You are on page 1of 12

Renewable and Sustainable Energy Reviews 135 (2021) 110083

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews


journal homepage: http://www.elsevier.com/locate/rser

Analyzing power and dynamic traffic flows in coupled power and


transportation networks
Zhe Zhou a, Xuan Zhang a, **, Qinglai Guo a, b, *, Hongbin Sun a, b
a
Smart Grid and Renewable Energy Laboratory, Tsinghua-Berkeley Shenzhen Institute, Shenzhen, 518055, China
b
State Key Laboratory of Power Systems, Department of Electrical Engineering, Tsinghua University, Beijing, 100084, China

A R T I C L E I N F O A B S T R A C T

Keywords: The increasing penetration of electric vehicles is tightly coupling the operation of power and transportation
Electric vehicle systems. The representation of the dynamics of network flows in response to real-time information and control
Dynamic traffic assignment strategies is a problem of considerable practical significance. However, most of work considers static network
Optimal power flow
models that are only suitable for long-term planning horizons. This paper seeks to develop an integrated
Charging station
Coupled infrastructure systems
modeling framework for the real-time analysis and operation of the coupled systems. In particular, a dynamic
traffic assignment model is exploited to account for the time-varying travel demand and flow dynamics. An
accelerated diagonalization algorithm is proposed to compute the traffic flows in an efficient manner. Moreover,
we show that the equilibrium flow pattern of the coupled systems can be characterized by a fixed-point problem,
thus motivating a decentralized approach to attain the flow solutions. Numerical studies demonstrate both the
spatial and temporal interconnections between the power and traffic flows. The comparison of network flow
solutions based on the dynamic, semi-dynamic and static transportation network models is presented.

1. Introduction system models, where the former is usually described by an optimal


power flow model, and the latter gives rise to a Wardrop equilibrium.
Electric vehicles (EVs) offer consumers affordable, efficient and high- Studies explore various objectives, such as examining the equilibrium
tech transportation [1,2]. Electric mobility continues to grow rapidly as flow pattern [12,13], achieving robust power dispatch [14], and secu­
a result of support policies, technology advances and cost reductions [3, rity operation of both systems [15,16]. The second relevant body of
4]. The global EV sales are expected to reach 23 million and the stock research endeavors to manage the power and transportation systems
achieves 130 million vehicles by 2030 [5]. The increasing penetration of towards a social optimum, i.e., minimize the total social cost in the
EVs will create more profound interactions between the power and coupled systems [17–23]. Electricity prices and congestion tolls are two
transportation systems [6,7]. On the one hand, traffic conditions affect most popular approaches to achieve this goal. Various pricing schemes
the temporal and spatial distributions of EV flows, thereby changing the are explored, such as locational marginal pricing [18–20], first-best
pattern of charging requirements and the operation of power systems [8, congestion pricing [17,18] and second-best congestion pricing [17].
9]. On the other hand, the associated charging expenses influence the Finally, the literature on charging station planning also addresses the
travel decisions of EV drivers, thus affecting the operation of trans­ deployment problem considering both power and transportation system
portation systems [10,11]. As such, it is imperative to examine the in­ constraints [26–31]. Examples include integrating EV drivers’ choice
fluence of EV charging and develop proper control mechanisms from an models [26], considering heterogeneous EV driving range constraints
interdisciplinary perspective. [27,28], and accounting for elastic charging demand [31]. A short
Previous studies have examined at least three important problems summary of some state-of-the-art interdisciplinary studies is summa­
related to the coupled power and transportation systems [12–25]. First, rized in Table 1. As shown in the table, to present, few studies have
many researchers have investigated the integrated flow pattern in the explicitly accounted for time-varying electricity and travel demands as
coupled systems. These studies leverage both power and transportation well as traffic flow dynamics in the interdisciplinary studies. In

* Corresponding author. State Key Laboratory of Power Systems, Department of Electrical Engineering, Tsinghua University, Beijing, 100084, China.
** Corresponding author.
E-mail address: guoqinglai@tsinghua.edu.cn (Q. Guo).

https://doi.org/10.1016/j.rser.2020.110083
Received 9 March 2020; Received in revised form 19 June 2020; Accepted 8 July 2020
Available online 18 August 2020
1364-0321/© 2020 Elsevier Ltd. All rights reserved.
Z. Zhou et al. Renewable and Sustainable Energy Reviews 135 (2021) 110083

Table 1 models give a rather coarse representation of traffic flows and are not a
Summary of some state-of-the-art studies on coupled power and transportation proper option for real-time traffic prediction and control [33]. Dynamic
networks. traffic assignment (DTA) models are more preferable in short-term
References System models Objectives traffic analysis [35]. In fact, the development of DTA models is moti­
Power Transportation
vated by the rapid evolution of communication technologies and re­
Alizadeh et al. DCOPF STA min total travel time & generation quirements for effective real-time control measures [32,36]. Since the
[18] cost pioneering work of Merchant and Nemhauster in 1970s [37], the DTA
Alizadeh et al. DCOPF STA max social welfare models have evolved substantially over the past few decades. Their
[24]
inherent complexity and mathematical intractability have spawned a
He et al. [17] DCOPF STA min total travel time & generation
cost variety of approaches that range from the analytical to simulation-based
or min transportation social cost ones [35]. A comparison between these traffic assignment models is
Manshadi et al. DCOPF STA Equilibrium listed in Table 2. However, to the authors’ best knowledge, the
[25] application of DTA on the coupled power and transportation systems has
He et al. [21] ACOPF STA min total travel time & power
losses
not been investigated.
Wei et al. [19] ACOPF STA min total travel cost & generation This paper advances the above research by examining the in­
cost terconnections between the power and transportation systems consid­
Jiang et al. [20] ACOPF STA min total social cost ering the spatial and temporal variations of power and traffic flows.
Geng et al. [23] ACOPF STA Reduce peak load and peak traffic
More specifically, our paper makes the following contributions:
flow
Wei et al. [15] ACOPF STA Quantify impacts of road capacity
degradation • First, an integrated modeling framework based on the multi-period
Wei et al. [14] ACOPF STA Robust power dispatch network models is developed. Specifically, a dynamic trans­
Wei et al. [12] ACOPF STA Equilibrium portation network model is exploited to consider time-varying travel
Lv et al. [22] ACOPF Semi-DTA min total travel cost & generation
demands and traffic flow dynamics.
cost
Sun et al. [16] SCUC STA min total operational cost • Second, the network equilibrium flow pattern is described by a fixed-
point problem. Moreover, a decentralized approach is proposed for
DCOPF: direct current optimal power flow ACOPF: alternating current optimal
the independent power and transportation system operators to attain
power flow SCUC: security-constrained unit commitment STA: static traffic
the network flow solutions with limited information exchange.
assignment Semi-DTA: semi-dynamic traffic assignment.
• Third, an accelerated diagonalization algorithm is proposed to
compute the dynamic traffic flows in an efficient manner. The per­
Table 2 formance of the dynamic traffic network model compared to its semi-
Comparison of traffic assignment models based on time representations. dynamic and static counterparts is investigated based on the theo­
Models Time Multi- Description Traffic dynamics
retic analysis and numerical experiments.
interval period
The rest of this paper is organized as follows. Section 2 introduces the
STA >90 min × User equilibrium −
dynamic traffic network model and its computational algorithm. After
Semi- 15–90 User equilibrium in Residual flow
specifying the individual traveler and power system models in Section 3,

