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

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
1

Collaborative EV Routing and Charging Scheduling with

Power Distribution and Traffic Networks Interaction
Jiayan Liu, Student Member, IEEE, Gang Lin, Sunhua Huang, Yang Zhou, Christian Rehtanz, Senior
Member, IEEE, and Yong Li, Senior Member, IEEE

Abstract—The increasing of electric vehicles (EVs) alleviates g Traffic node installed with charging station kn.
the faced environmental problems but brings challenges to the  (u)
T
Set of neighboring nodes of traffic node u
optimal operation of transportation network (TN) and GTN Graph of the traffic network
distribution network (DN). However, the most of existing research
works consider EV charging station assignment and navigation GDN Graph of the power distribution network
services in the TN separately from charging station power Capn Battery capacity of EV n
scheduling services in the DN. To overcome this research gap, this drs distance between node r and s
paper proposes a collaborative optimal routing and scheduling Ec Electricity consumption per km
(CORS) method, providing optimal route to charging stations and Mk Number of chargers in charging station k
designing optimized charging scheduling schemes for each EV. In T/t Set and index of time slot
the order of reporting, whenever an EV reports its charging
demand, a CORS optimization model is built and solved so that a Tn Operation time slot of EV n, from trn to tdn
specific charging scheme is designed for that EV. Then, the TN and trn Report time of EV n
DN status is updated to guide the subsequent EVs operating. The ten Expected charging time of EV n
proposed CORS integrates the real-time state of the TN and DN, tan Arrival time of EV n
and effects positive benefits in helping EVs to avoid traffic tdn Departure time of EV n
congestion, improving the utilization level of charging facilities Sini Initial battery SOC of EV n
n
and enhancing charging economy. The combined distributed
biased min consensus algorithm and generalized benders Sreq
n Requested battery SOC of EV n
decomposition algorithm are adopted to solve the complex Ereqn Requested energy of EV n
nonlinear optimization problem. Through comparing with the Capn Battery capacity of EV n
existing methods, better effectiveness is verified by simulation Pchmax Maximum charging power provide by a charger
results. Pcsj,t Expected charging power of a charging station
located at bus j.
Index Terms-- Electric Vehicle, Charging Scheduling, Route
λt TOU electricity price at time t
Navigation, Transportation Network, Distribution Network
μ Monetary price of time
NOMENCLATURE kn Integer variable, the charging station index
assigned for EV n
B/ij Set and indexes of distribution network bus an,t Binary decision variable, which is equal to 1 if EV
B(k) Set of distribution network buses installed with n connected to a charger at the time slot t,
charging station k. otherwise, zero
L/l Set and indexes of distribution network line Pn,t Decision variable, charged power of EV n at time
C/k Set and indexes of charging station slot t
E/n Set and indexes of EV charging task
E(t) Set of EVs at time slot t
E’(t) Set of unprocessed EV charging task at time slot t I. INTRODUCTION
En /n
E’n /n
Set and indexes of charging information of EV n
Set and indexes of unprocessed charging
information of EV n
C limate change is an important opportunity and challenge
facing the economic and social development of all
countries. In order to actively respond to climate change, many
E(k) Set of EVs staying in the station k. countries have proposed their own ambitious timetables for
T/uv Set and indexes of traffic network intersection carbon neutrality [1], for instance, Germany in 2050, China in
R/Auv Set and indexes of traffic network road 2060. Replacing fossil-fueled cars with electric vehicles (EVs)
T(k) Set of traffic nodes installed with charging station is a good solution to environmental protection in the
k. transportation field. The proliferation of EVs puts higher


This work is based upon work in the project KonVeEn (Grant No. 03EK3057A, supported by the German Federal Ministry of Education and Research), in
part by the “111” Project of China under Grant B17016, and supported in part by China Scholarship Council (No. 201806130201). (Corresponding author: Jiayan
Liu, Gang Lin.)
J. Liu, G. Lin Y. Zhou and C. Rehtanz are with the Institute of Energy Systems, Energy Efficiency and Energy Economics, TU Dortmund University, Dortmund
44227, Germany (e-mail: [email protected]; [email protected]; [email protected]; [email protected])
S. Huang is with the School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, Shanghai, 200240, China (e-mail:
[email protected])
Y. Li is with the College of Electrical and Information Engineering, Hunan University, Changsha 410082, China (e-mail: [email protected])

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
2

requirements for charging infrastructure and effects the impacts of charging facilities on integrated TN and DN system
operations of the distribution network (DN) and traffic network is discussed in [20] and corresponding reliability assessment
(TN) remarkably [2]. approach is proposed by [21]. In [22], an optimal traffic-power
Various studies have been conducted on the EV charging flow model with traffic user equilibrium constraints is proposed
station (CS) planning [3, 4]. However, the number of chargers to address the coordinated operation of TN and DN. The
that can work at the same time in a charging station is limited optimal pricing to manage EVs in coupled power and
by the grid structure and transformer capacity. In addition, the transportation networks is proposed in [23], which can optimize
insufficient number of chargers during peak hours will reduce the daily trip costs as well as refueling strategies. But the
users’ enthusiasm for using EVs, while the higher idle rate of capacity of charging stations and the waiting time consumption
chargers during non-peak hours will adversely affect CS of specific charging implementations are neglected. Wireless
constructing investment. Based on the above considerations, charging station enables in-motion charging of the EVs.
charging station structures such as one charger for multiple EVs Therefore, for the application scenarios of wireless charging,
[5] and multiple chargers for multiple EVs [6] are proposed, and wireless charging station planning and route design schemes for
the application case [7] is also built in Shenzhen, China. The electric buses is proposed in [24], and the optimal operation of
research work in this paper will be carried out according to this wireless charging stations considering the coordinated coupling
type of CS. of TN and DN is designed by [25]. However, due to the cost
Traffic flow situations and regulation strategies in the TN and technology level, the promotion of wireless charging
will influence the driving patterns of EVs, which further stations still faces great problems. The authors in [2] propose a
impacts spatial and temporal distributions of electrical loads as traffic-distribution coordination model to minimize the travel
well as the operation of DN. Thus, many corresponding cost of TN and the energy service cost of DN, which
researches for EV navigation to the charging station are simultaneously considers the dynamic optimal power flow and
implemented. An intention-aware routing system for EVs to the EVs dynamic user equilibrium. However, this approach still
minimize their expected journey time is proposed in [8]. is applied with fast charging stations and the flexibility gains to
However, the main influencing factor of journey time is the grid through charging station power scheduling
charging waiting time, thus, the impact of traffic congestion is optimization is neglected.
ignored. In [9], an integrated rapid-charging navigation strategy Different from the currently exist methods, this paper
that considers both the traffic conditions and the status of the proposes a collaborative optimal routing and scheduling (CORS)
power grid is described. But the time-of-use (TOU) electricity method. The framework is shown in Fig. 1. In order in which
price, which can affect user charging satisfaction, is not taken the EVs report charging requirements, the proposed algorithm
into account. Thereby, from the perspective of operating arranges specific navigation and charging schemes for each EV
economy, the dedicated EV routing models that can achieve in turn. The proposed CORS method can not only assign and
traffic and power network equilibrium are proposed [10, 11]. route charging stations for EVs, but also optimize the charging
However, the methods mentioned above are all based on the fast power based on the assigned charging stations. An optimization
charging stations. The fast charging stations in the grid is model that considering EV driving cost, electricity purchase
equivalent to a fixed power load, which means the flexibility cost and battery degradation cost is proposed to find the routing
provided by coordinated charging scheduling schemes is and scheduling scheme with the least comprehensive cost. The
utilized insufficiently. charging facilities limitation in a charging station is also
In the DN system, with activating the flexibility of the considered. In order to solve this complex optimization mode,
charging station, the potential of CSs for optimized operation we split the optimization model into upper layer and lower layer
and demand-side response has grasped the attention from both optimization. The upper layer mainly decides the charging
industry and academia. Nevertheless, a massive access of EVs station assignment process including determine the charging
can impact on the power grid, which will bring problems such station and plan the driving path. Meanwhile the lower layer
as voltage deviations, voltage drops and distribution system solves the coordinate charging scheduling scheme for the EV
losses [12]. Thus, the impact of transport electrification on with the charging station assignment results from the upper
electrical networks is discussed in [13], and the impact indices layer. The modified distributed biased min consensus (DBMC)
for performing charging is proposed in [14]. Vehicle-to-Grid and generalized benders decomposition (GBD) methods are
(V2G) describes a power scheduling system that allows the EVs introduced as the solver. The contributions of this paper are
inject power to or draw power from the grid. V2G can be summarized as follows:
applied in peak shaving and valley filling [15], frequency 1) To the best of our knowledge, most of previous work has
regulation [16], and profits enhancing [17, 18]. Besides, the considered charging station assignment [2, 8-11] and charging
privacy protection when applying the V2G technic is discussed power scheduling optimization [15-18] separately. This paper
in [19]. However, most of the works focusing on charging provides a complete charging service operation scheme that can
power scheduling in grid operation neglect the interaction relieve TN congestion for EV users, improve charging economy
between the TN and DN, and cannot provide charging station and utilization for charging stations and enhance stability for
assignment and navigation schemes that can avoid congestion. DN.
By considering the characteristics of both TN and DN, the 2) The multiple chargers for multiple EVs charging stations
coordinated operation methods are investigated. The combined are applied in this paper for power scheduling. Different from

