Optimal Energy Trading

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/JIOT.2019.2906186, IEEE Internet of
Things Journal
Optimal Energy Trading for Plug-in Hybrid Electric

Vehicles based on Fog Computing
Gang Sun, Feng Zhang, Dan Liao, Hongfang Yu, Xiaojiang Du, Senior Member, IEEE, Mohsen
Guizani, Fellow, IEEE
Abstract—A large number of plug-in hybrid electric vehicles

(PHEVs) have high mobility but a small battery capacity; thus, I. INTRODUCTION
these vehicles urgently need to make charging and discharging
decisions in real time. Our study proposes a new architecture
based on fog computing for an Internet of Vehicles energy ASdelivery
global energy demand continues to rise, traditional power
systems are encountering increasing difficulties in
trading system, which we call a vehicle-mounted energy fog. satisfying the growing energy trading requirements, which
This architecture includes a fog computing energy center include the ability to support large numbers of users, charging
(FCEC), which manages local energy trading and reduces the and discharging efficiency, and utility and safety. Moreover,
peak load energy trading for an external public energy rapid economic development and the acceleration of
company. We model the optimization problems for energy urbanization are leading to an increasing number of cars,
trading under two different types of FCECs: 1) a further exacerbating energy shortages and environmental
non-profit-driven FCEC whose goal is solely to benefit the pollution [1, 2]. Therefore, the world's leading auto producers
PHEV charging and discharging operations and 2) a have accelerated the deployment of new energy production
profit-driven FCEC whose goal is to maximize its own profits
technology and have developed national strategies to accelerate
while still guaranteeing that each PHEV achieves a
research on energy saving technologies and promote the
nonnegative utility. We also propose efficient algorithms for
increasing popularity of new energy electric vehicles [3].
these two types of FCECs to seek optimal pricing and make
supply-demand decisions. Simulation results show that our Compared with traditional vehicles, plug-in hybrid electric
proposed algorithms are superior to existing algorithms in vehicles (PHEVs) are both more efficient and friendly to the
terms of the convergence rate, the final objective value and the environment in terms of energy use. PHEVs are regarded as one
evenness of the Pareto solution set. Specifically, the evenness of of the cheapest future forms of urban transportation [4, 5]. One
the Pareto solution set is improved by 23% compared to the remarkable feature of PHEVs is that they are equipped with
results of the existing algorithm. converters that allow their batteries to act as mobile energy
devices in the power grid. In addition to drawing electricity
Index Terms—Plug-in hybrid electric vehicle; Energy from the grid, PHEVs can return electricity to the grid when
pricing; Charging and discharging decision; Fog computing;
required. In contrast, traditional purely electric vehicles (EVs),
Internet of Vehicles
hybrid electric vehicles (HEVs) and plug-in electric vehicles
(PEVs) all lack this feature. By joining the powerful Internet of
TABLE 1 Vehicles (IoV) [6,7], PHEVs can help to integrate the entire
DEFINITIONS OF ABBREVIATIONS process of distributed energy storage, transportation,
Abbreviation Definition distribution and consumption based on real-time, two-way
PHEV plug-in hybrid electric vehicle
flows of both electricity and information [8]. Thus, special
IoV Internet of Vehicles
trading models for the smart grid paradigm have been
DRM demand-response mechanism
V2G vehicle-to-grid
established in accordance with the demand-response
V2V vehicle-to-vehicle
mechanism ‎(DRM), which enables energy consumers and
V2B vehicle-to-building suppliers to flexibly dispatch their demand and supply [9]; one
CDC cloud data center example is the vehicle-to-grid (V2G) model, in which
VEF vehicle-mounted energy fog distributed PHEVs are used as mobile storage batteries to shift
FCEC fog computing energy center peak loads and achieve regional deployment.
EPEC external public energy company However, feeding energy back into the grid requires
ODS-IPM one-dimensional search - interior point method long-distance power transmission and distribution capabilities,
NSGA-II Non-dominated Sorting Genetic Algorithm II which result in higher scheduling costs and influence the
DP-NSGA-DLS dual population - NSGA - density local search
stability of the energy system. These negative factors are likely
NDX normal-distribution crossover operator
to reduce the operating profit of an external public energy
SBX simulated binary crossover operator
company‎ (EPEC) and hinder the development of V2G
Gang Sun and Hongfang Yu are with Key Lab of Optical Fiber Sensing and Communications
technology. This would cause PHEVs to be underutilized or
(Ministry of Education) and Center for Cyber Security, UESTC, Chengdu, P.R. China (e-mail: even useless. Therefore, addressing the development
[email protected]; [email protected]). bottlenecks of the V2G model is a significant challenge. There
Feng Zhang and Dan Liao are with Key Lab of Optical Fiber Sensing and Communications
(Ministry of Education), UESTC, Chengdu, China (e-mail: [email protected]; is a critical need to design a novel architecture for PHEVs to
[email protected]). reduce losses, relieve pressure and lower time costs in energy
Xiaojiang Du is with Department of Computer and Information Sciences, Temple University, trading.
Philadelphia, PA, USA (e-mail: [email protected]).
Mohsen Guizani is with Department of Computer Science and Engineering, Qatar University, Typically, there are many parked vehicles with surplus
Qatar (e-mail: [email protected]). energy concentrated at certain social hotspots, such as
2327-4662 (c) 2018 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.
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/JIOT.2019.2906186, IEEE Internet of
Things Journal
dedicated parking lots or underground parking garages at large The remainder of this paper is organized as follows. Section
shopping malls or office buildings, along with other vehicles in II introduces the related work. The system architecture and
urgent need of charging [10]. This motivates the need for a interactive process are elaborated in Section III. Section IV
vehicle-to-vehicle (V2V) model to realize real-time and presents the supply-demand formulation for charging and
low-loss energy trading. If there is no profit to be made, a discharging a PHEV. The optimization-problem models and
PHEV will not automatically provide the grid with services solution algorithms for non-profit-driven and profit-driven
such as recycling distributed excess energy or replenishing FCECs are explained in Sections V and VI, respectively.
supply when needed due to its rationality and self-interest. This Finally, Section VII concludes the paper.
fact further motivates a need to propose an incentive
mechanism that can comprehensively consider the interests of II. RELATED WORK
multiple entities, including the PHEVs and intermediaries
participating in energy trading. Specifically, the purpose of an A. Fog Computing based IoV
energy supplier is to adjust its energy availability based on
market requirements and the physical constraints of the energy It is estimated that approximately 50 billion ‘things’ will be
system [11], while an energy consumer modifies its demand in connected to the Internet by 2020 [12]. Transferring all data
response to the energy supply conditions. Due to the complex from all connected devices for processing in cloud data centers
space-time dimensions and balanced management requirements will require massive amounts of bandwidth and storage. Not all
of energy distribution, the question of how to handle energy things are connected to the controller via IP; instead, other IoT
pricing and charging and discharging allocation so as to industrial protocols [13] are also used, which necessitate a
optimize the profit of multiple entities poses another key translation process for processing or storing information from
challenge. IoT devices. The term ‘fog computing’, proposed in 2012 by
For a real-time perception and geographic distribution Cisco, is defined as a distributed computing paradigm that
architecture, cloud computing is obviously not the most extends computing power and data analysis applications to the
appropriate choice for providing communication and “edge” of the network [14]. The devices within the fog
environment are known as fog devices and can be deployed at
computation resources. A more advantageous alternative is
any location with connectivity to the network: on a power pole,
fog computing, which technically resides between the cloud on a factory floor, at the side of the road, in a vehicle, inside a
and Internet of Things (IoT) devices and handles processing shopping mall, etc. Fog computing relies on one or more
and storage tasks in close proximity to the user. Thus, we collaborative end users or near-user edge devices for
propose a novel system architecture based on fog computing performing storage, communication, control, configuration,
to support local V2V energy trading for PHEVs and reduce measurement and management functions [15] and offers the
the time cost and energy loss of detour charging in the V2G advantages of shorter delays, location-sensitive responses, and
scenario. To cope with multientity energy transactions under the ability to adapt to applications characterized by extensive
the new heterogeneous framework, we establish two geographic distribution and high mobility. Overall, fog
optimization models for different situations and propose computing contributes to solving various security, cognition,
corresponding algorithms. The main contributions of this paper agility, efficiency and delay problems.
are summarized as follows. An IoV environment requires big data processing, including
 In contrast to a remote cloud center, we propose a novel data acquisition, transmission, storage and computation [16].
system architecture based on fog computing for The IoV is a typical application scenario for fog computing, and
many researches have been conducted on this topic. The
real-time energy trading in the IoV and introduce a local
authors of [17] proposed a data analytics framework for fog
market manager, named the fog computing energy center
infrastructures in the fog layer of a traditional IoV architecture
(FCEC), to control energy trading. that offers context-aware real-time batch services at the edge of
 Considering typical real scenarios, we establish two the network. A feasible solution that enables offloading for
different optimization models, one for non-profit-driven real-time traffic management in fog-based IoV systems, aiming
FCEC and one for profit-driven FCEC, to maximize the to minimize the average response time for events reported by
interests of the two multilateral trading entities. vehicles, was put forward in [18]. To address location privacy
 For these two established models, we design a issues in the IoV, a privacy-preserved pseudonym (P3) scheme
single-objective optimization algorithm based on one- has been proposed in a new paradigm called the
dimensional search combined with the interior point fog-computing-supported IoV (F-IoV) [19]. Considering the
method (ODS-IPM) and a multiobjective optimization constraints on service latency, quality loss, and fog capacity,
algorithm (DP-NSGA-DLS) that is an improved version the process of task allocation across stationary and mobile fog
nodes in vehicular fog computing (VFC) has been formulated
of a well-known nondominated sorting genetic algorithm
into a joint optimization problem [20]. Furthermore, several
(NSGA-II).
papers (e.g., [21-24]) have studied the related energy saving
 We conduct extensive simulations and define several and security issues.
performance metrics to validate our scheme. The In summary, the distributed deployment of fog computing
numerical results verify the optimal pricing and energy nodes provides local computing power at the network edge,
allocation for PHEVs as well as the superiority of the enabling abundant IoV services for highly mobile vehicles,
proposed algorithms. including entertainment, safety alerts, traffic safety, analysis of
Things Journal
geographic information, and so on, while also achieving low decision-making has been confined by the grid environment
latency and context awareness. and has not yet been extended to IoV application scenarios in
urban areas. However, beyond charging stations, there are
B. Charging and Discharging Management many public locations with naturally high vehicle densities that
Charging and discharging management for energy trading offer considerable room for optimization to achieve distributed
with PHEVs are important concepts for the smart grid. They are energy management and balanced allocation, such as large
also a hot topic of research for future energy systems for the shopping mall parking lots and office building parking garages.
IoV. Researchers have performed a large amount of work in Therefore, this paper makes the essential contribution of
this area. Previous researches can be divided into three main introducing an FCEC to enable real-time V2V energy trading
categories: (1) charging station deployment strategies for for PHEVs at such social hotspots.
PHEVs [25-28], (2) ordered charging strategies and load
scheduling for multiple PHEVs [29,30], and (3) the joint III. SYSTEM ARCHITECTURE
optimization of the energy pricing strategy and distributed
scheduling for a V2G system [31-33]. A. Motivation
Considering comprehensive factors, the author of [25] At many social hotspots, such as dedicated parking lots,
focused on optimizing the power cost and energy management underground parking garages at large shopping malls or office
strategy for PHEVs. In [26], an optimal plan for distributing buildings and charging stations, many vehicles remain idle for
PHEV charging stations was proposed that considered both long periods of time. The authors of [34] classified these parked
demand-response programs and uncertainties. To reduce the vehicles into three types based on their energy states and travel
significant fluctuations in electricity consumption between plans, namely, vehicles that can complete their travel plans with
peak and off-peak periods in a given community, the author of surplus battery energy, vehicles that have insufficient energy
[27] proposed a novel software-defined network based but sufficient time to charge their batteries to complete their
approach for charging station allocation. The author of [28] trips, and vehicles that are underpowered but do not have time
introduced a type of heterogeneous IoV organized on the basis to charge their batteries. Obviously, the distribution of energy
of a wireless cellular network to collect charging station demand is extremely unbalanced at social hotspots with high
information and used oligopolistic game theory to seek a vehicle densities. Surplus energy exists that cannot be fully
balance between energy utility and owner preference factors. utilized. During the peak charging period, a large number of
For coordinated charging strategies, weight-aggregation PHEVs must wait in line for charging, and this scenario
multiobjective evolutionary algorithms were designed to presents the energy system with great challenges in terms of
optimize PHEV load scheduling in [29]. An autonomous power load and dispatch efficiency. Instead, most PHEVs
real-time charging strategy based on quadratic-constrained should have the option to adopt a more intelligent charging and
mixed integer linear programming for PHEVs was proposed in discharging mode—one that fully considers energy pricing and
[30]. A further consideration was to decentralize the grid mobility constraints to optimize energy transactions in both
infrastructure by using distributed EVs to support the grid, i.e., space and time.
a V2G scheme. This concept was originally presented in [31], Some scholars have considered the V2G model for balancing
in which the energy stored in the batteries of EVs was primarily energy distribution. However, the main grid is typically distant
used to reduce the peak electricity demand. A multilevel system from most social hotspots with high vehicle densities;
was established in [32] to capture charging and discharging consequently, transactions for repurchasing, centralizing and
behavior as well as energy trading between PHEVs and other reselling energy entail considerable energy loss. Meanwhile,
elements of the smart grid. This study utilized a double auction the PHEVs with surplus energy at these locations are not
mechanism to solve the problem of competition among large generally willing to contribute energy to other PHEVs in urgent
numbers of PHEVs. need of charging, which results in further mismatch of the
Even more applications for PHEVs are expected to emerge in energy supply and demand. Considering a time-sensitive and
the future. The V2G, vehicle-to-building (V2B) and V2V low-loss system, we propose a local V2V energy trading
models are all attracting considerable attentions from academia architecture for PHEVs based on fog computing that extends
and industry [33]. For these application scenarios, previous communication and computation resources to the proximity of
researchers have not clearly solved the problems of charge and end users. The proposed architecture enables a PHEV to
discharge decision-making and energy efficiency optimization connect to not only the EPEC but also the local energy network
for PHEVs. More specifically, many studies have merely maintained by the FCEC, thereby allowing PHEVs with surplus
proposed charge and discharge scheduling algorithms without energy to trade energy with other nearby PHEVs that require
seeking to optimize the overall performance of energy trading energy. The need to balance the interests of multiparty trading
at the system architecture level. In our work, we propose a new entities incents us to model and optimize the trading process for
architecture based on fog computing to close the distance the hybrid energy market.
between the control center and a terminal. Second, most
researches have focused merely on optimizing charging station
B. Fog Architecture Entities
deployment, while work focusing on actively attracting PHEVs
to participate in energy market balancing is lacking. Our As shown in Fig. 1, fog computing nodes are deployed at
models fully consider marginal effects and differential benefits social hotspots with high vehicle densities to provide wireless
to encourage more PHEVs to contribute surplus energy. In Internet services and V2V energy trading services for users
addition, research on PHEV charge and discharge within their coverage areas. For convenience, Table 1 defines
Things Journal
the abbreviations that are frequently used throughout this paper. wireless connections. Then, the FCEC sends messages to the
The proposed architecture mainly includes the following charging or discharging PHEVs based on the received
entities. certification messages (indicating whether the authentication
1) CDC: The cloud data center (CDC) is operated by an succeeded or failed) and classifies all PHEVs authenticated
authoritative organization that manages all system and entity within a given time slice into a group. If the prespecified time
registration and publishes the key materials used for system slice has not yet elapsed but the number of certified PHEVs has
initialization. reached a certain threshold, these certified vehicles will be
2) VEF: The vehicle-mounted energy fog (VEF) is an immediately divided into a group. On the other hand, when a
integrated entity abstracted from the distributed computing prespecified time slice has been completed but no PHEV has
infrastructure of the IoV to support energy trading. A VEF node authenticated as a buyer or seller, the time slice is doubled, and
is a fog node deployed at a social hotspot that provides so on. After certification, the PHEVs send information
communication and computation resources for the FCEC and concerning their energy demands or surpluses to the FCEC via
controls the normal operation and maintenance of the FCEC. wireless connections. After collecting all the information, the
3) PHEV: A PHEV is a new type of pollution-free, low-noise FCEC runs the energy trading scheduling algorithm to obtain
electric vehicle that is designed in accordance with green the optimal pricing and scheduling strategy. Finally, the FCEC
concepts of energy conservation and environmental protection. executes the transactions in accordance with the optimal
Equipped with a pluggable two-way charging-and-discharging decision.
interface, a PHEV is more efficient and more suitable for
EPEC Discharging PHEV FCEC Charging PHEV EPEC
operating in a city than a traditional gasoline-powered car.
Moreover, PHEVs play various roles in local V2V energy Register as an Register as an
energy supplier energy consumer
trading at social hotspots: a charging PHEV is an energy
consumer, while a discharging PHEV is an energy supplier. Note the energy surplus and the
EPEC’s energy prices
Each PHEV chooses its own role based on its current energy
state and travel plan when deciding whether to participate in Submit the Submit the
energy supply energy demand
energy trading.
Data flow
Energy flow Implement the optimal
Wireless Network pricing mechanism and
Charging PHEV scheduling algorithm
Discharging PHEV
Note the FCEC’s energy prices