DTA min each period propagation
DTA <15 min √ User optimality Flow propagation we describe the network equilibrium model and the decentralized
approach to compute the flow solutions in Section 4. Numerical exper­
iments are conducted in Section 5. Finally, Section 6 concludes this
aforementioned research, a common assumption is that static Wardrop work.
equilibrium is achieved in each period and traffic flows propagate
instantaneously over the entire (or majority of) journey. This assump­ 2. Transportation network model
tion is reasonable for long-term system planning purposes [32]. How­
ever, it may not be appropriate in cases where we design time-varying This section presents the dynamic transportation network model. An
electricity prices and congestion tolls and need to capture variations of illustrative example is employed to demonstrate the computational al­
traffic and power flows over time and space [33]. gorithm for the proposed DTA model. Please reference the nomenclature
Recently, a growing line of research has emerged to address the section in Appendix A for definitions.
dynamics of traffic and power flows in response to real-time informa­
tion. For example, Alizadeh et al. [24] captured the EV mobility patterns 2.1. Network flow constraints
as well as the electric load flexibility by an energy expanded trans­
portation graph. Lv et al. [22] proposed a multi-period optimal power Consider a transportation system graph G T (N , A ) with the node set
and semi-dynamic traffic flow model and considered the flow propa­ N and arc set A . Let W represent the set of origin-destination (O-D)
gation between adjacent time periods. A key challenge in these studies is pairs. In the following, we denote an O-D pair by an index w ∈ W or a
the modeling of transportation systems. Currently, popular frameworks pair (o, d) interchangeably. Let ua (t) represent the inflow to arc a at time
for capturing traffic flows are static traffic assignment (STA) and t and va (t) the exit flow from arc a at time t. Let xa (t) denote the traffic
semi-dynamic traffic assignment (semi-DTA) models, which typically flow traveling on arc a at time t. The flow variables ua (t) and va (t) are
model traffic flows over a relatively long period [34]. Therefore, these taken as control variables and xa (t) is regarded as a state variable. The

Fig. 1. Flow variables of arc a (a) and flow conservation at node j (b) [32].

2
Z. Zhou et al. Renewable and Sustainable Energy Reviews 135 (2021) 110083

relationship between the flow variables is shown in Fig. 1(a). The state of the subroute r̃ at time t + ta (t) or 2) reach destination d at time t +
equation of arc a is given by Ref. [32]. ta (t). We typically write these constraints based on the estimated travel
time and update the estimated values iteratively during the optimization
dxar,w
= uar,w (t) − var,w (t), ∀a, r, w (1) process, known as relaxation or diagonalization technique [38]. Define
dt
t a (t) as the estimated travel time on arc a at time t. Flow propagation can
where xar,w represents the traffic flow of arc a via route r between O-D be described as follows:
pair w, and uar,w (t) and var,w (t) represent the inflow and exit flow of arc a ∑{ } { }
xar,w (t) = xbr,w [t + ta (t)] − xar,w (t) + Er,w [t + ta (t)] − Er,w (t) ,
via route r between O-D pair w, respectively. It follows that
b∈r̃
∑ ∑ ∑
uar,w (t) = ua (t), var,w (t) = va (t), xar,w (t) = xa (t), ∀a (2) ∀a, j, r, w; j ∕
= o; a ∈ D (j)
r,w r,w r,w (11)
The traffic flow on arc a is initialized to zero at t = 0, i.e., There exist other network flow constraints for DTA models, such as
xar,w (0) = 0, ∀a, r, w (3) first-in-first-out (FIFO) constraints. However, as flow propagation con­
straints imply that inflows must stay on that arc for a time period ta (t),
At a midway node j(j ∕ = o, d) of route r connecting O-D pair (o,d), the FIFO constraints are automatically satisfied for continuous-time models
flow conservation law restricts the traffic flows leaving the arcs that [32].
point towards node j at time t to be identical to the flows coming onto the
arcs that depart from node j at time t, as shown in Fig. 1(b). Hence, the
flow conversation constraints at midway nodes are given by 2.2. The DTA model
∑ ∑
uar,w (t) = var,w (t), ∀j ∕= o, d; r, w (4)
a∈C (j) a∈D (j)
In this subsection, we establish the dynamic transportation network
model to describe the traffic flows [32]. We begin by defining the travel
where C (j) denotes the set of arcs where flows depart from node j, and time for DTA models. Different from the static traffic models, there exist
D (j) denotes the set of arcs where flows exit from node j. various definitions for dynamic travel time. Here, the concept of
Let fw (t) denote the travel request between O-D pair w at time t, instantaneous travel time is utilized. The travel time on arc a at time t is
which is a given input parameter for the traffic assignment problem. The defined as the time to traverse an arc if current traffic conditions do not
flow conservation law at origin o enforces the flows leaving origin o at change. The travel time on route r at time t is the summation of the travel
time t to be identical to the flows entering the arcs whose head node is time of the constituent arcs at time t. Suppose that travel time on arc a,
origin o at time t, namely, ta (t), is only dependent on the inflow ua (t), namely,
∑ ∑
uar,w (t) = fw (t), ∀o ∕
= d; d (5) ta (t) = ga [ua (t)], ∀a (12)
The travel time on route r connecting O-D pair w is given by
a∈C (o) r


where C (o) denotes the set of arcs where flows depart from node o. tr,w (t) = ta (t), ∀r, w (13)
Moreover, let ew (t) denote the traffic flows that reach destination d at {a∈r:r∈R w }
time t and er,w (t) the traffic flows that reach destination d via route r at
In static traffic analysis, the concept of user equilibrium plays an
time t. Both ew (t) and er,w (t) are defined as control variables. Flow
important role. It refers to a stabilized flow pattern based on day-to-day
conservation constraints at destination d require that the flows depart­
adjustments of travelers’ route choice behavior [36]. However, finding
ing from destination d at time t are identical to the flows exiting the arcs
long-term traffic equilibrium is no longer the focus of research in dy­
that point towards destination d at time t. Flow conservation at desti­
namic traffic analysis. We are more interested in affecting traffic flows
nation nodes is represented mathematically by
based on the time-varying travel information [32]. The concept of user
∑ ∑
var,w (t) = ew (t), ∀o; d ∕
=o (6) optimality is developed, which is extended from the static traffic equi­
a∈D (d) r librium and formalized in the following statement.

