LAUSR

Lithology information is critical to the adjustment of drilling control strategies, and can be identified by training a classification model from the well logging data. However, achieving accurate lithology identification is rather difficult owing to complex characteristics, such as data imbalance, data-overlapping, and multi-classification. In this brief, a hybrid lithology identification method is developed based on the Reducing Error Correcting Output Code algorithm with the Kernel Fisher Discriminant Analysis (RECOC-KFDA). The effectiveness of the proposed method is demonstrated based on case studies with the UCI machine learning database and the real logging data. The results show that the proposed method has superior performances compared to conventional methods.