PRIO Project Group_7

Abraham Peres (91674), Ariadne Monteiro (94134), Bárbara Ferreira (94629), Cristina Cunha (82245), Christoph
Geisendörfer (94631), José Cordeiro (93888).

1 Abstract

The Portuguese company PRIO is responsible for the distribution and commercialization of liquid fuels and the production
of biofuels. However, it aims to find ways to improve its supply and operational flow, aiming at optimizing logistics costs
and processes. For this, the use of mathematical modelling techniques and the use of software and servers such as:
GAMS, EXCEL and NEOS were used to assist in compiling the model and extracting the results. We can mention the
creation of a terminal (Setúbal), as an supply option to consider and the great impact that the purchasing cost has on the
choice of suppliers.

2 Introduction

As in others industries, the transportation has a huge weight on the logistics cost of the fuel market, how to reduce this
cost usually is the main target of the logistics sector of these companies. The development of strategies based on location,
fleet management and partners generate some trade-offs which needed detail analysis to achieve a decision. About the
fuel market it is few more complex, it involves territorial factors, path restrictions, agreements among companies and
government – local taxation – and the big investment required to build a new supply point. Nevertheless, this scenario is
one of those it has more data and equations available to compare the options using mathematical results.

At Portugal market scenario, the road fuel demand is supplied by 5 operators, being the Galp Energia the only company
able to refiner the fuel and the other 4 companies (Repsol, BP, Prio and CEPSA) working as supplies points, using logistics
strategies to supply the services stations affiliates. The Prio is the only Portuguese company on this market providing
diesel, gasoline 95 and 98 (without and with additives) for more than 250 services station. Therefore, the Prio sees on
the knowledge of logistics and distribution approached on this course the opportunity to analyse the Portugal’s market
fuel scenario to use this information to establish strategies to be more competitive in this market.

The Prio provided data about the market share of each company and the costs related to distribution by product and
district. All this information is to solve the problem of optimize routes identifying the scenarios: lower total costs of
supplying by each company, reducing of supply location, suggesting of an optimal location and different demand’s
scenario. To answer these questions the problem was defined by a balance to be equated changing the distance between
point as the main variable to achieve the optimal cost and location required by each scenario.

3 Model Formulation

For the problem two gas stations per district and company were chosen to represent all gas stations of the company in
this district. To achieve a better representation of all gas stations, the two gas stations were selected so they
approximately cover half of the district and can represent about half of the stations each.

We developed two scenarios: For the first scenario each gas station represents one big gas station that combines half of
the demand of the district and company. In this scenario a demand of 400 for this district and company would result in a
demand of 200 for each station. This scenario is the primarily used model in this project. In the second scenario the two
gas stations represent half of the gas stations. So if there are 20 gas stations of the company in the district, each of the
two stations represents 10 gas stations. This can change the results because in the first scenario you need 7 trucks per
location (200/32 = 6.25 → 7 trucks) while in the second scenario 10 trucks are needed (20/32 = 0.625 → 1 truck → 10
trucks because 10 stations need to be supplied).
1
The following model is the formulation of the first scenario. To describe PRIO problem in a mathematical model
formulation and reply to question 1 of the case [concerning to what would be the volume of throughput per each supply
location with the lower total (purchasing and transportation) cost for each of the considered oil companies (Repsol, BP,
Prio and CEPSA)] we started to identify the applicable sets, such as: 𝑆 for the supply locations{𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆, … },𝑆 for the
examined companies that operate gas station{𝑆𝑆𝑆𝑆𝑆𝑆, 𝑆𝑆𝑆𝑆𝑆, 𝑆𝑆, 𝑆𝑆𝑆𝑆},𝑆 for the districts of Portugal where examined
gas stations are locate{𝑆𝑆𝑆𝑆𝑆𝑆, 𝑆𝑆𝑆𝑆, … },𝑆 for the annotation to differentiate between gas stations of the same company
in the same district {𝑆1, 𝑆2}and 𝑆for the types of products {𝑆98, 𝑆95, 𝑆𝑆𝑆𝑆𝑆𝑆}. Each gas stations is explicitly defined by
the values of 𝑆, 𝑆 and 𝑆. The cardinality of these sets are S, C, D, G and P respectively.