where D (d) denotes the set of arcs where flows exit from node d. Definition 1. (Dynamic User-optimal State). The travel costs on
The accumulative flow that reaches destination d via route r by time t routes that are being utilized are identical and do not exceed the travel
is denoted by a state variable Er,w , which satisfies costs on any routes that are not being utilized [38].
It is noted that the major difference between the user equilibrium
dEr,w (t)
= er,w , ∀r, o, d ∕
=o (7) and user optimality states is that the latter requires the routes “being
dt
utilized” are equal since travel costs may change in the subsequent in­
The accumulative flow is zero at the initial time t = 0, i.e., tervals. However, due to the long tradition, the term “user equilibrium”
Er,w (0) = 0, ∀r, w (8) is still adopted in dynamic traffic analysis [36].
Based on the optimal control theory, the DTA model is established as
Furthermore, all the flow variables must remain non-negative, given follows:
by: ∫ T ∑{ ∫ ua (t) }
uar,w (t) ≥ 0, var,w (t) ≥ 0, xar,w (t) ≥ 0, ∀a, r, w (9) min ga (ω)dω dt (14)
u,v,x,e,E o 0
a

er,w (t) ≥ 0, Er,w (t) ≥ 0, ∀r, w (10)


s.t. Define ​ state ​ and ​ control ​ variables : ​ (1), ​ (7) (15)
Flow propagation constraints describe how traffic flow traverses
though the network over time and are crucial for distinguishing dynamic Definitional constraints: (2) (16)
traffic models from static ones [36]. For a midway node j(j ∕ = o) on route
Boundary conditions: (3), (8) (17)
r, define a subroute r̃ as the segment from node j to node d. The travel
time on arc a at time t is denoted by ta (t). The traffic flows traversing on Flow conservation constraints: (4)–(6) (18)
arc a ∈ D (j) via route r at time t are supposed to 1) traverse on the arcs
Non-negativity constraints: (9)–(10) (19)

3
Z. Zhou et al. Renewable and Sustainable Energy Reviews 135 (2021) 110083

• Add a super destination node that serves as the destination for all O-
D flows. The super destination is node 13 in Fig. 3.
• Add arcs pointing to the super destination node. These arcs are added
in two cases:
1. For layers 1, ⋯, K − 1, the destination nodes on these layers are
connected to the super destination node, e.g., arcs 7, 14, 21;
Fig. 2. A simple transportation graph. The numbers inside the circles represent
2. For the final layer K, all nodes on this layer are connected to the
the indexes of nodes and numbers marked next to arcs represent the indexes of
arcs. A single O-D pair is from nodes 1 to 3. There are two routes connecting the
super destination node, e.g., arcs 25, 26, 27.
O-D pair, namely, routes going through nodes 1→2→3 and through nodes 1→3. • These arcs are directed and point to the super destination node. It
takes no travel costs traversing these arcs.

We can seek the minimal-cost routes from the origins on each layer to
Flow propagation constraints: (11) (20)
the super destination node in the expanded network while satisfying the
The optimal control problem (14)–(20) yields a unique user-optimal flow propagation constraints. First, the estimated travel time should be
flow solution since the optimality conditions of the problem indicate the rounded to the closest integer, i.e.,
dynamic user-optimal state. Interested readers are referred to Refs. [39]
for a a detailed proof for this problem.

2.3. Accelerated diagonalization algorithm

To compute the dynamic traffic flows, we first convert the optimal


control problem (14)–(20) into a nonlinear optimization program by
discretizing the time period into K smaller intervals. Note that the
discrete time optimization program can become convex if the flow
propagation constraints are absent. To this end, a time-space expanded
transportation graph is constructed to handle the constraints [24,32].
We take a simple transportation graph G T (N , A ) shown in Fig. 2 as
an example. There is a single O-D pair from nodes 1 to 3. There exist two
routes connecting O-D pair (1, 3), i.e., routes going through nodes
1→2→3 and through nodes 1→3. The time-space expanded trans­
portation graph, ̂
G T(N̂, ̂A ), is constructed by the following steps (see
Fig. 3 for an example of 4 intervals):

• Duplicate the original graph G T (N , A ) in Fig. 2 into K layers. On


each layer, there is a duplicate of arcs and nodes that are identical to
G T (N , A ). These duplicated arcs maintain the same cost as that of
the original graph.
• Add arcs between nodes situated on adjacent layers that correspond Fig. 4. Flowchart of the accelerated diagonalization algorithm to compute
dynamic traffic flows. Initialize the estimated travel time t a (k) based on (21).
to the same nodes in the original graph. These arcs are directed and
The subset of minimal-cost routes in the original graph R is updated in each
point towards the layer of the higher interval. For example, arcs 4, 5,
iteration based on the newest travel costs. The set of minimal-cost routes in the
6 are added pointing from layer 1 to layer 2. Traversing these arcs
expanded network, R ̂ , is identified by satisfying the flow propagation
takes no travel costs and represents the flows staying on these arcs
constraints.
within that time interval.

Fig. 3. The time-space expanded transportation graph of Fig. 2 for 4 intervals. The introduction of layers helps handle the flow propagation constraints.

4
Z. Zhou et al. Renewable and Sustainable Energy Reviews 135 (2021) 110083

Fig. 5. Transportation graph (a) and extended transportation graph (b). Dummy nodes and arcs are added to represent the cost of charging, shown as the red node
and red arc in (b). A charging route from nodes O to D is through arcs (1, 3, 2’’). (For interpretation of the references to color in this figure legend, the reader is
referred to the Web version of this article.)

ta (k) = j, if ​ j − 0.5 ≤ ta (k) < j + 0.5, j = 0, ⋯, K (21) [ ]


ccs ev
j λj (t) = p λj (t) (23)
When arc a is searched and t a (k) = j(j ∕ = 0), the next node to be
searched is the tail node of this arc copy situated on the layer that is j-th where the charging station is connected to bus j of the power network,
layer higher. For example, suppose the travel time of all arcs equals one. ccs ev
j (t) is the associated cost of electricity at the charging station, p is the
For traffic flow entering route 1→2→3 at interval 1, the feasible route average EV charging power. and λj (t) is the locational marginal price
should be through nodes 1→4→5→8→9→13. It can be observed that (LMP) at bus j in period t.
traversing arcs 4 and 12 represents the delay during the propagation EV drivers are assumed to seek and travel on the routes with the
process. Moreover, if the estimated travel time t a (k) = 0, the route minimal travel cost. If we can capture the cost of travel and charging into
search for the next node remains on the same layer. a unified framework, the minimal-cost routes can be found based on the
We apply the iterative diagonalization technique to compute the traditional shortest path algorithms [12,18]. As such, we establish an
traffic flows. In each iteration, the estimated travel time is fixed. By extended transportation graph by adding dummy arcs at nodes where
searching the minimal-cost routes in the time-space expanded graph, the charging stations locate, shown in Fig. 5. In the extended graph, let A tt
diagonalized DTA problem is a linear-constrained convex problem and represent the set of travel time arcs (black arcs in Fig. 5(b)) and A cs the
much easier to solve. The estimated travel time is updated iteratively till set of charging cost arcs (red arc in Fig. 5(b)). The cost traversing an arc
convergence. However, enumerating all routes can be computationally is given by
expensive. Therefore, we extend the column generation technique [12, {
40,41] to solve the DTA problem. ca (t) =
βta (t), ∀a ∈ A tt ,
(24)
The overall iterative procedure is illustrated in Fig. 4. In the diago­ p λj (t), ∀a ∈ A cs , j ∈ M cs
ev

nalization step (blue box in Fig. 4), we first identify a subset of minimal-
cost routes in the original graph, R . Then, the set of routes in the where ca (t) is the travel cost on arc a in period t, β is the economic value
of time, and M cs is the set of coupled buses in the power network, where
expanded graph, R ̂ , is obtained based on the route set R and current arc
bus j ∈ M cs provides electrical energy for the nearest charging station on
cost t(k). Each time the estimated travel time is updated, the route sets
arc a ∈ A cs . We denote by the set of all arcs in the extended graph.
R and R ̂ are updated.

