ABSTRACT

Artificial intelligence (AI) techniques have been widely used in the oil
and gas industry.
In this study using artificial neural network (ANN) method with SPSS
program for predicting (ECD). This study was based on dataset taken from
the measured study. Three types of activation function with different number
of neurons are used in training artificial neural network.

These activation functions were used in six cases, in (one, three, four)
cases has been used hyperbolic tangent as activation function in input hidden
layer, with pureline function for output hidden layer, and in (Two, five, six)
cases hyperbolic as a logsigmoid ctivation function in input and output
hidden layer is used,

Correlation coefficient (R2) and sum of square error (SSE) are used to
evaluate performance of prediction. The results obtained shows all activation
function types are used in input and output hidden layer gave good
predicting for ECD compared. Lastly, artificial neural network method could
be used to directly estimate ECD with good engineering accuracy and
without complex procedures.

A
 Table of Contents
Subject Page
Abstract A
Table of Contents B
List of Figures C
List of Tables C
List of Abbreviations D
Chapter One: Introduction 1
1.1: Introduction 1
1.2; Objective of this project 2
Chapter Two: Literature Review 3
2.1:Predicting ECD by Employing Machine Learning
3
Techniques
2.2 : Artificial Neural Networks 8
2.2.1: Brain Neuron 9
2.2.2: Artificial Neuron 10
2.2.3: Strengths and Limitations of ANN 13
2.2.4: Strengths and Limitations of ANN 14
Chapter Three: Results and Discussion 17
3.1: Artificial Neural Network Results and Discussion 17
3.1.1: Developing a Neural Network Model Equivalent
17
circulation density
3.1.1.1: Acquisition and Analysis of Data Base 17
3.1.1.2: Network Architecture Design 18
3.1.1.3: Pre-processing Data 19
3.1.1.4: Regression Based on Artificial Neural Network 19
3.2: Prediction of ECD Based on ANN 19
Chapter Four: Conclusions And Recommendations 27
4.1: Conclusions 27
4.2: Recommendations for Future Work 27
References 28
Appendices 37

B
 List of Figures
Figure Page
Fig. (2.1): Brain neuron 10
Fig. (2.2): An artificial neuron model 10
Fig.(2.3): sigmoid function 11
Fig. (2.4): hyperbolic tangent function 12
Fig. (2.5): neural network architecture 13
Fig. (3.1): ANN model for three-layer feed-forward back
19
propagation networks
Fig. (3.2): Predicted and actual ECD vs. test number from training
22
network for hyperbolic tangent and pure line with 2 neurons.
Fig. (3.3): Predicted vs. actual ECD for all stages for hyperbolic
22
tangent and pure line with 2 neurons.
Fig. (3.4): Predicted and actual ECD vs. test number from training
25
network for log sigmoid and pure line with 2 neurons.
Fig. (3.5): Predicted vs. actual ECD for training stage for log
26
sigmoid and pure line with 2 neurons.

List of Tables
Table Page
Table (2.1): ECD prediction models using AI among the
7
literature
Table (3.1): Data Range for ECD. 18
Table (3.2): Results from BPNN for determining ECD with 2
23
neurons for hyperbolic tangent.
Table (3.3): Weights and biases of suggested empirical model
23
for Eqs. (3.1)&(3.3).
Table (3.4): Results from BPNN for determining ECD with 2
26
neurons for log sigmoid.
Table (3.5): Weights and biases of proposed empirical model for
26
eq. (3.4)

C
 List of Abbreviations
Abbreviation Description
ANN Artificial Neural Network
BPNN Back Propagation Neural Network
ECD Equivalent Circulating Density
Levenberg-Marquardt Algorithm-based Artificial Neural
LM-ANN
Network
NCP Normal Computer Program
RBFN Radial Basis Functional Networks
ROP Rate of Penetration
SPSS Statistical Package for the Social Sciences
SSE Sum of Squares Error
ANFIS Adaptive Neuro-Fuzzy Inference System
SVM Support Vector Machine
FM Functional Networks
LSSVM Least Square Support Vector Machine
PSO-ANFIS Particle Swarm Optimization
FIS Fuzzy Inference System
GA Genetic Algorithm
TFA Total Flow Area
WOB Weight On Bit
RBF Radial Basis Function
APL Annular Pressure Losses
MWD Measurement While Drilling
PWD Pressure While Drilling
AI Artificial Intelligence
SPP stand-pipe pressure
T Torque
MLP Multilayer perceptions
NCP normal computer program

D
Chapter One
Introduction
Chapter One Introduction

CHAPTER ONE
INTRODUCTION

1.1: Introduction
Introduction Equivalent circulating density is an important parameter for
monitoring the drilling operations especially for the narrow window between
the formation and the fracture pressure. ECD is the total pressure of the mud
hydrostatic column and the annular losses, and hence, it shows the mud
pressure against the formation in the case of mud circulation 1 . Therefore, it
is critical to estimate the ECD with a high degree of precision to avoid any
well control issues like loss of circulation, formation fracturing, and
underground blowout situations. During the drilling operations, several
factors were found to have an impact on the ECD, and among them, the
annular pressure losses, wellbore geometry, mud properties (density and
viscosity), mud pumping rate, downhole pressure and temperature, and
2-5
concentration of cuttings . ECD can be acquired by means of downhole
measurements, estimation using mathematical models, and/or predicting
with the help of artificial intelligence (AI) techniques. The new technology
in the drilling tools assisted in implementing a continuous circulating tool to
6
monitor the ECD and provide good control for the formation pressure .
Downhole measurements of the ECD are available using downhole sensors
as measurements while drilling and pressure while drilling 7,8. The downhole
measurement is considered accurate and robust for ECD values, however,
the implementation of these downhole tools is not common due to the
expensive daily charge and operational limitations such as downhole
pressure and temperature that cause the tool failures. Several mathematical

1
Chapter One Introduction

correlations exist in the literature for estimating the ECD that are different in
the fluid type and the parameters utilized as inputs. ECD estimation by
implementing the material balance calculation for the mud compositional
9,10
analysis was studied in the literature . However, the models had many
assumptions and limitations regarding the downhole pressure, temperature,
11
mud types. Bybee introduced a mathematical equation to calculate the
ECD. The model considers the effect of concentration of solids in the
annular, in addition to, the mud static density and other mud-related
parameters. The developed mathematical correlations are limited to some
applications, and it ignores a lot of other input parameters that have an
impact on the ECD values. Such ignored parameters as well geometry, fluid
rheological properties, the rotation of the drill string, downhole pressure and
temperature conditions that affect the mud density, cuttings dispersion, hole
12,13
cleaning, and swab and surge of drillpipe movements in the hole .
Ignoring these parameters will affect the ECD prediction and lead to the
inaccurate evaluation of ECD and causes well control problems during the
drilling operations 14,15 .

1.2: Objective of this project

The main objective of this project to develop a new approach for predicting
Equivalent Circulating Density ECD using artificial neural network
techniques from surface drilling parameters [mud weight, drill pipe pressure,
and rate of penetration (ROP)].

