Electrical Power and Energy Systems 156 (2024) 109698

Contents lists available at ScienceDirect

International Journal of Electrical Power and Energy Systems

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

A coordinated generation and voyage planning optimization scheme for

all-electric ships under emission policy
Jingjie Gao a , Hai Lan a , Xinan Zhang b , H.H.C. Iu b , Ying-Yi Hong c , He Yin a ,∗
a
College of Intelligent Systems Science and Engineering, Harbin Engineering University, Harbin 150001, China
b
School of Electrical, Electronic and Computer Engineering, The University of Western Australia, Perth WA 6009, Australia
c
Department of Electrical Engineering, Chung Yuan Christian University, Chung Li District 320, Taoyuan City, Taiwan

ARTICLE INFO ABSTRACT

Keywords: The International Maritime Organization has introduced the concept of Emission Control Area to reduce sulfur
Emission control area emissions from coastal areas. The expensive low-sulfur fuels are required to be used in Emission Control
All-electric ships Area, leading to increased sulfide emissions due to bypassing by ship operators. It creates new challenges for
Bi-objective joint optimization
navigation and power generation scheduling of all-electric ships. This article presents a new bi-objective joint
Joint voyage planning and energy management
optimization scheme for ship sailing routes and energy management. It considers emission control strategies to
Mixed integer linear programming
balance the total cost of operation and sulfur dioxide emissions throughout the ship’s voyage. Unlike previous
optimization problems with fixed routes and sailing time, the influence of emission policies on the operator’s
navigation decisions is considered. The route options and sailing time are treated as variables embedded in the
model. In addition, the optimization accuracy is improved by integrating the model into mixed integer linear
programming through a modified piecewise linearization model. The results show that the total operating
cost for the same route is reduced by 6.05%, and the SO2 emission is 8.36% after integrating the emission
policy influencing factor in the model. The effectiveness of the proposed scheme is validated, and the model’s
necessity for optimizing ship power generation and voyage within Emission Control Area is also illustrated.

1. Introduction optimization model of a vessel is influenced by a variety of factors [11],

such as solar power generation [12], fluctuations in onshore electricity
The development of society cannot be separated from the massive prices [13,14], and speed loss due to waves [15]. In [16], the loss
consumption of energy. According to statistics, nearly 90% of the of ship speed due to wave and wind resistance uncertainty is intro-
world’s overseas trade logistics by maritime transportation [1]. Due duced into the optimization model to avoid severe voyage delays. Joint
to increasing concerns about ships’ energy efficiency and pollutant optimization of energy storage system (ESS) sizing and ship voyage
emissions, replacing mechanical propulsion with electric propulsion is by considering solar power generation in AES containing shipboard
a promising solution [2]. The all-electric ship (AES) allocates power photovoltaics [12]. In [13], a two-stage optimal dispatch model is
uniformly to the output power of the power generation system through proposed based on shore power price forecasting to achieve voyage
an energy management system [3], effectively improving the ship’s optimization. To reduce diesel generator (DG) fuel consumption from
energy efficiency and mitigating environmental pollution. Previous
the origin, some studies have introduced energy storage batteries, com-
studies have focused on AES stability control [4], power scheduling
bined heat, and power in AES to jointly optimize energy management
considering reliability [5], shore-to-ship power technology [6], energy
strategies [17,18].
management based on demand response [7], and integrated energy
The above studies sufficiently integrate ship voyage and power
optimization allocation [8]. With the further development of trans-
allocation to realize a joint multi-objective optimization. However,
portation electrification, ship voyage planning is highly coupled with
there needs to be more focus on the impact of the new emission policies
power demand [9]. Therefore, studying power generation and voyage
coordinated optimization of AES is very important. on ship operators [19]. In 2006, the International Maritime Organi-
Integrating an all-electric ship’s power generation system and voy- zation established the Emission Control Area (ECA) to regulate sulfur
age planning ensures optimal ship performance under varying mission emissions from ships in specific navigation areas. These areas include
requirements and environmental conditions [10]. In fact, the operation the Baltic Sea, the North Sea, the North American coasts, and the US

∗ Corresponding author.
E-mail address: [email protected] (H. Yin).

https://doi.org/10.1016/j.ijepes.2023.109698
Received 24 April 2023; Received in revised form 10 October 2023; Accepted 30 November 2023
Available online 14 December 2023
0142-0615/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
J. Gao et al. International Journal of Electrical Power and Energy Systems 156 (2024) 109698

Nomenclature

AES All-electric ship

DECA Domestic emission control area
DECAI, DECAII Parts I and II of DECA
DG Diesel generator
ECA Emission control area
ESS Energy storage system
HFO Heavy fuel oil
MGO Marine gas oil
MILP Mixed integer linear programming
MPWL Modified piecewise linearization
MSE Mean square error
OECA Outside of DECA area
Fig. 1. The navigation route and configuration of the AES.
PWL Piecewise linearization
SOC State of charge

Caribbean coasts. The practice shows that the fuel conversion method is
the most widely used and least costly of all the effective ways to reduce
emissions in the context of green shipping. The Ministry of Transport
of the People’s Republic of China proposed the implementation of the
domestic emission control area (DECA) [20] in 2015 based on this.
Specifically, ships in the DECA region must use low sulfur fuels to
reduce the SO2 production. In 2018, China published the ‘‘Marine Air
Emission Control Areas Implementation Scheme’’ to extend the DECA
constraints to cover the entire mainland coastal area [21,22]. The
impact of emission control areas on coastal ships is a common challenge
Fig. 2. Diagram of the shipping routes between the two ports.
to be addressed by the international society.
Many institutions have developed studies to deal with the opti-
mization of ship voyages and power generation as a result of emission
2. Problem description
policies. In [23], a bi-objective mixed integer programming model
integrating fuel cost and SO2 emissions is developed to jointly optimize This study uses one of Shanghai Zhonggu Shipping Group Co. Ltd’s
the ship’s route and speed inside and outside the ECA. In [24], the routes. This route runs from Dalian to Shanghai, passing through Yantai
impact of the 0.1% sulfur fuel restriction policy on sulfur emissions in in between. The navigation route and configuration of the AES are
the DECA region of China is analyzed by calculating the fuel costs of shown in Fig. 1. Battery storage introduced in AES can effectively im-
vessels inside and outside the DECA shipping area [25]. In [25], a two- prove generator efficiency while mitigating the problem of propulsion
layer programming model combining DECA width design is proposed power fluctuations due to wind and wave resistance [15,27]. According
to mitigate environmental pollution in coastal areas, considering the to the emission policy, ships in the DECA use marine gas oil (MGO)
effect of ECA width on SO2 emissions. 0.5% low sulfur fuel oil, and the outside of DECA area (OECA) use
heavy fuel oil (HFO). As the price per unit weight of MGO is much
In summary, there needs to be more research on the joint opti-
higher than HFO, ship operators will detour to make ships sail less in
mization of voyage planning and energy management for AES under the ECA zone.
relevant emission policies. Pollutant emissions will not be reduced if the Fig. 2 shows the different path options for ships sailing from Port A
impact of emission policies and ship routes on operational optimization to Port B. Route ‘‘1’’ is within DECA and uses the expensive MGO. Both
results is not considered [26]. Therefore, this paper uses the emission routes ‘‘2’’ and ‘‘3’’ reduce fuel costs by bypassing the OECA area. The
policy of China’s emission control area as an example to construct first and second parts of the sailing distance within DECA are noted as
the AES generation and voyage planning model. The contributions are DECAI and DECAII.
summarized as follows: When a path is chosen, the ship’s speed must be designed according
to the voyage and the navigation area. There is an exponential rela-
(1) An optimization scheme for AES generation and voyage consid-
tionship between vessel speed and propulsion power, and speed design
ering emission policies is established to optimize energy management, is highly coupled with demand-side management. In addition, AES
path options, and voyage planning while balancing costs and SO2 contains various operating conditions compared to land-based micro-
emissions. grids: approaching, cruising, departing, and berthing. The service load
(2) The variable time window with scheduling flexibility is used, of the ship varies under different operating conditions. In summary,
and the model is transformed into mixed integer linear programming optimal energy management strategies considering route selection must
that a commercial solver can solve through model relaxation and be implemented with the ship’s voyage and shipboard power generation
linearization. system.