3. System models 3.2. Power network model

Consider a transportation network with charging stations located at Consider a radial distribution network G E (M , B ) consisting of the
different traffic nodes. EV drivers are assumed to stop at one charging set of buses M = {1, ⋯, n} and set of branches B . Let us index the slack
station during the journey. Their route choice is affected by the price of bus by 0 and denote a branch in B by the pair (i, j) if it points from buses
electricity and traffic conditions. Moreover, their aggregate route choice i to j. Denote the child set of bus j as δ(j) : = {i : (i,j) ∈ B }. In each period
also changes the temporal and spatial distributions of power and traffic t, the power flow is described by the branch flow equations under a
flows, thereby affecting the operation of power and transportation sys­ second-order cone relaxation [43–45]:
tems. In the following, we first describe the travel costs of the individual ∑
EV drivers. Then, we propose a network equilibrium model that ac­ pj (t) = Pij (t) − Rij Lij (t) − Pjk (t), ∀j ∈ M (25)
counts for the interdependency between the power and transportation
k∈δ(j)

systems. ∑
qj (t) = Qij (t) − Xij Lij (t) − Qjk (t), ∀j ∈ M (26)
k∈δ(j)

3.1. Travel costs for EV drivers ( )


( )
νj (t) = νi (t) − 2 Rij Pij (t) + Xij Qij (t) + R2ij + Xij2 Lij (t), ∀(i, j) ∈ B
The travel time on an arc increases with traffic inflows due to
(27)
congestion effects. The most widely adopted travel time function is
developed by the U.S. Bureau of Public Roads and given by Ref. [42]. Pij (t)2 + Qij (t)2
[ ( )4 ] Lij (t) ≥ , ∀(i, j) ∈ B (28)
ua (t) νi (t)
ta [ua (t)] = ta0 1 + κa (22)
Constraints (25) enforce the nodal active power balance. The left-
Qard

hand side of (25) is the active power injection and satisfies


where ta0 is the free-flow travel time, κa is the model parameter, and Qrd
a
is the capacity of arc a. pj (t) : = − pgj (t) + plj (t) + pev ua (t), ∀j ∈ M cs , a ∈ A cs
(29)
Assume that the electricity is provided at the locational marginal
g
pricing. When EV drivers get charged at a charging station en route, the where pj (t), plj (t) and pev ua (t) are the active power generation, base
cost of charging is given by electricity load and EV charging demand at bus j in period t,

5
Z. Zhou et al. Renewable and Sustainable Energy Reviews 135 (2021) 110083

Fig. 6. Interdependency between the power and transportation systems in period t. In each period t, the ED problem is solved once, while the DTA problem is
updated K times. To achieve the equilibrium state in period t, a fixed-point problem is formulated, where the LMP λ(t)(i) and charging traffic flow u(t)(i+1) are
exchanged in the i-th iteration.

respectively. Constraints (26) are nodal reactive power balance condi­ On the transportation side, the price of electricity affects the route
tions. Constraints (27) describe the forward voltage drop on each choice of EV travelers, resulting in the spatial and temporal variations of
branch. Constraints (28) denote the conic relaxation. charging loads. On the power side, the charging loads influence the
In addition to the power flow equations 25–28, we also impose the power system operation, thus affecting the LMPs. As such, the in­
following radiality constraints (30)–(31), ramp limits of the generators teractions and couplings between the two infrastructures can be char­
(32) and security constraints on the active power generation and nodal acterized by a fixed-point mapping problem.
voltages (33)–(34). Let λ(t) = [λj (t)]j∈M cs denote the vector of electricity prices at
charging stations and u(t) = [ua (t)]a∈A cs the vector of traffic flows
Rij Lij (t) − Pij (t) ≤ 0, ∀(i, j) ∈ B (30)
through charging stations in period t. With a slight abuse of the notation,
suppose that a time duration [0, T] is divided into |T | time periods and
Xij Lij (t) − Qij (t) ≤ 0, ∀(i, j) ∈ B (31)
T is the set of such time periods. Within each period t ∈ T , the ED
− pramp ≤ pgj (t) − pgj (t − 1) ≤ pramp , ∀j ∈ M (32) problem is solved once to determine the LMPs. Meanwhile, the DTA
problem runs K times such that the period t is subdivided into K smaller
j j

pgj ≤ pgj (t) ≤ pgj , ∀j ∈ M (33) time intervals.


Given the electricity price in the i-th iteration in period t, λ(t)(i) , the
νj ≤ νj (t) ≤ νj , ∀j ∈ M (34) distribution of traffic flow can be obtained by the following DTA
problem:
The power system operator solves the economic dispatch (ED)
K ∑ ∫
problem with an objective of minimizing the total energy production ∑ ua (k)
min ca (ω)dω (37)
costs, given by Ref. [12,45]: u,v,x,E
k=1 a∈A e 0
∑ [ g ] ∑
min
g
FE : = Gj pj (t) + λ0 (t) P0k (t) (35) s.t. ∀k = 1, ⋯, K :
pj (t)
j∈M k∈δ(0)

xar,w (k + 1) = xar,w (k) + uar,w (k) − var,w (k), ∀a, r, w (38)


s.t. (25) − (34) (36)
∑ ∑
g Ew (k + 1) = Ew (k) + var,w (k), ∀o; d ∕
=o (39)
where Gj (t)[pj (t)] is a convex cost function for generators at bus j in
∑ a∈D (d) r
period t, and λ0 P0k (t) is the energy purchase cost from the main grid
k∈δ(0) ∑ ∑
uar,w (k) = fw (k), =d
∀o ∕ (40)
with λ0 (t) being the contract energy price in period t. Note that the
a∈C (o) r
electricity is provided at LMPs, which are the dual variable associated
with constraints (25). ∑ ∑
var,w (k) = uar,w (k), ∀j, r, w; j ∕
= o, d (41)
a∈D (j) a∈C (j)
4. Network equilibrium analysis
∑{ } { }
xar,w (k) = xbr,w [k + ta (k)] − xar,w (k) + Er,w [k + ta (k)] − Er,w (k) ,
In this section, we first analyze the equilibrium flow pattern in the b∈r̃
coupled power and transportation networks. Then, we compare the ∀a ∈ D (j); j ∕
= o; r, w
network equilibrium solutions based on the dynamic, semi-dynamic and
(42)
static traffic assignment models.
uar,w (k) ≥ 0, var,w (k) ≥ 0, xar,w (k + 1) ≥ 0, xar,w (1) = 0, ∀a, r, w
4.1. Problem formulation (43)

The interdependency between the power and transportation systems Er,w (k + 1) ≥ 0, Er,w (1) = 0, ∀r, w (44)
is depicted in Fig. 6. While the power system operator typically solves
the ED problem on an hourly basis, the dynamic traffic flow is updated where the price of electricity affects the cost traversing arcs in (37). We
no longer than every 15 min. As demonstrated by the figure, the traffic average the traffic flow through charging stations over K time intervals
flow evolves a couple of times within a single period for power dispatch. and obtain the charging traffic flow u(t)(i+1) .

