ABSTRACT/ SUMMARY

Liquid loading being defined as operational constraint associated with production in gas wells

which becomes the most common operational problem associated with gas production in gas

wells. Liquid loading is characterized with the reversal flow of liquid film into the wellbore,

hence diminishing the gas production rate and subsequently affecting the production cost.

Although, liquid loading and critical velocity have been studied extensively, but the results

from previous investigation and the most commonly applied model in the industry still has a

high degree of inaccuracy, especially in predicting the minimum gas flow rate required to

prevent liquid loading into the wellbore. This research utilizes the knowledge of essemble

machine learning technique to investigate the possibility of its application in the prediction of

liquid loading by determining critical velocity through thorough diagnostic analysis of well

data.
 CHAPTER ONE

1.0 INTRODUCTION

1.1 Background Information

Liquid loading phenomenon is known as the inability of the produced gas to carry all the co-

produced liquid to the surface. Under such condition, the non-removed liquid accumulates at

the wellbore resulting in reduction of the production and sometimes causes the death of the

well (Vieira & Stanko, 2019). Characterized with the reversal flow of liquid film into the

wellbore, liquid loading is a genuine issue for gas wells as it diminishes the gas production

rate. In the event that fluid rate accumulating in the wellbore is excessively high, the gas

production rate will decrease fundamentally and for severe instances of accumulation, the

operating organization will relinquish the well which will cause immense budgetary

misfortunes. Subsequently, so as to maintain a strategic distance from the latter occurring, it is

proper for the working organization to anticipate and recognize the liquid loading status of the

gas wells in order to utilize viable apparatuses and pathways to avert it. Therefore, to

counteract those misfortunes, the forecast of the liquid reversal point is obligatory (Landjobo

& Wu, 2020).

Initially, shallow gas wells have low flowing bottomhole pressure. When liquid introduces

itself in the wellbore, bottomhole pressure starts to increase. Subsequently, gas velocity

decreases and reaches its critical rate, which below this rate liquid loading process expands in

the wellbore and the flow converts from unloaded to loaded regime.
Figure 1: Liquid Loading Steps in the Well

Source: Jafarov & Al-Nuaim, 2016

Figure 1 schematically illustrates the liquid loading process. As seen from the figure, liquid

rate in gas flow increases from mist to bubble by decreasing gas velocity. The gas well can

enter to one or all of these regimes of its life. Consequently, tubing filled with the liquid

increases bottomhole pressure and reduces production and ultimate recovery. Observation of

onset of liquid loading in most cases is very difficult, because loaded well may continue to

produce for long period without any symptoms. It can be determined through several ways in

later stages of production life of the well.

Liquid loading will not always lead to non-production. If a well is loaded, it still may produce

for long time. If liquid loading is recognized and reduced, higher producing rates are achieved.

Symptoms indicating liquid loading include the following:

i. Sharp drops in liquid decline curve

 ii. One of the most common methods to detect liquid loading is to measure the gas flow

rate through an orifice over time.

iii. Increasing difference between the tubing and casing flowing pressure (i.e., Pcf – Ptf)

with time, measurable without packer’s present

iv. Sharp changes in gradient on a flowing-pressure survey

1.1.1 Liquid Loading Mitigation Methods

Various methods can be used to mitigate liquid loading. Some of them include:

a. Venting: In this method, the surface pressure is reduced to atmospheric such that

the pressure differential between the bottomhole and the surface is increased, which

would increase the flow velocity.

b. Soaping: This involves the introduction of a surfactant at the bottomhole to induce

foaming of the liquids, which would make it easier to transport up the well.

c. Velocity Strings (Tubing Replacement): This involves replacing the tubing with a

smaller diameter tubing to increase the gas velocity.

d. Plunger Lift: A plunger pump is used to unload liquids from the bottomhole.