For the parameters the following notation was established:

➔ 𝒕𝒕 is the transportation cost travelled by the truck (in € per km);

➔ 𝒅𝒆𝒎𝒂𝒏𝒅 (𝒄, 𝒅, 𝒈, 𝒑)is the demand volume of gas station for product p (in m3);
➔ 𝒅𝒊𝒔𝒕 (𝒄, 𝒅, 𝒈, 𝒔)is the distance between supply location and gas station (in km).
➔ 𝒕𝒕𝒕𝒕𝒕𝒕𝒕 (𝒕, 𝒕) with the premium costs of products 𝒕 per supply locations 𝒕 (in € per m3).

To be able to easily implement 𝒕𝒕𝒕𝒕𝒕𝒕𝒕 (𝒕, 𝒕) without it causing problems in the model, the prices for G95 and G98 at
the supply location Repsol Sines (both products are unavailable there) were set to 1000 € per m 3 to make it impossible
to consider the location as a supply for those products.

The variables used are: 𝒕𝒕𝒕 (𝒕, 𝒕, 𝒕, 𝒕) that represents the number of trucks, being 𝒕𝒕𝒕 an integer variable; 𝒕𝒕𝒕(𝒕, 𝒕, 𝒕)
that represents the volume of product 𝒕 from each supply locations to a company; 𝒕(𝒕, 𝒕, 𝒕, 𝒕, 𝒕)that assume the value
one if s supply location with product 𝒕 and zero otherwise, being 𝒕 a binary variable and; 𝒕 that represents the objective
function.

The problem purpose is:

𝑆 𝑆 𝑆 𝑆 𝑆 𝑆 𝑆 𝑆 𝑆

(0) 𝑆𝑆𝑆 ∑ ∑ ∑ ∑ ∑ 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆 + ∑ ∑ ∑ ∑ 𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 2

𝑆=1 𝑆=1 𝑆=1 𝑆=1 𝑆=1 𝑆=1 𝑆=1 𝑆=1 𝑆=1

subject to the following constraints:

(1) 𝑆𝑆𝑆𝑆𝑆𝑆𝑆 ≤ ∑𝑆𝑆=1 𝑆𝑆𝑆𝑆𝑆𝑆 × 100000, ∀ 𝑆, 𝑆, 𝑆, 𝑆

(2) ∑𝑆𝑆=1 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆 ≤ 32 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆 , ∀ 𝑆, 𝑆, 𝑆, 𝑆
(3) ∑𝑆𝑆=1 𝑆𝑆𝑆𝑆𝑆𝑆 = 1, ∀ 𝑆, 𝑆, 𝑆, 𝑆.

The objective function (0) minimizes the total cost of purchasing and transportation, where the first component of the
equation represents the fixed cost of the purchasing (premium cost) when the product is supplied from supply locations
to company gas station and the second component represents the transportation cost. The constraint (1) ensures that
trucks can only drive between supply locations and gas station if there is a connection between them. The value 100000
on the right side is there to not limit the number of trucks. If for example∑𝑆𝑆=1 𝑆𝑆𝑆𝑆𝑆𝑆 = 1, up to 100000 trucks can drive
from 𝑆 to gas station 𝑆𝑆𝑆 which is much more than needed in any case. The constraint (2) ensures that enough trucks are
provided to 𝑆𝑆𝑆 if a part or all of the demand is supplied by the supply location 𝑆. The constraint (3) ensures that every
company gas station is supplied with all products.

