You are on page 1of 12

Engineering Management in Production and Services

Volume 14 • Issue 1 • 2022

received: 28 October 2021


accepted: 25 February 2022

Green last-mile route planning pages: 1-12

for efficient e-commerce distribution


© 2022 S. Kunnapapdeelert et al.
This work is published under the Creative
Commons BY-NC-ND 4.0 License.

Siwaporn Kunnapapdeelert
James Vincent Johnson Passarin Phalitnonkiat
ABSTRACT
This study aims to design vehicle routes based on cost minimisation and the
minimisation of greenhouse gasses (GHG) emissions to help companies solve the
vehicle routing problem with pickup and delivery (VRPPD) via particle swarm
optimisation (PSO). An effective metaheuristics search technique called particle swarm
optimisation (PSO) was applied to design the optimal route for these problems.
Simulated data from Li and Lim (2001) were used to evaluate the PSO performance for
solving green vehicle routing problems with pickup and delivery (Green VRPPD). The
findings suggest that green vehicle routing problems with pickup and delivery should
be used when distributing products to customers living in a specific area called
a cluster. However, the design of vehicle routes by Green VRPPD costs more when used
to distribute products to customers living randomly in a coverage service area. When
logistics providers decide to use Green VRPPD instead of VRPPD, they need to be
concerned about possible higher costs if an increase in the number of vehicles
is needed. PSO has been confirmed for solving VRPPD effectively. The study compared
the results based on the use of two different objective functions with fuel consumption
from diesel and liquefied petroleum gas (LPG). It indicates that solving VRPPD
Siwaporn Kunnapapdeelert
by considering the emissions of direct greenhouse gases as an objective function
provides cleaner routes, rather than considering total cost as the objective function for
all test cases. However, as Green VRPPD requires more vehicles and longer travel Burapha University
distances, this requires a greater total cost than considering the total cost as the International College, Thailand
ORCID 0000-0002-0579-3877
objective function. Considering the types of fuels used, it is obvious that LPG is more
environmentally friendly than diesel by up to 53.61 %. This paper should be of interest Corresponding author:
to a broad readership, including those concerned with vehicle routing problems, e-mail: siwapornk@go.buu.ac.th
transportation, logistics, and environmental management. The findings suggest that
green vehicle routing problems with pickup and delivery should be used when James Vincent Johnson
distributing products to a cluster. However, the design of vehicle routes by Green
VRPPD costs more when used to distribute products to customers living randomly Burapha University
in a coverage service area. When logistics providers decide to use Green VRPPD instead International College, Thailand
of VRPPD, they need to be concerned about possible higher costs if an increase in the ORCID 0000-0002-7971-7877
number of vehicles is needed.

KEY WORDS Passarin Phalitnonkiat


last-mile, routing, green, particle swarm optimisation, e-commerce
Burapha University
10.2478/emj-2022-0001 International College, Thailand
ORCID 0000-0003-3063-3055

Introduction
under the United Nations Framework Convention on
Environmental concerns are a serious issue in Climate Change (UNFCCC) is issued annually.
most developed countries. Transportation is one of According to Greenhouse Gas Emissions and Sinks
the primary sources of GHG emissions. The Inven- 1990–2017 (EPA 2019), light-duty vehicles were the
tory of US Greenhouse Gas Emissions and Sinks largest category source of GHG emissions in the
Kunnapapdeelert, S., Johnson, J. V., & Phalitnonkiat, P. (2022). Green last-mile route planning for efficient e-commerce
distribution. Engineering Management in Production and Services, 14(1), 1-12. doi: 10.2478/emj-2022-0001

1
Engineering Management in Production and Services
Volume 14 • Issue 1 • 2022

transportation sector, followed by medium-duty and The purpose of a last-mile delivery service is to
heavy-duty trucks with about 59 % and 23 %, respec- deliver products to customers as fast as possible.
tively. The GHG emissions created by the transporta- Therefore, various technologies have been developed
tion sector include carbon dioxide (CO2), methane to enhance the efficiency of last-mile delivery ser-
(CH4), and nitrous oxide (N2O). These emissions vices. Examples of such technologies include drones,
result from fossil fuel combustion. autonomous vehicles, delivery by car, and self-service
In the last decade, the e-commerce business lockers or smart lockers. The use of such varied tech-
growth rate has increased significantly in most parts nologies implies that last-mile delivery has become
of the world. Online retail markets have changed another key customer expectation in this era. How-
rapidly. The US was the largest e-commerce market in ever, last-mile delivery also impacts the environment
the world before it was overtaken by China in 2013. due to increased GHG emissions.
In today’s fiercely competitive business environment, Although many researchers have published stud-
online sellers need to adapt to outperform their busi- ies on the concept of last-mile delivery, the optimal
ness competitors. The literature on e-commerce gen- green route design for last-mile delivery has not been
erally considers issues addressing the interrelations addressed. The last-mile delivery concept coupled
between e-commerce companies and logistics pro- with green route optimisation can transform ordinary
viders, investigating solutions and measures in the last-mile delivery to be a much more efficient delivery
e-commerce environment, and evolutionary applica- method. The main idea of green route optimisation is
tions in solving e-commerce problems (Zhang et al., to minimise total GHG emissions, delivery cost, and
2020; Tsang et al., 2021; Mutinda Kitukutha, Vasa delivery time by considering several factors, such as
& Oláh, 2021; Al-Tit, 2020; Federko et al., 2018; Flo- the location of the depot and customers, the vehicles’
rek-Paszkowska, Ujwary-Gil & Godlewska-Dzioboń, capacity and number.
2021; Gulc, 2021). E-commerce for logistics service Subsequently, this study aims to propose an
providers has faced the challenge of delivery service optimisation search technique called particle swarm
reliability that can serve the customised requirements optimisation (PSO) that can be used to design opti-
of larger e-commerce enterprises while maintaining mal green routes for the delivery of products to online
low-cost operations (Yu et al., 2020). Liu, Zhang, shoppers. The remainder of this paper is organised
Chen, Zhou, and Miao (2018) suggested that logistics into four sections. Theory and experimental research
initiatives are among the main factors e-commerce are presented in Section 2. This section describes
businesses can use to leverage operational perfor- a PSO algorithm used to solve green vehicle routing
mance. Logistics initiatives have accelerated the speed problems with pickup and delivery. Computational
with which order volumes move both up and down results and discussion are presented in Sections 3 and
the supply chain. Numerous researchers have focused 4, respectively. Lastly, conclusions and future research
on topics related to online shopping behaviour (Fu et recommendations are discussed in Section 5.
al., 2019; Rita et al., 2019; Zhou et al., 2020), customer
experiences (Lemke, 2016; Yuen et al., 2019; Chen
& Yang, 2021), online marketing (Gregory, 2017), the
1. Literature review
design of smart locker banks (Faugère & Montreuil,
2018), and last-mile delivery (Vakulenko et al., 2019; The vehicle routing problem (VRP) is a generali-
van Lopik et al., 2020; Qin et al., 2021). Researchers sation of the travelling salesperson problem (TSP).
have focused on these topics because good products The VRP is considered a combinatorial optimisation
and efficient services are essential factors driving and integer programming problem. It is an NP-hard
demand and, therefore, boosting revenue in e-com- problem that is time-consuming when solved using
merce businesses. E-commerce businesses generate exact algorithms, such as branch and price, branch
revenue for online businesses and also create oppor- and cut, and branch price and cut methods (Yu et al.,
tunities for other service businesses in the supply 2019). However, such problems could be solved using
chain. One type of supply chain business is known as metaheuristics approaches. Simulated annealing (SA)
a last-mile delivery service. The Capgemini Research and genetic algorithm (GA) were applied to green
Institute (2018) reported that 74 % of satisfied cus- vehicle routing with a heterogeneous fleet, including
tomers intended to increase their spending by 12 % reverse logistics in the form of collecting returned
with preferred retailers that provide great last-mile goods. The experimental results showed that the
delivery service. proposed algorithms were able to find the near-opti-