1.1.2 Critical Gas Flow Rate

The onset of liquid loading in gas well can be calculated based on Liquid droplet transport

model. The droplet model for “critical velocity” or “critical rate” is based on the fact that in

mist flow, two forces act on liquid droplet – the drag force and the gravity force. With

decreasing reservoir pressure, the drag force declines and once it balances with the gravity

forces, a liquid particle would “float” (not move) in the gas stream. From this point onwards

the well starts to liquid load. The problem of gas well loading is serious one in gas wells. Due
to lack of sufficient energy to lift the produced fluids, liquid may accumulate in the wellbore,

and impose additional hydrostatic pressure on the reservoir and causes more reduction in the

transport energy. Thus, the accumulated liquids may completely kill the well or liquid

slugging or churning may take place and gives more chance for liquid accumulation,

eventually, the well may die. To solve such a problem, there are several methods of preventing

liquid load up.

The first technique is to maintain the production rate from the gas well above its critical gas

flow rate. The critical gas flow rat

e is the flow rate below which some of the liquid cannot be lifted to the surface; so, it

accumulates in the production string. Many authors studied this critical flow rate of critical

velocity. Analysis of research conducted in this area indicated that it is possible to prevent

load-up if the gas production rate (velocity) maintained above its critical velocity. This could

be achieved by selecting proper tubing size or maintaining low wellhead pressure. Predicting

that minimum gas flow rate is very crucial.

1.1.3 Liquid Loading Prediction Models

Several researchers have developed various mathematical models to calculate the critical flow

rate necessary to keep gas wells unloaded and/or predict liquid loading occurrence in future.

Many different models have been developed to deal with this problem and forecast future

performance of gas wells. Some of which are:

i. Liquid Droplet Reversal Models

Liquid droplet reversal models are based on the flow reversal of liquid drops toward the

wellbore, which infers that it is the source of the liquid loading. The most prevalent model is

that developed by Turner et al. in 1969, who established their model by applying Newton’s

law to the drag force, the buoyancy, and the gravity force acting on a stationary spherical

droplet, and by implementing the Hinze model of 1955 under a critical Weber number ranging

between 20 and 30. During the early 1990s, several researchers made modifications to the

model developed by Turner et al. in 1969 and therefore built other models (Arnold Landjobo

Pagou, 2020).

Turner’s Model: Turner et al. in 1969 derived a method of predicting the critical gas rate to

suspend a liquid droplet by equating drag forces with gravity forces. The Turner equation

further incorporated about 20% upward adjustment to fit relevant field data (Malik, Prakash,

Kumar, & Barot, 2017). For the removal of gas well liquids, Turner et al. (1969) proposed two

physical models: (1) the continuous film model (2) and entrained drop movement model.

ii. Liquid Film Reversal Models

Unlike the liquid drop model, the liquid film model suggests that liquid loading originates

from the reversal of a liquid film attached to the tubing wall (Veeken K., 2010); (Yonghui Liu,

2018); (Shekhar, Kelkar, Hearn, & Hain, 2017). Subsequently, through several experiments,

scientists proved that the transition from an annular regime to a slug or churn flow regime

initiates the loading phenomenon. While investigating the influence of a film on the flow

regime, Barnea in 1986-1987 found that the unsteadiness of a liquid film will cause a blockage

of the gas core, thus generating a change from an annular regime into a slug flow regime.

Furthermore, the author developed a model to predict the gas void fraction, and
experimentally found it to range between 0.65 and 1. Finally, the author developed a model

for both inclined and vertical gas wells based on the momentum balances by assuming a

uniform film thickness around the entire inner pipe wall (Arnold Landjobo Pagou, 2020).

1.1.4 Machine Learning

According to Wikipedia, Machine learning (ML) is a field devoted to understanding and

building methods that let machines "learn" – that is, methods that leverage data to improve

computer performance on some set of tasks. It is seen as a broad subfield of artificial

intelligence.

Machine learning algorithms build a model based on sample data, known as training data, in

order to make predictions or decisions without being explicitly programmed to do so.

Machine learning algorithms are used in a wide variety of applications, such as in engineering,

medicine, email filtering, speech recognition, agriculture, and computer vision, where it is

difficult or unfeasible to develop conventional algorithms to perform the needed tasks.

