Global Crude Oil Renery Models: Parametric and Nonparametric Methods

Global Crude Oil Renery Models:

Parametric and Nonparametric

Methods

Evar Umeozor

September 2023

Doi: 10.30573/KS--2023-MP02

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

Acknowledgments

The author thanks all KAPSARC colleagues and support staff for their constructive

discussions on the Oil Value-Chain Analyzer (KOVA) project under which this research

was funded and for their help with the data acquisition procedures, respectively.

About KAPSARC

KAPSARC is an advisory think tank within global energy economics and sustainability

providing advisory services to entities and authorities in the Saudi energy sector

to advance Saudi Arabia's energy sector and inform global policies through

evidence-based advice and applied research.

This publication is also available in Arabic.

Legal Notice

This Document (and any information, data or materials contained therein) (the

“Document”) shall not be used without the proper attribution to KAPSARC. The

Document shall not be reproduced, in whole or in part, without the written permission

of KAPSARC. KAPSARC makes no warranty, representation or undertaking whether

expressed or implied, nor does it assume any legal liability, whether direct or indirect,

or responsibility for the accuracy, completeness, or usefulness of any information that

is contained in the Document. Nothing in the Document constitutes or shall be implied to

constitute advice, recommendation or option. The views and opinions expressed in this

publication are those of the authors and do not necessarily reect the ofcial views or

position of KAPSARC.

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

Abstract: There is a dearth of publicly available literature on reliable, evidence-based modeling

representations of global crude oil reneries, which has led to a lack of transparency and consistency

regarding how business and government policy processes are informed in many energy systems and

energy transition modeling scenarios. This paper uses actual eld data to identify that 40 unique renery

congurations account for the global renery landscape. A machine learning algorithm training extremely

randomized tree regressor (ETR) predictors is applied to develop a nonparametric production model

of reneries. To facilitate computational convenience and suitable deployments in diverse use cases,

a multivariate linear regression (MLR) model of reneries is also presented. For all rened products

captured, the machine learning model demonstrates superior predictive performance, with coefcients of

determination of over 90%, but both the machine learning model and the regression model are useful. Other

performance metrics are assessed for both models.

Keywords: Crude oil, Renery, Conguration, Machine learning, Regression

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

Acronym Denition

CDU Crude distillation unit

CS Condensate splitter

HDTR Hydrotreater

LPG Liquied petroleum gas

RFMR Reformer

DSLR Desulfurizer

ISMR Isomerizer

VDU Vacuum distillation unit

RCC Resid catalytic cracker

FCC Fluid catalytic cracker

DH Distillate hydrocracker

RH Resid Hydrocracker

CKR Coker

NAP Naphtha

GAS Gasoline

KJF Kerosene/jet fuel

DGO Diesel/gas oil

HFO Heavy fuel oil

COK Coke

MLR Multivariate linear regression

ETR Extremely randomized trees regressor

HSO Heavy sour crude oil

HSW Heavy sweet crude oil

MSO Medium sour crude oil

MSW Medium sweet crude oil

LSO Light sour crude oil

LSW Light sweet crude oil

CAP Design capacity of renery

MMb/d Million barrels per day

Kb/d Thousand barrels per day

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

Crude oil reneries produce nished

petroleum products through a series of

processing steps known as renery unit

operations. The varieties and capacities of process

units, along with their operational characteristics,

differentiate one renery from another. Over the

years, the rening industry has adopted various

metrics to categorize reneries based on their

capital investment and upgrading capability.

These efforts have resulted in renery complexity

indicators such as the equivalent distillation

capacity, the Nelson complexity index and the

bottom of the barrel index (Kaiser 2017, Ogjresearch

2022). Although these metrics have helped quantify

the renery complexity and suitability for various

crude oil feed qualities, they fail to provide relevant

information for renery modeling applications

(Herce, Martini et al. 2022).

Renery models are essential for assessing the

economic and energy performance of reneries

and potentially exploiting opportunities to optimize

protability, efciency, and policy compliance.