2
Engineering Management in Production and Services
Volume 14 • Issue 1 • 2022

mal solutions for large instances. SA achieved rela- its own neighbour. The velocity and position of each
tively better results in terms of solution quality, while particle is then updated according to its own best
GA spent less computational time for all-sized test experience and the global best experience to reach
problems (Foroutan et al., 2020). A hybrid genetic the best solution.
algorithm (GA) was successfully developed to deter- The PSO algorithm consists of the following
mine multi depot capacitated vehicle routing problem steps:
with split delivery and vehicle selection (Mehlawat et • The PSO algorithm starts by initialising an array
al., 2019). A hybrid genetic algorithm with variable of particles with random position and velocities
neighbourhood search was developed to solve the on d dimensions.
problem multi-depot vehicle routing problem under • Initialise the inertia weight.
the time-varying road network. It was found that the • Evaluate the objective function. [The objective
Hybrid genetic algorithm was effective for solving function is to determine vehicle routing prob-
VRP (Fan et al., 2021). lems with pickup and delivery requests by mini-
The green vehicle routing problem with pickup mising the emissions of greenhouse gases
and delivery is an extended version of the vehicle including carbon dioxide (CO2), methane (CH4),
routing problem with a time window (VRPTW). nitrous oxide (N2O) in d variables.]
Vehicle routes are commonly designed to visit each • Compare the particle’s fitness evaluation with the
location on the route that requires pickups and deliv- particle’s pbest (previous best). If the current
eries, and the addition of this requirement transforms value is better than pbest, then set the pbest value
the VRPTW problem into the vehicle routing prob- equal to the current value and set the pbest loca-
lem with pickup and delivery. Practical applications tion equal to the current location in d-dimension
of VRPPD include postal deliveries, school bus rout- space.
ing, and urban newspaper distribution (Créput et al., • Compare the fitness evaluation with the popula-
2004; Gupta et al., 2021). VRPPD can also include the tion’s overall previous best. If the current value is
problem of on-demand delivery service where cus- better than gbest (global best) then set gbest to
tomers pick up a product at a specific location (e.g., the current particle’s array index and value.
a convenience store or smart locker). A solution to • Adjust the velocity and position of the particle
the VRPPD problem involves designing a set of routes according to Equations (20) and (21), respec-
by minimising total routing cost while meeting the tively.
following requirements: • Loop back to step 2 until a stopping criterion is
• Each route starts and ends at the depot. met. A stopping criterion is usually a sufficiently
• A pickup, and its corresponding delivery cus- good fitness or a maximum number of iterations
tomer, is visited by exactly one vehicle. (generation).
• The total demand of any vehicle route does not
exceed the capacity of the vehicle assigned to the
route. 2. Research methods
• The total duration of any route does not exceed 2. Research methods
the pre-set route duration bound.
• Time windows specified by the customer are sat- The following section presents a mathematical The
isfied. model for VRPPD, including the input parameters 𝑖𝑖𝑖𝑖, an
PSO is one of the most famous optimisation and variables used in the model. inter
search techniques for solving NP hard problems. PSO Input parameters consist of the set of pickup at n
was inspired by the social behaviour of animals, such nodes and delivery nodes. These nodes are defined as 𝑞𝑞𝑞𝑞𝑖𝑖𝑖𝑖 =
as bird flocking and fish schooling (Shi and Eberhart, 𝑃𝑃𝑃𝑃 = {1, … , 𝑛𝑛𝑛𝑛}, 𝐷𝐷𝐷𝐷 = {𝑛𝑛𝑛𝑛 + 1, … ,2𝑛𝑛𝑛𝑛}, respectively, each
1998). Numerous researchers have successfully where 𝑛𝑛𝑛𝑛 is the number of requests. emis
adopted PSO to solve the VRPs (Belmecheri et al., 𝐻𝐻𝐻𝐻𝑖𝑖𝑖𝑖 represents the penalty cost when request 𝑖𝑖𝑖𝑖 is not emis
2013; Goksal et al., 2013; Chen et al., 2016; Norouzi et served. emis
al., 2017; Lagos et al., 2018; Zhu et al., 2019; Harbaoui 𝑖𝑖𝑖𝑖 ∈ 𝑃𝑃𝑃𝑃, 𝐾𝐾𝐾𝐾 represents the set of all vehicles |𝐾𝐾𝐾𝐾| = 𝑚𝑚𝑚𝑚. Deci
Dridi et al., 2020; Bansal & Wadhawan, 2021). 𝐶𝐶𝐶𝐶𝑘𝑘𝑘𝑘 denotes the capacity of the vehicle 𝑘𝑘𝑘𝑘, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾. 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘
The concept is to determine the solution by let- 𝑓𝑓𝑓𝑓𝑘𝑘𝑘𝑘 is the fixed cost of the vehicle 𝑘𝑘𝑘𝑘, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾 if it is used. 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝑣𝑣𝑣𝑣
ting each particle search for the solution randomly. 𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘 is the start node of the vehicle 𝑘𝑘𝑘𝑘, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾. 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 is
The solution of each particle is then compared with 𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘′ is the end node of the vehicle 𝑘𝑘𝑘𝑘, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾. start
All nodes are set as 𝑉𝑉𝑉𝑉 = 𝑁𝑁𝑁𝑁 ∪ {𝜏𝜏𝜏𝜏1 , … 𝜏𝜏𝜏𝜏𝑚𝑚𝑚𝑚 } ∪ 𝑄𝑄𝑄𝑄𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 is
′ }.
{𝜏𝜏𝜏𝜏1′ , … 𝜏𝜏𝜏𝜏𝑚𝑚𝑚𝑚 𝐴𝐴𝐴𝐴 denotes the set of arcs from node 𝑖𝑖𝑖𝑖 to boun
node 𝑗𝑗𝑗𝑗(𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) 𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗 ∈ 𝑉𝑉𝑉𝑉. While 𝑑𝑑𝑑𝑑𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 is the nonnegative3 servi
distance from node 𝑖𝑖𝑖𝑖 to node 𝑗𝑗𝑗𝑗, 𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗 ∈ 𝑁𝑁𝑁𝑁, 𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 represents 𝑧𝑧𝑧𝑧𝑖𝑖𝑖𝑖 =
𝐶𝐶𝐶𝐶𝑘𝑘𝑘𝑘 denotes the capacity of the vehicle 𝑘𝑘𝑘𝑘, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾. 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 = {1 𝑖𝑖𝑖𝑖𝑓𝑓𝑓𝑓 𝑒𝑒𝑒𝑒𝑑𝑑𝑑𝑑𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑏𝑏𝑏𝑏𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑏𝑏𝑏𝑏𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑖𝑖𝑖𝑖 𝑎𝑎𝑎𝑎𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑗𝑗𝑗𝑗 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑢𝑢𝑢𝑢𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
𝑆𝑆𝑆𝑆 ≤ 𝑆𝑆𝑆𝑆 𝑘𝑘𝑘𝑘 ∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀𝑗𝑗𝑗𝑗 ∈ 𝑉𝑉𝑉𝑉
𝑓𝑓𝑓𝑓𝑘𝑘𝑘𝑘 is the fixed cost of the vehicle 𝑘𝑘𝑘𝑘, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾 if it is used. 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝑣𝑣𝑣𝑣𝑒𝑒𝑒𝑒ℎ𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙𝑘𝑘𝑘𝑘 𝑘𝑘𝑘𝑘; 0𝑛𝑛𝑛𝑛+𝑖𝑖𝑖𝑖 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑒𝑒𝑒𝑒.
Engineering Management in Production and Services
𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘 is the start node of the vehicle 𝑘𝑘𝑘𝑘, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾. 𝑆𝑆𝑆𝑆𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 is=a 1nonnegative integer that𝐾𝐾𝐾𝐾,represents
⇒ 𝑄𝑄𝑄𝑄𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 + 𝑞𝑞𝑞𝑞𝑖𝑖𝑖𝑖 ≤ 𝑄𝑄𝑄𝑄𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ∀𝑘𝑘𝑘𝑘 ∈ Volume ∀(𝑖𝑖𝑖𝑖,14𝑗𝑗𝑗𝑗)• Issue
the service
∈ 𝐴𝐴𝐴𝐴 1 • 2022
′ 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘
𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘 is the end node of the vehicle 𝑘𝑘𝑘𝑘, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾. start time of the vehicle 𝑘𝑘𝑘𝑘 at the location 𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾.
All nodes are set as 𝑉𝑉𝑉𝑉 = 𝑁𝑁𝑁𝑁 ∪ {𝜏𝜏𝜏𝜏1 , … 𝜏𝜏𝜏𝜏𝑚𝑚𝑚𝑚 } ∪ 𝑄𝑄𝑄𝑄𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 is a nonnegative 𝑄𝑄𝑄𝑄𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 ≤ 𝐶𝐶𝐶𝐶𝑘𝑘𝑘𝑘 ∀𝑘𝑘𝑘𝑘 integer ∈ 𝐾𝐾𝐾𝐾, ∀𝑖𝑖𝑖𝑖that
∈ 𝑉𝑉𝑉𝑉represents the(11) upper
{𝜏𝜏𝜏𝜏1′ , … 𝜏𝜏𝜏𝜏𝑚𝑚𝑚𝑚 ′ }.
𝐴𝐴𝐴𝐴 denotes the set of arcs from node 𝑖𝑖𝑖𝑖 to bound on the amount of goods on the vehicle 𝑘𝑘𝑘𝑘 after
node 𝑗𝑗𝑗𝑗(𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) 𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗 ∈ 𝑉𝑉𝑉𝑉. While 𝑑𝑑𝑑𝑑𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 is the nonnegative servicing node 𝑄𝑄𝑄𝑄𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘 = 𝑖𝑖𝑖𝑖 where
𝑄𝑄𝑄𝑄𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘′𝑘𝑘𝑘𝑘 =𝑖𝑖𝑖𝑖 0∀𝑘𝑘𝑘𝑘
∈ 𝑉𝑉𝑉𝑉,∈𝑘𝑘𝑘𝑘 𝐾𝐾𝐾𝐾∈ 𝐾𝐾𝐾𝐾. (12)
distance from node 𝑖𝑖𝑖𝑖 to node 𝑗𝑗𝑗𝑗, 𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗 ∈ 𝑁𝑁𝑁𝑁, 𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 represents 𝑧𝑧𝑧𝑧𝑖𝑖𝑖𝑖 = {1 𝑖𝑖𝑖𝑖𝑓𝑓𝑓𝑓 𝑡𝑡𝑡𝑡ℎ𝑒𝑒𝑒𝑒 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑝𝑝𝑝𝑝𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 𝑖𝑖𝑖𝑖𝑛𝑛𝑛𝑛 𝑡𝑡𝑡𝑡ℎ𝑒𝑒𝑒𝑒 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
the nonnegative travel time from node 𝑖𝑖𝑖𝑖 to node 𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎; 0 𝑘𝑘𝑘𝑘 ∈ {0,1}∀𝑘𝑘𝑘𝑘
𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝑒𝑒𝑒𝑒 , 𝑖𝑖𝑖𝑖∈∈𝐾𝐾𝐾𝐾,𝑃𝑃𝑃𝑃.∀(𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) ∈ 𝐴𝐴𝐴𝐴 (13)
𝑗𝑗𝑗𝑗, 𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗 ∈ 𝑁𝑁𝑁𝑁. In cases where travel time is included, the According to the assumptions above, green
𝑧𝑧𝑧𝑧 ∈ {0,1}∀𝑘𝑘𝑘𝑘 ∈ 𝑃𝑃𝑃𝑃 (14)
ds travel time must satisfy the triangle inequality where VRPPDP can 𝑖𝑖𝑖𝑖 be explained with a mathematical
𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ≤ 𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 + 𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 for all 𝑖𝑖𝑖𝑖. 𝑗𝑗𝑗𝑗. 𝑙𝑙𝑙𝑙 ∈ 𝑉𝑉𝑉𝑉. 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖 is service time model as follows: 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 ≥ 0∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉 (15)
section presents a mathematical spent for loading and unloading vehicles
The time windows are represented by [𝑎𝑎𝑎𝑎, 𝑏𝑏𝑏𝑏] for node at node 𝑖𝑖𝑖𝑖.
tical
including the Theinput time parameters windows are represented 𝑖𝑖𝑖𝑖, and by a visit [𝑎𝑎𝑎𝑎, 𝑏𝑏𝑏𝑏]tofor node node can only occur in𝑄𝑄𝑄𝑄this
𝑖𝑖𝑖𝑖Minimise 𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 ≥time 0∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉 (16)
eters
n the model.𝑖𝑖𝑖𝑖, and a visit to node 𝑖𝑖𝑖𝑖 can interval. only occur Theinquantity this time of goods loaded onto a vehicle
ers consist of the setThe of quantity pickup of goods at loaded node 𝛼𝛼𝛼𝛼 ∑i𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘 is represented a∑vehicle 𝑑𝑑𝑑𝑑𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖as
𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 (𝜀𝜀𝜀𝜀𝑞𝑞𝑞𝑞 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶2 + 𝜀𝜀𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶4 + 𝑖𝑖𝑖𝑖 ∈𝜀𝜀𝜀𝜀𝑁𝑁𝑁𝑁𝑃𝑃𝑃𝑃 )
interval. onto (𝑖𝑖𝑖𝑖,𝑖𝑖𝑖𝑖) 𝑖𝑖𝑖𝑖 when 2 𝐶𝐶𝐶𝐶 and

