Designing a Model for Predicting and Enhancing Drilling Performance

Abstract

In the oil and gas business, mining, and underground engineering, drilling activities are very important. It is very important to get the best results while lowering practical risks and costs. Rate of Penetration (ROP) and other traditional scientific and physics-based models for predicting drilling success have been limited because they can't adapt to settings that aren't linear, have a lot of dimensions, and change over time. This study suggests a strong data-driven approach that uses optimisation methods and machine learning techniques like Random Forest Regressor (RFR) and Artificial Neural Networks (ANN) to predict and improve drilling performance. Real-life drilling data was cleaned up, normalised, and put through feature engineering to find the most important practical factors. The suggested ANN model, which had two hidden layers and was trained using the ReLU activation function and the Adam optimiser, did much better than baseline models, with a R² score of 0.91 and much lower RMSE and MAE. The model was then combined with a Genetic Algorithm to find the best drilling settings. This led to an estimated 12% increase in the expected ROP. Visualisation tools like SHAP were added to make the model easier to understand, so it is both correct and easy to explain. The results show that smart prediction systems could change drilling operations into processes that are flexible, run in real time, and are optimised.