CDC
and allocation strategy
VEF Node VEF Node PHEV discharges

PHEV discharges PHEV charges
to the FCEC
to the EPEC from the FCEC
PHEV charges
from the EPEC
FCEC EPEC FCEC

Fig. 2. The energy trading process among PHEVs
Social hotspot 1 Social hotspot 2
Fig. 1. The proposed energy trading system based on fog computing
IV. MODELING OF CHARGING AND DISCHARGING
4) EPEC: An EPEC is an electric utility company with a PHEVS
limited geographical coverage density. Some EPECs are
collocated with the CDC, which is generally distant from social
hotspots with high vehicle densities. The EPEC is the entity that A. Heterogeneous Energy Trading Market
is mainly responsible for the production, transportation and This paper proposes a V2V energy trading mechanism based
distribution of public energy: the EPEC provides energy to on fog computing, which serves as the foundation for a hybrid
charging PHEVs, and it can repurchase surplus energy from energy trading market. Based on the energy states and travel
discharging PHEVs. plans of their PHEVs, two types of users can be identified: 1)
5) FCEC: The FCEC is maintained by the VEF node Consumers, whose batteries are in urgent need of recharging,
deployed at a social hotspot that acts as an energy exchanging namely, charging PHEVs, are denoted by CV={1, 2, 3…, C},
site. The FCEC is in charge of collecting real-time information where C is the total number of charging PHEVs. 2) Suppliers,
concerning energy demand and supply from the PHEVs in that
whose batteries have surplus energy, namely, discharging
location. Moreover, the FCEC can implement an optimal
PHEVs, are denoted by DV={1, 2, 3…, D}, where D is the total
pricing mechanism and scheduling algorithm to ensure fair and
number of discharging PHEVs. We mainly focus on the
real-time energy trading while maximizing the benefits of
multilateral entities. incremental benefits gained during a fixed period, in which
each charging PHEV i∈CV receives a certain amount of energy
C. Interaction Process to meet its demand, denoted by di, while each discharging
As shown in Fig. 2, the PHEVs involved in a transaction first PHEV j∈DV contributes a certain amount of energy, denoted by
post their own authentication messages to the FCEC through dj, to obtain corresponding economic benefits.
Things Journal
During the process of energy trading, the charging and PHEV j∈DV, the amounts of energy contributed to the EPEC
discharging PHEVs are interlinked through the FCEC. Each
charging PHEV can buy energy not only from the EPEC, whose and the FCEC are denoted by aj and bj , respectively. The total
unit energy selling price is denoted by mout, but also from the amount of contributed energy cannot exceed the amount of
FCEC, whose unit energy selling price is denoted by nout. surplus energy dj available to be sold. This constraint can be
Similarly, each discharging PHEV can contribute its surplus formulated as shown in Equation (5):
energy to either the EPEC or the FCEC, whose unit energy
repurchasing prices are denoted by min and nin, respectively. a j  (bj  l j (bj ))  d j , j  DV , (5)
Thus, at a social hotspot with a high density of vehicles and a where l j (b j )   j (b j )2   j b j is the transmission loss when
VEF node, the FCEC serves as an energy aggregator and
exchanging station, to help PHEVs to stagger their charging discharging PHEV j sells energy to the EPEC;  j and  j are
during peak charging periods of the EPEC as well as to reduce two constants that represent loss factors. As mentioned before,
their long-distance energy consumption. It should be the unit energy repurchasing prices of the EPEC and FCEC
emphasized that the PHEVs accept the energy losses produced when buying energy from each discharging PHEV are min and
by local transactions, whereas the EPEC bears the energy losses nin, respectively. Thus, the income of discharging PHEV j can
produced by external trading. be expressed as shown in Equation (6):
B. Supply-Demand Modeling of a Charging PHEV I j (a j , bj )  a j min  bj nin , j  DV . (6)
This section describes the process of energy trading from the In a simple case, the discharging PHEVs can contribute
perspective of the energy consumers. For each charging PHEV energy only to the EPEC, and the net margin is I j (d j ,0)  d j min .
i∈CV, the amounts of energy bought from the EPEC and Then, the net margin of each discharging PHEV when selling
FCEC are denoted by ai and bi, respectively. The total amount energy to the FCEC can be expressed in Equation (7):
of purchased energy minus the local energy loss is equal to the
total energy demand. The formula is as follows: M j (a j , bj )  I j (a j , bj )  I j (d j ,0)
ai  (bi  li (bi ))  di , i  CV ,  (a j  d j )min  b j nin , j  DV . (7)