6
Z. Zhou et al. Renewable and Sustainable Energy Reviews 135 (2021) 110083

analyzed theoretically. Section 5 will conduct numerical experiments to


validate our analysis.
The interactions between the power and traffic flows based on the
STA, semi-DTA and DTA models are illustrated in Fig. 7. These traffic
assignment models update traffic flows at different time scales, thus
interacting with the power flows at various time scales. The major dif­
ference between the DTA model and its static and semi-dynamic coun­
terparts is the inclusion of flow propagation constraints, which require
Fig. 7. Interactions between the power and traffic flows based on the static, that an inflow can not exit the arc unless the arc travel time elapses [36].
semi-dynamic and dynamic traffic assignment models. Moreover, both semi-DTA and STA models can only consider aggregate
travel requests over a certain time duration. The decentralized iterative
Given the charging load, the distribution of power flow is obtained procedures for computing the network flows are demonstrated in Fig. 8.
by the ED problem: If we assume that the travel requests and cost functions are identical for
all transportation models, the congestion level evaluated by STA and
min
g
FE (45) semi-DTA models is generally higher than that by DTA models. The
pj (t)
reason is that DTA models allow inflows entering the same arc at
different time intervals, resulting in lower estimation of travel costs. STA
s.t. pj (t) = − pgj (t) + plj (t) + pev ua (t)(i+1) , ∀j ∈ M cs , a ∈ A cs
(46)
and semi-DTA models, on the contrary, assume that the all inflows
propagate instantaneously over the entire (or majority of) journey,
(25) − (28), (30) − (34) (47)
entering all arcs along the routes at the same time, thus resulting in a
higher estimation of the congestion level.
where the Lagrange multipliers associated with constraints (25) are the
electricity prices λ(t)(i+1) .
Let us define the above i-th iterative mapping in period t, ψ t,i : ΩE ↦ → 5. Numerical examples and discussions
ΩE as
( ) This section conducts numerical experiments for two test systems.
λ(t)(i+1) = ψ t,i λ(t)(i) (48) We first examine the distinctions between static, semi-dynamic and
dynamic traffic assignment models based on a simple transportation
where ΩE is the feasible set of the power flow defined by (46)-(47). network in Fig. 2. Then, we consider the Sioux Falls transportation
Hence by contraction mapping theorem [46,47], the convergence re­ network coupled with a 141-node electric distribution network and
quires that the spectral radius of the Jacobian matrix ψ t,i ’ is smaller than investigate the integrated power and traffic flows based on various
1, i.e., traffic assignment models.
( )
ρ ψ t,i ’ < 1 (49)
5.1. Small network example
where ρ( ⋅) represents the spectral radius of a matrix.
In each time period t, an equilibrium flow pattern is achieved by We first consider a simple transportation network with 3 nodes and 3
iterating the power and traffic subproblems. The overall iterative pro­ arcs, as depicted in Fig. 2. The travel time function is given by
cedure for all periods t ∈ T is summarized in Algorithm LABEL:Alg:
DTACoupled. ta (t) = 1 + 0.1ua (t)2 , ∀a (50)

Algorithm 1. Decentralized Algorithm to Compute Power and Dy­ The assignment time period is divided into K = 4 intervals. The time-
namic Traffic Flows space expanded transportation graph is shown in Fig. 3. The O-D travel
demand is listed in Table 3. The optimal traffic flows and traffic time are
summarized in Table 4, where xa (k) denotes the traffic flow on arc a
4.2. Comparison with static and semi-dynamic counterparts during interval k, and ua (k) and va (k) denote the inflow and exit flow
during interval k, respectively. Table 4 shows clearly how traffic flow
In this subsection, comparisons of the computational procedures and propagates through the network during each time interval. For example,
equilibrium solutions based on the STA, semi-DTA and DTA models are there is 1.3246 unit of inflow entering arc 1 during interval 1 and the

Fig. 8. Decentralized iterative procedures to compute power and traffic flows using the static, semi-dynamic and dynamic traffic assignment models.

7
Z. Zhou et al. Renewable and Sustainable Energy Reviews 135 (2021) 110083

Table 3 travel time on arc 1 becomes 1.1755 intervals. Thus, this flow exits
O-D trip rates from nodes 1 to 3 of the small network (in p.u.). during interval 2. Note that at the same time, an inflow of 3.6754 enters
Time interval k 1 2 3 4 arc 3 during interval 1 and incurs travel cost of 2.3509 intervals, thereby
exiting arc 3 during interval 3. If we add up the inflows entering arcs 1
Flow/interval 5 5 0 0
and 3 during interval 1, they are equal to the O-D trip rates shown in
Table 3. Table 5 provides the dynamic travel time on the two routes
during each interval. The dynamic travel time on routes 1→2→3 and
Table 4 1→3 is very close during intervals 1 and 2 and satisfies the dynamic user
Optimal traffic flows and travel time of the small network.
optimality condition considering the round-off errors. Note also dy­
Interval Arc Vehicles Inflow Exit Vehicles Travel namic traffic time during intervals 3 and 4 is not identical since route
k a xa (k + 1) ua (k) flow xa (k) time
1→2→3 is only partially used, i.e., ua (k) > 0 does not hold for all arcs
va (k) ta (k)
The optimal traffic flows and travel time based on the STA and semi-
1 1 1.3246 1.3246 0.0000 0.0000 1.1755 DTA models are reported in Table 6. The optimal travel times evaluated
2 1 1.3246 1.3246 1.3246 1.3246 1.1755 by the STA and semi-DTA models are both higher than that estimated by
3 1 0.0000 0.0000 1.3246 1.3246 1.0000
the DTA model. This validates our analysis in Section 4.2 owning to the
4 1 0.0000 0.0000 0.0000 0.0000 1.0000
1 2 0.0000 0.0000 0.0000 0.0000 1.0000 fact that DTA allows inflows entering the arcs at different time intervals,
2 2 1.3246 1.3246 0.0000 0.0000 1.1755 thereby yielding lower travel costs. Note that the actual O-D trip rates
3 2 1.3246 1.3246 1.3246 1.3246 1.1755 for the semi-DTA model are less than the given O-D trip rates since semi-
4 2 0.0000 0.0000 1.3246 1.3246 1.0000
DTA models assume that certain O-D flow cannot reach the destination
1 3 3.6754 3.6754 0.0000 0.0000 2.3509
2 3 7.3508 3.6754 0.0000 3.6754 2.3509
within one time period. Interested readers may refer to Ref. [35,48] for
3 3 3.6754 0.0000 3.6754 3.6754 1.0000 more details on these models.
4 3 0.0000 0.0000 3.6754 3.6754 1.0000

5.2. Sioux Falls network example

Table 5 In this subsection, we consider the city of Sioux Falls [49] served by a
Dynamic route travel time of the small network. 141-node power distribution network [50] (see Fig. 9) to examine the
Interval k Route traffic time equilibrium solutions in the coupled networks. We assume that six
Route 1→2→3 Route 1→3 charging stations are located in the transportation network and their
1 2.1755 2.3509 electrical demand is provided by the nearest buses in the power network
2 2.3510 2.3509 [41]. The coupling nodes and arcs in the power and transportation
3 2.1755 1.0000 networks are denoted by the same colors in Fig. 9. The time-space
4 1.0000 1.0000 expanded graph of Sioux Falls transportation network for two in­
tervals is portrayed in Fig. 10.