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
3

the path planning approaches based on the FCSs [2, 11], the of the scheduling flexibility of their dwell time and formulate
flexibility offered by charging power scheduling is exploited in the optimal routing and charging schemes to reduce the overall
our designed system. Meanwhile, the detailed power scheduling cost.
and charger operating process are provided, beyond [18, 26], In order to provide charging service for EV users,
the interaction between TN and DN is considered. corresponding charging demand information of each EV is
3) A bi-layer solving technique is proposed to solve the required. When the n-th EV user has a charging demand and
proposed CORS optimization model. The upper layer is solved reports to the dispatch center, record the report time as trn . The
based on DBMC that can assign and route the optimal path for expected charging time ten , initial and required battery state of
the EV to appropriate charging station. According to the upper charge (SOC) value Sini req
n and Sn , and battery capacity Capn also

results, lower layer optimizes the scheduling strategy by the need to be provided. Let n be the set of unprocessed charging
E’

GBD algorithm. information of EV n, which is reported by the EVs with

4) Different common operation methods are designed as the charging demands in the transportation network:
comparisons with the present CORS method, and the
effectiveness of the proposed method is verified in an integrated

nE '  tnr , tne , Snini , Snreq , Capn  (1)
TN-DN system. Thus, the required energy of n-th EV can be obtained:
EV
Na
v GTN=(ΩT,ΩR)

Enreq  Snreq  Snini Capn  (2)
iga Input e
tio
n Network CORS Dispatch Center In addition, the expected charging time t set by users cannot
n
Structure
Arrange the operating scheme be shorter than the minimum charging time, which is:
according to the order of the EV
tne  Enreq Pmax
Traffic flow ch
report (3)
Power flow
Charging Station Assignment The following section will describe the charging station
EVs' · Determine the appropriate
Information charging station for each EV assignment process of how to distribute appropriate charging
B L · Plan an optimal driving path
GDN=(Ω ,Ω )
for each EV stations k and calculate corresponding arrival time tan for each
individual EV. Thus, the departure time of EV n can be obtained:
Coordinate Charging Scheduling
Output · Operate the connecting status tnd  tna  tne (4)
Assignment between chargers and EVs
Results · Optimize the charging power
of the charging station
Let  be the set of processed charging information of EV n
E
n
Scheduling
Results that needed to participate in scheduling, which can be expressed
g Sc heme
dulin as the union between the unprocessed set and results of station
Sche
assignment process: En ={E’n , Ereq
n , tn , tn , k}.  represents the
a d E
Fig.1. Framework of the proposed CORS approach
set of EV charging Task that  ={1 , 2 ,…, En }. For each
E E E

II. SYSTEM MODEL FORMULATION reported EV, formulate the corresponding charging station
allocation and scheduling plan according to the order of the
In the proposed CORS approach, the optimization object is
report, let Tn be the optimal interval of EV n, which is from trn
to minimize the overall operating costs, including the driving
to t dn . Notice that users can achieve different demands of
time consumption, electricity purchase cost and battery
charging service through the setting of the charging information.
degradation cost. When an EV has charging demand, the driver
For example, users with urgent charging requirement can set
can report instructions to the dispatch center through the mobile
their excepted charging time ten to a smaller value (the minimum
devices. After that, the dispatch center plans optimal route to
value should satisfy (4)), so the dispatch center will finish their
charging station and then perform power charging for the EV
charging as soon as possible. In contrast, users with economic
according to the actual traffic and power grid status.
charging needs can set the relatively longer excepted charging
EV charging users can usually be divided into two types:
time ten . Longer scheduling period brings more flexibility that
users who have urgent charging requirements and hope to
can benefit both charging cost and battery degradation cost.
complete the charging as soon as possible, and users who have
economic charging requirements and are willing to spend a A. Driving Consumption Model
longer time to complete the charging with a lower charging cost. The graph theory is adopted to represent the TN, GTN=(T,
Although the power scheduling is considered in the proposed  ), where T denotes the set of intersections and R is the set
R
CORS algorithm, both of these charging demands can be of roads. The origin–destination matrix is established to
satisfied by the CORS method. The CORS method proposed in indicate the relationship between the intersections of the TN,
this paper provides users with the right to choose. When users where r and s denote the origin and destination intersections
have urgent charging requirements, they can use remote devices respectively. Auv denotes the road between intersection u and v.
such as smart phones to set up that they want to complete the For an EV intends to charge, the driver starts from one origin
charging in the shortest charging time. Then, CORS algorithm intersection r to the destination intersection s, and the EV will
will allocate available the charging station and the charging be charged during the travel. In the TN, the traffic flow varies
station will provide maximum charging power during their in different time periods of a day. Dynamic road congestion
dwell time. For users with not urgent charging requirements, situations can change the optimal path to reach the destination
users can send their charging information with longer expected for each driver [11]. The travel time of different roads can be
charging time, so the proposed CORS method can make full use modeled by the latency function [23, 27]:

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
4

  uv (t )  and reactive power output of generation at bus j. Pload and Qload
tuv uv (t )   tuv
j,t j,t

1  0.15 , u, v  T , Auv   R (5)