2
 Chapter Two
Literature View
Chapter Two Literature view

CHAPTER TWO

LITERATURE REVIEW

2.1: Predicting ECD by Employing Machine Learning Techniques

Predicting the ECD from the drilling parameters is considered a new outlook
for drilling engineering practices in the petroleum industry and that because
of the limitations of the downhole ECD tools and the low accuracy of the
mathematical models. Artificial intelligence is a technique that utilized high
computing capabilities for processing advanced algorithms to solve
technical/problematic issues by simulating the human brain's thinking
manner16. AI has many tools like artificial neural networks (ANNs), adaptive
neuro-fuzzy inference system (ANFIS), support vector machine (SVM), and
functional networks (FN) that showed high performance and accuracy level
for prediction and classification problems17. The implementation of AI has
wide applications in many disciplines of engineering, economics, medicine,
military, marine sectors, etc.18,19. In the oil and gas industry, many studies
utilized machine learning techniques for finding solutions for practical
challenges20,23. Intelligent models were accomplished by artificial
intelligence tools for many purposes as identifying the formation lithology24,
predicting the formation and fracture pressures 25,26, estimating the properties
of reservoir fluids27, estimating the oil recovery factor 28,29
, predicting the
tops of the drilled formation30, ROP prediction and optimization for different
31,33
drilled formations and well profiles , determining the content of total

3
Chapter Two Literature view
34,36 37,40
organic carbon , and estimating the rock static Young’s modulus ,
predicting the compressional and shear sonic times41, determining the rock
failure parameters42, detecting the downhole abnormalities during horizontal
drilling43, determining the wear of a drill bit from the drilling parameters44,
45,49
and predicting the rheological properties of drilling fluids in real-time .
For ECD prediction, Table 1 represents recent works that were performed
for ECD prediction from the drilling and mud parameters. Ahmadi 50 utilized
the least square support vector machine (LSSVM), ANFIS, and enhanced
particle swarm optimization PSO-ANFIS tools to estimate the ECD from
only mud initial density, pressure, and temperature. The results showed the
outperformance of ANN than the other tools. Ahmadi et al51. studied
predicting ECD by employing PSO-ANN, fuzzy inference system (FIS), and
a hybrid of genetic algorithm (GA) and FIS (GA-FIS) from the initial mud
density, pressure, and temperature data. The PSO-ANN model presented a
high degree of prediction performance in terms of coefficient of
determination (R2 ) and sum of squares error (SSE) between the actual and
predicted values of ECD. Alkinani et al.52 predicted the ECD using the ANN
model that had only one hidden layer and 12 neurons and the study utilized
drilling parameters in addition to the hydraulics and mud properties as mud
pumping rate, properties of the mud (density, plastic viscosity, and yield
point), total flow area for the bite nozzles (TFA), revolutions per minute for
the drill pipe (RPM), and the weight on bit (WOB). Abdelgawad et al.5
provided a model for ECD prediction using two AI techniques ANN, and
ANFIS. The study provided an ECD-ANN model of one hidden layer with
20 neurons, while the ANFIS model was developed by utilizing five
membership functions with gaussian membership function (gaussmf) as the
input membership function and the output membership function was a linear
4
Chapter Two Literature view

type. Rahmati and Tatar53 employed radial basis function (RBF) to build an
ECD prediction model that showed a good prediction capability with R 2 of
0.98 and AAPE of 0.22%.

ECD is defined as the sum of the mud hydrostatic pressure and the annulus
pressure loss acting on the formation (Haciislamoglu 1994). The annular
clearance, mud weight, mud rheology, annular velocity (pump rates), cutting
concentration in the annulus, and hole depth are the main parameters which
affect the annular pressure losses (APL). The two main components that
affect the ECD are the cutting portion in the annulus expressed as equivalent
static density (ESD), and the mud-related parameters (Zhang et al. 2013;
Hemphill and Ravi 2011). Bybee (2009).

Such numerical evaluations for predicting ECD values did not take into
account other factors affecting ECD while drilling such as flow geometry
defined by well geometry, fluid resistance to flow defined by fluid rheology,
and drill string rotation. Ignoring these factors in the equation will increase
the error factors while estimating ECD (Caicedo et al. 2010; Costa et al.
2008).

Recently, in the oil industry, downhole tools are used to measure and

monitor changes of ECD to avoid well control issues such as gas kicks,

blowout, and formation fracturing (such as Erge et al. 2016; Rommetveit

et al. 2010). The main tools used now are measurement while drilling

(MWD) and pressure while drilling (PWD). These tools contain pressure

sensors that can independently measure the bottomhole pressure of the well

5
Chapter Two Literature view

during drilling, regardless of the factors controlling the ECD (such as

Ettehadi et al. 2013; Dokhani et al. 2016). The tools can give an accurate

reading for ESD and ECD from the total pressure acting on the bottom of the

well during circulation. Comparing the ESD with ECD will give a clear view

about the reasons for ECD changes (such as Vajargah et al. 2016; Osisanya

and Harris 2005; Lin et al. 2016). In addition to the expensive daily rates of

such tools, there are some operating limitations for its application such as

pressure, temperature, and tool failures. The objective of this paper is to use

different artificial intelligence (AI) techniques to develop a robust model to

predict the ECD using surface drilling parameters such as mud weight,

surface drill pipe pressure and rate of penetration. The models are developed

using artificial neural network (ANN) and adaptive neuro-fuzzy inference

system (ANFIS). In addition, an empirical correlation is extracted from the

ANN model which can be used to calculate the ECD from surface drilling

parameters.

6
 Chapter Two Literature view

Table (2.1): ECD prediction models using AI among the literature

Ref. Model Model Inputs Data R2 SSE

LSSVM 0.9999 0.000145

ANFIS Pressure 0.8502 35.002

Temperature
Ahmadi 50 Not Available Not
Initial density
PSO-ANFIS 0.869 Available

PSO-ANN 0.9964 0.0001374

Pressure
Ahmadi et al. FIS Temperature 664 points 0.7273 67.0907
51
Initial density from literature
GA-FIS 0.9397 0.091

Flow rate
Mud weight
Plastic viscosity
Yield point
Alkinani et al. TFA Not
52 ANN 2000 wells 0.982 Available
RPM
WOB

ANFIS 2376 data

points
Mud weight
Abdelgawad Drill pipe pressure
et al 5 8.5″ vertical 0.98 0.22
ANN ROP
hole section

Pressure
Radial Basis Temperature
Rahmati and Function 884 points MSE
Type of mud 0.99
Tatar 53 (RBF) from literature 0.00000166
Initial density

7
Chapter Two Literature view

It is clear from the literature that the AI models enhanced the ECD
prediction, however, the models are different in terms of the input
parameters, the data used to feed the models, and the methodology followed
for the ECD prediction. One of the shortcomings found from many studies in
the literature is that the downhole pressure and temperature are required as
inputs in the prediction models, and from an operational view, downhole
sensors are required to acquire these parameters with high accuracy for
better ECD prediction, and this will add operational cost and time for the
data collecting. Consequently, the new contribution of this study is to
employ available real-time drilling parameters from surface rig sensors to
build ECD prediction models using ANN and ANFIS techniques. The novel
approach in this study is that the AI models are mainly dependent only on
the mechanical drilling parameters that are mud pumping rate (GPM), rate of
penetration (ROP), drillstring speed in revolutions per minute (RPM), stand-
pipe pressure (SPP), weight on bit (WOB), and drilling (T). Besides, the
study presented an empirical correlation that can be easily utilized for ECD
estimation from only the drilling parameters. The AI models that were
presented in this study were validated from another data set to ensure high
and robust performance for ECD prediction.