ckup Theseatnodes
nodes. nodearei defined is represented as as𝑞𝑞𝑞𝑞𝑖𝑖𝑖𝑖 = 𝑞𝑞𝑞𝑞𝑖𝑖𝑖𝑖 −𝑞𝑞𝑞𝑞 when 𝑖𝑖𝑖𝑖−𝑛𝑛𝑛𝑛 for 𝑖𝑖𝑖𝑖 ∈𝑖𝑖𝑖𝑖 𝑃𝑃𝑃𝑃 ∈ 𝐷𝐷𝐷𝐷. and Lastly, an emissions The goal offactor greenforVRPPD is to design a vehicle emission (
Subject route to by minimising three GHG emissions (CO2, et al., 20
ed as
{𝑛𝑛𝑛𝑛 + 1, … ,2𝑛𝑛𝑛𝑛}, 𝑞𝑞𝑞𝑞𝑖𝑖𝑖𝑖 = −𝑞𝑞𝑞𝑞respectively, 𝑖𝑖𝑖𝑖−𝑛𝑛𝑛𝑛 for 𝑖𝑖𝑖𝑖 ∈ 𝐷𝐷𝐷𝐷. Lastly,each an emissions GHG is represented factor for as 𝜀𝜀𝜀𝜀 where 𝜀𝜀𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶2 denotes the
ber
vely,of requests. each GHG is represented as 𝜀𝜀𝜀𝜀emissions where 𝜀𝜀𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶∑2factor for CH , and N O) from the transportation vehicle, as multiplyin
denotes the 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖2𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘, 𝜀𝜀𝜀𝜀+𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑧𝑧𝑧𝑧4𝑖𝑖𝑖𝑖 = represents the
4 2
𝑘𝑘𝑘𝑘∈𝑘𝑘𝑘𝑘 ∑𝑖𝑖𝑖𝑖∈𝑁𝑁𝑁𝑁 1∀𝑖𝑖𝑖𝑖 ∈ 𝑃𝑃𝑃𝑃
enalty cost when emissions requestfactor 𝑖𝑖𝑖𝑖 is notfor 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶2emissions , 𝜀𝜀𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶4 represents factor forthe
𝑘𝑘𝑘𝑘
presented
𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻4 , and 𝜀𝜀𝜀𝜀𝑁𝑁𝑁𝑁2𝐶𝐶𝐶𝐶 refers to the in Equation (1). This minimisation ensures of GHG fo
s not emissions factor for 𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻4 , and emissions 𝜀𝜀𝜀𝜀∑𝑁𝑁𝑁𝑁𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉 refers
factor𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 − forto∑𝑁𝑁𝑁𝑁 the that the
2 𝐶𝐶𝐶𝐶. 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖 𝑛𝑛𝑛𝑛+𝑖𝑖𝑖𝑖 𝑘𝑘𝑘𝑘 = 0∀𝑖𝑖𝑖𝑖 ∈ 𝐾𝐾𝐾𝐾, ∀𝑖𝑖𝑖𝑖 ∈ 𝑃𝑃𝑃𝑃
pickup and delivery orders are performed by one of the
2 𝐶𝐶𝐶𝐶 𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉
the set of all emissions vehiclesfactor |𝐾𝐾𝐾𝐾| =for 𝑚𝑚𝑚𝑚.𝑁𝑁𝑁𝑁2 𝐶𝐶𝐶𝐶. Decision variables are explained the same below.vehicle, and the orders are implemented in calculating
apacity
= 𝑚𝑚𝑚𝑚. of the Decision vehiclevariables 𝑘𝑘𝑘𝑘, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾. are explained 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 below.
= {1 𝑖𝑖𝑖𝑖𝑓𝑓𝑓𝑓 𝑒𝑒𝑒𝑒𝑑𝑑𝑑𝑑𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 ∑ 𝑏𝑏𝑏𝑏𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑏𝑏𝑏𝑏𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
𝑖𝑖𝑖𝑖∈𝑃𝑃𝑃𝑃∪�𝜏𝜏𝜏𝜏 ′ 𝑥𝑥𝑥𝑥 Equations = 1∀𝑘𝑘𝑘𝑘 ∈ (2)
𝐾𝐾𝐾𝐾
𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑖𝑖𝑖𝑖 𝑎𝑎𝑎𝑎𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑗𝑗𝑗𝑗 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑢𝑢𝑢𝑢𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 and (3). The confirmations for the emissions
𝑘𝑘𝑘𝑘 �
∈f the 𝐾𝐾𝐾𝐾. vehicle 𝑥𝑥𝑥𝑥𝑘𝑘𝑘𝑘,𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘 = ∈ 𝐾𝐾𝐾𝐾 {1if𝑖𝑖𝑖𝑖𝑓𝑓𝑓𝑓it𝑒𝑒𝑒𝑒𝑑𝑑𝑑𝑑𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 is used. 𝑏𝑏𝑏𝑏𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑏𝑏𝑏𝑏𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
𝑣𝑣𝑣𝑣𝑒𝑒𝑒𝑒ℎ𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑎𝑎𝑎𝑎𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑗𝑗𝑗𝑗 𝑖𝑖𝑖𝑖0
𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 𝑘𝑘𝑘𝑘; 𝑖𝑖𝑖𝑖 𝑢𝑢𝑢𝑢𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑒𝑒𝑒𝑒. conditions that each vehicle departs from its starting distance b
∑𝑖𝑖𝑖𝑖∈𝐷𝐷𝐷𝐷∪{𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘} 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝜏𝜏𝜏𝜏′terminal = 1∀𝑘𝑘𝑘𝑘 ∈ and 𝐾𝐾𝐾𝐾 stops at its ending terminal are emissions
ode of the𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏vehicle
used. 𝑣𝑣𝑣𝑣𝑒𝑒𝑒𝑒ℎ𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 𝑘𝑘𝑘𝑘, 𝑘𝑘𝑘𝑘;𝑘𝑘𝑘𝑘0∈𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑒𝑒𝑒𝑒.
𝐾𝐾𝐾𝐾. 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 is a nonnegative integer that 𝑘𝑘𝑘𝑘 𝑘𝑘𝑘𝑘 represents the service
of∈ the 𝐾𝐾𝐾𝐾. vehicle𝑆𝑆𝑆𝑆𝑘𝑘𝑘𝑘, 𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 is𝑘𝑘𝑘𝑘 a∈ nonnegative 𝐾𝐾𝐾𝐾. integer that startrepresents time of thethe vehicle service performed
𝑘𝑘𝑘𝑘 at the location 𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾. in Equations (4) and (5). Equation (6) as fuel hea
∑𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 − ∑𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 = 0∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀𝑗𝑗𝑗𝑗 ∈ 𝑁𝑁𝑁𝑁 ′
e set as 𝑉𝑉𝑉𝑉start = 𝑁𝑁𝑁𝑁time ∪ {𝜏𝜏𝜏𝜏of 1 , the … 𝜏𝜏𝜏𝜏vehicle 𝑚𝑚𝑚𝑚 } ∪ 𝑘𝑘𝑘𝑘 at the 𝑄𝑄𝑄𝑄𝑖𝑖𝑖𝑖location
𝑘𝑘𝑘𝑘 is a nonnegative 𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉, 𝑘𝑘𝑘𝑘 ∈integer 𝐾𝐾𝐾𝐾. thatensures represents that consecutive the upper paths between 𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘 and 𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘 are fuel, and
es the ∪ set of 𝑄𝑄𝑄𝑄arcs 𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 is from node 𝑖𝑖𝑖𝑖 tointeger that bound used for each vehicle. Equations (7) and (8) guarantee somewhat
𝑚𝑚𝑚𝑚 } a nonnegative 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖on
represents the amount the upper of goods on the vehicle 𝑘𝑘𝑘𝑘 after
𝑘𝑘𝑘𝑘 = 1 ⇒ 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 + 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 + 𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ≤ 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀(𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) ∈ 𝐴𝐴𝐴𝐴
𝑉𝑉𝑉𝑉.𝑖𝑖𝑖𝑖 to While 𝑑𝑑𝑑𝑑𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖bound is the on the amount of goods
nonnegative servicing on the node vehicle 𝑖𝑖𝑖𝑖 where 𝑘𝑘𝑘𝑘 after𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉, 𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾. that 𝑆𝑆𝑆𝑆 𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 is set correctly along the paths within travel dist
𝑖𝑖𝑖𝑖 to node 𝑗𝑗𝑗𝑗, 𝑖𝑖𝑖𝑖,servicing
ative 𝑗𝑗𝑗𝑗 ∈ 𝑁𝑁𝑁𝑁, 𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖node represents 𝑖𝑖𝑖𝑖 where 𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉,𝑧𝑧𝑧𝑧𝑘𝑘𝑘𝑘 𝑖𝑖𝑖𝑖 = ∈ {1 𝐾𝐾𝐾𝐾. 𝑖𝑖𝑖𝑖𝑓𝑓𝑓𝑓 𝑡𝑡𝑡𝑡ℎ𝑒𝑒𝑒𝑒 𝑎𝑎𝑎𝑎𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
𝑖𝑖𝑖𝑖 ≤ 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 ≤𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑏𝑏𝑏𝑏𝑖𝑖𝑖𝑖𝑝𝑝𝑝𝑝𝑙𝑙𝑙𝑙 a
∀𝑘𝑘𝑘𝑘𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 particular
∈ 𝐾𝐾𝐾𝐾, 𝑖𝑖𝑖𝑖∀𝑗𝑗𝑗𝑗 𝑛𝑛𝑛𝑛 𝑡𝑡𝑡𝑡ℎ𝑒𝑒𝑒𝑒 time
∈ 𝑉𝑉𝑉𝑉 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 window. Further, these two for calcula
avel time from
sents 𝑧𝑧𝑧𝑧𝑖𝑖𝑖𝑖 =node {1 𝑖𝑖𝑖𝑖𝑓𝑓𝑓𝑓 𝑖𝑖𝑖𝑖𝑡𝑡𝑡𝑡ℎ𝑒𝑒𝑒𝑒 to 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
node 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑝𝑝𝑝𝑝𝑙𝑙𝑙𝑙𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎; 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 𝑖𝑖𝑖𝑖𝑛𝑛𝑛𝑛 0 𝑡𝑡𝑡𝑡ℎ𝑒𝑒𝑒𝑒
𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏
𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 , 𝑖𝑖𝑖𝑖 ∈ 𝑃𝑃𝑃𝑃. equations are used to make sure that sub-tours will calculation
node
where travel 𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎; time is 0included, 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 the 𝑒𝑒𝑒𝑒 , 𝑖𝑖𝑖𝑖 ∈ 𝑃𝑃𝑃𝑃. According 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖to 𝑘𝑘𝑘𝑘 ≤the 𝑘𝑘𝑘𝑘 ∀𝑘𝑘𝑘𝑘not
𝑆𝑆𝑆𝑆𝑛𝑛𝑛𝑛+𝑖𝑖𝑖𝑖assumptions ∈ 𝐾𝐾𝐾𝐾,be∀𝑗𝑗𝑗𝑗 generated. ∈ above,
𝑉𝑉𝑉𝑉 Equation (9) ensures that each
green main proc
, thethe triangle inequality
isfy Accordingwhere to the assumptions VRPPDP above, can be green explained with a mathematical pickup takes place before the corresponding delivery. travel dista
𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 = 1 ⇒ 𝑄𝑄𝑄𝑄𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 + 𝑞𝑞𝑞𝑞𝑖𝑖𝑖𝑖 ≤ 𝑄𝑄𝑄𝑄 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾,(10)
Equations ∀(𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) to ∈(12) 𝐴𝐴𝐴𝐴 confirm that load variability is mile), for
allhere𝑖𝑖𝑖𝑖. 𝑗𝑗𝑗𝑗. 𝑙𝑙𝑙𝑙 ∈ 𝑉𝑉𝑉𝑉.VRPPDP 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖 is service can be timeexplainedmodel with asa follows: mathematical
time
nd unloadingmodel vehicles as follows: at node 𝑖𝑖𝑖𝑖. precisely set along the path and confirm the use of fuel used.
𝑄𝑄𝑄𝑄𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 ≤ 𝐶𝐶𝐶𝐶𝑘𝑘𝑘𝑘 ∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉
de 𝑖𝑖𝑖𝑖. Minimise vehicle capacity constraints. Lastly, the nature of the converted
Minimise 𝑄𝑄𝑄𝑄𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘 = 𝑄𝑄𝑄𝑄𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘′𝑘𝑘𝑘𝑘 decision = 0∀𝑘𝑘𝑘𝑘 ∈ variables 𝐾𝐾𝐾𝐾 is set up in Equations (13) to (16). are determ
𝛼𝛼𝛼𝛼 ∑𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘 ∑(𝑖𝑖𝑖𝑖,𝑖𝑖𝑖𝑖) 𝑑𝑑𝑑𝑑𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 (𝜀𝜀𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶2 + 𝜀𝜀𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶4 + 𝜀𝜀𝜀𝜀𝑁𝑁𝑁𝑁2𝐶𝐶𝐶𝐶 ) (1) Various researchers have studied the alternative a distance-
𝑖𝑖𝑖𝑖) 𝑑𝑑𝑑𝑑 𝑥𝑥𝑥𝑥 (𝜀𝜀𝜀𝜀
𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶2 + 𝜀𝜀𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶4 + 𝜀𝜀𝜀𝜀 𝑁𝑁𝑁𝑁2 𝐶𝐶𝐶𝐶 ) 𝑥𝑥𝑥𝑥 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 ∈ {0,1}∀𝑘𝑘𝑘𝑘 ∈ fuel 𝐾𝐾𝐾𝐾, ∀(𝑖𝑖𝑖𝑖, for 𝑗𝑗𝑗𝑗) Green
∈ 𝐴𝐴𝐴𝐴 VRP by considering the GHG equations
The goal of green VRPPD Subject tois to design a vehicle emission (Xu et al., 2019; Bruglieri et al., 2019; Sruthi
Subject route to by minimising three GHG emissions (CO2𝑧𝑧𝑧𝑧, 𝑖𝑖𝑖𝑖 ∈ {0,1}∀𝑘𝑘𝑘𝑘 et al., ∈ 𝑃𝑃𝑃𝑃 2019). GHG emissions are calculated by
∑𝑘𝑘𝑘𝑘∈𝑘𝑘𝑘𝑘 ∑𝑖𝑖𝑖𝑖∈𝑁𝑁𝑁𝑁𝑘𝑘𝑘𝑘 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 + 𝑧𝑧𝑧𝑧𝑖𝑖𝑖𝑖 = 1∀𝑖𝑖𝑖𝑖 ∈ 𝑃𝑃𝑃𝑃 (2) 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶2 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
CH4, and N2O) from the transportation vehicle, as multiplying fuel consumption by the emissions factor
∑𝑖𝑖𝑖𝑖∈𝑁𝑁𝑁𝑁𝑘𝑘𝑘𝑘 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 + 𝑧𝑧𝑧𝑧𝑖𝑖𝑖𝑖 = 1∀𝑖𝑖𝑖𝑖 ∈ 𝑃𝑃𝑃𝑃 𝑆𝑆𝑆𝑆 ≥ 0∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉
presented ∑𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉 in Equation (1). This minimisation ensures
𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 − ∑𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖 𝑛𝑛𝑛𝑛+𝑖𝑖𝑖𝑖 𝑘𝑘𝑘𝑘 = 0∀𝑖𝑖𝑖𝑖 ∈ 𝐾𝐾𝐾𝐾, ∀𝑖𝑖𝑖𝑖 ∈ 𝑃𝑃𝑃𝑃 (3)
𝑖𝑖𝑖𝑖 𝑘𝑘𝑘𝑘 of GHG for each fuel type. The distance travelled is
𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
that the
∑𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖 𝑛𝑛𝑛𝑛+𝑖𝑖𝑖𝑖 𝑘𝑘𝑘𝑘 = 0∀𝑖𝑖𝑖𝑖 ∈ 𝐾𝐾𝐾𝐾, ∀𝑖𝑖𝑖𝑖 ∈ 𝑃𝑃𝑃𝑃 pickup and delivery orders are performed by one of the most significant influencing factors for4
𝑄𝑄𝑄𝑄𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 ≥ 0∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉
the same vehicle, ∑𝑖𝑖𝑖𝑖∈𝑃𝑃𝑃𝑃∪�𝜏𝜏𝜏𝜏 and′ the 𝑥𝑥𝑥𝑥orders are implemented in (4)
𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 = 1∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾
calculating GHG emissions. Consequently, GHG 𝑁𝑁𝑁𝑁2 𝐶𝐶𝐶𝐶 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
𝑘𝑘𝑘𝑘 �

∪�𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘 � 𝑥𝑥𝑥𝑥𝜏𝜏𝜏𝜏 𝑘𝑘𝑘𝑘
Equations
𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 = 1∀𝑘𝑘𝑘𝑘 ∈ (2)
𝐾𝐾𝐾𝐾 and (3). The confirmations for the emissions are calculated by multiplying the travel
conditions that∑each 𝑖𝑖𝑖𝑖∈𝐷𝐷𝐷𝐷∪{𝜏𝜏𝜏𝜏vehicle 𝑘𝑘𝑘𝑘 }
𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘′departs
𝑘𝑘𝑘𝑘 = 1∀𝑘𝑘𝑘𝑘from ∈ 𝐾𝐾𝐾𝐾 its starting (5) distance The by energy a distance-based consumption emissions (in litres/100 factor. km)The is pid is the b
∪{𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘 } 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘terminal

𝑘𝑘𝑘𝑘 = 1∀𝑘𝑘𝑘𝑘 ∈and 𝐾𝐾𝐾𝐾 stops at its ending terminal are emissions factoraccording
approximated for CO2 relies to Ubeda on several et factors al. (2014). such the particl
performed ∑𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉 in𝑥𝑥𝑥𝑥Equations 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 − ∑𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉 (4) 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖and = 0∀𝑘𝑘𝑘𝑘 (5).∈Equation 𝐾𝐾𝐾𝐾, ∀𝑗𝑗𝑗𝑗 ∈ 𝑁𝑁𝑁𝑁 (6) (6) as fuel heatthe content, fraction of oxidised carbon
However, fuel conversion factor for dieselinand the the dth dim
∑𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖ensures 𝑖𝑖𝑖𝑖 = 0∀𝑘𝑘𝑘𝑘 that ∈ 𝐾𝐾𝐾𝐾, ∀𝑗𝑗𝑗𝑗 ∈ 𝑁𝑁𝑁𝑁
consecutive paths between 𝜏𝜏𝜏𝜏 and 𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘′ are fuel, (kgCO and the carbon content from coefficient,
𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 = 1 ⇒ 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 + 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 + 𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ≤ 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾,𝑘𝑘𝑘𝑘∀(𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) ∈ 𝐴𝐴𝐴𝐴 (7) LPG 2/litre) is obtained DEFRAwhich (2013). is pgd is the b
used for each vehicle. Equations (7) and (8) guarantee somewhat
The approximation difficult to obtain.
of Therefore,
carbon dioxide only vehicle
(CO 2), the swarm
+ 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 + 𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ≤ 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀(𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) ∈ 𝐴𝐴𝐴𝐴
that 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 is set𝑎𝑎𝑎𝑎 ≤ correctly𝑆𝑆𝑆𝑆 ≤ 𝑏𝑏𝑏𝑏 along
∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, the
∀𝑗𝑗𝑗𝑗 ∈ paths
𝑉𝑉𝑉𝑉 within (8) travel distance
methane (CH 4 ), (distance-based
and nitrous oxide approach)(N 2 O) is applied
emissions d th
dimens
𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 𝑖𝑖𝑖𝑖
a
𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 ≤ 𝑏𝑏𝑏𝑏𝑖𝑖𝑖𝑖 ∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀𝑗𝑗𝑗𝑗 ∈ 𝑉𝑉𝑉𝑉 particular time window. Further, these two for
factors calculating for diesel GHGand LPG emissions are presented in this in study.
Table The
1. x id is th
equations are used 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 ≤to𝑆𝑆𝑆𝑆𝑛𝑛𝑛𝑛make +𝑖𝑖𝑖𝑖 𝑘𝑘𝑘𝑘 ∀𝑘𝑘𝑘𝑘 sure∈ 𝐾𝐾𝐾𝐾, that
∀𝑗𝑗𝑗𝑗 ∈ sub-tours
𝑉𝑉𝑉𝑉 will (9) calculations of GHG emissions are divided into two dimension
≤ 𝑆𝑆𝑆𝑆𝑛𝑛𝑛𝑛+𝑖𝑖𝑖𝑖 𝑘𝑘𝑘𝑘 ∀𝑘𝑘𝑘𝑘not ∈ 𝐾𝐾𝐾𝐾,be ∀𝑗𝑗𝑗𝑗 generated.
∈ 𝑉𝑉𝑉𝑉 Equation (9) ensures that each Tab.
main 1. processes. Approximation In the beginning, of carbon datadioxide
is collected (COfor 2), C1, C2 is eq
pickup𝑥𝑥𝑥𝑥takes 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 = place 1 ⇒ 𝑄𝑄𝑄𝑄before 𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘 + 𝑞𝑞𝑞𝑞𝑖𝑖𝑖𝑖 the ≤ 𝑄𝑄𝑄𝑄corresponding 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀(𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) ∈ 𝐴𝐴𝐴𝐴 (10)
delivery. methane
travel distance, (CH4),inand terms nitrous of freight oxidedistance (N2O) emissions (e.g., ton- rand( )
𝑘𝑘𝑘𝑘 + 𝑞𝑞𝑞𝑞𝑖𝑖𝑖𝑖 ≤ 𝑄𝑄𝑄𝑄 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾,(10)
Equations ∀(𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) to∈(12) 𝐴𝐴𝐴𝐴 confirm that load variability is factors
mile), for for different a 10-ton capacity vehicle types, truck sizes, and types of within (0,1
𝑄𝑄𝑄𝑄 ≤ 𝐶𝐶𝐶𝐶𝑘𝑘𝑘𝑘 ∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾, ∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉
precisely set along𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘the path and confirm the use of fuel used.
2.1. The particle swarm optimisation (PSO) The approximated freight distance is then As sh
≤ 𝐶𝐶𝐶𝐶𝑘𝑘𝑘𝑘 ∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾,4∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉
vehicle capacity constraints. 𝑄𝑄𝑄𝑄𝜏𝜏𝜏𝜏 𝑘𝑘𝑘𝑘 = 𝑄𝑄𝑄𝑄𝜏𝜏𝜏𝜏′ 𝑘𝑘𝑘𝑘 = 0∀𝑘𝑘𝑘𝑘 ∈ 𝐾𝐾𝐾𝐾
Lastly, the nature of the converted
algorithm for green VRPPD into GHG emissions. The GHG emissions particle in
decision variables is𝑘𝑘𝑘𝑘 set up𝑘𝑘𝑘𝑘in Equations (13) to (16). are determined by multiplying the freight distance by The first t
that the pickup and delivery orders are performed by one
However, of thethe most fuelsignificant conversioninfluencing factor for factors diesel and for the d
the same vehicle, and the orders are implemented in calculating
LPG (kgCO2GHG /litre) is emissions. obtained Consequently, from DEFRA (2013). GHG pgd is
The goal of green VRPPD is to design a vehicle emission (Xu et al., 2019; Bruglieri et al., 2019; Sruthi
Equations (2) and (3). The confirmations for the emissions
The Engineering approximation areManagement calculated of by incarbon multiplying
Production dioxide and theServices (COtravel 2), the s
route 14by• Issue
Volume minimising 1 • 2022 three GHG emissions (CO2, et al., 2019). GHG emissions are calculated by
conditions that each
The goal of green VRPPD is to design a vehicle vehicle departs from its starting distance
methane emission by
(CH (Xua distance-based
4), et and al., nitrous 2019; Bruglieri emissions
oxide (N et 2al., O) 2019;factor.
emissions The
Sruthi dth d
CH4, and N2O) from the transportation vehicle, as multiplying fuel consumption by the emissions factor
VRPPD is to design terminal route
a vehicle
Tab. 1. Approximation
and by stops
minimising at
emission
of carbon
its
three ending
dioxide (Xu GHG et al.,terminal
emissions
(CO2), methaneensures 2019; are
Bruglieri (CO
(CH4), and nitrous oxide 2 , et al., emissions
factors 2019;et al., for Sruthi
(N2O)for
factor
diesel
2019). emissions
for
and GHG CO LPG
factors 2 relies are
emissions for a The
on
presented 10-ton
several are factors
in
calculated
capacity truck Table such1. by xid
emission (Xu etpresented
al., The goal in Equation
of green (1).
VRPPD This minimisation
is to Equationdesignvehicle, a vehicle of GHG
emission (Xu each
et consumption
al., fuel 2019; type. Bruglieri distance
et al., 2019; travelled Sruthi is
hree GHG emissionsperformed CH2019; , and
4(CO in Bruglieri
, Equations
N O) from et et al., al.,
(4)
the 2019;
and
transportation
2019). Sruthi
(5). GHG emissions (6) as are ascalculated fuelmultiplying heat content,
by fuel fraction of oxidised
by the carbon
emissions in the
factor dime
a vehicle emission
that the (Xu
pickup
2 2
et al.,
andare 2019;delivery Bruglieri orders et al.,
are performed 2019; Sruthi
(COby et one al.,of the most
2019). significant influencing factors for
et al., 2019). route
GHG byemissions minimising three
calculated GHG emissions
by 2, Tab. 1. Approximation eachGHG emissions
of carbon are
dioxide calculated (CO2is ),isby