0
represent the active and reactive power load at bus j. Pcsj,t is the
  uv 
c
total charging power of a charging station located at bus j at
where u and v represent two connected intersections in TN; t0uv time slot t. Constraints (12)-(15) represent the active and
denotes the free flow traffic time on the road Auv;uv(t) is the reactive power bounds of generations, as well as the voltage
traffic flow at time t and cuv is the capacity of road Auv. Thus, magnitude bounds at bus j. (15) is the apparent power flow
the latency times for roads are also related to the traffic flow at constraint on power line Lij.
different time slots. When driving from the start traffic node u The charging of EVs affects the state of the distribution
to the end traffic node v, there will also be a time difference. network, which is the controllable variable in the proposed
Therefore, in this paper, when calculating the real-time traffic method. After charging stations assignment process finished,
flow, the time when an EV at the start node u is used as the the arrival information of the EVs will be allocated to different
benchmark. charging stations, then the coordinate charging scheduling
Let Trs denotes one path from origin s to destination r, where algorithm will be implemented. The flexibility of existing
rs={Aru, Auv,…, Avs}. T(k) represents the traffic node installed
T
charging facilities is fully tapped and detailed charger operating
with charging station k; Let kn be the charging station assigned scheme can be provided. Let E(k) be the set of EVs allocated
for EV n, gT(kn) be the traffic node installed with charging to charging station k for charging, the charging power of
station kn. Thus, the driving cost CTn for EV n from intersection charging station k located at bus j at time slot t is:
s to charging station at g can be obtained as:   a n ,t Pn ,t j   B k 
 t  t (uv ), u, v   , g   k n  (6) P  n k 
cs
CTn  (uv )  T T E
(16)
j   B k 
j ,t
0
uv uv
Auvrg
R
AuvRgv

where μ is the monetary price of time, which is set as 10$/h. where an,t is the binary variable that represents the connecting
When the EV user intends to charge and sends the charging status of each EV by considering the limitation of chargers,
information to the dispatch center, the arranged charging station which is:
should be within the driving range of its remaining power. 1 if n   E k n  connected to a charger at t
Assume the energy consumption is a linearly decreasing a n ,t   (17)
0 otherwise
function [4], the charging station assignment should satisfy:
Snini Ec Capn  d rg , r  T , u  T k n 
Let M(kn) be the number of chargers in charging station kn,
(7) due to the charging station service capability, only M(kn) EVs
where r is the departure node of EV n; g is the destination node can be charged at the same time:
with charging station; drg denotes the distance between node r
and g; Ec means the electricity consumption per km, which is
 a
n E
kn
n ,t
 M k (k n ) k  C (18)
set as 7 km/kWh [28]. Since EVs cannot connect to chargers when they were not at
B. Power Purchase Model the charging station, the following constraint should be satisfy:
Similarly, the DN could also be represented by graph theory, an ,t  0 , if t  tna , tnd , n   E k n  (19)
GDN=(B, L), where B and L denote the set of buses and the The charging power cannot exceed the maximum charging
set of lines. Let Lij denote the power line connecting bus i and j, power Pmax:
B(i) represents the collection of buses stemming from bus i. Pn,t  Pmax , n   E (20)
The constraints of the AC optimal power flow is given [29]:
The EV cannot be charged when not connected to a charger:
, t  rij I ij, t   Pjz, t  Pj , t
Pij,t  Pjgen  Pjcs,t , Lij   L (8)
2 load

z B
 j Pn,t  0 , if an,t  0, n   E (21)
Qij,t  Q gen
j ,t x I 2
ij ij, t  Q jz, t Q load
j ,t , Lij   L
(9) Energy requirement of each EV should be satisfied:
z B  j 

tnd
tPn ,t  Enreq , n   E
V V2
i ,t
2
j ,t   
 2 rij Pij,t  xijQij,t  r  x I , Lij   (10)
2
ij
2
ij  2
ij,t
L t  t na
(22)
Since the maximum charging power that each charger can
Vi ,2t I ij2,t  Pij2,t  Qij2,t , Lij   L (11) provide is limited, set to Pchmax , the number of chargers in the
gen
Pj
gen
 Pjgen  P j , j   B (12) charging station k is Mk, so the total charging power should
gen follow Pcs ch
j,t ≤Mk Pmax . Therefore, in the charging station
Qj
gen
 Q gen
j  Q j , j   B (13)
assignment model, the capacity limit should also be considered.
V j  V j  V j , j   B (14) Due to the flexibility of scheduling, the remaining charging
margin of the station should satisfy the EVs’ energy demand
Pij2,t  Qij2,t  Sij,t , Lij   L (15) during their dwell time. Assuming that the expected charging
The equations (8)-(11) describe the nodal power balance, power is superimposed by nE(k) EVs staying in the station
where the subscript t represents the value at time t. rij and xij k. For the newly arrived EV n+1, the charging station
mean the impedance and reactance of line Lij. Vj,t is the voltage assignment must follow:
at bus j and Iij,t is the current on line Lij. Pij,t and Qij,t denote the
active and reactive power on line Lij. Pgen gen
j,t and Qj,t are the active

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
5

t nd1 Constraints of power grid (8)-(15)

d a

1  t  Pj ,t  t t n 1  t n 1  1 M k Pmax
Enreq cs ch
 (23) Constraints of provided charging power (16)-(22)
t  t na1

The objective function of the AC optimal power flow (OPF) III. CORS SOLVING ALGORITHM AND SPECIFIC
problem is formulated [2, 30]: IMPLEMENTING PROCESS

CDn  min    a j Pjgen
tT  j
,t

2  bj Pjgen,t  t  P1 j,t  (24)  It can be found that the established model is a very complex
n
B
j (1)  B problem, including shortest path planning and optimal power
The energy cost in the distribution grid CDi including two flow solution containing integer decision variable kn, binary
parts. The first part is the generation cost and the second part is decision variables an,t and decision variable Pn,t. Let
the purchasing cost for buying electricity from the main grid. an={an,t|tTn} be the vector of binary decision variables and
This paper focuses on the impact of charging power on the grid, Pn={Pn,t|tTn} be the vector of decision variables. By
so the controllable components in the distribution network only deconstructing the optimization model, we can find that the
consider the charging power of EVs. The generators in the diving consumption CTn is only affected by kn. Meanwhile, the
distribution network are set as photovoltaic (PV) sources, so the different assignment results kn correspond to their respective
operating cost can be ignored. λt is the electricity price at time charging station scheduling results an and Pn. In contrast, when
t. P1j,t is the purchased power that derived from bus 1, which the value of kn is fixed, the optimization model (27) becomes a
can be modeled as a dummy generator at bus 1. After assigning mixed integer nonlinear programming (MINLP) h(Pn,an)
a charging station to this user, the charging demand Pcsj,t of the (deformation of f(kn,Pn,an) when CTn is regarded as a constant),
charging station will be update by (16). which usually solved by the generalized benders decomposition
Notice that when power scheduling is performed on the n-th (GBD) [34] algorithm.
EV, the EV reported before it has completed the charging Therefore, when only kn is considered, we can equate it to a
station assignment and optimal power scheduling process, shortest path planning problem, and then find the shortest path
which means {a1,t ,⋯,an-1,t } and {P1,t ,⋯,Pn-1,t } are fixed. to each different charging station and corresponding arrival
time. Then, by solving the problem model h(Pn,an), we can
C. Battery degradation model obtain the optimal scheduling method under all different
The LiFePO4 lithium-ion is topical and widely used battery charging station assignment kn, and then find out the optimal
material in automobile industries, with more thermal and charging station assignment plan kn with its corresponding
chemical stability physical characteristics [31]. A degradation power scheduling scheme an and Pn.
cost model for LiFePO4 battery cells is developed in [32], Therefore, the solving algorithm is divided into two layers.
which is related to the EVs’ charging power Pn,t at time slot t The upper layer solves the charging station assignment problem
and its battery capacity Capn: that regards kn as the decision variable. Lower layer considers
Φn Pn,t   1 ( Pn,t Capn ) 2   2 Pn,t Capn  3 (25) the kn as a fixed value and solves the an and Pn. The solving
technique of these two levels are introduced separately in the
where α1, α2 andα3 are parameters. Specifically, based on the
next chapters, and the combined solving approach is
price of battery cell ($/kWh) and battery’s internal parameters
comprehensively introduced. In addition, the detailed
such as voltage, current, and SoC, all of these three parameters
application process of the proposed CORS in the reality
can be determined. The calculating of the parameters is given
situation is also introduced.
in [33, 32]. Thus, the battery degradation cost can be calculated:
CBn  min  a n,t Φn Pn,t  (26) A. Charging station assignment upper layer solver based on
tTn distributed biased min consensus
Appropriate charging method can reduce battery Assigning and navigating an appropriate charging station for
degradation and improve battery life. By setting longer EVs is a shortest path optimization problem. Each EV has
excepted charging time t en allows EVs to avoid high-power individual optimal path and charging station selections, which
charging in a short period of time, which is benefit to the battery are determined not only by the time-series traffic flow but also
life. Users can set it according to their own charging by the charging station status. Therefore, inspired by [11, 35],
requirements. the distributed biased min-consensus (DBMC) algorithm that
can achieve the navigation results in a short time period is
D. Collaborative optimal routing and scheduling model adopted and corresponding adjustments are made to the
The spatial and temporal characteristics of the TN and the characteristics of our model. For the case that EV n is assigned
DN impact the assignment and scheduling efficiency of the EV. to kn, the weight of each edge of the graph GTN can be obtained
Therefore, for every EV with charging requirement, we orderly by:
establish the comprehensive optimization model that considers t   , u  T kn , v   T u 
driving cost, charging cost and battery degradation cost. buvn   uv uv (28)
tuv uv    uv , u   kn , v   u 
n T T

