Streamlining Data Integration and Decision Support in Refinery

Operations

Ocan Şahin

, Aslı Yasmal

, Mustafa Oktay Samur

and Gizem Kuşoğlu Kaya

Turkish Petroleum Refinery, Körfez, Kocaeli, 41780, Turkey

Keywords: Digital Twins, Process Optimization, Modelling, Simulation and Architecture, Real-Time Analysis,

Decision Support Systems.

Abstract: Refineries, operating with millions of dollars at stake, face significant economic consequences even with just

30 minutes of non-ideal operation. To address this challenge, this paper presents an industrial application of

seamless integration of two different data sources into a complicated decision support tool that enables

feedforward decisions. The integration is done in Node-RED, facilitating the data flow from two sources

leveraging SOAP calls and COM interfaces in Python to automate the model manipulation, thus generating

live estimates before operation takes place. A dashboard is developed, provides a user-friendly interface for

visualizing the data and making informed decisions on how to increase efficiency and feed the existing model

predictive control architecture. This use-case demonstrates the effectiveness of Node-RED in streamlining

data integration, automation, and decision-making processes in industrial settings is demonstrated,

contributing to improved operational efficiency and profitability in the refinery industry.

1 INTRODUCTION

The refining industry, being one of the oldest, stands

to gain greatly from advancements in technology

through automation and data science in the last

decades. Cracking and treatment units in an oil and

gas refinery are crucial to refining process, where

minor optimizations in their operation can lead to

improvements in overall efficiency and output. A

practical approach to improving the process would be

to focus on the final outcome and adjust the

operational parameters accordingly (Yasmal et al.,

2022; Kaya et al., 2023). One key process we focus

on is the Diesel Hydro Processing (DHP) unit, which

is critical due to its role in the catalytic conversion of

a naphtha and diesel mixture. The DHP unit plays an

essential part in the reactors, where this conversion

occurs, making it a central element in optimizing the

overall process. In this context, diesel

hydroprocessing is an important refining process that

consists of hydrodesulphurization to remove the

unwanted sulfur from the diesel cut. The process

https://orcid.org/0000-0002-5911-8076

https://orcid.org/0000-0002-8080-5591

https://orcid.org/0009-0005-0931-4190

https://orcid.org/0000-0003-2825-0143

consists of hydrocracking and hydrotreating to

produce a diesel product with the required

characteristics (Aydın et al., 2015). Related to this,

the plant model comprises two distinct parts:

hydrodesulphurization (HDS) (Kabe et al., 1999) and

hydrocracking (HC) (Ward, 1993). In accordance

with BS EN regulation (Automotive fuels, 2023), the

use of the HDS process is a key factor in achieving a

product with superior cleanliness and ultra-low sulfur

content, eliminating all negative environmental

impacts such as sulfur dioxide emissions and water

pollution (Safari & Vesali-Naseh, 2018). The

hydrocracking process breaks down hydrocarbon

molecules into lower molecular weight carbon chains.

This is a high temperature process, around 650K to

700K (Park et al., 2018), with strong dependency on

its catalysis’ performance and very difficult to

optimize due to being a black box. When done

correctly, it can convert fuels into high-value

products with a high hydrogen/carbon ratio and low

metal contaminants within one catalysis life cycle

(Rana et al. 2007).

¸Sahin, O., Yasmal, A., Samur, M. and Kaya, G.

Streamlining Data Integration and Decision Support in Reﬁnery Operations.

DOI: 10.5220/0012950500003822

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 21st International Conference on Informatics in Control, Automation and Robotics (ICINCO 2024) - Volume 2, pages 403-409

ISBN: 978-989-758-717-7; ISSN: 2184-2809

403

The aim of this study is to establish links between

a substantial quantity of data sources, implement live

estimation models in an automated manner, and

devise an interactive decision-making assistive

mechanism to the control of hydrocracking reactor

and separation units. The written manuscript is

structured as outlined below: Section 1, problem

description Section 2 presents the methods, system

design, representational state transfer, application

programming interface, historian database server, and

hardware implementation. In Section 3, the

performance and advantages of the experiment, and

the experimental results are described in detail.

Section 4 includes a brief discussion about

operational efficiency, technical and economic

benefits, and potential future directions. Finally,

Section 5 concludes the whole study and provides

potential future research directions.

