PO198646 DOI: 10.

2118/198646-PA Date: 15-July-20 Stage: Page: 1 Total Pages: 23

Scale-Prediction/Inhibition Design Using

Machine-Learning Techniques and
Probabilistic Approach
Nasser M. Al-Hajri and Abdullah Al-Ghamdi, Saudi Aramco, and Zeeshan Tariq
and Mohamed Mahmoud, King Fahd University of Petroleum & Minerals

Summary
This paper presents a data-driven methodology to predict calcium carbonate (CaCO3)-scale formation and design its inhibition program
in petroleum wells. The proposed methodology integrates and adds to the existing principles of production surveillance, chemistry,
machine learning (ML), and probability theory in a comprehensive decision workflow to achieve its purpose. The proposed model was
applied on a large and representative field sample to verify its results.
The method starts by collecting data such as ionic composition, pH, sample-collection/inspection dates, and scale-formation event.
Then, collected data are classified or grouped according to production conditions. Calculation of chemical-scale indices is then made
using techniques such as water-saturation level, Langelier saturation index (LSI), Ryznar saturation index (RSI), and Puckorius scaling
index (PSI). The ML part of the method starts by dividing the data into training and test sets (80 and 20%, respectively). Classification
models such as support-vector machine (SVM), K-nearest neighbors (KNN), gradient boosting, gradient-boosting classifier, and
decision-tree classifier are all applied on collected data. Prediction results are then classified into a confusion matrix to be used as
inputs for the probabilistic inhibition-design model. Finally, a functional-network (FN) tool is used to predict the formation of scale.
The scale-inhibition program design uses a probabilistic model that quantifies the uncertainty associated with each ML method. The
scale-prediction capability compared with actual inspection is presented into probability equations that are used in the cost model. The
expected financial impact associated with applying any of the ML methods is obtained from defining costs for scale removal and scale
inhibition. These costs are factored into the probability equations in a manner that presents incurred costs and saved or avoided
expenses expected from field application of any given ML model. The forecasted cost model is built on a base-case method (i.e., current
situation) to be used as a benchmark and foundation for the new scale-inhibition program.
As will be presented in the paper, the results of applying the preceding techniques resulted in a scale-prediction accuracy of 95%
and realized threefold cost-savings figures compared with existing programs.

Introduction
Scale precipitation causes several issues in oil and gas fields. There are different causes and sources behind the scale formation in differ-
ent wells. Seawater injection can be considered one of the primary causes of sulfate-scale precipitation because of the mixing of two
different incompatible waters. Scales can be formed in well tubulars as well as the formation. Scale formation in well tubulars can lead
to serious operational issues in which conventional workover rigs might not be sufficient to do the job because of the weight of the
tubing with the scale (Bayona 1993).
Scales also can be formed because of the change in thermodynamic conditions in the reservoir or in the well tubulars. Changes in
pressure, temperature, partial pressure of carbon dioxide (CO2)/hydrogen sulfide, and pH can also cause scale precipitation (Moghadasi
et al. 2003a, 2003b; Mackay et al. 2003).
Fig. 1 shows the constants of the solubility product for magnesium carbonate, calcium carbonate, barium carbonate, and strontium car-
bonate scales. At low temperatures (less than 150 F), strontium carbonate has very low solubility constant, while magnesium carbonate
has the highest one. CaCO3 scales are very common compared with other types of carbonate scales because CaCO3 scale forms anywhere
in the well. It can form inside the reservoir, inside the production tubing, and at the downhole-pump intakes (which cause plugging to the
electrical-submersible pump and degrades its efficiency). In addition, CaCO3 scale can form at the wellhead and in the surface flow lines
(Lakshmi et al. 2013). Calcite has a very low solubility constant at high temperature (Fig. 1), which makes it the most-stable scale type at
the reservoir pressure and temperature (Kamal et al. 2018). CaCO3 scale has different forms such as calcite, vaterite, and aragonite, and
calcite is the common type that exists in the reservoir. The formation of CaCO3 scale depends on several conditions, such as temperature;
pH of the medium; ion concentration, such as calcium and bicarbonate; and ionic strength. The abundance of CO2 in water and the pH of
the medium control the formation of CaCO3 scales. CO2 reacts with water and produces carbonic acid, which is a weak acid that will dis-
sociate to bicarbonate. The abundance of bicarbonate and calcium ions will promote the formation of CaCO3 scales. At higher pH values
and higher temperature, the formation of CaCO3 will be accelerated (Ramstad et al. 2005; Hamid et al. 2016). Iron carbonate (siderite)
can form in the reservoirs and downhole equipment (Amiri et al. 2013). Similar conditions that promote the formation of CaCO3 scales
promote the formation of iron carbonate scales as well. These conditions include temperature change, pressure change, and CO2-solubility
change with pressure and temperature. The liberation of CO2 from the solution will increase the pH of the medium, and this will promote
the carbonate-scale formation by reducing the solubility of these mineral scales at high pH values (Jordan et al. 2014).
CaCO3 is a type of oilfield scale that is classified as inorganic. CaCO3 scale commonly occurs in many fields around the globe. This
type of inorganic scale forms under different conditions (thermodynamic, kinetic, and hydrodynamic) because of the mixing of petro-
leum fluids. CaCO3-scale deposit is caused by a shift toward carbonate in the carbonate/bicarbonate/CO2 equilibrium. When equilib-
rium shifts in the other direction, the precipitation goes back into solution. In a chemical reaction, chemical equilibrium is the state in
which both reactants and products are present in concentrations that have no further tendency to change with time. Chemical equilib-
rium is achieved when the rate of forward reaction is the same as the rate of reverse reaction.

Copyright V
C 2020 Society of Petroleum Engineers

This paper (SPE 198646) was accepted for presentation at the SPE Gas & Oil Technology Showcase and Conference, Dubai, UAE, 21–23 October 2019, and revised for publication. Original
manuscript received for review 16 March 2020. Revised manuscript received for review 28 April 2020. Paper peer approved 5 May 2020.

2020 SPE Production & Operations 1

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 2 Total Pages: 23

1.00×10–6
MgCO3
CaCO3
BaCO3
SrCO3

Solubility-Product Constant
1.00×10–7

1.00×10–8

1.00×10–9

1.00×10–10
50 100 150 200 250 300
Temperature (°F)

Fig. 1—Solubility of carbonate scale (after Li et al. 1995). MgCO3 5 magnesium carbonate; BaCO3 5 barium carbonate;
SrCO3 5 strontium carbonate.

The ability to predict scale formation is a major challenge in the oil industry. According to Vetter et al. (1987), “the main variables
dictating the location and amount of CaCO3-scale deposition in an oil field are as follows:
• Pressures and temperatures at any location within the entire production system.
• The brine and oil compositions prior, during and after the reservoir fluids have been exposed to temperature and/or
pressure changes.
• The bubble point and flash behavior of the three-phase oil/brine/gas system as a function of pressure and temperature.
• The distribution of CO2 between oil and brine phases and the dramatic variations of this CO2 partitioning prior and during any
production operation.
• The constant variation of the water oil ratio, the gas oil ratio and the gas water ratio during any production operation.”
Langelier developed an equation setting forth the conditions of carbonate equilibrium. By using this equation, the pH value of water
at equilibrium can be calculated. If the actual pH is higher than the calculated pH, the water has a tendency to form scale. If it is lower,
the water has a tendency to be corrosive. Langelier formulated an equation relating the stability index with pH, calcium concentration,
and total alkalinity. Alkalinity is the ability of a solution to neutralize an acid to the equivalence point of CaCO3. Alkalinity can be cal-
culated as the sum of ion concentrations ½HCO 3  þ 2x½CO3
ð2Þ
 þ ½OH   ½Hþ . The stability index, if it is a positive index, indicates
scale formation. A negative index indicates corrosion. This equation has been shown to apply to waters with total solid concentration as
high as 4,000 ppm. Larson and Buswell (1942) improved the reliability of the Langelier index by adjusting their own index for the tem-
perature and salinity effects during the CaCO3 precipitation at atmospheric pressure.
Stiff and Davis (1952) proposed an empirical method to extend the application of the Langelier equation to waters of high salt con-
centration. This was done by experimentally deriving that the value of the K term in the Langelier equation applies to waters of high
salt content. Using this equation, the tendency of oilfield waters to deposit CaCO3 can be predicted.
SI ¼ pH  K  pCa  pAlk ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð1Þ

where K is obtained as function of ionic strength (l), and the ionic strength can be determined as

l ¼ 0:5ðC1 V12 þ C2 V22 þ       þ Cn Vn2 Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð2Þ

where pCa is the negative logarithm of the calcium concentration; pAlk is the negative logarithm of the total alkalinity; l is ionic
strength; C is the concentration of each ion (in g ion/1000 g solvent); and V is the valence of the ion.
Yeboah et al. (1993) developed the oilfield scale-prediction model, which predicts the potential and deposition profile of scale using
extensive thermodynamic and kinetic data. The model uses experimental solubility data to determine the saturation index. Critical satur-
ation indices beyond which scaling occurs have been established. The model uses the flow characteristics and experimental kinetic data
to predict the scale-deposition profile from the bottomhole to the surface once the critical saturation index is exceeded.
2
½Ca2þ ½HCO3
SI ¼ log ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð3Þ
Ksp PCO2

where SI is the CaCO3 saturation index, Ksp is the equilibrium constant, and PCO2 is the partial pressure of CO2.
Vetter et al. (1987) highlighted the importance of flashed gases from both the oil and the brine phases in CaCO3-scale formation,
which is ignored in many predictive models in the industry. In addition, the authors discussed the various effects of the gas distribution
(especially CO2) between oil and brine phases under reservoir and various production conditions on CaCO3 formation and provided
some algorithms for prediction. They also presented a methodology that can be used to predict CaCO3 formation under field conditions
as a function of water composition, pressure, temperature, water/oil ratio (WOR), gas/oil ratio (GOR), and total CO2, and its partition-
ing between the various liquid phases.
Hamid et al. (2016) developed an empirical model using weight-gain data from coupon tests and from a tube-plugging test. This
model was able to predict scale-growth rate at a given point on a solid surface with pressure, pressure gradient, temperature, fluid veloc-
ity, and brine concentration as independent variables. Artificial-intelligence (AI) methodology was used to develop the model. The
scale-growth field y at a given time is obtained by integrating the scaling-rate function,
y_ ¼ F1F2 þ F3F4; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð4Þ

2 2020 SPE Production & Operations

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 3 Total Pages: 23

where y_ is the rate of scale growth at a point on the solid surface (in in./D); F1 depends on an empirical model F0l F0, which is derived from
the artificial neural network of the scale deposition as a function of five variables (pressure, temperature, brine concentration, velocity, and
time); F2 is the locations at which scale would be deposited because of the surface position with respect to gravity and the surface being con-
cave or convex; and F3 and F4 account for boned scaling, which is independent of gravity and a strong function of the pressure gradient.
The model output was compared with experimental data and gave fair results. This model was used to predict scale formation at
inflow-control valves, allowing for a better design completion and fluid-handling system for presalt wells.
Bahadori (2011) created a simple predictive tool to predict CaCO3-scale precipitation as a function of pH, temperature, ionic
strength of the solution, calcium-cation concentration, bicarbonate-anion concentration, and CO2 mole fraction when the water mixture
is saturated with gas containing CO2 to evaluate the effects of solution conditions on the tendency and extent of precipitation. This
method covered concentrations of CaCO3 from 10 to 10 000 mg/L, with temperature ranging from 5 to 90 C, total ionic strength rang-
ing between 0.1 and 3.6, and pH values ranging from 5.5 and 8. This model was created for CO2 sequestration in saline aquifers to
reduce CO2 emissions into the atmosphere. The steps to predict CaCO3-scale precipitation are
lnðKÞ ¼ a þ bI þ cI2 þ dI3 ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð5Þ

where a, b, c, and d are constants, and

pCa ¼ 8:922ðCCa2þ Þ0:2708 ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð6Þ

pAlk ¼ 8:1997ðCHCO3 þCO3 Þ0:23638 ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð7Þ

pHs ¼ pCa  pAlk  K; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð8Þ

lnðSf Þ ¼ a þ bT þ cT2 þ dT3 ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð9Þ

R0 ¼ ðCHCO3  0:82Þ=ðXCO2  Sf Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð10Þ

pH ¼ 0:4341lnðR0 Þ þ 6:2964; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð11Þ

SI ¼ pH  pHs ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð12Þ

where K is the correction factor for total ionic strength and temperature; I is total ionic strength; pH is the actual pH value in the
system; pHs is the pH value when CaCO3 achieves saturation in the system; Sf is the solubility factor; T is temperature (in K); and XCO2
is the CO2 mole fraction in the water mixture saturated with a gas containing CO2.
The objectives of this paper are to develop a ML workflow to predict scale precipitation in oil and gas wells. Several data sets will
be used as inputs to the model and the model output will be either matrix prediction or an empirical equation that will be developed by
the FN ML technique. In addition, a probabilistic framework was developed to design the inhibition program according to cost.