1.1.5 Types of Machine Learning

Machine learning approaches are traditionally divided into three broad categories, which

correspond to learning paradigms, depending on the nature of the "signal" or "feedback"

available to the learning system:

 Supervised learning: The computer is presented with example inputs and their desired

outputs, given by a "teacher", and the goal is to learn a general rule that maps inputs to

outputs.
 Unsupervised learning: No labels are given to the learning algorithm, leaving it on its own

to find structure in its input. Unsupervised learning can be a goal in itself (discovering

hidden patterns in data) or a means towards an end (feature learning).

 Reinforcement learning: A computer program interacts with a dynamic environment in

which it must perform a certain goal (such as driving a vehicle or playing a game against

an opponent). As it navigates its problem space, the program is provided feedback that's

analogous to rewards, which it tries to maximize.

1.2 Problem Statement

Liquid loading is considered as major issue for wells delivering petroleum gas as it diminishes

and/or even stops production. When the liquid loading phenomenon occurs in gas wells, the

fluid accumulating in the wellbore will initiate a persistent increment of the formation

backpressure, thus generating a decrease in gas production rate. In the event that the

aggregating fluid attains a great volume, the gas production may stop, and if the gathered fluid

is not removed on schedule, the well will be flooded and abandoned causing a huge financial

loss to the producing company. Consequently, identifying and predicting liquid loading earlier

could be an advantage as it would sanction us to take felicitous actions to evade or lessen that

unbeneficial outcome. Most of the current methods for predicting liquid loading in gas wells

are mathematic approaches which are time consuming; financially cumbersome and specific to

a given type of well (either vertical, horizontal or inclined) therefore cannot be applied to

other well.

Many different models have been developed to deal with this problem and forecast future

performance of gas wells. All these models are either on theory or empirically based on

experiments.
1.2 Aim and Objectives

To investigate the possibility of applying ensemble machine learning technique in predicting

liquid loading in gas well

The objectives to achieve this aim are as follows;

i. To develop a simplified machine learning algorithm using ensemble machine learning

techniques

ii. To use simulation data of different types of well to test the validity of the developed

algorithm

iii. Model validation using existing models

1.4 Justification

Artificial intelligence and machine learning are the most emerging technologies that are

dominating the oil and gas industry and therefore the need to apply it in solving real time

problem and preventing future loses in the petroleum production industry.

1.5 Scope and Limitations

The scope of this research is to investigate the applicability of ensemble machine learning

technique in the prediction of liquid loading and is limited to the investigation of the

application of Ensemble method of the machine learning using both MatLab and Python

software and to validate the result obtained with other existing models.
1.6 Research Gap

Although machine learning technique is one of the leading technology in the oil and gas

industry today, the static aspect of liquid loading such as the onset of liquid slugs at the

surface of the well were extensively studied. No much work has been done in the investigation

of the combine effect of the dynamic aspect such as inclination angle, flow regime transitions

and tubing outflow behavior amongst others, as such the need to investigate the applicability

of the ensemble machine learning technique in predicting liquid loading in gas wells.
 CHAPTER TWO

2.0 LITERATURE REVIEW

2.1 Liquid Loading

Liquid loading is defined as the flow reversal of liquid films into wellbores. It is a

severe issue for gas wells because it decreases the production rate of gas. If the rate of

accumulation in the wellbore is extremely high, the production rate will significantly

decrease, and, under extreme cases of accumulation the operating company will

abandon the well, leading to substantial financial losses. Consequently, to avoid this, it

is appropriate for the operating company to predict and identify the liquid loading

status of the gas wells and use practical tools and pathways to prevent such a loading

(Arnold Landjobo Pagou, 2020). Several studies on liquid loading have been

published, including the two most well-known liquid loading models developed by

Turner et al. (1969) and Coleman et al. (1991). These models were developed using a

liquid droplet mechanism, whereby the minimum gas velocity to carry the liquid

(critical gas velocity) is based on the equilibrium between the drag force applied to the

liquid droplets and the gravity force.