2 PROBLEM DESCRIPTION

The DHP, which is the plant this study focuses on,

while producing LPG, naphtha as side products,

mainly produces vast amounts of clean and cracked

diesel. It should be noted that highly exothermic

reactions occur in the reactors. Due to the substantial

presence of hydrogen in the system, the highly

exothermic reaction is mitigated through the

introduction of cooling hydrogen quenches. This

cooling process necessitates additional hydrogen to

be fed to the circulation because hydrogen is

consumed. Simultaneously, catalytic reactivity is

maximized to maintain reactor temperatures above a

certain threshold, as higher temperatures enhance

reactivity. However, this cannot be pushed to the

extreme, because higher temperatures not only

increase reactivity, but it also favors multiple

processes that reduce catalysis life cycle such as

faster coke deactivation or support sintering (Gruia,

2006). It is a delicate balance that involves optimizing

the cracking process, ensuring reactor safety,

extending catalyst lifespan, and managing the

optimum hydrogen ratio. This equilibrium, while

crucial for the operation's success and profitability, is

complex to maintain, but the challenges that it brings

with it are also significant

For instance, the plant can encounter issues with

off-spec diesel product accumulating at the base of

the stripper column, resulting in significant product

mismatches. In addition to that, it can also encounter

significantly over-cracked diesel, which means the

catalyst and the reactors are unnecessarily

overworked. To mitigate these issues and maintain

the high quality of diesel produced, it is essential to

implement a feed-adaptive control system. This is

especially important because the characteristics of the

incoming diesel feed often vary, necessitating various

operating conditions to consistently achieve the

desired results.

In response to these problems, the application

described in this work presents a decision support tool

for estimating the remaining operational life of major

plant components. By offering this prediction, the

tool allows plant personnel to solve potential issues

before they have an impact on the final product's

profitability. This proactive strategy ensures that the

facility performs optimally and produces high-quality

products.

3 METHODOLOGY

This study's methodology focuses on integrating real-

time operational data with predictive modeling and

simulation tools to improve decision-making

capabilities in refinery operations. To accomplish

this, we have implemented a multi-faceted approach

involving live data extraction, predictive modeling

using hydroprocessing and separation simulations,

and a flow-based architecture for data handling and

model integration. The system leverages a

combination of SOAP API for secure and structured

data retrieval, MATLAB for hydroprocessing

estimations, commercial simulation software for

separation modeling, and Node-RED for

orchestrating data flows and presenting live results on

a user-accessible dashboard. This integrated

framework supports proactive adjustments in

processing operations, aligning closely with the

dynamic requirements of refinery environments to

improve operational efficiency and reduce potential

risks. The complete flow diagram of the solution can

be seen in Fig. 1.

3.1 Data Connections

Continuous operation required a continuous solution

that can support the decisions made live, so the

challenge ahead was obvious. The first step was to

feed live operational data into accurate models that

run faster than the decision support requirements.

These models are hydroprocessing estimation model

enhanced in MATLAB and separation simulation

models created using commercial simulating

software. To achieve that goal, we opted to use an

application programming interface (API) that utilizes

Simple Object Access Protocol (SOAP) calls. The

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

404

Figure 1: Flow Diagram of the complete application.

SOAP API is a widely used protocol for exchanging

structured information in web services. In the context

of our application, the SOAP API serves as a

communication bridge between the application

developed and the historian database server (World

Wide Web Consortium, 2010). As a large enterprise

we are using SOAP instead of RESTful APIs because

SOAP provides a more structured and standardized

approach that aligns well with our complex enterprise

requirements (Fielding, 2000). SOAP's reliance on

XML ensures consistent data representation, which is

crucial for our integration with diverse systems.

Additionally, SOAP's support for various transport

protocols allows us to seamlessly communicate with

different platforms within our enterprise architecture.

To retrieve the required data, the SOAP request is

constructed from an xml-based envelope and the

secure server parses the envelope, identifies the

action and its parameters, runs the query and prepares

response. The response contains the requested data

prepared by using the parameters supplied, such as

operation data point tag name of the operation data

point in Table 1. (e.g.: 10TIC5.PV), tag name prefix

represents the plant number, this table consists tags

from plant 10 and plant 15. A temperature controller

process value (PV) on first row. A flow controller's

set point (SP) on second row. A pressure indicator's

process value (PV) on third row and the same

pressure indicator’s respective valve opening (OP) on

fourth row. Confidience represents the sureness of the

collected value and very rarely reads something other

than 100 or 0 (100 represents correct, 0 represents

incorrect values). Due to data privacy, we cannot

share real operational data.