ML Techniques
The terms ML and AI are often used interchangeably, but there is a difference between the two techniques: ML is more related to pre-
diction, while AI is more concerned with the actions for creating machines using human intelligence. ML and AI both are captivating
fields that integrate computational power with human intelligence to produce smart and reliable solutions for extremely nonlinear and
highly complicated problems. In the last 2 decades, petroleum-engineering journals have been overwhelmed with articles using AI and
ML for regression, function approximation, and classification problems. More details about AI and ML can be found in our previous
publications (Tariq et al. 2018a, 2018b, 2019). The focus of our work is centered around the use of ML techniques such as the KNN
algorithm, SVM, gradient boosting, gradient-boosting classifier, and decision tree.
• KNN: In pattern recognition, the KNN algorithm is a method used for classification and regression. In both cases, the input con-
sists of the k closest training examples in the feature space. The output depends on whether KNN is used for classification
or regression.
• SVM: In ML, SVMs are supervised learning models with associated learning algorithms that analyze data used for classification
and regression analysis. Given a set of training examples, each marked as belonging to one or the other of two categories, an
SVM training algorithm builds a model that assigns new examples to one category or the other, making it a nonprobabilistic
binary linear classifier.
• Gradient boosting: An ML technique for regression and classification problems that produces a prediction model in the form of an
ensemble of weak prediction models, typically decision trees. It builds the model in a stagewise fashion, like other boosting meth-
ods do, and it generalizes them by allowing optimization of an arbitrary function.
• Gradient-boosting classifier: A view of boosting algorithms that optimize a cost function over function space by iteratively choos-
ing a weak hypothesis that points in the negative-gradient direction. This functional-gradient view of boosting has led to the devel-
opment of boosting algorithms in many areas of ML and statistics beyond regression and classification.
• Decision-tree classifier: Decision-tree learning uses a decision tree (as a predictive model) to go from observations about an item
(represented in the branches) to conclusions about the item’s target value (represented in the leaves). It is one of the predictive-
modeling approaches used in statistics, data mining, and ML.
The other technique used in this study, the FN tool, is to predict the existence of scale formation. An FN technique is the relatively
latest AI technique, first discussed by Castillo et al. (1999) for a function-approximation problem. The basis of FN is the combination
of artificial neural networks with the functional theory. It is a supervised data-learning technique mostly used for prediction and
regression purposes.

Methodology
The proposed workflow describes the CaCO3-scale-prediction methodology, from data collection to model building. Furthermore, the
process is extended to include a probabilistic framework by which fieldwide scale-inhibition programs are designed. The steps are
as follows:

2020 SPE Production & Operations 3

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 4 Total Pages: 23

1. Data collection. Water samples were collected from wet wells to extract the following inputs:
• Ionic composition of water; example ions are Ca2þ, Naþ, HCO 3 , and so forth.
• pH, which is the 10th logarithm of Hþ concentration.
• Sample-collection date.
• Scale-inspection date. Each case study must have a valid scale-inspection date to verify the scaling-tendency-prediction
results. The inspection location is the manifold-choke valve after dropping the valve spool. The manifold-choke valve is typi-
cally located a few feet away from the production wellhead.
• Scale-formation event: This is to record whether the scale has precipitated, according to the inspection results.
2. Classify the collected data by field. This is performed to achieve as much similarity as possible in the production conditions,
such as flow rate and pressure drop. In the presence of abundant data, it is recommended to group the wells according to their
production performance and water cut.
3. Define the output events. For each case study, denote the scaling event as follows:
• 1 if the scale has occurred in the well within X number of years.
• 0 if the scale has not occurred in the well within X number of years.
X is selected according to the scale-inhibition-program-design criteria and economics, such as the type of the inhibition chemi-
cal, the chemical-residue life, field size, and so forth. For the purposes of the data set used to demonstrate the concept of this
method, an X amount of 10 years was selected. This number represents the scale-protection time (when the inhibition is per-
formed correctly) for existing oilfield chemicals used at the fields under study. However, time-period applications other than
10 years can be used by using data sets within the desired period in the ML prediction model. For example, if the required scale-
prediction duration is 2 years, then the ML models would only be fed with geochemical-analysis data collected within 2 years
of the physical-inspection dates to train and test the models.
4. Calculate the water-saturation level. The equation to calculate the saturation level is

½Ca2þ  ½CO2
3 
SL ¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð13Þ
Ksp

where ½Ca2þ  is the molar concentration of the calcium cation (in mol/L), ½CO2
3  is the molar concentration of the carbonate
anion (in mol/L), and Ksp is the equilibrium constant.
5. Calculate the LSI. The equations to calculate the LSI are

LSI ¼ pH  pHs ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð14Þ

pHs ¼ ð9:3 þ A þ BÞ  ðC þ DÞ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð15Þ

½log10 ðTDSÞ  1
A¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð16Þ
10
B ¼ 13:12  log10 ½TðCÞ þ 273 þ 34:55; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð17Þ

C ¼ log10 ð½Ca2þ as CaCO3 Þ  0:4; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð18Þ

D ¼ log10 ðalkalinity as CaCO3 Þ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð19Þ

6. Calculate the RSI. The equation to calculate RSI is

RSI ¼ 2pHs  pH: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð20Þ

7. Calculate the PSI. The equation to calculate the PSI is

PSI ¼ 2pHs  pHeq ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð21Þ

pHeq ¼ 1:465  log10 ðalkalinityÞ þ 4:54: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð22Þ

8. Divide the data. After calculating all input parameters in Steps 4 through 7, divide the data randomly into 80% training and 20%
testing sets.
9. Train and test the ML models.
10. Generate the prediction results. For each method discussed previously, generate a confusion matrix that describes the method of
prediction performance. An example matrix is listed in Table 1.

Predicted Event: Predicted Event:

Scale No Scale
Actual event: scale X1 X2
Actual event: no scale X3 X4

Table 1—Example confusion matrix for scale prediction.

Define the probabilistic model. The following probabilistic equations are used to quantify the uncertainty associated with each
method, if implemented to select scale-inhibition candidates:

Xt ¼ Total number of cases or wells in a given field ¼ X1 þ X2 þ X3 þ X4 ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð23Þ

4 2020 SPE Production & Operations

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 5 Total Pages: 23

Number of cases that formed scale X1 þ X2

Ps ¼ Probability of scale formation ¼ ¼ ; . . . . . . . . . . . . . . . . . . . . . . . ð24Þ
Total number of cases in a given field Xt
Number of cases that did not form scale X3 þ X4
PNs ¼ Probability of no scale formation ¼ ¼ ¼ 1  Ps; . . . . . . . . . . . . ð25Þ
Total number of cases in a given field Xt
Number of correctly predicted scale-formation cases X1
Pcsp ¼ Probability of correct scale-formation prediction ¼ ¼
Total number of cases in a given field Xt :
                                       ð26Þ

In this case, scale has actually formed, and the model correctly predicted the event.
Number of incorrectly predicted scale-formation cases X2
PIsp ¼ Probability of incorrect scale-formation prediction ¼ ¼
Total number of cases in a given field Xt :
                                       ð27Þ

In this case, scale has actually formed, and the model did not correctly predict the event.
PCNsp ¼ Probability of correct no-scale-formation prediction
Number of correctly predicted no-scale-formation cases X4
¼ ¼                               ð28Þ
Total number of cases in a given field Xt :
In this case, scale has not actually formed, and the model correctly predicted the event.
PINsp ¼ Probability of incorrect no-scale-formation prediction
Number of incorrectly predicted no-scale-formation cases X3
¼ ¼                              ð29Þ
Total number of cases in a given field Xt :
In this case, scale has not actually formed, and the model did not correctly predict the event.
11. Define the cost model. Eqs. 30 through 34 are used to quantify the expected financial effect associated with each ML method, if
implemented to select scale-inhibition candidates. The positive sign indicates incurred cost and the negative sign indicates
saved/avoided cost.

Ccsp ¼ Expected net saving per well ¼ Pcsp ðCT  Cs Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð30Þ

where Cs is the cost of scale removal per well and CT is the cost of scale-inhibition treatment per well.
The logic of Eq. 30 is that the correct prediction of a scale-formation event (i.e., model predicted that scale will form, and
scale has actually formed) leads to avoiding the cost of a scale-removal operation and leads to incurring the cost of scale-
inhibition treatment.

CCNsp ¼ PCNsp CT : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð31Þ

The logic of Eq. 31 is that the correct prediction of no scale-formation event within time period X leads (i.e., model predicted
that no scale will form and scale actually has not formed) to saving the scale-inhibition-treatment operation.

CIsp ¼ PIsp CS : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð32Þ

The logic of Eq. 32 is that incorrect predication of a scale-formation event (i.e., model predicted no scale, although scale has
actually formed) leads to incurring the cost of a scale-removal operation.

CINsp ¼ PINsp CT : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð33Þ

The logic of Eq. 33 is that incorrect predication of no scale-formation event (i.e., model predicted that scale will form, while
scale has actually not formed) leads to incurring the cost of scale-inhibition treatment.

Cp ¼ Xt ðCcsp þ CIsp þ CCNsp þ CINsp Þ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð34Þ

Eq. 34 presents the overall expected net cost (Cp) of the scale-inhibition program.
12. Define the base case. In this step, the base-case expected cost of the scale-inhibition program is calculated. This cost is defined
using the current practice scenario before introducing any prediction workflow/model. A typical base case to benchmark against
is the conservative scenario, where all wet wells are treated to mitigate scale formation. In such a scenario, the base-case cost is
calculated as

Cbc ¼ Xt ðCT  Ps Cs Þ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð35Þ

Eq. 35 presents the overall base-case expected cost of the scale-inhibition program.
13. Apply Steps 11 through 13 on every method in Step 9. After producing a Cp value for each method in Step 9, select the method
that satisfies the following two conditions:
• Cp of the selected method is the minimum of all other methods.
• Cp of the method is less than Cbc.

2020 SPE Production & Operations 5

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 6 Total Pages: 23

14. If the Step 14 conditions are not applicable, repeat Steps 9 through 14 until the two conditions are satisfied. If no convergence is
achieved, proceed with the base-case scenario.
15. Repeat the complete workflow with every new data point collected from the field to update and enhance the prediction models.

Results and Discussion

The proposed methodology in this paper was tested on a total of 486 data sets that are representative of two different fields. The data
sets were from 84 wells (Field A has 40 wells and Field B has 44 wells). The data used for prediction include both single data sets asso-
ciated with single wells, as well as multiple data sets from the same wells. Table 2 presents the data used in this paper. This
would ensure representation of data for the ML prediction models. ML techniques were implemented on all data sets from both groups.
From a cost comparison of these data sets, the cost-model results show additional savings of USD 400 million in the data from Field A
and USD 1.4 million in the data from Field B when adopting the proposed methodology for the scale program. Table 3 presents the pre-
dicted cost analysis using ML.

