Intelligent Lithology Identification of Deep Coal Seam Driven by Drilling Parameters

ABSTRACT: Deep coal seams are characterized by complex lithology and developed pore fissures. Geosteering techniques are ill-suited to these conditions due to their high cost and the risk of tool burial resulting from wellbore collapses. Therefore, the formation of a reliable method for detecting the lithology of deep coal seams is conducive to improving the drilling encounter rate of horizontal well sections and ensuring that the well trajectory is drilled in the coal seams. In this paper, we take the No.8 deep coal seam of the Da Niu Di gas field in the Ordos Basin of China as the research object, and statistically organize six drilling data, namely drilling time, bottomhole torque, bottomhole pressure, rotational speed, drilling speed, and mechanical specific energy (MSE), and carry out data cleaning and quality control processing. Combined with the rock chip logging data to classify the lithology in the study area into three types of lithological labels: limestone, coal seam and mudstone. Combining Pearson correlation coefficient and mutual information method, the bottomhole torque, rotational speed, and mechanical specific energy are preferred as the lithology identification parameters of deep coal seams, and eight types of machine learning models, such as K-Nearest Neighbor, Support Vector Machines, Random Forest, Gradient Boosting, etc., are established to realize the quantitative classification prediction based on the lithology identification parameters of deep coal seams, and the application analysis of actual deep coal seam horizontal sections is carried out. Classifier models are further established to realize the quantitative classification prediction of lithology labels based on the lithology identification parameters of deep coal seams, and the application analysis of lithology identification in the horizontal section of actual deep coal seams is carried out.The findings indicate that, in comparison with the lithology observed in the borehole trajectory, the Random Forest model exhibits an accuracy of 96.59% in lithology identification of the horizontal section of the deep coal seam. This model demonstrates robust generalization capabilities and rapid computation speeds, enabling efficient identification of the deep coal seam in real time. It provides a foundation for subsequent, precise evaluation of the reservoir and hydraulic fracturing modification.