An Open Source Spreadsheet Solver for Vehicle Routing

Problems

Güneş Erdoğan
School of Management, University of Bath, BA2 7AY, Bath, UK

Abstract

The Vehicle Routing Problem (VRP) is one of the most frequently encountered
optimization problems in logistics, which aims to minimize the cost of trans-
portation operations by a fleet of vehicles operating out of a base. This paper
introduces VRP Spreadsheet Solver, an open source Excel based tool for solving
many variants of the Vehicle Routing Problem (VRP). Case studies of two real-
world applications of the solver from the healthcare and tourism sectors that
demonstrate its use are presented. The solution algorithm for the solver, and
computational results on benchmark instances from the literature are provided.
The solver is found to be capable of solving Capacitated VRP and Distance-
Constrained VRP instances with up to 200 customers within 1 hour of CPU
time.
Keywords: Vehicle Routing Problem, Metaheuristics, Spreadsheets, Open
Source Software

1. Introduction

The Vehicle Routing Problem (VRP) is one of the most frequently encoun-
tered optimization problems in logistics, which aims to minimize the cost of
transportation operations by a fleet of vehicles operating out of a base called
depot. It arises in many industries and contexts at tactical and operational
levels. The VRP has been introduced more than 50 years ago by Danztig [8],

Email address: G.Erdogan@bath.ac.uk (Güneş Erdoğan)

Preprint submitted to Computers & Operations Research February 16, 2017

and many variants of the VRP that incorporate additional features such as time
windows (intervals in which the customers may be visited) and fleet composition
[22] have been studied. Despite its operational nature, the VRP is considered
to be in the academic domain of Operations Research rather than Operations
Management. This is due to the inherent difficulty of solving a VRP, not only
due to the complexity of the associated solution algorithms but also practical
considerations regarding the implementation of the solution.
From a practitioner’s perspective, there exist a number of barriers to develop
an in-house VRP solver. Developing a solution algorithm for VRP is a daunting
task, and even if an open-source academic code is to be used as the solution
algorithm, most academics develop algorithms in C++ and the resulting codes
are not designed for the faint-hearted. The travel distance and duration data
have to be repeatedly retrieved from a GIS, due to their dynamic nature, which
introduces either a recurring cost of acquisition or the requirement for in-house
specialist knowledge. It is not straightforward to manually compute the existing
cost of the vehicle routes, much less so to visualize and compare the existing
and optimized solutions, which is important to demonstrate the benefit of an
optimization tool.
Although there exist many commercial software packages to solve VRPs,
any package must be integrated with the existing software infrastructure of the
company, and needs to be learned by the planning managers. Most commercial
VRP software packages have a black-box component, the algorithm determining
the vehicle routes, since the developers will want to protect their intellectual
property. Finally, every real-world application of the VRP has specific needs
to which the software should be custom-tailored, which requires a constant link
with the company that developed it. In case the company ceases to exist, the
software faces the risk to become obsolete in a few years.
In this study, we introduce VRP Spreadsheet Solver that overcomes the prob-
lems stated above through the familiarity of its interface, ease of use, flexibility,
and accessibility. Microsoft Excel is arguably the standard software for small to
medium scale quantitative analysis for businesses, and is being used in almost

2
every corner of the world, in both academia and industry alike [19]. Many soft-
ware packages have built-in functionality to exchange information with Excel,
which eases the integration of the solver. The code for the solver, developed
using Visual Basic for Applications (VBA), is open-source and can be under-
stood and modified by medium-level programmers. VRP Spreadsheet Solver
has built-in functions to query a GIS web service, from which the distances,
driving times, and maps can be retrieved. The solver is available for download
on an academic website at no cost [12], and has been downloaded over 2000
times.
VRP Spreadsheet Solver has been used in practice by multiple organizations
in diverse sectors and countries. The organizations that provided feedback in-
clude two US companies in the oil industry, an Argentinian company in the
agriculture industry, a Finnish company in the tourism sector, and two chains
of chilled food delivery in Taiwan and Turkey, all of which report significant
savings. We believe that VRP Spreadsheet Solver has the potential to be used
throughout the world and achieve savings for many Small and Medium-sized En-
terprises (SMEs), and consequently reduce CO2 emissions. Furthermore, new
VRP variants have emerged through our interaction with the users that are
relevant to other sectors, and they are our contribution to the literature.
The rest of this paper is organized as follows. In Section 2, we provide a
brief list of applications of the VRP. In Section 3, we present two case studies of
the application of VRP Spreadsheet Solver, and a resulting new VRP variant.
We provide a brief overview of how to use VRP Spreadsheet Solver in Section 4.
In Section 5, we present a unified formulation that encompasses all variants of
the VRP that VRP Spreadsheet Solver can handle, a metaheuristic optimiza-
tion algorithm that VRP Spreadsheet Solver utilizes, and the results of VRP
Spreadsheet Solver on a number of benchmark instances. Finally, in Section 6,
we give our concluding remarks.