Na Ca Mg SO4 HCO3 TDS Inspection

Field Well Number Sample Date (ppm) (ppm) (ppm) (ppm) (ppm) (ppm) pH Inspection Date Result
A 2 21 February 2012 29900 9460 1190 569 172 107091 7 2 June 2013 1
A 2 27 February 2011 30400 11200 1380 786 190 114056 6.6 2 June 2013 1
A 2 3 January 2010 32800 11600 1440 365 209 121414 6.9 2 June 2013 1
A 2 11 September 2007 4446 1450 226 932 509 16663 7.5 2 June 2013 1
A 29 28 April 2013 36600 13400 1620 271 239 136630 7.2 27 May 2015 1
A 29 11 June 2012 24700 7670 917 4720 250 88957 7 27 May 2015 1
A 29 6 July 2011 24400 7400 920 665 146 86331 6.9 27 May 2015 1
A 29 5 July 2010 26700 8160 1010 518 227 94615 7.5 27 May 2015 1
A 29 27 May 2006 16400 5410 671 647 153 59581 7.3 27 May 2015 1
A 29 21 May 2005 17700 5430 705 693 189 63117 7 27 May 2015 1
A 56 12 February 2013 41800 13300 1710 842 72 150024 6.5 17 November 2015 1
A 56 25 January 2006 43400 15700 1950 344 154 161548 6.1 17 November 2015 1
A 135 7 November 2013 36600 12900 1680 1240 467 135887 6.9 26 May 2013 1
A 135 31 May 2013 36900 12300 1410 110 135 133555 6.8 26 May 2013 1
A 135 7 February 2006 25700 9430 1290 556 455 96831 7 26 May 2013 1
A 179 17 December 2013 31900 10300 1280 287 473 114940 7 14 April 2014 1
A 179 1 December 2012 42700 11300 1710 731 345 146886 6.7 14 April 2014 1
A 179 4 October 2007 42700 11800 1730 682 218 148230 6.6 14 April 2014 1
A 179 22 March 2006 40700 10100 1500 734 196 137530 6.3 14 April 2014 1
A 179 26 December 2005 40000 11400 1560 473 204 139637 6.4 14 April 2014 1
A 194 30 December 2009 7213 2150 274 520 250 25607 7.7 27 March 2014 1
A 194 3 November 2013 22600 6080 853 580 412 77925 6 27 March 2014 1
A 204 25 January 2011 5959 2340 311 552 744 23306 7 15 May 2016 1
A 204 1 February 2010 18900 6580 886 543 336 70045 6.9 15 May 2016 1
A 204 3 January 2010 12000 4080 556 346 455 44337 7.1 15 May 2016 1
A 204 14 January 2009 10600 3340 480 796 232 38348 7.1 15 May 2016 1
A 204 3 February 2007 12900 4170 590 799 174 46933 7 15 May 2016 1
A 218 6 August 2011 48000 14700 1880 553 191 170324 6.4 12 May 2016 1
A 218 4 November 2009 16500 12000 1580 691 287 81658 7.1 12 May 2016 1
A 218 25 May 2009 18100 6740 897 1100 307 68644 7.4 12 May 2016 1
A 218 26 November 2008 49200 16600 2100 326 183 179409 6.6 12 May 2016 1
A 218 1 August 2007 52100 17800 2260 508 255 190923 6.3 12 May 2016 1
A 225 11 May 2013 29300 8660 1160 315 188 103223 7.3 1 August 2013 1
A 225 28 April 2013 39900 12600 1550 437 206 142593 6.8 1 August 2013 1
A 225 28 April 2013 37000 14500 1590 259 174 140623 7.1 1 August 2013 1
A 225 1 February 2012 17600 5110 780 270 371 62231 7.5 1 August 2013 1
A 225 21 December 2010 36100 12600 1500 890 168 132858 7.1 1 August 2013 1
A 225 16 June 2010 36000 13100 1590 380 140 134210 6.5 1 August 2013 1
A 225 23 February 2009 34900 10100 1350 1020 445 122515 7.1 1 August 2013 1
A 225 25 May 2005 28400 8520 1150 607 226 100603 6.9 1 August 2013 1

Table 2—Geochemical-analysis and physical-scale-inspection data. Na ¼ sodium; Ca ¼ calcium; Mg ¼ magnesium; SO4 ¼ sulfate;
HCO3 ¼ bicarbonate; TDS ¼ total dissolved solids. Inspection result of 1 means scale has formed and a result of 0 means no scale has formed.

6 2020 SPE Production & Operations

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 7 Total Pages: 23

Na Ca Mg SO4 HCO3 TDS Inspection

Field Well Number Sample Date (ppm) (ppm) (ppm) (ppm) (ppm) (ppm) pH Inspection Date Result
A 234 23 January 2013 5490 1930 277 558 310 20665 7.1 21 May 2013 1
A 234 25 January 2012 5244 431 83 1220 247 15275 7.7 21 May 2013 1
A 234 8 May 2011 5230 1990 273 315 262 20070 7.3 21 May 2013 1
A 234 23 June 2010 24700 8660 1060 963 67 91250 6.7 21 May 2013 1
A 234 22 June 2010 5509 1890 283 897 354 20733 7.4 21 May 2013 1
A 234 7 July 2009 4687 1700 267 530 217 17901 7.3 21 May 2013 1
A 234 6 July 2008 5469 1780 275 586 94 20104 7.2 21 May 2013 1
A 234 21 June 2008 5228 1420 229 840 42 18359 7.2 21 May 2013 1
A 234 3 April 2007 6039 1890 293 564 168 21954 7.1 21 May 2013 1
A 234 26 February 2006 6130 2020 305 621 622 22798 7 21 May 2013 1
A 250 20 November 2012 17200 4010 654 398 343 57605 6.9 28 August 2016 1
A 250 22 November 2011 30500 9070 1200 244 317 107631 6.9 28 August 2016 1
A 250 22 January 2011 21700 5020 772 3030 677 73199 7.9 28 August 2016 1
A 250 3 January 2010 31700 7400 1120 591 316 105727 6.6 28 August 2016 1
A 250 2 February 2009 28000 7610 1250 913 476 97649 7.5 28 August 2016 1
A 250 30 May 2005 28400 7470 1270 580 73 98093 6.5 28 August 2016 1
A 250 2 February 2009 28000 7610 1250 913 476 97649 7.5 28 August 2016 1
A 250 3 January 2010 31700 7400 1120 591 316 105727 6.6 28 August 2016 1
A 250 22 November 2011 30500 9070 1200 244 317 107631 6.9 28 August 2016 1
A 250 20 November 2012 17200 4010 654 398 343 57605 6.9 28 August 2016 1
A 250 30 May 2005 28400 7470 1270 580 73 98093 6.5 28 August 2016 0
A 250 22 January 2011 21700 5020 772 3030 677 73199 7.9 28 August 2016 0
A 254 16 March 2013 19300 4290 848 314 294 64446 7.5 14 April 2014 1
A 254 6 March 2013 29700 7430 1160 268 218 100776 7.4 14 April 2014 1
A 254 27 March 2012 26800 7610 1260 231 445 94346 6.7 14 April 2014 1
A 254 26 November 2011 29200 7180 1060 361 520 98521 7.2 14 April 2014 1
A 254 22 January 2011 17700 4290 677 3280 769 60716 7.4 14 April 2014 1
A 254 7 December 2009 28000 6920 1060 627 303 94810 7.1 14 April 2014 1
A 254 8 November 2008 27800 7630 1170 298 422 96620 6.8 14 April 2014 1
A 254 25 August 2007 29600 7690 1180 779 323 101572 6.5 14 April 2014 1
A 255 14 October 2012 19100 3970 861 615 448 63294 7.5 14 April 2014 1
A 255 26 November 2011 25400 4710 1080 636 763 82389 7.2 14 April 2014 1
A 255 30 January 2011 22000 5050 1050 1650 195 74545 7.5 14 April 2014 1
A 255 9 January 2010 18300 5110 682 372 477 63641 7.1 14 April 2014 1
A 255 22 March 2009 28100 6300 1290 780 537 94307 7.4 14 April 2014 1
A 258 1 January 2014 30000 9730 1300 365 490 108585 6.9 15 July 2014 1
A 258 1 January 2013 30300 9200 1310 499 195 107804 7.1 15 July 2014 1
A 258 22 November 2011 29500 8360 1120 293 314 102687 7 15 July 2014 1
A 258 3 January 2010 30500 7340 1090 466 296 102392 6.6 15 July 2014 1
A 258 2 February 2009 27200 8900 1380 810 464 99654 7.5 15 July 2014 1
A 258 18 November 2007 31400 8190 1180 542 395 107407 6.5 15 July 2014 1
A 258 20 September 2006 26800 8860 1230 575 226 97791 6.6 15 July 2014 1
A 258 6 July 2005 27400 9060 1300 618 131 100009 6.9 15 July 2014 1
A 267 29 May 2013 62900 14700 1530 462 240 206832 6.7 27 January 2014 1
A 283 28 April 2013 36600 14200 1600 376 223 138799 7.1 19 May 2014 1
A 283 7 December 2009 14200 3880 503 733 600 49216 7.7 19 May 2014 1
A 283 27 March 2006 19700 6320 721 810 178 70729 6.6 19 May 2014 1
A 283 25 December 2005 16200 5290 618 465 450 58523 7.1 19 May 2014 1
A 317 19 June 2010 3170 1080 179 1160 409 12228 7.5 27 January 2014 1

Table 2 (continued)—Geochemical-analysis and physical-scale-inspection data. Na ¼ sodium; Ca ¼ calcium; Mg ¼ magnesium; SO4 ¼ sulfate;
HCO3 ¼ bicarbonate; TDS ¼ total dissolved solids. Inspection result of 1 means scale has formed and a result of 0 means no scale has formed.

2020 SPE Production & Operations 7

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 8 Total Pages: 23

Na Ca Mg SO4 HCO3 TDS Inspection

Field Well Number Sample Date (ppm) (ppm) (ppm) (ppm) (ppm) (ppm) pH Inspection Date Result
A 317 17 March 2009 2587 1040 161 795 573 10536 7.4 27 January 2014 1
A 317 17 October 2005 4662 1280 198 922 367 16569 7.4 27 January 2014 1
A 342 10 February 2013 17400 5130 825 214 196 61865 7.2 25 June 2013 1
A 342 27 February 2012 40400 12500 1720 132 322 144174 6.8 25 June 2013 1
A 342 19 March 2007 35900 13800 1770 539 199 136708 6.6 25 June 2013 1
A 342 22 November 2005 37400 12100 1570 486 162 135018 6.2 25 June 2013 1
A 361 4 November 2013 17600 4810 655 522 312 60899 7.3 2 April 2014 1
A 361 1 September 2013 15700 4600 685 580 61 55526 6.7 2 April 2014 1
A 361 1 July 2013 14700 4680 465 346 222 52413 7.7 2 April 2014 1
A 361 4 October 2010 17000 4490 576 397 177 58040 6.5 2 April 2014 1
A 361 12 October 2008 17400 4400 650 634 196 59280 7.3 2 April 2014 1
A 361 27 February 2006 10500 3340 465 382 523 38110 6.6 2 April 2014 1
A 361 27 February 2006 10500 3340 465 382 523 38110 6.6 2 April 2014 1
A 361 12 January 2005 43200 6080 790 723 445 130138 7 2 April 2014 1
A 390 19 May 2013 26800 6940 1250 320 293 92403 7.6 14 May 2013 1
A 390 27 October 2010 25700 6650 2520 245 262 93777 7 14 May 2013 1
A 391 28 November 2005 16400 5090 733 486 251 58860 6.9 24 December 2016 1
A 391 31 May 2006 33400 9310 1400 686 140 116436 7.1 24 December 2016 1
A 391 10 September 2007 33700 9560 1420 478 210 117868 6.2 24 December 2016 1
A 391 7 April 2010 26000 6790 1160 461 253 89664 7.3 24 December 2016 1
A 391 10 April 2010 23300 10600 1350 199 350 94099 7 24 December 2016 1
A 391 10 April 2011 25800 6610 1060 441 185 88296 6.8 24 December 2016 1
A 391 6 March 2012 29700 6960 1000 317 337 98914 7.1 24 December 2016 1
A 391 27 April 2013 25600 5550 931 652 630 84563 7.5 24 December 2016 1
A 391 24 March 2010 26200 7850 1070 656 27 92703 6.4 24 December 2016 0
A 391 12 March 2013 16200 5410 754 841 683 59588 7.2 24 December 2016 0
A 397 18 March 2009 41400 14900 2010 2 421 154533 7 15 June 2016 1
A 397 21 August 2007 22400 7590 975 366 381 82112 5.1 15 June 2016 1
A 397 4 December 2006 40200 16100 2060 538 253 155151 6.7 15 June 2016 1
A 397 25 November 2006 39800 15100 1870 642 313 150625 6.6 15 June 2016 1
A 397 5 January 2005 7433 1980 253 804 195 25665 7.7 15 June 2016 1
A 398 28 May 2013 36400 11500 1310 421 64 129695 6.4 17 July 2013 1
A 398 13 May 2013 14900 6970 841 452 275 60638 7.5 17 July 2013 1
A 398 5 March 2012 28800 6030 3050 2080 331 102591 7.1 17 July 2013 1
A 398 20 December 2011 26100 6300 2300 1910 365 93475 7.2 17 July 2013 1
A 398 27 February 2011 25700 5220 5000 3070 379 100269 7.1 17 July 2013 1
A 398 31 March 2010 21400 5840 760 392 351 73843 7.2 17 July 2013 1
A 398 13 November 2007 60600 13200 1410 1020 397 196627 6.6 17 July 2013 1
A 398 16 July 2007 20500 5080 611 661 221 68873 7 17 July 2013 1
A 402 13 May 2013 18300 4350 1040 381 203 62774 7.2 1 July 2013 1
A 402 17 April 2012 26100 7530 1180 1000 490 92300 7 1 July 2013 1
A 402 23 February 2011 18900 4960 890 890 378 65618 7.5 1 July 2013 1
A 402 31 March 2010 33100 10800 1220 401 304 119125 6.7 1 July 2013 1
A 402 23 March 2010 36400 11500 1580 271 123 130674 6.7 1 July 2013 1
A 402 9 October 2007 30600 11800 1390 376 199 116065 6 1 July 2013 1
A 402 4 October 2007 27600 11400 1320 384 211 107115 6.7 1 July 2013 1
A 402 7 February 2006 24600 10900 1310 438 215 98063 6.4 1 July 2013 1
A 447 28 April 2013 32100 10700 1570 521 329 117620 6.7 18 January 2016 1
A 447 16 March 2013 25000 5910 970 399 504 84083 7.3 18 January 2016 1

Table 2 (continued)—Geochemical-analysis and physical-scale-inspection data. Na ¼ sodium; Ca ¼ calcium; Mg ¼ magnesium; SO4 ¼ sulfate;
HCO3 ¼ bicarbonate; TDS ¼ total dissolved solids. Inspection result of 1 means scale has formed and a result of 0 means no scale has formed.