f k n , Pn , an   min  CT n  CDn  CBn  (27)

inf , nE  constriant (5), (21)
k n , a n ,t , Pn ,t
tTn
 uvn   (29)
subject to:
0 , nE  constriant (5), (21)
Constraints of charging station assignment (7), (23)
where T(k) is the set of traffic nodes installed with charging

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
6

station kC. T(u) is the set of neighboring nodes of traffic calculated arrival time could at any time. By considering the
node u.  nuv is auxiliary parameter, indicating whether the energy requirement constraint (22), the energy requirement can
planned path for EV n meets the remaining battery capacity be satisfied during the power scheduling period from tan to tdn .
constraint (7) and charging station margin constraint (23). However, if the actual stay time of the EV is shorter than the
For each EV user n, the distributed biased min consensus power scheduling period, it may cause the incomplete charging.
algorithm is implemented to find the optimal driving path with In contrast, if we postpones the arrival time tan to the next time
the minimum driving time: slot, the power scheduling period from tan to tdn will be at least
Cu q  1  Cu 0 , u  T k  shorter than the actual stay time of the EV, which will also
C q  1  min
 u v u 
Cv q  buv  , u  T k  (30)
T
n satisfy the energy requirement during the actual stay time of the
EV. Therefore, this paper postpones the arrival time tan to the
where Cu(q) is the state value of node u and k means the number next time slot, which can leave margin for EV scheduling.
of iterations. The traffic nodes with charging station are the Equation (35) indicates that the arrival time t an for power
leader nodes while the others are follower nodes. scheduling will be the nearest time slot after the actual
Then, all nodes of the graph globally and asymptotically calculated arrival time.
converge to the equilibrium point C *u , which satisfies the
B. Coordinate charging scheduling Lower layer solver based
following equation:
on generalized benders decomposition
Cu 0 , u  T k 
Cu*   According to the previous discussion, each fixed kn value
T
* n

minv u  Cv  buv , u   k 
T
 (31)
corresponds to a MINLP problem h(Pn,an) which can be solved
The proofs of the optimality of (30) and the convergence of by the GBD algorithm. Here, define the variable z=[Pn,an].
(31) are described in [11, 35, 36]. Then, after the equilibrium To adopt GBD, the optimization problem h(Pn,ân) should be
point of each traffic node is found, the parent nodes can be a convex optimization problem with variable P, where â
obtained: denotes the vector of fixed an,t to any feasible region. The object
, u  T k 
is convex if the Hessian matrix of h(Pn,ân) was positive.

 (u )  
arg minv u  Cv  buv  , u   k 
n T
(32) Equation (36) and (37) describes the first-order and second-
T
order partial derivatives of h(Pn,ân).
h Pn ,t , aˆ n ,t  
 aˆ n,tΦn Pn,t   CDn  CTn 
Then, the optimal navigation path to the determined charging

station is discovered by finding the parent nodes in a recursively Pn ,t Pn ,t tT
manner. Let Trs (n)={r→u→⋯g→⋯v→s} be the optimal route
n
(36)
 21 2 
for EV n form origin r to the destination s while charging at ˆ 
 a n ,t  Pn ,t  
gT(kn), thus the driving cost under this kn can be obtained:  Capn Capn 
CTn   tuv uv  (33) h 2 Pn ,t , aˆ n ,t     21  
u ,vTrs n   aˆ n ,t  Pn ,t  2 
P 2
n ,t Pn ,t   Capn Capn 
As mentioned above, the time when an EV at the start node
21 (37)
u is used as the benchmark when calculating the real-time traffic 
aˆ n ,t , diagonal elements
flow uv(t). However, the calculated driving time by (3) can be  Capn
any value, but the traffic flow data provided in this paper is 
0 , off - diagonal elements
based on the time slot’s resolution. Therefore, the real-time
traffic flow used when calculating from u to v depends on the It can be found that the Hessian matrix is a diagonal matrix.
traffic flow of the closest time slot when the vehicle is at node For the values on the left side of the diagonal elements: α1 is a
u, which can be formulated as: constant related to the battery, which is larger than 0 according
uv (tslot
bef
) if t  tslot
bef
 t  tslot
aft
to [32]; ân,t is the fixed binary value; Capn is the battery capacity
uv (t )   (34) that can only be a positive value. Therefore, the values of the
uv (tslot ) if t  tslot  t  tslot
aft bef aft

second-order partial derivatives of h(Pn,ân) are always greater

where tbefslot means the nearest time slot before t, tslot means the
aft
than or equal to zero, so the Hessian matrix is positive semi-
nearest time slot after t, and t right here represent the time when definite.
EV at traffic start node u. In constraints, since the kn is fixed, the constraints, (7) and
Let Trg (n)={r→u→⋯g} be the optimal route for EV n from (23), of charging station assignment can be ignored. Among the
origin r to the node of charging station gT(kn), thus the remaining constraints, the nonconvex situation is caused by
arrival time can be computed: (11), which can be relaxed as [37]:
  Vi ,2t I ij2,t  Pij2,t  Qij2,t , Lij   L (38)
tna  tnr   tuv uv  (35)
 u , v n  T  Accordingly, the formulated convex optimization problem
rg
slot
when charging station assignment kn is fixed can be obtained:
where ⌈*⌉slot means the function that to find the nearest time slot
after *. min han , Pn   min  CT n  CDn  CBn  (39)
a n ,t , Pn ,t
tTn
The power scheduling part is also based on each time slot
(Pn,t is the charging power of EV n at time slot t), but the actual s.t. (8)-(10), (12)-(22), (38)