3
2. Applications of the VRP

With the progress of computer hardware and optimization software, better

algorithms and implementations for the VRP emerge every year. The compu-
tational reach of exact algorithms for the VRP is limited to 200 customers for
the most studied basic variants of the VRP, e.g. the column-and-cut generation
algorithm of [2], and decreases significantly for more realistic variants that in-
clude features such as a heterogeneous fleet or distance constraints. On the other
hand, state-of-the-art metaheuristic algorithms e.g. Adaptive Large Neighbor-
hood Search [29], Iterated Local Search [31], and Unified Hybrid Genetic Search
[34] can handle much larger instances and detailed operational constraints but
cannot offer a mathematical guarantee of performance.
As stated in the introduction, many commercial and free solvers exist for
the VRP. A recent survey [28], based on the answers to a questionnaire by 15
software vendors, have provided a number of characteristics of available VRP
software packages. More recent surveys [4, 35] list a number of commercial and
free VRP software packages, the latter including VRP Spreadsheet Solver, and
provide features required of VRP software packages. We refer the interested
reader to the comprehensive book of Toth and Vigo [32] for critical reviews of
many variants of the VRP and the associated solution algorithms.
In the rest of this section, we provide a brief list of applications of the VRP.
The list is by no means complete, but is provided to give an impression of the
generality of VRP and the diversity of the industries and contexts it arises in.
The most straightforward applications of the VRP are found in the logistics
sector. Companies in the small package shipping industry, for example, aim to
minimize the routing cost while keeping the routes of the drivers as consistent
as possible [16]. A decision support tool utilized by Toyota for selecting third
party logistics service providers based on optimized vehicle routes is presented
in [30]. The problem of a large Benelux logistics service provider that aims to
minimize the total transportation cost in a multi-depot system, which consists of
the speed-related, distance-related, and vehicle-related costs of transportation

4
is analyzed in [9].
Some examples of the VRPs arising in urban transportation are the efficient
routing of the school buses [3], and the design of tourist tours for visiting multiple
points of interest in a city [15]. The joint problem of bin allocation and vehicle
routing to optimize solid waste collection is studied in [18]. There is also a
growing body of literature on optimizing the rebalancing operations for shared
bicycle systems that aim to minimize the total cost and maximize the user
satisfaction [14].
The importance of the use of VRP models in humanitarian logistics have
been underlined in the survey by Van Wassenhove [33]. Optimizing post-disaster
relief operations by minimizing estimated total travel time of vehicles is studied
in [27]. The problem of finding the optimal routes for teams that survey a
disaster area to assess damage and relief needs is analyzed in [21]. Planning of
fuel distribution operations in the case of a domestic disaster is studied in [26],
where the authors use VRP Spreadsheet Solver to optimize the routes.
With the increasing emphasis on climate change and environmental concerns,
a venue of research within the VRP has emerged in the past decade, named
as Green Vehicle Routing Problems (G-VRP). The objective function of these
problems focus on minimizing CO2 emissions, noise pollution, and accidents, as
stated in the recent survey on G-VRP [10]. The problem of minimizing the risk
of running out of fuel, which is prominent in the route planning for Alternative
Fuel Vehicles, is studied by [13]. The problem of determining the optimal size
and mix of a vehicle fleet of electric vehicles is analyzed by [20].
Applications of the VRP are not limited to companies with a sole focus on
logistics. Indeed, variants of VRP may arise in any context where a pickup or
delivery service is performed. Specifically, examples from the healthcare sector
include routing of nurses for home health care [24] and the transportation of
blood donations to storage centers [36, 7], and the delivery services for biological
samples collected from patients to testing laboratories [1].

5
3. Case studies

VRP Spreadsheet Solver can solve more than 64 variants of the VRP, based
on features related to selective visits to customers, simultaneous pickups and
deliveries, time windows, fleet composition, distance constraint, and the final
destination of the vehicles. Some of these variants are relevant in practice but
have not been formally studied. VRP Spreadsheet Solver can hence provide a
starting point and a benchmark result for future studies on such problems. In
the rest of the section we go over two case studies in which VRP Spreadsheet
Solver was used, and a new VRP variant that we introduce as a result.

3.1. Healthcare sector

A non-profit organization based in Istanbul, Turkey provides a set of health-

care services at home, including visits by medical doctors and nurses providing
physiotherapy and psychotherapy, as well as logistic services such as patient
transport, domestic cleaning, and personal hygiene. The services are performed
at no cost and are provided for the poor, elderly, and disabled citizens. The
number of registered patients is over 3000, and an average of 1000 patients in
separate locations are served each day. The organization operates out of three
bases and owns a fleet of 90 vehicles, each of which can by driven by the health
specialists and contains a small medical inventory.
The logistics planning manager of the organization intended to use the nurse
service as a test case for VRP Spreadsheet Solver and apply it to the planning
of other services if the results were successful. This service operates out of one
of the bases on the Anatolian side of the city with 20 vehicles and serves 150
patients per day on the average. The visits are planned in advance with no time
window specified, and all patients scheduled for a visit on a given day must
be visited. The service time per patient, although there are slight variations,
was assumed to be constant. The vehicles are refueled during the night and
their range is enough to cover a day trip, so the distance limit is not a binding
constraint. Even though the vehicles are from different makers and models, they

6
are considered identical in terms of operational parameters. Each nurse has a
driving time limit of 8 hours as per the general regulations about driving, and
a working time limit of 9 hours including the lunch break.
Capacity of the vehicles does not seem to be a binding constraint, since there
is no patient transport service involved and the medical supplies in a vehicle can
last a day. However, concerns of equal work allocation to the nurses were brought
up that were not addressed by the working time limit. Given the size of the city
and heavy traffic, a solution may contain routes with long driving times and a
small number of patient visits, and other routes with short driving times and a
significantly larger number of visits. To prohibit unequal work distribution in
terms of patients visited, each vehicle was assigned a capacity that is slightly
larger than the ratio of the number of patients to be served to the number of
vehicles, and each patient location was assigned a demand value of 1.
Initial runs with this setting seemed to satisfy the workload equity concerns.
However, problems with the solution were realized upon a detailed analysis of
the results. The driving durations retrieved from the GIS web service are based
on using the main arteries of the transportation infrastructure (i.e. highways),
and although the speed of transportation on the arteries are acceptable, getting
into and out of the arteries is time consuming. Hence, a solution that visits
multiple districts of the city accumulates a higher driving time than the GIS
driving durations provide. To overcome this problem, the users at the organiza-
tion modified the driving durations by adding a penalty term to the durations
between locations in separate districts to prohibit the routes from changing dis-
tricts multiple times. After a few experiments with this penalty parameter, the
users found a setting for which the resulting routes were observed to be more
realistic. The visualization of the resulting solution for a day with 150 patient
locations is provided in Figure 1. The return arcs to the depot are omitted for
a clearer visualization.
The organization refrained from providing figures regarding cost savings,
but the feedback indicated that the planning process was more transparent
to their staff and the use of the tool increased their awareness of the finer

7
 Figure 1: Visualization of the result for the case study in the healthcare sector

details of the transportation component of their operation. This case study

serves to show that the decision makers often have managerial concerns that
are not directly addressed in the routing literature, the GIS data may not give
an accurate reflection of the reality, and parameters of VRP Spreadsheet Solver
can be used to generate solutions that can address managerial concerns as well
as shortcomings of the available data.

