Matheuristic Algorithms
for Green Vehicle Routing Problems
Thesis Project
Alejandro Fernández Gil
Director: Dr. Carlos Castro
Co-Director: Dr. Eduardo Lalla-Ruiz
Departamento de Informática
Universidad Técnica Federico Santa María
July 10, 2020, Valparaíso, Chile
1/72
Contents
Introduction
Problem Context
Theoretical Framework
Research Proposal
Motivation
Hypothesis
Research Goals
Expected Results
Previous Work
Work Schedule
Questions
2/72
Problem Context
3/72
Freight Transportation
• People demands products • Logistics systems
- Food, clothing, everything - Connect suppliers to customers
else (nice to have) cost-effectively, using freight
transportation networks and
• Supply chain networks logistics facilities
- Procure/produce products
- Get them to people 4/72
Freight Transportation
• All forms of transportation contribute to greenhouse gases and
congestion, trucking has the biggest impact.
• The burning of fossil fuels causes approximately 80% of environmental
pollution worldwide, and around 60% of this fuel is consumed in the
freight transportation sector 1.
• In Chile, road freight transport produces the highest emissions 2.
1An approach for analysing transportation costs and a case study [SYU+09]
2Second Biennial Update Report on Climate Change. MMA-Chile, 2016
5/72
Freight Transportation
In the period from 2002 to 2035, experts estimate that transportation volumes
will grow in the following ways 3:
• This level of growth places stress and increasing congestion on a network
that’s already in heavy use.
To move toward sustainability, the transportation industry needs sustainable
practices:
3United States Environmental Protection Agency | US EPA
6/72
Freight Transportation
In the period from 2002 to 2035, experts estimate that transportation volumes
will grow in the following ways 4:
• This level of growth places stress and increasing congestion on a network
that’s already in heavy use.
To move toward sustainability, the transportation industry needs sustainable
practices:
4United States Environmental Protection Agency | US EPA
7/72
Theoretical Framework
8/72
Related Optimization Routing Problems
Traveling Salesman Problem [Men32]
• Shortest path, visit each city exactly once and return to the origin city.
• Minimizing the total cost.
9/72
Related Optimization Routing Problems
Vehicle Routing Problem [DR59] [CW64]
• Find effective routes for multiple vehicles visiting a set of locations.
• multi-depot [CE69], time-dependent [VRP66], time-windows [Rus77]
10/72
Related Optimization Routing Problems
Green Vehicle Routing Problems [KKY07] [EMH12] [LCH+14]
• Harmonizing the environmental and economic costs by implementing
effective routes to meet the environmental concerns and financial indexes.
11/72
Green VRPs
Green Logistics
Describes all attempts to measure and minimize the ecological impact of
logistics activities.
Green VRPs Area [LCH+14]
1. Energy consumption
(fuel consumption)
2. Pollution emissions
(reduction of carbon
emissions)
3. Recycling and reverse
logistics
12/72
Green VRPs
Green Logistics
Describes all attempts to measure and minimize the ecological impact of
logistics activities.
Green VRPs Area [LCH+14]
1. Energy consumption
(fuel consumption)
2. Pollution emissions
(reduction of carbon
emissions)
3. Recycling and reverse
logistics
Sensitive margin between
G-VRP and PRP:
energy consumption, then
reduce emissions
13/72
G-VRP related problems
Energy Minimizing VRP (Kara et al. [KKY07])
• Minimize emissions, by minimizing energy consumption.
14/72
G-VRP related problems
Cumulative VRP (Kara et al. [KKY08])
• Minimize the cumulative cost (distance× flow on each arc).
15/72
G-VRP related problems
Electric VRP (Gonçalves et al. [GCRBP11])
• Fleet of EVs, refueling is done at the location closest to where
the need for fuel occurred.
16/72
G-VRP related problems
Fuel Consumption VRP (Xiao et al. [XZKX12])
• Model based on minimizing fuel consumption.
17/72
G-VRP related problems
Green-VRP (Erdoğan and Miller-Hooks [EMH12])
• AFVs can refuel on the route in the AFSs. Minimization of the
distance, eliminating the risk of running out of fuel.
18/72
G-VRP related problems
Electric VRPTW (Schneider et al. [SSG14])
• E-VRP extension, considers time windows constraints, for the
delivery of goods to customers.
19/72
G-VRP related problems
Green-VRP with Multiple Trips (Ayadi et al. [AEBE14])
• Each vehicle can perform multiple routes during a working day.
20/72
G-VRP related problems
Green VRP and Scheduling Problem (Xiao and Konak. [XK16])
• Minimize emissions through better scheduling of
delivery/collections by a fleet of vehicles.
21/72
G-VRP related problems
Green Time-Dependent VRPTW (Kazemian et al. [KRFA18])
• Model based on time dependency, TW, scheduling and speed
limits.
22/72
G-VRP related problems
Multi-Depot G-VRP (Li et al. [LSZ19])
• Multiple deposits. Minimization of costs, maximization of
profits, minimization of travel time and emissions.
