Well Log Data Preparation and Effective Utilization of Drilling Parameters Using Data Science Based Approaches

Abstract

Researchers in the early 1990s utilized neural networks for different applications in the oil and gas sector. Gradually, with the development of powerful algorithms and computing power applications of random forests, clustering, isolation forests, XGboost increased. The digital transformation in the oil and gas sector began because of two downturns. Literature review suggests that not much work has been published on outlier identification in the geophysical well logs. The proposed workflow uses a machine learning (ML) approach to clean and condition the data and handle missing values. The planned workflow will construct the cleaned composite log data in quick time which can then directly be consumed by 1D geomechanical workflows. The proposed study has two-fold objectives to utilize drilling parameters. Firstly, use an unsupervised machine learning (ML) algorithm like K-means and Hierarchical clustering to classify drilling data of 6 wells into meaningful clusters. These clusters are further classified based on their relationship with rate of penetration feature which defines the drillability. Secondly, the proposed workflow will utilize predictive machine learning algorithms like XGboost and random forest to estimate geophysical logs indicating rock type using the drilling parameters. The results and conclusions of this study are expected to save time in data preparation steps of 1D geomechanical modelling and lead to effective utilization of drilling parameters. This study will redound to the benefit of geoscientists and drilling data analysts in their day-to-day work. Geoscientists spend an enormous amount of time in data preparation before getting to the modelling stage. This workflow saves more than 50% of the time invested in the identification of outliers and hence reducing the overall turnaround of time of the modelling. Furthermore, the study gives insight of hidden trends in drilling parameters to the drilling optimization engineers. The clustering algorithms helped in the identification of high-performance and low-performance clusters and knowledge of their characteristics can optimize drilling operations in the field. Lastly, the predictive algorithms will aid in the understanding of downhole rock lithofacies in near-real-time.