(3) The applicability and effectiveness of the model are demon- 3. Mathematical formulation
strated based on multiple case studies of route data within China’s
Emission Control Area. Moreover, the optimization accuracy is en- The AES’s route planning and optimal energy management problem
hanced by a modified piecewise linearization (MPWL) method. is modeled as a bi-objective optimization scheme considering coastal

2
J. Gao et al. International Journal of Electrical Power and Energy Systems 156 (2024) 109698

emission control policies. The total operating costs and SO2 emissions Eq. (4) indicates that the ship can only be in one mode at each
of the ship are minimized through various constraints such as voyage, moment in all legs. Eq. (5) represents that if at the moment 𝑡 of the r-th
𝑡
leg the ship is sailing in DECAI (𝐼𝑑𝑒𝑐𝑎𝐼,𝑟 =1), then it is possible that at
emission policy, speed, operating conditions, and power balance.
𝑡+1 the ship will be in the OECA, DECAI, and DECAII regions of the r-th
3.1. Objective function leg. (6) represents the next moment of sailing in the OECA, where the
ship may still be in OECA or have entered the DECAII. (7) indicates that
after arrival at the DECAII, the ship has only two options, DECAII and
The optimization objectives of this study include the economic costs
the berth mode. (8) shows that the ship travels in the DECAI region of
and SO2 emissions along the AES voyage, as shown in (1)–(2). 𝑓𝑑𝑒𝑐𝑎
the 𝑟+1 leg after berthing. (9) represents that when the ship is traveling
and 𝑓𝑜𝑒𝑐𝑎 represent the total fuel consumed by a ship operating in
to the last leg (𝑟 = 𝑟_𝑚𝑎𝑥), if it is currently in the berthing condition, it
the DECA and OECA regions, respectively. Economic expenses include
should also be in the berthing mode at the next moment.
fuel consumption costs 𝐸𝑜𝑝 and generator start-up costs 𝐸𝑠𝑡 . 𝑆𝑑𝑒𝑐𝑎 , 𝑆𝑜𝑒𝑐𝑎
represent the sulfur emission factors for the DECA and OECA areas, 3.2.2. Operation conditions constraints
respectively. 𝐶𝑑𝑒𝑐𝑎 , 𝐶𝑜𝑒𝑐𝑎 are fuel price factors. 𝐶𝑠𝑡 is the start-up cost.
𝑡
𝑃𝐺,𝑘 , 𝑃𝐺,𝑘 are rated and output power of the 𝑘th DG. 𝑎𝑘 , 𝑏𝑘 are fuel ⎧ ∑𝑇 𝑡
consumption factors. 𝑂𝑘𝑡 is the on/off state of DG. The 𝑟 represents the ⎪ 𝑡=1 𝜇𝑑𝑒𝑝,𝑟 =1
index of the leg between ports. For example, a ship’s voyage is from ⎪
⎨ 𝑡
𝐼𝑏𝑒𝑟,𝑟 𝑡+1
≤ 𝜇𝑑𝑒𝑝,𝑟+1 (10)
port A to B and finally to port C. The value of 𝑟 is {1, 2}, where the leg ⎪
𝑡
from port A to port B is denoted as 𝑟 = 1. The binary variables 𝐼𝑑𝑒𝑐𝑎𝐼,𝑟 , ⎪ 𝑡+1
𝜇𝑑𝑒𝑝,𝑟1 𝑡
≤ 𝜇𝑑𝑒𝑝,𝑟1 𝑡
≤ 𝐼𝑑𝑒𝑐𝑎𝐼,𝑟1
⎩
𝑡 𝑡
𝐼𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 and 𝐼𝑜𝑒𝑐𝑎,𝑟 are introduced to indicate whether the ship is sailing ⎧ ∑𝑇 𝑡
in the DECAI, DECAII, and OECA regions of the r-th leg at t-th period. ⎪ 𝑡=1 𝜇𝑎𝑝𝑝,𝑟 =1
⎪
𝑚𝑖𝑛 𝐹1 = 𝐸𝑜𝑝 + 𝐸𝑠𝑡 (1) ⎨ 𝑡+1
𝐼𝑏𝑒𝑟,𝑟 𝑡
≤ 𝜇𝑎𝑝𝑝,𝑟 , 𝑟 < 𝑟_max (11)
⎪
⎪ 𝑡+1
𝐼𝑏𝑒𝑟,𝑟_max ≤ 𝑡
𝜇𝑎𝑝𝑝,𝑟_max 𝑡
+ 𝐼𝑏𝑒𝑟,𝑟_max
⎩
𝑚𝑖𝑛 𝐹2 = 𝑆𝑑𝑒𝑐𝑎 ⋅ 𝑓𝑑𝑒𝑐𝑎 + 𝑆𝑜𝑒𝑐𝑎 ⋅ 𝑓𝑜𝑒𝑐𝑎 (2)
As each phase’s service load and the sailing speed are related to the
AES’s operating conditions, relevant constraints are established based
⎧ 𝐸𝑜𝑝 = 𝐶𝑑𝑒𝑐𝑎 ⋅ 𝑓𝑑𝑒𝑐𝑎 + 𝐶𝑜𝑒𝑐𝑎 ⋅ 𝑓𝑜𝑒𝑐𝑎 on its sailing characteristics. Eq. (10) indicates that at the beginning
⎪
⎪ ∑ ∑ of the leg, the ship is departing from port. (11) indicates that the last
⎪ 𝐸𝑠𝑡 = 𝑇𝑡=1 𝐾 𝑡 𝑡−1
𝑘=1 𝑚𝑎𝑥(0, 𝑂𝑘 − 𝑂𝑘 ) ⋅ 𝐶𝑠𝑡 moment of the berthing mode should be the approaching condition;
⎪ 𝑡 𝑡
⎪ 𝐶𝑓𝑡 𝑢𝑒𝑙,𝑘 = 𝑎𝑘 ⋅ (𝑃𝐺,𝑘𝑡 ∕𝑃
𝐺,𝑘 ) + 𝑏𝑘 otherwise, the ship is in cruising operation. 𝜇𝑑𝑒𝑝,𝑟 , 𝜇𝑎𝑝𝑝,𝑟 are binary
⎪ ∑𝑇 ∑𝐾 variables that are equal to 1 if the ship is in the departure/approach
⎨ 𝑡
𝑓𝑑𝑒𝑐𝑎 = 𝑡=1 𝑘=1 𝐶𝑓 𝑢𝑒𝑙,𝑘 𝐼𝑑𝑒𝑐𝑎 𝑡 ⋅ △𝑡 (3) mode at time t in the r-th leg. 𝑟1 represents the minimum value of index
⎪ ∑𝑇 ∑𝐾
⎪ 𝑡 𝑡 𝑟.
𝑓𝑜𝑒𝑐𝑎 = 𝑡=1 𝑘=1 𝐶𝑓 𝑢𝑒𝑙,𝑘 𝐼𝑜𝑒𝑐𝑎 ⋅ △𝑡
⎪
⎪ 𝑡 ∑ 𝑅 𝑡 𝑡 3.2.3. Ship speed constraints
⎪ 𝐼𝑑𝑒𝑐𝑎 = 𝑟=1 (𝐼𝑑𝑒𝑐𝑎𝐼,𝑟 + 𝐼𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 )
⎪ ∑ 𝑅 { }
⎩ 𝑡
𝐼𝑜𝑒𝑐𝑎 𝑡
= 𝑟=1 𝐼𝑜𝑒𝑐𝑎,𝑟 𝑡 𝑡 𝑡 𝑡
𝜇𝑑𝑒𝑝,𝑟 𝛾 + (𝐼𝑑𝑒𝑐𝑎𝐼,𝑟 − 𝜇𝑑𝑒𝑝,𝑟 ) ⋅ 𝑉𝑛𝑜𝑚 (1 − 𝛿) ≤ 𝑉𝑑𝑒𝑐𝑎𝐼,𝑟
{ } (12)
3.2. Problem constraints 𝑡
≤ 𝜇𝑑𝑒𝑝,𝑟 𝑡
𝛾 + (𝐼𝑑𝑒𝑐𝑎𝐼,𝑟 𝑡
− 𝜇𝑑𝑒𝑝,𝑟 ) ⋅ 𝑉𝑛𝑜𝑚 (1 + 𝛿)
{ }
𝑡 𝑡 𝑡 𝑡
The AES route optimization and energy management problem are 𝜇𝑎𝑝𝑝,𝑟 𝛾 + (𝐼𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 − 𝜇𝑎𝑝𝑝,𝑟 ) ⋅ 𝑉𝑛𝑜𝑚 (1 − 𝛿) ≤ 𝑉𝑑𝑒𝑐𝑎𝐼𝐼,𝑟
not only restricted by physical constraints such as voyage distance and { } (13)
𝑡 𝑡 𝑡
≤ 𝜇𝑎𝑝𝑝,𝑟 𝛾 + (𝐼𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 − 𝜇𝑎𝑝𝑝,𝑟 ) ⋅ 𝑉𝑛𝑜𝑚 (1 + 𝛿)
ship speed but also by electrical constraints such as operating condi-
tions and power balance. To ensure the safe and stable operation of 𝑡 𝑡 𝑡
𝐼𝑜𝑒𝑐𝑎,𝑟 ⋅ 𝑉𝑛𝑜𝑚 (1 − 𝛿) ≤ 𝑉𝑜𝑒𝑐𝑎,𝑟 ≤ 𝐼𝑜𝑒𝑐𝑎,𝑟 ⋅ 𝑉𝑛𝑜𝑚 (1 + 𝛿) (14)
the ship during its voyage, the following constraints must be followed.
∑
𝑅
𝑡 𝑡 𝑡
𝑉𝑑𝑒𝑐𝑎𝐼,𝑟 + 𝑉𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 + 𝑉𝑜𝑒𝑐𝑎,𝑟 =𝑉𝑡 (15)
3.2.1. Voyage constraints 𝑟=1
Generally, the shipping area a ship navigates is divided into two
According to the vessel’s operating area and the working mode, the
scenarios. The first is route ‘‘1’’ in Fig. 2, which is always in the speed is divided into three categories depending on the shipping areas,
DECA and without any bypass behavior. The second scenario is ‘‘2’’ 𝑡
namely 𝑉𝑑𝑒𝑐𝑎𝐼,𝑟 𝑡
, 𝑉𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 𝑡
, and 𝑉𝑜𝑒𝑐𝑎,𝑟 . There are different constraints
and ‘‘3’’ in Fig. 2. The ship crosses the DECA as quickly as possible, on the speed of a ship under various operating conditions. The speed
sailing out of the emission control area and arriving at the target port range is [𝑉𝑛𝑜𝑚 (1 − 𝛿)𝛾, 𝑉𝑛𝑜𝑚 (1 + 𝛿)𝛾] for approach and departure; and
through a detour. Therefore, in order to constrain the navigation zone [𝑉𝑛𝑜𝑚 (1 − 𝛿), 𝑉𝑛𝑜𝑚 (1 + 𝛿)] for cruising conditions. 𝛿 is the maximum
in which the ship is operating during its entire voyage, Eqs. (4)–(9) allowable cruise speed variation. 𝑉𝑛𝑜𝑚 denotes the nominal speed, and
are introduced. All route options must follow the sequence constraints 𝛾 denotes the ratio of the speeds at departure and approach relative
described below. to the cruising condition [7]. In fact, there is a coupling between the
∑
𝑅 ship’s navigation area and its operating conditions. For example, when
𝑡 𝑡 𝑡 𝑡 the vessel departs from port (𝜇𝑑𝑒𝑝,𝑟 𝑡 = 1) it must be in DECAI, that is,
(𝐼𝑑𝑒𝑐𝑎𝐼,𝑟 + 𝐼𝑜𝑒𝑐𝑎,𝑟 + 𝐼𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 + 𝐼𝑏𝑒𝑟,𝑟 )=1 (4)
𝑟=1 𝑡
𝐼𝑑𝑒𝑐𝑎𝐼,𝑟 = 1. Eq. (12) represents the departure speed constraint in this
𝑡 𝑡+1 𝑡+1 𝑡+1 𝑡
case. Otherwise, the speed 𝑉𝑑𝑒𝑐𝑎𝐼,𝑟 should conform to the cruise speed
𝐼𝑑𝑒𝑐𝑎𝐼,𝑟 ≤ 𝐼𝑜𝑒𝑐𝑎,𝑟 + 𝐼𝑑𝑒𝑐𝑎𝐼,𝑟 + 𝐼𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 (5)
constraint. Similarly, when the ship approaches the port (𝜇𝑎𝑝𝑝,𝑟 𝑡 = 1),
𝑡 𝑡+1 𝑡+1
𝐼𝑜𝑒𝑐𝑎,𝑟 ≤ 𝐼𝑜𝑒𝑐𝑎,𝑟 + 𝐼𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 (6) it is in DECAII and Eq. (13) represents the approach speed constraint.
𝑡
Otherwise, 𝑉𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 should satisfy the cruising requirement. In addition,
𝑡 𝑡+1 𝑡+1
𝐼𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 ≤ 𝐼𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 + 𝐼𝑏𝑒𝑟,𝑟 (7)
a ship in the OECA is definitely in cruising condition and is also
𝑡 𝑡+1
𝐼𝑏𝑒𝑟,𝑟 ≤ 𝐼𝑑𝑒𝑐𝑎𝐼,𝑟+1 (8) constrained by the speed constraints in the cruising state as in Eq. (14).
𝑡 𝑡+1 Eq. (15) shows that 𝑉 𝑡 is the sum of the three categories of speeds at
𝐼𝑏𝑒𝑟,𝑟_𝑚𝑎𝑥 ≤ 𝐼𝑏𝑒𝑟,𝑟_𝑚𝑎𝑥 (9)
the 𝑡 moment.

