Application of Adaboost-Transformer Algorithm for Lithology Identification Based on Well Logging Data

In the field of oil and gas exploration, accurately predicting lithology during well logging is crucial. This research introduces a novel approach, the Adaboost-Transformer method, which utilizes data mining techniques to enhance logging lithology prediction. The first step involves applying the Adaboost algorithm for selecting features, which develops a strong classifier by focusing on weighted observations and particularly challenging samples. This approach not only boosts accuracy and durability but also lessens the likelihood of overfitting. Following this, we implement the Transformer as the principal classifier in creating the Adaboost-Transformer model. The Transformer is specifically designed for sequential data and is highly efficient in modeling sequences, especially in capturing temporal links within well logging data. When tested on two separate well logging datasets, this innovative model exhibited superior lithology identification capabilities, achieving accuracy rates of 95.20% and 95.50%. These figures notably surpass those of the standalone Transformer model, which scored 91.50% and 91.10% in accuracy. In comparison, the random forest (RF) model displayed more limited performance with accuracy rates of 88.80% and 87.10%.