2.2: Artificial Neural Networks

Simply, an artificial neural network (ANN) imitates the human brain as the
natural brain can learn new things and adapt to new and changing
environments. Brain can remarkably analyze incomplete, and vague
information, as well as making its own judgments based on it. For instance,
other’s handwriting can be read in spite of being totally different from
others; the child can describe the ball and orange shapes as circle, know its
8
Chapter Two Literature view

mother via her voice and smell; even a famous person can be identified from
a fuzzy picture. Brain represents a highly complicated organ, controlling the
whole body. Even brain of most primitive animals is more skilled compared
to the most innovative computers. It is not only controlling the physical body
parts, but also the most difficult activities including visualizing, thinking and
learning activities difficult to be explained using physical expressions.
Hence, the most innovative processors still lack the ability to create an
artificial machine of thinking.

2.2.1: Brain Neuron

Brain contains cells known as neurons whose interconnections generate the

neural network. There are about 1011 neurons in the human brain and about
10000 connections with each other. ANN imitates the natural neural network
(ANN) in which connection among artificial neurons is done the same way
as the network of brain. The natural neuron contains cell body, axon and
dendrite. Electro biological signals from other neurons are transmitted into
the cell body by dendrite. The cell body (known as Soma) is composed of
nuclei and other biological structures needed to support the cell. Signals are
transported from one neuron to others by axon. Synapse refers to a junction
between two neurons ' dendrites, or a neuron to muscle cells 54

9
Chapter Two Literature view

Fig. (2.1): Brain neuron 55.

2.2.2: Artificial Neuron

ANN includes processing units known as neurons, attempting to imitate the

natural neuron structure and behavior. A neuron has inputs (dendrites) and
one output (synapse via axon), to determine the activation of the neuron.

Fig. (2.2): An artificial neuron model 56.

11
Chapter Two Literature view

Where (x1...xn) are the inputs to the neuron. A bias is also added to the
neuron along with inputs. Usually, bias value is initialized to 1; (W0...Wn)
are the weights, a weight is the connection to the signal. Product of weight
and input gives the signal strength. A neuron receives multiple inputs from
different sources and has a single output. There are many transfer functions
are used for treatment data and the sigmoid function is used one of the most
frequently, given by:

( ) (2-1)

( ) (2-2)

Where:

Fig. (2.3): Sigmoid function 57.

11
Chapter Two Literature view

Step, Linear, Ramp and Hyperbolic tangent functions are also used.

( ) (2.3)

( ) (2.4)

Function of hyperbolic tangent (tanh) resembles sigmoid in terms of shape,

but differs from it in terms of limits (-1 to +1) unlike sigmoid (0 to 1).

Fig. (2.4): Hyperbolic tangent function.

Sum represents the inputs weighted sum multiplied by weights between two
layers. Sigmoid is the function of activation, representing a continuous and
58
differentiable approximation of function of step . The neural network is
created by the interconnections of individual neurons. The architecture of
ANN involves:
1. Input layer to receive the values of input
2. A group of neurons between input and output layers represents hidden
layer(s). These layers have single or multiple.

12
Chapter Two Literature view

3. Usually one neuron in output layer with output ranging between (0

and 1). However, there could be multiple outputs 59.

Fig. (2.5): Neural network architecture 56.

Weights represent inter-unit connection strengths where the ability of

56
processing is stored . Strength of Input depends on the value of weight,
which could be zero, negative or positive. Negative weight refers to reduced
or inhibited signal, and the lack of coupling between each two neurons refer
zero value. These weights are modified to get required output. ANN weights
are adjusted using algorithms to obtain the output required. The process of
modifying weights is known as learning or training 58.

2.2.3: ANN Training Process

ANN categories depend on supervised and unsupervised

methods of learning. Two inputs and one output with one
neuron considered the simplest system of ANN design.
Either step or ramp is used as the function of activation. Data
are classified into two separate categories by using
perceptions. Multilayer perceptions (MLP) are used for

13
Chapter Two Literature view

complex applications, and consist of one or more hidden

layers with one input and one output layer.
Back propagation algorithm is the most common method used in neural netw
ork training. The difference in the targeted output and the obtained output is
propagated back to the layers and the adjusted weights. A supervised method
of learning and feed-forward architecture is used in back propagation neural
network (BPNN), which represents the most common techniques of neural
network utilized for prediction and classification. The hidden layer outputs
in algorithm of BP are propagated to the output layer where the output is
calculated and then compared to the desired output for specific input.

This difference determines the back propagation of error from the output
layer to the hidden one, and then from the hidden layer to the input one.
With flow of the back movement, weights between neurons are changed.
The going forward cycle from input to output and from output to input is
called Epoch. Firstly, a set of identified input data is given to the neural
network, then training it to obtain on output known. This process is termed
training the network. Several such epochs are experienced by the network
until the error reaches specific tolerance so that the network becomes
trained. Weights among all neurons in all layers are set by this training
process, and then are employed in calculating the network response to
unknown data.

2.2.4: Strengths and Limitations of ANN

There are several ways defining the differences between ANN and a normal
computer program (NCP), including 59:

14
Chapter Two Literature view

1. Adaptive learning: ANN imitates human brain regarding the

performance of tasks while learning. Other input types cannot be
adapted by NCP.
2. Self-organization: ANN could organize itself while learning. Whereas
NCP adheres to its fixed tasks and will not perform other tasks beyond
its capacity.
3. Parallel process: ANN similar work a human brain, unlike NCP which
works in sequence.
4. Fault tolerance: ANN can work even with incomplete, noisy and
vague data; however, NCP cannot deal with incomplete and vague
data and will stop working when encountering data with a smallest
wrong.
5. Compared to the brain of human, when the processing time of brain is
slower the ANN is relatively fast.
6. Compared to NCP, the output calculation method in ANN is unclear.
The time taken remains changing with different inputs sets regardless
of their similarity.
7. ANN could be applied to classify data, recognize patterns, and
applications with unclear data.
8. Cannot be applied ANN with nature of input and output of known,
and with the obviously known tasks.

For training and testing using ANN, normalization of input parameters

ranged between [-1,1] consistent with the chosen function of activation.
Therefore, the equation below was utilized 59:

( ) (2,5)

15
Chapter Two Literature view

or between [0,1] by using the following equation.

(2.6)

Where xn is the normalized parameter, x is the actual parameter, xmin and xmax
represent the minimum and the maximum values of the actual parameters,
respectively.

Finally, for evaluating performance, the coefficient of correlation (R 2) and

the Sum of Squares Error (SSE) were utilized and defined using these
equations, respectively:

∑ ( ̅) ∑ ( )
∑ ( ̅)
(2.7)

∑ | | (2.8)

Where yact. is the measured (actual) value, ypred. is the predicted (estimated)
value, ̅ represents average value and N refers to the samples number.
Higher values of R with lower values of (SSE) denote higher accuracy and a
better prediction performance.

16
Chapter Three
Results and
Discussion
Chapter Three Results and Discussion

CHAPTER THREE
RESULTS AND DISCUSSION

In This chapter we covers using artificial neural network (ANN)

method to ECD and to find relationship between the inputs and target.
Artificial neural network (ANN) was achieved by SPSS program.

3.1: Artificial Neural Network Results and Discussion

3.1.1: Developing a Neural Network Model Equivalent circulation
density
In this section suggested ANN method was used to ECD. This work is
conducted by SPSS.

3.1.1.1: Acquisition and Analysis of Data Base

A 14 observation from measurement field using for building ECD network
form. Table (3.1) presents Ranges of data were used in this study. Also in
this work we are selection three effective as input parameters which
include: rate of penetration(RPM) , mud weight(MW) , drill pipe
pressure(DPP).

17
Chapter Three Results and Discussion

Table (3.1): Data Range for ECD.