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
7

Let g(P,a)≤0 be the set of constrains of the proposed in Algorithm 1, where u is the vector of Lagrange multiplier
optimization model (39), the detailed solving process is shown and s is the introduced relaxation variable. Combining the
Algorithm 1: GBD solving process for the optimal scheduling model
charging station assignment results in previous chapter and the
Input: E, GTN, GDN, C, Pmax
charging scheduling method in this chapter, the collaborative
Step 1: Number of iterations: q←0; Lower boundary: LB0←−∞; Upper operation of TN and DN are obtained.
boundary: UB0←+∞; â0(24)-(25) is the arbitrary but fixed vector
Step 2: LBq < UBq do
C. Comprehensive solving algorithm and real-time operation
Step 2: Primal problem: mode
min h(P, âq) To better illustrate the implement of the proposed CORS,
s.t. g(P, âq)≤0
if Primal is feasible
detailed solving and operating process is shown in Fig. 2. The
Solve the relaxed problem: state of the system is updated as a cycle is executed. According
min L(P,âq,u)= h(P,âq)+uTg(P,âq) to the order of EVs report, the specific routing and charging
Obtain: {Pq, uq}←{P*, u*} schemes for each EV is designed in turn in every execution
Update: UBq←min{UBq, h(Pq,âq)}
Add cut: h ≥L(Pq,âq,uq)+∇aTL(Pq,âq,uq) (a− âq)
cycle. First, the algorithm will update the status of the TN and
else DN based on the CORS results of the previous EV n−1. Among
Solve the new primal problem: them, TN is to update the path and charging station selected by
min s the EV n−1, and DN is to update the charging power of the
s.t. g(P, a) − s≤0; s≥0
Obtain: {Pq, uq}←{P*, u*}
charging station Pchj,t by (16). It should be noted that the proposed
Update: UBq ← UBq−1 CORS algorithm takes the formulation of one EV’s charging
Add cut: 0 ≥uqT [g(Pq, âq)+ ∇aTg(Pq, âq) (a− âq)] scheme as a complete execution cycle, and the system state will
end if be updated after each execution cycle. That is, in (16), only an,t
Step 3: Solve master problem:
min h(Pq,a)
and Pn,t are the decision variables. {a1,t.,…, an-1,t} and{P1,t,…,
s.t. cuts Pn-1,t} are the known values that are calculated in the previous
Obtain:{ hq,âq+1}←{ h *,a*} execution cycle. When then EV n reported its charging demand
Update: LBq ← fq  E’n , the travel distance constraint (7) and the remaining
Step 4: if LBq < UBq
q←q+1
capacity constraint (23) can be obtained, then the corresponding
jump to Step 2 weight of each edge of the graph GTN can be computed by (28)
else and (29). Next, compare the optimal result (27) under all the
Problem convergence different charging station selection condition. By activating the
Output: {Pq, uq, hq }
end if
DBMC algorithm, to get the optimal route to the charging
End station kn for EV n and the corresponding driving cost CTn can

Network status processing Optimization model solving algorithm

CORS results Traffic flow update

EV n reports its charging For kn=1
of EV n−1

assignment upper layer scheduling Lower layer

ηuv={ηuv+1|u,v∈ΩrsT (n)} demand ΩnE’

Charging station
Power flow update Active distributed biased min consensus
cs cs algorithm (30),(31),(32) to find the optimal
Pj,t={Pj,t+(an−1,t·Pn−1,t)
routing ΩrsT (n) and CTn under current kn
a
|t∈[tn-1 d
,tn-1 ], j∈ΩT(kn-1)}

Arrival time tna calculation based on (35)

Charging station assignment constraints
Remaining capacity Travel distance
Coordinate charging

constraint (23) constraint (7) Coordinate charging scheduling (39)

Solver: Algorithm 1
Operating scheme

Operating scheme

Operating scheme

Calculate weight of each edge of the

Connecting Charging
for: EV n−1

for: EV n+1

graph GTN by (28),(29)

status: an power: Pn
for: EV n

Compare the optimal results under

different charging station kn selection Obtained corresponding CDn and CBn

Record the optimal results N

Min f >
Optimal Routing of Optimal charging of CTn+CDn+CBn?
EV n: kn and ΩrsT (n) EV n: an and Pn
Y
Navigation Scheduling Let Min f = CTn+CDn+CBn
Destination

kn= kn+1
...

...
...

...
...
...

Y N
...

Terminate?
Origin CORS results
of EV n
Update TN and DN situation based on
the CORS results of EV n

Fig. 2. The flow chart of the proposed collaborative EV routing and charging scheduling process

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
8

be found. Then the time tan when EV n arrived at the charging 1 12

station kn is found by (35), with its corresponding power
scheduling period from tan to tdn . Thereby, solve the coordinate CS2
charging scheduling scheme under this kn condition by CS4
Algorithm 1, and obtain the time-series connecting status an and 4 5 6 7 8
charging power Pn of EV n. Meanwhile, the electricity purchase
cost and battery degradation cost are also computed. By CS5 CS3
comparing the driving, electricity purchase and battery
degradation costs under all different kn conditions, the kn with 9 10 11 2
the lowest cost is selected as the assigned charging station for CS1
EV n, and the corresponding scheduling scheme an and Pn are
adopted for optimal charging after it arrives at the 13 3
corresponding charging station. Finally, update the TN and DN
Nguyen–Dupius Transpiration Network CS1
situation according to the CORS results of EV n for the next
execution cycle. 1 2 3 4 5 6 7 DG4
This method will arrange the specific routing and charging
18
scheduling method for each EV that uploads its charging
demand, and continuously update the grid status with the CORS 8 17
CS319 23
results of finished EVs to adjust the operation plan of the
CS4 26
subsequent EVs. Therefore, it not only avoids the overcrowding CS2 9 16
20 24
of EVs in some charging stations to reduce the insufficient 27
scheduling margins, which leads to an increase in charging 10 15
21 25
costs, but also increases the utilization rate of charging facilities. 28
DG3 11 14
22 DG2
IV. CASE STUDY 29
DG1 12 13
A. Case Overview and Basic Settings 33 32 31 30
In this paper, we consider the Nguyen–Dupuis TN CS5
IEEE 33-bus Distribution Network
interaction with IEEE 33-bus DN to illustrate the proposed
Fig. 3. Topology of the interacted TN and DN system
routing and scheduling method. Nguyen–Dupuis network is a
traffic network model developed by [38] and is widely used in
various studies related to transportation. The parameters and
traffic flow of TN are adopted in [2] and appropriate
adjustments are made. The top half of Fig. 3 displays the
topology of the TN with the installation location of the charging
station. The basic data of the original IEEE 33-bus network can
be found in [39], with extra distributed generations (DGs) and
charging stations, the topology of the DN is shown in the Fig. 4. Overall Traffic flow and TOU electricity price.
bottom half of Fig. 3. In total there are 5 charging stations in the
case study and let the number of in charging stations be the
same: Mk=6, the maximum charging power provided by each
charger is Pmax=45kW. The time slot duration is set as 15 minute.
Since time-series is taken into account in this paper, the
overall traffic flow variation according to the actual data in one
day is shown in Fig. 4, along with TOU electricity price [40]
purchase from main grid. The vehicles depart from starting
point 1 or 4, then navigating appropriate charging stations for
EVs with charging demands. After EVs finish charging, the Fig. 5. Profile of the DGs and different types of loads.
algorithm will navigate from the corresponding charging In order to demonstrate the effect of the proposed
stations to the ending point 2 or 3. Number of EVs with Collaborative Optimal Routing and Scheduling (CORS)
charging demand in the traffic flow are set to 0.1% It includes method, we consider the Nearby Routing with Optimal
both users with urgent charging requirements and users with Scheduling (NROS), Optimal Routing with Uncontrolled
economic charging requirements. Fig. 5 shows the DGs and Scheduling (ORUS) and the Nearby Routing with Uncontrolled
loads (not including EV charging loads) variation in DN. The Scheduling (NRUS) as the comparisons. In the NROS approach,
DGs and loads profile adopted in the simulation are the real there is no unified dispatching center to arrange and navigate
daily characteristic curve DGs represent photovoltaic daily the charging station for EVs, EVs will only the nearest charging
output characteristics. The load contains four typical profles station by themselves. Then the charging stations will
and is evenly distributed to the nodes in the distribution network. independently optimize the charging power scheduling to