Brito et al. (2005) studied the liquid film reversal mechanism to develop a robust

model for predicting liquid loading instead of using the liquid droplet mechanism.

They postulate that liquid loading occurs when the shear stress exerted by the gas flow

at the liquid-gas interface is insufficient to generate sufficient drag force to carry the

liquid film surrounding the pipe wall. This model predicts the in-situ critical gas

velocity along the wellbore, which allows us to determine the liquid loading locations.
It requires several data inputs, such as gas and liquid production rates, well geometry,

production tubing and casing diameters, and fluid properties. It shows that liquid

loading potentially occurs at 1,400 - 10,600 ft MD, as the actual in-situ critical gas

velocity is lower than the in-situ critical gas velocity at this depth. It is necessary to

predict when the liquid loading will start happening, known as the liquid loading onset,

in order to develop a liquid loading management strategy. This can be achieved by

performing the liquid loading analysis with various gas production rates based on the

production decline analysis. The liquid loading onset can be predicted by determining

the time when the actual in situ gas velocity drops to the in-situ critical gas velocity

(Wenq K., 2021).

2.1.1 Signs of Liquid Loading in Gas Wells

Liquid loading is not always obvious. If a well is loaded, it may still produce at a lower

rate that is not favorable to the production company. If recognized and reduced, higher

production rates are achieved. Symptoms indicating liquid loading in a well includes:

a. Sharp drops in a decline curve

b. Onset of liquid slugs at the surface of the well

c. Increasing difference between the tubing and casing flowing pressures (i.e Pcf-Ptf)

with time measurable without parkers present

d. Sharp changes in gradient on flowing pressure survey

2.2 Critical Gas Velocity

The accumulation of the liquid in the wellbore will cause a decline in the well

production rate or the well might cease to flow. To avoid liquid loading in gas wells,

the well should be produced at or exceed a certain minimum rate. This particular rate is

termed as the Critical Gas Rate, which is defined as Minimum Gas Rate required to lift

the produced condensate liquid or water to the surface without liquid accumulation

downhole. There are several correlations to estimate the critical gas rate and most are

based on the pioneering work proposed by Turner (1969), Coleman et al. (1991),

and Li et al. (2001).

The minimum gas velocity equation is:

1 /4
υ g=1.593 δ ¿¿ ..……………………. 1

The minimum gas flow rate equation is:

P υg A
q g=3.067 ………………………….. 2
TZ

A = Flow Area ft2

P = Wellhead Pressure Psia

qg = Gas Rate MMscf/d

ρg = Gas Density lbm/ft3

ρl = Liquid Density lbm/ft3

σ = Surface Tension dyne/cm

T = Flowing Temperature ºR
Υg = Gas Velocity ft/sec

Z = Gas Compressibility Factor (Dimensionless)

2.3. Prediction Models for Critical Velocity

Liquid load-up in gas wells is not always obvious; therefore, a thorough diagnostic

analysis of well data needs to be carried out to adequately predict the rate at which

liquids will accumulate in the well. Although this subject has been studied extensively

but the results from previous investigators and the most commonly applied model in

the industry still has a high degree of inaccuracy, especially in predicting the minimum

gas flow rate required to prevent liquid loading into the well bore (Princewell

Maduabuchi, 2018).

2.3.1 Turner model

Turner, Hubbard, and Dukler, observations, proposed two physical models for the

removal of gas well liquids. The models are based on: (1) the liquid film movement

along the walls of the pipe and (2) the liquid droplets entrained in the high velocity gas

core. They used field data to validate each of the models and concluded that the

entrained droplet model could better predict the minimum rate required to lift liquids

from gas wells. This is because the film model does not provide a clear definition

between adequate and inadequate rates as satisfied by the entrained droplet model

when it is compared with field data. A flow rate is determined adequate if the observed

rate is higher than what the model predicts and inadequate if otherwise. Again, the film

model indicates that the minimum lift velocity depends upon the gas-liquid ratio while
no such dependence exists in the range of liquid production associated with field data

from most of the gas wells (1 - 130 bbl/MMSCF).

