Probabilistic logging lithology characterization with random forest probability estimation

Abstract Borehole lithology discrimination is the foundation of formation evaluation and reservoir characterization. Due to the limitation of costing or accuracy, direct discrimination methods such as borehole core and drilling cutting analysis are unable to be deployed to every well, while logging lithology interpretation provides an alternative solution for this. Recently, several machine learning algorithms such as the neural network, support vector machine, decision tree, and random forest have already been employed by researchers for automatic logging lithology interpretation. However, the vast majority of these studies belong to the category of deterministic lithology characterization. In this article, we propose a probability based fuzzy characterization method for more effective logging lithology interpretation. Moreover, to improve the accuracy of lithology probability estimation, we propose the probabilistic random forest algorithm and investigate its advantages referred to 8 existing probability estimation algorithms. Through the comparative experiments on 9 real-world logging lithology interpretation tasks, the feasibility and advantage of the proposed method are confirmed. Application case demonstrates that compared with traditional deterministic lithology characterization methods, probabilistic lithology characterization is able to provide more information about rhythm, heterogeneity, and formation properties, which worths further application and promotion to improve the fineness of formation evaluation and reservoir characterization.