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
9

reduce the overall costs. The ORUS can assignment the is closer to the starting node at the TN. Thus, unless the capacity
charging station for each EV according to the state of power limit of CS2 is reached, the EVs with nearby routing will intend
grid and traffic network. After the EVs arrive at the charging to charge at this charging station. CS3 is at the end node of TN
station, the first come first charge mode is adopted to charge the while at the upstream node 19 in the DN. Therefore, with the
EV with the fixed maximum charging power [11, 2]. In contrast, same charging power, the power loss when charge at CS3 is
NRUS approach represents a disordered state that EVs are find smaller. Thereby, in the optimal routing mode, CS3 is assigned
the nearest charging station by themselves and charging in the to more charging EVs, while CS2 is assigned fewer EVs
station without any optimal scheduling technic. compared with nearby routing.
B. Simulation Results
The simulation results of the designed case is shown in Table
I. Obviously, under the proposed CORS method, the charging
cost of the entire system and the total driving cost of the EVs
are the smallest. Voltage mean square error represents the
square error to which the voltage of all nodes deviates from the
rated voltage at different times, where the rated voltage is set as
1 p.u. Due to the high ratio of R/X of the distribution network,
the larger voltage deviation will lead to higher grid loss, which
Fig. 6. The number of EVs assigned to each charging station with and without
is also reflected in Table I. It can be found that by CORS, less
optimal routing.
voltage mean square error is obtained, along with the smallest The flow of EVs with charging demand under different
grid loss. NROS method ignores the routing planning process, routing methods at 10 a.m. is illustrated in Fig. 7. When
thus the driving cost is significantly higher than CORS and implement the nearby routing, each independent EV ignores the
ORUS. Because the scheduling process is implemented current traffic status and will only choose the route with shortest
independently at each charging station, the overall charging distance, thus ignoring the traffic congestion. By optimal
cost of NROS is relatively lower than ORUS and NRUS. routing, traffic jams can be avoided. For instance, the roads 4-
Adopting ORUS brings less driving cost but higher charging 9, 5-6, 9-13 take more driving time, so the optimal routing
cost. This is because the optimal routing process reduce the method reduces the number of EVs on these roads, thereby
driving cost while EVs adopt the first-come first-charge after reducing the overall driving cost. Through the above discussion,
arriving at the station method. Therefore, the flexibility of the it can be seen that the optimal routing process comprehensively
charging station is not fully utilized. Similarly, the voltage considers the operation status of the DN and the TN, which not
mean square error and grid loss are also higher due to the only reduces the voltage fluctuation of the power grid but also
neglecting of charging power scheduling. CORS as the improves the transportation efficiency.
comparison without route and power optimization, all Road travel
indicators under this method are better. Battery saving indicator time (min)
209 Traffic flow with optimal routing
represents the ability of the algorithms to cope with battery 1 12
Traffic flow with nearby routing 35
degradation. The battery saving indicator shown in Table I
means the battery cost saving rate of different algorithm 231 22 CS2 143 CS4 66
comparing with the uncontrolled scheduling situation. It can be 28
148 97 240 84
found that for algorithms with optimal power scheduling 4 5 6 7 8
83 314 314 152
(CORS and NROS), the battery saving rate is significantly
higher than that with uncontrolled scheduling (ORUS and 30 73 20
162 156 152 150
NRUS). This means the battery life with CORS and NROS will 95 CS5

be longer. 8 8 7
9 10 11 2 12
Table I Simulation results under different algorithms 5
CS1 87
Total Voltage mean 95 157 165
Total driving Battery CS3
charging cost square error Grid loss (%)
cost (€) Saving (%) 4
(€) (p.u.) 13 87 3
CORS 2077 3454.8 0.0014 4.05% 7.03% 95
NROS 2133 4289.7 0.0014 4.06% 7.11% Fig. 7. Status of the TN and the flow of EVs with charging demand under
different routing methods at 10 a.m.
ORUS 2349 3454.8 0.0016 4.10% 0%
The heat-map of power provided by each charger of each
NRUS 2388 4289.7 0.0018 4.13% 0%
charging station at different times is shown in Fig 8. Because
To verify the effect of the route optimization, the charging the power scheduling process is included in CORS and NROS,
station assignment results is shown in Fig. 6. The optimal the charging power distribution throughout the day is more
routing is considered in ORUS and our proposed CORS while extensive. By contrast, the ORUS and NRUS methods are
nearby by routing is adopted in NROS and NRUS. Comparing executed with first come first charge mode, which leads to a
with nearby routing, the optimal routing method can divert the relatively concentrated charging power distribution in the heat-
overcrowded charging stations to ensure the flexibility margin map. The working status of the charging station is closely
for power scheduling. For instance, according to Fig. 3, the CS2

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
10

related to the traffic flow, and the high charging power cannot
avoid periods of high electricity prices. Therefore, by CORS
and NROS methods, the flexibility brought by the dwell time of
electric vehicles has been fully exploited, and the charger
utilization rate of each charging station has also been improved.
CORS Charging
Power
CS1
(kW)

CS2 45

CS3

CS4
40
CS5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time
NROS
35
CS1

CS2
Fig. 9. Charging power comparisons under different charging methods (a)
CS3 30 overall charging power at different time slot and the TOU price (b) cumulative
charging cost
CS4

CS5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Time
ORUS
CS1

20
CS2

CS3

CS4
15

CS5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time
NRUS 10
CS1

CS2

5
CS3

CS4

CS5
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time
Fig. 8. Heat-map of the provided energy at different times
Fig. 9 shows the charging power comparisons under
different charging methods. From Fig. 9 (a), the charging
methods CORS and NROS reduce the charging power when the
TOU price is relatively higher, such as from 9:00-12:00. And
can dispatch peak charging power to the time slot when the Fig. 10. Voltage profiles of the DN at different charging methods
TOU price is lower. The cumulative charging cost at different The voltage profiles of the DN at different charging methods
time slot is shown in fig. 9 (b). Obviously, the charging costs of is illustrated in Fig. 10. It can be seen that the grid voltages are
charging methods with scheduling process (CORS and NROS) mainly affected by the operation status of the power grid. In the
are lower than with uncontrolled charging process (ORUS and daytime, the output of the PV power supply increases, which is
NRUS). Due to the consideration of optimal routing process, the main reason for the voltage to rise. PV cannot supply energy
the assignment of charging stations selection is more at night while the residential electricity load increases, which
appropriate, so the cumulative charging cost under CORS is causes the voltage to decrease. Even so, the voltages are staying
lower than NROS while charging cost under ORUS is lower within the regular limits under all operating methods. The
than NRUS. maximum and minimum voltage under the proposed CORS in

