. Jalilzadeh Hamidi, H. Livani / Electric Power Systems Research 143 (2017) 522532

based on the exiting situation and without any load or electricity


price forecast.
Fig. 9 shows the Max., Ave., and Min. ACC curves during the
simulation period based on the MCSs. It is noticed that the Max.
ACC curve reduces from 0.16 and 0.124 [$/kWh], and the Ave. ACC
curve decreases from 0.12 to 0.0983 [$/kWh], therefore, the Max.
and Ave. of ACCs are reduced by 0.036 and 0.0217 [$/kWh], respec-
tively. However, the minimum ACC increases from 0.08 to 0.0835
[$/kWh]. During 18 : 10 < t < 00 : 00 , both Min. and Ave. ACC
curves increase. This shows that a considerable number of SOCs (i.e.,
more than 60%) are less than SOCmin , and they require to be charged
regardless of the electricity price. However, during this time inter-
val, the Max. ACC curve decreases slowly, mainly because of a small
number of discharging PHEVs. During 00 : 00 < t < 3 : 30 , all the
curves sharply decrease according to the low electricity price and
high percentage of charging PHEVs. For 3 : 30 < t, Ave. and Min.
ACC curves increase less sharply as the percentage of the idle charg-
ers rises.
Fig. 13. Block diagram of the central controller for comparison purposes.
Fig. 10 shows the voltage and current measurements at the crit-
ical points during the simulation time. It is noticed that all the
PHEVs are connected to the power system and start charging at
t = 17 [h]. Therefore, a sharp decrease and increase respectively
occur in voltage and current measurements at all critical point mea-
surements between t = 17 : 00 and t = 17 : 10 . However, none of
the measurements passes its corresponding limitation owing to the
cooperative-based controller. As Fig. 10(b) shows, the currents at
Current Critical Points I1 and I2 reach their maximum allowable
values (45 and 105 [A], respectively) at t = 17 : 10 because the
voltages at the related voltage critical points (V1-1, V2-1, V2-2, and
V2-3) are higher than the minimum allowable value (0.95 [pu]).
However, I3 does not reach 150 [A] at t = 17 : 10 since the volt-
ages at the related voltage critical points (V3-1, V3-2, and V3-3)
drop. Therefore, the controller limits the charging rates in Section
three to retain the voltages at the minimum value (0.95 [pu]) as
shown in Fig. 10(a).

4.3. Comparison

In this section, the performance of the proposed algorithm is


compared with the exiting PHEV charging management methods
in the literature. The proposed cooperative control-based controller
is able to prevent the system constraints from being violated if only Fig. 14. The currents at current critical points for distributed and centralized con-
technical objectives are considered (i.e., the economical objective is trollers.

neglected). However, the valley lling concept is no longer available


since PHEVs start charging after connection and continue charging Distributed controllers are often compared to centralized con-
until they become fully charged, as shown in Fig. 11. According trollers in term of settling time. Fig. 13 illustrates the block
to 1000 MCSs, almost 6.5% of PHEVs have SOCs less than 50% at diagram of a centralized networked controller equivalent to the
the end of the simulation period (9 am). Therefore, the percent- proposed decentralized method, where 1 = [1, 1, . . ., 1]T1m , m
is the number of voltage measurements in the group, P max =
age of PHEVs without desired amount of charge is greater than the
max max C
one in the case studies with economical objective since discharging Pc1 , Pc2 , . . ., Pcmax
n
is the vector of the maximum charging
or idle modes of some PHEVs provide more charging opportunity capacities, n is the number of the chargers in the group, and K
to the other PHEVs. With respect to the charging costs, the Ave. is the gain factor which is selected identical to the gain factor in
ACC increases from 0.12 to 0.141 [$/kWh] while in the previous the decentralized controller. All the initial conditions and parame-
simulation it reduced from 0.12 to 0.0983 [$/kWh]. ters related to PHEVs and the distribution system, communication
The proposed method is also compared with the results in 0 frequency, the number and location of the PHEVs, voltage critical
with respect to the valley lling capability. Fig. 12 shows the total points, current critical points, and the Max. and Min. allowable val-
load for different PHEV penetration levels in 0. It can be observed in ues of the critical points are the same. PHEVs are connected to
Fig. 12(a) and (b) that On-line Decentralized methodology con- the grid at time t = 17 [h]. The currents at current critical points
siderably lls the valley. However, the maximum capacity of the are linearly proportional to the total load, and their variations
substation transformer is stated as 2.5 [MVA] in the IEEE 37-bus during each hour are resulted only by the PHEV load as Fig. 8(a)
test feeder, and the transformer is overloaded for several hours shows. Therefore, the currents at current critical points are shown
using the method presented in 0. While, referring to Figs. 8 and 11, in Fig. 14. The current measurements during 17 : 00 t 17 : 10
the proposed decentralized method in this paper prevents the resulted by centralized and the proposed decentralized controllers
substation transformer from being overloaded since the technical are depicted in Fig. 14. The settling times (Ts ) are stated in Table 5.
constraints are considered. As it can be observed from the simulation results, the centralized
R. Jalilzadeh Hamidi, H. Livani / Electric Power Systems Research 143 (2017) 522532 531