Elements Range
Rate of penetration (ft/hr) 42-55
Mud weight (ppg) 8.6-8.9
Drill pipe pressure (psi) 160-1100

3.1.1.2: Network Architecture Design

The feed-forward neural networks with the back-propagation (BPNN)
learning algorithm and the radial basis functional networks (RBFN)
represents two different models in artificial neural network. Artificial neural
network with three-layer feed-forward back propagation networks has been
used to ECD in horizontal and inclined wells see in Figure (3.1). Two types
of transfer function with Different number of neurons of hidden were used in
this study.
In this study, Levenberge-Marquardt (trainlm) is training algorithm
with two activation function and different number of neurons has been used
to achieve this job and considered the fastest method for training moderate-
sized feed-forward neural networks reach to several hundred weights.

18
Chapter Three Results and Discussion

Fig. (3.1): ANN model for three-layer feed-forward back propagation networks.
3.1.1.3: Pre-processing Data
Preprocessing data it is necessary considerations before applying the
data base in neural network. Normalization all input data were used in this
study between [-1, 1] and [0, 1] by eqs. (2.5)& (2.6) for hyperbolic tangent
and log-sigmoid function respectively.

3.1.1.4: Regression Based on Artificial Neural Network

The relationship between system variables in many cases it is difficult
known. The ANN is able to connect and give the relationship between input
and output parameters. In this study supervised networks which are used and
consist of training examples.
ANN models are consist three steps always observed: training, validation,
and testing. Also to build ANN model the dataset in appendix B can be
divided into three parts: 70%, 15%, and 15% for training, validation, and
testing respectively.

3.2: Prediction of ECD Based on ANN

In this section, dataset in Table (3.1) were used for generating ANN model.
SPSS code were used in the application of the Levenberg-Marquardt

19
Chapter Three Results and Discussion

algorithm-based ANN model (LM-ANN). Many cases for (ECD) were using
for hyperbolic tangent and log sigmoid function with different number of
neuron.
Moreover, plotting the predicted value (output), versus the actual value.
Lastly, the correlation coefficient (R) and (SSE) are used for performance
evaluation the results that were obtained during training process.

Case-1: Using hyperbolic tangent as activation function in layer one

(Hidden layer) and pureline function in layer two (out layer) with 2
neuron in hidden layer.
The multilayer perceptron technique was using to estimate the ECD. In this
section the results are extracted and discussed here from training and testing
regression with multilayer perceptron technique. The multilayer perceptron
has been implemented for carried out non-linear regression to obtain
predicted ECD which plotted with (measured) and versus the test number
see in Figure (3.2) for the training stage. Also regression results between the
actual and predicted data for all stages show in Figure (3.3).
Table (3.2) presents evaluation tools for this model which include (R) (R 2)
and (SSE) for all stages regression, we are observation high correlation
coefficient value in all stage with maximum value in validation stage and
SSE is accepted value. This is indicate the model performance is a
satisfactory. Also, results observed excellent match between predicted and
actual data.
Empirical model in this study based on weights and biases show in table
(3.3) associated with the input layer/hidden layer and hidden layer/output
layer has been developed by incorporating the dataset shown in appendix B.
Before presenting the input-output data to the LM-ANN model, all the
21
Chapter Three Results and Discussion

datasets were normalized between the ranges [-1, 1] using Eq. (2.5) for
tangent function.
Also, the weights and the biases was extracted from empirical correlation for
SPSS program have been used to convert the ANN technique to a white box.
Equation (3.1) to calculate equivalent circulating density (ECD) which is
normalized form of ECD . To convert normalize the output to obtain the de-
normalize values of ECD used Equation (3.2) for this purpose.
*∑ ( ( ) ) + (3.1)
( )( )
(3.2)
Where:
N: the number of neurons.
W1: hidden layer cofficients.
W2: output layer cofficients.
b1: hidden layer bais
b2: output layer bais
Table (3.3) lists the input parameters for eq. (3.1)&(3.3).
Other expression for eq. (3.1).
∑ ( )
(3.3)

21
Chapter Three Results and Discussion

0
200
400
600
800
Depth (ft)

1000
1200
1400
1600
1800
2000
8.65 8.7 8.75 8.8 8.85 8.9 8.95 9 9.05

ECD (ppg)
ECD Predicted (ppg) ECD Actual (ppg)

Fig. (3.2): Predicted and actual ECD vs. test number from training network for
hyperbolic tangent and pureline with 2 neurons.

9.05

9 y = 0.9733x + 0.2375
R² = 0.9837
8.95
ECD Actual (ppg)

8.9

8.85

8.8

8.75

8.7

8.65
8.65 8.7 8.75 8.8 8.85 8.9 8.95 9 9.05
ECD Predicted (ppg)

Fig. (3.3): Predicted vs. actual ECD for all stages for hyperbolic tangent and
pureline with 2 neurons.

22
Chapter Three Results and Discussion

Table (3.2): Results from BPNN for determining ECD with 2 neurons for hyperbolic
tangent.
Stage R R2 SSE (%)
Training 0.991 0.9837 0.004

Table (3.3): Weights and biases of suggested empirical model for Eqs. (3.1)&(3.3).

Input Hidden
Output
Hidden Layer Layer
Input Layer Weight Matrix Layer
Layer Bias Weight Bias
Neurons ( ) Vector
(b1) Vector
(i) ( ) (b2)
1 -0.062 0.517 -0.152 0.307 0.341
-0.067
2 -0.236 -0.987 -0.432 0.199 -0.848

Results for the same case with 3 number neurons are given in appendixes
A&B

23
Chapter Three Results and Discussion

Case-2: Using logsigmoid as activation function in layer one (Hidden

layer) and pureline function in layer two (out layer) with 2 neuron in
hidden layer.
In this case using hyperbolic tangent as activation function in hidden and
output layers. In this section the results are extracted and discussed here
from training and testing regression with BPNN technique. The BPNN has
been implemented for carried out non-linear regression to obtain predicted
(ECD) which plotted with (measured) and versus the test number see in
Figure (3.4) for the training stage. Also regression results between the actual
and predicted data for all stages show in Figure (3.5).
Table (3.4) presents evaluation tools for this model which include (R) (R 2)
and (SSE) for all stages regression, we are observation high correlation
coefficient value in all stage with maximum value in validation stage and
SSE is accepted value. This is indicate the model performance is a
satisfactory. Also, results observed excellent match between predicted and
actual data.
Empirical model in this study based on weights and biases show in table
(3.5) associated with the input layer/hidden layer and hidden layer/output
layer has been developed by incorporating the dataset shown in appendix B.
Before presenting the input-output data to the LM-ANN model, all the
datasets were normalized between the ranges [0, 1] using Eq. (2.6) for
tangent function.
Also, the weights and the biases was extracted from empirical correlation for
SPSS program have been used to convert the ANN technique to a white box.

42
Chapter Three Results and Discussion

Equation (3.4) to calculate (ECD) which is normalized form of ECD. To

convert normalize the output to obtain the de-normalize values of ECD used
Equation (3.5) for this purpose. Moreover Figure (3.4) presents the
architecture of ANN constructed for ECD prediction from SPSS.
*∑ ( ( ) ) + (3.4)

( ) (3.5)

0
200
400
600
800
Depth (ft)

1000
1200
1400
1600
1800
2000
8.65 8.7 8.75 8.8 8.85 8.9 8.95 9 9.05
ECD (ppg)
ECD Predicted (ppg) ECD Actual (ppg)

Fig. (3.4): Predicted and actual ECD vs. test number from training network for logsigmoid
and pureline with 2 neurons.