3.2. Tourism sector

A tourism company based in Finland offers numerous types of travel pack-

ages, the main one being a ferry trip between the cities of Helsinki and Tallinn
(Estonia). The customers use the ferry for a day trip or an overnight visit to
Tallinn, or to other cities in Estonia / Baltic countries for 1-7 days. These routes
are operated from 4 to 7 days during the week, depending on the season. The
company aims to maximize profitability by planning the travel packages based
on demand forecasts and vehicle capacities. A significant portion of their cost is
due to the bus service they outsource from subcontractors to pick up customers
from their houses on the morning of their trip, and return them back to their
houses at the end of their trip.
The buses subcontracted by the company are based in 7 depots, one of them

8
in Helsinki. The buses are of different models and make, and can have different
carrying capacities. The subcontractors charge a fixed price per day of use,
so the problem becomes minimizing the number of buses needed, a problem
of packing as well as routing. The customers need to be at the ferry terminal
in Helsinki 15 minutes before the ferry departs, and some depots are located
quite far away from Helsinki, which pushes the driving time and working time
limits to be binding constraints. There are no time windows for the customers,
and all locations containing customers must be serviced. The buses based in
Helsinki perform closed tours, in which they pick up customers and return to
the ferry port. On the other hand, the buses based at the other depots perform
“open” tours, i.e. finish their trip in a location that is not their depot. These
buses eventually go back to their respective depots using the exact same route
traversed in the reverse direction. A sample solution is depicted in Figure 2,
where locations that do not contain any customers are not visited.

Figure 2: Visualization of the result for the case study in the tourism sector

The problem at hand is then an instance of the the Close–Open Mixed Vehi-
cle Routing Problem (COMVRP) introduced in [23]. We agree with the authors
of this paper that the COMVRP has applications in many sectors in which the

9
transportation service is subcontracted to companies based at multiple cities.
An extension of the COMVRP is the Close-Open Mixed Team Orienteering
Problem, in which the customers are serviced selectively based on the capacity
and the distance limit of the vehicles. A formulation that is capable of solving
both these new variants as well as others is presented in Section 5.

4. How to use VRP Spreadsheet Solver

In this section, we will briefly describe the structure of the worksheets and
the menu of VRP Spreadsheet Solver. We will be focusing on its usability rather
than the technical details, for which the interested user can refer to the user’s
manual.

4.1. Structure of the spreadsheets

VRP Spreadsheet Solver keeps the data about the elements of a VRP in
separate worksheets, and adopts an incremental flow of information. Initially,
the workbook only contains the worksheet named VRP Solver Console. The re-
maining worksheets, 1.Locations, 2.Distances, 3.Vehicles, 4.Solution, and 5.Vi-
sualization, are generated in the sequence denoted by their indices. Figure 3
depicts the information flow between the spreadsheets, where the arrows signify
the dependence of a worksheet on another worksheet.

Figure 3: Spreadsheet structure of VRP Spreadsheet Solver

To guide the user about which cells of a spreadsheet to work on, we have
adopted the following color scheme. The cells with a black background are

10
set by the worksheets and should not be modified. The cells with a green
background are parameters or decisions to be set by the user. The cells with
a yellow background are to be computed by the worksheets, but they can be
edited by the user for what-if analysis. The cells with an orange background
signal a warning, e.g. a vehicle arriving before the beginning of the time window
of a customer. The cells with a red background signal an error, e.g. a vehicle
violating the capacity constraint. The worksheets described below utilize this
color scheme.

4.1.1. VRP Solver Console

This worksheet stores and provides information to the rest of worksheets.
It contains various parameters regarding the size of the instance being solved
and its characteristics including the number of depots and customers, number
of vehicle types, the width of time windows. In addition, the user can set the
options about GIS data retrieval, and the time that the user allows the solver
to work on the problem. A screenshot of the worksheet is presented in Figure
4.

Figure 4: VRP Solver Console spreadsheet

4.1.2. 1.Locations
The details about the locations including their names, addresses, coordi-
nates, time windows, and pickup and delivery service requirements are kept in

11
this worksheet. The coordinates can be input manually, or copied and pasted
from an external source, or populated using the GIS web service based on the
addresses input by the user. It is good practice to provide a postcode with
every address, since vague addresses may correspond to unreachable points e.g.
the address of a park being resolved to be in the middle of a lake. It is possi-
ble to prohibit the vehicles from visiting certain customers using the options in
this worksheet, for quick what-if analysis without data modification. Figure 5
displays a screenshot of the worksheet.

Figure 5: 1.Locations spreadsheet

4.1.3. 2.Distances
This worksheet contains the distances and travel durations between every
two points that are specified in the 1.Locations worksheet. As of the time of
this writing, using the GIS web service to populate the distances and driving
distances takes about 5 minutes for 50 locations and 45 minutes for 150 locations.
The number of locations for which the distance matrix can be computed is
limited by the GIS web service and the type of access the user has to it. VRP
Spreadsheet Solver provides an estimate the time requirement for this step by
simply multiplying the number of entries in the distance matrix by a factor of
0.1 seconds.
The parameter about the type of route (shortest or fastest) is crucial. Choos-
ing the shortest route usually finds routes that go through city centers, which
are subject to strict speed limits and heavy traffic. Hence, using the fastest
route is usually a better option for large distance delivery operations. On the
other hand, fastest routes may end up using peripheral highways of the city too

12
frequently, and consequently the shortest paths may be better suited to com-
panies performing intra-city delivery operations. It is also possible to retrieve
real-time driving durations based on the traffic, which is computed and pro-
vided by the GIS web service. The user can prohibit the vehicles from traveling
between two given locations by manually setting the relevant distance to a high
value. A screenshot of the 2.Distances worksheet is depicted in Figure 6.