CH4 FUEL CONVERSION FACTOR FOR

CH4 FUEL CONVERSION FACTOR FOR


ensures that consecutive paths between 𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘 and 𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘′ensures are fuel, of and the for carbon fuelcontent coefficient, which

CO2 EMISSIONS FACTOR FOR DIESEL

CH4 EMISSIONS FACTOR FOR DIESEL


presented in Equation (1). This minimisation GHG type. The distance travelled C 1, C

N2O EMISSIONS FACTOR FOR LPG


CO2 EMISSIONS FACTOR FOR LPG
FUEL CONSUMPTION ( LITRE/100

CH4 EMISSIONS FACTOR FOR LPG


he transportation vehicle, as multiplying fuel consumption by the emissions factor

N2O FUEL CONVERSION FACTOR

N2O FUEL CONVERSION FACTOR


CO2 FUEL CONVERSION FACTOR

CO2 FUEL CONVERSION FACTOR


ons (CO2, etthe al., same 2019). vehicle, GHG and emissions orders are
are calculated
implemented by in calculating GHG emissions. Consequently, GHG

N2O EMISSIONS FACTOR FOR


CH , and N O) from the transportation vehicle, as multiplying fuel consumption by the emissions factor

FOR DIESEL (KG N2O/ LITRE)


FOR DIESEL (KG CO2/ LITRE)
multiplying fuel consumption byand the emissions factor methane of(CH ), and nitrous oxide (N2O) emissions

FOR LPG (KG N2O/ LITRE)


used forensures each vehicle. Equations (7) and are (8) guarantee somewhat difficult tosignificant obtain. Therefore, only vehicle

FOR LPG (KG CO2/ LITRE)


1). This minimisation that 4
the pickup 2
of(3). delivery
GHG for orders each fuelperformed type.for Thethe bydistance one
travelled the is4most influencing factors for rand

DIESEL (KG CH4/ LITRE)


vehicle, as multiplying fuel consumption by the emissions factor

LPG (KG CH4/ LITRE)

DIESEL (KG N2O/KM)


Equations (2) and The confirmations emissions are calculated by multiplying the travel
WEIGHT LADEN (%)
presented invehicle,
Equation (1). This minimisation ensures of GHG for each fuel type. The distanceis travelled is
STATE OF VEHICLE

very of GHG
orders are performed for that each the fuel
𝑆𝑆𝑆𝑆 same is type. by set Thecorrectly distance
and
onevehicle the along
of the travelled
orders the
most are is
paths
implemented
significant within influencing in factors
travel calculating
factors for
distance a for 10-ton (distance-based
GHG capacity emissions. truck approach)
Consequently, applied GHG with

(KG N2O/KM)
(KG CO2/KM)

(KG CO2/KM)

(KG CH4/KM)