42
Chapter Three Results and Discussion

9.05

9 y = 1.0346x - 0.3034
R² = 0.9639
8.95
ECD Actual (ppg)

8.9

8.85

8.8

8.75

8.7

8.65
8.65 8.7 8.75 8.8 8.85 8.9 8.95 9 9.05
ECD Predicted (ppg)

Fig. (3.5): Predicted vs. actual ECD for training stage for logsigmoid and
pureline with 2 neurons.

Table (3.4): Results from BPNN for determining ECD with 2 neurons for
logsigmoid.
Stage R R2 SSE (%)
Training 0.981 0.9639 0.011

Table (3.5): Weights and biases of proposed empirical model for eq. (3.4).

Input Hidden
Output
Hidden Layer Layer
Input Layer Weight Matrix Layer
Layer Bias Weight Bias
Neurons ( ) Vector
(b1) Vector
(i) ( ) (b2)
1 0.618 0.793 0.986 -0.167 -4.090
-5.949
2 0.360 1.280 0.693 -0.492 13.378

42
 Chapter Four
Conclusions and
Recommendations
Chapter Four Conclusions and Recommendations

CHAPTER FOUR
CONCLUSIONS AND RECOMMENDATIONS

The most important observations and conclusions of this study has

been summarized and presented in this chapter. Also, in this chapter
Recommendations for future work is presented.

4.1: Conclusions
1. The BPNN learning algorithm is a good technique for training this
network and equivalent circulating density.
2. All activation function gave excellent match between actual and
predicted data.
3. All activation function gave higher correlation coefficient with
minimum Sum of Squares Error.
4. Two neurons gave better results as compared with four and six
neurons in all cases and for all activation function.
5. In all cases the tan function is better than the log function.

4.2: Recommendations for Future Work

Future work recommendations are including the followings points:
Using other parameters effects on (ECD) with artificial neural network such
as;(temperature),(viscosity).

27
References
 References

1. Haciislamoglu, M. 1994. Practical Pressure Loss Predictions in Realistic

Annular Geometries. Paper presented at the Annual Technical Conference
and Exhibition, New Orleans, Louisiana, 25-28 September. SPE-28304-MS.
https://doi.org/10.2118/28304-MS.
2. Osman, E.A. and Aggour, M.A., 2003. Determination of drilling mud
density change with pressure and temperature made simple and accurate by
ANN. Paper presented at the Middle East Oil Show, Bahrain, 9-12 June.
SPE-81422-MS. https://doi.org/10.2118/81422-MS.
3. Hemphill, T., and Ravi, K. 2011. Improved Prediction of ECD with Drill
Pipe Rotation. Paper presented at the International Petroleum Technology
Conference, Bangkok, Thailand, 15-17 November. IPTC-15424-MS.
https://doi.org/10.2523/IPTC-15424-MS.
4. Zhang, H., Sun, T., Gao, D. and Tang, H., 2013. A new method for
calculating the equivalent circulating density of drilling fluid in deepwater
drilling for oil and gas. Chemistry and technology of fuels and oils, 49(5),
pp.430-438.
5. Abdelgawad, K.Z., Elzenary, M., Elkatatny, S., Mahmoud, M.,
Abdulraheem, A. and Patil, S., 2019. New approach to evaluate the
equivalent circulating density (ECD) using artificial intelligence techniques.
Journal of Petroleum Exploration and Production Technology, 9(2),
pp.1569-1578. https://doi.org/10.1007/s13202-018-0572-y.
6. Ataga, E., Ogbonna, J. and Boniface, O., 2012, January. Accurate
estimation of equivalent circulating density during high pressure high
temperature (HPHT) drilling operations. Paper presented in Nigeria Annual
International Conference and Exhibition. Nigeria Annual International

28
 References

Conference and Exhibition, Lagos, Nigeria, 6-8 August. SPE-162972-MS.

https://doi.org/10.2118/162972-MS.
7. Rommetveit, R., Odegard, S.I., Nordstrand, C., Bjorkevoll, K.S., Cerasi,
P.R., Helset, H.M., Fjeldheim, M. and Havardstein, S.T., 2010, January.
Drilling a challenging HP/HT well utilizing an advanced ECD management
system with decision support and real-time simulations. Paper presented at
the IADC/SPE Drilling Conference and Exhibition, New Orleans, Louisiana,
USA, 2-4 February. SPE-128648-MS. https://doi.org/10.2118/128648-MS.
8. Erge, O., Vajargah, A. K., Ozbayoglu, M. E., & van Oort, E. 2016.
Improved ECD Prediction and Management in Horizontal and Extended
Reach Wells with Eccentric Drillstrings. Paper presented at the IADC/SPE
Drilling Conference and Exhibition, Fort Worth, Texas, USA, 1-3 March.
SPE-178785-MS. https://doi.org/10.2118/178785-MS.
9. Hoberock, L.L., Thomas, D.C. and Nickens, H.V., 1982. Here's how
compressibility and temperature affect bottom-hole mud pressure. Oil Gas
J.;(United States), 80(12) 159-164.
10. Peters, E.J., Chenevert, M.E. and Zhang, C., 1990. A model for
predicting the density of oil-base muds at high pressures and temperatures.
SPE drilling engineering, 5(02), pp.141-148.
11. Bybee, K., 2009. Equivalent-circulating-density fluctuation in extended-
reach drilling. J Petrol Technol 61:64–67. https://doi.org/10.2118/0209-
0064-JPT.
12. Hemphill, T., Ravi, K., Bern, P.A. and Rojas, J., 2008. A simplified
method for prediction of ECD increase with drillpipe rotation. Paper
presented at the Annual Technical Conference and Exhibition, Denver,

29
 References

Colorado, USA, 21-24 September. SPE-115378-MS,

https://doi.org/10.2118/115378-MS.
13. Ahmed, R.M., Enfis, M.S., El Kheir, H.M., Laget, M. and Saasen, A.,
2010. The effect of drillstring rotation on equivalent circulation density:
modeling and analysis of field measurements. Paper presented at the Annual
Technical Conference and Exhibition, Florence, Italy, 19-22 September.
SPE-135587-MS. https://doi.org/10.2118/135587-MS.
14. Costa, S.S., Stuckenbruck, S., Fontoura, S.A. and Martins, A.L., 2008,
January. Simulation of transient cuttings transportation and ECD in wellbore
drilling. Paper presented at Europec/EAGE Conference and Exhibition,
Rome, Italy, 9-12 June 2008. SPE-113893-MS.
https://doi.org/10.2118/113893-MS.
15. Caicedo, H. U., Pribadi, M. A., Bahuguna, S., Wijnands, F. M., &
Setiawan, N. B. 2010. Geomechanics, ECD Management, and RSS to
Manage Drilling Challenges in a Mature Field. Paper presented at SPE Oil
and Gas India Conference and Exhibition, Mumbai, India, 20-22 January.
SPE-129158-MS. https://doi.org/10.2118/129158-MS.
16. Kalogirou, S. 2003. Artificial intelligence for the modeling and
control of combustion processes: a review.Progress in Energy and
Combustion Science,29(6),pp.515-566https://doi.org/10.1016/S0360-
1285(03)00058-3.
17. Shahab, M. 2000. Virtual-Intelligence Applications in Petroleum
Engineering: Part 1 Artificial Neural Networks. Journal of Petroleum
Technology, 52(9). https://doi.org/10.2118/58046-JPT.

18. Hag Elsafi, S., 2014. Artificial Neural Networks (ANNs) for flood
forecasting at Dongola Station in the River Nile, Sudan. Alexandria
Engineering Journal 53, 655-662. https://doi.org/10.1016/j.aej.2014.06.010.