Figure 6: 2.Distances spreadsheet

4.1.4. 3.Vehicles
The data about the vehicle types are kept in this worksheet. The user can
set the number of vehicles of each type that are kept at each depot. The data
includes cost parameters such as the cost per unit distance and the cost per trip,
as well as operational parameters e.g. the depot, capacity, driving time limit,
and the distance limit of the vehicle. Only one capacity parameter exists, which
may correspond to the weight capacity of trucks in the case of an excavation
operation, the volume capacity of tanker trucks in the case of oil transport, or
the maximum number of passengers in the case of school bus routing. Figure 7
displays a screenshot of the worksheet.

13
 Figure 7: 3.Vehicles spreadsheet

4.1.5. 4.Solution
This worksheet is generated to contain the list of stops for each vehicle
specified in 3.Vehicles, and it uses the information in 1.Locations to regarding
service times and pickup / delivery amounts, as well as the distance and duration
in 2.Distances to compute the departure / arrival times the cost of traveling
between customers. The worksheet computes the net profit rather than cost, to
accommodate variants of the VRP that accumulate profits when customers are
selectively visited. This worksheet contains a number of conditional formatting
features that are designed to visually identify infeasible solutions and facilitate
manual solution construction. For example, a vehicle exceeding its capacity or
distance limit, or a customer being visited out of its time window are highlighted
in red. It is also possible to copy and paste lists of customers between vehicle
routes for the purpose of manual modification of the routes. A screenshot of
the 4.Solution worksheet is provided in Figure 8.

Figure 8: 4.Solution spreadsheet

14
4.1.6. 5.Visualization
The locations and the routes of the vehicles can be visually inspected by
generating this optional worksheet. Options in the VRP Solver Console may
be set to display various details about the locations including their pickup /
delivery amounts or service times. This worksheet simply contains a scatter
graph with the map of the region retrieved from the GIS web service. It can
be formatted, e.g. made smaller or larger, or display axes, for the needs of the
user as any other graph object of Excel. Figure 9 displays a screenshot of the
worksheet.

Figure 9: 5.Visualization spreadsheet

4.2. Structure of the menu of VRP Spreadsheet Solver

The menu is located in the “Add-ins” tab of the ribbon, and consists of
5 core and 3 optional commands to set up the worksheets. It also includes
the command to engage the solver, as well the optional calls to a feasibility
checker for manually modified solutions and an external solver that advanced
users may develop and compile into a Dynamically Linked Library (DLL) file.
The numerical indices of the commands match the numerical indices of the
worksheets for ease of use.

15
 Figure 10: VRP Spreadsheet Solver menu

5. A unified formulation for the VRP and the solution algorithm

The field of VRP research is mature and many solution algorithms have
been developed. The best known heuristic algorithm is arguably the savings
algorithm Clarke and Wright [6]. Many metaheuristic algorithms have been
proposed in the last decade, the most successful being the Adaptive Large Neigh-
bourhood Search [29], Iterated Local Search [31], and Genetic Algorithms [34].
In the rest of this section we provide a unified formulation for the VRP that
encompasses all variants of the VRP that VRP Spreadsheet Solver can handle,
the pseudocode of the metaheuristic solution algorithm implemented with VRP
Spreadsheet Solver, the details of how the infeasible solutions are handled, and
the computational results of our algorithm on benchmark instances.

5.1. A unified formulation for the VRP

We first provide the notation that we will use to state the formulation. Let
us define the vertex set VD to contain the depots, VC to contain the customers,
and V = VD ∪VC . Furthermore, we define VM ⊆ VC as the set of customers that
must be visited. Let G = (V, A) be the complete directed network on which we
will solve the VRP. We define the profit of servicing a customer i ∈ VC as pi ,
the pickup service amount for the customer as qi , the delivery service amount
as q̂i , and the service time required by the customer as si . Furthermore, we

16
define the time interval for the customer as [ai , bi ]. Note that there is also a
time interval for each depot vertex.
Let us denote the set of vehicles as K, and define for each vehicle k ∈ K
the origin depot of the vehicle as ok ∈ VD , the work start time of the vehicle
as τ k , the fixed cost of using the vehicle as f k the capacity of the vehicle as
Qk , the distance limit as Dk , the driving time limit as D̂k , the working time
limit as W k , and the return depot of the vehicle as rk . Associated with each
arc (i, j) ∈ A, there is a distance dij and driving duration dˆij . In addition, for
each vehicle k ∈ K, there is a travel cost ckij on arc (i, j).
Next, we present the parameters related to the operational constraints. Let
us define Ω to be equal to 1 if the vehicles have to return to their specified
return depots and 0 otherwise. Similarly, let us define β to be 1 if there is a
backhaul constraint, and 0 otherwise. In addition, we define Θ to be equal to
1 if the time windows can be violated at the cost of a penalty Π per unit time,
and 0 otherwise.
We are now ready to define the decision variables. Let xkij be equal to 1
if vehicle k traverses arc (i, j) and 0 otherwise. Furthermore, let yik be equal
to 1 if vehicle k visits and serves vertex i, and 0 otherwise. The amount of
the pickup commodity and the delivery commodity carried by vehicle k on arc
k k
(i, j) is defined as wij and zij , respectively. We also define tki as the time at
which vehicle k arrives at vertex i, and vi as the amount of violation of the time
window of vertex i. The formulation for the unified VRP is then:

