ReFormer: Lithology Identification via the Improved Transformer Model with Well Logging Data

Predicting lithology with high accuracy is crucial for effective oil and gas extraction, influencing operational strategies significantly. This study introduces the Recurrent Transformer (ReFormer) model, an innovative blend of Transformer frameworks and recurrent dynamics, engineered for superior processing of well logging data. It integrates a unique Recurrent Scale-wise Attention (RSA) feature, designed specifically for well logging applications. This feature efficiently harnesses multi-level data details, enhancing the model’s contextual analysis capabilities. It relies on crucial well log curves to link lithological features with logging measurements, and comparative assessments reveal its superior performance and adaptability in varying geological settings. This model marks a substantial leap in applying machine learning to lithology prediction within well logging, providing a robust tool for advancing resource identification and extraction, illustrating the transformative impact of sophisticated machine learning in earth science studies.