31
 References

19. Babikir, H.A., Abd Elaziz, M., Elsheikh, A.H., Showaib, E.A.,
Elhadary, M., Wu, D., and Liu, Y., 2019. Noise prediction of axial piston
pump based on different valve materials using a modified artificial neural
network model. Alexandria Engineering Journal 58, 1077-1087,
https://doi.org/10.1016/j.aej.2019.09.010.

20. Rolon, L., Mohaghegh, S.D., Ameri, S., Gaskari, R. and McDaniel, B.,
2009. Using artificial neural networks to generate synthetic well logs.
Journal of Natural Gas Science and Engineering, 1(4-5), pp.118-133.
https://doi.org/10.1016/j.jngse.2009.08.003.

21. Tariq, Z., Elkatatny, S., Mahmoud, M., Ali, A.Z. and Abdulraheem, A.,
2017, May. A new technique to develop rock strength correlation using
artificial intelligence tools. This paper was presented at the SPE Reservoir
Characterisation and Simulation Conference and Exhibition.SPE
186062-MS.https://doi.org.extoljp.kfupm.edu.sa/10.2118/186062-MS.

22. Elkatatny, S., Mahmoud, M., Tariq, Z. and Abdulraheem, A., 2017.
New insights into the prediction of heterogeneous carbonate reservoir
permeability from well logs using artificial intelligence network. Neural
Computing and Applications, 30(9), pp.2673-2683.
https://doi.org/10.1007/s00521-017-2850-x.

23. Mousa, T., Elkatatny, S.M., Mahmoud, M.A. and Abdulraheem, A.,
2018. Development of new permeability formulation from well log data
using artificial intelligence approaches. Journal of Energy Resources
Technology. https://doi.org/10.1115/1.4039270.

24. Ren, X., Hou, J., Song, S., Liu, Y., Chen, D., Wang, X., and Dou, L.,
2019. Lithology identification using well logs: A method by integrating
artificial neural networks and sedimentary patterns. Journal of Petroleum
Science and Engineering 182, 106336.
https://doi.org/10.1016/j.petrol.2019.106336.

25. Ahmed, A.S., Mahmoud, A.A., and Elkatatny, S., 2019. Fracture
Pressure Prediction Using Radial Basis Function. In Proceedings of the

31
 References

AADE National Technical Conference and Exhibition, Denver, CO, USA, 9-

10 April. AADE-19-NTCE-061.

26. Ahmed, A.S., Mahmoud, A.A., Elkatatny, S., Mahmoud, M., and
Abdulraheem, A., 2019. Prediction of Pore and Fracture Pressures Using
Support Vector Machine. In Proceedings of the 2019 International
Petroleum Technology Conference, Beijing, China, 26-28
March. IPTC-19523-MS.https://doi.org/10.2523/IPTC-19523-MS.

27. Elkatatny, S. and Mahmoud, M., 2018. Development of new

correlations for the oil formation volume factor in oil reservoirs using
artificial intelligent white box technique. Petroleum, 4(2),
pp.178-186. https://doi.org/10.1016/j.petlm.2017.09.009.

28. Mahmoud, A.A., Elkatatny, S., Abdulraheem, A., and Mahmoud, M.,
2017. Application of Artificial Intelligence Techniques in Estimating Oil
Recovery Factor for Water Drive Sandy Reservoirs. This paper was
presented at the 2017 SPE Kuwait Oil & Gas Show and Conference, Kuwait
City, Kuwait, 15-18 October, SPE-187621-MS.
https://doi.org/10.2118/187621-MS.

29. Mahmoud, A.A., Elkatatny, S., Chen, W., and Abdulraheem, A., 2019.
Estimation of Oil Recovery Factor for Water Drive Sandy Reservoirs
through Applications of Artificial Intelligence. Energies 12, 3671.
https://doi.org/10.3390/en12193671.

30. Elkatatny, S., Al-AbdulJabbar, A., and Mahmoud, A.A., 2019. New
Robust Model to Estimate the Formation Tops in Real-Time Using
Artificial Neural Networks (ANN). Petrophysics 60, 825-837.
https://doi.org/10.30632/PJV60N6-2019a7.

31. Al-Abduljabbar, A., Gamal, H. and Elkatatny, S. 2020. Application of

artificial neural network to predict the rate of penetration for S-shape well
profile. Arab J Geosci 13, 784. https://doi.org/10.1007/s12517-020-05821-
w.

32. Gamal, H., Elkatatny, S. and Abdulraheem, A., 2020, November. Rock
Drillability Intelligent Prediction for a Complex Lithology Using Artificial
32
 References

Neural Network. Paper presented in Abu Dhabi International Petroleum

Exhibition & Conference, Abu Dhabi, UAE, 9-12 November.
SPE-202767-MS.https://doi.org/10.2118/202767-MS.

33. Mahmoud, A.A., Elkatatny, S., Al-AbdulJabbar, A., Moussa, T., Gamal,
H. and Shehri, D.A., 2020, September. Artificial Neural Networks Model for
Prediction of the Rate of Penetration While Horizontally Drilling Carbonate
Formations. In 54th US Rock Mechanics/Geomechanics Symposium.
American Rock Mechanics Association. ARMA-2020-1694.

34. Mahmoud, A.A., Elkatatny, S., Abdulraheem, A., Mahmoud, M.,

Ibrahim, O., and Ali, A., 2017. New Technique to Determine the Total
Organic Carbon Based on Well Logs Using Artificial Neural Network
(White Box). This paper was presented at the SPE Kingdom of Saudi Arabia
Annual Technical Symposium and Exhibition, Dammam, Saudi Arabia, 24-
27 April. SPE-188016-MS. https://doi.org/10.2118/188016-MS.

35. Mahmoud, A.A., Elkatatny, S., Ali, A., Abouelresh, M., and
Abdulraheem, A., 2019. New Robust Model to Evaluate the Total Organic
Carbon Using Fuzzy Logic. This paper was presented at the SPE Kuwait Oil
& Gas Show and Conference, Mishref, Kuwait, 13-16 October. SPE-
198130-MS. https://doi.org/10.2118/198130-MS.

36. Mahmoud, A.A., Elkatatny, S., Mahmoud, M., Abouelresh, M.,

Abdulraheem, A., and Ali, A., 2017 Determination of the total
organic carbon (TOC) based on conventional well logs using artificial neural
network. International Journal of Coal Geology 179, 72-80.
https://doi.org/10.1016/j.coal.2017.05.012.

37. Mahmoud, A.A., Elkatatny, S., and Al-Shehri, D., 2020. Application of
Machine Learning in Evaluation of the Static Young’s Modulus for
Sandstone Formations. Sustainability 12(5).
https://doi.org/10.3390/su12051880.

38. Mahmoud, A.A., Elkatatny, S., Ali, A., and Moussa, T., 2019.
Estimation of Static Young’s Modulus for Sandstone Formation Using

33
 References

Artificial Neural Networks. Energies 12, 2125,

https://doi.org/10.3390/en12112125.

39. Tariq, Z., Elkatatny, S., Mahmoud, M. and Abdulraheem, A., 2016,
November. A holistic approach to develop new rigorous empirical
correlation for static Young's modulus. This paper was presented at Abu
Dhabi International Petroleum Exhibition & Conference. Abu Dhabi, UAE,
7-10 November. SPE-183545-MS. https://doi.org/10.2118/183545-MS.