For question 1 three variable and three equations were added. For the volume throughput per supply location and
product for each company the variable 𝑆𝑆𝑆𝑆𝑆𝑆 . To calculate these volumes the following equation was added:

(4) 𝑆𝑆𝑆𝑆𝑆𝑆 = ∑𝑆𝑆=1 ∑𝑆𝑆=1 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆 , ∀ 𝑆, 𝑆, 𝑆

For the fuel cost and transportation cost per company the variables 𝑆𝑆𝑆𝑆𝑆𝑆 and 𝑆𝑆𝑆𝑆𝑆𝑆 were added. The equations for
calculating the values of these variables are the following:
2
 (5) 𝑆𝑆𝑆𝑆𝑆𝑆 = ∑𝑆𝑆=1 ∑𝑆𝑆=1 ∑𝑆𝑆=1 ∑𝑆𝑆=1 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆
(6) 𝑆𝑆𝑆𝑆𝑆𝑆 = ∑𝑆𝑆=1 ∑𝑆𝑆=1 ∑𝑆𝑆=1 𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 2

For question 2 one variable and two equations had to be added. The binary variable 𝑆𝑆𝑆𝑆𝑆𝑆𝑆 signifies whether supply
location 𝑆 is open (value is 1) or closed (value is 0). The equations are as follows:

(7) ∑𝑆𝑆=1 ∑𝑆𝑆=1 ∑𝑆𝑆=1 ∑𝑆𝑆=1 𝑆𝑆𝑆𝑆𝑆𝑆 ≤ 𝑆𝑆𝑆𝑆𝑆𝑆𝑆 ∗ 1000000

(8) ∑𝑆𝑆=1 𝑆𝑆𝑆𝑆𝑆𝑆𝑆 = 2

In contrast to (4), (5) and (6) which just function as data collection functions (7) and (8) are real constraints. Equation (7)
ensures that gas stations can only be supplied by 𝑆 if it is open (𝑆𝑆𝑆𝑆𝑆𝑆𝑆 = 1). (8) limits the number of open supply
locations to 2. To look for the optimal two supply locations for each company the objective function had to be changed
as well. In both sums of the objective function the condition was introduced that 𝑆 must equal the respective company
name.

To find what would be the optimal supply locations without taking into consideration the product cost difference per
location for each oil company (question nr. 3 of the case), we did some changes in the original model. The parameter
𝒕𝒕𝒕𝒕𝒕𝒕𝒕 (𝒕, 𝒕) was eliminated, as well as the first component of the objective function. The rest of the model remained
unchanged, so the objective function now had the following formulation:

𝑆 𝑆 𝑆 𝑆

𝑆𝑆𝑆 ∑ ∑ ∑ ∑ 𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 2

𝑆=1 𝑆=1 𝑆=1 𝑆=1

At last, to run a “stress test” considering a volume reduction up to 15% equally spread along all regions we reduced the
demand data by 15% for each product and district in the excel file and converting that to a GDX file, leaving the model
unchanged.

For the second scenario the model above only needed to be slightly changed. The new parameter 𝑆𝑆𝑆𝑆𝑆𝑆 , which
represents the number of gas stations of a company 𝑆 in a district 𝑆, was introduced. Additionally, the demand data was
adjusted. Each gas station 𝑆𝑆𝑆 now had a demand per product calculated by the demand per district and product times
the market share of company 𝑆 divided by 𝑆𝑆𝑆𝑆𝑆𝑆 . Furthermore, the objective function was adjusted to
𝑆 𝑆 𝑆 𝑆 𝑆 𝑆 𝑆 𝑆 𝑆

𝑆𝑆𝑆 ∑ ∑ ∑ ∑ ∑ 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆 × 0.5 + ∑ ∑ ∑ ∑ 𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆 × 𝑆𝑆𝑆𝑆𝑆𝑆