(1)
where li(bi) is the transmission loss when the charging PHEV i Mj(aj,bj) is the incremental energy trading gain of PHEV j
acquires energy from the FCEC. As mentioned earlier, the unit after introducing the FCEC. Each energy supplier is also
energy selling prices when each charging PHEV buys energy rational and self-interested. The purpose of participating in
from the EPEC and FCEC are mout and nout, respectively. Thus, local energy trading is to obtain a nonnegative economic
the total cost incurred by charging PHEV i when buying energy benefit, namely,
is given by Equation (2): M j (a j , b j )  0, j  DV . (8)
Pi (ai , bi )  ai mout  bi nout , i  CV . (2)
D. Different Types of FCECs
Consider a simple case in which each charging PHEV can
buy energy only from the EPEC; the cost of this energy demand The FCEC controls local energy trading between charging
and discharging PHEVs by adjusting the energy prices (nin, nout)
is Pi(di,0) = dimout, ∀i∈CV. Then, the net margin of each based on the demand {bi, ci} of each charging PHEV i and the
charging PHEV i when it buys energy from the FCEC can be supply {bj, cj} of each discharging PHEV j. As the scheduler for
expressed as shown in Equation (3): energy trading, the FCEC focuses on balancing the interests of
M i (ai , bi )  Pi (di ,0)  Pi (ai , bi ) the consumers and suppliers who are participating in local
energy trading. We mainly explore the relationship between the
 (di  ai )mout  bi nout , i  CV . (3) pricing mechanism and the supply and demand strategy for the
FCEC; therefore, we assume that the energy prices (min, mout) of
Mi(ai,bi) is the reduced energy trading cost of PHEV i after the EPEC remain fixed.
introducing the FCEC. Because every energy consumer is More specifically, relative to the specific energy pricing of
rational and self-interested, the purpose of participating in local the EPEC, the FCEC can approach the maintenance of the local
energy trading is to obtain a nonnegative net income, namely, energy trading market from two different perspectives: the
M i (ai , bi )  0, i  CV . (4) non-profit-driven perspective and the profit-driven perspective.
A non-profit-driven FCEC aims at balancing energy supply and
According to [35], the energy transmission losses in a local demand, fully invoking the potential of energy distribution, and
energy network are mainly due to transmission line resistance. maximizing the benefits of the PHEVs; whereas the purpose of
The author of [35] quantified the energy transmission loss by a profit-driven FCEC is to benefit from local energy trading
means of a quadratic function li(bi)=ai(bi)2+βibi, in which ai and while ensuring the interests of the PHEVs.
βi are constants that represent loss factors. In this research, we
adopt this function to quantify the energy losses caused by
energy trading between charging PHEVs and the FCEC. V. ENERGY TRADING OF A NON-PROFIT-DRIVEN FCEC
C. Supply-Demand Modeling of a Discharging PHEV A. Problem Modeling
This section describes the process of energy trading from the This section mainly considers the trading problem managed
perspective of the energy suppliers. For each discharging by a non-profit-driven FCEC that does not seek economic
Things Journal
benefits from the local market. Instead, a non-profit-driven Ui(·) and Uj(·) are both increasing and concave, and their initial
FCEC adjusts the energy prices (nin, nout) solely to benefit the values are Ui(0)=0 and Uj (0)=0. The increasing assumption is
PHEVs. However, it is important to ensure that the FCEC does based on the rationale that each PHEV will be more satisfied
not suffer economic losses; this constraint can be formulated as when it achieves a greater net margin during energy trading.
shown in Equation (9): The assumed concavity means that the marginal utility of each
PHEV decreases as the net income increases during energy
nout iCV bi  nin  jDV bj . (9) trading. A similar utility function, featuring both an increasing
Equation (9) means that the FCEC will not sell repurchased trend and concavity, has been applied in [35] and [36]. In this
energy at a loss. Once informed of the FCEC’s energy prices work, the utility functions for the charging and discharging
(nin, nout), each charging PHEV i must determine its energy PHEVs are U(x)=ln(1+x) and U ( x)  ln(1  x) , respectively. It is
demand allocations {ai, bi}, and each discharging PHEV j must worth noting that the non-profit-driven FCEC is the optimal
choose its energy supply allocations {aj, bj} to maximize the net case in which PHEVs can achieve the highest net gain from
benefit of both sides. The FCEC should maintain a local energy trading.
supply-demand balance when making scheduling decisions, as
B. Algorithm
shown in Equations (10):
In this section, first, two significant properties of Problem
iCV bi   jDV bj . (10)
(OP) are identified. Then, an efficient algorithm is proposed to
By combining Equations (9) and (10) with the constraints solve Problem (OP) based on these properties.
mentioned in Section IV, the optimization problem for the Property 1: When the optimal solution of Problem (OP) is
non-profit-driven FCEC, which seeks the maximum total utility obtained, Constraint (5) is strictly fulfilled, namely,
for the charging and discharging PHEVs, is formulated as
follows, where the capital letters “OP” stand for “optimization a j  (bj  l j (bj ))  d j , j  DV . (13)
problem”.
Property 2: When the optimal solution to Problem (OP) is
(OP): max iCV Ui (M i (ai , bi ))   jDV U j (M j (a j , b j )) *
obtained, nout =nin* always holds.
s.t. (1), (4), (5), (8), (9), (10) The two properties above are proven in Appendix I. Property
mout  nout (11) 2 indicates that the FCEC does not benefit from local energy
trading, which is consistent with the goal of benefiting only the
nin  min (12) PHEVs. Based on Property 1, for each discharging PHEV j, a j
can be replaced with b j to update the net margin (Equation (3))
var. (nin , nout ),{ai , bi }iCV ,{a j , bj } jDV of that PHEV, as shown in Equation (14):
In Problem (OP), the goal is to maximize the sum of the M j (bj )=(nin  min )bj  minl j (b j ) . (14)
transaction cost reductions for the charging PHEVs and the
increased incomes of the discharging PHEVs. Both Ui(·) and Similarly, for each charging PHEV i, ai can be replaced with
Uj(·) are logarithmic utility functions whose properties will be bi to update the net margin (Equation (7)) of that charging
explained later. Ui(·) represents the utility of charging PHEV i, PHEV, as shown in Equation (15):
which is related to the net margin Mi(ai, bi), while Uj (·)
M i (bi )=(mout  nout )bi  mout li (bi ) . (15)
represents the utility of discharging PHEV j, which is related to
the net margin Mj(aj, bj). Constraint (11) restricts the selling By combining Equations (14) and (15) and Property 2, Problem
price of the FCEC to be no higher than that of the EPEC (OP) can be converted to the following equivalent optimization
(otherwise, the charging PHEVs would be reluctant to problem (the capital “E” stands for “equivalent”).
participate in local energy trading). Similarly, Constraint (12)
restricts the FCEC’s repurchasing price to be no lower than that (OP-E): max iCV Ui (ci )   jDV U j (c j )
of the EPEC to induce discharging PHEVs to participate in s.t. (mout  n)bi  mout li (bi )  ci  0, i  CV (16)
local energy trading. Problem (OP) has three group decision
variables: 1) the FCEC’s energy prices (nin, nout), 2) each (n  min )b j  minl j (b j )  c j  0, j  DV (17)
charging PHEV’s energy demand allocations {bi, ci}i∈CV, and 3)
each discharging PHEV’s energy supply allocations {bj, cj}j∈DV.
di  bi  li (bi )  0, i  CV (18)
We use the notations (nin* , nout
*
) , {ai* , bi* }iCV , and {a*j , b*j } jDV d j  bj  l j (bj )  0, j  DV (19)
to represent the optimal solution of Problem (OP), where the
superscript “*” indicates optimality. Notably, Problem (OP) is a iCV bi   jDV bj (20)
typical nonconvex optimization problem; therefore, finding its
globally optimal solution is extremely difficult. Nonetheless, mout  n  min (21)
Problem (OP) is always feasible because all constraints are true var. n,{bi ,ci }iCV ,{bj ,c j } jDV
when bi = 0 and bj = 0, without introducing the FCEC.
Considering the rational constraints for consumers and In Problem (OP-E), the goal is to maximize the total utility of
suppliers, we assume that for each PHEV, its utility functions the charging and discharging PHEVs. Based on the properties
Things Journal
of the optimal solution, the equivalent transformation reduces Proof: Given n, all constraints of Problem (OP-EE-Sub)
the dimensionality of the decision variables. ci and cj represent form a convex feasible set. According to [37], Problem
the net margins of charging PHEV i and discharging PHEV j, (OP-EE-Sub), along with the convex feasible set and convex
respectively, and they are less than or equal to the net gain for objective function, is a convex optimization problem. ■
the optimal solution. Based on Property 2, the optimal energy In summary, Problem (OP-EE-Sub) is a convex optimization
price n is equal to both nout and nin. Constraints (18) and (19) problem with equality constraints and inequality constraints
ensure that the value of ai for each charging PHEV i and the that can be solved using the interior point method [35].
value of aj for each discharging PHEV j are both nonnegative. Therefore, for each n[min, mout], we propose an algorithm
Constraint (21) is deduced from Constraints (11) and (12). based on a one-dimensional search combined with the interior
The FCEC’s energy price n is a key factor in local energy point method, named ODS-IPM, for solving Problem (OP-EE)
trading. A smaller n means that the selling price is lower; and Problem (OP-E).
consequently, energy consumers gain a larger net margin,
which motivates charging PHEVs to buy more energy from the ODS-IPM algorithm: for solving Problem (OP-EE) and (OP-E)
FCEC. Conversely, a smaller n discourages discharging Input: [min , mout ] , optimization model.
PHEVs from contributing energy to the FCEC because an *
excessively low repurchasing price affects their net margin. Output: n , {bi* , ci* }iCV , {b*j , c*j } jDV .
Thus, the FCEC will have insufficient energy to sell to the 1: Initialization: set an appropriate step size  1 , let n  min , and
charging PHEVs. Such an imbalance of supply and demand
will make Constraint (20) invalid, which will eventually set a utility threshold 1 ;
adversely affect the interests of the charging PHEVs. 2: while n<mout do
Consequently, either an excessively small or an excessively 3: Use the interior point method to solve Problem
large n runs counter to the interests of both the charging and (OP-EE-Sub) based on the input and record the optimal
discharging PHEVs. Hence, the FCEC needs to properly solution as {bi , ci }iCV and {b j , c j } jDV ;
regulate the local energy price.
However, it is difficult for the FCEC to find the optimal 4: if V (n)  1 , then
Let n  n , 1  V (n) , {b*j , c*j } jDV  {b j , c j } jDV ;
energy price because Problem (OP-E) is a nonconvex *
5:
optimization problem [37] associated with three sets of decision
6: end if
variables: n, {bi , ci }iCV , and {b j , c j } jDV . To solve Problem
7: Let n  n   1 ;
(OP-E), we present the following proposition.
8: end while
Proposition 1: For each charging PHEV i  CV , if di  bi
*
*
9: return n , {bi* , ci* }iCV , {b*j , c*j } jDV
always holds, then Problem (OP-E) has the same optimal
solution set as the following problem: As shown above, the main idea of the ODS-IPM algorithm is
(OP-EE): max iCV Ui (ci )   jDV U j (c j ) to perform a one-dimensional search with a step size of 1 over
a given interval [min, mout] for n (the WHILE loop from Line 2
s.t. (16), (17), (19), (20), (21) to Line 8). For each n, the ODS-IPM algorithm obtains the
di  bi  0, i  CV (22) utility value V(n) by solving Problem (OP-EE-Sub) using the
interior point method (line 3). In Line 4, the ODS-IPM
var. n,{bi , ci }iCV ,{bj ,c j } jDV
algorithm obtains the optimal solution, denoted by {bi* , ci* }iCV
*
The above proposition is proven in Appendix II. Here, bi is and {b*j , c*j } jDV , based on the output of Line 3. Once bi and
*
* *
the optimal amount of energy for charging PHEV i to buy from b j have been obtained, a for charging PHEV i and a j for
*
the FCEC. Proposition 1 indicates that Problem (OP-E) can be discharging PHEV j can be derived according to Property 1 and
solved by solving Problem (OP-EE), in which each charging Constraints (1) and (5). Meanwhile, the FCEC’s energy prices
PHEV’s energy demand is sufficiently large that it needs to buy satisfy nin  nout  n . Thus, Problem (OP) can be solved.
* * *
energy from both the FCEC and the EPEC. This is a common The computational complexity of the ODS-IPM algorithm
case in reality. mainly depends on the efficiency of the one-dimensional search
Given that the FCEC’s energy prices satisfy n[min, mout], and the interior point method. The time complexity of the
indicating that the decision variable n has a fixed value, one-dimensional search is O(logn), and the search space for n,
Problem (OP-EE) can be solved by solving the following which is the non-profit FCEC’s energy price within an interval
subproblem: [min, mout], is small in Problem (OP-EE-Sub). The interior point
method used in the ODS-IPM algorithm mainly relies on a
(OP-EE-Sub): V(n)= max iCV Ui(ci )+ jDV U j(c j ) penalty function (obstacle function) to impose a high “obstacle”
s.t. (16), (17), (19), (20), (22) at the boundary of the feasible domain. When the iteration point
is close to the boundary, the objective function increases to
var. {bi , ci }iCV ,{b j ,c j } jDV prevent the iteration point from crossing the boundary and
To solve Problem (OP-EE-Sub), we present the following ensure that the solution remains within the feasible domain. The
proposition. interior point method has polynomial time complexity and
Proposition 2: Given the FCEC’s energy prices n[min, mout], possesses advantages in terms of convergence and complexity.
Problem (OP-EE-Sub) with the decision variables {bi , ci }iCV
and {b j , c j } jDV is a convex optimization problem. C. Simulation Results and Analysis
Things Journal
In this section, we test the fairness index, average energy loss decreases, thus confirming the previous findings. Moreover, the
ratio and time cost to verify the low computation complexity, differences among the utilities of each PHEV are small,
strong fairness and energy savings of the designed architecture indicating that the transactions are relatively fair.
1.2
and approach. Furthermore, we present numerical results to
validate the influence of the number of vehicles, the market charging PHEV
1.0 discharging PHEV
price and the demand on the optimal charging and discharging
The total utility of the PHEVs

