A Novel Method for Real-Time Identification of Formation Lithology Based on Machine Learning

: Real-time identification of formation lithology is of great significance to optimization of drilling rate and directional drilling. However, the experimental observation method and the statistical analysis method require researchers with rich experience and professional knowledge, and the identification efficiency is low and the cost is high. The goal of this study is to employ machine learning for identifying lithology from only the real-time drilling parameters without any downhole measurements. First, the real-time drilling data were cleaned, and then the correlation between each drilling parameter and formation lithology was analyzed by correlation analysis algorithm, and 13 drilling parameters were selected as model inputs. And then random forest (RF) and XGBoost algorithms are used to develop lithology identification models respectively. The results show that the XGBoost model has the best result in identifying formation lithology, with an accuracy of 79.21%. Finally, feature importance analysis shown that MWI, MWO and MFI have important effects on model performance. This study has important implications for improving the probability of drilling into reservoirs and reducing drilling costs. 1. INTRODUCTION Lithology identification is an important basic problem in the fields of geology, oil and gas exploitation, resource exploration, geotechnical engineering, etc (Xu et al., 2021). In drilling engineering, accurate and timely determination of the formation lithology at the positive bit is one of the most important factors to ensure safe and efficient drilling (Al-AbdulJabbar et al., 2019; Zhang et al., 2017; Shan et al., 2015). The classical lithology identification methods mainly include two kinds, one is the experimental observation method, through the direct observation of rock specimens or rock cast thin slices, the rocks can be classified according to their mineral composition, color, etc (Guojun et al., 2010; Hu et al., 2010). The other is the statistical analysis method, the logging cross-section chart is drawn by measuring the logging response characteristics of pure rocks, and then the rock type is roughly determined according to the actual logging values (Teama et al., 2016; Zhou et al., 2016). The above two traditional lithology identification methods require researchers with rich experience and professional knowledge, and the identification efficiency is low and the cost is high. Therefore, a fast and accurate method for lithology identification is urgently needed.