𝑆=1 𝑆=1 𝑆=1 𝑆=1 𝑆=1 𝑆=1 𝑆=1 𝑆=1 𝑆=1

4 Data Collection & Results

The data that was first collected was the coordinates of the supply stations and of gas stations of the inspected
companies. This was done directly from the companies websites 1234 which provided ample data in the case of the gas
stations. The supply locations were extracted by searching on “Google Maps”. It was assumed that the supply locations
“Galp Leixões” and “Cepsa Matosinhos” as well as “Galp Sines” and “Repsol Sines” are respectively in the exact same
location. The location data was stored in a latitude and longitude format. This data was then used to calculate the
distances between supply stations and gas stations. To achieve this calculation a free-of-charge API (“Application
Programming Interface”) from Microsoft was used 5. This API uses “Bing Maps” to calculate the driving distance between
two locations in latitude and longitude format. Because of complexity reasons the limitation of not being able to use the

1
https://www.prio.pt/pt/postos/rede-de-postos_289.html
2
http://bpplusmaps.bp.com/
3
https://www.repsol.com/SA/ProductosyServicios/EESS/EESS_Portugal/privado/formulario.aspx
4
https://pt.cepsa.com/pt/utilizacao/produtos-servicos/postos-de-abastecimento/postos-de-abastecimento
5
https://docs.microsoft.com/en-us/bingmaps/rest-services/routes/calculate-a-distance-matrix
3
bridge “Ponte 25 de abril” was not considered. Therefore, there might be some errors in the distance data, especially for
routes that run through the Lisbon area.

The demand data was obtained by extracting it from the PDF “LD_Project_PRIO”. These figures per district and product
where then multiplied by the market share of the inspected companies to obtain the demand figures per district, product
and company.

Demand data as well as the distance data were then stored in an EXCEL and not in the GAMS code. To be usable for the
GAMS code the EXCEL file needs to be converted into an GDX file. This was done by running the code locally. In order
that this can be done the “model” and the “solve” statement lines need to be commented out. Otherwise the limitation
of the demo license will terminate the execution of the code. Lines 2-4, 23 and 26 need to be activated by deleting “*”
and line 28 commented out (those are the lines regarding Gams file “Prio_Case_question1”, in other files this might vary).
Then the code can be run and a GDX file is created from the selected EXCEL file. To run the code on an online server all
the previous steps have to be undone and the GDX file uploaded together with the code.

There are some of the selected gas stations that don't sell all products (e.g. G98 - Repsol in Guarda) but since all
companies have a demand for all products in each district (e.g. Repsol has a demand of approximately 126 m 3 G98 in
Guarda) we just assumed those stations have all products for problem simplification. When running the problem on an
online server on NEOS the solver automatically sets the upper bound of integer variables to 100. Since 𝑆𝑆𝑆𝑆𝑆𝑆𝑆 can easily
surpass this figure we included the line 𝑆𝑆𝑆𝑆𝑆𝑆 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 = 0; in the code. This sets the upper limit of all integer
variables in the programm to +𝑆𝑆𝑆.

For the problem model compilation the solver used was GAMS Version win64 25.1.2 (milp:CPLEX:GAMS) with recourse
to the NEOS Server Version 5 for model running due to model size limitation. Table 1 show the computational results for
model used for each question taking into account the changes we did to answer each question.

Table 1. Computational results printed by NEOS Server

Question Model Compilation Time Execution Time Model Status Solver Status Equations Variables
Size [seconds] [seconds]

1 6MB 0.001 0.017 1 optimal solution 1 normal completion 3433 5881

2 6MB 0.002 0.020 1 optimal 1 normal completion 3324 5771

3 6MB 0.002 0.016 1 optimal 1 normal completion 3313 5761

4 6MB 0.002 0.016 1 optimal 1 normal completion 3313 5761

Concerning the volume of throughput per each supply location with the lower total cost (purchasing and transportation)
for each of company, the output of model gives to us seven supply locations as the optimal solution for the lower total
costs. Figure 1 show the seven supply location and amount of volume throughput per each supply location for each oil
company by product. This figure shows that the new terminal in setubal is one choice for each company taking into
account lower total costs.