allocation and the maximal utility of the PHEVs, thus further 0.8
proving the validity and feasibility of our proposed model for a
non-profit-driven FCEC. 0.6
The experiments were conducted using MATLAB on a PC

0.4
with an Intel Core i5 CPU and 8 GB of RAM. According to [34],
the battery capacity of a typical PHEV is approximately 20-30 0.2
kWh. For example, the battery capacity of BYD’s latest PHEV
is 23 kWh. Therefore, for each PHEV in the simulation, the 0.0
(min, mout)=(0.6,1.0) (min, mout)=(0.7,1.0) (min, mout)=(0.8,1.0)
demand and supply are both set to 10 kWh. Furthermore, n*in= n*out=0.794 n*in= n*out=0.838 n*in= n*out=0.896
according to [38], the typical selling and repurchasing prices of
the EPEC at peak times in a V2G system are 0.6 yuan/kWh and Fig. 4. Each PHEV’s utility when the FCEC sets the optimal price n*.
1 yuan/kWh, respectively. Similar to [35], the energy loss
functions’ constant parameters {ai}i∈CV and {aj}j∈DV were set To further analyze the fairness of the local energy trading, we
to random values in the interval [0.0025, 0.0075], and we set define a fairness index [39], denoted by FI and calculated as
i   j  0.005, i  CV , j  DV . shown in Equation (23):
 iCV ci*   jCV c*j  .

2
Fig. 3 depicts the relationship between the total utility of the 1
PHEVs participating in energy transactions and the FCEC’s FI= (23)
| CV |  | DV | iCV (ci* ) 2   jCV (c*j ) 2
energy price. To provide a clear demonstration, this numerical
example includes five charging PHEVs and five discharging Equation (23) is derived from the average inequality. FI is a
PHEVs. Three different sets of energy prices (min, mout) for the positive number that is always less than 1. If and only if
EPEC are tested: (0.6,1.0), (0.7,1.0) and (0.8,1.0). When the ci*  c*j , i  CV , j  DV , FI=1 holds. A value of FI close to 1
PHEVs acquire the maximum total utility, the FCEC’s optimal indicates that the net gains of all PHEVs are more similar to
energy prices are nin* =nout *
=n*=0.794 , nin* =nout
*
=n*=0.838 , and each other, it means that the transactions are more equitable.
* * * 1.00
nin =nout =n =0.896 , respectively. A comparison of the curves
reveals that both the smallest n and the largest n, which are 0.99
equal to the EPEC’s prices, are detrimental to the total utility of 0.98
all PHEVs; these findings support the previous argument.

The fairness index
0.97
Moreover, the larger the difference is between the EPEC’s min=0.6, mout=1.0
0.96 min=0.7, mout=1.0
selling and repurchasing prices, the greater the utility that the min=0.8, mout=1.0
0.95
PHEVs can obtain. One reason may be that a greater EPEC
price variance allows the FCEC more freedom to choose the 0.94
optimal energy price to benefit the charging and discharging 0.93
PHEVs. Stated otherwise, for a given FCEC energy price, a 0.92

1 2 3 4 5 6 7 8 9 10 11 12
greater EPEC price variance allows two trading parties to gain a The number of charging (discharging) PHEVs
greater net income. Fig. 5. The fairness of the PHEVs’ profits vs. the number of PHEVs.
14
13
min=0.6, mout=1.0 nin*=nout*=0.794 25
12
min=0.7, mout=1.0 nin*=nout*=0.838 min=0.6, mout=1.0
11
10 min=0.8, mout=1.0 nin*=nout*=0.896 min=0.7, mout=1.0

20
9 min=0.8, mout=1.0
8
7 15
6
5
10
4
3
2 5
1
0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00
The FCEC's energy price n 0
1 2 3 4 5 6 7 8 9 10 11 12
Fig. 3. The PHEVs’ total utility vs. the FCEC’s energy price. The number of charging (discharging) PHEVs
Fig. 6. The total utility of PHEVs vs. the number of PHEVs.

