Revolutionizing fluid identification in well logging data with a novel framework of progressive gated transformers and multi-scale temporal features.

The article focuses on the development and evaluation of a novel machine learning model, the multi-scale temporal feature (MTF)-progressive gated transformer (PGT) module, designed for fluid classification using well logging data in oil and gas exploration. The MTF module extracts features from well logging signals at multiple temporal scales, capturing both local fluctuations and long-term trends, while the PGT module integrates convolutional operations with transformer encoders to effectively model complex sequential data by combining coarse-grained and fine-grained features. Tested on well logging datasets from the Tarim Oilfield, the MTF-PGT model demonstrated superior accuracy and robustness in classifying subsurface fluids compared to other models, including traditional recurrent neural networks and ensemble methods. The study highlights the model’s potential applicability across diverse geological settings and suggests its extension to related fields such as mineral exploration and environmental monitoring.