4
 Figure 1. Volume throughput per each supply location for the considered oil companies by product

Considering the volume purchasing cost and transportation cost, the model gives to us the total lower cost in euros per
company demonstrated in figure 2. With this result we found interesting to give a description of the total cost of figure
2, in relation to the cost of transport and cost of acquisition, which is show in table 2.

Figure 2. Total costs for each oil company

5
 Table 2. Fuel and transportation costs by company

Purchasing Cost Transportation Cost Total Cost

Repsol 9 378 215,24 € 5 360 016,97 € 14 738 232,21 €

Cepsa 5 063 440,55 € 2 702 003,11 € 7 765 443,66 €

BP 8 802 082,52 € 4 724 482,53 € 13 526 565,05 €

Prio 8 659 011,04 € 5 857 719,67 € 14 516 730,71 €

Total Cost: 50 546 971,62 €

Limiting the number of supply locations to two resulted in a significant rise of costs. When minimizing the total cost of all
companies, the additional cost amounted to 7 097 494,71 €. The two supply locations chosen were Galp Leixões and the
new terminal in Setubal. For the next part of the question, the objective function was restricted to only consider the cost
of the respectively observed company. In each case the two optimal supply locations were Galp Leixões and the new
terminal in Setubal. The companies Repsol, Cepsa, BP and Prio had an additional cost of 2 050 449,23 €, 1 065 510,79 €,
2 095 830,79 € and 1 885 703,91 € respectively. Naturally those additional costs add up to the total additional cost
obtained in the problem with the complete objective function because the optimal supply locations remain the same.

Disregarding the difference in product cost per location for each oil company, we obtained as a result that all supply
locations were included in the optimal solutions for the companies Cepsa, BP and Prio. For Repsol the total number of
used supply locations amounted to 9 - Repsol Sines was not included. Changes were observed by analyzing the number
of trucks between supply locations and gas stations. It could be observed that supply flows moved partially from Galp
Leixões to CEPSA Matosinhos and from Galp Sines to Repsol Sines. The biggest loss of supply was registered for the new
terminal in Setubal. The supply mostly moved to Alkion Tanquipor Barreiro and CLC Aveiras. For example, for Repsol the
number of trucks from Setubal to their gas stations amounted to 8 086 units (original prices considered) and from Alkion
Tanquipor Barreiro and CLC Aveiras 0 and 5 624 respectively. This changed to 568, 6 653 and 6 488 trucks respectively
when the price difference was removed. Disregarding minor changes, the rest of the supply allocation stayed the same.

The volume reduction of 15% for all demand led to a significant reduction in costs, number of trucks and volume
throughput per supply location. Apart from that no changes could be observed. All gas stations were still supplied by the
same supply locations without exception.

For scenario 2 an optimal solution of 50 663 216,65 € was achieved. This is a difference of 116 245,03 € to scenario 1. The
supply situation for Repsol and Cepsa stayed exactly the same. At BP 997,68 𝑆3 of G95 were supplied by CLH Huelva
instead of Galp Sines. At Prio 6814,24 𝑆3 of G95 and 415,04 𝑆3 of G98 were supplied by CLC Aveiras instead of Setubal and
83,36 𝑆3 of G98 by CLH Vigo instead of Galp Leixões. The number of trucks rose marginally. For example, to supply both
gas stations of Repsol in Aveiro a total of 2562 trucks are needed in scenario 1. In scenario 2 this rises to 2583. Because
of the marginal difference to scenario 1 in regard to costs and routing and the limited space in this report scenario 2 was
not further examined.