(KG CH4/KM)
𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘
on ensures of GHG
conditions for thateach each fuel type. The departs distance from travelled
its starting is onedistance of theare by most a distance-based significant emissions
influencing factor.
factors The
heormed one ofare
orders the amost that
particular
Equations the pickup
significant time and
(2)influencingand window.delivery
calculating(3). The orders
factorsFurther, confirmations
GHG
are these
for performed
emissions. two the
for by
Consequently, for emissions
calculating
TheGHG GHG calculated emissions by multiplying in this study. the Thefor
travel
by implemented inand 2.1.calculating particle swarm optimisation (PSO)
KM)
one
theterminal same of the vehicle, moststops significant
and theat its
orders influencing
ending are implemented terminal factors for are
in emissions factor GHG for CO
emissions. 2 relies on several
Consequently, factors GHG such
calculating
The confirmations GHG
equations conditionsfor emissions. are the used that to Consequently,
each make
emissions vehicle suredeparts are GHG
thatcalculated from its starting
sub-tours bywillmultiplying calculations distance
the travel by of green aGHG distance-based emissions are emissions dividedfactor. into two The parti
mented in calculating
performed GHG
in and Equations emissions. (4) and Consequently, (5). terminal Equation GHG (6) algorithm as fuel heat
emissions for are content, VRPPD fraction by ofmultiplying oxidised carbon in the
emissionsfrom arenotEquations calculated
terminal
be generated. (2)by
and multiplying
stops
Equation (3). atThe
distance its
(9) the confirmations
atravel
ending
ensures that each for the are main emissions processes. factor Incalculated for CO
the beginning, 2 relies on data several is collected factors the such travel
for The
hicle
ns fordeparts the its
emissions
ensures
conditions
starting that are
that calculated
consecutive
each vehicle byby
paths multiplying
departs between distance-based
from 𝜏𝜏𝜏𝜏 itsthe
and startingtravel
𝜏𝜏𝜏𝜏 ′ emissions factor. The
are fuel,
distance and by the a carbon
distance-based content emissionscoefficient, factor. which Theis
distance bypickup a distance-basedperformed takesare place in before Equations
emissions (4)factor
factor.
the corresponding andThe (5). Equation
delivery. (6)several travel as fuel distance, heat
suchcontent, in terms fraction ofisfreight of oxidised distance carbon ton- in the
with(e.g.,
𝑘𝑘𝑘𝑘 𝑘𝑘𝑘𝑘
t starting
ts its ending terminal distance by a distance-based emissions emissions for CO 2 reliesThe
factor. on factors The PSO algorithm initialised random parti
used
terminal
ensures for eachand vehicle.
stops Equations
at its ending (7) and (8)
terminal guarantee ′ are somewhat
emissions difficult
factor for to
CO obtain.
relies Therefore,
on several only
factors vehicle
such
emissions factor
Equations for CO (10) 2that
relies toconsecutive
(12)on several confirm paths
factorsthat betweensuchvariability
load 𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘 and 𝜏𝜏𝜏𝜏is are mile), fuel,for and different the carbon vehicle content
2 types, coefficient, sizes,The andvelocity whichofis
types of th
s (4) and
are (5). Equation (6) as fuel heat content, fraction of 𝑘𝑘𝑘𝑘 oxidised carbon
position, indistance the
minal
as fuel heat content,
precisely
emissions
that
performed
usedfraction 𝑆𝑆𝑆𝑆for
′𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘0 each
set
factor
isinset
along of
for
Equations
vehicle. correctly
oxidised
the
CO
path
2 relies
Equations (4)
carbon
and
along on(7)
andconfirm in
severalthe
(5).
theandEquation paths
(8)
the
factors guarantee
use
within
of
such (6) fuel as travel
somewhat fuel velocity,
used. heatThe content,
difficult
approximated
and(distance-based inertia
tofraction obtain.
weight.
freight ofTherefore, oxidisedapproach)
distance carbon
only is
isvehicle
then inisthe
applied
euation paths (6) between 𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘as Empty
and 𝜏𝜏𝜏𝜏 are 29.6 fuel, and the carbon content coefficient, constantly 0.786442 which 0.45291 is
adjusted 0.000266
according 0.000207 0.005654
to inthe 0.000533 It is
athat
ensures fuelparticular 𝑘𝑘𝑘𝑘heat content,
𝑆𝑆𝑆𝑆that 𝑘𝑘𝑘𝑘 isconsecutive set time fraction
correctly window. paths of along oxidised
between Further, 𝜏𝜏𝜏𝜏carbon
𝑘𝑘𝑘𝑘 and
thesewithin 𝜏𝜏𝜏𝜏in′ two
𝑘𝑘𝑘𝑘 are
the for calculating
fuel, and the carbon GHG emissions
content coefficient, thisparticle’s isstudy. which Theis
uations fuel, andand
(7) the vehicle
(8)fuel, carbonLowcapacity
guarantee 𝑖𝑖𝑖𝑖content constraints. coefficient,
somewhat Lastly, which
difficult thethe is
nature to
paths
obtain. of the Therefore, converted
experience, only
travel
vehicle
distance
into and GHG (distance-based
its emissions.
group’s The
experience,
approach) GHG emissions to
applied
move of th
and 𝜏𝜏𝜏𝜏𝑘𝑘𝑘𝑘′ are equations
used a forand
particular 25 each the
are 32 carbon
used
vehicle. time to content
make
Equations
window. sure coefficient,
(7)
Further,that
and sub-tours
(8) which
guarantee
these will
two is 0.850208calculations
somewhat
for calculating 0.489632 difficult of GHG
GHG0.000288 to emissions
obtain.
emissions0.000224 Therefore, are
in 0.006112 divided
this only
study.0.000576 into
vehicle
The two
tly somewhat
along thedifficult decision
paths
Load to
within obtain. Therefore,
variables is settravel up indistance only
Equations vehicle (13) to (16). approach)
(distance-based are determined
is applied by multiplying the freight distance by parti
guarantee somewhat
not difficult toEquation obtain. Therefore, only vehicle towards the better Insolution, as are described in applied the
travel distance that Half be
equations
(distance-based
Various 𝑆𝑆𝑆𝑆𝑖𝑖𝑖𝑖𝑘𝑘𝑘𝑘50generated.
is are
researcherssetused correctly
approach) tohave make along
is
studied sure(9)
applied ensures
the
that
the paths
sub-tours
alternative
that within each
will a
main
travel
calculations
distance-based
processes.
distance ofemissions (distance-based
GHG the beginning,
emissions factor, as approach)data
divided
explained
is collected is
intoin thetwofor
ndow.
hs within Further, travel these
Load distance two 34.4 2.6569
(distance-based for calculating
1.5301 0.0009
approach) GHG 0.0007 emissions
isthat 0.0191 0.0018
applied in this following study.
0.913974 The
0.526354
equations: 0.00031 0.000241 0.00657 0.000619 mov
for calculating a pickup
not particular be takes generated. place
time before window.
Equation the corresponding
Further,
(9) ensures these delivery. two
each fortravel
main calculating distance,
processes. In in
GHG the terms of
emissions
beginning, freight datain distance
this
is study.
collected (e.g., ton-
The
for
make fuel GHG for Green emissions VRPin bythisconsideringstudy. The the GHG are divided equations below: (sear
hese sure two that sub-tours for Highcalculating
Equations
pickup
will
takes (10) place GHG
to calculations
(12) before emissions
confirm the
of
that
corresponding in GHG this
load emissions
study.
variability delivery. The is mile),
travel
into for twodifferent vehicle types, sizes, and types of
distance, in terms of freight distance (e.g., ton-
ation calculations (9) ensures ofequations GHGLoad
that each
75
emissions are used
36.7 are todivided
main
make sure
processes. into that two In
sub-tours will
the beginning, datavis
calculations of GHG emissions are divided into two
0.975082
collected
0.561547 0.00033 0.000257 0.00701 0.000661
in la
tours will calculations
notprecisely
Equations be generated.set ofalong
(10) GHG to (12) the
Equation emissions
path
confirm and
(9) areensures
that confirmdivided
load the
variability
thatintouse each two of
is id = w*vidused.
fuel
mile),
main + for
for
processes. Cdifferent
rand(approximated
1The
In)*(p vehicle
the id-xid) types,
beginning, +C2freight rand( data)*(p
sizes, distance
isand gd-xtypes
collected id) (20)
is thenof
for
ethat main
the corresponding processes. InFull the
delivery. beginning, data
travel is collected
distance, infor terms of freight distance 1.036191 ton-
(e.g., wou
each main
vehicle
pickup precisely
Load processes.
100
takescapacity set 39 In the beginning, data is collected for
place along constraints.
before the path
the Lastly,
and
corresponding confirm the nature the delivery. of
use the of converted
fuel
travel used. 0.596739 0.000351 0.000273 0.007449 0.000702
distance, The into GHG
approximated
in terms emissions.
of freight freight The GHG
distance
distance emissions
is
(e.g., thenton-
travelthat distance, in terms is of freightmile), distance (e.g., ton- vehicle types, 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
of xGHG
𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑒𝑒𝑒𝑒
id = xid + vid*Δt
× 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑓𝑓𝑓𝑓𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶
(21) (17)
suita
gonfirm load variability offor indifferent sizes, and aretypes
2 2
delivery. travel
decision
vehicle
Equations distance, variables
capacity
(10) inconstraints.
to terms
(12) issizes, set up
confirm freight Lastly, thatdistance
Equations the nature
load (13) (e.g.,
variability toof(16). ton- the is converted
mile), determined for into different by multiplying emissions.
vehicle types, thesizes,
The freight
GHGand distance
emissions types of by
path mile), for
andisconfirmmile), different the vehicle
use types, and types of where a bal
riability Tab. decision
precisely 2.Various forof
Comparison
setdifferent
variables along of results
researchers fuel
isvehicle
the set used.
from
up
path have types,
in The
two
and studied
Equations approximated
sizes,the
different
confirm and
objective
(13) the types
alternative
to use
freight
functions
(16).
𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻 4 of of
𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚
distance
for theare
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 a
fuel
=
is𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖then
cluster-distributed
distance-based
determined
used.
𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖The by 𝑒𝑒𝑒𝑒customer
approximated
𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑒𝑒𝑒𝑒emissions
multiplying
𝑒𝑒𝑒𝑒 × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 dataset factor,
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚 the
freight
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 as
freight explained
distance
𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 distance
𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻 is in by
(18)
thenthe
ints. fuel Lastly, used.the The nature approximated of theThe freight converted distanceinto is then GHG emissions. The GHG vid a emissions is the velocityemissions of the factor, ith particle in theindthe
4 th
the use of fuel
fuel
vehicle used. Various
for capacity Green approximated
researchers
constraints.VRP by have freightstudied
considering
Lastly, M INIMISING the distance thenature theisofGHG
COSTalternative then the distance-based
equations
converted MINIMISINGinto below:
GHGGHG EMISSIONS emissions. DThe as explained
ISTANCE GHG INCREASE emissions (%)
up converted into GHG emissions. The GHG emissions dimension. size,
ureinofEquations the (13) converted
fuel
decision
to I(16). NSTANCE
forvariables intoGreen GHG are
VRP determined
emissions. byin Equations
DISTANCE considering The byGHG multiplying emissions
the to𝑁𝑁𝑁𝑁(16).
NV
the freight
2 𝐶𝐶𝐶𝐶 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 =
GHG distance
𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖by
equations
DISTANCE 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 ×NV
below: 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑁𝑁𝑁𝑁2 𝐶𝐶𝐶𝐶 (19)
s3)have are studied determined the by multiplying
alternative theisafreight set up
distance-based distance by emissions
(13)
factor, as w
explained is
are
the
determined
inertia
in989.91 the weight.
by multiplying the freight distance by
and
to (16). are determined Various LC101 by
researchers multiplying have the
828.94
studied freight the distance
alternative by 10 a distance-based emissions factor, 12 as explained 19.42 in the
bya distance-based
lternative considering athe Theemissions
distance-based energy
GHG factor,
consumption as explained
equations (inbelow: in
litres/100the km) isthe 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶210𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚p𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 is = the𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖828.94
best previous𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒position 𝑚𝑚𝑚𝑚(the
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚position giving itera
Green emissions VRP by factor, as explained the in
id𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑒𝑒𝑒𝑒 × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚10 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑓𝑓𝑓𝑓𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶20 (17)
fuel for LC102 828.94
considering GHG equations below:
equations below: approximated according to Ubeda et al. (2014). the particle’s best fitness value) of the i th
particle in that
he GHG equations below: LC103 827.86 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 2 10 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖
827.86𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑒𝑒𝑒𝑒 × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚
10 𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑓𝑓𝑓𝑓𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶2 0 (17)
However, LC104 the fuel conversion factor 861.95 for diesel and 𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻49𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚the 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚d𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 dimension.
th
= 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 903.85 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚9𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻 4.86 4 (18)
LPG (kgCO LC105 2/litre)
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 is𝑚𝑚𝑚𝑚obtained
2 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖from
𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖828.94 DEFRA
𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 ×(2013). 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚10 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚p 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚
4𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 the
is
𝑓𝑓𝑓𝑓𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 best
828.94𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖(17)
𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
previous𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒position × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚10 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚
(the 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎
position 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻
giving 4 0 (18)
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑓𝑓𝑓𝑓𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶2 (17)
(17) gd 2
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶2 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑓𝑓𝑓𝑓𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶2 (17)
𝑁𝑁𝑁𝑁2 𝐶𝐶𝐶𝐶10𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚the 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖best
𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
fitness𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒value) 𝑒𝑒𝑒𝑒 × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚the
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚
ith𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛in 𝐶𝐶𝐶𝐶0 (19)
𝑁𝑁𝑁𝑁2the
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶2 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚The
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 =approximation 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒of
𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖LC106 𝑒𝑒𝑒𝑒 × carbon 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚dioxide
𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚828.94 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑓𝑓𝑓𝑓𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 (CO𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶22), (17) swarm’s 828.94 of10 particle
𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻4),𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑁𝑁𝑁𝑁𝑚𝑚𝑚𝑚2 𝐶𝐶𝐶𝐶
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚th𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚
𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛=𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
(18) 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑁𝑁𝑁𝑁2 𝐶𝐶𝐶𝐶 (19)
methane𝑒𝑒𝑒𝑒(CH
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒The
LC107
𝑒𝑒𝑒𝑒 ×energy 4 and nitrous oxide
𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚consumption
828.94
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻(in
(N2(18) O)𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒emissions 𝑒𝑒𝑒𝑒 × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚
(18) 𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻10
𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎
d dimension. 4 828.94 10
4 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻4
0 (18)
4 litres/100 km)
𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻1.4 (18)10 xid is the
is p id is the best previous position (the position giving
𝐶𝐶𝐶𝐶𝐻𝐻𝐻𝐻4 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚factors
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 =for
The
LC108
𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖diesel and LPG
𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 ×are𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚
presented 827.61
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 in Table
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎
pthe
827.61 position of the ith10particle inth thegiving d0th
approximated 𝑁𝑁𝑁𝑁2 𝐶𝐶𝐶𝐶energy 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚consumption
𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚according to𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖827.82
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 Ubeda (in litres/100
𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒et
𝑒𝑒𝑒𝑒 ×al.𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 km)
(2014). is 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 id 𝑁𝑁𝑁𝑁is𝐶𝐶𝐶𝐶the (19)
2particle’s
best previous
best fitness position value) (the of position
the i particle in
LC109 (19) 10
𝑁𝑁𝑁𝑁2 𝐶𝐶𝐶𝐶 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚dimension. 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 =particle’s 827.82
𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 × value) 10
𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚of 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 the th 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑁𝑁𝑁𝑁2 𝐶𝐶𝐶𝐶 0 (19)
𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎
approximated
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
However,
𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 × 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚according
the
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛to𝑁𝑁𝑁𝑁2Ubeda
fuel𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒conversion
𝐶𝐶𝐶𝐶 (19) et al. (2014). the
the dth 591.56 best fitness i particle in
𝑚𝑚𝑚𝑚factor for𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛diesel 𝑁𝑁𝑁𝑁2 𝐶𝐶𝐶𝐶2),and dimension.
th
𝑁𝑁𝑁𝑁2 𝐶𝐶𝐶𝐶 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚Tab.
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 1. 𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
= Approximation
LC201
𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑇𝑇𝑇𝑇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑒𝑒𝑒𝑒 of carbon
× 𝐸𝐸𝐸𝐸𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚dioxide
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚591.56 𝐹𝐹𝐹𝐹𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 (CO (19) 3 C , C 3 0
mption (in litres/100 However, km) is the fuelpidconversion is the bestfrom factor
previous for diesel position and(the position 1 the 2 is d giving equal
dimension. to 2.
ptoid Ubeda is the best methane LPG The (kgCO
(CH
LC202 energy ), /litre)