Figure 2, Turner et al. model calculated minimum flow rates mapped against the test

flow rates (Guo et al. 2018)

Turner et al. (1969) discovered that the drop model prediction in most cases was too

low and he blamed this on the values of critical weber number and drag coefficient

used in the development of the model and also on the fact that the mathematical

development predicts stagnation velocity, which must be exceeded by some finite

quantity to guarantee removal of the largest droplets. Analysis of the Turner’s data

reveals that the total contribution of these factors requires an upward adjustment of

approximately 20%. Mapping Turner’s calculated model against actual test data

reveals large discrepancies at lower flowrates; hence adjusted droplet model is given

by the following equation to accommodate these discrepancies:

1 /4
υ crit−T =1.92 δ ¿ ¿ ………………………….. 3

3060 P V crit−T A
q crit−T = ……………………………..4
TZ
A = Flow Area ft2

P = Wellhead Pressure Psia

V= Volume M3

qcrit-T = Gas Rate at critical Temperature MMscf/d

ρg = Gas Density lbm/ft3

ρl = Liquid Density lbm/ft3

σ = Surface Tension dyne/cm

T = Flowing Temperature ºR

Z = Gas Compressibility Factor (Dimensionless)

2.3.2 Coleman model

Using the Turner model but validating with field data of lower reservoir and wellhead

flowing pressures all below approximately 500 psia, Coleman et al. were convinced

that a better prediction could be achieved without a 20% upward adjustment to fit field

data with the following expressions:

1/ 4
υ crit−C =1.593 δ ¿ ¿ ………………………5

3060 P V crit−C A
q crit−T = ……………………………….6
TZ

2.3.3 Li’s model

Li et al. (2001) in their research posited that Turner and Coleman’s models did not

consider deformation of the free falling liquid droplet in a gas medium, and

furthermore, contended that as a liquid droplet is entrained in a high velocity gas

stream, a pressure difference exists between the fore and aft portions of the droplet

leading to its deformation and its shape changes from spherical to convex bean with

unequal sides (flat) as shown in figure 3. Spherical liquid droplets have a smaller

efficient area and need a higher terminal velocity and critical rate to lift them to the

surface. However, flat-shaped droplets have a more efficient area and are easier to be

carried to the wellhead.

Fig. 3, Flat- Shaped droplet model and Shape of entrained moving in high velocity gas

(Li et al. 2001)

This lead to the formulation of the following expression:

1/ 4
υ crit−L =0.7241 δ ¿ ¿ …………………….7
 3060 P V crit− L A
q crit−T = ………………………………8
TZ

2.3.4 Nosseir’s model

Nosseir et al. (2000) focused his study mainly on the impact of flow regime and

changes in flow conditions on gas well loading. Their work was similar to that of

Turner but the difference between the both was that Turner did not consider the effect

of flow regime on the drag coefficient (used in the derivation stage), and thereby

making use of the same drag coefficient (0.44) for laminar, transient and turbulent

flows. Nosseir derived the critical flow equations by assuming drag coefficient value

of 0.44 for Reynolds number (Re) 2×10 5 to 104 and for Re value greater than 10 6 he

took the drag coefficient value to be 0.2. Representation of the critical velocity

equation by Nosseir’s model is summarized as:

1/ 4
υ crit−N =1.938 δ ¿ ¿ …………………….9

3060 P V crit−N A
q crit−T = ……………………………..10
TZ

2.4 Machine Learning

According to Arthur Samuel Machine learning is defined as the field of study that gives

computers the ability to learn without being explicitly programmed. Arthur Samuel was

famous for his checkers playing program. Machine learning (ML) is used to teach machines

how to handle the data more efficiently. Sometimes after viewing the data, we cannot interpret

the extract information from the data. In that case, we apply machine learning. With the

abundance of datasets available, the demand for machine learning is in rise. Many industries
apply machine learning to extract relevant data. The purpose of machine learning is to learn

from the data. Many studies have been done on how to make machines learn by themselves

without being explicitly programmed. Many mathematicians and programmers apply several

approaches to find the solution of this problem which are having huge data sets (Mahesh,

2018). Machine Learning relies on different algorithms to solve data problems. Data scientists

like to point out that there’s no single one-size-fits-all type of algorithm that is best to solve a

problem. The kind of algorithm employed depends on the kind of problem you wish to solve,

the number of variables, the kind of model that would suit it best and so on. Here’s a quick

look at some of the commonly used algorithms in machine learning (ML).