8 2020 SPE Production & Operations

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 9 Total Pages: 23

Na Ca Mg SO4 HCO3 TDS Inspection

Field Well Number Sample Date (ppm) (ppm) (ppm) (ppm) (ppm) (ppm) pH Inspection Date Result
A 447 5 December 2009 42800 12100 1680 362 498 149240 7 18 January 2016 1
A 447 5 November 2008 48400 15400 2000 679 378 173857 6.8 18 January 2016 1
A 447 25 August 2007 40700 12800 1800 564 333 146197 6.2 18 January 2016 1
A 447 20 September 2006 34800 11500 1610 420 289 126819 6.5 18 January 2016 1
A 447 6 February 2006 28500 9570 1380 300 407 104657 6.5 18 January 2016 1
A 447 11 July 2005 35100 12300 1660 476 259 130095 6.6 18 January 2016 1
A 450 5 December 2012 49000 15700 2040 379 155 176274 6.3 25 June 2013 1
A 450 20 December 2011 15100 4110 333 1110 361 51514 7.7 25 June 2013 1
A 450 20 February 2011 55200 16800 2370 811 214 196395 6.7 25 June 2013 1
A 450 10 March 2010 27300 7900 1020 468 473 95661 7.3 25 June 2013 1
A 450 24 February 2006 51400 16000 2020 569 242 183231 6.1 25 June 2013 1
A 450 8 January 2006 44000 16200 2030 524 89 164843 6.1 25 June 2013 1
A 457 10 November 2013 20500 2710 970 963 284 63827 7.5 25 August 2016 1
A 457 23 June 2012 40500 14200 1960 1140 168 150368 6.3 25 August 2016 1
A 457 26 May 2012 37000 12700 2020 336 425 137481 6.7 25 August 2016 1
A 457 18 January 2011 41900 11100 1530 450 636 143716 6.9 25 August 2016 1
A 457 5 December 2009 42500 12400 1740 378 516 149534 6.9 25 August 2016 1
A 457 8 November 2008 42300 12700 1850 316 447 150213 6.6 25 August 2016 1
A 457 25 August 2007 39600 12500 1740 618 312 142470 6.8 25 August 2016 1
A 457 11 July 2005 22800 8040 1080 349 300 84669 7.5 25 August 2016 1
A 458 28 April 2013 33200 10400 1540 342 206 119388 6.7 28 January 2014 1
A 458 7 March 2012 36700 10300 1340 219 350 127309 7 28 January 2014 1
A 458 25 May 2011 27500 7930 1050 494 615 96389 6.9 28 January 2014 1
A 458 5 December 2009 41400 13400 1990 365 472 150527 6.9 28 January 2014 1
A 458 8 November 2008 37200 12800 1680 435 414 136829 6.7 28 January 2014 1
A 458 13 August 2007 46800 13400 1760 1040 336 163336 6.6 28 January 2014 1
A 458 13 January 2007 38600 13000 1640 803 212 140855 6.7 28 January 2014 1
A 458 6 December 2006 40300 14500 1840 190 346 150076 5.9 28 January 2014 1
A 458 31 May 2006 32600 11800 1520 412 232 121664 7 28 January 2014 1
A 468 26 February 2013 16900 4840 663 708 548 59459 7.6 23 November 2015 1
A 468 20 December 2011 18200 5150 642 530 354 63276 7.4 23 November 2015 1
A 468 27 February 2011 13500 4570 568 642 379 49459 7.4 23 November 2015 1
A 468 31 March 2010 21100 6060 778 350 356 73744 7 23 November 2015 1
A 468 6 November 2007 22600 8650 1100 626 434 86110 6.6 23 November 2015 1
A 468 28 May 2006 26500 8880 1080 531 214 96405 6.8 23 November 2015 1
A 468 24 April 2006 33100 8800 1100 756 367 113223 7 23 November 2015 1
A 468 30 November 2005 26700 9130 1050 450 406 97536 6.9 23 November 2015 1
A 473 25 May 2013 38400 4210 1310 623 156 114599 6.6 14 May 2013 1
A 473 10 February 2013 20500 7840 1050 149 221 78060 7.5 14 May 2013 1
A 473 20 February 2012 21800 7130 1010 688 239 79367 6.7 14 May 2013 1
A 473 13 November 2011 23200 6600 830 215 348 80693 7.4 14 May 2013 1
A 473 29 January 2011 39800 13600 1720 601 703 146024 7.2 14 May 2013 1
A 473 21 June 2010 22100 8430 1080 666 195 84071 6.8 14 May 2013 1
A 473 4 July 2009 22300 8770 1150 570 221 85711 7 14 May 2013 1
A 473 15 September 2008 22700 7640 1090 696 222 83348 6.9 14 May 2013 1
A 473 18 March 2007 21500 8920 1100 627 253 83900 7 14 May 2013 1
A 473 27 May 2006 16300 6830 836 486 73 63725 6.5 14 May 2013 1
A 473 14 March 2006 24200 9310 1130 817 245 92102 7.1 14 May 2013 1
A 478 19 April 2006 39900 14800 1970 489 302 150361 7 22 December 2016 1

Table 2 (continued)—Geochemical-analysis and physical-scale-inspection data. Na ¼ sodium; Ca ¼ calcium; Mg ¼ magnesium; SO4 ¼ sulfate;
HCO3 ¼ bicarbonate; TDS ¼ total dissolved solids. Inspection result of 1 means scale has formed and a result of 0 means no scale has formed.

2020 SPE Production & Operations 9

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 10 Total Pages: 23

Na Ca Mg SO4 HCO3 TDS Inspection

Field Well Number Sample Date (ppm) (ppm) (ppm) (ppm) (ppm) (ppm) pH Inspection Date Result
A 478 18 April 2007 47100 15400 1870 406 192 169968 6.4 22 December 2016 1
A 478 24 February 2008 42100 15300 1910 297 231 157138 6.6 22 December 2016 1
A 478 25 May 2011 29900 9000 1170 462 251 105783 6.9 22 December 2016 1
A 478 17 April 2012 25900 6890 1160 1040 213 89803 6.7 22 December 2016 1
A 478 14 April 2013 57800 7630 1050 583 380 172443 7.4 22 December 2016 1
A 478 22 May 2013 59700 8660 2410 400 356 185526 6.7 22 December 2016 0
A 496 10 November 2007 37700 13800 1860 437 365 141662 7.3 3 June 2013 1
A 496 15 February 2006 34700 12200 1690 430 215 128835 6.3 3 June 2013 1
A 496 27 September 2005 37400 10700 1480 426 292 130798 6.8 3 June 2013 1
A 498 8 September 2013 15000 6880 835 305 293 60713 7 26 August 2013 1
A 498 7 September 2011 22700 5050 741 615 394 74900 7.5 26 August 2013 1
A 498 25 May 2009 43100 16600 2020 870 270 163860 6.8 26 August 2013 1
A 498 20 September 2006 45500 16100 2000 478 232 168310 6.4 26 August 2013 1
A 498 30 May 2005 44700 15300 1850 384 247 163481 6.3 26 August 2013 1
A 500 10 March 2013 12200 2850 479 1240 384 41253 7.6 26 May 2013 1
A 500 5 June 2012 32400 10600 1460 1540 387 117987 6.8 26 May 2013 1
A 500 21 December 2011 23700 6580 1020 564 383 82747 7.1 26 May 2013 1
A 500 30 January 2011 30000 11600 1430 566 377 114373 7 26 May 2013 1
A 500 25 May 2009 49400 27900 3130 640 293 215363 6.5 26 May 2013 1
A 522 6 January 2013 33000 8920 1150 556 139 113365 7.5 2 September 2013 1
A 522 11 January 2012 11500 9480 1140 751 345 60316 7.5 2 September 2013 1
A 522 1 December 2010 25900 11300 1430 622 305 103057 6.5 2 September 2013 1
A 522 17 March 2009 28500 10000 1220 560 293 105273 7.3 2 September 2013 1
A 523 2 February 2013 30400 9090 1190 818 175 107473 6.6 16 July 2013 1
A 523 15 February 2012 27300 8150 1090 280 368 96488 7.2 16 July 2013 1
A 523 20 February 2011 13200 3700 469 862 580 46111 7.9 16 July 2013 1
A 523 10 March 2010 26900 7880 1010 328 492 94510 7 16 July 2013 1
A 523 17 March 2009 20200 6240 870 580 409 72299 7.5 16 July 2013 1
A 526 17 March 2012 15200 4540 710 256 558 54264 7.3 23 June 2013 1
A 526 17 January 2012 41700 11200 1310 666 442 142518 7.1 23 June 2013 1
A 526 14 October 2006 484 103 42 223 340 1881 8.4 23 June 2013 1
A 550 31 May 2013 30700 11000 1380 103 354 114137 6.9 9 July 2013 1
A 550 12 March 2013 36700 11600 1610 291 415 131916 7.1 9 July 2013 1
A 550 25 December 2011 50800 13200 1720 695 380 172795 7.1 9 July 2013 1
A 550 29 December 2009 39500 11600 1530 415 235 138780 6.8 9 July 2013 1
A 550 19 May 2006 51 16 0.77 10 120 228 8.4 9 July 2013 1
A 550 5 December 2005 76600 1220 220 1110 345 199495 7.5 9 July 2013 1
A 556 2 March 2010 22800 6930 1090 442 577 81739 7.2 1 April 2014 1
A 556 19 May 2012 41100 14800 1900 360 187 153147 6.6 1 April 2014 1
A 556 15 October 2012 36000 13200 1640 747 192 134779 6.3 1 April 2014 1
A 560 7 December 2009 25800 6090 958 567 333 86548 7.2 28 January 2014 1
B 35 4 January 2005 15300 6530 896 478 198 60702 6.8 26 May 2014 1
B 35 19 April 2005 12700 4490 637 455 124 47406 6.6 26 May 2014 1
B 35 31 July 2006 14500 5960 827 537 179 56903 6.7 26 May 2014 1
B 35 1 June 2009 18100 7820 1060 320 176 71976 6.6 26 May 2014 1
B 35 10 July 2010 17400 6360 857 615 366 65598 7.5 26 May 2014 1
B 35 13 March 2011 14200 6160 816 297 403 56676 7.5 26 May 2014 1
B 35 8 December 2012 16600 5170 744 567 411 59692 6.9 26 May 2014 1
B 35 24 November 2013 16700 10200 1150 465 311 75526 6.3 26 May 2014 1

Table 2 (continued)—Geochemical-analysis and physical-scale-inspection data. Na ¼ sodium; Ca ¼ calcium; Mg ¼ magnesium; SO4 ¼ sulfate;
HCO3 ¼ bicarbonate; TDS ¼ total dissolved solids. Inspection result of 1 means scale has formed and a result of 0 means no scale has formed.