5 Results Discussion

The results show that in the optimal solution each observed oil company used the same supply locations. Those were the
supply locations Galp Leixões, PRIO Aveiro, CLC Aveiras, the new terminal in Setúbal, Galp Sines, CLH Vigo and CLH Huelva.
The other three supply location didn’t satisfy any demand in the optimal solution whatsoever. For Alkion Tanquipor
Barreiro this is likely due to the fact that the cost of purchase (premium costs) is so high that possible distance advantages
over other supply locations are easily outweighed. The supply locations CEPSA Matosinhos and Repsol Sines don’t satisfy
any demand because there location was assumed to be same as Galp Leixões and Galp Sines respectively. All products in

6
Galp Leixões are cheaper than in CEPSA Matosinhos and the price of the only available product in Repsol Sines, Diesel, is
undercut by Galp Sines.

When limiting the amount of suppliers to two, the results obtained show a significant increase in cost which was
expected. The optimal supply locations were Galp Leixões and the new terminal in Setúbal. This choice optimizes the
minimum costs in terms of purchasing costs and transportation costs, and among all suppliers. These two are probably
the optimal supplies because of their central geographic location and low prices. This conjecture is reinforced by the fact
that even when minimizing the costs only for one of the companies the two optimal supply locations remained to be Galp
Leixões and Setúbal. In terms of supply allocations, Galp Leixões would supply northern and a share of central Portugal
while the new Setúbal terminal would provide the south and the other share of the center of Portugal.

The new terminal in Setúbal is a promising supply location since it is chosen by all four considered companies as a high-
volume supplier. This is probably because it is in a good geographic location and the prices are lower than the nearest
other supplier, Alkion Tanquipor Barreiro. This results in the situation mentioned in the previous paragraph. So, for any
gas station near these two suppliers, the lower purchasing price outweighs the potential extra cost of transportation. In
the case that we disregard the difference in purchasing costs between suppliers, the new terminal in Setúbal loses almost
all of volume throughput to Alkion Tanquipor Barreiro. This implies that the location of the latter is more advantageous
than the one of the potential new terminal. The observation that supply flows moved partially from Galp Leixões to CEPSA
Matosinhos and from Galp Sines to Repsol Sines can be easily explained. As the transportation costs from those two pairs
of supply locations is respectively the same, the equalization of prices leads to an equalization of total costs of purchasing
from Galp Leixões/CEPSA Matosinhos and Galp Sines/Repsol Sines respectively. It is probable that in a case like this the
solver chooses the supply location to use arbitrarily.

The robustness of the solution was tested with a 15% reduction in volume for all demands. The supply allocation remains
the same with all gas station to being supplied by the same supply locations only with a reduction of costs (purchasing
and transportation). This reduction of costs is self-explanatory since the reduction of demand leads to a reduction of
volume which leads to reduced purchasing cost and less needed trucks for transportation of the products, which will
reduce transport costs. At last we looked at scenario 2 to verify whether the approach in relation to the number of gas
stations has a great impact. We anticipated it would require larger quantities of trucks, which would increase transport
costs and amplify its effect since the purchasing costs would be unchanged since the purchasing volume stays the same.
But as the results show the cost differed only by 116 245,03 € between the two scenarios. This amounts to a difference
of only 0.23% which is quite small.

6 Conclusions

The model clearly implies that the supply locations CEPSA Matosinhos and Repsol Sines are redundant. Furthermore, the
supply location Alkion Tanquipor Barreiro would be superseded by the new terminal in Setúbal. From this several
conclusions can be drawn. The locations CEPSA Matosinhos and Repsol Sines should be closed if the supply locations Galp
Leixões and Galp Sines respectively can handle the combined volume in reality. The situation for the new terminal in
Setúbal is different. The establishment of such a terminal is probably accompanied by high construction costs. The results
from the model with equalized prices show that the location of the new terminal is unfavourable compared to Alkion
Tanquipor Barreiro. If the new terminal’s purpose is to reduce the acquisition cost of fuel in the area, an agreement with
Alkion Tanquipor Barreiro could be reached and the prices there lowered. In this case the construction of the new
terminal would become obsolete. On the other hand, if the potential operator of the new terminal wants to enter the
market aggressively, he could exploit the high prices of Alkion Tanquipor Barreiro and absorb its market share. In this
case the investment cost could probably be easily recuperated.