Fig. 4 depicts the maximum utility of each PHEV under three
different types of energy prices from the EPEC. Each PHEV’s As shown in Fig. 5, the fairness of the PHEVs’ profits first
utility decreases as the EPEC’s energy price difference decreases as the number of PHEVs increases but then levels off.
Things Journal
A comparison of the three curves for (min, mout) values of reason is that the wider price range of the EPEC makes the
(0.6,1.0), (0.7,1.0) and (0.8,1.0) reveals that they tend to search space for the optimal local pricing larger, and thus, the
stabilize at approximately 0.988, 0.976, and 0.931, respectively, search takes more time.
indicating that when the EPEC’s price variance (i.e., mout-min) is Fig. 9 depicts how the total utility of the PHEVs varies with
larger, the fairness of local trading becomes much better. the energy demand (supply) of the charging (discharging)
Fig. 6 depicts how the PHEVs’ total utility changes with the PHEVs. With an increase in the demand di (or supply dj), the
number of PHEVs under three different energy pricing schemes. total utility of the PHEVs also increases because more energy is
As shown in the figure, the total utility of the PHEVs increases exchanged in the local trading process. A comparison of the
as the number of PHEVs increases because when more energy
three curves reveals that the larger the EPEC’s energy price
can be traded, the net gain of the PHEVs increases. Moreover,
variance is, the more utility the PHEVs can obtain, and the rate
the larger the EPEC’s energy price variance is, the more utility
of increase is also much larger. The primary reason for these
the PHEVs can obtain.
findings is that an excessively small price variance (i.e.,
0.08
mout-min) results in a small profit space; therefore, the cost of
min=0.6, mout=1.0
0.07 compensating for the local energy loss when a large amount of
min=0.7, mout=1.0
energy is exchanged through local transactions is high.
The average energy loss ratio
0.06 min=0.8, mout=1.0

12
0.05
11 min=0.6, mout=1.0
0.04 min=0.7, mout=1.0
10

min=0.8, mout=1.0
0.03 9
0.02 8
7
0.01
6
0.00
1 2 3 4 5 6 7 8 9 10 11 12 5
The number of charging (discharging) PHEVs
4
Fig. 7. Average energy loss ratio of PHEVs vs. the number of PHEVs. 3
5 6 7 8 9 10 11 12 13 14 15
The energy demand (supply) of each PHEV
To analyze the average energy loss ratio of the PHEVs in Fig. 9. Total utility of PHEVs vs. the energy demand (supply) of each
local energy trading, the following metric, denoted by AL, is PHEV.
defined:
18
 
2
iCV fi (bi )   jCV fi (bi )
* *
1 16
AL  . (24)
| CV |  | DV | iCV (bi* )2   jCV (bi* )2 14
12
As shown in Fig. 7, as the number of PHEVs increases, the 10
average loss ratio fluctuates within only a small range, and the 8
overall loss ratio remains relatively stable. Comparisons 6
between the three sets of EPEC prices show that they have an 4
insignificant influence on the consequent energy transmission 2

loss ratio. 0
100 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
The EPEC's energy repurchasing price min
90 min=0.6, mout=1.0
The time cost of the ODS-IPM algorithm
(a)
80 min=0.7, mout=1.0
14
min=0.8, mout=1.0
70
12
60
50 10
40
8
30
20 6
10
4
0
1 2 3 4 5 6 7 8 9 10 11 12 2
The number of charging (discharging) PHEVs
0
Fig. 8. Time cost of ODS-IPM algorithm vs. the number of PHEVs. 0.6 0.7 0.8 0.9 1.0 1.1 1.2
The EPEC's energy selling price mout
(b)
As shown in Fig. 8, the time cost for determining the solution
in which the participating PHEVs gain the maximal utility is Fig. 10. Total utility of PHEVs vs. the EPEC’s energy prices.
roughly linear in the number of PHEVs, demonstrating that the
time complexity of the proposed algorithm is polynomial and Fig. 10 depicts the relationship between the total utility of the
acceptable. In addition, the larger the EPEC’s price variance is, PHEVs and the energy prices of the EPEC. As shown in Fig. 10
the higher the time cost of the ODS-IPM algorithm is; the (a), when the energy selling price mout of the EPEC is fixed, the
Things Journal
total utility of the PHEVs decreases as the EPEC’s energy (OP-MOS): min f1 (B)  (nout iCV bi  nin  jDV b j )
repurchasing price min increases. This is because that the valid
range for the FCEC’s energy price n is smaller, and the net gain min f 2 ( B)  (iCV Ui (M i (bi ))   jDV U j (M j (b j )))
from local trading decreases as the price variance (i.e., n-min) (mout  nout )bi  m out li (bi )  0, i  CV , p  1, 2,..., C
decreases. Similarly, in Fig. 10 (b), when the energy (n  m )b  m l (b )  0, j  DV , p  C  1,..., C  D
repurchasing price min of the EPEC is constant, the total utility  in in j in j j
bi  di  0, i  CV , p  C  D  1,..., 2C  D
of the PHEVs increases as the EPEC’s energy selling price mout 
increases. This behavior occurs because the valid range for the bi  li (b i )  0, i  CV , p  2C  D  1,...,3C  D (31)

FCEC’s energy price n becomes larger and the net gain of local g p ( B)= d j  b j  l j (b j )  0, j  DV , p  3C  D  1,...,3C  2 D
b  0, j  DV , p  3C  2 D  1,...,3C  3D
trading increases as the price variance (i.e., mout-n) increases.  j
nout  mout  0, p  3C  3D  1