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
11

Fig. 11. Boxplot of (a) driving cost with different traffic scenarios (b) charging cost with different traffic scenarios (c) battery saving rate with different traffic
scenarios (d) driving cost with different energy demand scenarios (e) charging cost with different energy demand scenarios (f) battery saving rate with different
energy demand scenarios
one day is also marked in this figure. In comparison, the charging cost and battery saving under different methods with
proposed CORS algorithm effectively reduces the range of different energy demand scenarios. Similarly, let the overall
voltage fluctuations, so it has a positive effect on maintaining traffic flow unchanged, set the energy demand to 20%, 40%,
the stability of the power grid and reducing grid losses. 60%, 80% and 100% of the energy demand in previous case
study as different scenarios. It can be seen that the driving cost
C. Scene analysis based on Monte Carlo algorithm
also increases with the increase in charging demand. This is
The above part mainly discusses various technical indicators because the overall number of EVs with charging demand has
in a specific case. To verify the universality of the proposed increased. The driving cost of the proposed CORS is relatively
algorithm, set up different scenarios to conduct Monte Carlo lower than NROS and NRUS. And the charging cost
analysis on cost indicators. Traffic flow status has a significant undoubtedly increase with the energy demand increased. But
impact on path planning. Therefore, let the overall charging with the CORS method, the cost increase rate is smaller than
demand unchanged (the number of EVs charged in a day and other methods. As mentioned before, the battery saving rate of
their required power are fixed), set the traffic flow to 20%, 40%, ORUS and NRUS with uncontrolled control are zero. CORS
60%, 80% and 100% of the total traffic flow in Fig. 4 as and NROS have similar battery savings rates, and they both
different scenarios. For each scenario, repeat the simulation decrease as energy demand increases. This because the number
1000 times, and record the driving and charging cost of each of charging EVs increased, and more chargers have to be
time. The boxplot reveals the cost and battery saving rate occupied, resulting in the reducing of the scheduling flexibility.
distribution of different methods under different scenarios, Combining the Monte Carlo analysis of the two situations, the
which is shown in Fig. 11 (a)-(c). It can be seen that with the proposed CROS method is in an advantageous position in terms
increase in traffic flow, driving cost have increased of reducing driving and charging cost and increasing the battery
significantly. When the number of traffic is small, the effect of life.
optimal routing (CORS and ORUS) on driving cost reduction is
not obvious, and the greater the overall traffic flow, the stronger V. CONCLUSION
the superiority of optimal routing. In contrast, the impact of
Considering the coupling mechanism of transportation
traffic flow on charging costs is small, and the proposed CORS
network and distribution network, the CORS method to
method is still the optimal. Similarly, traffic flow effects less
improve the economic operation of EV charging is proposed in
influence on the battery saving rate, this is because the number
this paper. By considering the driving cost, electricity purchase
of charged EVs and their required energy is fixed. The battery
cost and battery degradation cost, the optimal charging station
saving rate is the ratio of battery degradation that can be saved
assignment and navigation plan and coordinate charging
compared with uncontrolled charging, because the ORUS and
scheduling scheme are calculated for each EV in turn. By
NRUS are adopted uncontrolled charging strategy, their battery
receiving the charging demand information sent by each EV
saving rate in any case are zero. The optimization scheduling is
user to the dispatching center, the proposed CORS algorithm
executed in CORS and NROS, so the battery saving rate are
can formulate a specific optimal charging scheme for each EV.
higher.
The multiple chargers for multiple EVs charging station, which
Fig. 11 (d)-(f) illustrated the boxplot of driving cost,
can tap the scheduling flexibility, is applied, and corresponding

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2022.3142256, IEEE
Transactions on Power Systems
12