40. Elkatatny, S., Mahmoud, M., Mohamed, I. and Abdulraheem, A., 2018.
Development of a new correlation to determine the static Young’s modulus.
Journal of Petroleum Exploration and Production Technology, 8(1), pp.17-
30. https://doi:10.1007/s13202-017-0316-4.

41. Elkatatny, S., Tariq, Z., Mahmoud, M., Mohamed, I. and Abdulraheem,
A., 2018. Development of new mathematical model for compressional and
shear sonic times from wireline log data using artificial intelligence neural
networks (white box). Arabian Journal for Science and Engineering,
43(11), pp.6375-6389. https://doi.org/10.1007/s13369-018-3094-5.

42. Tariq, Z., Elkatatny, S., Mahmoud, M., Ali, A.Z. and Abdulraheem, A.,
2017, June. A new approach to predict failure parameters of carbonate rocks
using artificial intelligence tools. This paper was presented at SPE Kingdom
of Saudi Arabia Annual Technical Symposium and Exhibition.
Dammam, Saudi Arabia, 24-27 April. SPE-187974-MS.
https://doi.org/10.2118/187974-MS.

43. Alsaihati, A., Elkatatny, S., Mahmoud, A.A. and Abdulraheem, A.,
2020. Use of Machine Learning and Data Analytics to Detect Downhole
Abnormalities While Drilling Horizontal Wells, With Real Case Study.
Journal of Energy Resources Technology, 143(4).
https://doi.org/10.1115/1.4048070.

44. Arehart, R.A., 1990. Drill-bit diagnosis with neural networks. SPE
Computer Applications 2, 24-28.https://doi.org/10.2118/19558-PA.

45. Abdelgawad, K., Elkatatny, S., Moussa, T., Mahmoud, M., and Patil, S.,
2018. Real Time Determination of Rheological Properties of Spud Drilling
34
 References

Fluids Using a Hybrid Artificial Intelligence Technique. Journal of Energy

Resources Technology. https://doi.org/10.1115/1.4042233.

46. Elkatatny, S.M., 2017. Real Time Prediction of Rheological Parameters

of KCl Water-Based Drilling Fluid Using Artificial Neural Networks.
Arabian Journal of Science and Engineering 42, 1655-1665.
https://doi.org/10.1007/s13369-016-2409-7.

47. Alsabaa, A., Gamal, H., Elkatatny, S. and Abdulraheem, A., 2020. Real-
Time Prediction of Rheological Properties of Invert Emulsion Mud Using
Adaptive Neuro-Fuzzy Inference System. Sensors, 20(6), p.1669.
https://doi.org/10.3390/s20061669.

48. Alsabaa A, Gamal H, Elkatatny SM, Abdulraheem A. Real-Time

Prediction of Rheological Properties of All-Oil Mud Using Artificial
Intelligence. This paper was presented at the 54th US Rock
Mechanics/Geomechanics Symposium 2020 Sep 18. American Rock
Mechanics Association. ARMA-2020-1645.

49. Elkatatny, S., Tariq, Z. and Mahmoud, M., 2016. Real time prediction of
drilling fluid rheological properties using Artificial Neural Networks visible
mathematical model (white box). Journal of Petroleum Science and
Engineering, 146, pp.1202-1210.
https://doi.org/10.1016/j.petrol.2016.08.021.

50. Ahmadi, M.A., 2016. Toward reliable model for prediction Drilling
Fluid Density at wellbore conditions: A LSSVM model. Neurocomputing,
211, pp.143-149. https://doi.org/10.1016/j.neucom.2016.01.106.

51. Ahmadi, M.A., Shadizadeh, S.R., Shah, K. and Bahadori, A., 2018. An
accurate model to predict drilling fluid density at wellbore
conditions. Egyptian Journal of Petroleum, 27(1), pp.1-10.
http://dx.doi.org/10.1016/j.ejpe.2016.12.002.

52. Alkinani, H.H., Al-Hameedi, A.T.T., Dunn-Norman, S., Al-Alwani,

M.A., Mutar, R.A. and Al-Bazzaz, W.H., 2019, October. Data-Driven
Neural Network Model to Predict Equivalent Circulation Density ECD.
Paper presented at the Gas & Oil Technology Showcase and
35
 References

Conference, Dubai, UAE, 21-23 October. SPE-198612-MS.

https://doi.org/10.2118/198612-MS.

53. Rahmati, A.S. and Tatar, A., 2019. Application of Radial Basis Function
(RBF) neural networks to estimate oil field drilling fluid density at elevated
pressures and temperatures. Oil & Gas Science and Technology-Revue
d’IFP Energies nouvelles, 74, p.50.

54.Carlos Gershenson. “Artificial Neural Networks for Beginners,”

arxiv.org.
55.Kuldeep S, Dr. Anitha G S. “Neural Network Approach for Processing
Substation Alarms,” International Journals of Power Electronics Controllers
and Converters.
56.M. Abdelrahman. “Artificial neural networks based steady state security
analysis of power systems,” Thirty-Sixth Southeastern Symposium on
System Theory, 2004.
57.O.S. Eluyode, Dipo Theophilus A. “Comparative Study of Biological and
Artificial Neural Networks,” European Journal of Applied Engineering and
Scientific Research, 2(1), pp. 36-46, 2013.
58.K Y Lee, Y T Cha, J H Park. “Short Term Load Forecasting Using an
Artificial Neural Network,” Transactions on Power Systems, Vol.1.
59.Demuth H, Beale M. “Neural network toolbox for use with MATLAB,”
2002.

36
 APPENDIX
Artificial Neural
Network Graphs
 APPENDIX

CASE 3: Using hyperbolic tangent as activation function in layer one

(Hidden layer) and pureline function in layer two (out layer) with 4
neuron in hidden layer.

0
200
400
600
800
Depth (ft)

1000
1200
1400
1600
1800
2000
8.65 8.7 8.75 8.8 8.85 8.9 8.95 9 9.05

ECD (ppg)
ECD Predicted (ppg) ECD Actual (ppg)

Fig. (A.1): Predicted and actual ECD vs. test number from training network for
hyperbolic tangent and pureline with 4 neurons.
9.05

9 y = 0.9858x + 0.1277
R² = 0.9736
8.95
ECD Actual (ppg)

8.9

8.85

8.8

8.75

8.7

8.65
8.65 8.7 8.75 8.8 8.85 8.9 8.95 9 9.05
ECD Predicted (ppg)

Fig. (A.2): Predicted vs. actual ECD for all stages for hyperbolic tangent and
pureline with 4 neurons.

37
 APPENDIX

Table (A.1): Results from BPNN for determining ECD with 4 neurons for
hyperbolic tangent.

Stage R R2 SSE (%)

Training 0.986 0.9735 0.046

Table (A.2): Weights and biases of suggested empirical model for Eqs. (3.1)&(3.3)

Input Hidden
Output
Hidden Layer Layer
Input Layer Weight Matrix Layer
Layer Bias Weight Bias
Neurons ( ) Vector
(b1) Vector
(i) ( ) (b2)
1 -0.284 -0.496 -0.077 0.393 -0.632

2 -0.214 -0.502 -0.365 0.495 -0.580

0.313
3 0.352 -0.638 -0.143 0.597 -0.455

4 -0.22 0.107 0.248 0.259 0.216

38
 Appendix

CASE 4: Using hyperbolic tangent as activation function in layer one (Hidden layer)
and pureline function in layer two (out layer) with 6 neuron in hidden layer.