Table 1: Dummy Sample Plant 10 and 15 data response.

Timestamp Tag Name Value Confidence

10.01.2023T12:30:00 10TIC5.PV 35°C 100

10.01.2023T12:30:00 10FIC2.SP 500

100

10.01.2023T12:30:00 15PI12.PV 12

bar

100

10.01.2023T12:30:00 15PI12.OP 25% 0

To achieve the level of predictive control we

needed, feeding live operational data into our models

was not enough. A traditional feedback loop makes

adjustments in response to system output, which can

cause delays and inefficiencies. The goal was a feed-

forward system that would proactively adjust on the

basis of input data before problems occurred. Hence,

an online analyzer was installed to the input feed of

the system to model the chemical properties of the

incoming liquid. Analyzer uses Near-Infrared (NIR)

(Falla et al., 2006) spectroscopy technology. It

estimates feed total boiling point using the internally

developed statistical models and sends the data to its

own on-site computer. From behind the firewall, a

trivial bat script writes the data to an intermediate

server that has one-directional communication with

the analyzer computer and to our solution server.

Although the analyzer can take measurements once

every couple of seconds, it is set to work once every

five minutes. Results are sent to the solution server

within a similar frequency. This is due to the time

required to run the other models reliably. There was

no need to create more input data if the models cannot

run fast enough.

Streamlining Data Integration and Decision Support in Reﬁnery Operations

405

3.2 Hydro Processing Estimation

Model

The solution models’ first part consists of

distinct hydrodesulphurization calculations and

hydrocracking calculations. While we are modelling

the reactor, due to the different bed types and bed

lengths of the reactors, each bed has been considered

as a separate reactor in series and calculations have

been made accordingly. In the HDS reactor, sulfur

compounds are removed from the feed based on the

sulfur specification. Cracking of the feed carries out

in the HC reactor. One of the underlying purposes of

this study is to keep the boiling point of the product

within the desired values of the unit. In the MATLAB

part (The MathWorks Inc., 2022), models are

constructed to predict the composition of reactor exit

and reactor bed exit temperatures. Certain properties

of the feed are used as inputs. These include TBP

values at different temperatures, flow rate, sulfur

content, and bed inlet temperatures. During the

calculations, some assumptions are made for both

HDS and HC reactions. In the HDS beds, it is

assumed that the formation and effects of Hydrogen

Sulfide (H

S) are ignored, and no cracking reactions

occur. For both reactors, it is assumed that reactions

are adiabatic, homogeneous, and liquid phase,

reactions are first order, heat capacities of

components are constant, and activation energies for

the beds are constant. For this estimation, feed

properties are used, and the pseudo-true boiling point

is calculated.

Feed characterization data is taken from the

summary day of each month. Summary day is a

special day where extra examples are taken from the

plant to be analysed in the laboratory, like an offline

snapshot of the operation. Using this data, the cost

function is tried to be minimized. The cost function

for optimizing the kinetic parameters considers the

reactor bed outlet temperatures and the weight

fractions of the reactor effluent. Thus, optimum

parameters are determined through an iterative study

for each month.

The models are fed 32 different tags, including the

estimated incoming feed characteristics combined

with the current operating conditions of the plant.

From these 32 tags, 10 represents the incoming feed

characteristics; however, the other 22 tags are

selected after careful field tests to encapsulate the

maximum amount of operational meaning while

using the least amount of tag load possible. They are

pulled for the last 15 minutes of operation and the

mean values of the 15 minutes are used in our models.

The model successfully estimates the creation of the

major products of the plant, but because of the black

box nature of the system, results cannot be validated

until the products are separated from each other. From

the first model results, only the reactor temperatures

are something that can be meaningful to use, the

cracked diesel compositions consist of 145 features

cannot be used anywhere before second model runs.

These temperature values can be further used to fine-

tune the overall solution in an iterative way to find the

optimal operating conditions for the desired outcome.

The MATLAB model is compiled using the

MATLAB Compiler Toolbox, and it is hosted as a

web application using Microsoft Internet Information

Services (IIS) (The MathWorks Inc., 2022). We opted

for this approach because it meant similar HTTP

requests for both models.