From a process design perspective, these models

facilitate the development of optimal renery

congurations under prevailing operating and

logistical conditions in local and/or regional

markets. Business and government policymakers

draw insights from modeling and analysis for

investment selection, downstream diversication

and regulatory policy design and compliance.

Identifying the fundamental renery conguration

is crucial for realizing reliable and deployable

renery models.

Abella, Motazedi et al. (2015) used a predominantly

North American database to identify 10 unique

renery congurations for model development. Their

conguration-based modeling approach enabled

the deployment of models for estimating renery

energy consumption and greenhouse gas (GHG)

emissions. Doing so aided the identication and

exploration of efciency improvements and emission

reduction opportunities (Motazedi, Abella et al.

2017). Such models are also useful for assessing

crude feed blends and renery conguration options

to maximize the netback (Nduagu, Umeozor et al.

2018). They have also been extended to quantify

the environmental impact of the renery life cycle

(Young, Hottle et al. 2019). However, for the effective

application of conguration-based modeling, the

accurate identication of rening process units and

their attributes is crucial.

In 2020, there were 867 global crude oil reneries

with active capacities of 103.32 million barrels

per day (MMb/d), of which approximately 80%

were located outside North America (Platts 2022).

Identifying the unique congurations across

the global crude oil rening landscape would

facilitate individual, national, regional, and global

renery modeling. Such modeling could promote

various application studies related to, for example,

efciency improvement, emission reduction, policy

design, netback optimization, and investment

competitiveness analysis. The rest of this paper is

organized as follows: global renery classication

into design congurations is presented in the next

section. In section 3, modeling methods and the

data for both the parametric and nonparametric

modeling of renery congurations are presented.

The modeling results are presented in section 4, and

the conclusions and further research opportunities

are discussed in section 5.

1. Introduction

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

In this work, the S&P Global Platts World Renery

Database (Platts 2022) is used to investigate

global crude oil renery congurations using

process unit-level data for all operating reneries

during the 2005-2020 period. A clustering technique

is developed based on the availability of major

process units and their capacities relative to other

contiguous units in each renery plant. For the entire

study period, 40 unique renery congurations are

identied to represent all possible congurations

that operated each year over the 16-year period.

Essentially, 30 of the identied congurations

have not been previously reported and represent

those found outside the North American region,

particularly in Asia. All identied renery

congurations are presented in supplementary

information section SI.A.

Figure 1 shows the global rening capacity in

2020 by the identied renery congurations. The

renery congurations are classied so that design

complexity increases in relative terms from the

2. Global Renery Congurations

Figure 1: Global crude oil renery capacities by renery conguration (2020).

Source: KAPSARC Oil Value-Chain Analyzer

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

lower-numbered simpler reneries to the higher-

numbered complex reneries. For example, the

simplest renery – also known as a hydroskimming

renery – is labeled conguration 0 (Cong_0),

while the most complex deep-conversion renery

is labeled Cong_39. At approximately 16.4

MMb/d, the Cong_33 renery has the highest

total capacity in the world, whereas there has

been no existing Cong_22 renery capacity since

2016 due to either shutdowns or retrotting to a

different conguration class. The last Cong_22

renery was in operation in 2015, with a capacity of

approximately 190.8 Kb/d.

The global rening capacity processes approximately

100 MMb/d of crude oil blends, which are normally

categorized into 6 grades based on their sulfur

content and density. Figure 2 shows the average

global crude oil blend runs by renery conguration

between 2005 and 2020. Heavier (higher density)

2. Global Renery Congurations

Figure 2: Global crude oil blend runs by renery conguration (2005-2020 average).

Source: KAPSARC Oil Value-Chain Analyzer

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

and sour (higher sulfur content) crudes, such as

heavy sour crude oil (HSO), heavy sweet crude oil

(HSW), and medium sweet crude oil (MSO), are

mainly processed by higher complexity reneries,

whereas light and medium crude grades (light sweet

crude oil (LSW), medium sweet crude oil (MSW)

and light sour crude oil (LSO)) are mostly rened at

simpler and medium conversion reneries (see Table 1

for a description of crude grades). Furthermore,

crude blends are often matched to each renery

conguration to maximize the production of the

most desired products at the expense of the

least desirable products. Other reasons for crude

oil blending include satisfying environmental

requirements such as sulfur content limits, exploiting

the price differentials among crude grades in the

market, utilizing locally available crude oil products,

and optimizing the crack spread of rened products.