10 2020 SPE Production & Operations

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 11 Total Pages: 23

Na Ca Mg SO4 HCO3 TDS Inspection

Field Well Number Sample Date (ppm) (ppm) (ppm) (ppm) (ppm) (ppm) pH Inspection Date Result
B 54 22 January 2008 19100 10300 1410 607 481 82898 7.3 9 July 2013 1
B 54 5 December 2005 19100 10800 1410 604 292 84206 6.9 9 July 2013 1
B 77 27 June 2011 20300 4790 895 958 238 68781 7 18 August 2014 1
B 90 26 June 2013 17700 4280 827 495 284 60286 7.1 26 May 2014 1
B 90 23 June 2012 15700 3800 693 1330 354 53677 7.5 26 May 2014 1
B 90 27 June 2011 14900 3980 659 853 433 51925 7.1 26 May 2014 1
B 90 18 September 2007 15100 4640 673 830 350 54193 6.9 26 May 2014 1
B 90 18 September 2007 15100 4640 673 830 350 54193 6.9 26 May 2014 1
B 90 27 June 2011 14900 3980 659 853 433 51925 7.1 26 May 2014 1
B 90 23 June 2012 15700 3800 693 1330 354 53677 7.5 26 May 2014 1
B 90 26 June 2013 17700 4280 827 495 284 60286 7.1 26 May 2014 1
B 119 23 May 2011 24800 9940 1760 680 218 97798 6.9 28 May 2011 1
B 119 10 April 2011 11400 7550 966 293 262 53871 6.7 28 May 2011 1
B 119 21 February 2010 18700 9390 1220 317 270 78497 7 28 May 2011 1
B 119 14 February 2009 19100 9910 1330 640 238 81418 6.6 28 May 2011 1
B 119 16 July 2007 19800 11600 1480 452 247 88479 6.4 28 May 2011 1
B 119 24 December 2005 18400 11900 1460 388 262 85610 6.4 28 May 2011 1
B 119 19 November 2005 17400 11400 1470 504 246 81820 6.4 28 May 2011 1
B 159 23 June 2012 8871 2110 464 1290 547 30782 7.7 23 January 2014 1
B 159 15 April 2012 10600 4260 621 605 634 41620 7.7 23 January 2014 1
B 159 12 April 2011 9623 2410 442 1110 478 33363 7.7 23 January 2014 1
B 159 22 December 2009 15200 2410 617 1300 559 48286 7.8 23 January 2014 1
B 159 26 February 2007 10500 3570 557 527 348 39002 7.3 23 January 2014 1
B 159 28 November 2006 11000 3860 575 403 554 41192 7.5 23 January 2014 1
B 169 3 July 2005 27200 14300 1900 421 336 116457 6.8 21 January 2014 1
B 169 26 June 2009 27800 15400 2000 460 417 121477 7.1 21 January 2014 1
B 169 9 June 2010 25300 15200 1810 553 189 113752 6.7 21 January 2014 1
B 169 22 August 2011 24300 13800 1840 266 214 107420 7.2 21 January 2014 1
B 169 29 August 2012 18100 15600 1860 788 70 96718 6.4 21 January 2014 1
B 169 16 September 2013 28800 13200 1900 446 320 117466 6.5 21 January 2014 1
B 174 24 February 2013 11900 5750 795 1210 286 49741 6.9 20 May 2013 1
B 174 15 December 2012 12300 5560 781 859 161 49961 7.1 20 May 2013 1
B 174 13 December 2011 10700 5140 695 347 465 44447 7.2 20 May 2013 1
B 174 4 April 2011 12400 5480 790 782 199 50051 6.9 20 May 2013 1
B 174 19 April 2010 12100 6300 827 722 165 51714 6.5 20 May 2013 1
B 180 23 October 2013 18800 7270 1120 1170 563 72923 7.3 20 May 2013 1
B 180 20 February 2013 16000 7660 1190 376 260 66686 6.7 20 May 2013 1
B 180 27 October 2012 16100 7360 1140 631 395 66126 7.1 20 May 2013 1
B 180 8 October 2011 18400 6400 1010 320 352 68682 7.4 20 May 2013 1
B 180 1 August 2010 15900 7600 1130 649 226 66205 6.9 20 May 2013 1
B 207 22 July 2006 22800 7770 1050 680 205 83905 7.5 11 March 2014 1
B 276 21 December 2005 41000 15900 1900 367 305 155972 6.7 11 October 2014 1
B 276 30 September 2006 11800 4160 560 575 548 44043 7.4 11 October 2014 1
B 276 8 September 2007 37900 13900 1740 672 355 141967 6.1 11 October 2014 1
B 276 13 May 2009 34400 13800 1760 380 287 132727 6.6 11 October 2014 1
B 276 24 May 2010 33200 11000 1290 250 244 120084 6.6 11 October 2014 1
B 276 29 June 2011 36600 12300 1750 670 403 134323 6.6 11 October 2014 1
B 276 7 August 2011 45100 21600 3000 530 134 186364 6.3 11 October 2014 1
B 276 26 May 2012 34700 12300 1560 385 304 128649 6.4 11 October 2014 1

Table 2 (continued)—Geochemical-analysis and physical-scale-inspection data. Na ¼ sodium; Ca ¼ calcium; Mg ¼ magnesium; SO4 ¼ sulfate;
HCO3 ¼ bicarbonate; TDS ¼ total dissolved solids. Inspection result of 1 means scale has formed and a result of 0 means no scale has formed.

2020 SPE Production & Operations 11

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 12 Total Pages: 23

Na Ca Mg SO4 HCO3 TDS Inspection

Field Well Number Sample Date (ppm) (ppm) (ppm) (ppm) (ppm) (ppm) pH Inspection Date Result
B 276 24 June 2012 33000 13600 1680 641 259 128380 6.7 11 October 2014 1
B 276 27 January 2013 31200 11200 1420 569 555 116344 7.1 11 October 2014 1
B 276 18 February 2013 33700 12300 1490 457 537 125984 6.8 11 October 2014 1
B 278 17 February 2013 41200 16600 2080 198 294 159072 6.4 30 April 2014 1
B 278 27 January 2013 12900 3840 631 834 528 46333 7.2 30 April 2014 1
B 278 24 June 2012 14700 4360 676 1370 365 52571 7.3 30 April 2014 1
B 278 26 May 2012 13000 3560 568 504 454 45386 7 30 April 2014 1
B 278 13 November 2011 12900 4230 615 421 413 47279 7.4 30 April 2014 1
B 278 7 August 2011 13300 4220 656 640 608 48424 7.7 30 April 2014 1
B 278 27 October 2010 12700 3850 579 751 189 45569 7 30 April 2014 1
B 278 29 September 2010 11700 4660 622 754 464 45500 6.5 30 April 2014 1
B 278 23 November 2008 12600 4410 597 719 506 46932 7.7 30 April 2014 1
B 278 5 December 2007 16000 4170 571 604 474 54819 7 30 April 2014 1
B 278 3 December 2006 14900 5140 692 290 537 55059 7.5 30 April 2014 1
B 278 5 June 2006 15300 5460 746 589 183 57178 6.9 30 April 2014 1
B 278 3 October 2005 18200 5280 766 571 490 64307 7.1 30 April 2014 1
B 278 10 September 2005 18200 5570 740 461 401 64872 6.8 30 April 2014 1
B 279 15 May 2005 8690 2570 390 379 188 30917 7.2 11 November 2014 1
B 279 5 August 2007 7560 2660 385 720 276 28401 6.6 11 November 2014 1
B 279 23 November 2008 7227 2830 390 901 563 28211 7.5 11 November 2014 1
B 279 4 December 2012 7634 2680 464 616 543 29037 7.1 11 November 2014 1
B 291 10 February 2009 12600 2030 3160 4330 519 51339 6.7 10 November 2014 1
B 291 9 June 2010 8580 3470 519 925 183 33777 7 10 November 2014 1
B 291 20 October 2010 9272 3580 511 886 299 35848 6.7 10 November 2014 1
B 292 29 November 2005 21300 6670 844 585 534 76333 6.2 16 November 2014 1
B 292 28 November 2006 19600 7000 939 551 546 73236 6.6 16 November 2014 1
B 292 6 November 2007 19100 6360 959 675 611 70405 6.2 16 November 2014 1
B 292 22 November 2008 20800 7280 1050 615 565 77510 6.6 16 November 2014 1
B 292 27 June 2009 5045 1370 339 840 644 18438 7.3 16 November 2014 1
B 292 2 June 2010 7456 1500 524 1670 604 25854 7.6 16 November 2014 1
B 299 15 November 2011 52900 19100 2360 142 263 196765 6.5 21 January 2014 1
B 299 4 December 2012 50100 18700 2360 129 249 188538 7 21 January 2014 1
B 310 25 April 2005 27700 10300 1250 645 97 103992 6.3 17 August 2014 1
B 310 1 May 2005 28600 10700 1300 527 394 107721 6.8 17 August 2014 1
B 310 21 November 2005 22700 7830 1080 383 537 84030 6.8 17 August 2014 1
B 310 2 September 2007 20000 7460 966 585 508 75719 7.3 17 August 2014 1
B 310 11 September 2007 21800 7780 985 567 524 81156 7 17 August 2014 1
B 310 11 July 2011 18300 2770 1160 673 428 59031 7.7 17 August 2014 1
B 311 11 October 2013 19200 2170 1180 2040 282 60072 8.1 11 March 2014 1
B 311 11 September 2012 18800 3410 1420 740 257 63127 6.8 11 March 2014 1
B 311 7 September 2011 19600 2580 1020 1220 416 61536 7.6 11 March 2014 1
B 311 6 April 2009 42900 22600 2790 310 165 182765 6.7 11 March 2014 1
B 311 17 April 2010 17700 3250 1120 940 540 58850 7.6 11 March 2014 1
B 311 14 November 2006 37800 24100 2880 400 86 174266 5.9 11 March 2014 1
B 311 14 January 2008 17400 3910 840 810 535 58795 7.5 11 March 2014 1
B 311 16 May 2006 18900 4320 1000 818 250 64288 7.2 11 March 2014 1
B 311 11 March 2006 19100 5530 1170 995 428 68823 7.1 11 March 2014 1
B 311 15 January 2006 21000 5790 1120 478 556 74144 7.6 11 March 2014 1
B 330 2 May 2005 7887 3160 457 716 450 30970 7.3 17 August 2014 1

Table 2 (continued)—Geochemical-analysis and physical-scale-inspection data. Na ¼ sodium; Ca ¼ calcium; Mg ¼ magnesium; SO4 ¼ sulfate;
HCO3 ¼ bicarbonate; TDS ¼ total dissolved solids. Inspection result of 1 means scale has formed and a result of 0 means no scale has formed.

12 2020 SPE Production & Operations

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 13 Total Pages: 23

Na Ca Mg SO4 HCO3 TDS Inspection

Field Well Number Sample Date (ppm) (ppm) (ppm) (ppm) (ppm) (ppm) pH Inspection Date Result
B 330 14 December 2005 8533 3150 421 689 439 32432 7.1 17 August 2014 1
B 330 14 January 2006 8844 3250 434 408 442 33478 7.9 17 August 2014 1
B 330 17 February 2007 10900 3830 521 695 398 40744 7.5 17 August 2014 1
B 330 2 April 2008 17300 6010 783 612 199 63904 6.8 17 August 2014 1
B 330 2 June 2009 14800 5180 696 330 192 54798 6.6 17 August 2014 1
B 330 25 July 2009 50500 31400 3500 770 160 229330 6.3 17 August 2014 1
B 333 20 January 2007 20000 6760 860 315 375 73110 7 17 August 2014 1
B 333 2 May 2005 19300 6980 909 586 398 72273 7.2 17 August 2014 1
B 343 18 March 2008 25700 7600 976 568 353 90497 7.2 5 May 2014 1
B 343 28 March 2011 22200 6890 852 279 189 79010 6.5 5 May 2014 1
B 343 9 October 2012 32700 11000 1290 321 292 118903 6.9 5 May 2014 1
B 343 6 October 2013 32300 10400 1330 1640 280 116650 6.6 5 May 2014 1
B 361 9 October 2005 28600 8020 1030 475 506 99331 6.8 7 May 2014 1
B 361 18 October 2005 27700 8110 988 438 495 97131 7 7 May 2014 1
B 361 11 September 2007 22700 6190 965 642 304 79001 6.5 7 May 2014 1
B 361 9 December 2007 20800 7190 1180 803 610 77883 7 7 May 2014 1
B 361 26 November 2008 22500 6170 1030 611 523 78734 7.3 7 May 2014 1
B 361 2 November 2009 22500 5540 1010 855 458 76863 7.5 7 May 2014 1
B 361 30 November 2010 10500 3140 525 1220 298 37883 6.9 7 May 2014 1
B 361 6 March 2011 11300 3340 583 543 305 40571 6.8 7 May 2014 1
B 361 1 August 2011 21700 5940 1060 1310 570 76280 6.9 7 May 2014 1
B 361 10 March 2012 22300 4860 1030 783 505 74678 6.7 7 May 2014 1
B 361 3 September 2013 21600 4740 1000 495 534 72369 6.9 7 May 2014 1
B 369 2 February 2005 30900 10100 1360 592 459 112211 7 20 April 2014 1
B 369 26 May 2013 18900 6590 863 290 353 69896 7.5 20 April 2014 1
B 381 2 August 2006 35300 12500 1590 399 138 130727 6.6 26 May 2011 1
B 381 2 October 2007 34000 11800 1480 521 412 125213 6.8 26 May 2011 0
B 381 22 November 2009 24700 6970 938 608 598 86214 6.9 26 May 2011 0
B 381 7 April 2011 30200 9560 1310 832 427 108829 7 26 May 2011 0
B 381 4 July 2011 24500 7990 962 751 437 88540 6.5 26 May 2011 0
B 381 16 July 2012 25500 9800 1240 437 325 97102 6.8 26 May 2011 0
B 389 5 January 2005 44700 16100 1820 699 371 165690 6.7 23 November 2014 0
B 389 1 August 2006 30900 11300 1340 381 344 115365 6.9 23 November 2014 0
B 389 20 September 2006 30100 10800 1240 361 198 111499 6.2 23 November 2014 0
B 389 26 September 2006 29700 10700 1240 462 146 110248 6.7 23 November 2014 0
B 389 30 October 2006 29300 10400 1190 278 399 108267 6.9 23 November 2014 0
B 389 27 April 2010 50200 16000 1610 522 107 178439 6.6 23 November 2014 0
B 389 19 April 2011 13700 5180 610 731 537 52058 7 23 November 2014 0
B 389 14 December 2011 29400 8330 977 225 341 101873 6.9 23 November 2014 0
B 389 1 April 2012 17200 5710 758 628 256 62852 6.4 23 November 2014 0
B 389 26 September 2012 35700 14100 1640 670 245 136455 6.4 23 November 2014 0
B 389 28 October 2013 13700 4570 665 1020 578 50533 7.5 23 November 2014 0
B 403 22 June 2013 22800 8900 1220 212 271 87503 6.8 23 January 2014 0
B 403 18 June 2012 24300 10700 1360 676 156 96992 7 23 January 2014 0
B 403 3 January 2011 24400 10700 1250 1690 534 97274 7.2 23 January 2014 0
B 403 25 June 2011 23200 10600 1380 429 200 93909 6.8 23 January 2014 0
B 403 14 May 2011 19600 9440 1290 803 243 81376 7 23 January 2014 0
B 439 11 October 2006 32500 12300 1540 467 317 122924 7 18 August 2014 0
B 439 31 October 2006 33000 13400 1600 443 297 127540 6.9 18 August 2014 0

Table 2 (continued)—Geochemical-analysis and physical-scale-inspection data. Na ¼ sodium; Ca ¼ calcium; Mg ¼ magnesium; SO4 ¼ sulfate;
HCO3 ¼ bicarbonate; TDS ¼ total dissolved solids. Inspection result of 1 means scale has formed and a result of 0 means no scale has formed.