X X X X X X X
Maximize pi yik − ckij xkij − f k xkok ,j − Π vi (1)
i∈VC k∈K (i,j)∈A k∈K j∈VC k∈K i∈V
X
subject to yik = 1 ∀i ∈ VM , (2)
k∈K
X
yik ≤ 1 ∀i ∈ VC \ VM , (3)
k∈K
X X
xkij ≤ xkji ∀j ∈ VC , k ∈ K, (4)
j∈V \{i} j∈V \{i}

17
 X
xkpq ≥ yik ∀i ∈ VC , k ∈ K, S ⊂ V : ok ∈ S, i ∈ V \ S,
p∈S,q∈V \S

(5)
X
xkpq ≥ Ωyik ∀i ∈ VC , k ∈ K, S ⊂ V : i ∈ S, rk ∈ V \ S,
p∈S,q∈V \S

(6)
X
xkok ,j ≤ 1 ∀k ∈ K, (7)
j∈VC
X
xkij ≤ 1 − β ∀(i, j) ∈ A : qi > 0 and q̂j > 0 (8)
k∈K
X X
k k
wij − wij = qi yik ∀i ∈ VC , k ∈ K, (9)
j∈V \{i} j∈V \{i}
X X
k
wi,r k = qj yjk ∀k ∈ K, (10)
i∈VC j∈VC
X X
k k
zji − zij = q̂i yik ∀i ∈ VC , k ∈ K, (11)
j∈V \{i} j∈V \{i}
X X
zokk ,j = q̂i yik ∀k ∈ K, (12)
i∈VC i∈VC

tki + (dˆij + si )xkij − W k (1 − xkij ) ≤ tkj ∀(i, j) ∈ A : j ∈ VC , k ∈ K,

(13)

ai ≤ tki ≤ bi − si + vi ∀i ∈ VC , k ∈ K, (14)

vi ≤ M.Θ ∀i ∈ VC , (15)

tkok = τ k ∀k ∈ K, (16)

tki + (si + dˆij )xki,rk ≤ brk + vrk + M (1 − Ω) ∀(i, j) ∈ A : i ∈ VC , k ∈ K,

(17)
k k
wij + zij ≤ Qk ∀(i, j) ∈ A, k ∈ K, (18)
X
dij xkij ≤ Dk ∀(i, j) ∈ A, k ∈ K, (19)
(i,j)∈A
X
dˆij xkij ≤ D̂k ∀(i, j) ∈ A, k ∈ K, (20)
(i,j)∈A
X X
si yik + dˆij xkij ≤ W k ∀(i, j) ∈ A, k ∈ K, (21)
i∈VC (i,j)∈A

18
 xkij ∈ {0, 1} ∀(i, j) ∈ A, k ∈ K, (22)

yik ∈ {0, 1} ∀i ∈ VC , k ∈ K, (23)

vi ≥ 0 ∀i ∈ VC , (24)
k
wij ≥ 0 ∀(i, j) ∈ A, k ∈ K, (25)
k
zij ≥ 0 ∀(i, j) ∈ A, k ∈ K. (26)

The objective function (1) maximizes the total profit collected minus the
travel cost of vehicles, fixed cost of using vehicles, and the penalty for violating
time windows. We first state the constraints set the visit rules for the customers
by the vehicles. Constraint (3) enforces a visit to the customers that must
be visited, and constraint (2) ensures that every customer is visited at most
once. Constraint set (4) is a weak form of the well-known flow conservation
constraints, which require an inflow if there is an outflow, and accommodates
the VRP variants in which the vehicle does not have to return to its depot.
Constraints (5) provide the connectivity between the origin depot of vehicle
k and the customers visited by this vehicle, and constraints (6) dictate the
vehicle to return to its depot if it is required to. Constraints (7) state that each
vehicle can be used at most once, whereas the backhaul constraint is enforced
by constraint (8).
Next, we present the constraints that set the customer requirements. The
flow conservation for the pickup commodity is provided by constraints (9) and
(10). Similarly, the flow conservation for the delivery commodity is provided by
constraints (11) and (12). Constraints (13) are formulated based on the Miller-
Tucker-Zemlin subtour elimination constraints [25] and provide the framework
for the time windows. The lower and upper limits of the time window for each
customer, and the variable to account for violation are stated in constraints (14)
and (15).
The final set of constraints state the restrictions related to vehicles. Con-
straints (16) and (17) set the start of the working time for vehicle k, and ensures
that the vehicle returns to its depot on time if it is required to. Constraint (18)

19
prohibit the violation of the vehicle capacities. Constraints (19), (20), and (21)
state the distance, driving time, and working time limits for each vehicle, re-
spectively. Finally, constraints (22) – (26) are integrality and nonnegativity
constraints.
To the best of our knowledge, there has not been any attempts to formu-
late a VRP with all the constraints stated above. Although the formulation
can be solved to optimality only for small instances, it defines the problem pre-
cisely, demonstrates its complexity, and will serve as a reference formulation for
the future studies on the VRP. Next, we provide our algorithm to solve this
formulation.

5.2. Pseudocode of the solution algorithm

We have opted to implement a variant of the Adaptive Large Neighbour-

hood Search of [29] within VRP Spreadsheet Solver, due to its flexibility to
accommodate many variants of the VRP. The algorithm diversifies the search
by randomly removing customers from the solution at hand, and intensifies
through re-insertion of the customers and local search. A high level pseudocode
is provided below, named as Algorithm 1.
Four local search operators have been utilized in Algorithm 1, namely EX-
CHANGE, 1-OPT, 2-OPT, and VEHICLE-EXCHANGE. The EXCHANGE op-
erator searches all possible pairs of customers in a given solution and checks if
exchanging them would result in a better objective function value. The opera-
tor 1-OPT examines the possibility of removing every customer within a given
solution and re-inserting it to a different position within the routes to improve
the objective value. The 2-OPT operator attempts to remove two arcs from the
solution at a time, e.g. the arc from customer a to customer b and the arc from
customer c to customer d. To retain feasibility, it then adds the arc from cus-
tomer a to customer d and the arc from customer c to customer b, and checks if
the resulting solution has a better objective value. All three operators described
so far have a neighborhood size of O(|V |2 ), and we refer the interested reader
to the review by [17] for their details. The operator VEHICLE-EXCHANGE