Currently, the global crude run data for Cong_9 and

Cong_19 reneries have yet to be captured in the

source database of world reneries.

The combinations of renery designs and their

operations processing various crude oil blends

yield various arrays of rened product slates.

Figure 3 depicts the global average rened product

2. Global Renery Congurations

Figure 3: Global rened product slates by renery conguration (2005-2020 average).

Source: KAPSARC Oil Value-Chain Analyzer

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

number of modeling variables required to accurately

specify the appropriate model for a given renery

conguration. The models utilize the larger volume

of operational datasets from all reneries of similar

congurations, capturing the variabilities in design

capacities and uncertainties in operational schemes.

Given the proprietary nature of renery data,

information on the applicable renery conguration

can inform the development of process simulator-

based designs for further assessment studies.

Moreover, conguration-based modeling can drive

best-practice information dissemination among

rening plants operating in similar congurations

through comparative analysis of performances

across operators, countries, and regions. Such

information dissemination can support transparent

and standard renery energy efciency and

emission assessments with a view to guiding

toward opportunities to optimize relative operational

performances focused on efciency, economy, and

emission reduction. Such models are also valuable

for developing realistic energy outlook scenarios

and exploring future potentials for the co-processing

of renewable and hydrocarbon feedstocks in

conventional renery plants.

slates by renery type between 2005 and 2020. In

contrast to simpler renery congurations, more

complex reneries achieve deeper conversion of the

crude feed to yield more lighter to middle distillate

products (such as liqueed petroleum gas (LPG),

naphtha (NAP), gasoline (GAS), kerosene/jet fuel

(KJF) and diesel/gas oil (DGO)) and fewer bottom

distillates and residues (such as heavy fuel oil

(HFO)). Doing so also translates into higher yields of

petrochemical feedstocks (such as LPG and NAP).

The composition of products from complex reneries

enables them to achieve better crack spreads

through higher yields of more valuable products and

the processing of heavier and/or sour crude grades,

which are normally sold at discounts compared

to benchmark lighter and/or sweet crude grades.

At the time of this study, the source database had

not reported other bottom products, such as resid

and coke. Additionally, global production data for

Cong_9, Cong_12, and Cong_19 have not yet

been captured in the database.

The conguration-based modeling of reneries has

several benets. It reduces the modeling task for

large-scale studies encompassing multiregional

and multitemporal analyses. It also reduces the

2. Global Renery Congurations

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

global crude oil renery database is used to

assemble design and operational data for

the 2005-2020 period, covering over 11,000

observations of process unit-level capacities, crude

blend runs and product slate yields (Platts 2022).

The dataset is split into two samples for model

training and testing at proportions of 70% and

30%, respectively.

A nonparametric model is developed by

implementing the extremely randomized tree

regressor (ETR) method in the Python programming

language. Details on the ETR algorithm are

available in (Geurts, Ernst et al. 2006). However, in

this context, for a renery learning data sample of

size N, the application of the algorithm is expressed

as follows:

ls xy









,: ,,1 (1)

where xi is the vector of explanatory variables

(called features) and yi is the vector of

corresponding output variables (called targets). For

the case of n features and p targets, both can be

denoted as follows:

xx x





1,, (2)

yy y





1,, (3)

The sample values of the jth feature organized in

the order of increasing value are

N()

()

1



for features (,

,)jn

1. An innite ensemble of

extremely random trees estimates the value of the

target variable (y

ˆp) in the following form:

"



() = ' … ' 

!"

()



"

 ' 

!"

, 



#

!"



!

(4)

where

n1,,



is the characteristic function of the

hyper-interval, as given by the following:

[, [, ,,

xx xx























[ . 



0, ,Nn

[, [,

xx xx























[ . 



