Using Artificial Neural Networks to Predict Formation Stresses for Marcellus Shale with Data from Drilling Operations

Abstract

Artificial neural networks have been applied to different petroleum engineering disciplines. This is contributed to the powerful prediction capability in complex relationships with enough data available. The objective of this study is to develop a new methodology to predict the vertical and horizontal stresses using artificial neural networks for Marcellus shale well laterally drilled in Monongalia County, WV. This approach coupled the drilling surface measurements with the recorded well logging data. Drilling parameters included depth, WOB, RPM, standpipe pressure, torque, pump flow rate and rate of penetration. Well logging data included gamma ray and bulk density. The model output was the minimum horizontal stress and vertical stress. The well trajectory was divided into two main parts, the vertical and lateral section since the change in the drilling direction along with changing structural geology and sedimentation impacted the resultant stresses. Several neural networks were designed with different number of feedforward backpropagation architectures. The collected data was filtered and normalized before neural networks were trained using part of data. A percentage of the data was used to validate the trained model. Finally, a blind data set aside was used to test the model prediction accuracy and to estimate error percentages. Preliminary results show that adding logging data such as gamma ray and bulk density improves the model accuracy. Also, increasing the number of hidden layers and neurons improved the efficiency. However, higher the number of neurons and hidden layers higher was the computational cost due to increased model convergence time. The correlation coefficients of the predicted and observed values ranged between 0.76 and 0.99. This approach is beneficial regarding hydraulic fracturing design and fracture orientation prediction in unconventional shales. iii Dedication Dedicated to my dear God, without Him guiding me in every step, nothing would be possible for me. This is also dedicated to my father Salem Abusurra who's presence would make this achievement even more special. Also, I would like to present this to my beloved mother Fatima Omar Mousa whose constant sacrifices give me a better life and led me to this point. iv ACKNOWLEDGEMENT I would like to express my deepest thanks and appreciations to my research advisor Dr. Ilkin Bilgesu for his continuous support and encouragement throughout this research. His contributions are numerous and valuable. I would also like to thank all members of my research committee for their guidance and helpful suggestions. I heartily appreciate the support and advice from Professor Sam Ameri, Chair of Petroleum and Natural Gas Engineering Department. He has been a father and friend throughout my study. My thank is to my academic advisor Dr. Kashy Aminian for his support and advice and special thanks to all family of Petroleum and Natural Gas Engineering Department especially to Ms. Beverly Matheny for her friendly ambiance and continuous help. My sincere appr...

Ensemble Learning Based Sustainable Approach to Carbonate Reservoirs Permeability Prediction

“…They used their data, which was gathered from five distinct wells located in the Middle Eastern region and obtained an improvement in the coefficient of correlation using the ANN correlation-based technique with 93.76%. Abusurra [17] used ANN while developing a new method to predict the vertical and horizontal stress for Marcellus shale well drilled in the County of Monongalia, West Virginia. The data used is from the drilling surface calibration measurements combined with the recorded well logging data over time.…”