3.3 Separation Simulation Model

The solution models’ second part consists of

separation processes that are simulated using

commercial simulation software. The simulator

model consists of three separator columns and

estimates the separation, and thus it creates DHP

unit’s four different end products. Results from the

first model are fed to the second model as significant

inputs and the remaining operation parameters that

the model requires are pulled with a similar simple

SOAP call as explained previously. Because of this,

22 different features are pulled from the historian on

top of the 145 features coming from the first model.

A Python script utilizing the simulator’s COM

interface existing on its backend was used to feed

these features to the simulator. The script consists of

4 different sections and employs the Win32com

library to manipulate simulator classes and objects.

Figure 2: Live dashboard of the solution.

The application runs with the simulator already

opened in the operating system, so the first section

does not open the simulator but only attaches to the

respective process, finding the flowsheet of the

simulator and assigning individual stream and

equipment to respective Python variables. Then, in

the second section, the output of the first model,

which is the input of the second model, is

characterized as an oil mixture using the 145 features

mentioned earlier. After that, it is attached to the input

stream in the simulation flowsheet. The paused

simulation is run to steady state in Section 3, and any

errors are caught and handled in this section. If the

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

406

simulation breaks due to any reason, such as bad data

or a momentary server downtime, the simulation file

is re-launched from a safe point for the next run in 15

minutes. And finally, in Section 4, the end results of

the separator columns (temperatures and pressures

mainly) and the product specifications that are

significant for the daily operation are read from the

simulator and fed back to Node-RED flow as a JSON

file.

Though navigating the object-oriented topology

of the simulator was challenging, the COM interface

provided a crucial role in automating the software and

streamlining task management. By automating this

part of the solution, we were able to eliminate the

most repetitive manual section. In addition, since the

Python script manipulates the simulation and runs it

to a steady state, we can read any important results

from the simulator and send them back to the Node-

RED flow as a JSON file. This operation takes less

than 30 seconds, and the same simulator file can be

used indefinitely if the simulation does not break. In

which case, the simulation file is closed without being

saved, reopened, and restarted from a safe point.

The simulator model is hosted as a web

application using flask framework. Flask was chosen

because it requires no additional tools and offers

simplicity and flexibility through the implementation

of a minimal web server. (Flask Documentation

User's Guide, 2010).

3.4 Node-RED Flow

Node-RED is an open-source, versatile, flow-based

development ETL tool (Rymaszewska et. al., 2017).

JavaScript based interface, modifiable and flexible

nature, low overhead are why Node-RED was used in

this study. The explained communications in

previous sections, besides the analyzer data

communication, all occur in the respective Node-

RED flow by the help of different processes.

Analyzer data is directly written to a specified

directory in the Node-RED server by a bat script

operating on its own computer. The latest written csv

file in the analyzer directory is read by a node in the

flow. SOAP calls that bring the process data to the

flow is handled within a simple Python process for

easier code maintenance. Both are combined to

generate the total input data.

The Node-RED flow uses the input data in HTML

request nodes to send requests to the models that are

hosted at a local IP address as flask applications.

Results are sent back to the Node-RED flow as

JSONs. The direct estimated results that have

significant and urgent operational meaning are fed to

a simple Node-RED dashboard to be used as a

decision support measure. The complete flow is on a

loop that repeats itself every five minutes, thus the

dashboard refreshes itself every five minutes. Four

product specifications can be seen in color coded

gauge graphs showing the operation engineers the

estimated outcome of the current state (Fig. 2.).

The dashboard is hosted at a specific IP address

and port that can be accessed by the authorized users

on the company intranet. With this method, we aimed

to let users see the live results directly. The entire

results are written to a SQL database for further use

and additional statistical analysis applications.

4 RESULTS AND DISCUSSION

This study demonstrated that we achieved our

objective of creating a system that helps the operation

make accurate decisions before any operational errors

happen. Operating at 5-minute frequencies, the

solution works with the previous 5-minute average

unit data that has been collected from temperature,

flow, and pressure indicators. The frequency was

chosen as 5 minutes because the unit's operating

procedures and operational parameters cannot change

significantly faster than 5 minutes. Also, the overall

flow of the solution (Fig.2) runs for 2 minutes before

writing its results, so it could not run faster than that.

However, 5 minutes was a balanced midpoint, as

running the model more frequently would not provide

any benefit.