0
200
400
600
800
Depth (ft)

1000
1200
1400
1600
1800
2000
8.65 8.7 8.75 8.8 8.85 8.9 8.95 9 9.05
ECD (ppg)
ECD Predicted (ppg) ECD Actual (ppg)

Fig. (A.3): Predicted and actual ECD vs. test number from training network for
hyperbolic tangent and pureline with 6 neurons.

9.05

9 y = 0.9308x + 0.6119
R² = 0.9545
8.95
ECD Actual (ppg)

8.9

8.85

8.8

8.75

8.7

8.65
8.65 8.7 8.75 8.8 8.85 8.9 8.95 9 9.05
ECD Predicted (ppg)

Fig. (A.4): Predicted vs. actual ECD for all stages for hyperbolic tangent and
pureline with 6 neurons.

39
 Appendix

Table (A.3): Results from BPNN for determining ECD with 6 neurons for
hyperbolic tangent.

Stage R R2 SSE (%)

Training 0.9769 0.9545 0.085

Table (A.4): Weights and biases of suggested empirical model for Eqs. (3.1)&(3.3).

Input Hidden
Output
Hidden Layer Layer
Layer
Layer Input Layer Weight Matrix Weight
Bias Bias
Neurons ( ) Vector Vector
(i) (b1) ( ) (b2)

1 -0.301 0.343 0.3 0.455 -0.003

2 -0.128 0.594 -0.043 -0.222 0.509

3 -0.009 0.41 0.232 -0.418 0.572

4 0.203
-0.2 -0.31 -0.512 -0.01 -0.616

5 -0.302 -0.158 -0.152 0.465 -0.371

6 -0.503 0.048 0.243 -0.034 -0.091

41
 Appendix

CASE 5: Using logsigmoid as activation function in layer one (Hidden layer) and
pureline function in layer two (out layer) with 4 neuron in hidden layer.

0
200
400
600
800
Depth (ft)

1000
1200
1400
1600
1800
2000
8.65 8.7 8.75 8.8 8.85 8.9 8.95 9 9.05
ECD (ppg)
ECD Predicted (ppg) ECD Actual (ppg)

Fig. (A.5): Predicted and actual ECD vs. test number from training network for
logsigmoid and pureline with 4 neurons.

41
 Appendix

9.05

9
y = 1.0157x - 0.1367
R² = 0.9252
8.95
ECD Actual (ppg)

8.9

8.85

8.8

8.75

8.7

8.65
8.65 8.7 8.75 8.8 8.85 8.9 8.95 9
ECD Predicted (ppg)

Fig. (A.6): Predicted vs. actual ECD for training stage for logsigmoid and
pureline with 4 neurons.

Table (A.5): Results from BPNN for determining ECD with 4 neurons for
logsigmoid.
Stage R R2 SSE (%)
Training 0.961 0.9252 0.028

42
 Appendix

Table (A.6): Weights and biases of proposed empirical model for eq. (3.4).
Input Hidden
Output
Hidden Layer Layer
Input Layer Weight Matrix Layer
Layer Bias Weight Bias
Neurons ( ) Vector
(b1) Vector
(i) ( ) (b2)
1 -0.136 -1.467 0.172 -0.122 -1.202

2 -0.46 -0.05 -0.31 -0.452 -0.064

0.372
3 0.124 1.142 0.577 -0.521 1.289

4 -0.105 -0.334 -0.801 0.57 -0.795

CASE 6: Using logsigmoid as activation function in layer one (Hidden

layer) and pureline function in layer two (out layer) with 6 neuron in
hidden layer.

0
200
400
600
800
Depth (ft)

1000
1200
1400
1600
1800
2000
8.6 8.65 8.7 8.75 8.8 8.85 8.9 8.95 9 9.05
ECD (ppg)
ECD Predicted (ppg) ECD Actual (ppg)

Fig. (A.7): Predicted and actual ECD vs. test number from training network for
logsigmoid and pureline with 6 neurons.

43
 Appendix

9.05

9 y = 0.9284x + 0.6381
R² = 0.9358
8.95
ECD Actual (ppg)

8.9

8.85

8.8

8.75

8.7

8.65
8.6 8.65 8.7 8.75 8.8 8.85 8.9 8.95 9 9.05
ECD Predicted (ppg)

Fig. (A.8): Predicted vs. actual ECD for training stage for logsigmoid and
pureline with 6 neurons.

Table (A.7): Results from BPNN for determining ECD with 6 neurons for
logsigmoid.
Stage R R2 SSE (%)
Training 0.967 0.9358 0.019

44
 Appendix

Table (A.8): Weights and biases of proposed empirical model for eq. (3.4).

Input Hidden
Output
Hidden Layer Layer
Input Layer Weight Matrix Layer
Layer Bias Weight Bias
Neurons ( ) Vector
(b1) Vector
(i) ( ) (b2)
1 0.411 0.587 0.411 -0.433 1.063

2 0.31 -0.468 -0.386 -0.05 -1.118

3 0.295 0.394 0.09 -0.486 0.685

-0.822
4 0.534 0.775 0.401 -0.548 0.913

5 -0.019 1.029 -0.247 -0.431 0.423

6 -0.223 -0.338 0.717 0.375 0.043

45
 ‫ستخلص‬
‫ال ُم َ‬
‫حسخخذو حقُيبث انزكبء األصطُبعي عهى َطبق واسع في صُبعت انُفط وانغبص‪.‬‬

‫حى في هزِ انذساست اسخخذاو حقُيبث انشبكبث انعصبيت االصطُبعيت ببسخخذاو بشَبيج ‪spss‬‬
‫نهخُبؤ بُسبت كثبفت انخذويش انًكبفئت ‪ .‬هزِ انذساست أعخًذث عهى بيبَبث يأخىرة يٍ دساست عًهيت‬
‫نذساست حأثيش انعىايم انًهًت عهى كثبفت انخذويش انًكبفئت‪ .‬حى إسخخذاو ثالد دوال وبعذد َيشوَبث‬
‫يخخهفت في حذسيب انشبت انعصبيت األصطُبعيت‪ .‬هزِ انذوال حى إسخخذايهب في سخت حبالث‪ ,‬في انحبنت‬
‫األونى وانثبنثت وانشابعت حى اسخخذاو دانت انظم انضائذيت نهخعبيم يع انًذخالث بيًُب أُسخخذيج انذانت‬
‫انخطيت نهخعبيم يع يخشجبث انبشَبيج‪ ,‬أيب في انحبنت انثبَيت وانخبيست وانسبدست فقذ أُسخخذيج انذانت‬
‫انهىغبسحًيت نهخعبيم يع انًذخالث بيًُب حى إسخخذاو انذانت انخطيت نهخعبيم يع انًخشجبث‪.‬‬

‫حى إسخخذاو يعبيم األسحببط ويعذل َسبت انخطأ نخقييى انُخبئج انًسخحصهت‪ .‬حبيٍ يٍ انُخبئج أٌ‬
‫جًيع انذوال وجًيع انحبالث انًسخخذيت أعطج َخبئج جيذة يٍ خالل قيى األسحببط انعبنيت انًسخحصهت‬
‫أخيشا ‪ ،‬يًكٍ اسخخذاو طشيقت انشبكت انعصبيت االصطُبعيت نخقذيش‬
‫ً‬ ‫يٍ جًيع انذوال وجًيع انحبالث‪.‬‬
‫يعقذة‪.‬‬ ‫إجشاءاث‬ ‫وبذوٌ‬ ‫جيذة‬ ‫هُذسيت‬ ‫وبذقت‬ ‫انًكبفئت‬ ‫انخذويش‬ ‫كثبفت‬ ‫َسبت‬