Table 6
Optimal traffic flows and travel time based on the semi-DTA and STA models for the small network.
Traffic model Arc traffic flow Route traffic flow Route traffic time

1 2 3 1→2→3 1→3 1→2→3 1→3

Semi-DTA 2.5167 2.5167 4.7610 2.5167 4.7610 3.2667 3.2667


STA 3.7840 3.7840 4.8638 3.7840 6.2160 4.8638 4.8638

Fig. 9. The Sioux Falls transportation network (a) and a 141-node radial power network (b) coupled by fast charging stations. The coupled arcs and buses of one
charging station are labeled by the same colors. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of
this article.)

8
Z. Zhou et al. Renewable and Sustainable Energy Reviews 135 (2021) 110083

Consider a time duration of 2 h. We assume that DTA runs every 10


min and divides the duration into K = 12 intervals. The ED problem in
the power system is solved hourly, i.e., every 6 intervals. Semi-DTA and
STA models are assumed to update traffic flows every 6 and 12 intervals,
respectively. The O-D travel requests are given in Table 7.
Fig. 11 reports the spatial and temporal distributions of traffic flows
at six charging stations during intervals 1 to 25. The location and time
EV flows traveling through charging stations are clearly illustrated in the
figure. It can be observed during intervals 1–25, the earliest charging
takes place at charging station 4 during interval 12 occurs during in­
terval 22 at charging stations 5 and 6. Moreover, the majority of EV
travelers choose to charge at charging station 4 from intervals 12 to 20.
No charging is fulfilled at charging station 1.
To acquire insight into the differences of several traffic assignment

Fig. 10. Time-space expanded graph of the Sioux Falls transportation network
for 2 intervals.

Table 7
O-D trip rates of the Sioux Falls network (in p.u.).
Origin Destination Flow/interval Departure time interval k

1 24 1 1
1 24 1 2
1 24 1 4
1 24 1 7
1 24 1 9

Fig. 12. Electricity prices at six charging stations based on the DTA, semi-DTA
and STA models for the Sioux Falls network.

Fig. 11. The spatial and temporal distributions of EV traffic flows at six charging stations (CSs) based on the DTA model for the Sioux Falls network.

Table 8
EV traffic flows at charging stations based on the semi-DTA and STA models for the Sioux Falls network.
Traffic flows CS 1 CS2 CS 3 CS 4 CS 5 CS 6

Semi-DTA Intervals 1-6 0.0000 0.0002 0.0000 2.7518 0.0001 0.0003


Intervals 7-12 0.0000 0.0000 0.0000 2.0898 0.0000 0.0000
STA Intervals 1-12 0.0000 0.4923 0.0000 2.9866 0.0000 1.5209

9
Z. Zhou et al. Renewable and Sustainable Energy Reviews 135 (2021) 110083

Table 9
Optimal travel cost of EV users based on the DTA, semi-DTA and STA models for the Sioux Falls network.
Travel cost Time interval k

1 2 3 4 5 6 7 8 9 10 11 12

DTA 12.17 12.17 12.17 12.17 12.18 12.17 12.17 12.18 12.17 12.17 12.18 12.17
Semi-DTA 13.20 13.20 13.20 13.20 13.20 13.20 12.61 12.61 12.61 12.61 12.61 12.61
STA 13.79 13.79 13.79 13.79 13.79 13.79 13.79 13.79 13.79 13.79 13.79 13.79

models, we compare the traffic flows at charging stations obtained by transportation system operators in a decentralized fashion. Numerical
semi-DTA and STA models in Table 8. First, semi-DTA and STA models results validate our analysis and demonstrate that employing DTA
can only compute traffic flow distributions for longer periods and give a models can provide a more accurate estimation of electric and traffic
rather coarse representation of traffic flows. Second, one can see that the conditions.
flow solutions differ greatly from that obtained by the DTA model. Semi- The proposed framework is expected to provide off-line dynamic
DTA and STA models take aggregate travel requests as input, resulting in forecasting, estimation and monitoring for the coupled systems. This
greater differences when the network is more congested. work can be generalized in many ways, such as security analysis during a
We further compare the price of electricity at charging stations power blackout, and design of real-time electricity prices and congestion
estimated by the DTA, semi-DTA and STA models, shown in Fig. 12. As tolls. Note that the strengths of DTA models come at the cost of high
expected, the electricity prices at charging stations evaluated by the DTA computational burdens and the requirement for accurate O-D trip data.
model are always the lowest during all intervals. Both semi-DTA and In cases where the computational resources are limited or detailed O-D
STA models neglect details on the exact time EVs arrive at charging trip data is absent, semi-DTA or STA models would be a better option.
stations, thus overestimating the charging demand. STA has least in­ Future work involves exploring efficient algorithms and simulation-
formation on traffic dynamics, resulting in the highest estimation of based tools for DTA problems.
electricity prices. Moreover, the optimal travel cost for EV travelers
evaluated by the three traffic assignment models is listed in Table 9. The Credit author statement
optimal travel cost obtained by the STA model remains unchanged and
the highest for the entire 12 intervals, while the travel cost estimated by Zhe Zhou: Writing- Original draft preparation, Methodology, Soft­
the DTA model is time-varying and always lower than those obtained by ware, Visualization, Data curation, Xuan Zhang: Conceptualization,
the semi-DTA and DTA models. Investigation, Validation, Writing- Reviewing and Editing, Qinglai Guo:
Writing- Reviewing and Editing, Formal analysis, Funding acquisition,
6. Conclusion Project administration, Hongbin Sun: Resources, Supervision

Large-scale fast charging of EVs inevitably couples the operation of Declaration of competing interest
the power and transportation systems. This paper develops a network
equilibrium model that captures the spatial and temporal variations of None.
power and traffic flows for real-time analysis and management of both
systems. Specifically, a dynamic traffic assignment model is established Acknowledgments
to consider the time-varying travel demand and traffic flow dynamics.
An accelerated diagonalization algorithm is proposed to solve the dy­ This work was supported in part by Beijing Natural Science Foun­
namic traffic assignment problem in a more tractable manner. We show dation under Grant No. JQ18008 and in part by Shenzhen Science and
that the equilibrium flow solutions can be attained by the power and Technology Program under Grant No. KQTD20170810150821146.