2020 SPE Production & Operations 13

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 14 Total Pages: 23

Na Ca Mg SO4 HCO3 TDS Inspection

Field Well Number Sample Date (ppm) (ppm) (ppm) (ppm) (ppm) (ppm) pH Inspection Date Result
B 439 1 December 2006 27600 10600 1300 252 350 104902 7 18 August 2014 0
B 439 14 January 2007 28200 11000 1340 584 345 107769 7.8 18 August 2014 0
B 452 12 June 2013 12700 1550 585 356 320 39111 8.1 26 September 2013 0
B 452 23 June 2012 10700 2770 451 1300 536 37257 7.9 26 September 2013 0
B 452 28 June 2011 11000 3250 489 598 482 39219 7.9 26 September 2013 0
B 452 11 September 2007 12800 4750 648 602 392 48592 6.9 26 September 2013 0
B 452 18 October 2005 19000 5560 775 543 160 66938 6.5 26 September 2013 0
B 452 30 April 2005 19100 7330 1000 538 443 73111 6.8 26 September 2013 0
B 452 4 January 2005 18800 8740 1040 620 514 76414 7.3 26 September 2013 0
B 455 10 October 2011 18900 6090 762 145 359 68156 7.4 22 January 2014 0
B 455 14 December 2011 16700 4910 699 390 456 59055 7.3 22 January 2014 0
B 455 2 June 2012 35200 211 8.7 611 515 90546 6.9 22 January 2014 0
B 455 23 June 2012 24500 8660 987 1320 432 90699 7 22 January 2014 0
B 455 18 December 2013 22300 8140 932 321 456 83149 7.2 22 January 2014 0
B 457 19 September 2013 20900 4400 800 608 290 68798 7.7 26 May 2013 0
B 457 29 January 2013 19700 6120 786 481 244 70331 7.2 26 May 2013 0
B 457 16 July 2012 19700 6900 980 440 143 73163 7 26 May 2013 0
B 457 14 December 2011 24200 6290 835 305 417 82447 7.1 26 May 2013 0
B 457 27 June 2011 24200 7890 1030 669 235 87624 6.8 26 May 2013 0
B 457 11 June 2011 22000 8570 1130 824 415 84539 6.8 26 May 2013 0
B 457 29 June 2010 24600 7950 936 596 515 88597 6.8 26 May 2013 0
B 480 21 July 2013 23900 7120 1040 546 601 85007 7.6 17 August 2014 0
B 480 16 July 2012 22100 7150 966 360 393 79969 6.9 17 August 2014 0
B 480 11 July 2011 23700 6900 975 320 183 83278 7.2 17 August 2014 0
B 480 22 December 2009 13200 3480 451 379 475 45285 7.8 17 August 2014 0
B 480 12 November 2008 25400 8160 1170 423 280 91933 7.2 17 August 2014 0
B 480 2 September 2007 17200 5860 816 704 232 63512 6.7 17 August 2014 0
B 480 12 December 2006 30600 9710 1270 692 395 110067 7.1 17 August 2014 0
B 480 13 November 2006 28900 10200 1350 410 245 107205 6.6 17 August 2014 0
B 480 26 September 2006 30900 10400 1350 674 249 112973 6.8 17 August 2014 0
B 480 7 June 2006 31100 10500 1410 781 92 113883 6.8 17 August 2014 0
B 480 10 May 2006 17800 11700 1560 483 61 83904 6.8 17 August 2014 0
B 480 20 March 2006 28300 11100 1450 795 346 108691 6.6 17 August 2014 0
B 480 20 March 2006 28300 11100 1450 795 346 108691 6.6 17 August 2014 0
B 480 10 May 2006 17800 11700 1560 483 61 83904 6.8 17 August 2014 0
B 480 7 June 2006 31100 10500 1410 781 92 113883 6.8 17 August 2014 0
B 480 26 September 2006 30900 10400 1350 674 249 112973 6.8 17 August 2014 0
B 480 13 November 2006 28900 10200 1350 410 245 107205 6.6 17 August 2014 0
B 480 12 December 2006 30600 9710 1270 692 395 110067 7.1 17 August 2014 0
B 480 2 September 2007 17200 5860 816 704 232 63512 6.7 17 August 2014 0
B 480 12 November 2008 25400 8160 1170 423 280 91933 7.2 17 August 2014 0
B 480 22 December 2009 13200 3480 451 379 475 45285 7.8 17 August 2014 0
B 480 11 July 2011 23700 6900 975 320 183 83278 7.2 17 August 2014 0
B 480 16 July 2012 22100 7150 966 360 393 79969 6.9 17 August 2014 0
B 480 21 July 2013 23900 7120 1040 546 601 85007 7.6 17 August 2014 0
B 489 23 October 2013 23300 9630 1280 805 300 91315 7.1 6 August 2014 0
B 489 24 February 2013 11800 5670 806 1080 590 49446 7.2 6 August 2014 0
B 489 12 October 2012 10900 5750 768 586 358 46962 7 6 August 2014 0
B 489 17 April 2012 11000 6280 854 732 451 49117 7.2 6 August 2014 0

Table 2 (continued)—Geochemical-analysis and physical-scale-inspection data. Na ¼ sodium; Ca ¼ calcium; Mg ¼ magnesium; SO4 ¼ sulfate;
HCO3 ¼ bicarbonate; TDS ¼ total dissolved solids. Inspection result of 1 means scale has formed and a result of 0 means no scale has formed.

14 2020 SPE Production & Operations

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 15 Total Pages: 23

Na Ca Mg SO4 HCO3 TDS Inspection

Field Well Number Sample Date (ppm) (ppm) (ppm) (ppm) (ppm) (ppm) pH Inspection Date Result
B 489 24 August 2011 8440 4200 591 812 128 35671 7.3 6 August 2014 0
B 489 9 March 2011 10100 4880 710 607 409 42306 7.6 6 August 2014 0
B 489 1 August 2010 12200 7260 1010 611 275 55356 7.5 6 August 2014 0
B 489 2 June 2010 17400 11800 1590 706 122 83318 6.8 6 August 2014 0
B 489 22 December 2009 19000 4730 710 450 318 64408 7.3 6 August 2014 0
B 490 3 August 2010 12900 5420 749 604 427 51100 7.6 21 May 2013 0
B 508 18 November 2009 19900 9520 1240 433 275 81968 7 12 August 2014 0
B 508 15 February 2009 17400 9100 1230 650 390 74570 7 12 August 2014 0
B 517 24 September 2013 55100 27000 3370 350 529 228349 6.1 21 May 2013 0
B 517 18 September 2012 43000 34700 3670 484 143 219997 6.1 21 May 2013 0
B 517 6 March 2011 27300 13700 1290 132 85 112407 6.2 21 May 2013 0
B 517 3 August 2010 46800 34400 3810 229 110 229349 6.1 21 May 2013 0
B 518 29 August 2013 30000 9040 1490 428 419 107477 6.8 25 May 2013 0
B 518 6 August 2012 24800 13700 1610 506 165 107481 6.5 25 May 2013 0
B 518 16 October 2011 22300 10200 1190 159 210 89659 6.9 25 May 2013 0
B 518 13 March 2011 29500 13500 1610 213 329 118952 6.6 25 May 2013 0
B 518 15 November 2009 43600 18400 2270 451 298 171019 6.7 25 May 2013 0
B 519 3 August 2010 44400 31400 3480 245 146 213671 6.3 11 March 2014 0
B 519 19 January 2012 47300 25200 2750 667 249 201166 6.5 11 March 2014 0
B 519 19 June 2012 24000 13800 1420 1280 193 105193 6.8 11 March 2014 0
B 519 19 June 2012 24000 13800 1420 1280 193 105193 6.8 11 March 2014 0
B 519 25 June 2013 45600 28100 3150 83 197 206130 6.3 11 March 2014 0
B 523 18 June 2012 35900 18600 2240 527 167 151834 6.4 21 May 2013 0
B 523 2 October 2013 35700 15500 1960 1550 368 141878 6.6 21 May 2013 0
B 523 9 October 2012 37600 17100 2110 825 224 151559 6.7 21 May 2013 0
B 523 29 June 2011 39700 18800 2470 736 384 163090 6.3 21 May 2013 0
B 523 20 June 2010 40900 18800 2090 443 281 164514 6.4 21 May 2013 0
B 523 24 May 2010 8120 3870 465 182 334 33371 7.2 21 May 2013 0
B 524 9 March 2013 34100 16700 1940 154 232 140626 6.9 3 November 2014 0
B 524 27 March 2012 35800 15300 1920 657 461 141338 7.3 3 November 2014 0
B 524 17 March 2009 38000 23800 2940 236 159 174135 6.2 3 November 2014 0
B 524 3 August 2010 25400 10700 1380 464 250 99794 7.3 3 November 2014 0
B 524 28 October 2006 50400 30900 3370 270 132 227072 6.4 3 November 2014 0
B 526 27 January 2013 19000 8920 1140 644 354 77758 6.9 11 March 2014 0
B 526 28 October 2012 20200 9140 1180 1170 458 81748 7.1 11 March 2014 0
B 526 27 May 2012 21200 8950 1110 543 505 83408 6.7 11 March 2014 0
B 526 20 October 2010 15200 9690 1220 1150 67 70527 5.8 11 March 2014 0
B 526 22 October 2011 16300 6990 893 407 121 64511 6.9 11 March 2014 0
B 526 7 November 2009 13900 7200 982 546 500 59528 6.7 11 March 2014 0
B 526 11 November 2008 14300 7880 1120 606 369 62875 6.8 11 March 2014 0
B 531 18 February 2013 19000 12300 1510 593 539 88642 6.9 4 August 2014 0
B 531 12 January 2013 17600 7680 964 656 588 70188 7.1 4 August 2014 0
B 531 12 January 2013 19600 8260 1030 889 650 77229 7 4 August 2014 0
B 531 2 January 2012 19500 10100 1250 674 254 82678 6.6 4 August 2014 0
B 531 20 August 2011 16000 9100 1420 273 222 71615 7 4 August 2014 0
B 531 9 March 2011 18500 10300 1380 747 244 81271 6.6 4 August 2014 0
B 531 16 February 2011 16400 9400 1090 622 201 72313 6.5 4 August 2014 0
B 531 15 January 2007 40800 27100 3230 297 232 191659 6.3 4 August 2014 0
B 531 26 January 2010 19400 9860 1330 328 207 82025 6.8 4 August 2014 0

Table 2 (continued)—Geochemical-analysis and physical-scale-inspection data. Na ¼ sodium; Ca ¼ calcium; Mg ¼ magnesium; SO4 ¼ sulfate;
HCO3 ¼ bicarbonate; TDS ¼ total dissolved solids. Inspection result of 1 means scale has formed and a result of 0 means no scale has formed.

