Abstract
Rate of Penetration (ROP) optimization has played a key role in the economic return and value of both onshore and offshore wells by decreasing drilling time thereby increasing the net present value (NPV) of the localized field. In this study, an Artificial Neural Network (ANN) model is developed to accurately model the ROP of a well in SW Oklahoma to optimize the drilling process.
A parametric study was conducted to showcase the effect of operational parameters on the ROP-ANN model and to minimize error and increase accuracy. The number of neurons and hidden layers within the model are optimized based on the lowest Mean Square Error (MSE) and highest R2. A comparative study was comprised of one well in Southern Oklahoma targeting the Caney Shale. The well is about 10,000″ vertical with a 2-mile lateral with a maximum inclination of 78° and Dogleg Severity (DLS) of 12°/100ft. UCS was incorporated into the model to geomechanically differentiate between shale, sandstone, and limestone.
The input drilling data is weighed against ROP showcasing the impact of each parameter on ROP. From this and further proven in the results, RPM, WOB, and UCS greatly effect ROP per foot based on the sensitivity analysis but steadily decline as the critical value is achieved.
The major advantage of this study is developing an accurate ANN model for onshore North American shale plays in understanding the lithological impact of UCS and high lateral deviation on ROP which can be used in pre-planning to optimize the drilling processes.