20
Algorithm 1 LNS algorithm implemented within VRP Spreadsheet Solver
1: procedure LNS(depots, customers, distances, durations, vehicles)
2: Construct an incumbent solution by adding customers to the routes, choosing the customer
that results in the maximal profit increase (equivalently, minimal cost increase) at every
step
3: Improve the incumbent solution using local search with the EXCHANGE, 1-OPT, 2-
OPT, and VEHICLE-EXCHANGE operators
4: Record the incumbent solution as the best known solution
5: repeat
6: Destroy the incumbent solution by randomly removing vertices
7: Repair the incumbent solution heuristically by adding vertices
8: Improve the incumbent solution using local search with the EXCHANGE, 1-OPT,
2-OPT, and VEHICLE-EXCHANGE operators
9: if the incumbent solution is better than the best known solution then
10: Record the incumbent solution as the best known solution
11: else
12: Replace the incumbent solution by the best known solution with probability p
13: until time elapsed is larger than the CPU time allowed
14: return best known solution
15: end LNS

attempts to exchange all the customers in the routes of two vehicles with dif-
ferent types, has a neighborhood size of O(|K|2 ), and is particularly useful for
the case of heterogeneous fleets.
Two constructive heuristics are employed in step 7 are greedy insertion and
max regret. The latter heuristic is based on selecting the customer, for which the
difference between the cost of the cheapest insertion and the second cheapest
insertion decisions is the largest. Both heuristics are chosen with equal prob-
ability at each iteration. Each heuristic finds a number of best candidates (a
parameter set by the algorithm) and chooses randomly among them at each
step. The probability p of rejecting an incumbent solution is set at 10% in the
beginning and decreases linearly with time to reach 0% at the end of the CPU
time allowance.

21
5.3. Handling infeasibility
Infeasibility of a solution refers to the case when one or more constraints are
violated by a solution, whereas infeasibility of an instance refers to the case when
it is not possible to find any feasible solution for the instance. Infeasibility is a
common occurrence in the field of algorithm development. However, a software
package that returns an infeasibility message with no suggestions for remedial
action is of little use to a practitioner, and the computer (or real) time spent
waiting for the result is perceived as wasted time.
Mathematically, a solution is either feasible or infeasible, and there is no
comparison between two infeasible solutions. However, many solutions that
would be declared infeasible by a computer could be useful in practice. For
example, an infeasible solution with a single route exceeding the capacity of the
vehicle by a fraction of the vehicle capacity can be converted into a feasible one
by renegotiating the delivery amount with a customer on the route. Similarly,
a vehicle exceeding the working time limit can be made feasible by paying the
driver for the extra time. As a consequence, all solutions may be infeasible but
in practice, some are less infeasible. This gives rise to the need of a method of
penalizing infeasibility based on its severity.
An intuitive way of penalizing infeasibility is to include it into the objective
function with a penalty coefficient. Let us denote the capacity of a vehicle in
a homogeneous fleet by Q, the capacity required for route to be Q′ , and the
penalty coefficient to be a large constant M . We then need to include a penalty
term of max{Q′ − Q, 0} × M to the objective function to ensure that violating
the capacity constraint will be penalized. However, minimizing this term does
not necessarily make the resulting solution useful. Consider the case with k
vehicles of capacity Q, and a a number of customers with a total demand of
(k + 1) × Q. In this case, the penalty value of a solution with the first vehicle
containing 2Q units and the rest of the vehicles containing Q units is equal to
the penalty value of a second solution with all vehicles carrying Q + 1 units.
However, many practitioners would find the latter solution to be more useful,
since there is a smaller degree of modification required on each vehicle route.

22
 The only hard constraint within the solution algorithm of VRP Spreadsheet
Solver is to visit customers that must be visited, and this constraint is enforced
on all solutions throughout the algorithm. The rest of the constraints are all
treated as soft constraints, and their violations are penalized. To prioritize
infeasible solutions with less severe violations, the solution algorithm uses a
quadratic scaling method for the penalty. Following the example in the previous
paragraph, the penalty term for a vehicle would be (max{Q′ − Q, 0}/Q)2 × M .
As a result, a violation of the capacity constraint by 5% of the capacity would
be penalized by 0.0025×M whereas a violation of 10% would be penalized by
0.01×M . Similar formulas apply for the violation of the time windows, distance
limit, driving time limit, and working time limit.
The solver component of VRP Spreadsheet Solver first performs a feasibility
check of the data and searches for possible reasons of infeasibility. The search
entails customers that must be visited but cannot be reached or serviced by
any vehicle within the given time limit, as well as pickup / delivery amounts
that cannot fit in any of the vehicle types. It also compares the overall carrying
capacity of the fleet to the total pickup / delivery requirement of the customers.
If any of these issues are found, the user is alerted with a message, and given a
choice to stop or proceed. If the user decides to proceed, the resulting solution
will certainly be infeasible but may still be useful.