0, ,Nn (5)



n1,,



are real-valued parameters that depend on

examples xi and yi, as well as the parameters for the

minimum sample size for splitting a node (nmin) and

the number of randomly selected features at each

node (K). For our renery model, the set

of features consists of the plant design capacity

(CDU basis), the runs of grades of crude oil

feedstock, as categorized using density and sulfur

contents (see Table 1) into LSW, LSO, MSW,

MSO, HSW and HSO, and the types of renery

congurations. Overall, there are 47 features in the

model, including the plant capacities, the 6 crude

grades and the 40 renery congurations. The

ETR algorithm is implemented using the PyCaret

modeling package in Python (Ali 2020).

3. Material and Methods

Table 1. Categories of crude oil grades based on density and sulfur characteristics.

Crude Grade Density (API) Sulfur (%)

LSW > 34 < 0.6

LSO > 34 > 0.6

MSW > 25 and < 34 < 0.6

MSO > 25 and < 34 > 0.6

HSW < 25 < 0.6

HSO < 25 > 0.6

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

Given the nonparametric nature of the ETR model,

deployment in renery linear programming (LP)

applications faces tractability challenges due to

the uncertain number of operations needed to

obtain the unique optimum solution of the problem.

Consequently, a multivariate linear regression

(MLR) model that furnishes a model for closed-

form mathematical optimization use cases is

also presented. The model is also implemented

in the Python programming language using the

statsmodels library. Using the same dataset and

explanatory variables, the MLR model is estimated

as follows:

! ",!  $,!

$&'

$

(6)

For the categorical variables (xj') representing

renery congurations,

'{,

}





01 (7)

In analyzing the model parameters for Equation 6,

heteroscedasticity is observed to be present due

to the nature of the dataset. Signicant sparsity

results from actual realizations of crude runs and

product yields for the array of crude grades

and product slates. For this reason, a Yeo-

Johnson transformation (Yeo and Johnson

2000) is applied to the formulation to achieve

heteroscedasticity robustness. This transformation

necessitates a reformulation of the model as

follows:

log$&!()+, = ",!

$+/%!,!

%!+ / %,!

%&%!

log$%+,

logl

yxx

jp j









 







 



(8)

For the transformation parameter value,



1

, the

model is heteroscedasticity robust. In addition to

improving homoscedasticity, the transformation

supports desirable attributes in the lower bound

of the variable space required for mathematical

optimization modeling applications. Furthermore,

other statistical tests are performed to assess

normality, autocorrelation, etc. Details on the

statistical tests and results are provided in the

supplementary material (Section SI.C).

3. Material and Methods

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

The models are trained with source data

from the Platts World Renery Database.

The database has approximately 15,000

observations of crude runs and product slates

covering every crude oil renery plant in the world

that operated between 2005 and 2020. Some

of the observations had incomplete information

about either crude runs or product yields, but all

observations reported individual renery capacities.

The database was preprocessed to remove the

incomplete entries, bringing the nal number

of observations for modeling to approximately

11,000. Model training and testing data sample

splits of 70% and 30%, respectively, are applied.

Figures 4 to 9 show parity plots for all the rened

products predicted with the ETR model.

4. Results and Discussion

Figure 4: Parity plot for LPG product prediction with the ETR model.

Source: KAPSARC Oil Value-Chain Analyzer

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

4. Results and Discussion

Source: KAPSARC Oil Value-Chain Analyzer

Figure 5: Parity plot for NAP product prediction with the ETR model.

Source: KAPSARC Oil Value-Chain Analyzer

Figure 6: Parity plot for GAS product prediction with the ETR model.

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

4. Results and Discussion

Figure 7: Parity plot for KJF product prediction with the ETR model.

Source: KAPSARC Oil Value-Chain Analyzer

Figure 8: Parity plot for DGO product prediction with the ETR model.

Source: KAPSARC Oil Value-Chain Analyzer

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

4. Results and Discussion

The coefcient of determination (R2) is above 90%

for all renery products predicted with the ETR

model, meaning that over 90% of the variation

in the target variables is accounted for by the

model. Using the information from the coefcient of

determination for each target variable and the values

of the coefcients of partial determination, which

measures the proportion of total variation in the

target variable that is explained by each individual

feature (explanatory variable) in the model, it is

possible to evaluate and rank all features based on

their importance in predicting the target. Figures 10

to 15 present the top ten features accounting for the

total variation in the predicted rened products. The

gures conrm that the renery capacity, crude oil

blend runs, and renery congurations are essential

variables for quantifying variations in the amounts

of the rened products but with different levels of

importance for each estimation of the volumes of

rened products.