23/72
G-VRP related problems
CumVRP with TW (Fernández et al. [FGLRC20], accepted)
• Presents CumVRP with TW. Propose an analysis of trade-off
between environmental cost and customer’s dissatisfaction.
24/72
Factors in energy consumption
Factors [DBL14]
Vehicle Environment Traffic (travel)
related related related
Vehicle Ambient Speed
curb weight temperature Acceleration /
decceleration
Engine Wind Congestion
temperature conditions
25/72
Fuel tipe /
composition
Emission estimate
Fuel-based approach [PRO08]
1. If the fuel consumption is known:
• CO2emission = fuel_consumption ×
heating _value ×
emission_factor
2. If fuel consumption is not known, it can be calculated by
distance:
• fuel_consumption = distance × fuel_economy _factor
Distance-based approach [DP10]
Efficiency factor data are not available:
• CO2emissions = distance × payload × emission_factor
26/72
Current study areas
• Consider constraints on availability and fuel capacity in the
AFSs [LCH+14].
• Explore the trade-off between a greater travel distance joined to the
weight of the vehicle load on each route (economic costs and the
environmental impact) [LCH+14].
• Explore the trade-off described in the previous point taking into
account the use of time windows that may incur in reducing energy
consumption [LCH+14].
• Integrate Green Logistics problems to Vehicle Routing and
Scheduling Problem (VRSP) models [SE07].
27/72
Resolution Algorithms
Exact Constraint Programming [BB19]
algorithms Integer Linear Programming [XK16], [XK17]
Branch-and-Price [DDW17]
Column Generation [GS15]
Constructive C&W Savings [CCEGP17]
algorithms GRASP [NGABMSCF18], [EEBM18]
Algorithms Trajectory
algorithms
Improvement TS [PSS17], [Goe19]
algorithms SA [CGP15], [FZJ17]
(Local Search) GRASP [NGABMSCF18], [EEBM18]
Approximate
algorithms
Constructive ACO [LSZ19], [ZZGA19]
algorithms
Population GA [Hsu16], [XK17]
algorithms PSO [NSATM17], [LLT19]
Improvement
algorithms
28/72
Resolution Algorithms
Matheuristic algorithms
Heuristics that incorporate phases where MILP or, more generally,
mathematical programming (MP) models are solved [AS14].
Classification [AS14]
1. Decomposition
approaches
2. Branching approaches
3. Improvement heuristics
approaches
29/72
Resolution Algorithms
Decomposition-based matheuristic algorithms [AS14]
• Cluster first-route second: divide the two main decisions that
characterize routing problems: assignment of customers to
vehicles and the sequencing of the customers visited by each
route.
• Two-phase: decompose the problem in two phases and solve
them separately. There are many different ways of
decomposing problems.
• Partial optimization: one or more MILP models are used to
handle a part of the problem.
• Rolling horizon: decisions must be taken in a short time period
and solve the corresponding subproblem.
30/72
Resolution Algorithms
Matheuristic algorithms in literature
• In [FMJL17], a matheuristic algorithm based of two-phase is
developed, ILS as first stage and Benders decomposition as second
stage) to solve an extension of E-VRP focused on capacitated
charging stations.
• In [KSVdAFC15], proposed a matheuristic approach combining a
local search method with MILP for the PRP.
• In [MGMV16], develop a matheuristic based two-phase for G-VRP.
The first phase applies randomized heuristics, and the second phase
is based on the solution of a SPF.
• In [LRV19], proposed a POPMUSIC matheuristic for solving the
MDCCVRP. First decomposing a large problem into parts, and
subsequently, group some parts to build subproblems. The
subproblems are solved with an exact approach.
31/72
Research Proposal
32/72
Motivation
Problem (Bektaş et al. [BEPP19])
• High levels of environmental pollution.
• The freight tranportation industry is one of the most sector
that cause environmental pollution.
• The energy used by the global transport sector has been fossil
fuel, the demand for its is expected to increase by 30%-82%
between 2010 and 2050 over the 2010 levels.
Resolution method (Archetti & Speranza [AS14])
• Robustness and quickness provided by matheuristics.
• The decomposition-based matheuristic approaches are
particularly suitable for the resolution of complex and
integrated problems.
33/72
Hypothesis
1. There are several investigations in G-VRP caused by the
growing concern for the protection of the environment.
Several aspects have not been considered in the literature:
• Availability and fuel capacity in the AFSs.
• Deposit relocation or use of alternative facilities.
• Time windows and AFVs in green routing issues related to
cumulative cost.
• The trade-off between energy consumption and customer
satisfaction.
34/72
Hypothesis
2. The problems related to G-VRP have been solved in the
literature using exact and metaheuristic algorithms.
- Exact: provide robustness for small-scale scenarios, but in
large-scale scenarios these will require too much memory from
the computer, and also too much execution time,
- Metaheuristics: provide approximate solutions in relatively
small times.