Using live data streams in combination with

predictive modelling can greatly improve the

effectiveness of operations too. The properties of the

feed, which include but are not limited to True

Boiling Point, are fully utilized through the decision

support tool to correct possible deviations in process

variation before they occur. This is in great contrast

to the backward mechanisms of feedback that mostly

respond too late to avoid losses. An easy-to-use

dashboard allows plant operators to make fast,

informed decisions, reducing out-of-specification

products and unnecessary reactor loadings.

Strong data management in the system is due to

flow development by Node-RED and proper, secure

data movement structure by SOAP APIs. The

intuitive interface of Node-RED made the co-

ordination possible with varying sources of data to

run complex decision-making automated smoothly

and efficiently. The choice of technologies, such as

Node-RED, is based on the specific operational

requirements of the organization technology

structure, ensuring system efficiency and reliability.

Streamlining Data Integration and Decision Support in Reﬁnery Operations

407

Collecting all available data further enabled

statistical modelling, data analysis, machine learning,

and physical modelling work. One can exploit similar

ease-of-access to live data to transfer their offline

practices to an online context. In addition, by

connecting the physical models we have developed to

live data and automating them, we have enabled

further simulations or mathematical models to work

with live data with the methods we have developed

in-house automatically. In addition, by linking our

physical models to live data and automating them, we

have facilitated the use of additional simulations or

mathematical models with live data automatically

using our proprietary methods. Given that

simulations and models integrated with live data or

databases are often sold commercially as separate

packages or licences, this capability represents a

significant economic advantage of our approach.

Several issues arose, mainly, how to ensure that

models could run faster than the decision support

requirements and how complex data flows could be

managed securely. For instance, an online analyzer

was installed to model the chemical properties of the

feed coming in, and an intermediate server was also

installed for the secure handling of the data. Such is

the kind of careful planning that has gone into

building a balance between real-time processing

capacity and data surety (Aldoseri et. al.,2023).

Economically, massive savings could be realized

through real-time optimization of DHP unit

operations by minimizing off-spec diesel and

extending catalyst life (Aydin, 2015). Even hydrogen

consumption is lowered under optimal conditions in

the reactor, producing further decreases in operational

costs. Environmentally, more controlled sulfur

removal processes yield diesel products that meet and

surpass-stringent environmental regulations, thereby

limiting harmful emissions and producing greater

sustainability.

Forthcoming, future research efforts might focus

on refining the models to further enhance accuracy

and improve the response times to higher levels.

Predicting long-term trends in addition to predicting

trends of potential issues would be a big added value

toward the decision support. Further enhancement of

the system to include interaction with other units

within the refinery could provide a more

comprehensive approach toward the refinery

optimization by extending the benefits realized in the

DHP unit across the facility.

5 CONCLUSIONS

In this paper, we tried to describe the automated

application we developed by combining multiple

different software and data sources to improve and

support the current operation and reduce potential

errors by giving them the ability to react before errors

occur. We developed a data connection to two

different data sources through the unit firewall to our

server using SOAP calls and a simple bat script to

access the data. By feeding the pre-processed

versions of this data to the two models we developed

in MATLAB and using commercial process

simulators, we produced results to predict the course

of the current operation. The complete connection

between data points, and models and databases are

done via the open-source project, Node-RED and we

automated commercial simulators using the COM

interface of the Windows operating system in Python

and delivered live results to users in Node-RED

interfaces.

In summary, with this decision support system,

unit engineers will be able to make more controlled

interventions, intervene with prior knowledge of

product characteristics, operate in a manner that is

more aligned with maintenance schedules, and follow

production planning objectives.

REFERENCES

Aldoseri, A., Al-Khalifa, K. N., & Hamouda, A. M. (2023).

Re-Thinking Data Strategy and Integration for

Artificial intelligence: Concepts, opportunities, and

challenges. Applied Sciences, 13(12), 7082.

https://doi.org/10.3390/app13127082

Aydın, E., et al. “Dynamic modeling of an industrial diesel

hydroprocessing plant by the method of continuous

lumping,” Computers & Chemical Engineering,

vol. 82, pp. 44–54, 2015, doi: 10.1016/j.

compchemeng.2015.06.005.

Aydin, Erdal. (2015). Plant-Wide Modeling, Optimization

and Control of an Industrial Diesel Hydroprocessing

Plant. 10.13140/RG.2.2.23698.86725.