2.5 Supervised Learning

Supervised learning is the machine learning task of learning a function that maps an input to

an output based on example input-output pairs. It infers a function from labelled training data

consisting of a set of training examples. The supervised machine learning algorithms are those

algorithms which needs external assistance. The input dataset is divided into train and test

dataset. The train dataset has output variable which needs to be predicted or classified. All

algorithms learn some kind of patterns from the training dataset and apply them to the test

dataset for prediction or classification. The workflow of supervised machine learning

algorithms is given in figure 4.

Fig. 4 Supervised Learning Workflow (https://pub.aimind.so/supervised-machine-learning-

algorithms-and-techniques-explained-in-depth)

2.5.1 Decision Tree

Decision tree is a graph to represent choices and their results in form of a tree. The nodes in

the graph represent an event or choice and the edges of the graph represent the decision rules

or conditions. Each tree consists of nodes and branches. Each node represents attributes in a

group that is to be classified and each branch represents a value that the node can take.

Fig. 5 Decision Tree Workflow (https://www.researchgate.net/figure/Workflow-of-Decision-

Tree-New-instances-are-classified-from-the-root-node-and-sorted)

2.5.2 Support Vector Machine

Another most widely used state-of-the-art machine learning technique is Support Vector

Machine (SVM). In machine learning, support-vector machines are supervised learning

models with associated learning algorithms that analyze data used for classification and

regression analysis. In addition to performing linear classification, SVMs can efficiently

perform a non-linear classification using what is called the kernel trick, implicitly mapping

their inputs into high-dimensional feature spaces. It basically, draw margins between the

classes. The margins are drawn in such a fashion that the distance between the margin and the

classes is maximum and hence, minimizing the classification error.

Fig. 6 Support Vector Machine Workflow (https://medium.com/@avicsebooks/ml-part12-

classification-support-vector-machines)

2.5.3 Navie Bayes

It is a classification technique based on Bayes Theorem with an assumption of independence

among predictors. In simple terms, a Naive Bayes classifier assumes that the presence of a

particular feature in a class is unrelated to the presence of any other feature. Navie Bayes

mainly targets the text classification industry. It is mainly used for clustering and classification

purpose depends on the conditional probability of happening.

2.6 Unsupervised Learning

The unsupervised learning algorithms learn few features from the data, when new data is

introduced; it uses the previously learned features to recognize the class of the data. It is

mainly used for clustering and feature reduction.

Fig. 7 Unsupervised Learning Workflow ( (Shatha G., 2022)

2.6.1 K-Means Clustering

K-means is one of the simplest unsupervised learning algorithms that solve the well-known

clustering problem. The procedure follows a simple and easy way to classify a given data set

through a certain number of clusters. The main idea is to define k centers, one for each cluster.

These centers should be placed in a cunning way because of different location causes different

result. So, the better choice is to place them is much as possible far away from each other.

Fig. 8 K-Means Clustering (https://www.ibm.com/topics/k-means-clustering)

2.7 Ensemble Machine Learning

Ensemble learning is the process by which multiple models, such as classifiers or experts, are

strategically generated and combined to solve a particular computational intelligence problem.

Ensemble learning is primarily used to improve the performance of a model, or reduce the

likelihood of an unfortunate selection of a poor one. Other applications of ensemble learning

include assigning a confidence to the decision made by the model, selecting optimal features,

data fusion, incremental learning, non-stationary learning and error-correcting. It is a general

approach to machine learning that seeks better predictive performance by combining the

predictions from multiple models.