VI. ENERGY TRADING OF A PROFIT-DRIVEN FCEC min  nin  0, p  3C  3D  2

nin  nout  0, p  3C  3D  3
A. Problem Modeling
This section mainly considers the trading problem managed h( B)  iCV bi   jDV b j  0 (32)
by a profit-driven FCEC, which must control the local energy T
Here, B={nin, nout, {bi}i∈CV, {bj}j∈DV} is a (D+C+2)-dimensional
pricing to balance the interests of the charging and discharging decision variable, and gp(B) is the standardized constraint set,
PHEVs while optimizing its own profits. This issue is a typical with p being the index of the constraints. Constraint (32) is the
multiobjective optimization problem in which the first standardized version of Constraint (28), which ensures the
optimization goal is to maximize the net income of the EPEC supply-demand balance.
and the second one is to maximize the total utility of the PHEVs
B. Algorithm Design
during local trading. Because the FCEC also seeks to maximize
its own economic benefits, Property 2, presented in Section V, In profit-driven local energy trading, conflicts of interest
is no longer true; however, the remaining constraints must still exist between the FCEC and the PHEVs. Hence, there is no
be satisfied. Thus, the optimization problem for a profit-driven optimal solution that can achieve both maximization goals of
FCEC is modeled as follows, where “OP” stands for Problem (OP-MOS); however, the solution approach can seek a
“optimization problem” and “MO” stands for “multiobjective”. set of noninferior solutions (also known as nondominated
solutions or Pareto solutions) [37]. To obtain the Pareto
(OP-MO): max nout iCV bi -nin  jDV b j solutions for Problem (OP-MOS), we improve the Nondomin-
max iCV Ui(M i(bi ))+ jDV U j(M j(b j )) ated Sorting Genetic Algorithm II (NSGA-II) [40], a fast
algorithm with an elitist strategy, which is one of the most
s.t. （mout -nout）bi  mout（
li bi） 0, i  CV (25) popular multiobjective genetic algorithms. The improved
version of this algorithm is called DP-NSGA-DLS, an
（nin  min）b j  min（
l b j） 0, j  DV (26) abbreviation for dual population - NSGA - density local search.
d j  bj  l（
j b j） 0, j  DV (27) Compared with the original NSGA-II algorithm, the
improvements are as follows: 1) We introduce a normal-
iCV bi   jDV bj (28) distribution crossover operator (NDX) in place of the original
simulated binary crossover operator (SBX); 2) We add a
d i  bi  0, i  CV (29) population of infeasible solutions, resulting in dual populations;
mout  nout  nin  min (30) 3) We add a local search of the sparse solutions in each iteration.
The details of the improvements are described below.
var. nin , nout ,{bi }iCV ,{bj }jDV Normal-Distribution Crossover Operator
In Problem (OP-MO), the first objective function seeks to To enhance the spatial search capability of the NSGA-II
maximize the net gain of the FCEC, which is obtained by algorithm, we replace SBX with NDX. In [41], in a
subtracting its costs from its revenues; the second objective one-dimensional search, NDX generated a better exploration
function seeks the maximum total utility of the charging and space than SBX did, enabling it to escape local optima and
discharging PHEVs. Based on Property 1, Constraints (25) and obtain Pareto solutions faster. Based on the chromosome vector
(26) guarantee that the net gains of the charging and B={nin, nout, {bi}i∈CV, {bj}j∈DV}T, which consists of several
discharging PHEVs are nonnegative. Constraint (27) indicates variables (called genes), we first select two parent
that the amount of energy sold to charging PHEVs from chromosomes B1 and B2 that have different genes, such as
discharging PHEV j can be no greater than the amount of different nin or nout values. Then, we cross them to form
available energy of this discharging PHEV. Constraint (28) daughter chromosomes. The computational procedure of NDX
represents the law of supply and demand equilibrium. As in for gene i is described as follows.
Proposition 1, Constraint (29) represents the common case in 1) Generate a random number u ∈ U(0,1), where U(0,1)
which each charging PHEV’s energy demand is sufficiently represents the uniform distribution on the interval [0,1].
large that it needs to buy energy from both the FCEC and the 2) If u  0.5 , then the daughter chromosomes are:
EPEC. To match the standard form of a convex optimization
B +B 1.481( B1,i  B2,i ) | i |
problem [38], the original maximization-based optimization Bˆ1,i = 1,i 2,i 
goals are reformulated as minimization-based goals. Thus, 2 2 . (33)
B + B 1.481( B  B2,i ) | i |
Problem (OP-MO) is converted into the following Problem Bˆ 2,i = 1, i 2 , i
 1, i
(OP-MOS), where the capital letter “S” stands for “standard”: 2 2
Things Journal
3) If u  0.5 , then the two daughter chromosomes are: Let us assume that the sum of the population is T, meaning that
B +B 1.481( B1,i  B2,i ) | i | there are T solutions, and let the objective vector for the ith
Bˆ1,i = 1,i 2,i  solution Bi to be denoted by (f1(Bi), f2(Bi)). Then, the normalized
2 2 . (34) formula for the target function is as follows:
ˆB = B + B 1.481( B  B2,i ) | i |
1, i 2 , i
 1, i
f j ( Bi )  f j min
, j   1,2  ,
2, i
2 2 f j ( Bi )  (36)
In the above NDX formulas, i  N (0,1) is a random variable f j max  f j min
that follows a normal distribution, the index i indicates different
where fjmin and fjmax are the maximum and minimum values,
genes, and B̂1 and B̂2 are the two new individuals that result respectively, of the jth objective function. After normalization,
from applying the NDX crossover operation on parent chrom- the degree of sparsity of Bi can be calculated as
osomes B1 and B2.
ti
Infeasible Solution Population SP( Bi )  , (37)
Constraint-processing technology plays an important role in T
multiobjective optimization algorithms. Currently, the main where T is the number of solutions and ti is the number of target
constraint-processing method is the penalty function method vectors whose Euclidean distances from the objective function
[37]. However, the basic penalty function method is not f(Bi) are less than r (0 < r <1). Similar to [42], we set r = 0.1.
particularly effective for multiple optimization targets. SP(Bi) represents the degree of sparsity of solution Bi and can
Therefore, some scholars have proposed the idea of using a dual help to determine a better solution search direction.
population structure [41] that includes an infeasible solution A search algorithm based on sparse solutions has two
population. The dual population storage mechanism avoids advantages. One is that it effectively enhances the search
direct comparisons between feasible and infeasible solutions, process in the space near sparse solutions when the solutions
and it retains a certain number of infeasible solutions to are unevenly distributed. The other is that the minimum density
promote better population diversity. We introduce an infeasible corresponds to the solutions at both ends of the distribution, and
solution population into NSGA-II to obtain the improved thus, focusing on sparse solutions allows the algorithm to
DP-NSGA algorithm, which searches a wider space. For the explore beyond the edges of the current solution region, thus
evaluation of the infeasible solutions, we propose the following increasing the solution space.
two definitions. To further improve the DP-NSGA algorithm, a mutation
Definition 1: A violating function measures the degree to strategy based on extremal optimization [42] is used to produce
which an infeasible solution violates the constraints of the a new population around each sparse solution to enhance the
problem. Such a function can be formulated as shown in solution accuracy. To obtain each local solution, it is necessary
Equation (35): to change only one decision variable in the sparse solution
3C  3 D  3 when using the extremal-optimization-based mutation
G ( B)   (max(0, g p ( B))2  (max(0,| h( B) |))2 (35) algorithm. We denote the current sparse solution by
p 1
where gp(B) is the standardized constraint set, with p being the B=(b1,b2,,bk), where k is the number of decision variables.
index of the constraints, and h(B) is the supply-demand balance Then, the number of new solutions generated by the
constraint. We use the value of the above expression to evaluate extremal-optimization-based mutation strategy is k, and the
the degree of violation of an infeasible solution, i.e., the degree variation formula is as follows:
to which gp(B) and h(B) are not met. For a feasible solution, Bi  (b1 ,..., bi,..., bk ), 0<i  k
G(B) is equal to 0.
bi  bi     max (bi ), 0<i  k
Definition 2: An excellent infeasible solution is an infeasible
solution BIF with the following properties: i) ∃B∈P such that  1
 (2 )  1, if   0.5
q 1
BIF < B or B  P such that B < BIF; and ii) G ( B )   . Here, (38)
  1
P is a Pareto-optimal solution set, B is a Pareto-optimal solution, 
and  is a small constant that represents the acceptable degree 1  [2(1   )] q 1
, otherwise
of violation. In this paper, we set ε = 0.5.  max (bi )  max(bi  llimit , ulimit  bi ), 0<i  k
Based on the above two definitions, several excellent
infeasible solutions are chosen and stored as an additional Here, bi is the ith decision variable, which is changed to generate
population; then, these solutions are used in combination with solution Bi, and bi is its optimal mutation; λ∈U(0,1) is a
the feasible solution population in the next iteration to produce random number that follows a uniform distribution; q is a
individuals with better traits. positive real number called the shape parameter, which is set to
Local Search of Sparse Solutions 11, in accordance with [42]; and βmax(bi) is the maximal
A multiobjective problem has more than one optimal perturbation allowed between the original and mutated values
solution; thus, it is necessary to guarantee the distribution of the decision variable bi.
properties of its solutions. Here, we evaluate a solution based In the extremal optimization variation method, only one
on its density. A solution with a low density in the Pareto set is decision variable is changed at any time. This approach endows
defined as a sparse solution. Before defining a formula for the the method with a strong fine-tuning ability but a small search
degree of sparsity, we need to normalize the objective functions scope. To increase the algorithm’s convergence rate, a local
to obtain their standard Euclidean distance, which can be used random search strategy is used to generate a new population.
to determine the degree of similarity between two target values. We denote the current sparse solution by B=(b1, b2,, bk) and
Things Journal
the total number of solutions in the population by T. Similar to 6: while it  MaxIt do

[43], the number of local solutions is set to 20% of the total 7: Perform selection, crossover, and mutation for Fp :
number of solutions in the population (rounded up to the next
popfs  tournament _ selection  popf , nPop / 2, 2  ,
integer if necessary). The mutation formula for the local
random search is as follows: Fs  genetic _ operator  popfs,1,11, l _ limit , u _ limit  ;
B =(b1,...,bi,..., bk ),  =(1,2,..., 0.2T  ) 8: Cross Fp and Fq :
bi   bi , (i  1, 2,..., k ), 0<  1.2

. (39) Fd  mix _ pop  popf , popc,1, 11 ;
9: Identify sparse solutions in Fp and then perform a local
In (39), bi is the mutated value of decision variable bi, and γ
search:
∈U(0,1.2) follows a uniform distribution.
F1  popf  Ff {1} ,
Together, the extremal-optimization-based mutation
algorithm and the local random search strategy described above Fm  local _ search  F1, nPop, l _ limit, u _ limit  ;
produce a total of k+0.2T local solutions. 10: Combine the populations Fm , Fs , Fd , and Fp :
At this point, the NSGA-II algorithm has been improved into
the DP-NSGA-DLS algorithm by replacing the SBX operator Fn  [ Fm Fs Fd Fp ] ;
with the NDX operator, introducing a dual population structure 11: Separate the feasible solutions and excellent infeasible
and adding a local search. The entire process of the DP-NSGA- solutions in Fn into different populations:
DLS algorithm is described below.  popf _ child , popc _ child   find _ islegal  Fn  ;
Problem (OP-MOS) has two objective functions, and
accordingly, the time complexity of the DP-NSGA-DLS Fq  check _ not _ fea  F1, popc _ child  ;
algorithm has two main components, as follows. First, the time Fp   popf ; popf _ child  ;
complexity of executing competitive selection among the two 12: Repeat Steps 3 and 4;
groups of individuals is O(NlogN), where N is the total number
13: Truncate Fp and Fq in accordance with the sorted
of individuals in both groups. Second, the time complexity of
producing new individuals from both populations by applying solutions:
the crossover and mutation operations is O(N). Subsequently, Fp  Fp 1: min  end , nPop   ;
the time complexity of selecting excellent individuals is
Fq  Fq 1: min  end , nPop / 4   ;
O(mN2), where m is the number of objective functions; here, m
is equal to 2. Therefore, the time complexity of the 14: Let it  it  1 ;
DP-NSGA-DLS algorithm is O(N2), which is the same as that 15: end while
of the NSGA-II algorithm. Furthermore, the improved 16: return Pareto solution set of Problem (OP-MOS)
DP-NSGA-DLS algorithm has various advantages over the
original NSGA-II algorithm, such as its improved spatial search C. Simulation Results and Analysis
capability, population diversity, solution uniformity and
convergence rate, enabling it to escape local optima and obtain In this section, we implement the NSGA and
Pareto solutions faster. DP-NSGA-DLS algorithms to solve Problem (OP-MOS). The
goal of Problem (OP-MOS) is to maximize two targets; thus, a
DP-NSGA-DLS Algorithm: for solving Problem (OP-MOS) solution closer to the outside of the corresponding
two-dimensional diagram is better. We compare the two
Input: Population size nPop , mutation probability mp , algorithms’ Pareto planes, time costs and convergence to verify
crossover probability cp , optimization model. the superiority of our proposed DP-NSGA-DLS algorithm.
Output: Pareto solution set of Problem (OP-MOS). Moreover, we define a spacing metric to compare the
1: Initialization: Let nPop=N , mp  0.1 , cp  0.9 , and it  1 , distribution characteristics of the two algorithms’ solution sets
where it is the current number of iterations, and set the and present graphical results to facilitate the analysis of the
mutation step to  and the maximum number of iterations relationship between the two optimization objectives of
Problem (OP-MOS).
to MaxIt ;
The 1000th iteration
2: Generate the initial population: 14
pop  repmat  empty _ individual , nPop,1 ; DP-NSGA-DLS
12 NSGA-II
3: Separate feasible solutions and excellent infeasible solutions
10
into their own populations:
The FCEC's total profit
 popf , popc  find _ islegal  pop  , Fp  SortPopulation  popf  , 8
Fq  SortPopulation  popc ; 6
4: For the initial feasible solution population Fp : 4
 popf , Ff   NonDominatedSorting  popf  , 2
popf  CalcCrowdingDistance  popf , Ff  ; 0