Table 5
PHEV discharging leads to a decrease in ACC if ACC(t) < 0.
Settling times. t
a
t
Control method Settling time [s]
N
ACCi (t) = D
, where N = ACCi (t1 ) SOCi (t1 ) BCi Pdi ()
I1 Centralized method 55
Proposed method 89.4 t1
t
I2 Centralized method 94.2  
Proposed method 134.4 Pri () d and D = SOCi (t1 ) BCi 1/di Pdi () d.
I3 Centralized method 110.4 t1
Proposed method 272.4
a
98% of the nal value. P 
di (t)
(Pd i (t)Pri (t))(D)+ d (N)
ACC(t) i ACC(t)
= . for < 0,
(D)P 
t 2 t
controller converges faster since each charger receives the charg- di (t)
(Pdi (t) Pri (t)) (D) + d (N) < 0. If d 0, [(Pdi (t)
ing rate directly from the control center as shown in Fig. 13. In the i
case of the proposed decentralized method, as the number of agents Pd (t)
Pri (t))(SOCi (t1 ) BCi ) + ( d
i
)(ACCi (t1 ) SOCi (t1 ) BCi )] < 0.
(chargers) increases, the settling time increases since data propa- i

gation time rises. For the centralized controller, the settling time 
Thus, 
discharging mode leads to a reduction in ACC if
ACCi (t1 )
varies from section to section as the power system topologies are d < Pri (t).
i
different in each section. Therefore, the PHEV charging differently
affects the measurements in each section which results in different
feedback gains, and consequently, settling times. Appendix B.

Feeder transformers are three-winding with ratios of 4.8


5. Conclusion to 0.12 and to 0.12 [kV], their capacities are 75 [kVA]. X12 =
 X13 = 3.22%,X23 = 2.12%. Feeder Lengths
3.22%,  are 1000 [ft] with
This paper proposes a myopic decentralized control method for 0.082 0.028 0.172 0.058
charging management of Plug-in Hybrid Electric Vehicles (PHEVs). R= ,X = [pu/mile].
0.028 0.086 0.058 0.178
In the proposed method, the charging rates of PHEV chargers are
determined through a distributed cooperative control-based algo-
rithm. One of the controller objectives is to retain the current and Appendix C.
voltage measurements at critical points in normal range. The other
objective is to decrease the charging costs of PHEVs. In this regard, In the simulations, the minimum communication links among
PHEV chargers are switched among the charging, discharging, or the chargers are assumed.
idle operating mode based on PHEV State of Charges (SOCs), PHEV
Average Cost of Charges (ACCs), and the electricity prices at the For the illustrative
testcase, the graph Laplacian matrix is
connection nodes. The proposed control method is evaluated on a 1 1 0 0
modied IEEE 37-Node Test Feeder. The distribution grid is sim- 1 2 1 0
L=
1
. As only Charger 1 receives the con-
ulated using OpenDSS and the proposed charging management 0 1 2
algorithm is implemented using MATLAB. The simulation results 0 0 1 1
verify the performance of the proposed method in preventing the
T
trol signal (u), B = 1 0 0 0 . As all eigenvalues of (L) =
occurrence of over-current and under-voltage conditions at critical  
points and reducing the charging costs of PHEVs. As the proposed 3.41, 0.58, 2, 0 . As all eigenvalues are negative except
for exactly one zero-eigenvalue, the system is BIBO stable and
method is completely myopic and distributed, it partially lls the
T
load valley, and does not provide the most optimum charging costs, converges to 1 1 1 1 where is a real number 0. The
while it reduces the charging costs. integrator gain factors ki = 0.25, i = 1, 2, 3, 4.
For the entire system, the cooperative control input vectors