Although there are a seemingly unlimited number of ensembles that can be develop for

predictive modeling problem, there are three methods that dominate the field of ensemble

learning. So much so, that rather than algorithms per se, each is a field of study that has

spawned many more specialized methods.

The three main classes of ensemble learning methods are bagging, stacking, and boosting.

2.7.1 Random Forest Classifier

Random forest classifier (RFC) is one of the most successful ensemble learning techniques

which has been proven to be an effective technique in pattern recognition and ML for high-

dimensional classification and skewed problems. RFC does have a drawback which is tree

classifiers have high variance. In practice, it is common for a slight change in the training

dataset to be in a different tree. The reason behind this is the hierarchical nature of the tree

classifiers.

2.7.2 Boosting
The term „Boosting‟ refers to a family of algorithms which converts weak learner to strong

learners. Boosting is a technique in ensemble learning which is used to decrease bias and

variance. Boosting is based on the question posed by Kearns and Valiant “Can a set of weak

learners create a single strong learner?" A weak learner is defined to be a classifier; a strong

learner is a classifier that is arbitrarily well-correlated with the true classification.

2.7.3 Bagging

Bagging or bootstrap aggregating is applied where the accuracy and stability of a machine

learning algorithm needs to be increased. It is applicable in classification and regression.

Bagging also decreases variance and helps in handling over-fitting.

Table 2.1: Summary of some previous works done on liquid loading

Author(s) Work done Finding(s) Limitation(s)

(Title of paper)

(Landjobo & Liquid film mode for Liquid film thickness is 1- Limited to

Wu, 2020) prediction and identification more sensitive to the tubing vertical wells
 of liquid loading in vertical inner diameter, the tubing 2- Machine

gas well pressure gradient, change in learning

liquid properties and the technique was

film and gas gravitational not applied.

forces

(Vieira & Applicability of models for Critical gas velocity for 1- Machine

Stanko, 2019) liquid loading in gas wells viscous oil was smaller learning

compared with other fluids technique was

not applied

2- Limited to

inclined wells

(Jafarov & Al- Critical review of the There is no any applicability 1- Machine

Nuaim, 2016) existing liquid loading ranges that can be applied to learning

prediction models for all vertical wells with technique was

vertical wells different wellheads not applied

pressure. 2- Limited to

vertical wells
(Nagoo, A Simplified Critical Gas Artificial variables (such as 1- Limited to

Kulkani, Velocity Equation as Direct interfacial friction factor) horizontal

Arnold, Function of Diameter and are not necessarily valid in wells

Dunham, Sosa, Inclination for Horizontal the determination of the 2- Machine

& Oyewole, Well Liquid Loading onset of fluid flow reversal learning
2018) Prediction: Theory and technique was

Extensive Field Validation not used.

(Shekhar, Improved Prediction of Critical gas velocity is a Machine learning

Kelkar, Hearn, Liquid Loading in Gas Wells function of inclination angle technique was not

& Hain, 2017) (as the angle increases, the used.

velocity also increases)

(Nallaparaju, Prediction of Liquid Loading The Turner model gives the Machine learning was

2012) in Gas Wells most conservative value of only involved to

critical velocity of all the capture the

other methods for any value relationship between

of pressure friction factors and

flowing conditions.

(Bolujo et al. A New Model for Predicting Numerical integration Machine learning was

2017 Liquid Loading in method was used while not involved

Multiphase Gas Wells considering the introduction

of valve equation along the

functional nodes to the

fundamental equation

El-sayed A. Prediction of Critical Gas The model used artificial Ensemble machine
Osman, 2002 Flow Rate for Gas Wells neural network (ANN) for learning technique was

Unloading predicting the minimum not tested

flow rate for continuous

removal of liquids from the

wellbore

CHAPTER THREE
3.0 MATERIALS AND METHOD

3.1 Materials

i. Python Software

ii. Mathlab Software

iii. Excel Sheet

iv. Production Data Set

v. Laptop

3.2 Method

A simple learning algorithm will be developed to construct a set of classifier using the

“Bayesian Averaging Method (Ensemble Technique)” and will be used for the prediction in

Python Software. Production data from different wells and literatures will be used to check the

applicability of the developed algorithm in predicting liquid loading.