Figure 9: Parity plot for HFO product prediction with the ETR model.

Source: KAPSARC Oil Value-Chain Analyzer

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

4. Results and Discussion

Figure 10: Feature importance plot showing the top 10 features accounting for the total variation in LPG product

prediction.

Source: KAPSARC Oil Value-Chain Analyzer

Figure 11: Feature importance plot showing the top 10 variables accounting for the total variation in NAP product

prediction.

Source: KAPSARC Oil Value-Chain Analyzer

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

4. Results and Discussion

Figure 12: Feature importance plot showing the top 10 variables accounting for the total variation in GAS product

prediction.

Source: KAPSARC Oil Value-Chain Analyzer

Figure 13: Feature importance plot showing the top 10 variables accounting for the total variation in KJF product

prediction.

Source: KAPSARC Oil Value-Chain Analyzer

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

4. Results and Discussion

Figure 14: Feature importance plot showing the top 10 variables accounting for the total variation in DGO product

prediction.

Source: KAPSARC Oil Value-Chain Analyzer

Figure 15: Feature importance plot showing the top 10 variables accounting for the total variation in HFO product

prediction.

Source: KAPSARC Oil Value-Chain Analyzer

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

In addition to the coefcient of determination

(R2), mean absolute error (MAE) and root mean

square error (RMSE) are two other performance

metrics that are evaluated for both the parametric

and nonparametric models of rened products.

Table 2 presents all performance metrics for the

ETR and MLR models. Evidently, the machine

learning-based, nonparametric ETR model

demonstrates higher predictive performance

for the target rened products. For a given

renery conguration, the potential sources of

deviations between the predicted and observed

values of rened product ows could be from

the differences in the process subunit-level

designs, process technology, the operational

expertise of plant operators (human factors), and

model suitability. Consequently, 15 other training

algorithms were applied to the datasets, and

none outperformed the ETR model on all target

rened product predictions. On the R2 basis, the

MLR model shows the lowest performance on

NAP prediction, whereas the ETR model has the

lowest performance on HFO prediction. Additional

statistical analysis results for both models are

available in supplementary information sections

SI.B and SI.C.

Renery conguration-based modeling not only

is useful for estimating the production of rened

products but also can facilitate the quantication of

operational energy consumption, GHG emissions and

economic performance for each type of renery plant

design or conguration. For a given plant capacity

and conguration, variabilities in the properties and

economics of crude oil blend feedstocks determine

the product yield, product types and the overall

renery economic margin performance. However,

the subjects of renery energy requirements and

economics are not the focus of this paper; rather,

they constitute the objectives of future research

leveraging the renery conguration framework and

models presented in this work. By identifying global

renery congurations, which encapsulates the

attributes of each renery plant design in the form of

the conguration variable in the production model, the

approach presented in this paper avoids the need to

explicitly represent and specify design and operating

conditions – such as temperature and pressure – in

the presented production model. The use of actual

eld data on global reneries in the model training

ensures that the effects of design and operational

differences across the renery conguration classes

are captured.

4. Results and Discussion

Table 2. Prediction performance metrics for the test of the ETR and MLR models.

Target

Test Performance – ETR Model Test Performance – MLR Model

MAE RMSE R2MAE RMSE R2

LPG 0.62 1.43 0.95 2.07 3.43 0.77

NAP 1.12 2.72 0.96 4.64 8.53 0.57

GAS 2.53 6.29 0.96 8.32 14.71 0.84

KJF 0.94 2.11 0.97 3.21 5.82 0.83

DGO 2.32 5.34 0.98 6.41 11.43 0.88

HFO 1.93 4.54 0.93 7.46 12.30 0.65

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

Until now, no comprehensive representation

of the global crude oil rening landscape

has been modeled or reported in the

publicly available literature. In this paper, real

renery design and operating information from over

800 existing crude oil reneries globally is used

to establish that there are essentially 40 unique

congurations. On this basis, renery plant data are

used to train a machine learning model to predict

rened product yields given plant capacities,

congurations, and crude blend feedstocks. Due

to the nonparametric form of the machine learning

algorithm and the computational challenges for

renery LP applications, a parametric model

based on MLR is also proposed. Although the

machine learning model demonstrates superior

predictive performance, the parametric model

is convenient for the closed-form representation

of renery models as part of a wider energy

system for various purposes or simply for optimal

decision-making.