T
(B = b1 b2 . . . bi ) are: Section 1, Group a, b1 , b10 = 1. Sec-
Appendix A.
tion 2, Group a, b1 , b10 , b20 = 1; Group b, b1 , b10 , b20 = 1; Group
c, b1 , b10 , b20 = 1. Section 3, Group a, b1 , b10 , b20 , b30 , b40 = 1;
PHEV charging leads to a decrease in ACC if ACC(t) < 0.
t Group b, b1 , b10 , b20 , b30 , b40 = 1; Group c, b1 , b10 , b20 , b30 = 1.
t Graph Laplacian matrices are
N
ACCi (t) = , where N = ACCi (t1 ) SOCi (t1 ) BCi + Pci ()
D
t1
1 1 0 0 0
t 1 2 1 0 0
0 1 2 1 0
L= .
for all the groups. Eigenval-
Pri () d and D = SOCi (t1 ) BCi + ci Pc i () d. .. .. .. ..
.. . . .
..
. .
t1
0 0 0 0 1 1
    ues of (L) are all negative except for exactly one zero-eigenvalue.
ACC(t) Pc (t)Pri (t) (D) c Pc i (t) (N)
i i The integrator gain factors ki = 0.25, i N.
= . for
t (D)2
ACC(t)
   
< 0, Pci (t) Pri (t) (D) < ci Pc i (t) (N).
t   References
If d 0, Pci (t) Pri (t) (SOCi (t1 ) BCi ) <
  [1] S. Studli, E. Crisostomi, R. Middleton, R. Shorten, A exible distributed
ci Pc i (t) (ACCi (t1 ) SOCi (t1 ) BCi ). Thus, charging mode
framework for realizing electric and plug-in hybrid vehicle charging policies,
leads to a reduction in ACC if ACCi (t1 ) ci > Pri (t). Int. J. Control 85 (8) (2012) 11301145.
532 R. Jalilzadeh Hamidi, H. Livani / Electric Power Systems Research 143 (2017) 522532