2020 SPE Production & Operations 15

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 16 Total Pages: 23

Na Ca Mg SO4 HCO3 TDS Inspection

Field Well Number Sample Date (ppm) (ppm) (ppm) (ppm) (ppm) (ppm) pH Inspection Date Result
B 531 27 May 2009 18200 10400 1390 350 156 80696 6.5 4 August 2014 0
B 531 10 February 2009 20200 10700 1440 750 201 86891 6.7 4 August 2014 0
B 531 26 January 2008 20100 11000 1450 520 300 87570 7 4 August 2014 0
B 531 13 November 2006 19000 4020 1020 976 388 63804 7.1 4 August 2014 0
B 541 20 April 2013 21900 10700 1370 451 1434 91355 6.8 4 June 2013 0
B 541 23 February 2013 18600 11600 1520 427 451 85598 6.7 4 June 2013 0
B 541 29 August 2012 17700 12900 1640 606 588 87534 6.9 4 June 2013 0
B 541 5 February 2012 18400 10800 1410 669 409 82488 7 4 June 2013 0
B 541 6 August 2011 20700 12800 1830 920 352 95602 6.5 4 June 2013 0
B 541 9 March 2011 23100 9900 1680 475 268 92923 6.7 4 June 2013 0
B 541 16 February 2011 20900 11900 1510 1140 281 92431 6.5 4 June 2013 0
B 541 1 August 2010 20400 12800 1810 525 299 94734 6.9 4 June 2013 0
B 541 21 December 2009 30200 11100 1180 629 275 112384 7 4 June 2013 0
B 541 14 January 2007 25400 16800 2160 684 329 119873 6.5 4 June 2013 0
B 541 6 January 2007 24800 16400 2160 1030 336 117326 6.1 4 June 2013 0
B 543 18 June 2012 64300 22200 2560 1330 26 235416 6 19 August 2014 0
B 543 18 March 2008 592 81 23.5 189 153 1936 7.9 19 August 2014 0
B 543 18 March 2008 592 81 23.5 189 153 1936 7.9 19 August 2014 0
B 543 18 June 2012 64300 22200 2560 1330 26 235416 6 19 August 2014 0
B 555 23 February 2013 17600 10500 1350 76 756 79482 7.3 12 August 2014 0
B 555 20 February 2011 21900 15100 1690 543 55 104288 6.1 12 August 2014 0
B 555 6 February 2012 21000 12300 1450 379 420 93449 7.1 12 August 2014 0
B 555 15 February 2009 33900 17100 2160 637 268 142265 6.5 12 August 2014 0
B 555 22 February 2010 28200 16700 2090 234 289 126313 6.6 12 August 2014 0

Table 2 (continued)—Geochemical-analysis and physical-scale-inspection data. Na ¼ sodium; Ca ¼ calcium; Mg ¼ magnesium; SO4 ¼ sulfate;
HCO3 ¼ bicarbonate; TDS ¼ total dissolved solids. Inspection result of 1 means scale has formed and a result of 0 means no scale has formed.

Field Best-Prediction Results Cp (USD) Base-Case Cost (USD)

Group A 96.0% 6.9 million (saving) 6.5 million (saving)
Group B 92.6% 5.7 million (saving) 4.3 million (saving)

Table 3—Predicted cost analysis using ML. Cost of scale removal per well and cost of scale-
inhibition treatment per well were assumed to be USD 50 and USD 15 million, respectively, for cost-
comparison purposes.

Confusion Matrix
A neural-network model was trained using a pattern-recognition module inside MATLABV R software (The MathWorks, Inc., Natick,

Massachusetts, USA). A total of 486 data points was fed into the AI models. Of the 486 data points, 143 were instances when scale was not
formed and the remaining 343 were the instances when the scale was actually deposited. A confusion matrix is plotted between the actual and
predicted values for the scale-formation problem. The confusion matrix is used to evaluate the performance outcome of the classification prob-
lem. The matrix is applicable when the output has at least two classes. The matrix can be formed with four different combinations of actual and
predicted values. The diagonal values of the matrix represent the values that are correctly classified, while the off-diagonal values of the matrix
indicate values that are incorrectly classified. In fact, the rows represent the actual values, while columns represent the predicted values. The
column on the extreme right and at the bottom of the plot shows the accuracy values. The cell in the bottom right of the plot shows the overall
accuracy. The three main parameters that are derived from the confusion matrix are recall, precision, and F1 score criterion. These parame-
ters define which classification algorithm is superior in terms of performance compared with others. The definitions of these parameters are
True positive
Recall ¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð36Þ
True positive þ false negative
True positive
Precision ¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð37Þ
True positive þ false positive
ðRecall  precisionÞ
F1 score ¼ 2  ; .......................................................... ð38Þ
ðRecall þ precisionÞ
where “true positive” is an outcome where the model correctly predicts the positive value. Similarly, a “true negative” is an outcome where
the model correctly predicts the negative value. A “false positive” is an outcome where the model incorrectly predicts the positive value,
and a “false negative” is an outcome where the model incorrectly predicts the negative value. Fig. 2 shows the confusion matrix for our
problem using five ML techniques, such as KNN, SVM, gradient boosting, gradient-boosting classifier, and decision tree. For all five tech-
niques, recall, precision, and F1 were calculated and are given in Fig. 3. All techniques resulted in very good accuracy along with recall,
precision, and F1 score. KNN was chosen for further investigation because of its accuracy and ability to develop further classification plots.

16 2020 SPE Production & Operations

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 17 Total Pages: 23

Scale Formation Using KNN Confusion Matrix Scale Formation Using SVM Confusion Matrix

132 11 92.3% 117 2 98.3%

1 27.2% 2.3% 7.7% 1 24.1% 0.4% 1.7%

Predicted Values

Predicted Values
11 332 96.8% 26 341 92.9%
2 68.3% 3.2% 2
2.3% 5.3% 70.2% 7.1%

92.3% 96.8% 95.5% 81.8% 99.4% 94.2%

7.7% 3.2% 4.5% 18.2% 0.6% 5.8%

1 2 1 2
Actual Values Actual Values
(a) (b)

Scale Formation Using Gradient-Boosting Scale Formation Using Gradient-Boosting-

Confusion Matrix Classifier Confusion Matrix

134 5 96.4% 130 9 93.5%

1 1 26.7% 1.9% 6.5%
27.6% 1.0% 3.6%
Predicted Values

Predicted Values
9 338 97.4% 13 334 96.3%
2 2 68.7% 3.7%
1.9% 69.5% 2.6% 2.7%

93.7% 98.5% 97.1% 90.9% 97.4% 95.5%

6.3% 1.5% 2.9% 9.1% 2.6% 4.5%

1 2 1 2
Actual Values Actual Values
(c) (d)

Scale Formation Using Decision-Tree

Confusion Matrix

130 9 93.5%
1 26.7% 1.9% 6.5%
Predicted Values

13 334 56.3%
2 68.7% 3.7%
2.7%

90.9% 97.4% 95.5%

9.1% 2.6% 4.5%

1 2
Actual Values
(e)

Fig. 2—Scale-formation confusion matrix using five ML algorithms: (a) KNN, (b) SVM, (c) gradient boosting, (d) gradient-boosting
classifier, and (d) decision tree.

Classification Using KNN Algorithm

The KNN algorithm was implemented inside MATLAB software. The dimension of the KNN network was 10  10 in a hexagonal
grid. The topology plot is shown in Fig. 4. The 100 neurons are arranged in 10  10 2D hexagonal grids. The topology plot is created to
visualize the high-dimensional input space in the 2D network topology.

2020 SPE Production & Operations 17

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 18 Total Pages: 23

1.00

0.80

Performance Indicator
0.60 Recall

0.983
Precision

0.964
0.950
0.937

0.935

0.935
0.923
0.923
0.923

0.922

0.922
0.909

0.909
0.893
F1

0.818
0.40

0.20

0
KNN SVM Gradient Boosting Gradient-Boosting Decision Tree
Classifier

Fig. 3—Performance comparison of five techniques.

Self-Organizing Map Topology

–1
0 2 4 6 8 10

Fig. 4—2D hexagonal-grid network topology.

Fig. 5 shows the input-weight-planes plots. The input-weight-plane plot shows the weights of input features, with the weights of
each 100 neurons in the 10  10 hexagonal-grid structure. The weight planes, which look similar, have a highly dependent correlation,
while dissimilar planes indicate the independent features. Darker color is an indication of larger weights.
Fig. 6 shows the plot of neighbor-weight distances. A blue hexagonal-shaped grid indicates neurons. The red lines connect the
neighboring neurons. The distances between the neurons are represented by the colors of the regions where the red lines are seated. The
light color shows smaller distances between the neurons, while the dark color represents the largest distances. The distances given are
Euclidian distances.
Fig. 7 shows the self-organizing map hits plot for the output layer of size 10  10, in terms of neurons. The sample hit plot shows
how many of the training samples fall in each cluster. The adjacent clusters have learned similar features. The gap with no values
shows the indication of separation between the clusters. It means data are classified into two sets (i.e., either scale is formed or scale is
not formed). There is not any intermediate case. The sample hit plot calculates the classes for the deposition of scale formation and
shows the number of possibilities in each class. The regions with higher number of hit values show the similar highly populated regions
of input-feature spaces.

Development of Scale-Formation Correlation Using the FN Tool

A systematic workflow diagram of the proposed FN model to predict scale formation is given in Fig. 8. The workflow diagram high-
lights the four mandatory stages for the FN model development (i.e., input-parameter selection, parametric learning, model selection,
and model testing). For some steps, where one of the available choices must be made, the opted one is shown in light blue. A total of
486 data points were obtained from the real fields of a Middle Eastern carbonate reservoir. A comprehensive quality check of the col-
lected data set was performed before the data were fed into the AI models. After removal of outliers, the data set was divided into two
sets. A random 70% of the data were used for the model training and the remaining 30% was used for testing of the trained models. The
FN model was trained on 343 data points. During training, an average absolute percentage error (AAPE) of 6.5% was observed between
the actual and predicted values. During testing on the 143 data points, an AAPE of 5.4% was noted. The FN model was trained with
third-degree polynomial function. A backward-elimination method was chosen from the available training modules, such as exhaustive
search, forward-backward, backward-forward, and forward selection. The backward-elimination algorithm was selected because of its

18 2020 SPE Production & Operations

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 19 Total Pages: 23

higher accuracy and less computational time. The minimum-description length is adopted as a fitness criterion for the model-selection
purpose. This is a concept from information theory, which makes the optimal choice using both the network size and accuracy of pre-
dictions. Mathematically, the minimum-description-length measure ðdÞ is given by
plogNp Np
d¼ þ logp2 ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð39Þ
2 2
where Np is the number of functions in the optimal set (number or parameters), and p is the root-mean-square error given by
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u Np
u1 X
p¼t ei 2 : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð40Þ
N i¼1

Weights from Input 1 Weights from Input 2 Weights from Input 3

8 8 8
6 6 6
4 4 4
2 2 2
0 0 0
0 5 10 0 5 10 0 5 10

Weights from Input 4 Weights from Input 5 Weights from Input 6

8 8 8
6 6 6
4 4 4
2 2 2
0 0 0
0 5 10 0 5 10 0 5 10

Weights from Input 7 Weights from Input 8 Weights from Input 9

8 8 8
6 6 6
4 4 4
2 2 2
0 0 0
0 5 10 0 5 10 0 5 10

Fig. 5—Input-weight-plane plots. Input 1 is Na, the molar concentration of sodium cations (in mol/L); Input 2 is Ca, the molar con-
centration of calcium cations (in mol/L); Input 3 is Mg, the molar concentration of magnesium cations (in mol/L); Input 4 is SO4, the
molar concentration of sulfate cations (in mol/L); Input 5 is HCO3, the molar concentration of bicarbonate ions (in mol/L); Input 6
is total dissolved solids; Input 7 is pH; Input 8 is T, which is the normalized inspection time; and Input 9 is scale formed
within 10 years.

Self-Organizing Map Neighbor Weight Distances

–1
0 2 4 6 8 10

Fig. 6—Neighbor-weight-distances plot.

2020 SPE Production & Operations 19

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 20 Total Pages: 23

Hits

8
6 4 2 1 5 6 1 6 8 4

7 1 5 5 0 1 6 3 8 11 10

6 1 6 4 6 4 7 5 5 1 9

5 8 8 1 4 4 2 5 6 10
5
10 6 2 2 9 3 4 9 11 9
4
3 6 5 5 5 6 10 2 4 12
3
7 3 9 8 3 1 6 5 7 3
2
7 3 8 5 6 10 4 4 5 6
1 1 4 2 1 8 5 2 5 7 3

0 1 1 1 10 1 1 6 7 3 9

–1
0 2 4 6 8 10

Fig. 7—SOM hits plot.

Start
Input-Parameter

Data collection
Selection

Selection of input parameters

Selection of output parameters

Selection of parametric
Training learning method
data set Exact Approximate
learning learning
Parametric
Learning

Selection of basis family for neural functions

Polynomial Exponential Trigonometric Hybrid

Creation of initial function network

Initial neural functions of Model parameters with each
selection order neural function

Method for model selection

Exhaustive Forward Backward Forward- Backward-
search selection elimination backward forward
Model Selection

Optimal model selection by backward elimination

Calculation of MDL Compare MDL after Remove suboptimal

each elimination functions

Optimal functional network

Optimal neural functions Model parameters with each

neural function

Testing of the functional Scale

network precipitation
Training
Testing
Model

data set
Error analysis Error indices

Fig. 8—Model-flow diagram of the FN tool. MDL 5 minimum-description length.