0 1 2 3 4 5 6 7 8 9
5: To obtain an excellent infeasible solution population Fq : The total utility of the PHEVs
popc  CalcCrowdingDistance  popc, Fc  ; Fig. 11. A comparison of the Pareto planes.
Things Journal
Fig. 11 compares Pareto solution sets obtained by the obtained by DP-NSGA-DLS is larger than that obtained by
DP-NSGA-DLS and NSGA-II algorithms after 1000 iterations. NSGA-II.
As seen from the figure, the Pareto plane of DP-NSGA-DLS As shown in Fig. 13 (b), the FCEC’s average total profit in
lies significantly closer to the outside of the diagram than the the Pareto solution set obtained by DP-NSGA-DLS always
Pareto plane of NSGA-II does, meaning that the values exceeds that in the Pareto solution set of NSGA-II. As in the
obtained by DP-NSGA-DLS for both objectives are case of the average total utility of the PHEVs, DP-NSGA-DLS
substantially larger. Moreover, the relationship between the has converged after approximately 600 iterations, whereas
profit-driven FCEC’s total profit and the total utility of the NSGA-II has not yet converged. At 1000 iterations, the value
PHEVs is negatively correlated, indicating that the FCEC’s achieved by DP-NSGA-DLS is better than that achieved by
total profit decreases as the PHEVs’ total utility increases. This NSGA-II, demonstrating the superiority of the improved
is consistent with common sense. In summary, the results of the algorithm.
14
improved DP-NSGA-DLS algorithm are better than those of DP-NSGA-DLS
NSGA-II. This finding can be attributed to the introduction of a 12 NSGA-II
population of infeasible solutions, which allows the infeasible 10

solutions and feasible solutions to be crossed. Therefore, some 8
of the good genes in infeasible solutions near the edges of
6
infeasible regions can be inherited by the next generation. In
contrast, NSGA-II uses penalty function by which all infeasible 4
solutions are immediately eliminated. Hence, the genes of 2

infeasible solutions are difficult to preserve in the next
0
generation. 0 1 2 3 4 5 6 7 8 9
To evaluate the evenness of the Pareto solution set [40], a (a) The 1st iteration
spacing metric is defined, as shown in Equation (40): 14
DP-NSGA-DLS
12 NSGA-II
1 | F1 |
S  (d  di ) ,
2 (40)
| F1 | 1 i 1 The FCEC's total profit 10
8
where P is the achieved Pareto solution set and d is the
average of all di values. If the spacing metric is equal to 0, the 6
solution set is completely uniform: 4

2
di  min j {  | f m ( Bi )  f m ( B j ) |}, Bi , B j  P, i, j = 1,2,...,| P | . (41) 2
m 1
0
For the Pareto solution sets generated by the NSGA-II and 0 1 2 3 4 5 6 7 8 9
DP-NSGA-DLS algorithms after 1000 iterations, the values of (b) The 250th iteration
the spacing metric are 0.0674 and 0.0516, respectively, 14
indicating that DP-NSGA-DLS achieves a 23% improvement 12

DP-NSGA-DLS
NSGA-II
over NSGA-II.
10
Fig. 12 compares the Pareto solution sets generated by the

two algorithms after 1, 250, 500 and 750 iterations. A clear gap 8
appears between the results of the two algorithms at 250 6

iterations. The Pareto solution set obtained by the
4
DP-NSGA-DLS algorithm is far superior to that obtained by
NSGA-II. At 500 iterations, the Pareto solution set generated 2
by the DP-NGSA-DLS algorithm is approaching uniformity, 0

0 1 2 3 4 5 6 7 8 9
while the Pareto solution set generated by the NSGA-II The total utility of the PHEVs
algorithm is still uneven. At 750 iterations, the results of the (c) The 500th iteration
14
DP-NSGA-DLS algorithm show little difference from the DP-NSGA-DLS
results after 1000 iterations that are shown in Fig. 11 12 NSGA-II
demonstrating that the DP-NSGA-DLS algorithm has 10
converged by 750 iterations. By contrast, NSGA-II’s Pareto

8
solution set after 750 iterations still shows some differences
from its Pareto solution set after 1000 iterations, illustrating that 6
NSGA-II has not yet converged by 750 iterations. 4

As shown in Fig. 13 (a), the average total utility of the
2
PHEVs in the Pareto solution set obtained by DP-NSGA-DLS
is always above that in the Pareto solution set of NSGA-II. 0
0 1 2 3 4 5 6 7 8 9
After 600 iterations, the DP-NSGA-DLS curve shows no The total utility of the PHEVs
(d) The 750th iteration
obvious changes in value, indicating that the algorithm has
converged. By contrast, NSGA-II does not converge until Fig. 12. Convergence comparisons between the DP-NSGA-DLS and
approximately 950 iterations. Moreover, the ultimate value NSGA-II algorithms.
Things Journal
8
two groups of variables show significant differences. These
DP-NSGA-DLS
7
NGSA-II
tests further demonstrate the superiority of the improved
The PHEVs' average total utility
6 algorithm.
5
4 VII. CONCLUSIONS AND FUTURE WORK

3
We studied the optimal energy transaction problem in a
2 hybrid energy market composed of an external market managed
1 by an EPEC and a local market managed by an FCEC. The
0 FCEC provides new opportunities for charging and discharging
0 100 200 300 400 500 600 700 800 900 1000
The number of iterations PHEVs to actively participate in cooperative V2V energy
(a) trading in search of profit. We quantified the respective utilities
8
gained by charging and discharging PHEVs through local
DP-NSGA-DLS
7
NGSA-II energy trading and investigated their optimal energy allocation
in response to the FCEC’s pricing. A non-profit-driven FCEC
The FCEC's average total profit
5
and a profit-driven FCEC were both considered. For each type
of FCEC, the corresponding trading optimization problem was
4
modeled, and a relevant algorithm was proposed to efficiently
3
solve for the optimal solution. The numerical results reveal that
2 PHEVs benefit more from local trading managed by an FCEC
1 than they do from trading directly with the EPEC. In addition,
0
the larger the EPEC’s energy price variance is, the greater the
0 100 200 300 400 500 600 700 800 900 1000 net gain the PHEVs can obtain. For a profit-driven FCEC, the
The number of iterations
(b) proposed DP-NSGA-DLS algorithm is superior to the existing
Fig. 13. Pareto solution sets vs. number of iterations. NSGA-II algorithm in terms of the convergence rate, the final
objective value and the evenness of the Pareto solution set. In
14 particular, the evenness of the Pareto solution set obtained by
DP-NSGA-DLS DP-NSGA-DLS is 23% better than that of the Pareto solution
13 NSGA-II set obtained by NSGA-II. Compared with a pure V2G scheme,
The maximum social welfare
12 our scheme can allocate energy more evenly and greatly reduce
energy losses while optimizing the overall welfare of multiple
11 trading entities.
10
In future work, we will fully consider the privacy of users
and design a secure energy trading mechanism that can resist
9 single-point attack resulting in a compromised FCEC. Further-
more, a combined management approach in which the FCEC
8
and EPEC competitively (or cooperatively) adjust their own
0 100 200 300 400 500 600 700 800 energy prices to benefit both themselves and the PHEVs will be
The time cost
studied.
Fig. 14. Maximum social welfare vs. the time cost.
To compare the time costs of the two algorithms, we define ACKNOWLEDGEMENTS