[2] S. Han, S. Han, Kaoru Sezaki, Development of an optimal vehicle-to-grid [26] M.G. Vaya, G. Andersson, S. Boyd, Decentralized Control of Plug-in Electric
aggregator for frequency regulation, IEEE Trans. Smart Grid 1 (June (1)) Vehicles under Driving Uncertainty, ISGT-Europe, Istanbul, Turkey, 2014, pp.
(2010) 6572. 16.
[3] S. Abedi, M. He, S.M. Fatemi, Employing Price-responsive PHEVs in Microgrid: [27] T. Li, R. Cui, F. Franchetti, D. Ilic, On-line Decentralized Charging of Plug-in
Optimal Operations and Security Management, ISGT, MN, USA, 2015, pp. 15. Electric Vehicles in Power Systems, 2011 [Online]. Available: arXiv:1106.5063.
[4] I. Niazazari, H.A. Abyaneh, M.J. Farah, F. Safaei, H. Nasi, Voltage Prole and [28] M. Gillie, G. Nowell, The Future for EVs: Reducing Network Costs and
Power Factor Improvement in PREV Charging Station Using a Probabilistic Disruption, HEVC, London, UK, 2013, pp. 15.
Model and Flywheel, EPDC, Tehran, Iran, 2014, pp. 100105. [29] E. Saunders, T. Butler, J. Quiros-Tortos, L.F. Ochoa, R. Hartshorn, Direct Control
[5] N. Leemput, J. Van Roy, F. Geth, P. Tant, B. Claessens, J. Driesen, Comparative of EV Charging on Feeders with EV Clusters, CIRED, Lyon, France, 2015.
Analysis of Coordination Strategies for Electric Vehicles, ISGT Europe, [30] R. Mahmud, A. Nejadpak, R. Ahmadi, Cooperative Load Sharing in V2G
Manchester, UK, 2011, pp. 18. Application, EIT, IL, USA, 2015, pp. 451456.
[6] K. Zhou, L. Cai, Randomized PHEV charging under distribution grid [31] S. Mansour, I. Harrabi, M. Maier, G. Jos, Co-simulation study of performance
constraints, IEEE Trans. Smart Grid 5 (2) (2014) 879887. trade-offs between centralised, distributed, and hybrid adaptive PEV charging
[7] M. Ghofrani, A. Arabali, M. Etezadi-Amoli, M. Sami-Fadali, Smart scheduling algorithms, Comput. Netw. 93 (2015) 153165.
and cost-benet analysis of grid-enabled electric vehicles for wind power [32] A. Bidram, A. Davoudi, F.L. Lewis, J.M. Guerrero, Distributed cooperative
integration, IEEE Trans. Smart Grid 5 (September (5)) (2014) 23062313. secondary control of microgrids using feedback linearization, IEEE Trans.
[8] A. Bedir, B. Ozpineci, J.E. Christian, The Impact of Plug-in Hybrid Electric Power Syst. 28 (3) (2013) 34623470.
Vehicle Interaction with Energy Storage and Solar Panels on the Grid for a [33] T.A. Short, Electric Power Distribution Handbook, CRC Press LLC, Florida, USA,
Zero Energy House, IEEE PES T&D, LA, USA, 2010, pp. 16. 2004.
[9] C. Tushar, M. Maier, M.F. Uddin, Smart microgrids: optimal joint scheduling [34] San Diego Regional Plug-in Electric Vehicle (PHEV) Readiness Plan, January
for electric vehicles and home appliances, IEEE Trans. Smart Grid 5 (January 2014.
(1)) (2014) 239250. [35] California Center for Sustainable Energy, Electric Vehicle Charging Station
[10] K. Clement-Nyns, E. Haesen, J. Driesen, The impact of charging plug-in hybrid Installation Guidelines, 2016, Available on-line https://energycenter.org.
electric vehicles on a residential distribution grid, IEEE Trans. Power Syst. [36] I. Rahman, P.M. Vasant, B.S.M. Singh, M. Abdullah-Al-Wadud, N. Adnan,
(2010) 371380. Review of recent trends in optimization techniques for plug-in-hybrid and
[11] J.A. Pecas Lopes, C.L. Moreira, A.G. Madureira, Dening control strategies for electric vehicle charging infra structures, Renew. Sustain. Energy Rev. 58
microgrids islanded operation, IEEE Trans. Power Syst. 21 (May (2)) (2006) (2016) 10391047.
916924. [37] H.A. Ortega Carrascal, J.R. Garca, Review of NTC 2050 and NFPA 70 on Their
[12] S. Shao, T. Zhang, M. Pipattanasomporn, S. Rahman, Impact of TOU Rates on Section Dedicated to Equipments for Electrical Vehicles Recharge Systems,
Distribution Load Shapes in a Smart Grid with Phev Penetration, Technical WEA, Teatro Mxico, 2015, pp. 15.
Report, Advanced Research Institute, VT, U.S.A, 2010. [38] The Massachusetts Department of Energy Resources, Installation Guide for
[13] L.P. Fernandez, T.G. San Roman, R. Cossent, C.M. Domingo, P. Frias, Electric Vehicle Supply Equipment (EVSE), 2014 https://www.mass.gov.
Assessment of the impact of plug-in electric vehicles on distribution [39] K. Rahimi, B. Chowdhury, A Hybrid Approach to Improve the Resiliency of the
