Artificial Lift Selection Using Machine Learning

When it comes to capturing something crucial from the past experiences than which is faster and more accurate — a human brain or a Machine Learning model trained using those past experiences? The Oil industry is witnessing the application of Artificial Intelligence in various areas, Lift selection is one such area where Machine Learning can be leveraged.

Why Artificial Lifts

When The Reservoir Pressure has become so less that the production rate has become too low to be called economical or when well stops producing at all then a pump provides a draw down or a lift to allow/improve the production. It is considered as primary recovery method as we are not interfering the reservoir but exploiting the well using natural energy.

Popular Artificial Lift Methods

• SRP(Sucker Rod Pump)- most popular, simple design, not good for high GOR.
• ESP(Electrical Submersible Pump)-high production rate, not suitable for high temperature wells.
• PCP(Progressive Cavity Pump)- Energy efficient, Capable of Pumping high Viscous oils.
• Gas Lift- Highly recommended for solids production and high GOR wells.
• Hydraulic Pump
• Jet Pump

Selection Criteria

1. Well and Reservoir Characteristics.
2. Field Location.
3. Operational Problems.
4. Economics.

Data

To identify the suitable lift method the data set consists of 4 parameters- Formation depth, Flow rate, reservoir temperature and API gravity of oil. The lift selection depends upon various other variables like water cut, Gas oil ratio, economics(cost of installation,maintenance) etc. But here we are limited by the data.

Let’s Go

The data seems pretty simple for a ML model to train on it because it is simple as it is balanced and has only 4 variables. For a production technologist it could be imbalanced and will have a lot more parameters but take this one just for learning.

Splitting the data

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)

Fitting the model

Decision Tree classifier is found suitable but for a complex data ensembling may be useful. . Decision tree algorithm provides not only a good prediction performance and training speed but also intuitive understanding.

from sklearn.tree import DecisionTreeClassifier
dtree = DecisionTreeClassifier()
dtree.fit(X_train,y_train)

Prediction

y_pred = dtree.predict(X_test)
y_pred = pd.DataFrame(y_pred)

Performance

from sklearn.metrics import classification_report, confusion_matrix
print(classification_report(y_test,y_pred))

The reason for high accuracy is less complex and balanced data.

Benefit of ML approach

The Artificial Lift selected by machine learning was performed on 9 wells resulted in lowering the normalized cost as seen in yellow dots.

Source — Thanawit Ounsakul, Thum Sirirattanachatchawan, Wiwat Pattarachupong, Yaovanart Yokrat, and Peerapong Ekkawong, PTTEP. ” Artificial Lift Selection Using Machine Learning”, IPTC-19423-MS

References-

Thanawit Ounsakul, Thum Sirirattanachatchawan, Wiwat Pattarachupong, Yaovanart Yokrat, and Peerapong Ekkawong, PTTEP. ” Artificial Lift Selection Using Machine Learning”, IPTC-19423-MS

Motivation- Divyanshu Vyas(https://www.linkedin.com/in/divyanshu-vyas/)

I will try to improve this blog.

Until then happy reading.