3
J. Gao et al. International Journal of Electrical Power and Energy Systems 156 (2024) 109698

3.2.4. Route planning constraints Eq. (22) expresses the shipboard generator unit combination con-
straint, where 𝑞𝑘𝑡 is loading factor of DG. 𝑙𝑘,𝑚𝑖𝑛 , 𝑙𝑘,𝑚𝑎𝑥 are maximal and
⎧ ∑
𝑊 minimal ramp up rates of DG. Equations (23) present generators’ mini-
⎪ 𝑢𝑟,𝑤 = 1 mum on and off time [15]. Where 𝑇 𝑁𝑘𝑡−1 , 𝑇 𝐹𝑘𝑡−1 are number of periods
⎪ 𝑤=1 𝑘th DG has been online and offline continuously; 𝑇 𝑁𝑘,𝑚𝑖𝑛 , 𝑇 𝐹𝑘,𝑚𝑖𝑛 are
⎪
⎪ ∑𝑇 ∑
𝑊 minimum on and off period. The equivalent mixed integer linearization
𝑡
⎪ 𝑉𝑑𝑒𝑐𝑎𝐼,𝑟 ⋅ 𝛥𝑡 = (𝑢𝑟,𝑤 𝐷𝑑𝑒𝑐𝑎,𝑟,𝑤 ∕2) formulation is referenced in [31].
⎪ 𝑡 𝑤=1
⎨ (16)
⎪ ∑
𝑇 ∑
𝑊
3.2.7. Battery energy storage constraints
𝑡
⎪ 𝑉𝑑𝑒𝑐𝑎𝐼𝐼,𝑟 ⋅ 𝛥𝑡 = (𝑢𝑟,𝑤 𝐷𝑑𝑒𝑐𝑎,𝑟,𝑤 ∕2)
⎪ 𝑡 𝑤=1
⎪ ⎧
⎪ ∑
𝑇 ∑
𝑊