and consumption is
nitrous obtained oxide (in
591.56 (N O) DEFRA
litres/100 emissions (2013).
km) is 3 p p is the
the 591.56 best
best previous
previous position
position 3 (the
(the position
position giving
giving
0
et previous position (the position giving
2 gd
of3therand( ) is in a uniform previousrandom positionnumber generated
4 2 id
ng00 km) is pal.
The
LPG is (2014). the (kgCO
approximation best 2/litre)the
previous is obtained
particle’s
position
of carbon
from
best
(the DEFRA
fitness value)
position
dioxide
(2013).
giving
(CO ), ithpthe gd is the best
particle swarm’s best fitness value) of
(the position
the i th th
particle
giving
in
0 the
the particle’s factors approximated
id LC203
for approximation a 10-ton value)capacity according itruck to 591.17
Ubeda et al. (2014). the particle’s 591.17 best
(0,1). best fitness value) of the ith particle in thein fitness value) 3 of the i particle
factor forbest fitness ofthethe particle indioxide
th 2
version The
diesel and(CH dthof carbon
dimension. (COin 2), within theth swarm’s
l. (2014). the
methane particle’s LC204
However, the fuel conversion factor for diesel and best ), and fitness nitrous value) oxide
590.60 of the(N iO) th
particle
emissions 3 thed dimension.
dshown th 590.60
dimension. 3 0
btained the dfrom th
dimension. DEFRA2.1.the methane
The (2013). (CH4), and
particle
4
swarm
pgd LPG nitrous
is theare optimisation
best oxide previous (N2O) position
2
emissions(the
(PSO) position dthAsdimension. giving in Equation (21), theth velocity of the ith th
diesel and LPGfactors d th LC205
(kgCO dimension.
for diesel
2/litre)(the and is obtained 588.88
presented
from DEFRA in Table(2013). 1. 3 x
pxgdid isisin is 588.88
thethe the best position
thprevious position of the 3 i (the particle in the 0 d
ofA (2013). pcarbon
gd is the best previous
dioxidealgorithm factors (CO for position
for), diesel VRPPD
green and the position
LPG
swarm’s are 588.29 giving
presented
best fitness invalue) Table 1. the
of i thparticle
particle id in the
the dposition dimension of the consists i3th particle of positionthree in terms. the givingdth
p
The gd is the LC206
approximation
2best previous position
thof carbon dioxide (CO2), (the position giving 3
thedimension. swarm’s 588.49
best fitness value) of the i th
particle 0.04
in the
trous
e (CO
theoxide swarm’s (N2best O)the fitness
emissions
Tab. 1.LC207
value) of the
Approximation dth idimension. particle588.29
of oxide carbon
in theth
i dioxide (CO 2), 3
Thedimension. first term is the momentum3 of the part of the
th1, C2 588.29 0
th 2), methane swarm’s (CH
Tab. 1. Approximation of carbon dioxide (COth2),best ), fitness
and value)
nitrous of the (N particle
O) emissions in the d C dimension. is equal to 2.
G are d dimension. presented indmethane Table
The PSOLC2081. (CH
4
algorithm xid nitrous isisinitialised
theoxide position
588.32 with
2
ofrandom the i particle 3 particle. Cin 1, Cthe 2 Theis588.32 dthinertia
equal to 2.weight (w) represents 3 the degree 0
emissions th
factors dimension. for diesel ),thand and LPG are presented (N O) in Table emissions 1. rand(
xrand( is))the is positiona uniform of random
the ith previous numberingenerated
particle the dth
xid is the position, position methane of (CH44i),
the andparticle nitrous in oxide
the d th 22O)
(N emissions of the id momentum is a uniformof the particle’s
random number velocity.
generated
Table 1. factors
xidfactors velocity,
for
is the aposition a 10-ton and dimension.
inertia
capacity
of the truck weight. trucki particle in the d
th The velocity is th
within(0,1). (0,1).
Average 1.43
n ofdimension. carbon dioxide constantly (COfor ), 10-ton
2adjusted
capacity
Caccording
1, C2 is equal totothe 2.dioxide particle’s It isdimension. within
a control parameter used to control the influence th
dimension.
Tab.
2.1. 1. Approximation of carbon (CO 2), C1, previous CAs As
2 isshown shown
equal toinEquation 2.Equation (21),the thevelocity velocity of the i
trous
de C(CO 1,oxideC2),is equal (Nexperience,
2O) to 2. The
emissions
2.1. The The particle
and
particle
particle
its rand(
group’s swarm
swarm
swarm ) is
experience, optimisation
optimisation
optimisation
a uniform to random
move
(PSO)
(PSO) number of the generated in
velocity on the(21), current velocityofofthe theith
2 C(PSO)
methane1, C2 is equal (CH to
), 2.
and nitrous oxide (N O) emissions particle
rand( inisthe the dth dimension
th
dimension consists ofthree three terms.
algorithm algorithm for green for(0,1).
VRPPD green VRPPD A) in a duniform random number generated
4 2
acity rand( truck) istowards a uniform algorithm the random for green number VRPPD generated particle. particle larger inertia weight consists would ofpressure terms.
the
emissions rand(
factors )forisabetter a uniform
10-ton within
solution,
capacity random as described
truck number generated in the The
within
Theoffirst first (0,1). term is the momentum of the part of of
part thethe
warm optimisation
within (0,1).following equations:
within (PSO) (0,1). As shown in Equation (21), the velocity movement
particle. the of iterm
The
th particles is the momentum
inertia
towards global
weight (w)
of the
represents
exploration
the degree
As shown in Equation (21), the velocity of fly i
the th
The The PSO algorithm isis initialised with random random particle. for Thea inertia weight (w) represents thecan degree
As shown 2.1. in Equation The (21),
particle the velocity
swarm of the
optimisation dthivelocity (PSO)
th
PSO algorithm particle ininitialisedthe dimension with consists of (searching three terms. new area) because the particle
RPPD
n particle (PSO) in the algorithm As shown in Equation (21), the of the i th
of the momentum dcontrast, of the particle’s previous velocity.
in particle in the dimension consists of three terms.
th
vid =particle dposition,
th
w*v dimension
position, velocity,
for
velocity, consists
green and
and VRPPD inertia
ofinertiathree weight.terms.
weight. The
The velocity isis
-xvelocity
of
large the areas. momentum
the Inparameter
of the particle’s a smaller previous
inertia velocity.
weight
id + in C1rand( dth)*(p The id-xfirstid) +C term2rand( is )*(p the momentum id) (20)terms. of the The part
The first term constantly isconstantly
the momentum
theadjusted dimension
of according
the part
consists
of to
the
ofthe three
gd
particle’s ItItisisafirst aof control
control term is
parameter the momentum
usedusedtotocontrol control of the the the part influence
influence of the
adjusted particle. according The inertia to the weight particle’s
(w) represents would thethe move degree the particle in a smaller search area. The
m is initialised with The random
experience, first termand is the its momentum
group’s experience, of the part to ofmove the ofof
particle. the previous
previous The inertia velocity
velocity weight on on the the
(w) current current
represents velocityvelocity the of
degree
of thethe
particle. The inertia experience, The weight PSO(w) xandid =represents
algorithm xof id +
its the is
vidmomentum
group’s *Δt the
initialised degree
experience, ofwith the random
to (21)
particle’s move previous suitable velocity. selection of the inertia weight should provide
hnertia random weight. Theparticle. velocity The is inertia weight (w) represents the degree of particle.
particle.the momentum A larger inertia
of the weight
particle’s wouldprevious pressure velocity. the
of the momentum where towards
position,
towards of thevelocity, thethe better
particle’s better and Itprevious
solution,
inertia
solution,
isthe a control velocity.
weight. as described
parameter The velocity used
in the
in to is
the
controla the balance influence between the global and local search areas.the
A larger inertia weight would pressure
ccordingis to theof
velocity the
particle’s
following momentum equations: of particle’s previous velocity. It movement
movementis a control of of parameterparticles
particles usedtowards
towards to control
globalglobal the exploration
exploration influence
It is a control vidparameter following
constantly
is the velocity used equations:
adjusted to control of
of the the the
according ithinfluence
previous particle to
velocity thein the particle’s dth current velocity The of computation of PSO depends on population
oup’s
particle’s experience,It is toa move control parameter used to control theon the
influence of (searching
(searching
the previous
the for a new
for a new
velocity area)area) onbecause because
thevelocity, current the the particleparticle
velocity can can
offly fly
the
of the previous dimension. velocity
experience, on the
and current its particle. velocity
group’s A larger of
experience, the inertia to weight move would pressure size, inertia the weight, maximum maximum
ution, as described of the in previous the velocity on the current velocity of the in
inminimum large
large areas. areas. In
In contrast, contrast, a smaller inertia weight
to particle. move A larger w isvtowards vthe
id id=inertia
=w*v w*vididthe
inertia +weight
+C rand(
C1better
weight. 1rand( would
movement
)*(p
)*(p
solution, -x-xidid)) +C
ididpressure +C as22rand(
of the
described
particles )*(ptowards gd-x-xinidid))(20) (20) andparticle.
(20)
the A larger positions inertia andweight aamaximum smaller would inertia pressure
number weight ofthe
pressure the global exploration
gd
ed movement in the particle. A larger inertia weight would would
would
movement move move of the the particle
particles particle in in
towardsa a smallersmaller globalsearch search area.
explorationarea. The The
of following particles equations: towards (searching global exploration for aglobal new area) because the particle iterations. canThe fly initial population size was chosen so
movement of particles x x id= = xx idtowards
++ vv id*Δt*Δt exploration (21)
(21) suitable
suitable
(searching selection
selection of of the the inertia
inertia weight weight should should provide
provide
(searching for a new area) becausein id thelarge particle
id id
areas. canInflycontrast, a smaller that inertia it was weight large enough to cover the search space fly
for a new area) because the particle can
id-xid) +C2rand( )*(p (searching
gd-xid) (20)
where
where for a new area) because the particle can fly a
in a balance
balance
large between
areas. between In the theglobal
contrast, globaland a and smaller local localsearch search
inertia areas.
areas.
weight
in large areas.vidIn= w*v contrast, id + C1a rand( smaller )*(pidinertia
would -xmove id) +C weight2rand( )*(pgd
ththe particle in-xthe ) (20)
aidsmaller search area. The
pgd-xwould id) (20)
inv v large is is areas.
thethe velocityIn
velocity contrast, of
of the
the a smaller
ii th
particle inertia in weight
the d d thth The The computation
computation
would move the particle in a smaller search area. The of of PSO PSO depends depends on on population
population
+ vid*Δt move the particle in a smaller searchselection area. The the inertia weight should
id id
would (21)
move the xparticle suitable smaller of size,provide
suitable selection
dimension.
dimension.
of the inertia weight id = xidshould +invaid*Δt provide
search area. (21) The size,
suitable inertia
inertia selection weight,
weight, of the maximum
maximum
inertia weight velocity,
velocity, should maximum maximumprovide 5
(21) suitable
w w is is the the selection
inertia
inertia of
weight.
weight. a the balance inertia between weight the
should global provide and local search and
and areas.
minimum
minimum positions
positions and and a maximum
a maximum number number of of
a balance between where a balance between the global and local search areas.
the ith particle in the the global dth and local The search computation areas. of PSO depends on population
decreased to 0.4 to balance the global and local learning and attitude change. The limit also
exploration based on a linear function of time determines the search of the problem space. Next, vid
Engineering
(iteration) forManagement
improving in Production
convergence and Services
rate is set to be within the boundary ofVolume [−𝑣𝑣𝑣𝑣𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 so
14 • Issue 1 • ]2022
, 𝑣𝑣𝑣𝑣𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚
(Kennedy, 1997) were tested and compared. It was that the search space of each particle is limited, and
within
found thatthe iteration
when inertialimit based
weight is on
setthe
to trial
be a runs
largeand the toward
particlethe position
cannot moveit isout
multiplying.
of this range. In this If vcase, id is C1
literature.
value, The population
it is difficult for particlessize of 50 was
to perform globalthen and than
greater C2 are set, then
vmax to 2 set
for vall PSO to
id equal runs
vmaxto . But, balance if vid the
selected. The
exploration duringmaximum number of
the beginning of the
iterations
searchand impact
is less than of its, then
-vmax own set trajectory
vid equalwithto -vmax its. neighbours’
The vmax
other Then,
process. parameters were weight
the inertia also initialised
value is by using the
gradually trajectory.
parameter is important because it determines the
same rationales.
decreased so that theThey
good should ensurecan
region found thatbethe
foundsearch resolution Also,withrand(
which) theis aregions
random number
around the current uniformly
spaces
by the are never
search process violated and the
to provide thesolutions obtained
best searching distributed
solutions within theIf range
are searched. vmax is(0,1). rand( then
too high, ) makes the the
are always valid. The same parameter settings were
performance. PSOsystem less predictable
facilitates and more
a global search, flexible
which means so that thateach
used
The for all datasets.
second term in Equation (21) is called the particle
particles is stochastically
might fly past good accelerated
solutions.towardsHowever, its ifown
“cognition In this experiment,
part” because the thedistance
performancebetweenof PSO the for vmaxprevious
is tooposition small,andthen the global
the bestPSOposition. facilitates
solvingprevious
particle’s Green VRPPD
best andbasedcurrent on position
a different (pidvalue
-xid) of Furthermore,
a local search, which vmeans
id is limited
that to keep the
particles may computer not
inertia
would weight
provide withfor
a path thethe
range fromto
particle 0.1return
to 0.9toand its the from beyond
explore overflowing. locallyThisgood
limitregions.
makes itIn more caserealistic of
bestinertia weight sostarted
value achieved far. Thefrom 0.9 in
third term and gradually
Equation to simulate
maximum the incremental
and minimum positions of changes
the variables of human in
(21)decreased
is called theto “social
0.4 to part”
balance the global
because and local
the difference eachlearning
dimension, and they attitude change.to represent
were chosen The limitthealso
exploration
between based on
the swarm’s a linear
previous bestfunction
and currentof time determines
suitable search the space,search
whichof the space. Next, vid
problemdependence.
is problem
(iteration)
position (pgd-xid) for improving
provides a path for convergence
the swarm torate isThe
set to be within
positions ofthe boundary
particles of [−𝑣𝑣𝑣𝑣
in the equation are] so
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 , 𝑣𝑣𝑣𝑣𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚
(Kennedy,
return to their 1997) were tested and compared. It was
best value. that the
updated basedsearchon space of each particle
their movement over isa limited, discrete-and
found
C1 and that Cwhen
2 areinertia weightconstants
positive is set to be a large
called timethe particle
interval cannot
(Δt), with Δtmove out set
usually of this
to 1 as range. depicted If vid is
value, it is difficult
“acceleration for particles
coefficients”. They to areperform
used globalto greater than
in Equation (21).vmax , then set vid equal to vmax. But, if vid
exploration
determine during“pull”
the relative the beginning
of gbest andofpbest. the Thesearch is less than -vmax , then set vid equal to -vmax. The vmax
process.
higher the Then, the inertia
constant, the greaterweightthevalue is gradually
acceleration 3. Research results
parameter is important because it determines the
decreased so that the good region found can be found resolution with which the regions around the current
by the search process to provide the best searching solutions are searched. If vmax is too high, then the
performance. The experiment
PSO facilitates a was conducted
global by using
search, which means simu-that
The second term in Equation (21) is called the lated data from Li and Lim (2001). One
particles might fly past good solutions. However, hundred task if
“cognition part” because the distance between the instances
vmax isweretoo usedsmall,
as a benchmark
then the toPSO evaluate the
facilitates
particle’s previous best and current position (pid-xid) performance
a local search, of the PSO means
which for solving green vehicle
that particles may not
would provide a path for the particle to return to its routing
exploreproblemsbeyond withlocally
pickupgood and delivery.
regions.Each In case task of
best value achieved so far. The third term in Equation is either for a pickup or a delivery.
maximum and minimum positions of the variables in Four different
(21) is called the “social part” because the difference types
eachof datasets
dimension, (LC1, LC2,
they were LR1, LR2) to
chosen were representused to the
between the swarm’s previous best and current testsuitable
the proposed
search space,model.whichThe isfirst set ofdependence.
problem problem
position (pgd-xid) provides a path for the swarm to instances The(LCpositions
problems)ofrepresent
particles customers
in the equation located are
return to their best value. in clusters
updated based on their movement over aservice
that are distributed in the coverage discrete-
C1 and C2 are positive constants called area. Thisinterval
time kind of(Δt),distribution is similar
with Δt usually setto tocustomers
1 as depicted
“acceleration coefficients”. They are used to located in a town
in Equation or in a city. Numerous people stay
(21).
determine the relative “pull” of gbest and pbest. The in certain areas forming clusters. The second set of
higher the constant, the greater the acceleration problem instances (RC problems) represents cus-
and toward the position it is multiplying. In this case, C1 tomer locations that are randomly distributed in the
then and C2 are set to 2 for all PSO runs to balance the coverage service area. This kind of distribution is
and impact of its own trajectory with its neighbours’ similar to customers that live in rural areas. The LC1
g the trajectory. and LR1 problems have a long scheduling horizon,
arch Also, rand( ) is a random number uniformly while LC2 and LR2 problems have a shorter schedul-
ined distributed within the range (0,1). rand( ) makes the ing horizon. Next, the PSO parameter settings of the
were system less predictable and more flexible so that each experiment are explained. The acceleration coeffi-
particle is stochastically accelerated towards its own cients (C1 and C2) were equally set at 2 to balance the
O for previous position and the global best position. impacts of exploratory and exploitative learning
ue of Furthermore, vid is limited to keep the computer experiences. The inertia weight (w) was set to linearly
d the from overflowing. This limit makes it more realistic decrease from 0.9 to 0.4. This decrease allowed the
ually to simulate the incremental changes of human particles to perform global searches at the beginning
local learning and attitude change. The limit also of the search and then gradually decrease the scope of
time determines the search of the problem space. Next, vid the search space to the good region. The population
rate is set to be within the boundary of [−𝑣𝑣𝑣𝑣𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 , 𝑣𝑣𝑣𝑣𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 ] so size and the maximum number of iterations were set
was that the search space of each particle is limited, and at 50 and 500, respectively. These parameter settings
large the particle cannot move out of this range. If vid is
lobal greater than vmax , then set vid equal to vmax. But, if vid
6
arch is less than -vmax , then set vid equal to -vmax. The vmax
ually parameter is important because it determines the
High
75 36.7 0.975082 0.561547 0.00033 0.000257 0.00701 0.000661
Load
Full
100 39
Engineering Management in Production and Services
1.036191 0.596739 0.000351 0.000273 0.007449 0.000702
Volume
Load 14 • Issue 1 • 2022