the maximum social welfare, which is calculated as the total of This research was partially supported by the National Natural
the FCEC’s average total profit and the PHEVs’ average total Science Foundation of China (61571098) and the Fundamental
utility, as an indicator of the overall gains obtained by the Research Funds for the Central Universities (ZYGX2016J217), the
multiparty entities participating in local energy trading. As 111 Project (B14039).
shown in Fig. 14, the maximum social welfare obtained by the
DP-NSGA-DLS algorithm is always greater than that obtained REFERENCES
by the NSGA-II algorithm. Starting at 349.681 s, the DP-NSGA
-DLS curve shows no obvious changes in value, indicating that [1] M. Yaghmaee, A. Leon-Garcia. A Fog-Based Internet of Energy
Architecture for Transactive Energy Management Systems. IEEE
the algorithm has converged. By contrast, the NSGA-II Internet of Things Journal, 2018, 5(2): 1055-1069.
algorithm does not converge until approximately 439.696 s. [2] Q. Kang, J. Wang, M. Zhou, et al. Centralized charging
Moreover, the final converged value obtained by scheduling for electric vehicles with battery-swapping on spot
DP-NSGA-DLS is larger than that obtained by NSGA-II. pricing via heuristic algorithms. IEEE Transactions on Intelligent
Hence, the improved algorithm can achieve larger benefits for Transportation Systems, 2016, 17(3): 659-669.
both the FCEC and the PHEVs with a lower time cost. [3] S. Pal, R. Kumar. Electric Vehicle Scheduling Strategy in
Residential Demand Response Programs with Neighbor
Furthermore, we conducted Whitney nonparametric
Connection. IEEE Transactions on Industrial Informatics, 2018,
hypothesis tests for the two groups of optimized target vectors 14(3): 980-988.
obtained by the two algorithms after 750 iterations. In both [4] F. Ye, Y. Qian, R. Hu. Incentive Load Scheduling Schemes for
hypothesis testing experiments for the two sets of optimization PHEV Battery Exchange Stations in Smart Grid. IEEE Systems
targets, the original hypothesis was rejected, indicating that the Journal, 2017, 11(2): 922-930.
Things Journal
[5] Wang, N. Zheng, D. Cao, et al. Parallel driving in CPSS: a unified [25] X. Hu, S. Moura, N. Murgovski, et al. Integrated Optimization of
approach for transport automation and vehicle intelligence. Battery Sizing, Charging, and Power Management in Plug-In
IEEE/CAA Journal of Automatica Sinica, 2017, 4(4): 577-587. Hybrid Electric Vehicles. IEEE Transactions on Control Systems
[6] G. Sun, Y. Zhang, D. Liao, et al. Bus Trajectory-based Street- Technology, 2016, 24(3): 1036-1043.
Centric Routing for Message Delivery in Urban Vehicular Ad hoc [26] S. Shojaabadi, S. Abapour, M. Abapour, et al. Optimal planning
Networks. IEEE Transactions on Vehicular Technology, 2019, of plug-in hybrid electric vehicle charging station in distribution
67(8): 7550-7563. network considering demand response programs and
[7] G. Sun, M. Yu, D. Liao, et al. Analytical Exploration of Energy uncertainties. IET Generation, Transmission & Distribution,
Savings for Parked Vehicles to Enhance VANET Connectivity. 2016, 10(13): 3330-3340.
IEEE Transactions on Intelligent Transportation Systems, 2018. [27] D. Chekired, L. Khoukhi, H. Mouftah. Decentralized Cloud-SDN
DOI: 10.1109/TITS.2018.2834569 Architecture in Smart Grid: A Dynamic Pricing Model. IEEE
[8] Zeng, S. Yang, F. Peng. Design Consideration and Comparison of Transactions on Industrial Informatics, 2018, 14(3): 1220-1231.
Wireless Power Transfer via Harmonic Current for PHEV and [28] J. Xu, P. Yi, T. Xie, et al. Charge station placement in electric
EV Wireless Charging. IEEE Transactions on Power Electronics, vehicle energy distribution network. IEEE International
2017, 32(8): 5943-5952. Conference on Communications, Paris, France, 2017, 1-6.
[9] T. Chiu, Y. Shih, A. Pang, et al. Optimized Day-Ahead Pricing [29] Q. Kang, S. Feng, M. Zhou, et al. Optimal Load Scheduling of
with Renewable Energy Demand-Side Management for Smart Plug-In Hybrid Electric Vehicles via Weight-Aggregation
Grids. IEEE Internet of Things Journal, 2017, 4(2): 374-383. Multi-Objective Evolutionary Algorithms. IEEE Transactions on
[10] G. Sun, L. Song, H. Yu, et al. V2V routing in a VANET based on Intelligent Transportation Systems, 2017, 18 (9): 1-12.
the autoregressive integrated moving average model. IEEE [30] S. Xia, S. Bu, X. Luo, et al. An Autonomous Real-Time Charging
Transactions on Vehicular Technology, 2019, 68(1): 908-922. Strategy for Plug-In Electric Vehicles to Regulate Frequency of
[11] B. Jiang, Y. Fei. A PHEV Power Management Cyber-Physical Distribution System with Fluctuating Wind Generation. IEEE
System for On-Road Applications. IEEE Transactions on Transactions on Sustainable Energy, 2018, 9(2):511-524.
Vehicular Technology, 2017, 66(7): 5797-5807. [31] W. Kempton, J. Tomić. Vehicle-to-grid power fundamentals:
[12] S. Jose. Fog computing and the Internet of Things: Extend the Calculating capacity and net revenue. Power Sources, 2005, 144
cloud to where the things are. White Paper, 2015. (1): 268-279.
https://www.cisco.com/c/dam/en_us/solutions/trends/iot/../comp [32] X. Wang, Q. Liang. Energy Management Strategy for Plug-In
uting-overview.pdf. Hybrid Electric Vehicles via Bidirectional Vehicle-to-Grid. IEEE
[13] R. Naha, S. Garg, D. Georgakopoulos, et al. Fog Computing: Systems Journal, 2017, 11(3): 1789-1798.
Survey of Trends, Architectures, Requirements, and Research [33] M. Togou, L. Khoukhi, A. Hafid. An altruistic service channel
Directions. IEEEAccess, 2018, 6(1): 47980-48009. selection scheme for V2V infotainment applications. IEEE
[14] F. Bonomi, R. Milito, J. Zhu, et al. Fog computing and its role in International Conference on Communications, 2017, 1-6.
the Internet of Things. ACM MCC Workshop Mobile Cloud [34] R. Alvaro-Hermana, J. Fraile-Ardanuy, P. Zufiria, et al. Peer to
Computing, 2012, 13-16. Peer Energy Trading with Electric Vehicles. IEEE Intelligent
[15] P. Zhang，M. Zhou，G. Fortino. Security and trust issues in Fog Transportation Systems Magazine, 2016, 8(3): 33-44.
computing: A survey. Future Generation Computer Systems, [35] C. Wei, Z. Fadlullah, N. Kato, et al. GT-CFS: A game theoretic
2018, 88(1): 16-27. coalition formulation strategy for reducing power loss in micro
[16] G. Sun, S. Sun, J. Sun, et al. Security and Privacy Preservation in grids. IEEE Transactions on Parallel and Distributed Systems,
Fog-based Crowd Sensing on the Internet of Vehicles. Journal of 2014, 25(9): 2307-2317.
Network and Computer Applications, 2019, 134: 89-99. [36] J. Kang, R. Yu, X. Huang, et al. Enabling Localized Peer-to-Peer
[17] R. Iqbal, T. Butt, M. Shafique, et al. Context-Aware Data-Driven Electricity Trading Among Plug-in Hybrid Electric Vehicles
Intelligent Framework for Fog Infrastructures in Internet of Using Consortium Blockchains. IEEE Transactions on Industrial
Vehicles. IEEE Access, 2018, 8(1): 58182-58194. Informatics, 2017, 13(6): 3154-3164.
[18] X. Wang, Z. Ning, L. Wang. Offloading in Internet of Vehicles: A [37] S. Boyd, L. Vandenberghe. Convex optimization. Beijing:
Fog-Enabled Real-Time Traffic Management System. IEEE Tsinghua university press, 2013.
Transactions on Industrial Informatics, 2018, 14(10): 4568-4578. [38] M. Zeng, S. Leng, S. Maharjan et al. Group bidding for
[19] J. Kang, R. Yu, X. Huang, et al. Privacy-Preserved Pseudonym guaranteed Quality of Energy in V2G smart grid networks. IEEE
Scheme for Fog Computing Supported Internet of Vehicles. IEEE International Conference on Communications, 2015, 5266-5271.
Transactions on Intelligent Transportation Systems, 2018, 19(8): [39] I. Ahmed, A. Mohamed. Power control and group proportional
2627-2637. fairness for frequency domain resource allocation in
[20] C. Zhu, J. Tao, G. Pastor, et al. Folo: Latency and Quality L-SC-FDMA based LTE uplink. Wireless Networks, 2015,
Optimized Task Allocation in Vehicular Fog Computing. IEEE 21(6):1819-1834.
Internet of Things Journal, 2018, DOI [40] S. Wang, S. Ali, T. Yue, et al. Integrating Weight Assignment
10.1109/JIOT.2018.2875520. Strategies with NSGA-II for Supporting User Preference
[21] X. Du and H. H. Chen. Security in Wireless Sensor Networks. Multi-Objective Optimization. IEEE Transactions on
IEEE Wireless Communications Magazine, 2008, 15(4): 60-66. Evolutionary Computation, 2018, 22(3): 378-393.
[22] Y. Xiao, V. Rayi, B. Sun, et al. A Survey of Key Management [41] W. Gao, G. Yen, S. Liu. A Dual-Population Differential Evolution
Schemes in Wireless Sensor Networks. Journal of Computer with Coevolution for Constrained Optimization. IEEE
Communications, 2007, 30(11-12): 2314-2341. Transactions on Cybernetics, 2015, 45(5): 1108-1121.
[23] G. Sun, K. Wang, H. Yu, et al. Priority-based medium access [42] Q. Zeng, J. Chen, M. Li, et al. An improved multi-objective
control for wireless body area networks with high-performance population-based extremal optimization algorithm with
design. IEEE Internet of Things Journal, Early Access, 2019. polynomial mutation. Information Sciences, 2016, 330(C):49-73.
DOI: 10.1109/JIOT.2019.2900661 [43] P. Palar, T. Tsuchiya, G. Parks. A comparative study of local
[24] X. Du, M. Guizani, Y. Xiao et al. A Routing-Driven Elliptic search within a surrogate-assisted multi-objective memetic
Curve Cryptography based Key Management Scheme for algorithm framework for expensive problems. Applied Soft
Heterogeneous Sensor Networks. IEEE Transactions on Wireless Computing, 2016, 43(1): 1-19.
Communications, 2009, 8(3): 1223-1229.

Optimal Energy Trading

Uploaded by

Copyright:

Available Formats

Optimal Energy Trading

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Optimal Energy Trading

Uploaded by

Copyright:

Available Formats

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

Optimal Energy Trading for Plug-in Hybrid Electric

Abstract—A large number of plug-in hybrid electric vehicles

Note the FCEC’s energy prices

VEF Node VEF Node PHEV discharges

FCEC EPEC FCEC

B. Supply-Demand Modeling of a Charging PHEV I j (a j , bj )  a j min  bj nin , j  DV . (6)

ai  (bi  li (bi ))  di , i  CV ,  (a j  d j )min  b j nin , j  DV . (7)

The total utility of the PHEVs

The experiments were conducted using MATLAB on a PC

 iCV ci*   jCV c*j  .

all PHEVs; these findings support the previous argument.

optimal energy price to benefit the charging and discharging 0.93

PHEVs. Stated otherwise, for a given FCEC energy price, a 0.92

10 min=0.8, mout=1.0 nin*=nout*=0.896 min=0.7, mout=1.0

Fig. 6. The total utility of PHEVs vs. the number of PHEVs.

0.06 min=0.8, mout=1.0

The total utility of the PHEVs

As shown in Fig. 7, as the number of PHEVs increases, the 10

overall loss ratio remains relatively stable. Comparisons 6

insignificant influence on the consequent energy transmission 2

(OP-MOS), where the capital letter “S” stands for “standard”: 2 2

the total number of solutions in the population by T. Similar to 6: while it  MaxIt do

bi   bi , (i  1, 2,..., k ), 0<  1.2

 popf , popc  find _ islegal  pop  , Fp  SortPopulation  popf  , 8

4: For the initial feasible solution population Fp : 4

 popf , Ff   NonDominatedSorting  popf  , 2

popf  CalcCrowdingDistance  popf , Ff  ; 0

population of infeasible solutions, which allows the infeasible 10

The FCEC's total profit

solutions are immediately eliminated. Hence, the genes of 2

solution set is completely uniform: 4

indicating that DP-NSGA-DLS achieves a 23% improvement 12

Fig. 12 compares the Pareto solution sets generated by the

appears between the results of the two algorithms at 250 6

by the DP-NGSA-DLS algorithm is approaching uniformity, 0

converged by 750 iterations. By contrast, NSGA-II’s Pareto

NSGA-II has not yet converged by 750 iterations. 4

4 VII. CONCLUSIONS AND FUTURE WORK

To compare the time costs of the two algorithms, we define ACKNOWLEDGEMENTS

You might also like

10 min=0.8, mout=1.0 nin=nout=0.896 min=0.7, mout=1.0