Real-Time Prediction of Formation Types from Drilling Data Using Artificial Intelligence

Abstract

This study addresses the challenge of predicting formation types in petroleum field through a machine learning approach. Utilizing around 10,000 datasets from two wells in the Hassi Messaoud field, the model, named "Ama" (AI Mud Logger Assistant), demonstrates a mean accuracy of 70% in the training phase and high precision in predicting formation classes. Testing data confirms its effectiveness. The research overcomes limitations such as time lags and the lack of real-time data, employing seven drilling parameters rather than the traditional sole focus on the rate of penetration (ROP). The model was trained on a carefully preprocessed dataset from the Hassi Messaoud field to ensure unbiased and balanced samples. The primary purpose is to develop an machine learning model to accurately predict of geological formation types from drilling data. This leads to improve precision, reduce drilling time and costs, and enhance efficiency by minimizing uncertainties. This approach aspires to offer significant insights into geosciences, formation type detection and exploration practices