Speaker at Oil, Gas and Petroleum Engineering 2022 - Oyindamola Obisesan
University of Oklahoma , United States
Title : Optimization of ROP with Drilling Parameters and Petrophysical Data Using Machine Learning Approach


Rate of Penetration (ROP) is an important parameter during oil and gas drilling which influences drilling performance. Methods to optimize ROP involve selecting optimum drilling variables prior to a run or real-time optimization of ROP by varying Weight on Bit (WOB), Revolutions per Minute (RPM), Flowrate, and Total Flow Area (TFA).  
Most of the methods for optimizing ROP using machine learning techniques involve only the drilling parameters without consideration of the formation. The method presented in this paper utilizes petrophysical data in addition to drilling data for ROP optimization. The data collected was cleaned to remove missing data and other data inconsistencies. Data pre-processing measures which included data scaling, feature correlation, feature selection, etc. were applied prior to data modeling. ?? 
The 17.5in and 12.25in well sections were analyzed and the formation was divided into clusters based on the formation evaluation data using an unsupervised algorithm (K-means). Five machine learning algorithms which include Linear Regression, Random Forest, Support Vector Machine, K Nearest Neighbors (KNN), and Xtreme Gradient Boosting in addition to sensitivity analysis were trained and tested on data from 10 wells to optimize the drilling parameters to optimize the ROP. The best model (Random Forest) which accurately predicts ROP for each formation layer was selected based on the least Root Mean Square Error (RMSE). ?  
The analysis of our models show that the best model which optimizes ROP is that which incorporates both the formation and drilling data, as this had the highest model accuracy and lower RMSE when compared to the model which utilized only drilling data. 

What will the audience learn from your presentation? 
•    Explain how the audience will be able to use what they learn?
o    The lessons from this work would help the audience to utilize the huge volume of drilling data generated. 
o    The knowledge of machine learning approaches and algorithms that would be gained from this work would be useful for petroleum engineers to use in various areas of petroleum operations. 
o    The approaches utilized in this work can be extended to practical drilling operations to optimize drilling parameters and safely maximize ROP and reduce drilling time. 
•    How will this help the audience in their job? 
o    This research would help drilling engineers and other members of the drilling team to optimally plan wells and reduce downtime. 
•    Is this research that other faculty could use to expand their research or teaching? 
o    The findings from this work can be extended to drilling systems and software that would suggest drilling parameters in real-time during drilling operations. 


Oyindamola holds a bachelor's degree in Petroleum Engineering from the Afe Babalola University Ado-Ekiti, Nigeria, she also holds a dual master's degree in Petroleum Engineering and Data Science & Analytics from the University of Oklahoma. She currently works as a Data Scientist at Visa.