A. Gruia, Hydrotreating, in: D.J.S. Jones, P. Pujadó (Eds.)

Handbook of Petroleum Processing, Springer,

Netherlands, 2006, pp. 330-347.

A. Rymaszewska, P. Helo, and A. Gunasekaran, “IoT

powered servitization of manufacturing – an

exploratory case study,” International Journal of

Production Economics, vol. 192, pp. 92–105, 2017, doi:

10.1016/j.ijpe.2017.02.016.

A. Safari and M. Vesali-Naseh, “Design and optimization

of hydrodesulfurization process for liquefied petroleum

gases,” Journal of Cleaner Production, vol. 220, pp.

1255–1264, 2019, doi: 10.1016/j.jclepro.2019.02.226.

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

408

A. Yasmal, G. K. Kaya, E. Oktay, C. Çölmekci, and E.

Uzunlar, “Data Driven Leak Detection in a Real Heat

Exchanger in an Oil Refinery,” in Computer Aided

Chemical Engineering, vol. 52, Elsevier, 2023, pp. 3091–

3096. doi: 10.1016/B978-0-443-15274-0.50493-5.

F. S. Falla, C. Larini, G. A. C. Le Roux, F. H. Quina, L. F.

L. Moro, and C. A. O. Nascimento, “Characterization

of crude petroleum by NIR,” Journal of Petroleum

Science and Engineering, vol. 51, no. 1–2, pp. 127–137,

2006, doi: 10.1016/j.petrol.2005.11.014.

Flask Documentation User’s Guide (2024) Flask

Documentation (3.0.x). Available at: https://flask.

palletsprojects.com/ (Accessed: 24 June 2024).

G. K. Kaya, P. Döloğlu, Ç. O. Özer, O. Şahin, A. Palazoğlu,

and M. Külahçı, “A Study of Spectral Envelope Method

for Multi-Cause Diagnosis using Industrial Data,” in

Computer Aided Chemical Engineering, vol. 50,

Elsevier, 2021, pp. 1331–1337. doi: 10.1016/B978-0-

323-88506-5.50205-9.

Halili, F., & Ramadani, E. (2018c). Web services: A

comparison of soap and rest services. Modern Applied

Science, 12(3), 175. https://doi.org/10.5539/mas.

v12n3p175

H.-B. Park, K.-D. Kim, and Y.-K. Lee, “Promoting

asphaltene conversion by tetralin for hydrocracking of

petroleum pitch,” Fuel, vol. 222, pp. 105–113, 2018,

doi: 10.1016/j.fuel.2018.02.154.

J. W. Ward, “Hydrocracking processes and catalysts,” Fuel

Processing Technology, vol. 35, no. 1–2, pp. 55–85,

1993, doi: 10.1016/0378-3820(93)90085-I.

M. S. Rana, V. Sámano, J. Ancheyta, and J. A. I. Diaz, “A

review of recent advances on process technologies for

upgrading of heavy oils and residua,” Fuel, vol. 86, no.

9, pp. 1216–1231, 2007, doi: 10.1016/j.fuel.

2006.08.004.

“Automotive fuels. Unleaded petrol. Requirements and Test

Methods,” knowledge.bsigroup.com, Apr. 30, 2023.

https://knowledge.bsigroup.com/products/automotive

-fuels-unleaded-petrol-requirements-and-test-methods-5

?version=standard&tab=preview

R. T. Fielding, “Architectural styles and the design of

network-based software architectures,” Ph.D.

dissertation, Inf. Comput. Sci., Univ. California, Irvine,

CA, USA, 2000.

T. Kabe, A. Ishihara, and W. Qian, “Hydrodesulfurization

and hydrodenitrogenation. Chemistry and

engineering,” Jul. 1999, Accessed: Oct. 19, 2023.

[Online]. Available: https://www.osti.gov/etdeweb/

biblio/20101494

The MathWorks, Inc. (2022). MATLAB version: 9.13.0

(R2022b). Accessed: January 01, 2023. Available:

https://www.mathworks.com

Word Wide Web Consortium, SOAP Version 1.2, W3C

recommendation, second ed., http://www.w3.org/

TR/soap, Feb. 2010.

APPENDIX

Figure 3.

Streamlining Data Integration and Decision Support in Reﬁnery Operations

409