Appendix A

Table 1
Nomenclature.

The Transportation Network Indices and Sets

A cs Set of arcs representing costs in charging stations


a∈Ae Indices for arcs in the extended graph, A e = A tt ∪ A cs
w∈W Indices for O-D pairs
(o, d) An O-D pair with o being the origin and d the destination
r ∈ Rw Indices for routes connecting O-D pair w
R Set of routes in the original graph, R : = ∪w∈W R w
̂
R Set of routes in the time-space expanded network
C (j) Set of arcs where flows depart from node j
D (j) Set of arcs where flows exit from node j
State Variables
xa (t) Traffic flow on arc a at time t
Er,w Cumulative flow arriving at destination over route r by time t
Control Variables
ua (t) Inflow to arc a at time t
va (t) Exit flow from arc a at time t
er,w (t) Traffic flow arriving at destination over route r at time t
ew (t) Traffic flow arriving at destination at time t
(continued on next page)

10
Z. Zhou et al. Renewable and Sustainable Energy Reviews 135 (2021) 110083

Table 1 (continued )
The Transportation Network Indices and Sets

Other Problem Variables


ta (t) Travel time entering arc a at time t
ca (t) Travel cost on arc a at time t
Parameters
fw (t) Travel request for O-D pair w at time t
The Power Network Indices and Sets
M Set of buses in the power network
M cs Set of buses connected to charging stations
B Set of branches in the power network
Optimization Variables
λj (t) Locational marginal prices at bus j in period t
g
pj (t) Active power generation in period t
νj (t) Square magnitude of complex voltage at bus j in period t
pj (t) Active net load at bus j in period t
qj (t) Reactive net load at bus j in period t
Lij (t) Square magnitude of current from buses i to j in period t
Pij (t) Active power flowing from buses i to j in period t
Qij (t) Reactive power flowing from buses i to j in period t
FE Total energy production costs
Parameters
pev EV charging power for unit traffic flow
Rij Resistance of branch (i, j)
Xij Reactance of branch (i, j)
pramp
j
Ramp limits of generators at bus j
νj /νj Lower/upper limit of voltage square at bus j
g
pgj /pj Lower/upper limit of active power generation at bus j

References [14] Wei W, Mei S, Wu L, Wang J, Fang Y. Robust operation of distribution networks
coupled with urban transportation infrastructures. IEEE Trans. Power Syst. 2017;
32(3):2118–30.
[1] Ji Z, Huang X. Plug-in electric vehicle charging infrastructure deployment of China
[15] Wei W, Wang J, Wu L. Quantifying the impact of road capacity loss on urban
towards 2020: policies, methodologies, and challenges. Renew Sustain Energy Rev
electrified transportation networks: an optimization based approach. Int J
2018;90:710–27. https://doi.org/10.1016/j.rser.2018.04.011. http://www.scienc
Transport Sci Technol 2016;5(4):268–88. https://doi.org/10.1016/j.
edirect.com/science/article/pii/S136403211830220X.
ijtst.2017.03.003. http://www.sciencedirect.com/science/article/pii/S2046043
[2] Hu X, Li SE, Yang Y. Advanced machine learning approach for lithium-ion battery
016300375.
state estimation in electric vehicles. IEEE Trans Transp Electri 2016;2(2):140–9.
[16] Sun Y, Chen Z, Li Z, Tian W, Shahidehpour M. Ev charging schedule in coupled
[3] Das H, Rahman M, Li S, Tan C. Electric vehicles standards, charging infrastructure,
constrained networks of transportation and power system. IEEE Trans Smart Grid
and impact on grid integration: a technological review. Renew Sustain Energy Rev
2019;10(5):4706–16. https://doi.org/10.1109/TSG.2018.2864258.
2020;120:109618. https://doi.org/10.1016/j.rser.2019.109618. http://www.sci
[17] He F, Yin Y, Zhou J. Integrated pricing of roads and electricity enabled by wireless
encedirect.com/science/article/pii/S1364032119308251.
power transfer. Transport Res C Emerg Technol 2013;34:1–15.
[4] Hu X, Zou C, Tang X, Liu T, Hu L. Cost-optimal energy management of hybrid
[18] Alizadeh M, Wai H-T, Chowdhury M, Goldsmith A, Scaglione A, Javidi T. Optimal
electric vehicles using fuel cell/battery health-aware predictive control. IEEE Trans
pricing to manage electric vehicles in coupled power and transportation networks.
Power Electron 2020;35(1):382–92.
IEEE Trans. Automat. Contr. 2017;4(4):863–75.
[5] EVI, Global EV outlook. Scaling-up the transition to electric mobility [Online].
[19] Wei W, Mei S, Wu L, Shahidehpour M, Fang Y. Optimal traffic-power flow in urban
Available:. 2019. Accessed Feb. 3, 2020]. (May 2019, https://www.iea.org/report
electrified transportation networks. IEEE Trans Smart Grid 2017;8(1):84–95.
s/global-ev-outlook-2019.
[20] Jiang H, Zhang Y, Chen Y, Zhao C, Tan J. Power-traffic coordinated operation for
[6] Muratori M, Kontou E, Eichman J. Electricity rates for electric vehicle direct
bi-peak shaving and bi-ramp smoothing - a hierarchical data-driven approach. Appl
current fast charging in the United States. Renew Sustain Energy Rev 2019;113:
Energy 2018;229:756–66.
109235. https://doi.org/10.1016/j.rser.2019.06.042. http://www.sciencedirect.
[21] He F, Yin Y, Wang J, Yang Y. Sustainability SI : optimal prices of electricity at
com/science/article/pii/S1364032119304356.
public charging stations for plug-in electric vehicles. Network Spatial Econ 2016;
[7] Shaukat N, Khan B, Ali S, Mehmood C, Khan J, Farid U, Majid M, Anwar S,
16(1):131–54.
Jawad M, Ullah Z. A survey on electric vehicle transportation within smart grid
[22] Lv S, Wei Z, Sun G, Chen S, Zang H. Optimal power and semi-dynamic traffic flow
system. Renew Sustain Energy Rev 2018;81:1329–49. https://doi.org/10.1016/j.
in urban electrified transportation networks. IEEE Trans Smart Grid 2019:1.
rser.2017.05.092. http://www.sciencedirect.com/science/article/pii/S1364032
https://doi.org/10.1109/TSG.2019.2943912.
117307190.
[23] Geng L, Lu Z, He L, Zhang J, Li X, Guo X. Smart charging management system for
[8] Qin Y, Tang X, Jia T, Duan Z, Zhang J, Li Y, Zheng L. Noise and vibration
electric vehicles in coupled transportation and power distribution systems. Energy
suppression in hybrid electric vehicles: state of the art and challenges. Renew
2019;189:116275. https://doi.org/10.1016/j.energy.2019.116275. http://www.
Sustain Energy Rev 2020;124:109782. https://doi.org/10.1016/j.
sciencedirect.com/science/article/pii/S036054421931970X.
rser.2020.109782. http://www.sciencedirect.com/science/article/pii/S1364032
[24] Alizadeh M, Wai H, Goldsmith A, Scaglione A. Retail and wholesale electricity
120300782.
pricing considering electric vehicle mobility. IEEE Trans Control Netw Syst 2019;6
[9] Tu H, Feng H, Srdic S, Lukic S. Extreme fast charging of electric vehicles: a
(1):249–60.
technology overview. IEEE Trans Transp Electri 2019;5(4):861–78.
[25] Manshadi SD, Khodayar ME, Abdelghany K, Uster H. Wireless charging of electric
[10] fei Chen C, de Rubens GZ, Noel L, Kester J, Sovacool BK. Assessing the socio-
vehicles in electricity and transportation networks. IEEE Trans Smart grid 2017.
demographic, technical, economic and behavioral factors of nordic electric vehicle
1–1.
adoption and the influence of vehicle-to-grid preferences. Renew Sustain Energy
[26] He F, Wu D, Yin Y, Guan Y. Optimal deployment of public charging stations for
Rev 2020;121:109692. https://doi.org/10.1016/j.rser.2019.109692. http://www.
plug-in hybrid electric vehicles. Transport Res B-Meth 2013;47:87–101.
sciencedirect.com/science/article/pii/S1364032119308974.
[27] Zhang H, Moura SJ, Hu Z, Qi W, Song Y. A second-order cone programming model
[11] Kester J, Noel L, de Rubens] GZ, Sovacool BK. Policy mechanisms to accelerate
for planning PEV fast-charging stations. IEEE Trans. Power Syst. 2018;33(3):
electric vehicle adoption: a qualitative review from the nordic region. Renew
2763–77. https://doi.org/10.1109/TPWRS.2017.2754940.
Sustain Energy Rev 2018;94:719–31. https://doi.org/10.1016/j.rser.2018.05.067.
[28] Zhang H, Moura SJ, Hu Z, Song Y. PEV fast-charging station siting and sizing on
http://www.sciencedirect.com/science/article/pii/S136403211830426X.
coupled transportation and power networks. IEEE Trans Smart Grid 2018;9(4):
[12] Wei W, Wu L, Wang J, Mei S. Network equilibrium of coupled transportation and
2595–605. https://doi.org/10.1109/TSG.2016.2614939.
power distribution systems. IEEE Trans Smart Grid 2017. 1–1.
[29] Zhang H, Hu Z, Xu Z, Song Y. An integrated planning framework for different types
[13] Amini MH, Karabasoglu O. Optimal operation of interdependent power systems
of PEV charging facilities in urban area. IEEE Trans Smart Grid 2016;7(5):2273–84.
and electrified transportation networks. 2017. arXiv:1701.03487.