networks, IEEE Trans. Power Syst. 26 (May (1)) (2010) 206213. Power Distribution System, NAPS, Washington, USA, 2014, pp. 16.
[14] E. Sortomme, M.M. Hindi, S.D.J. MacPherson, S.S. Venkata, Coordinated [40] E. Kim, K.G. Shin, J. Lee, Real-time Discharge/Charge Rate Management for
charging of plug-in hybrid electric vehicles to minimize distribution system Hybrid Energy Storage in Electric Vehicles, RTSS, Rome, Italy, 2014, pp.
losses, IEEE Trans. Smart Grid (2011) 198205. 228237.
[15] R.J. Hamidi, H. Livani, S.H. Hosseinian, G.B. Gharehpetian, Distributed [41] H.T. Mouftah, M. Erol-Kantarci, Smart Grid: Networking, Data Management,
cooperative control system for smart micro-grids, Electr. Power Syst. Res. 130 and Business Models, CRC Press, 2016 (Chapter 13).
(2016) 241250. [42] S. Sojoudi, S.H. Low, Optimal Charging of Plug-in Hybrid Electric Vehicles in
[16] A. Mohsenian-Rad, V.W.S. Wong, J. Jatskev, R. Schrober, A. Leon-Garcia, Smart Grids, PES GM 2011, Michigan, USA, 2011, pp. 16.
Autonomous demand side management bases on game-theoretic energy [43] Siemens, Siemens VesriCharge Manual, 2014.
consumption scheduling for the future smart grid, IEEE Trans. Smart Grid [44] S. Mohajeryami, P. Schwarz, P.T. Baboli, Including the Behavioral Aspects of
(2010) 320331. Customers in Demand Response Model: Real Time Pricing Versus Peak Time
[17] R. Lasseter, Smart distribution: coupled microgrids, Proc. IEEE 99 (6) (2011) Rebate, NAPS, Charlotte, NC, USA, 2015, pp. 16.
10741082. [45] S. Mohajeryami, M. Doostan, A. Asadinejad, P. Schwarz, Error analysis of
[18] A. Vaccaro, V. Loia, G. Formato, P. Wall, V. Terzija, A self organizing customer baseline load (CBL) calculation methods for residential customers,
architecture for decentralized smart microgrids synchronization, control and IEEE Trans. Ind. Appl. (2016), early access.
monitoring, IEEE Trans. Ind. Inf. 11 (July (1)) (2014) 289298. [46] S. Mohajeryami, I.N. Moghaddam, M. Doostan, B. Vatani, P. Schwarz, A novel
[19] M. Zhongjing, D. Callaway, I. Hiskens, Decentralized charging control for large economic model for price-based demand response, Electr. Power Syst. Res.
populations of plug-in electric vehicles: application of the nash certainty 135 (2016) 19.
equivalence principle, Proceedings of CCA (2010) 191195. [47] Z. Qu, Cooperative Control of Dynamic Systems, Springer Science & Business
[20] Z. Ma, D.S. Callaway, I.A. Hiskens, Decentralized charging control of large Media, 2009.
populations of plug-in electric vehicles, IEEE Trans. Control Syst. Technol. 21 [48] R. Olfati-Saber, J.A. Fax, R.M. Murray, Consensus and cooperation in
(1) (2013) 6778. networked multi-agent systems, Proc. IEEE 95 (1) (2007) 215233.
[21] B. Jiang, Y. Fei, Decentralized Scheduling of PHEV On-street Parking and [49] M. Mahmoodi, P. Shamsi, B. Fahimi, Economic dispatch of a hybrid microgrid
Charging for Smart Grid Reactive Power Compensation, ISGT, Shanghai, China, with distributed energy storage, IEEE Trans. Smart Grid 6 (November (6))
2013, pp. 16. (2015) 26072614.
[22] C.K. Wen, J.C. Chen, J.H. Teng, P. Ting, Decentralized Energy Management [50] P.M. Sotkiewicz, J.M. Vignolo, Nodal pricing for distribution networks:
System for Charging and Discharging of Plug-in Electric Vehicles, WCSP, efcient pricing for efciency enhancing DG, IEEE Trans. Power Syst. 21 (May
Huangshan, China, 2012. (2)) (2006) 10131014.
[23] C.K. Wen, J.C. Chen, J.H. Teng, P. Ting, Decentralized plug-in electric vehicle [51] Electric Power Research Institute (EPRI), Smart Grid Resource Center,
charging selection algorithm in power systems, IEEE Trans. Smart Grid 3 Simulation Tool, 2016, Available on-line http://smartgrid.epri.com/
(December (4)) (2012) 17791789. SimulationTool.aspx.
[24] K. Turitsyn, N. Sinitsyn, S. Backhaus, M. Chertkov, Robust [52] M.F. Shaaban, M. Ismail, E.F. El-Saadany, W. Zhuang, Real-time PEV
Broadcast-Communication Control of Electric Vehicle Charging, charging/discharging coordination in smart distribution systems, IEEE Trans.
SmartGridComm, MD, USA, 2010, pp. 203207. Smart Grid 5 (July (4)) (2014) 17971807.
[25] I. Harrabi, M. Maier, Performance Analysis of a Real-time Decentralized
Algorithm for Coordinated PHEV Charging at Home and Workplace with PV
Solar Panel Integration, PES GM, WA, USA, 2014, pp. 15.

Vous aimerez peut-être aussi