power scheduling and charger operating process are provided. [19]D. Abbasinezhad-Mood, A. Ostad-Sharif, S. M. Mazinani and M.
Nikooghadam, "Provably Secure Escrow-Less Chebyshev Chaotic Map-
By introducing the proposed modified combination of DBMC Based Key Agreement Protocol for Vehicle to Grid Connections With
and GBD algorithm, the established complex nonlinear Privacy Protection," IEEE Trans. Ind. Electron., vol. 16, no. 12, pp. 7287-
optimization problem is solved, and real-time implementing 7294, Dec. 2020.
[20]J. Xiong, K. Zhang, Y. Guo and W. Su, "Investigate the Impacts of PEV
process is provided. By introducing the NROS, ORUS and
Charging Facilities on Integrated Electric Distribution System and
NRUS methods as comparisons, the advantages of the proposed Electrified Transportation System," IEEE Trans. Transp. Electrification,
CORS method that can reduce the driving and charging cost, vol. 1, no. 2, pp. 178-187, Aug. 2015.
improve the utilization rate of chargers, mitigate voltage [21]K. Hou, X. Xu, H. Jia, X. Yu, T. Jiang, K. Zhang and B. Shu, "A Reliability
Assessment Approach for Integrated Transportation and Electrical Power
violations, and decrease the grid loss are demonstrated. Systems Incorporating Electric Vehicles," IEEE Trans. Smart Grid, vol. 9,
no. 1, pp. 88-100, Jan. 2018.
REFERENCES [22]W. Wei, S. Mei, L. Wu, M. Shahidehpour and Y. Fang, "Optimal Traffic-
Power Flow in Urban Electrified Transportation Networks," IEEE Trans.
[1] "Wikipedia," [Online]. Available: Smart Grid, vol. 8, no. 1, pp. 84 - 95, Jan. 2017.
https://en.wikipedia.org/wiki/Carbon_neutrality. [Accessed 24 5 2021]. [23]M. Alizadeh, H.-T. Wai, M. Chowdhury, A. Goldsmith, A. Scaglione and
[2] G. Sun, G. Li, S. Xia, M. Shahidehpour, X. Lu and K. W. Chan, "ALADIN- T. Javidi, "Optimal Pricing to Manage Electric Vehicles in Coupled Power
Based Coordinated Operation of Power Distribution and Traffic Networks and Transportation Networks," IEEE Trans. Control. Netw. Syst., vol. 4,
With Electric Vehicles," IEEE Trans. Ind. Appl., vol. 56, no. 5, pp. 5944 - no. 4, pp. 863-875, Dec. 2017.
5954, Oct. 2020. [24]Y. Jin, J. Xu, S. Wu, L. Xu and D. Yang, "Enabling the Wireless Charging
[3] X. Wang, M. Shahidehpour, C. Jiang and Z. Li, "Coordinated Planning via Bus Network: Route Scheduling for Electric Vehicles," IEEE trans.
Strategy for Electric Vehicle Charging Stations and Coupled Traffic- Intell. Transp. Syst. , vol. 22, no. 3, pp. 1827-1839, Mar. 2021.
Electric Networks," IEEE Trans. Power Syst., vol. 34, no. 1, pp. 268 - 279, [25]S. D. Manshadi, M. E. Khodayar, K. Abdelghany and H. Üster, "Wireless
Jan. 2019. Charging of Electric Vehicles in Electricity and Transportation Networks,"
[4] W. Yang, W. Liu, C. Y. Chung and F. Wen, "Joint Planning of EV Fast IEEE Trans. Smart Grid, vol. 9, no. 5, pp. 4503-4512, Sept. 2018.
Charging Stations and Power Distribution Systems With Balanced Traffic [26]J. Liu, G. Lin, S. Huang, Y. Zhou, Y. Li and C. Rehtanz, "Optimal Logistics
Flow Assignment," IEEE Trans. Ind. Electron., vol. 17, no. 3, pp. 1795- EV Charging Scheduling by Considering the Limited Number of
1809, Mar. 2021. Chargers," IEEE Trans. Transp. Electrification, vol. 7, no. 3, pp. 1112-
[5] H. Zhang, Z. Hu, Z. Xu and Y. Song, "Optimal Planning of PEV Charging 1122 Sept. 2021
Station With Single Output Multiple Cables Charging Spots," IEEE Trans. [27]A. Moradipari and M. Alizadeh, "Pricing and Routing Mechanisms for
Smart Grid, vol. 8, no. 5, pp. 2119-2128, Sept. 2019. Differentiated Services in an Electric Vehicle Public Charging Station
[6] H. Chen, Z. Hu, H. Luo, J. Qin, R. Rajagopal and H. Zhang, "Design and Network," IEEE Trans. Smart Grid, vol. 11, no. 2, pp. 1489-1499, Mar.
Planning of a Multiple-Charger Multiple-Port Charging System for PEV 2020.
Charging Station," IEEE Trans. Smart Grid, vol. 10, no. 1, pp. 173 - 183, [28]P. Sadeghi-Barzani, A. Rajabi-Ghahnavieh and H. Kazemi-Karegarc,
Jan. 2019. "Optimal fast charging station placing and sizing," Appl. Energy, vol. 125,
[7] "OFweek," [Online]. Available: https://power.ofweek.com/2020-12/ART- pp. 289-299, Jul. 2014.
35009-8110-30472922.html. [Accessed 24 5 2021]. [29]S. Mhanna, G. Verbič and A. C. Chapman, "Adaptive ADMM for
[8] M. M. d. Weerdt, S. Stein, E. H. Gerding, V. Robu and N. R. Jennings, Distributed AC Optimal Power Flow," IEEE Trans. Power Syst., vol. 34,
"Intention-Aware Routing of Electric Vehicles," IEEE trans. Intell. Transp. no. 3, pp. 2025 - 2035, May. 2019.
Syst., vol. 17, no. 5, pp. 1472-1482, May. 2016. [30]Q. Peng and S. H. Low, "Distributed Optimal Power Flow Algorithm for
[9] Q. Guo, S. Xin, H. Sun, Z. Li and B. Zhang, "Rapid-Charging Navigation Radial Networks, I: Balanced Single Phase Case," IEEE Trans. Smart
of Electric Vehicles Based on Real-Time Power Systems and Traffic Data," Grid, vol. 9, no. 1, pp. 111-121, Jan. 2018.
IEEE Trans. Smart Grid, vol. 5, no. 4, pp. 1969-1979, Jul. 2014. [31]X. Tan, G. Qu, B. Sun, N. Li and D. H. K. Tsang, "Optimal Scheduling of
[10]W. Wei, L. Wu, J. Wang and S. Mei, "Network Equilibrium of Coupled Battery Charging Station Serving Electric Vehicles Based on Battery
Transportation and Power Distribution Systems," IEEE Trans. Smart Grid, Swapping," IEEE Trans. Smart Grid, vol. 10, no. 2, pp. 1372-1384, Mar.
vol. 9, no. 6, pp. 6764 - 6779, Nov. 2018. 2019.
[11]X. Shi, Y. Xu, Q. Guo, H. Sun and W. Gu, "A Distributed EV Navigation [32]Z. Ma, S. Zou, L. Ran, X. Shi and L. A. Hiskens, "Efficient decentralized
Strategy Considering the Interaction Between Power System and Traffic coordination of large-scale plug-in electric vehicle charging," Automatica,
Network," IEEE Trans. Smart Grid, vol. 11, no. 4, pp. 3545-3557, Jul. vol. 69, pp. 35-47, Jul. 2016.
2020. [33]J. Forman, J. Stein and H. Fathy, "Optimization of dynamic battery
[12]J.-M. Clairand, J. Rodríguez-García and C. Álvarez-Bel, "Assessment of paramter characterization experiments via differential evolution," in
Technical and Economic Impacts of EV User Behavior on EV Aggregator American Control Conference, Washington, DC, USA, Jun. 2013.
Smart Charging," J. Mod. Power Syst. Clean Energy, vol. 8, no. 2, pp. 356- [34]A. Geoffrion, "Generalized Benders Decomposition," J. Optimiz. Theory
366, Mar. 2020. App., vol. 10, no. 4, pp. 237-260, 1972.
[13]K. J. Dyke, N. Schofield and M. Barnes, "The Impact of Transport [35]Y. Zhang and S. Li, "Distributed Biased Min-Consensus With Applications
Electrification on Electrical Networks," IEEE Trans. Ind. Electron., vol. to Shortest Path Planning," IEEE Trans. Automat. Contr., vol. 62, no. 10,
57, no. 12, pp. 3917-3926, Dec. 2010. pp. 5429 - 5436, Oct. 2017.
[14]M. S. Islam, N. Mithulananthan and K. Y. Lee, "Development of Impact [36]Y. Zhang and S. Li, "Perturbing consensus for complexity: A finite-time
Indices for Performing Charging of a Large EV Population," IEEE Trans. discrete biased min-consensus under time-delay and asynchronism,"
Veh. Technol., vol. 67, no. 2, pp. 866-880, Feb. 2018. Automatica, vol. 85, pp. 441-447, Nov. 2017.
[15]A. Sangswang and M. Konghirun, "Optimal Strategies in Home Energy [37]M. Farivar and S. H. Low, "Branch Flow Model: Relaxations and
Management System Integrating Solar Power, Energy Storage, and Convexification," IEEE Trans. Power Syst., vol. 28, no. 3, pp. 2554-2564,
Vehicle-to-Grid for Grid Support and Energy Efficiency," IEEE Trans. Ind. Aug. 2013.
Appl., vol. 56, no. 5, pp. 5716 - 5728, Sept. 2020. [38]S. Nguyen and C. Dupuis, "An Efficient Method for Computing Traffic
[16]X. Chen and K.-C. Leung, "Non-Cooperative and Cooperative Equilibria in Networks with Asymmetric Transportation Costs,"
Optimization of Scheduling With Vehicle-to-Grid Regulation Services," Transportation Science, vol. 18, no. 2, pp. 185-202, May. 1984.
IEEE Trans. Veh. Technol., vol. 69, no. 1, pp. 114-130, Jan. 2020. [39]M. Baran and F. Wu, "Network reconfiguration in distribution systems for
[17]A. Ghosh and V. Aggarwal, "Menu-Based Pricing for Charging of Electric loss reduction and load balancing," IEEE Trans. Power Deliv., vol. 4, no.
Vehicles With Vehicle-to-Grid Service," IEEE Trans. Veh. Technol. , vol. 2, pp. 1401 - 1407, Apr. 1989.
67, no. 11, pp. 10268-10280, Nov. 2018. [40]"smard," 12 2020. [Online]. Available:
[18]Z. Wei, Y. Li, Y. Zhang and L. Cai, "Intelligent Parking Garage EV https://www.smard.de/home/strommarkt-aktuell/alle-artikel.
Charging Scheduling Considering Battery Charging Characteristic," IEEE
Trans. Ind. Electron., vol. 65, no. 3, pp. 2806-2816, Mar. 2018.

0885-8950 (c) 2021 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: University of South Florida. Downloaded on June 27,2022 at 15:21:21 UTC from IEEE Xplore. Restrictions apply.