11
Z. Zhou et al. Renewable and Sustainable Energy Reviews 135 (2021) 110083

[30] Zhou B, Chen G, Huang T, Song Q, Yuan Y. Planning pev fast-charging stations [41] Zhang H, Hu Z, Song Y. Power and transport nexus: routing electric vehicles to
using data-driven distributionally robust optimization approach based on promote renewable power integration. IEEE Trans Smart Grid 2020:1. https://doi.
φ-divergence. IEEE Trans Transp Electri 2020;6(1):170–80. org/10.1109/TSG.2020.2967082.
[31] Gan X, Zhang H, Hang G, Qin Z, Jin H. Fast-charging station deployment [42] Sheffi Y. Urban transportation networks: equilibrium analysis with mathematical
considering elastic demand. IEEE Trans Transp Electri 2020;6(1):158–69. programming methods. Prentice-Hall; 1984.
[32] Ran B, Boyce D. Dynamic urban transportation network models: theory and [43] Low SH. Convex relaxation of optimal power flow—Part I: formulations and
implications for intelligent vehicle-highway systems. Springer, Verlag Berlin relaxations. IEEE Trans Control Netw Syst 2014;1(1):15–27.
Heidelberg; 1994. [44] Li N, Chen L, Low SH. Exact convex relaxation of OPF for radial networks using
[33] Mahmassani HS. Dynamic network traffic assignment and simulation methodology branch flow model. In: Proc. Of IEEE SmartGridComm Conf.; 2012. p. 7–12.
for advanced system management applications. Network Spatial Econ 2001;1(3): [45] Münsing E, Mather J, Moura S. Blockchains for decentralized optimization of
267–92. energy resources in microgrid networks. In: 2017 IEEE Conference on control
[34] Nakayama S-i, Takayama J-i, Nakai J, Nagao K. Semi-dynamic traffic assignment technology and applications. CCTA); 2017. p. 2164–71. https://doi.org/10.1109/
model with mode and route choices under stochastic travel times. J Adv Transport CCTA.2017.8062773.
2012;46(3):269–81. https://doi.org/10.1002/atr.208. [46] Li Z, Guo Q, Sun H, Wang J. A new LMP-sensitivity-based heterogeneous
[35] Peeta S, Ziliaskopoulos AK. Foundations of dynamic traffic assignment: the past, decomposition for transmission and distribution coordinated economic dispatch.
the present and the future. Network Spatial Econ 2001;1(3):33–265. https://doi. IEEE Trans Smart Grid 2018;9(2):931–41.
org/10.1023/A:1012827724856. [47] Olver PJ, Shakiban C. Applied mathematics [Online]. Available:, Accessed May 14,
[36] Chen H-K. Dynamic travel choice models: a variational inequality approach. 2019, http://ft-sipil.unila.ac.id/dbooks/Applied%20Mathematics.pdf.
Springer; 1999. [48] Bliemer MCJ, Raadsen MPH, Brederode LJN, Bell MGH, Wismans LJJ, Smith MJ.
[37] Merchant DK, Nemhauser GL. Optimality conditions for a dynamic traffic Genetics of traffic assignment models for strategic transport planning. arXiv:
assignment model. Transport. Sci. 1979;12(3):200–7. 10.1080/01441647.2016.1207211 Transport Rev 2017;37(1):56–78. 10.1080/
[38] Ran B, Boyce D, LeBlanc L. A new class of instantaneous dynamic user-optimal 01441647.2016.1207211.
traffic assignment models. Oper. Res. 1993;41:192–202. https://doi.org/10.1287/ [49] OSM. Open street map of sioux falls [Online]. Available:, Accessed Jan. 24, 2019,
opre.41.1.192. https://www.openstreetmap.org/relation/194968.
[39] Zhou Z, Zhang X, Guo Q, Sun H. Supplementary material for “Analyzing power and [50] Khodr H, Olsina F, Jesus PDO-D, Yusta J. Maximum savings approach for location
dynamic traffic flows in coupled power and transportation networks [Online]. and sizing of capacitors in distribution systems. Electr Power Syst Res 2008;78(7):
Available:, Accessed June 17, 2020, https://github.com/ZhouZhe0825/DynamicT 1192–203. https://doi.org/10.1016/j.epsr.2007.10.002. http://www.sciencedirec
rafficSup. t.com/science/article/pii/S0378779607002143.
[40] He F, Yin Y, Lawphongpanich S. Network equilibrium models with battery electric
vehicles. Transport Res B-Meth 2014;67:306–19. https://doi.org/10.1016/j.
trb.2014.05.010.

12

You might also like