The robustness check showed that the supply allocation is untouched by variations in demand. This is helpful as the the
results of the model (specifically which gas station is supplied by which supply location) can be used medium to long-
term as the real-world variations have little influence on the optimal supply allocation.

7
One of the limitations of this model was not being able to use the bridge “Ponte 25 de Abril”. Because we didn’t consider
this there might be some errors in the distance data, especially for routes that run through the Lisbon area. Thus, we
looked into a small sample of flows that might choose a route crossing the bridge “Ponte 25 de Abril” and we observed
that one in three crossed the bridge. This resulted in an underestimated distance by approximately 10 km. Since the
Lisbon area has a high density of gas stations this restriction should be included in future models although a distance
discrepancy of 10 km might be too small to make a difference. An additional limitation is that each truck can only supply
one gas station. In the real world one truck can probably approach more than one gas station and, in that case, the total
costs could be lower than in the model.

The simplification of demands is another limitation to the validity of the model. It is highly likely that only a few (if at all)
gas stations can store the volume that they supply to customers over the year. In this case trucks have to supply the gas
stations more than once per year which could be inefficient and would therefore increase the cost. To maximize efficiency
the ability of a truck to supply more than one gas station per trip should be introduced as mentioned above. If this was
introduced, the model would come to be a variation of the traveling salesman problem as subroutes would need to be
considered. In this case we would have to change our approach and use a heuristic method which could not guarantee
the optimization of the model but could be more easier to implement.

Two other major limitations relate to the simplification regarding the gas stations. Because we only looked at two gas
stations per company and district we only calculated two real distances. Taking those two distances as representative for
the other gas stations in the district could have led to a significant difference to the real transportation costs.
Furthermore, it was assumed that all gas stations of a company in one district have the same demand which is highly
unlikely.

Possibly the biggest misconception of the model is the missing limitations of the supply locations. We know that the
supply locations CEPSA Matosinhos and Repsol Sines exist although according to the model they should have been
superseded by their neighbouring supply locations. That they still exist despite of higher prices can be only explained by
limitations that the suppliers are governed by.

Additionally, to the model result we need to consider the risk involved in the operations to make a reliable decision - be
it about change of supply locations or the choice whether to create a new supply point. For this it is necessary to conduct
a detailed risk analysis of the implications of the changes, the efficiency of the operational work of the new terminal, the
life cycle order and the impacts on the gas stations. Those factors are probably only with difficulty simulatable by
equations and need to be observed by studying the company strategies and its operational plan. Other points to be
analyzed are the road access around the localizations, the infrastructure to handle dangerous material and the laws and
restrictions involving it.

Another proposal to better evaluate the problem in question would be through the use of simulation models aiming to
predict the best possible scenarios with the use of statistical techniques and concepts. Using this method, we could
estimate the best possible solution while considering the variability of the market.

References

1. Prio - Combustíveis, Mobilidade elétrica e Gás (2019). Avaible in: https://www.prio.pt/pt/postos/rede-de-

postos_289.html
2. Bp Portugal, Energia com Futuro (2019). Avaible in: http://bpplusmaps.bp.com/
3. Repsol exploração e produção (2019). Avaible in:
https://www.repsol.com/SA/ProductosyServicios/EESS/EESS_Portugal/privado/formulario.aspx
4. Cepsa - Gás, gasóleo e outros serviços para particulares (2019). Avaible in:
https://pt.cepsa.com/pt/utilizacao/produtos-servicos/postos-de-abastecimento/postos-de-abastecimento
5. Microsoft - Centro de Desenvolvimento do Windows Matriz de Distância (2019). Avaible in:
https://docs.microsoft.com/en-us/bingmaps/rest-services/routes/calculate-a-distance-matrix