⎪
𝑡
𝑉𝑜𝑒𝑐𝑎,𝑟 ⋅ 𝛥𝑡 = 𝑢𝑟,𝑤 𝐷𝑜𝑒𝑐𝑎,𝑟,𝑤 ⎪ 𝑡 ≤ 𝑆𝑡 ⋅ 𝑃
0 ≤ 𝑃𝑐,𝑛 𝑐,𝑛,𝑚𝑎𝑥
⎩ 𝑡 𝑤=1 ⎪ 𝑐,𝑛
⎪ 𝑡
0 ≤ 𝑃𝑑𝑐,𝑛 𝑡 )⋅𝑃
≤ (1 − 𝑆𝑐,𝑛
The 𝑢𝑟,𝑤 binary variable is introduced to determine which bypass ⎪ 𝑑𝑐,𝑛,𝑚𝑎𝑥
⎨ 𝑡
𝐸𝐵,𝑛 (24)
option is chosen. If the w-th route is selected when the ship is sailing ⎪ 𝑆𝑂𝐶𝑛,𝑚𝑖𝑛 ≤ 𝑆𝑂𝐶𝑛𝑡 = ≤ 𝑆𝑂𝐶𝑛,𝑚𝑎𝑥
𝐸𝐵,𝑛
on the r-th leg, then 𝑢𝑟,𝑤 equals 1, otherwise 0. (16) denote the distance ⎪
constraints in the DECA and OECA regions for each leg of the voyage, ⎪ 𝑡 ⋅𝛥𝑡
𝜂𝑐,𝑛 ⋅𝑃𝑐,𝑛 𝑡 ⋅𝛥𝑡
𝑃𝑑𝑐,𝑛
⎪ 𝑆𝑂𝐶𝑛𝑡 = 𝑆𝑂𝐶𝑛𝑡−1 + 𝐸𝐵,𝑛
− 𝜂𝑑𝑐,𝑛 ⋅𝐸𝐵,𝑛
respectively. 𝐷𝑑𝑒𝑐𝑎,𝑟,𝑤 , 𝐷𝑜𝑒𝑐𝑎,𝑟,𝑤 are the distance in DECA and OECA for ⎩
sailing in the r-th leg under the w-th route. Eq. (24) is the battery energy storage’s charging and discharging
constraints. The binary variable 𝑆𝑐,𝑛 𝑡 is introduced to optimize the
3.2.5. Power balance constraints operating state of the battery at each moment, avoiding simultaneously
charging and discharging. When 𝑆𝑐,𝑛 𝑡 = 1, the battery is in charge or idle

∑
𝐾 ∑
𝑁 mode. When 𝑆𝑐,𝑛 𝑡 = 0, the battery operates in a discharge or idle state.
𝑡
𝑃𝐺,𝑘 𝜂𝑘𝑡 + 𝑡
(𝑃𝑑𝑐,𝑛 𝑡
− 𝑃𝑐,𝑛 ) = 𝑃𝑝𝑙𝑡 + 𝑃𝑠𝑒𝑟
𝑡
(17) Where 𝑃𝑐,𝑛,𝑚𝑎𝑥 , 𝑃𝑑𝑐,𝑛,𝑚𝑎𝑥 are maximum charge and discharge power of
𝑘=1 𝑛=1
the 𝑛th battery; 𝑆𝑂𝐶𝑛,𝑚𝑖𝑛 , 𝑆𝑂𝐶𝑛,𝑚𝑎𝑥 are maximum and minimum state of
𝑒1 𝑥 2 + 𝑒2 𝑥 + 𝑒3 charge; 𝜂𝑐,𝑛 and 𝜂𝑑𝑐,𝑛 represent the charging and discharging efficiency;
𝜂𝑘𝑡 = (18)
𝑒4 𝑥 + 𝑒5 𝐸𝐵,𝑛 is the rated capacity.
𝑡 𝑒1 𝑥3 + 𝑒2 𝑥2 + 𝑒3 𝑥
𝑃𝐺,𝑘 𝜂𝑘𝑡 = 𝑃𝐺,𝑘 ⋅ 4. Model analysis and solution method
𝑒4 𝑥 + 𝑒5 (19)
= 𝑃𝐺,𝑘 ⋅ 𝑓𝑘𝑡
In this study, the proposed model is a mixed integer non-linear
Eq. (17) represents the power balance constraint. Where DG’s effi- programming problem with dual objectives that is difficult to solve with
ciency 𝜂𝑘𝑡 is introduced as functions, it is transformed as the loading rate commercial solvers. The following methods transform the model into
changes [28]. Eq. (18) represents the generator operating efficiency, MILP while maintaining the accuracy requirements.
𝑡
𝑃𝐺,𝑘
where 𝑥 = 𝑃𝐺,𝑘
. The specific parameters are taken from the litera-
ture [29]. In (19), the active power of the generator 𝑃𝐺,𝑘 𝑡 times the 4.1. Bi-objective optimization method
efficiency 𝜂𝑘𝑡 can be equated to a single-variable polynomial function
with the independent variable being the generator load factor 𝑥. 𝑓𝑘𝑡 is There is a conflict between the total operating cost of the voyage
a variable defined for the convenience of the solution, with no physical and the sulfur emissions in the optimization objective, the Pareto
interpretation, and is referred to in this paper as the efficiency function. optimal solution of which is difficult to obtain. This paper uses the
𝑡 ,𝑃𝑡
𝑃𝑐,𝑛 are charge and discharge power of 𝑛th battery; 𝑃𝑝𝑙𝑡 , 𝑃𝑠𝑒𝑟𝑡 are weighted comprehensive criterion method [32] to transform the bi-
𝑑𝑐,𝑛
propulsion and service power. objective into a deterministic single-objective problem. First, 𝐹1 , 𝐹2
should be normalized to the same scale as in Eq. (25). The 𝐹𝑒,𝑚𝑖𝑛 and
∑
𝑅 ∑
𝑅 ∑
𝑅
𝐹𝑒,𝑚𝑎𝑥 represent the minimum and maximum value of the 𝑒 objective,
𝑡 𝑡 𝑡 𝑡
𝑃𝑠𝑒𝑟 = 𝑃𝑑𝑒𝑝 𝜇𝑑𝑒𝑝,𝑟 + 𝑃𝑎𝑝𝑝 𝜇𝑎𝑝𝑝,𝑟 + 𝑃𝑏𝑒𝑟 𝐼𝑏𝑒𝑟,𝑟
𝑟=1 𝑟=1 𝑟=1
and 𝐹𝑒 is the original objective function. Secondly, the normalized
(20) function 𝑓𝑒∗ is given the corresponding weight 𝜔𝑒 , and all weights sum
∑
𝑅
𝑡 𝑡 𝑡 to 1. 𝜔2 = 1-𝜔1 , therefore only the coefficient 𝜔1 is adjusted in the latter
+𝑃𝑐𝑟𝑢 (1 − (𝜇𝑑𝑒𝑝,𝑟 + 𝜇𝑎𝑝𝑝,𝑟 + 𝐼𝑏𝑒𝑟,𝑟 ))
𝑟=1 section. The weights can be adjusted flexibly according to the decision
𝑐1 (𝑉 ) 𝑡 𝑐2
= 𝑃𝑝𝑙𝑡 (21) maker’s preferences.
𝐹𝑒 − 𝐹𝑒,min
(20) shows that the AES service load changes accordingly un- 𝑓𝑒∗ = , ∀𝑒 ∈ {1, 2} (25)
𝐹𝑒,max − 𝐹𝑒,min
der different operating conditions. 𝑃𝑑𝑒𝑝 , 𝑃𝑎𝑝𝑝 , 𝑃𝑏𝑒𝑟 , 𝑃𝑐𝑟𝑢 are service loads
∑
at departing, approaching, berthing, and cruising mode. (21) repre- min 𝐹 = 𝜔𝑒 ⋅ 𝑓𝑒∗ , ∀𝑒 ∈ {1, 2}, 0 ≤ 𝜔𝑒 ≤ 1 (26)
sents an exponential relationship between ship speed and propulsion 𝑒
power [30]. Eqs. (19) and (21) are both nonlinear functions, further
transformed by the piecewise linearization method in the following 4.2. Model relaxation and linearization
section.
(1) The AES propulsion power function (21) and the dynamic gen-
3.2.6. Generator constraints erator output power in (19) are exponential and polynomial functions.
The piecewise linearization (PWL) is used to transform the above
⎧ 𝑃 𝑡 = 𝑂𝑡 ⋅ 𝑃 ⋅ 𝑞 𝑡 function into MILP [33], and (21) as an example to demonstrate the
𝐺,𝑘
⎪ 𝐺,𝑘 𝑡
𝑘 𝑘
linearization process. According to the speed 𝑉 𝑡 variation interval
⎨ 𝑞𝑚𝑖𝑛 ≤ 𝑞𝑘 ≤ 1 (22)
⎪ 𝑙 𝑡 𝑡−1 [𝑉𝑚𝑖𝑛 , 𝑉𝑚𝑎𝑥 ], M+1 points are identified to split it into 𝑀 subintervals,
⎩ 𝑘,𝑚𝑖𝑛 ≤ 𝑞𝑘 − 𝑞𝑘 ≤ 𝑙𝑘,𝑚𝑎𝑥
{ and the formulas are shown below.
(𝑂𝑘𝑡 − 𝑂𝑘𝑡−1 )(𝑇 𝑁𝑘𝑡−1 − 𝑇 𝑁𝑘,𝑚𝑖𝑛 ) ≤ 0
(23) 𝑉min = 𝑥∗0 < 𝑥∗1 < ⋯ 𝑥∗𝑀 = 𝑉max (27)
(𝑂𝑘𝑡−1 − 𝑂𝑘𝑡 )(𝑇 𝐹𝑘𝑡−1 − 𝑇 𝐹𝑘,𝑚𝑖𝑛 ) ≤ 0

