Data-Driven Neural Network Model to Predict Equivalent Circulation Density ECD

Abstract

Equivalent circulation density (ECD) is vital in drilling operations. Poor management of ECD can lead to many drilling obstacles that can increase the delivery time of the well. The aim of this study is to utilize artificial neural networks (ANNs) to create a model to estimate ECD prior to drilling. Data of more than 2000 wells collected from multiple sources worldwide were utilized in this work. An ANN model was created with one hidden layer and 12 neurons in the hidden layer. The data were clustered into three data sets; training (70% of the data), verification (15% of the data), and testing (15% of the data). Ten training algorithms were utilized to train the network, the training algorithm with the lowest mean squared error (MSE) and the highest R2 was selected to achieve the best predictive model. Bayesian Regularization (BR) algorithm was selected to train the model because it had the highest R2 and the lowest MSE. The results showed that the created model can predict ECD within an acceptable margin of error. The overall R2 of the model was 0.982 which is considered very good. Alternatively, given a target of ECD, the created model can be utilized in reverse to obtain the desired ECD by altering the key drilling parameters affecting the ECD model (the inputs). This will help the drilling personnel to optimize ECD in the field. Intelligent systems and machine learning have proven their effectiveness in solving complicated problems that cannot be solved analytically. With the large historical drilling data available in the oil and gas industry, machine learning and intelligent systems can be used to make better future decisions that will help to optimize the drilling operations.

Meeting Name

SPE Gas and Oil Technology Showcase and Conference 2019, GOTS 2019 (2019: Oct. 21-23, Dubai, UAE)

Department(s)

Geosciences and Geological and Petroleum Engineering

Research Center/Lab(s)

Center for Research in Energy and Environment (CREE)

Keywords and Phrases

Gas industry; Intelligent systems; Learning algorithms; Machine learning; Mean square error; Neural networks, Bayesian regularization; Drilling operation; Drilling parameters; Equivalent circulation density; Neural network model; Oil and Gas Industry; Predictive modeling; Training algorithms, Infill drilling

International Standard Book Number (ISBN)

978-161399704-8

Document Type

Article - Conference proceedings

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2019 Society of Petroleum Engineers (SPE), All rights reserved.

Publication Date

01 Oct 2019

Link to Full Text

Share

 
COinS