The result will be validated using other existing models such as:

1- Turner et al. model (1969)

2- Li et al. model (2001)

3- Colemen’s Model

4- Nossier’s Model

5- Liu et al. model (2018)

Expected Result
REFERENCES

Abhulimen K.E, O. A. (2022). Modelling of Liquid Loading in Gas Wells Using a Software-
Based Approach. Journal of Petroleum Exploration and Production Technology.

Adesina Fdairo, F. O. (2014). A New Model for Predicting Liquid Loading in a Gas Well.
Journal of Natural Gas Science and Engineering.

Arnold Landjobo Pagou, G. H. (2020). Liquid Loading Prediction And Identification Model
For Vertical And Inclined Gas Wells. Journal of Natural Gas Science and
Engineering, 2-5.

Bolujo E.O, F. A. (2017). A New Model for Predicting Liquid Loading in Multiphase Gas
wells. International Journal of Applaied Engineering Research, ISSN 0973-4562
Volume 12.

Chemmakh A, T. O. (2023). Evaluation of Liquid Loading in Gas Wells Using Machine

Learning. Petroleum and Petrochemical Engineering Journal, ISSN; 2578-4846.

Jafarov, T., & Al-Nuaim, S. (2016). Critical Review of the Existing Liquid Loading Prediction
Models for Vertical Gas Wells. Society of Petroleum Engineers.

Landjobo, A., & Wu, X. (2020). Liquid Film Mode for Prediction and Identification of Liquid
Loading in Vertical Gas Wells. Society of Petroleum Engineers.

Mahesh, B. (2018). Machine Learning Algorithms - A Review. International Journal of

Science and Research, ISSN: 2319 - 7064.

Malik, A., Prakash, R., Kumar, M., & Barot, M. (2017). Predicting Start-Up Liquid Loading
in a Matured Oil and Gas Field: A Case Study. Society of Petroleum Engineers.

Nagoo, A. S., Kulkani, P. M., Arnold, C., Dunham, M., Sosa, J., & Oyewole, P. O. (2018). A
Simplified Critical Gas Velocity Equation as Direct Function of Diameter and
Inclination for Horizontal Well Liquid Loading Prediction: Theory and Extensive Field
Validation. Society of Petroleum Engineers.

Nallaparaju, Y. D. (2012). Prediction of Liquid Loading in Gas Wells. Society of Petroleum

Engineers.

Osman, E.-S. A. (2012). Prediction of Critical Gas Flow Rate for Gas Wells Unloading.
Society of Petroleum Engineers.
Princewell Maduabuchi, M. O. (2018). Li and Turner Modified Model for Predicting Liquid
Loading in Gas Wells. Journal of Petroleum Exploration and Poduction Technology.

Shatha G., A. J. (2022). Evaluating Student Levelling Based on MAchine Learning Model's
Performance. Researchgate.net.

Shekhar, S., Kelkar, M., Hearn, W. J., & Hain, L. L. (2017). Improved Prediction of Liquid
Loading in Gas Wells. Society of Petroleum Engineers.

Veeken K., B. H. (2010). Gas-Well Liquid Loading Field Data Analysis and Multiphase Flow.
OnePetro.Org, SPE-123657-PA.

Vieira, C., & Stanko, M. (2019). Applicability of Models for Liquid Loading Prediction in
Gas Wells. Society of Petroleum Engineers.

Wenq K., L. H. (2021). Research on Critical Liquid Carrying Model in Wellbore and
Laboratory Experimental Verification. Processes MDPI-JOURNAL.

Yonghui Liu, C. L. (2018). Experimental and Modelling Studies on the Prediction of Liquid
Loading Onset in Gas Wells. Journal of Natural Gas Science and Engineering,
Volume 57, Pages 349-358.