4
J. Gao et al. International Journal of Electrical Power and Energy Systems 156 (2024) 109698

Table 1
Parameters for generators and ESS.
DG 𝑃𝐺 𝑇 𝑁𝑚𝑖𝑛 𝑇 𝐹𝑚𝑖𝑛 a b 𝑙𝑚𝑎𝑥 , 𝑙𝑚𝑖𝑛 𝐶𝑠𝑡 ($)
(MW) (△𝑡) (△𝑡) ($/p.u.) ($/p.u.)
1 15 5 5 2485 29 +0.5/−0.5 800
2 15 5 5 2485 29 +0.5/−0.5 800
3 10 3 3 1833 10 +0.5/−0.5 400
ESS 𝑃𝑐,𝑚𝑎𝑥 𝑃𝑑𝑐,𝑚𝑎𝑥 𝐸𝐵 𝑆𝑂𝐶𝑚𝑎𝑥 𝑆𝑂𝐶𝑚𝑖𝑛 𝜂𝑐 𝜂𝑑𝑐
(MW) (MW) (MWh)
1 15 15 30 0.95 0.1 0.95 0.97

∗
𝑃𝑝𝑙,𝑚 = 𝑐1 ⋅ (𝑥∗𝑚 )𝑐2 , ∀𝑚 ∈ {0, 1, 2, … , 𝑀} (28)
Then, the binary variable 𝑡𝑚
is introduced and the nonlinear func-
tion (21) is replaced by (29)–(33). In (30), the ship speed 𝑉 𝑡 is divided
into 𝑀 discrete variables 𝑥𝑡𝑚 at each moment. The linearized expression
for the propulsion power is shown in (32).
∑
𝑀
𝑡𝑚 = 1,  ∈ {0, 1} (29)
𝑚=1 Fig. 3. Path options graphic (left: from Dalian to Yantai; right: from Yantai to
∑𝑀 Shanghai).
𝑥𝑡𝑚 = 𝑉 𝑡 (30)
𝑚=1
Table 2
𝑥∗𝑚−1 ⋅ 𝑡𝑚 ≤ 𝑥𝑡𝑚 ≤ 𝑥∗𝑚 ⋅ 𝑡𝑚 , ∀𝑚 ∈ {1, 2, … , 𝑀} (31) Sailing distance and voyage model parameters.
Leg Route 1 Route 2 Route 3 Route 4 Route 5
∑
𝑀
𝑃̂𝑝𝑙𝑡 = ∗
(𝑡𝑚 ⋅ 𝑃𝑝𝑙,𝑚−1 + 𝑘∗𝑚 (𝑥𝑡𝑚 − 𝑡𝑚 ⋅ 𝑥∗𝑚−1 )) (32) Leg1 183/0 164/21 144/42 123/63 102/84
Leg2 113/216 184/315 153/346 125/375 96/403
𝑚=1
∗
𝑃𝑝𝑙,𝑚 ∗
− 𝑃𝑝𝑙,𝑚−1 𝑉𝑛𝑜𝑚 = 20 nm, 𝛿 = 0.3, 𝛾 = 0.7, 𝑐1 = 0.003, 𝑐2 = 3
𝑘∗𝑚 = (33)
𝑥∗𝑚 − 𝑥∗𝑚−1
The setting of the segmentation point 𝑥𝑡𝑚 is usually done using
𝑉 −𝑉
an equally spaced split, i.e., 𝑥∗𝑚 = max𝑀 min + 𝑥∗𝑚−1 . 𝑘∗𝑚 denotes the The 𝐶𝑑𝑒𝑐𝑎 =0.75$/kg and 𝐶𝑜𝑒𝑐𝑎 =0.405$/kg, respectively [23].
slope of the 𝑚th segment. However, the linearization accuracy is in- 𝑆𝑑𝑒𝑐𝑎 =0.01 and 𝑆𝑜𝑒𝑐𝑎 =0.07. This study uses one of Shanghai Zhonggu
fluenced by the number and location of segmentation points. In this Shipping Group Co. routes. Fig. 3 shows all the available paths for
study, a differential evolutionary algorithm is used to optimize the 𝑥∗𝑚 ‘‘Dalian-Yantai-Shanghai’’. Set the leg from Dalian to Yantai as Leg1
location [34]. The optimization objective (34) is the sum of squared and the one from Yantai to Shanghai as Leg2. In each leg, there are
errors between the original and linearized functions, where 𝑁𝑢𝑚 rep- five options for routes. ‘‘1-1’’ denotes that the first route is selected for
resents the number of sampled data. Therefore, the Eqs. (34)–(35) both legs. The specific sailing distance and voyage model parameters
optimize 𝑥∗𝑚 location to improve the accuracy of the model. 𝑃𝑝𝑙𝑡 (𝑉𝑖𝑡 ) and are shown in Table 2. The ‘‘183/0’’ means that the distance within
𝑃̂𝑝𝑙𝑡 (𝑥∗𝑚 , 𝑉𝑖𝑡 ) represent the non-linear function of propulsion power and the DECA is 183 nm, and the distance in the OECA region is 0 nm.
the optimized piecewise linearization function respectively. In addition, the △𝑡=1 h and 𝑇 = 35.
∑ The simulation is conducted on a computer with an Intel core
2
min
∗
(𝑃𝑝𝑙𝑡 (𝑉𝑖𝑡 ) − 𝑃̂𝑝𝑙𝑡 (𝑥∗𝑚 , 𝑉𝑖𝑡 )) (34) i7 CPU (2.30 GHz) and 32G memory. The optimization model is
𝑥𝑚
𝑖∈𝑁𝑢𝑚 implemented in Python 3.9 based on the open-source optimization
𝑠.𝑡., 𝑥∗𝑚−1 < 𝑥∗𝑚 , ∀𝑚 ∈ {0, 1, … , 𝑀}, 𝑥∗0 = 𝑉min , 𝑥∗𝑀 = 𝑉max (35) framework Pyomo with Cplex solver.