In this class of problems:
• It is not only important to provide quick solutions, but also to
provide a certain level of robustness, aspects that can be
obtained by using matheuristic algorithms, specifically those
based on decomposition and classified as cluster first-route
second and partial optimization.
Matheuristic algorithms may be able to provide quick and ro-
bust solutions to emerging problems related to G-VRP.
35/72
Thesis proposal
36/72
Thesis proposal
37/72
Thesis proposal
38/72
Thesis proposal
39/72
Thesis proposal
40/72
Research Goals
Main goal
• Investigate and study the related problems to the G-VRP. Design,
develop and analyze matheuristic integration approaches, to exploit
the advantages of exact and metaheuristic algorithms for solving
these problems.
Specific goals
• Study and formalize G-VRP problems that take into account
aspects not considered in the literature.
• Define a framework on which decomposition-based matheuristic
approaches will be designed and implemented for solving problems
related to the G-VRP, considering the aspects previously identified.
• Propose benchmarks that contain the aspects previously identified
and that serve as a reference for the analysis of the algorithms in
G-VRPs.
• Evaluate the performance of the proposed matheuristic integration
approaches for solving the G-VRPs studied.
41/72
Results
Expected results
• Define a framework that incorporates decomposition matheuristic
integration approaches for problems related to G-VRP, that
contributes to the Green VRPs research area and to Computational
Intelligence in Logistics methods.
• It is expected to communicate these results through publications in
specialized conferences:
• ICCL, LION, CAiSE, ICTAI, EUROCAST, Matheuristics, HM,
OPTIMA.
and also, in an indexed ISI WoS journal:
• ITOR, ANOR, EJOR, CAIE, ESWA, JAT.
42/72
Validation
Validation
• The results will be evaluated empirically, through experimental
tests on the results obtained from the proposed
implementations.
• Set of instances proposed in this investigation and also the
well-known state of art benchmarks for vehicle routing
problems, such as PRPLib, Solomon instances, Augerat
instances, others.
• Results will be compared using statistical tests (e.g., Wilcoxon
test, Friedman test).
43/72
Resources
Available
• ANID National Doctorate grant, until February, 2021.
• Advanced Laboratory of Informatics Doctorate of the
Department of Informatics (DII-Lab).
• Access to the Informatic Data Center to run the algorithms
and all types of experiments.
• Access to scientific journals subscribed by UTFSM.
Request
• Registration fees, transportation and accommodation for
conferences.
• Support for doctoral internship.
44/72
Previous Work
Articles
• 11th International Conference on Computational Logistics
(ICCL 2020), to be held from 27 to 30 September 2020, in
Enschede, The Netherlands. “Cumulative VRP with Time
Windows: a trade-off analysis".
• III International Conference on Agro BigData and Decision
Support Systems in Agriculture (BigDSSAgro 2019), was held
from 25 to 27 September 2019, in Valparaíso, Chile. “A Mixed
Integer Linear Programming approach for the 2D Strip
Packing Problem with different size options for plots of land in
Smart Floating Farms".
45/72
Previous Work
Projects
• Programa de Incentivo a la Iniciación Científica (PIIC) 2020
DGIIP-UTFSM. “Hybrid algorithms for the green vehicle
routing problems”.
• Proyecto de Línea de Investigación Regular 2020
DGIIE-UTFSM. Director: Carlos Castro.
46/72
Collaboration
• Department of Industrial Engineering and Business Information
Systems (IEBIS) at the University of Twente (Enschede, The
Netherlands). Contact: Eduardo Lalla-Ruíz.
• Most entrepreneurial university in the Netherlands.
• Part of the Technical Federation of Dutch universities of
technology and partner of European Consortium of Innovative
Universities.
• Best Industrial Engineering and Management programs in the
Netherlands (Elsevier ranking).
47/72
Work Schedule
48/72
Work Schedule I
Stage 1
1. Bibliographic study referring to problems related to the G-VRP to
identify aspects not considered, as well as the exact algorithms,
metaheuristics and matheuristics that exist to solve these problems.
[completed]
2. Write a research project to be presented to the Scientific Initiation
Incentives Program as part of item 1. [completed]
3. Design and implement a decomposition matheuristic integration
approach to solve an extension of G-VRP, as an experiment.
[completed]
4. Write a scientific article to be sent to an international conference
with scientific rigor. [completed]
49/72
Work Schedule II
Stage 2
5. Design a framework to incorporate decomposition matheuristic
integration approaches. [2 months]
6. Develop decomposition matheuristic integration approaches on the
framework designed in the previous point. [6 months]
7. Propose benchmarks that contain the aspects previously identified.
[2 months]
8. Evaluate the performance of the proposed matheuristic integration
approaches using the previous point and benchmarks known in the
literature. [6 months]
9. Write a scientific article to be sent to a journal on points 5-8.
[3 months]
10. Analyze the feasibility of applying matheuristic integration
approaches in real environments. [4 months]
50/72