An Improved Deep Forest Model Combining Adaptive Synthetic Sampling for Automatic Lithology Identification

As lithology is an important attribute describing reservoir characteristics, lithology identification is of great significance for oil exploration and development. Because of the shortcomings of low precision and high cost in traditional methods, machine learning (ML) technology has become the focus of logging lithology identification research. However, the lithology samples imbalance and unextracted complementary information in logging curves limit the performance of ML models. In this paper, an improved deep forest (DF) model combining adaptive synthetic sampling (ADASYN) is proposed to realize automatic lithology identification. To balance lithology samples, ADASYN is adopted to minority lithologies oversampling. Extreme learning machine is utilized as base learner instead of random forest to advance the classification ability and generalization of the DF algorithm. The implied complementary information in logging curves is mined by multi-grained scanning of the algorithm. The effectiveness of the proposed model is verified on the logging datasets of Jianghan oilfield. It is proved that the model can realize accurate lithology identification of unbalanced logs.