(2) The 𝑓𝑑𝑒𝑐𝑎 and 𝑓𝑜𝑒𝑐𝑎 in Eq. (3) are nonlinear constraints on the 5.2. Linearization results
multiplication of binary and continuous variables, which are linearized
by the introduction of auxiliary variables. With the 𝑓𝑑𝑒𝑐𝑎 as an example, According to the modified piecewise linearization method proposed
𝑡
the variable 𝑑𝑒𝑐𝑎,𝑘 is introduced. It represents the product of the binary in this paper, the linearization result of the ship speed-propulsion
𝑡
variable 𝐼𝑑𝑒𝑐𝑎 and the continuous variable 𝐶𝑓𝑡 𝑢𝑒𝑙,𝑘 . 𝑓𝑑𝑒𝑐𝑎 is equated to power curve is shown in Fig. 4, and the number of segments of the
Eq. (36), where 𝐶𝑓 𝑢𝑒𝑙,𝑘 and 𝐶𝑓 𝑢𝑒𝑙,𝑘 indicate the upper and lower bounds function is 16. The result of piecewise linearization of the generator
of 𝐶𝑓𝑡 𝑢𝑒𝑙,𝑘 . The linearization process of 𝑓𝑜𝑒𝑐𝑎 is omitted due to space efficiency function 𝑓𝑘𝑡 curve is shown in Fig. 5 with the number of
limitations. segments 5. As can be seen in Figs. 4 and 5, the nonlinear functions
⎧ ∑𝑇 ∑𝐾 are well fitted.
𝑡
⎪ 𝑓𝑑𝑒𝑐𝑎 = 𝑡=1 𝑘=1 𝑑𝑒𝑐𝑎,𝑘 ⋅ △𝑡 The segment points and slopes of the above linearization functions
⎪ 𝑡 𝑡 𝑡 are shown in Tables 3 and 4. Mean square error (MSE) measures
⎨ 𝐶𝑓 𝑢𝑒𝑙,𝑘 − 𝐶𝑓 𝑢𝑒𝑙,𝑘 (1 − 𝐼𝑑𝑒𝑐𝑎 ) ≤ 𝑑𝑒𝑐𝑎,𝑘 ≤ 𝐶𝑓𝑡 𝑢𝑒𝑙,𝑘 (36)
⎪ the deviation between the actual and estimated values. The results
⎪ 𝐶𝑓 𝑢𝑒𝑙,𝑘 𝐼 𝑡 𝑡
≤ 𝑑𝑒𝑐𝑎,𝑘 𝑡
≤ 𝐶𝑓 𝑢𝑒𝑙,𝑘 𝐼𝑑𝑒𝑐𝑎 indicate that the MSE is less than 0.01 for all segments in Table 3.
⎩ 𝑑𝑒𝑐𝑎
In Table 4, the mean square errors after linearization are equal to 0,
5. Simulation results which illustrates that the linearization results are sufficiently accurate
and can be applied to the optimization model proposed in this paper.
5.1. Parameter description
5.3. Joint optimization results for AES
The parameters of the generator and ESS are shown in Table 1 [10,
17]. In addition, the 𝑞𝑚𝑖𝑛 =0.6 and 𝜇𝑅 =0.2. The service loads for dif- The AES sails from Dalian to Shanghai, passing through the DECA
ferent operating modes are [𝑃𝑑𝑒𝑝 , 𝑃𝑎𝑝𝑝 , 𝑃𝑏𝑒𝑟 , 𝑃𝑐𝑟𝑢 ] = [8.69, 8.69, 3.5, 9.845]. area on the voyage. This study aims to provide an optimal route for

5
J. Gao et al. International Journal of Electrical Power and Energy Systems 156 (2024) 109698

Table 3 Table 5
Piecewise linearization results for propulsion power of AES. Comparison of optimization results.
Segment (nm/h) Slope MSE Segment (nm/h) Slope MSE Case1 Case2 Case3 Case4 Case5
[0, 2.88) 0.02 0.0002 [14.35, 15.76) 2.04 0.0036 Total cost 91 507 87 877 85 970 102 528 82 000
[2.88, 4.73) 0.13 0.0003 [15.76, 16.93) 2.41 0.0019 Fuel cost(DECA) 63 009 59 978 57 727 34 111 39 446
[4.73, 6.10) 0.27 0.0005 [16.93, 18.17) 2.77 0.0009 Fuel cost(OECA) 26 098 25 899 26 243 66 017 40 154
[6.10, 7.51) 0.42 0.0002 [18.17, 19.45) 3.19 0.0021 Start-up cost 2400 2000 2000 2400 2400
[7.51, 9.09) 0.62 0.0004 [19.45, 20.87) 3.66 0.0035 Total SO2 emissions 5788 5276 5304 11 865 7466
[9.09, 10.82) 0.89 0.0010 [20.87, 22.39) 4.21 0.0053 SO2 emissions(DECA) 1278 799 769 454 525
[10.82, 12.65) 1.24 0.0022 [22.39, 23.80) 4.80 0.0051 SO2 emissions(OECA) 4510 4476 4535 11 410 6940
[12.65, 14.35) 1.64 0.0045 [23.80, 25.00) 5.36 0.0022 Route options 1-1 1-1 1-1 5-5 5-1
Sailing time 29 29 30 38 29

Table 4
Piecewise linearization results for efficiency function of DG.
Segment(100%) Slope MSE
[0, 2.83) 0.008 0
[2.83, 42.72) 0.010 0
[42.72, 77) 0.009 0
[77, 97.27) 0.009 0
[97.27, 100) 0.009 0

Fig. 6. Results of AES voyage scheduling under different routes.

Fig. 4. Piecewise linearization curve for propulsion power.

Case 4: A cost analysis of an AES by the same optimization model

as in Case 3, the shortest route for the DECA area ‘‘5-5’’ is selected.
Case 5: A cost analysis of AES with the proposed optimization
model.
For comparative purposes, the optimization results for 𝜔1 = 1 are
shown in Table 5.
Case 1: In Case 1, the route ‘‘1-1’’ with the shortest navigation
distance is selected from the 25 (5*5) options. Due to the shortest
distance, the total sailing time for this route is the minimum at 29 h.
However, there is no reduction in total operating costs, especially the
fuel cost of 63009$ in the DECA area, which is the highest of all cases.
Therefore, it is necessary to design the voyage for AES.
Case 2: The propulsion and service load profiles are optimized by
Fig. 5. Piecewise linearization curve for efficiency function. adjusting the speed and operating conditions at each moment. The
results show the total cost is reduced from 91507$ to 87877$ and the
SO2 emissions from 5788 kg to 5276 kg. However, the difference in
operators through joint optimization of vessel voyage scheduling and fuel prices within various zones affects the total cost. The impact of
energy management, considering the emission policies within DECA. emissions policy on optimizing AES operations must be addressed.
Case 3: The results show a 6.05% reduction in total cost compared to
5.3.1. Case study Case 1 due to considering the effects of emission policies. The primary
The following five cases are used to verify the efficiency of the reason for this result is the decrease in the fuel cost in DECA. Unlike
optimization model proposed in this study. The route paths for Case previous studies, the varying time window is used in this article to
1–Case 4 are pre-defined. improve economic efficiency by optimizing AES’s speed and sailing
Case 1: A cost analysis for the AES with optimal energy management time in different navigation zones. The total sailing time is increased
but without speed scheduling, the shortest route ‘‘1-1’’ is selected. by 1 h due to the reduction in cost by lower sailing speed within the
Case 2: A cost analysis of AES with combined energy management DECA.
and speed optimization but without considering the effects of emissions Case 4: To investigate the effect of route design on the optimization
policies, the shortest route ‘‘1-1’’ is selected. results, the shortest DECA route, ‘‘5-5’’ is chosen for analysis. Table 5
Case 3: A cost analysis of AES for the joint optimization of speed indicates that the fuel cost within DECA is the lowest in Case 4, at
and energy management considering the effects of emission policies, 34111$. However, the total sailing time is increased to 38 h due to
the shortest route ‘‘1-1’’ is selected. bypassing the DECA area. Therefore, it is not sufficient to optimize AES