5.4. Computational results on benchmark instances

Before presenting our computational results on benchmark instances, we

would like to make a brief comparison of the computational tests carried out
in the routing literature and the use of VRP optimization software in practice.
The academic studies are usually run on a state-of-the art computer dedicated
to the task, with the algorithms coded in C++. Additionally, the academic
studies limit the CPU time by restricting the algorithm a number of iterations,
which may result in different CPU times on different computers. On the other
hand, a practitioner is more likely to run the VRP optimization algorithm on an
ordinary laptop computer, possibly with other programs running in the back-

23
ground. Furthermore, a practitioner will need a solution in a given amount of
time, since the output of the software will be an input of the decision process
rather than its result.
We also would like to emphasize that the speed of VBA is orders of magni-
tude lower than that of C++. To the best of our knowledge, there is no academic
source that provides a speed comparison, so we have performed the following
small experiment. We have created an array of integers with 10000 elements, and
we have filled this array with ⌊U [0, 1] ∗ 1000⌋, and repeated this process 10000
times. On the average of 10 runs of this small program, C++ implementation
of this small program has taken 1.56 seconds, whereas the VBA implementa-
tion has required 4.20 seconds, approximately 2.7 times that of C++. In terms
of memory management and pointers, C++ is known to be very efficient, the
importance of which is realized for problems with large memory requirements
and results in larger performance deviations.
Based on the reasons stated above, we have decided to test the solution
algorithm using a laptop computer with an Intel i7 CPU running at 2.5 GHz
with 8 GB of RAM, a configuration that would reflect the computers used
in practice. We have set the CPU time limit for the solution algorithm VRP
Spreadsheet Solver to 15 minutes for instances with 50 customers, and increased
it linearly with the number of customers for larger instances. We do not claim
that a single algorithm can solve all the variants of the VRP to near-optimality,
and we believe that a computational experiment to solve all existing variants is
beyond the scope of this paper. Hence, we have opted to use the well-known and
widely used benchmark data set by [5] that contains two of the main variants
of the VRP, the Capacitated VRP and the Distance Constrained VRP.
The computational results are provided in Table 1 and show that the algo-
rithm performs very well for up to 100 customers, and returns acceptable results
for larger instances. The performance of the algorithm is better for instances
with a distance constraint, due to the reduced search space. The performance
is slightly degraded for the last four instances, which consist of artificially con-
structed clusters of customers. We recommend a larger CPU time allowance for

24
 Table 1: Computational results on benchmark instances
Instance Number of Fleet Vehicle Distance Best known VRP Spreadsheet Solver
name customers size capacity limit solution value Average Average gap Best Best gap
vrpnc1 50 5 160 N/A 524.61 524.61 0.00% 524.61 0.00%
vrpnc2 75 10 140 N/A 835.26 840.67 0.65% 835.26 0.00%
vrpnc3 100 8 200 N/A 826.14 841.05 1.80% 831.28 0.62%
vrpnc4 150 12 200 N/A 1028.42 1052.22 2.31% 1040.81 1.20%
vrpnc5 199 17 200 N/A 1291.29 1341.19 3.86% 1323.08 2.46%
vrpnc6 50 6 160 200 555.43 556.77 0.24% 555.43 0.00%
vrpnc7 75 11 140 160 909.68 913.13 0.38% 909.68 0.00%
vrpnc8 100 9 200 230 865.94 876.40 1.21% 865.94 0.00%
vrpnc9 150 14 200 200 1162.55 1181.77 1.65% 1170.81 0.71%
vrpnc10 199 18 200 200 1395.85 1435.27 2.82% 1415.02 1.37%
vrpnc11 120 7 200 N/A 1042.11 1047.82 0.55% 1047.61 0.53%
vrpnc12 100 10 200 N/A 819.56 821.29 0.21% 821.29 0.21%
vrpnc13 120 11 200 720 1541.14 1565.01 1.55% 1554.51 0.87%
vrpnc14 100 11 200 1040 866.37 886.41 2.31% 869.96 0.41%

such instances.
A copy of VRP Spreadsheet Solver containing a real-world Pickup-and-
Delivery VRP instance with 27 customer locations has been made available for
researchers to verify the solution algorithm and use the instance as a benchmark
for future studies [11]. As a final note, we would like to state that we welcome
contributions from the researchers developing algorithms for the VRP, in the
form of DLL files containing their solution algorithms. Contributed implemen-
tations will be hosted online jointly with VRP Spreadsheet Solver, with the full
credit of each DLL file being attributed to the researchers who contributed it.

6. Concluding remarks

In this paper, we have introduced VRP Spreadsheet Solver, an open source

spreadsheet solver for the VRP and its many variants. We have demonstrated
its realized and potential benefits on two case studies, one from the healthcare
sector and the other one from the tourism sector. We have shown the per-
formance of the solution algorithm on a set of benchmark instances from the
literature. We believe that VRP Spreadsheet Solver has the potential to be

25
used throughout the world due to its ease of use, flexibility, and accessibility,
and achieve savings for many SMEs as well as CO2 emissions.
Any decision support tool should be able to generate alternative solutions
for the decision maker. VRP Spreadsheet Solver currently returns and displays
a single solution, for the sake of simplicity. As future work, we plan to add a
parameter for the number of alternative solutions required by the user, record
the corresponding number of best solutions encountered during the solver run,
and return the results.
Acknowledgment: We thank Maria Battarra for her constructive comments
on a draft of this paper. Thanks are also due to the two anonymous reviewers,
whose comments have helped to improve the paper.

References

[1] J. L. Andrade-Pineda, P. L. Gonzalez-R, J. M. Framinan, A decision-

making tool for a regional network of clinical laboratories, Interfaces 43 (4)
(2013) 360–372.

[2] R. Baldacci, A. Mingozzi, R. Roberti, New route relaxation and pricing

strategies for the vehicle routing problem, Operations Research 59 (5)
(2011) 1269–1283.

[3] T. Bektaş, S. Elmastaş, Solving school bus routing problems through in-
teger programming, Journal of the Operational Research Society 58 (12)
(2007) 1599–1604.

[4] O. Bräysy, G. Hasle, Software tools and emerging technologies for vehicle
routing and intermodal transportation, in: P. Toth, D. Vigo (Eds.), Vehicle
Routing: Problems, Methods, and Applications, vol. 18, SIAM, 351–380,
2014.

[5] N. Christofides, A. Mingozzi, P. Toth, Exact algorithms for the vehicle

routing problem, based on spanning tree and shortest path relaxations,
Mathematical Programming 20 (1) (1981) 255–282.