Tab. 2. Comparison of results from two different objective functions for the cluster-distributed customer dataset

MINIMISING COST MINIMISING GHG EMISSIONS DISTANCE INCREASE (%)


INSTANCE
DISTANCE NV DISTANCE NV
LC101 828.94 10 989.91 12 19.42
LC102 828.94 10 828.94 10 0
LC103 827.86 10 827.86 10 0
LC104 861.95 9 903.85 9 4.86
LC105 828.94 10 828.94 10 0
LC106 828.94 10 828.94 10 0
LC107 828.94 10 828.94 10 0
LC108 827.61 10 827.61 10 0
LC109 827.82 10 827.82 10 0
LC201 591.56 3 591.56 3 0
LC202 591.56 3 591.56 3 0
LC203 591.17 3 591.17 3 0
LC204 590.60 3 590.60 3 0
LC205 588.88 3 588.88 3 0
LC206 588.29 3 588.49 3 0.04
LC207 588.29 3 588.29 3 0
LC208 588.32 3 588.32 3 0
Average 1.43

Tab. 3. Comparison of results from two different objective functions for the randomly distributed customer dataset

MINIMISING COST MINIMISING GHG EMISSIONS


INSTANCE DISTANCE INCREASE (%)
DISTANCE NV DISTANCE NV
LR101 1543.38 17 1650.80 19 6.96
LR102 1361.93 13 1555.64 17 14.22
LR103 1071.23 10 1329.99 13 24.16
LR104 1013.99 9 1080.51 10 6.56
LR105 1295.14 12 1393.35 14 7.58
LR106 1221.29 12 1293.30 12 5.90
LR107 1174.83 11 1257.08 12 7.00
LR108 1085.18 10 1204.41 11 10.99
LR109 1263.96 12 1563.00 13 23.66
LR110 1135.66 10 1224.67 12 7.84
LR111 1156.54 11 1179.63 11 2.00
LR112 1151.38 11 1159.13 11 0.67
LR201 1266.25 4 1266.57 4 0.03
LR202 1162.40 4 1316.46 4 13.25
LR203 934.53 3 1153.83 3 23.47
LR204 912.40 2 1025.59 3 12.41
LR205 1118.70 3 1248.82 4 11.63
Average 10.49

were4.set
Tab. based on
Comparison trial-and-error
of results methods.
from the best-known The algo-
solution the from
and the results minimum total
PSO for the cost as the customer
cluster-distributed objectivedataset
function
rithms were coded using the C# version of Visual (VRPPD). The PSO was then applied to design the
MINIMISING COST BEST-KNOWN SOLUTION DEVIATION
Studio 2019.
INSTANCE The comparison of the results based on optimal route Rby considering the minimum
EFERENCES OF DISTANCEGHG
DISTANCE NV DISTANCE NV
cost minimisation and GHG emissions minimisation emissions as the objective function (Green(%) VRPPD).
LC101
for the LC and LR 828.94 10
problems are presented 828.94
in Tables 2 10 Li & Lim
The results from the(2001) 0.00
two different objective functions
LC102 828.94 10 828.94 10 Li & Lim (2001) 0.00
and 3, respectively. were compared using the per cent increase of dis-
LC103 827.86 10 1035.35 9 Bent & Van Hentenryck (2003) -20.04
Table 2 presents
LC104 the results of the
861.95 9 PSO 860.01
for design- tance.
9 TheHasle
reason is that
& Kloster the travel distance is directly
(2007) 0.23
ing LC105
vehicle routes828.94
from 17 benchmark
10 instances 10
828.94 proportionalLito& the total cost. The results reveal0.00
Lim (2001) that
LC106 828.94 10 828.94
where customers are cluster-distributed. The PSO 10 Li & Lim (2001)
designing the vehicle routes by using VRPPD and 0.00
wasLC107 828.94
applied to design 10
vehicle routes 828.94
by considering 10
Green VRPPD Li & Lim (2001)
are comparable when customers 0.00
are
LC108 827.61 10 826.44 10 Li & Lim (2001) 0.14
LC109 827.82 10 1000.6 9 Bent & Van Hentenryck (2003) -17.27
LC201 591.56 3 591.56 3 Li & Lim (2001) 0.00
LC202 591.56 3 591.56 3 Li & Lim (2001) 0.00 7
LC203 591.17 3 591.17 3 Bent & Van Hentenryck (2003) 0.00
LC204 590.6 3 590.6 3 Bent & Van Hentenryck (2003) 0.00
Tab. 3. Comparison of results from two different COST
MINIMISING objective functions for the randomly distributed
MINIMISING customer dataset
GHG EMISSIONS
INSTANCE DISTANCE INCREASE (%)
Engineering ManagementDin Production
MINIMISINGand
ISTANCE COST Services
NV DISTANCE NV
MINIMISING GHG EMISSIONS
INSTANCE Volume 14 • Issue 1 •(%)
2022
LR101 1543.38
DISTANCE NV 17 1650.80
DISTANCE NV 19 DISTANCE INCREASE6.96
LR102
LR101 1361.93
1543.38 13 1555.64 17 14.22
cluster-distributed. The number of vehicles used for17 GHG emissions 1650.80 requires a greater19 number of vehicles 6.96
LR103
LR102 1071.23
1361.93 10
13 1329.99
1555.64 13
17 24.16
14.22
delivering products to customers
LR104 1013.99
using VRPPD or 9 and greater travel distances 10
1080.51
than route planning 6.56
LR103 1071.23 10 1329.99 13 24.16
Green VRPPD
LR105 is the same in most
1295.14 cases. However, 12 based on minimum
1393.35 total cost as
14 the objective func-7.58
LR104 1013.99 9 1080.51 10 6.56
the percentage
LR106
LR105
difference of1221.29
benchmark instance12
1295.14 12
tion in most cases.
1293.30
1393.35 12
14 5.90
7.58
LC101 is 19.42, which is quite1174.83
LR107
LR106 high when compared11
1221.29 12 These1257.08
results reveal that Green
1293.30 12
12 VRPPD requires 7.00
5.90
to two other
LR107benchmark instances:
LR108 1085.18
1174.83LC104 and LC206.10 11 travel distances
1257.08 that are 10.49%
1204.41 12 greater than 10.99
11 those
7.00
On average,
LR108 the Green VRPPD
LR109 requires a travel dis-12
1263.96
1085.18 10 required 1563.00
by VRPPD on average.13
1204.41 11Furthermore, 11 out
23.66
10.99
tance thatLR110
LR109is 1.43% greater 1135.66
than that of VRPPD.10
1263.96 12 of 17 cases 1224.67 12
require a greater number
1563.00 13 of vehicles 7.84to
23.66
LR111
It is suggested
LR110 that Green VRPPD 1156.54
1135.66should be used for11 1179.63
10 deliver products
1224.67 to customers.11
12 Tables 4 and 2.00
7.84 5
LR112 1151.38 11 1159.13 11
designing LR111
vehicle routes in the 1156.54
last mile of delivery11 presents 1179.63 the comparison of the 11 results from the 0.67 2.00
best-
LR201
LR112 1266.25 4 1266.57 4 0.03
when customers’ locations are1151.38
cluster-distributed. 11 known solution 1159.13 and PSO for the 11 0.67
cluster-distributed
LR202
LR201 1162.40
1266.25 4 1316.46 4 13.25
Vehicle
LR203
routes were designed 934.53
by the PSO using 34 customer1153.83 1266.57 4
dataset and randomly distributed
3
0.03
customer
23.47
LR202 1162.40 4 1316.46 4 13.25
17 benchmark
LR204 instances for problems
912.40 where custom- 2 dataset, respectively.
1025.59 It found that
3 the results12.41from
LR203 934.53 3 1153.83 3 23.47
ers are randomly
LR205
LR204
distributed. In this
1118.70
912.40
case, the location 3
2
PSO are comparable
1248.82
1025.59
to the best-known
4
3
solution 11.63
12.41
for
of customers
LR205 is randomly dispersed
1118.70 around the ser- 3 LC1, LC2, and
1248.82 RC1. However, the
Average4 results of PSO10.49
11.63for
vice area. The experimental results in Table 3 show solving RC2 problems areAverage not as good as the10.49 best-
that vehicle
Tab. 4. route
Comparison planning
of results based
from the on minimising
best-known known
solution and the results fromsolution.
PSO for the cluster-distributed customer dataset
Tab. 4. Comparison of results from the best-known solution and the results from PSO for the cluster-distributed customer dataset
MINIMISING COST BEST-KNOWN SOLUTION DEVIATION
INSTANCE MINIMISING COST BEST-KNOWN SOLUTION REFERENCES DEVIATION
OF DISTANCE
DISTANCE NV DISTANCE NV
INSTANCE REFERENCES (%)
OF DISTANCE
DISTANCE NV DISTANCE NV
LC101 828.94 10 828.94 10 Li & Lim (2001) (%) 0.00
LC101
LC102 828.94 10 828.94 10 Li & Lim (2001) 0.00
LC102
LC103 828.94
827.86 10 828.94
1035.35 10
9 Li & Hentenryck
Bent & Van Lim (2001) (2003) 0.00
-20.04
LC103
LC104 827.86
861.95 10
9 1035.35
860.01 9 BentHasle
& Van& Hentenryck (2003)
Kloster (2007) -20.04
0.23
LC104
LC105 861.95
828.94 9
10 860.01
828.94 9
10 Hasle
Li & Lim
Kloster (2007)
(2001) 0.23
0.00
LC105
LC106 828.94 10 828.94 10 Li & Lim (2001) 0.00
LC106
LC107 828.94 10 828.94 10 Li & Lim (2001) 0.00
LC107
LC108 828.94
827.61 10 828.94
826.44 10 Li & Lim (2001) 0.00
0.14
LC108
LC109 827.61
827.82 10 826.44
1000.6 10
9 Li & Hentenryck
Bent & Van Lim (2001) (2003) 0.14
-17.27
LC109
LC201 827.82
591.56 10
3 1000.6
591.56 9
3 Bent & Van
Li & Hentenryck
Lim (2001) (2003) -17.27
0.00
LC201
LC202 591.56 3 591.56 3 Li & Lim (2001) 0.00
LC202
LC203 591.56
591.17 3 591.56
591.17 3 Li & Hentenryck
Bent & Van Lim (2001) (2003) 0.00
LC203
LC204 591.17
590.6 3 591.17
590.6 3 Bent & Van Hentenryck (2003) 0.00
LC204
LC205 590.6
588.88 3 590.6
588.88 3 Bent & Van
Li & Hentenryck
Lim (2001) (2003) 0.00
LC205
LC206 588.88
588.29 3 588.88
588.49 3 Li & Lim (2001) 0.00
-0.03
LC206
LC207 588.29 3 588.49
588.29 3 Li & Lim (2001) -0.03
0.00
LC207
LC208 588.29
588.32 3 588.29
588.32 3 Li & Lim (2001) 0.00
LC208 588.32 3 588.32 3 Li & Lim (2001) 0.00

Tab. 5. Comparison of results from best-known solution and the results from PSO for the randomly distributed customer dataset

MINIMISING COST BEST-KNOWN SOLUTION DEVIATION


INSTANCE REFERENCES OF DISTANCE
DISTANCE NV DISTANCE NV
(%)
LR101 1543.38 17 1650.8 19 Li & Lim (2001) -6.51
LR102 1361.93 13 1487.57 17 Li & Lim (2001) -8.45
LR103 1071.23 10 1292.68 13 Li & Lim (2001) -17.13
LR104 1013.99 9 1013.39 9 Li & Lim (2001) 0.06
LR105 1295.14 12 1377.11 14 Li & Lim (2001) -5.95
LR106 1221.29 12 1252.62 12 Li & Lim (2001) -2.50
LR107 1174.83 11 1111.31 10 Li & Lim (2001) 5.72
LR108 1085.18 10 968.97 9 Li & Lim (2001) 11.99
LR109 1263.96 12 1208.96 11 Hasle & Kloster (2007) 4.55
LR110 1135.66 10 1159.35 10 Li & Lim (2001) -2.04
LR111 1156.54 11 1108.9 10 Li & Lim (2001) 4.30
LR112 1151.38 11 1003.77 9 Li & Lim (2001) 14.71
LR201 1266.25 4 591.56 3 Li & Lim (2001) 114.05
LR202 1162.4 4 591.56 3 Li & Lim (2001) 96.50
LR203 934.53 3 591.17 3 Hasle & Kloster (2007) 58.08
LR204 912.4 2 590.6 3 Hasle & Kloster (2007) 54.49
LR205 1118.7 3 588.88 3 Li & Lim (2001) 89.97

8
LR203 934.53 3 591.17 3 Hasle and Kloster (2007) 58.08
LR204 912.4 2 590.6 3 Hasle and Kloster (2007) 54.49
LR205 1118.7 3 588.88 3 Engineering
Li & LimManagement
(2001) 89.97Services
in Production and
Volume 14 • Issue 1 • 2022

Tab. 6. Comparison of GHG emissions from both diesel and LPG fuel tion is to design vehicle routes by minimising the
total cost. Next, then PSO was applied to design
DIFFERENCE vehicle routes by minimising GHG emissions as the
DIESEL LPG
INSTANCES OF GHG
EMISSIONS EMISSIONS
EMISSION (%)
objective function. Three main GHGs were consid-
LC101 101.43 58.55 53.61 ered in this study: carbon dioxide (CO2), methane
LC102 97.84 56.48 53.60 (CH4), and nitrous oxide (N2O). Two different types
LC103 103.4 59.69 53.60 of customer locations were tested. The first type of
LC104 93.27 53.84 53.61
LC105 69.23 39.96 53.61
customer location is cluster-distributed, where the
LC106 49.98 28.85 53.61 location of customers in the coverage service area are
LC107 181.23 104.62 53.60 clustered in villages, towns, or cities. The results
LC108 86.5 49.93 53.61 reveal that vehicle route design based on Green
LC109 125.58 72.49 53.61
LC201 119.93 69.23 53.61
VRPPD is comparable to that of VRPPD, as presented
LC202 92.72 53.52 53.61 in Table 2. Only three out of seventeen cases show
LC203 301.41 173.99 53.61 that designing vehicle routes by considering GHG
LC204 202.55 116.92 53.61 emissions requires more travel distance than by con-
LC205 117.44 67.8 53.60
sidering minimum total cost as the objective func-
LC206 162.08 93.56 53.61
LC207 187.54 108.26 53.60 tion. The results for Green VRPPD instance LC101
LC208 148.5 85.72 53.61 require 19.42 % more travel distance than the results
LR101 51.85 29.93 53.61 for VRPPD. The increased travel distance is because
LR102 105.24 60.75 53.61
the computational result from Green VRPPD requires
LR103 79.31 45.78 53.61
LR104 47.13 27.2 53.63 two more vehicles to deliver products to customers.
LR105 61.39 35.44 53.60 The average difference in travel distance is equal to
LR106 150.54 86.9 53.61 1.43 %, which does not have much effect on the total
LR107 143.72 82.96 53.61
cost. Therefore, Green VRPPD is recommended for
LR108 147.21 84.98 53.60
LR109 124.78 72.03 53.60 designing vehicle routes when the location of cus-
LR110 139.27 80.4 53.60 tomers is cluster-distributed in the coverage service
LR111 74.23 42.85 53.60 area.
LR112 114.39 66.03 53.61
The second type of customer location is randomly
LR201 300.71 173.58 53.61
LR202 97.76 56.43 53.61 distributed around the coverage service area. The
LR203 580 334.81 53.60 results presented in Table 3 show that Green VRPPD
LR204 181.81 104.95 53.61 requires more travel distance compared to VRPPD.
LR205 473.12 273.11 53.61
The average travel distance difference is equal to 10.43
LR206 260.5 150.37 53.61
LR207 342.92 197.95 53.61 %. Green VRPPD for three instances, LR103, LR109,
LR208 536.67 309.8 53.60 and LR203, require 20 % more travel distance com-
LR209 226.6 130.81 53.60 pared to VRPPD. When considering GHG emissions,
LR210 249.36 143.94 53.61
most of the instances require more vehicles to service
LR211 312.26 180.25 53.61
Average 53.61 customers than when considering the total cost as the
objective function. The implication is that environ-
Considering greenhouse gases emissions, the mental concerns would cost 10.49 % more for trans-
results indicate that the use of liquefied petroleum gas portation. Fuel selection is another option that
(LPG) fuel is more environmentally friendly than the companies can use to provide more environmentally
use of diesel fuel. Table 6 presents the comparison of friendly transportation. The computational results
GHG emissions from both diesel and LPG fuels. The indicate that the use of LPG fuel emits 53.61 % less
data show that LPG fuel emits 53.61 % less GHG GHGs than does diesel fuel.
emissions than does diesel fuel. The results reveal that Green VRPPD is suitable
for the situation that the customer location is clus-
tered. Generally, cluster-distributed customer loca-
4. Discussion of the results tion can save both transportation costs and time for
the delivery of the product to the customers. Further,
Particle swarm optimisation (PSO) was applied this would lead to the reduction of GHG from trans-
to determine optimal vehicle routes based on two portation. It also causes less local air pollution to use
different objective functions. The first objective func- LPG as the fuel rather than using petrol gasoline or