6
J. Gao et al. International Journal of Electrical Power and Energy Systems 156 (2024) 109698

Table 6
Route optimization results under different weight coefficient of cost.
Weight 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Route option 2–1 1–1 1–1 1–1 3–1 4–1 4–1 5–1 5–1 5–1 5–1

5.3.2. Effect of different weighting coefficient for bi-objective optimization

As seen from the optimization results of Case 5, the sulfur emissions
in the DECA area are reduced by 58.92% compared to Case 1, but
the total SO2 emissions increase from 5788 kg to 7466 kg. It shows
that the current emission policies are ineffective in mitigating pollutant
emissions considering only the economic costs. Therefore, the dual
objectives of economic and SO2 emissions must be considered simul-
taneously. The Pareto frontier in Fig. 8 visualizes the trade-off between
total operating cost and SO2 emissions during the voyage. The figure
shows that the total operating cost in the voyage tends to decrease
with the increase of SO2 emission. By modifying the weights 𝜔1 , the
ship operator can obtain the optimal decision corresponding to the
Fig. 7. Energy management of ship power system during voyage. frontier line. At 𝜔1 = 0.5, the cost and sulfur emissions are 87872$ and
6,169 kg, respectively. To further illustrate the effect of the weighting
factor on the optimization results, Table 6 displays the variation of the
operation by considering only the voyage and energy management, the path selection decision with the weight. The distance for the first route
route selection is very significant. in Leg2 is much shorter than the other routes at 329 nm. Therefore Leg2
Case 5: The results reveal that the optimal route is ‘‘5-1’’ for 𝜔1 = is consistently chosen as route ‘‘1’’. The route number option for the
1, where the total operating cost is 82000$. Under the designed route, first leg varies between ‘‘1’’ and ‘‘5’’, depending on demand preferences.
operating cost is significantly reduced through the flexible optimization
of the AES speed and navigation zone coordinated with the energy 5.3.3. Sensitivity analysis of fuel prices
management system at each period. Fig. 6 shows the voyage scheduling The primary reason for the change in the AES route option is the
of AES under routes ‘‘1-1’’, ‘‘5-5’’ and ‘‘5-1’’ respectively. The figure difference in fuel prices within and outside the DECA region. The
illustrates that the shipping speed can be optimized according to each horizontal coordinate of Fig. 9 is the price ratio of the MGO used in
the DECA to the HFO applied in the OECA area. The figure shows a
route’s DECA and OECA distances. Due to the higher fuel prices within
rapid reduction in the DECA area’s SO2 emissions as the price ratio
the DECA than OECA, the fuel consumption is decreased by reducing
increases from 1.2 to 1.4, with the variation leveling off at ratios greater
ship speed in DECA. Conversely, the speed will be appropriately raised
than 1.5. However, the total SO2 emissions in the voyage are displayed
outside the control area to avoid extended sailing time.
in Fig. 9 increase as the price ratio rises. It results from increased fuel
To further verify the proposed method, Fig. 7 presents the energy consumption due to bypassing emission zones to reduce the cost of
management of the AES during the voyage. The ESS charge and dis- low-sulfur fuel. Herefore, expanding the current DECA width to avoid
charge state is optimized according to the load demand at various bypasses by ship operators might reduce SO2 emissions.
operating modes and sailing speeds to reduce energy consumption
and emissions. For example, the battery charging increases the DG 5.3.4. Comparative analysis of different PWL methods
units load factor at 11 h. At periods such as 12–14 and 26–29, the The accuracy of the piecewise linearization method is related to the
battery takes on the part of the load to reduce the number of DGs number of segments. From Fig. 10, it can be seen that the relative error
online. Therefore, the DG units’ operating efficiency can be improved of both linearization methods gradually decreases as the number of
effectively through generation scheduling optimization. segments increases. The MPWL is more accurate than the PWL, with

Fig. 8. Pareto frontier for total costs and SO2 .

7
J. Gao et al. International Journal of Electrical Power and Energy Systems 156 (2024) 109698

CRediT authorship contribution statement

Jingjie Gao: Investigation, Methodology, Software, Writing – orig-

inal draft. Hai Lan: Conceptualization, Supervision, Project admin-
istration, Writing – review & editing. Xinan Zhang: Conceptualiza-
tion, Supervision, Methodology, Writing – review & editing. H.H.C.
Iu: Investigation, Supervision, Writing – review & editing. Ying-Yi
Hong: Investigation, Validation, Writing – review & editing. He Yin:
Conceptualization, Data curation, Writing – review & editing.

Declaration of competing interest

We declare that we have no financial and personal relationships

with other people or organizations that can inappropriately influence
our work, there is no professional or other personal interest of any
nature or kind in any product, service and company that could be
Fig. 9. Economic costs at different fuel prices.
construed as influencing the position presented in, or the review of,
the manuscript entitled.

Data availability

The authors do not have permission to share data.

Acknowledgments

This research was funded by Science and Technology Projects of

Innovation Laboratory for Sciences and Technologies of Energy Ma-
terials of Fujian Province (IKKEM) under Grant RD2021020504 and
the Nature Scientific Foundation of Heilongjiang Province under Grant
LH2020E069.

References
Fig. 10. Relative errors of different PWL methods.
[1] Fang S, Wang Y, Gou B, Xu Y. Toward future green maritime transportation: An
overview of seaport microgrids and all-electric ships. IEEE Trans Veh Technol
2019;69(1):207–19. http://dx.doi.org/10.1109/TVT.2019.2950538.
an error of 0.09% and 0.17%, respectively at 16 segments. To reduce [2] Ahamad NBB, Guerrero JM, Su C-L, Vasquez JC, Zhaoxia X. Microgrids technolo-
gies in future seaports. In: 2018 IEEE international conference on environment
the number of binary variables and constraints, the number of segments
and electrical engineering and 2018 IEEE industrial and commercial power
is set to 16 to lower the computational burden. systems Europe. IEEE; 2018, p. 1–6.
[3] Feng X, Butler-Purry KL, Zourntos T. A multi-agent system framework for real-
time electric load management in MVAC all-electric ship power systems. IEEE
6. Conclusion
Trans Power Syst 2014;30(3):1327–36. http://dx.doi.org/10.1109/TPWRS.2014.
2340393.
The option of routes, the design of speed within different navigation [4] Haseltalab A, Wani F, Negenborn RR. Multi-level model predictive control for
all-electric ships with hybrid power generation. Int J Electr Power Energy Syst
zones, and the energy dispatch of the ship’s power system all impact
2022;135:107484. http://dx.doi.org/10.1016/j.ijepes.2021.107484.
economic costs and pollutant emissions. In this study, a bi-objective [5] Hasanvand S, Rafiei M, Gheisarnejad M, Khooban M-H. Reliable power schedul-
mixed integer programming framework for joint optimization of all- ing of an emission-free ship: Multiobjective deep reinforcement learning. IEEE
electric ship generation and voyage is developed, considering emission Trans Transp Electr 2020;6(2):832–43. http://dx.doi.org/10.1109/TTE.2020.
control policies. In order to optimize the scheduling of ships more 2983247.
[6] Kumar J, Kumpulainen L, Kauhaniemi K. Technical design aspects of harbour
flexibly, the variable time window is used to simulate the sailing time area grid for shore to ship power: State of the art and future solutions. Int J
and the navigation area in which the ship is located. In addition, to Electr Power Energy Syst 2019;104:840–52. http://dx.doi.org/10.1016/j.ijepes.
improve the accuracy of the optimization model, the modified piece- 2018.07.051.
wise linearization method is used to linearize the nonlinear constraints. [7] Shang C, Srinivasan D, Reindl T. Economic and environmental genera-
tion and voyage scheduling of all-electric ships. IEEE Trans Power Syst
The effectiveness of the proposed methodology is demonstrated by
2016;31(5):4087–96. http://dx.doi.org/10.1109/TPWRS.2015.2498972.
analyzing and comparing several cases with the China Emission Control [8] Song T, Li Y, Zhang X-P, Wu C, Li J, Guo Y, et al. Integrated port energy
Area route data. The simulation results show that the optimal voyage system considering integrated demand response and energy interconnection. Int
and power generation planning with a minimum cost and emission J Electr Power Energy Syst 2020;117:105654. http://dx.doi.org/10.1016/j.ijepes.
2019.105654.
can be obtained based on the proposed method, considering the ship
[9] Xia S, Tian Y, Wang Z, Fu X, Li G, Zhang F, et al. An energy scheduling method
operator’s preferences. In addition, the sensitivity analysis results reveal for clustering islands with shared power exchanging vessels. Int J Electr Power
that raising the price of low-sulfur fuels can effectively reduce sulfide Energy Syst 2023;152:109200. http://dx.doi.org/10.1016/j.ijepes.2023.109200.
emissions in coastal areas. However, the total amount of emissions will [10] Li Z, Xu Y, Fang S, Zheng X, Feng X. Robust coordination of a hybrid AC/DC
increase. In future research, the uncertainty factors in the navigation multi-energy ship microgrid with flexible voyage and thermal loads. IEEE Trans
Smart Grid 2020;11(4):2782–93. http://dx.doi.org/10.1109/TSG.2020.2964831.
will also be considered in the optimization problem. Maximizing the
[11] Wang K, Yan X, Yuan Y, Jiang X, Lin X, Negenborn RR. Dynamic optimization
system operation reliability and improving the optimal dispatching of ship energy efficiency considering time-varying environmental factors. Transp
model is necessary. Res D 2018;62:685–98. http://dx.doi.org/10.1016/j.trd.2018.04.005.