In addition to operational scheduling and crack

spread optimization use cases, renery models are

needed in energy outlook and energy transition

scenario analysis. They can provide business and

government decision makers with information on

market opportunities, environmental performance,

and regulatory compliance under specied operating

environments and conditions. This work addresses

the production modeling aspect of renery models.

Extensions of the modeling approach to energy

requirements and emissions constitute further

research opportunities.

5. Conclusions

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

Abella, Jessica, Motazedi Kavan, John Guo, and

Joule Bergerson. 2015. “Petroleum Renery Life Cycle

Inventory Model (PRELIM) PRELIM v1.5.”

Ali, Moez. 2020. “PyCaret: An open source, low-code

machine learning library in Python.”

Geurts, Pierre, Damien Ernst, and Louis Wehenkel. 2006.

“Extremely randomized trees.” Machine Learning 63,

no. 1: 3-42. DOI: https://doi.org/10.1007/

s10994-006-6226-1

Herce, Carlos, Chiara Martini, Marcello Salvio, and Carlo

Toro. 2022. “Energy Performance of Italian Oil Reneries

Based on Mandatory Energy Audits.” Energies 15,

no. 2: 532. DOI: https://doi.org/10.3390/en15020532

Kaiser, Mark. 2017. “A review of renery complexity

applications.” Petroleum Science 14. DOI: https://doi.

org/10.1007/s12182-016-0137-y

Motazedi, Kavan, Jessica Abella, and Joule Bergerson.

2017. “Techno–Economic Evaluation of Technologies to

Mitigate Greenhouse Gas Emissions at North American

Reneries.” Environmental Science & Technology 51,

no. 3: 1918-1928. DOI: https://doi.org/10.1021/acs.

est.6b04606

References

Nduagu, Experience, Evar Umeozor, Alpha Sow, and

Dinara Millington. 2018. “An Economic Assessment of the

International Maritime Organization Sulphur Regulations

on Markets for Canadian Crude Oil.” Canadian Energy

Research Institute.

Ogjresearch. 2022. “Worldwide Renery Survey and

Complexity Analysis.” Oil and Gas Journal. Online

Research Center.

S&P Global Platts. 2022. “Platts World Renery

Database.” Accessed August 15, 2022. https://www.

spglobal.com/commodityinsights/en/products-services/

oil/platts-world-renery-database.

Yeo, In-Kwon, and Richard Johnson. 2000. “A New

Family of Power Transformations to Improve Normality or

Symmetry.” Biometrika 87, no. 4: 954-959. DOI: https://

doi.org/10.1093/biomet/87.4.954

Young, Brett, Troy Hottle, Troy Hawkins, Matt Jamieson,

Greg Cooney, Kavan Motazedi and Joule Bergerson.

2019. “Expansion of the Petroleum Renery Life Cycle

Inventory Model to Support Characterization of a

Full Suite of Commonly Tracked Impact Potentials.”

Environmental Science & Technology 53(4): 2238-2248.

DOI: https://doi.org/10.1021/acs.est.8b05572

Supplementary Information

Global Crude Oil Renery Models:

Parametric and Nonparametric

Methods

Evar Umeozora, †

a King Abdullah Petroleum Studies and Research Center, Riyadh, Saudi Arabia

Email: evar.umeozor@kapsarc.org

Telephone: +966 507 029 871

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

Section SI.A: Summary process

diagrams of the world’s crude oil

reneries

Global Crude Oil Renery Models: Parametric and Nonparametric Methods

Section SI.A: Summary process diagrams of the world’s crude oil reneries