9
Engineering Management in Production and Services
Volume 14 • Issue 1 • 2022

diesel. However, Green VRPPD is suitable for design- duction Management - EPPM2021 was financed in
ing the route of vehicles in some cases when the cus- the framework of the contract no. DNK/
tomer location is randomly distributed around the SN/465770/2020 by the Ministry of Science and
coverage service area because it would require more Higher Education within the "Excellent Science"
travel distance to be travelled. programme. This work was financially supported by
the Burapha University International College.
Conclusions
The purpose of this study is to design a green last-
mile route that supports efficient e-commerce distribu- Literature
tion. A metaheuristics approach called particle swarm
optimisation (PSO) was applied to design the route for
Al-Tit, A. A. (2020). E-commerce drivers and barriers
last-mile delivery. This study used 34 benchmark and their impact on e-customer loyalty in small
instances from Li and Lim (2001) to test the perfor- and medium-sized enterprises (SMES). Business:
mance of the search technique. First, the PSO was Theory and Practice, 21(1), 146-157. doi: 10.3846/
btp.2020.11612
developed to solve VRPPD for transportation cost
Bansal, S., & Wadhawan, S. (2021). A hybrid of sine cosine
minimisation. Then, the PSO was applied to design
and particle swarm optimization (HSPS) for solv-
vehicle routes based on GHG emissions minimisation ing heterogeneous fixed fleet vehicle routing prob-
(Green VRPPD). lem. International Journal of Applied Metaheuristic
Two different types of customer location distribu- Computing (IJAMC), 12(1), 41-65.
tions were evaluated. The LC datasets contain clusters Belmecheri, F., Prins, C., Yalaoui, F., & Amodeo, L. (2013).
Particle swarm optimization algorithm for a vehicle
of customers, whereas the LR datasets contain custom- routing problem with heterogeneous fleet, mixed
ers that are randomly distributed. The results of this backhauls, and time windows. Journal of Intelligent
study reveal that the use of Green VRPPD is very suit- Manufacturing, 24(4), 775-789.
able for LC problems because the total distances and Bent, R., & Van Hentenryck, P. (2006). A two-stage hybrid
the number of vehicles used are the same whether algorithm for pickup and delivery vehicle routing
problems with time windows. Computers & Opera-
considering GHG emissions or minimum total cost as tions Research, 33(4), 875-893.
the objective function in most cases. For RC problems, Bruglieri, M., Mancini, S., & Pisacane, O. (2019). The
the decision of vehicle routing should depend on the green vehicle routing problem with capacitated al-
logistics provider’s situation because a design based on ternative fuel stations. Computers & Operations Re-
search, 112, 104759.
Green VRPPD is more expensive and requires more
Chen, M. C., Hsiao, Y. H., Reddy, R. H., & Tiwari, M. K.
vehicles than a design based on VRPPD in most cases. (2016). The self-learning particle swarm optimiza-
This study also looks at the use of two different fuel tion approach for routing pickup and delivery of
types. The study compares GHG emissions from LPG multiple products with material handling in multiple
fuel and diesel fuel. It was found that using LPG cross-docks. Transportation Research Part E: Logis-
tics and Transportation Review, 91, 208-226.
appears to be more environmentally friendly than
Chen, N., & Yang, Y. (2021). The impact of customer expe-
using diesel fuel. rience on consumer purchase intention in cross-bor-
The limitations of this work are the problem der E-commerce – taking network structural embed-
instances. It would be more realistic to use real data dedness as mediator variable. Journal of Retailing
and Consumer Services, 59, 102344.
from third-party logistics providers. However, it is
Créput, J. C., Koukam, A., Kozlak, J., & Lukasik, J. (2004).
somewhat difficult to get such data because most
An evolutionary approach to pickup and delivery
logistics companies consider this data to be confiden- problem with time windows. In International Con-
tial information. Suggested future research would be to ference on Computational Science (pp. 1102-1108).
improve the metaheuristics technique used to design Springer, Berlin, Heidelberg.
routes for delivering products to customers. Fan, H., Zhang, Y., Tian, P., Lv, Y., & Fan, H. (2021). Time-
dependent multi-depot green vehicle routing prob-
lem with time windows considering temporal-spatial
Acknowledgements distance. Computers & Operations Research,  129,
105211.
Faugère, L., & Montreuil, B. (2020). Smart locker bank
The publication of the article for 11th Interna- design optimization for urban omnichannel logistics:
Assessing monolithic vs. modular configurations.
tional Conference on Engineering, Project, and Pro- Computers & Industrial Engineering, 139, 105544.

10
Engineering Management in Production and Services
Volume 14 • Issue 1 • 2022

Fedorko, R., Fedorko, I., Riana, I. G., Rigelský, M., research in Polish Cities. Transportation Research
Oleárová, M., & Obšatníková, K. (2018). The impact Procedia, 16, 272-287.
of selected elements of e-commerce to e-shop recom- Li, H., & Lim, A. (2003). A metaheuristic for the pickup
mendation. Polish Journal of Management Studies, and delivery problem with time windows. Inter-
18(1), 107-120. doi: 10.17512/pjms.2018.18.1.09 national Journal on Artificial Intelligence Tools,
Florek-Paszkowska, A., Ujwary-Gil, A., & Godlewska- 12(02), 173-186.
Dzioboń, B. (2021). Business innovation and critical Liu, X., Zhang, K., Chen, B., Zhou, J., & Miao, L. (2018).
success factors in the era of digital transformation Analysis of logistics service supply chain for the
and turbulent times. Journal of Entrepreneurship, One Belt and One Road initiative of China. Trans-
Management, and Innovation, 17(4), 7-28. doi: portation Research Part E: Logistics and Transpor-
10.7341/20211741 tation Review, 117, 23-39.
Foroutan, R. A., Rezaeian, J., & Mahdavi, I. (2020). Green Mehlawat, M. K., Gupta, P., Khaitan, A., & Pedrycz, W.
vehicle routing and scheduling problem with hetero- (2019). A hybrid intelligent approach to integrated
geneous fleet including reverse logistics in the form fuzzy multiple depot capacitated green vehicle rout-
of collecting returned goods. Applied Soft Comput- ing problem with split delivery and vehicle selection.
ing, 94, 106462. IEEE Transactions on Fuzzy Systems, 28(6), 1155-
Fu, H., Manogaran, G., Wu, K., Cao, M., Jiang, S., & 1166.
Yang, A. (2020). Intelligent decision-making of on- Mutinda Kitukutha, N., Vasa, L., & Oláh, J. (2021). The Im-
line shopping behavior based on internet of things. pact of COVID-19 on the economy and sustainable
International Journal of Information Management, e-commerce. Forum Scientiae Oeconomia, 9(2), 47-
50, 515-525. 72. doi: 10.23762/FSO_VOL9_NO2_3
Goksal, F. P., Karaoglan, I., & Altiparmak, F. (2013). A hy- Norouzi, N., Sadegh-Amalnick, M., & Tavakkoli-Moghad-
brid discrete particle swarm optimization for vehicle dam, R. (2017). Modified particle swarm optimiza-
routing problem with simultaneous pickup and de- tion in a time-dependent vehicle routing problem:
livery. Computers & Industrial Engineering, 65(1), minimizing fuel consumption. Optimization Letters,
39-53. 11, 121-134.
Gregory, G. D., Ngo, L. V., & Karavdic, M. (2019). De- Qin, X., Liu, Z., & Tian, L. (2021). The optimal combina-
veloping e-commerce marketing capabilities and ef- tion between selling mode and logistics service strat-
ficiencies for enhanced performance in business-to- egy in an e-commerce market. European Journal of
business export ventures. Industrial Marketing, 78, Operational Research, 289(2), 639-651.
146-157.
Ready, C. (2013). Environmental reporting guidelines:
Gulc, A. (2021). Multi-stakeholder perspective of courier Including mandatory greenhouse gas emissions re-
service quality in B2C e-commerce. PLoS ONE, porting guidance. Retrieved from https://www.gov.
16(5), 1-18. doi: 10.1371/journal.pone.0251728 uk/government/publications/environmental-report-
Gupta, P., Govindan, K., Mehlawat, M. K., & Khaitan, A. ing-guidelines-including-mandatory-greenhouse-
(2021). Multiobjective capacitated green vehicle gas-emissions-reporting-guidance
routing problem with fuzzy time-distances and de- Rita, P., Oliveira, T., & Farisa, A. (2019). The impact of
mands split into bags. International Journal of Pro- e-service quality and customer satisfaction on cus-
duction Research, 1-17. tomer behavior in online shopping. Heliyon, 5(10),
Harbaoui Dridi, I., Ben Alaïa, E., Borne, P., & Bouchriha, e02690.
H. (2020). Optimisation of the multi-depots pick-up Shi, Y., & Eberhart, R. (1998). A modified particle swarm
and delivery problems with time windows and multi- optimizer. In IEEE international conference on evo-
vehicles using PSO algorithm. International Journal lutionary computation proceedings. IEEE world
of Production Research, 58(14), 4201-4214. congress on computational intelligence (Cat. No.
Hasle, G., & Kloster, O. (2007). Industrial vehicle routing. 98TH8360) (pp. 69-73). IEEE.
In G. Hasle, K.-A. Lie, E. Quak (Eds.), Geometric Sruthi, A., Anbuudayasankar, S. P., & Jeyakumar, G.
modelling, Numerical Simulation, and Optimization (2019). Energy-efficient green vehicle routing prob-
(pp.397-435). Berlin, Heidelberg: Springer. lem. International Journal of Information Systems
Jacobs, K., Warner, S., Rietra, M., Mazza, L., Buvat, J., and Supply Chain Management (IJISSCM), 12(4),
Khadikar, A., Cherian, S., & Khemka, Y. (2019). 27-41.
The last-mile delivery challenge: Giving retail and Tsang, Y. P., Wu, C. H., Lam, H. Y., Choy, K. L., & Ho,
consumer product customers a superior delivery ex- G. T. S. (2021). Integrating Internet of Things and
perience without impacting profitability. Retrieved multi-temperature delivery planning for perishable
from https://www.capgemini.com/wp-content/up- food E-commerce logistics: a model and applica-
loads/2019/01/Report-Digital-%E2%80%93-Last- tion. International Journal of Production Research,
Mile-Delivery-Challenge1.pdf 59(5), 1534-1556.
Lagos, C., Guerrero, G., Cabrera, E., Moltedo, A., Johnson, Úbeda, S., Faulin, J., Serrano, A., & Arcelus, F. J. (2014).
F., & Paredes, F. (2018). An improved particle swarm Solving the green capacitated vehicle routing prob-
optimization algorithm for the VRP with simultane- lem using a tabu search algorithm. Lecture Notes in
ous pickup and delivery and time windows. IEEE Management Science, 6(1), 141-149.
Latin America Transactions, 16(6), 1732-1740.
United States. Environmental Protection Agency. Office
Lemke, J., Iwan, S., & Korczak, J. (2016). Usability of the of Policy. (1999). Inventory of US Greenhouse Gas
parcel lockers from the customer perspective – the Emissions and Sinks: 1990-1997. The Agency.

11
Engineering Management in Production and Services
Volume 14 • Issue 1 • 2022

Vakulenko, Y., Shams, P., Hellström, D., & Hjort, K.


(2019). Service innovation in e-commerce last mile
delivery: Mapping the e-customer journey. Journal
of Business Research, 101, 461-468.
van Lopik, K., Schnieder, M., Sharpe, R., Sinclair, M.,
Hinde, C., Conway, P., West, A., & Maguire, M.
(2020). Comparison of in-sight and handheld navi-
gation devices toward supporting industry 4.0 sup-
ply chains: First and last mile deliveries at the human
level. Applied Ergonomics, 82, 102928.
Xu, X., Wang, C., Li, J., & Shi, C. (2019). Green Transpor-
tation and Information Uncertainty in Gasoline Dis-
tribution: Evidence from China. Emerging Markets
Finance and Trade, 57(11), 1-19.
Yu, Y., Wang, S., Wang, J., & Huang, M. (2019). A branch-
and-price algorithm for the heterogeneous fleet
green vehicle routing problem with time windows.
Transportation Research Part B: Methodological,
122, 511-527.
Yu, Y., Yu, C., Xu, G., Zhong, R. Y., & Huang, G. Q. (2020).
An operation synchronization model for distribution
center in E-commerce logistics service. Advanced
Engineering Informatics, 43, 101014.
Yuen, K. F., Wang, X., Ma, F., & Wong, Y. D. (2019). The
determinants of customers’ intention to use smart
lockers for last-mile deliveries. Journal of Retailing
and Consumer Services, 49, 316-326.
Zhang, X., Zhou, G., Cao, J., & Wu, A. (2020). Evolving
strategies of e-commerce and express delivery enter-
prises with public supervision. Research in Trans-
portation Economics, 80, 100810.
Zhou, M., Zhao, L., Kong, N., Campy, K. S., Xu, G., Zhu,
G., Cao, X., & Wang, S. (2020). Understanding con-
sumers’ behavior to adopt self-service parcel ser-
vices for last-mile delivery. Journal of Retailing and
Consumer Services, 52, 101911.
Zhu, L., & Hu, D. (2019). Study on the vehicle routing
problem considering congestion and emission fac-
tors. International Journal of Production Research,
57(19), 6115-6129.

12

You might also like