26
 [6] G. Clarke, J. W. Wright, Scheduling of vehicles from a central depot to a
number of delivery points, Operations Research 12 (4) (1964) 568–581.

[7] F. G. Şahinyazan, B. Y. Kara, M. R. Taner, Selective vehicle routing for a

mobile blood donation system, European Journal of Operational Research
245 (1) (2015) 22 – 34.

[8] G. B. Dantzig, J. H. Ramser, The truck dispatching problem, Management

Science 6 (1) (1959) 80–91.

[9] E. Demir, T. V. Woensel, T. de Kok, Multidepot distribution planning at

logistics service provider nabuurs b.v., Interfaces 44 (6) (2014) 591–604.

[10] R. Eglese, T. Bektaş, Green vehicle routing, in: P. Toth, D. Vigo (Eds.),
Vehicle Routing: Problems, Methods, and Applications, vol. 18, SIAM,
437–458, 2014.

[11] G. Erdoǧan, Glasgow Bike Share – Rebalancing Problem, Data in Brief

Submitted.

[12] G. Erdoǧan, VRP Spreadsheet Solver, available for download at

http://verolog.deis.unibo.it/vrp-spreadsheet-solver, 2013.

[13] S. Erdoǧan, E. Miller-Hooks, A green vehicle routing problem, Transporta-

tion Research Part E: Logistics and Transportation Review 48 (1) (2012)
100 – 114.

[14] I. A. Forma, T. Raviv, M. Tzur, A 3-step math heuristic for the static
repositioning problem in bike-sharing systems, Transportation Research
Part B: Methodological 71 (2015) 230–247.

[15] D. Gavalas, C. Konstantopoulos, K. Mastakas, G. Pantziou, A survey on

algorithmic approaches for solving tourist trip design problems, Journal of
Heuristics 20 (3) (2014) 291–328.

[16] C. Groër, B. Golden, E. Wasil, The consistent vehicle routing problem,

Manufacturing & Service Operations Management 11 (4) (2009) 630–643.

27
[17] C. Groër, B. Golden, E. Wasil, A library of local search heuristics for the
vehicle routing problem, Mathematical Programming Computation 2 (2)
(2010) 79–101.

[18] V. C. Hemmelmayr, K. F. Doerner, R. F. Hartl, D. Vigo, Models and

algorithms for the integrated planning of bin allocation and vehicle routing
in solid waste management, Transportation Science 48 (1) (2014) 103–120.

[19] R. Hesse, D. H. Scerno, How electronic spreadsheets changed the world,

Interfaces 39 (2) (2009) 159–167.

[20] G. Hiermann, J. Puchinger, S. Ropke, R. F. Hartl, The electric fleet size and
mix vehicle routing problem with time windows and recharging stations,
European Journal of Operational Research 252 (3) (2016) 995–1018.

[21] M. Huang, K. R. Smilowitz, B. B. k, A continuous approximation approach

for assessment routing in disaster relief, Transportation Research Part B:
Methodological 50 (2013) 20 – 41.

[22] G. Laporte, Fifty years of vehicle routing, Transportation Science 43 (4)

(2009) 408–416.

[23] R. Liu, Z. Jiang, The close-open mixed vehicle routing problem, European
Journal of Operational Research 220 (2) (2012) 349–360.

[24] D. S. Mankowska, F. Meisel, C. Bierwirth, The home health care routing

and scheduling problem with interdependent services, Health Care Man-
agement Science 17 (1) (2014) 15–30.

[25] C. E. Miller, A. W. Tucker, R. A. Zemlin, Integer programming formulation

of traveling salesman problems, Journal of the ACM 7 (4) (1960) 326–329.

[26] A. Nerg, K. Stuckenschneider, Domestic Disasters and Geospatial Tech-

nology for the Defense Logistics Agency, MBA thesis, Naval Postgraduate
School, Monterey, California, 2014.

28
[27] L. Özdamar, O. Demir, A hierarchical clustering and routing procedure for
large scale disaster relief logistics planning, Transportation Research Part
E: Logistics and Transportation Review 48 (3) (2012) 591 – 602.

[28] J. Partyka, R. Hall, Vehicle routing software survey: VR delivers the goods,
OR/MS Today 41 (2014) 40–46.

[29] D. Pisinger, S. Ropke, A general heuristic for vehicle routing problems,

Computers & Operations Research 34 (8) (2007) 2403 – 2435.

[30] P. Schittekat, K. Sörensen, OR practice–supporting 3PL decisions in the au-

tomotive industry by generating diverse solutions to a large-scale location-
routing problem, Operations Research 57 (5) (2009) 1058–1067.

[31] A. Subramanian, L. M. A. Drummond, C. Bentes, L. S. Ochi, R. Farias, A

parallel heuristic for the vehicle routing problem with simultaneous pickup
and delivery, Computers & Operations Research 37 (11) (2010) 1899–1911.

[32] P. Toth, D. Vigo, Vehicle routing: problems, methods, and applications,

vol. 18, SIAM, 2014.

[33] N. L. Van Wassenhove, Humanitarian aid logistics: supply chain manage-

ment in high gear, Journal of the Operational Research Society 57 (5)
(2006) 475–489.

[34] T. Vidal, T. G. Crainic, M. Gendreau, C. Prins, A unified solution frame-

work for multi-attribute vehicle routing problems, European Journal of
Operational Research 234 (3) (2014) 658 – 673.

[35] X. Wang, M. Battarra, B. Golden, E. Wasil, Vehicle routing and schedul-

ing, in: D. Teodorovic (Ed.), The Routledge Handbook of Transportation,
Routledge, London, U. K., 238–256, 2015.

[36] J. Yi, Vehicle routing with time windows and time-dependent rewards: A
problem from the american red cross, Manufacturing & Service Operations
Management 5 (1) (2003) 74–77.

29