20 2020 SPE Production & Operations

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 21 Total Pages: 23

The mathematical model for scale formation is given

$ N ’
X p
r1i r2i r3i r4i r5i r6i r7i r8i
Scale formation ¼ xi Na Ca Mg SO4 HCO3 TDS pH T ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð41Þ
i¼1

where Na is the molar concentration of sodium cations (in mol/L); Ca is the molar concentration of calcium cations (in mol/L); Mg is
the molar concentration of magnesium cations (in mol/L); SO4 is the molar concentration of sulfate cations (in mol/L); HCO3 is the
molar concentration of bicarbonate ions (in mol/L); TDS is the total dissolved solids; pH represents the acidity or alkalinity of a solu-
tion; and T is the normalized inspection time. The output of the model is either 1 or 2: 1 means precipitation and 2 means no precipita-
tion (previously, we have used 0 for no scaling). The coefficients of Eq. 41 are listed in Table 4.

Order
Np σ1 σ2 σ3 σ4 σ5 σ6 σ7 σ8 Coefficient ω
–5
1 0 0 0 1 1 0 0 0 1.32×10
–5
2 0 0 1 0 1 0 0 0 3.54×10
–5
3 0 1 0 0 1 0 0 0 2.76×10
–5
4 1 0 0 0 1 0 0 0 2.51×10
–6
5 0 0 0 0 1 1 0 0 –9.87×10
–4
6 0 0 0 0 1 0 1 0 2.73×10
–5
7 1 0 0 0 0 0 1 0 1.10×10
–6
8 0 0 0 0 0 1 0 1 7.27×10
–5
9 0 1 0 0 0 0 0 1 –6.69×10
–9
10 0 0 0 1 2 0 0 0 7.57×10
–9
11 0 0 2 0 1 0 0 0 2.09×10
–10
12 1 0 1 0 1 0 0 0 7.48×10
–11
13 2 0 0 0 1 0 0 0 5.36×10
–17
14 0 0 0 0 0 3 0 0 –4.86×10
–12
15 0 0 0 0 1 2 0 0 6.50×10
–11
16 0 0 0 1 1 1 0 0 –1.87×10
–10
17 0 0 1 0 1 1 0 0 –2.55×10
–11
18 1 0 0 0 1 1 0 0 –3.73×10
–5
19 0 0 0 0 0 1 2 0 7.31×10
–6
20 0 0 0 1 0 0 2 0 –3.39×10
–6
21 0 0 0 1 1 0 1 0 –2.00×10
–4
22 0 0 1 0 0 0 2 0 –2.69×10
–4
23 0 1 0 0 0 0 2 0 –2.02×10
–4
24 1 0 0 0 0 0 2 0 –1.88×10
–6
25 0 0 0 0 0 1 0 2 8.17×10
–8
26 0 0 1 0 0 1 0 1 –3.92×10
–5
27 0 1 0 0 0 0 0 2 –2.02×10
–7
28 0 1 1 0 0 0 0 1 1.30×10
–5
29 1 0 0 0 0 0 0 2 –2.30×10
–8
30 1 0 1 0 0 0 0 1 9.65×10

Table 4—Coefficients for scale-formation model.

The AAPE is defined as

P 100
j ðtmeasured  tpredicted Þ j 
tmeasured
AAPE ¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð42Þ
k
where tmeasured is the measured value of t and tpredicted is the estimated value from the model, and k is the total number of data points.

Conclusions
In this paper, methods for predicting and inhibiting CaCO3 scale in hydrocarbon wells using ML techniques are presented. Further, a
conceptualized workflow to quantify the overall cost associated with implementing a scale-inhibition/treatment program is presented.
The results of applying such methodology on different data sets showed high prediction accuracy on two data groups. In addition, cost
savings were realized compared with a base-case scenario.

2020 SPE Production & Operations 21

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 22 Total Pages: 23

Future work on this subject can expand ML techniques to include other available methods in the industry to test its accuracy com-
pared with the methods used in this paper. In addition, the applicability of this method on types of scale other than CaCO3 can be stud-
ied and analyzed.

Nomenclature
C ¼ concentration of each ion, g/1000 g
Cbc ¼ base-case cost
Ccsp ¼ expected net saving of correctly predicted scale formation cases per well ¼ Pcsp(CT  Cs)
CCNsp ¼ expected saving of correctly predicted no scale formation cases per well ¼ PCNsp * CT
CINsp ¼ cost of incorrectly predicted no scale formation cases per well ¼ PIsp * Cs
CIsp ¼ cost of incorrectly predicted scale formation cases per well ¼ PIsp * CT
Cp ¼ overall expected net cost of the scale inhibition program ¼ Xt (Ccsp þ CIsp þ CCNsp þ CINsp)
Cs ¼ cost of scale removal per well
CT ¼ cost of scale-inhibition treatment per well
F0 ¼ derived from artificial neural network of the scale deposition as function of five variables (pressure, temperature, brine concen-
tration, velocity, and time)
F1 ¼ depends on an empirical model, F0
F2 ¼ locations at which scale would be deposited because of the surface position with respect to gravity and the surface being con-
cave or convex
F3 ¼ accounts for boned scaling, which is independent of gravity and a strong function of the pressure gradient
F4 ¼ accounts for boned scaling, which is independent of gravity and a strong function of the pressure gradient
I ¼ total ionic strength
K ¼ constant, depends on the total salt concentration and temperature
Ksp ¼ equilibrium constant
pAlk ¼ negative logarithm of total alkalinity
pCa ¼ negative logarithm of calcium concentration
pH ¼ pH of water samples as actually determined
pHs ¼ pH value when the CaCO3 achieves saturation in the system
PCNsp ¼ probability of correct no-scale-formation prediction
Pcsp ¼ probability of correct scale-formation prediction
PINsp ¼ probability of incorrect no-scale-formation prediction
PIsp ¼ probability of incorrect scale-formation prediction
PNs ¼ probability of no-scale formation
Ps ¼ probability of scale formation
PCO2 ¼ partial pressure of CO2
Sf ¼ solubility factor
T ¼ temperature
V ¼ valence of the ion
XCO2 ¼ CO2 mole fraction in water mixture saturated with a gas containing CO2
Xt ¼ total number of wells or cases in a given field
X1 ¼ represents the number of cases in which the scale prediction was correct
X2 ¼ represents the number of cases in which the scale prediction was incorrect
X3 ¼ represents the number of cases in which the no-scale prediction was correct
X4 ¼ represents the number of cases in which the no-scale prediction was incorrect
y_ ¼ scaling-rate function
l ¼ ionic strength

References
Amiri, M., Moghadasi, J., and Jamialahmadi, M. 2013. Prediction of Iron Carbonate Scale Formation in Iranian Oilfields at Different Mixing Ratio of
Injection Water with Formation Water. Energ Source Part A 35 (13): 1256–1265. https://doi.org/10.1080/15567036.2010.514596.
Bahadori, A. 2011. Estimation of Potential Precipitation from an Equilibrated Calcium Carbonate Aqueous Phase Using Simple Predictive Tool. SPE
Proj Fac & Const 6 (4): 158–165. SPE-132403-PA. https://doi.org/10.2118/132403-PA.
Bayona, H. J. 1993. A Review of Well Injectivity Performance in Saudi Arabia’s Ghawar Field Sea Water Injection Program. Paper presented at the
Middle East Show, Bahrain, 3–6 April. SPE-25531-MS. https://doi.org/10.2118/25531-MS.
Castillo, E., Cobo, A., Gutiérrez, J. M. et al. 1999. Functional Networks with Applications: A Neural-Based Paradigm, Vol. 473. Boston, Massachusetts,
USA: Springer International Series in Engineering and Computer Science, Springer.
Hamid, S., De Jesús, O., Jacinto, C. et al. 2016. A Practical Method of Predicting Calcium Carbonate Scale Formation in Well Completions. SPE Prod
& Oper 31 (1): 1–11. SPE-168087-PA. https://doi.org/10.2118/168087-PA.
Jordan, M. M., Williams, H., Linares-Samaniego, S. et al. 2014. New Insights on the Impact of High Temperature Conditions (176 C) on Carbonate and
Sulphate Scale Dissolver Performance. Paper presented at the SPE International Oilfield Scale Conference and Exhibition, Aberdeen, Scotland,
14–15 May. SPE-169785-MS. https://doi.org/10.2118/169785-MS.
Kamal, M. S., Hussein, I., Mahmoud, M. et al. 2018. Oilfield Scale Formation and Chemical Removal: A Review. J Pet Sci Eng 171 (December):
127–139. https://doi.org/10.1016/j.petrol.2018.07.037.
Lakshmi, D. S., Senthilmurugan, B., Drioli, E. et al. 2013. Application of Ionic Liquid Polymeric Microsphere in Oil Field Scale Control Process. J Pet
Sci Eng 112 (December): 69–77. https://doi.org/10.1016/j.petrol.2013.09.011.
Larson, T. E. and Buswell, A. M. 1942. Calcium Carbonate Saturation and Alkalinity Interpretation. J Am Water Works Ass 34 (11): 1667–1684. https://
www.jstor.org/stable/41233315.
Mackay, J. E., Collins, R. I., Jordan, M. M. et al. 2003. PWRI: Scale Formation Risk Assessment and Management. Paper presented at the International
Symposium on Oilfield Scale, Aberdeen, Scotland, UK, 29–30 January. SPE-80385-MS. https://doi.org/10.2118/80385-MS.

22 2020 SPE Production & Operations

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030

 PO198646 DOI: 10.2118/198646-PA Date: 15-July-20 Stage: Page: 23 Total Pages: 23

Moghadasi, J., Jamialahmadi, M., Müller-Steinhagen, H. et al. 2003a. Scale Formation in Iranian Oil Reservoir and Production Equipment during Water Injec-
tion. Paper presented at the International Symposium on Oilfield Scale, Aberdeen, Scotland, UK, 29–30 January. SPE-80406-MS. https://doi.org/10.2118/
80406-MS.
Moghadasi, J. M., Jamialahmadi, M., Müller-Steinhagen, H. et al. 2003b. Scale Formation in Oil Reservoir and Production Equipment during Water
Injection (Kinetics of CaCO4 and CaCO3 Crystal Growth and Effect on Formation Damage). Paper presented at the SPE European Formation
Damage Conference, The Hague, The Netherlands, 13–14 May. SPE-82233-MS. https://doi.org/10.2118/82233-MS.
Ramstad, K., Tydal, T., Askvik, K. M. et al. 2005. Predicting Carbonate Scale in Oil Producers from High Temperature Reservoirs. SPE J. 10 (4):
363–373. SPE-87430-PA. https://doi.org/10.2118/87430-PA.
Stiff, H. A. Jr. and Davis, L. E. 1952. A Method for Predicting the Tendency of Oil Field Waters To Deposit Calcium Carbonate. J Pet Technol 4 (9):
213–216. SPE-952213-G. https://doi.org/10.2118/952213-G.
Tariq, Z., Abdulraheem, A., Mahmoud, M. et al. 2018a. A Rigorous Data-Driven Approach To Predict Poisson’s Ratio of Carbonate Rocks Using a
Functional Network. Petrophysics 59 (6): 761–777. SPWLA-2018-v59n6a2.
Tariq, Z., Mahmoud, M., Abdulraheem, A. et al. 2018b. An Intelligent Solution To Forecast Pressure Drop in a Vertical Well Having Multiphase Flow
Using Functional Network Technique. Paper presented at the PAPG/SPE Pakistan Section Annual Technical Conference and Exhibition, Islamabad,
Pakistan, 10–12 December. SPE-195656-MS. https://doi.org/10.2118/195656-MS.
Tariq, Z., Mahmoud, M., and Abdulraheem, A. 2019. An Intelligent Data-Driven Model for Dean–Stark Water Saturation Prediction in Carbonate
Rocks. Neural Comput & Applic 2019: https://doi.org/10.1007/s00521-019-04674-z.
Vetter, O. J. and Farone, W. A. 1987. Calcium Carbonate Scale in Oilfield Operations. Paper presented at the SPE Annual Technical Conference and
Exhibition, Dallas, Texas, USA, 27–30 September. SPE-16908-MS. https://doi.org/10.2118/16908-MS.
Yeboah, Y. D., Somuah, S. K., and Saeed, M. R. 1993. A New and Reliable Model for Predicting Oilfield Scale Formation. Paper presented at the SPE
International Symposium on Oilfield Chemistry, New Orleans, Louisiana, USA, 2–5 March. SPE-25166-MS. https://doi.org/10.2118/25166-MS.

2020 SPE Production & Operations 23

ID: jaganm Time: 15:25 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-PO##200030