8
J. Gao et al. International Journal of Electrical Power and Energy Systems 156 (2024) 109698

[12] Wen S, Zhao T, Tang Y, Xu Y, Zhu M, Huang Y. A joint photovoltaic-dependent [23] Zhen L, Hu Z, Yan R, Zhuge D, Wang S. Route and speed optimization for
navigation routing and energy storage system sizing scheme for more efficient liner ships under emission control policies. Transp Res C 2020;110:330–45.
all-electric ships. IEEE Trans Transp Electr 2020;6(3):1279–89. http://dx.doi.org/ http://dx.doi.org/10.1016/j.trc.2019.11.004.
10.1109/TTE.2020.3015983. [24] Wang S, Peng C. Model and analysis of the effect of China’s potential domestic
[13] Wen S, Zhao T, Tang Y, Xu Y, Zhu M, Fang S, et al. Coordinated optimal energy emission control area with 0.1% sulphur limit. Marit Bus Rev 2019;4(3):298–309.
management and voyage scheduling for all-electric ships based on predicted http://dx.doi.org/10.1108/MABR-03-2019-0012.
shore-side electricity price. IEEE Trans Ind Appl 2021;57(1):139–48. http://dx. [25] Tian X, Yan R, Qi J, Zhuge D, Wang H. A bi-level programming model for China’s
doi.org/10.1109/TIA.2020.3034290. marine domestic emission control area design. Sustainability 2022;14(6):3562.
[14] Zhao T, Qiu J, Wen S, Zhu M. Efficient onboard energy storage system sizing http://dx.doi.org/10.3390/su14063562.
for all-electric ship microgrids via optimized navigation routing under onshore [26] Chen L, Yip TL, Mou J. Provision of Emission Control Area and the impact
uncertainties. IEEE Trans Ind Appl 2022;58(2):1664–74. http://dx.doi.org/10. on shipping route choice and ship emissions. Transp Res D 2018;58:280–91.
1109/TIA.2022.3145775. http://dx.doi.org/10.1016/j.trd.2017.07.003.
[15] Fang S, Xu Y, Wang H, Shang C, Feng X. Robust operation of shipboard micro- [27] Kim M, Hizir O, Turan O, Day S, Incecik A. Estimation of added resistance and
grids with multiple-battery energy storage system under navigation uncertainties. ship speed loss in a seaway. Ocean Eng 2017;141:465–76.
IEEE Trans Veh Technol 2020;69(10):10531–44. http://dx.doi.org/10.1109/TVT. [28] Nuchturee C, Li T, Xia H. Energy efficiency of integrated electric propulsion for
2020.3011117. ships–A review. Renew Sustain Energy Rev 2020;134:110145. http://dx.doi.org/
[16] Fang S, Xu Y. Multi-objective robust energy management for all-electric ship- 10.1016/j.rser.2020.110145.
board microgrid under uncertain wind and wave. Int J Electr Power Energy Syst [29] Ghimire P, Zadeh M, Thorstensen J, Pedersen E. Data-driven efficiency modeling
2020;117:105600. http://dx.doi.org/10.1016/j.ijepes.2019.105600. and analysis of all-electric ship powertrain: A comparison of power system
[17] Li Z, Xu Y, Fang S, Wang Y, Zheng X. Multiobjective coordinated energy dispatch architectures. IEEE Trans Transp Electr 2022;8(2):1930–43. http://dx.doi.org/
and voyage scheduling for a multienergy ship microgrid. IEEE Trans Ind Appl 10.1109/TTE.2021.3123886.
2020;56(2):989–99. http://dx.doi.org/10.1109/TIA.2019.2956720. [30] Fan F, Aditya V, Xu Y, Cheong B, Gupta AK. Robustly coordinated operation of
[18] Huang Y, Lan H, Hong Y-Y, Wen S, Fang S. Joint voyage scheduling and a ship microgird with hybrid propulsion systems and hydrogen fuel cells. Appl
economic dispatch for all-electric ships with virtual energy storage systems. Energy 2022;312:118738. http://dx.doi.org/10.1016/j.apenergy.2022.118738.
Energy 2020;190:116268. http://dx.doi.org/10.1016/j.energy.2019.116268. [31] Carrion M, Arroyo J. A computationally efficient mixed-integer linear for-
[19] Kanellos FD. Multiagent-system-based operation scheduling of large ports’ power mulation for the thermal unit commitment problem. IEEE Trans Power Syst
systems with emissions limitation. IEEE Syst J 2019;13(2):1831–40. http://dx. 2006;21(3):1371–8. http://dx.doi.org/10.1109/TPWRS.2006.876672.
doi.org/10.1109/JSYST.2018.2850970. [32] Caramia M, Dell’Olmo P. Multi-objective management in freight logistics.
[20] Ministry of Transport. Implementation plan on domestic Emission Control Areas Springer; 2008.
in waters of the pearl River Delta, the YangtzeRiver delta and bohai rim.. 2015, [33] Tong B, Zhai Q, Guan X. An MILP based formulation for short-term hydro
Available online: https://www.chim-pn.com/Doc/implementation_plan_eng.pdf. generation scheduling with analysis of the linearization effects on solution
[21] Ministry of Transport. Marine air Emission Control Areas implementation feasibility. IEEE Trans Power Syst 2013;28(4):3588–99. http://dx.doi.org/10.
scheme. 2018, Available online: https://www.msa.gov.cn/html/xxgk/tzgg/wgfw/ 1109/TPWRS.2013.2274286.
20191025/7917B172-1CB6-421E-881C-25E5D00001B3.html. [34] Tian H, Zhao H, Liu C, Chen J. Iterative linearization approach for opti-
[22] Zhao Y, Fan Y, Zhou J, Kuang H. Bi-objective optimization of vessel speed and mal scheduling of multi-regional integrated energy system. Front Energy Res
route for sustainable coastal shipping under the regulations of emission control 2022;10:208. http://dx.doi.org/10.3389/fenrg.2022.828992.
areas. Sustainability 2019;11(22):6281. http://dx.doi.org/10.3390/su11226281.