ADLM-LWD: Adaptive Deep Learning Model for Lithology Identification from Logging- While-Drilling Data with Online Learning Mechanism

Accurate lithology identification from Logging-While-Drilling (LWD) data is crucial for optimizing drilling operations and enhancing reservoir characterization in the oil and gas industry. However, the lack of standardized datasets and the need for real-time adaptability in varying geological environments pose significant challenges. This study addresses these issues through three key contributions. First, a comprehensive dataset for lithology identification is constructed, combining LWD features with corresponding lithology labels, providing a valuable resource for deep learning applications in this field. Second, a novel deep learning model is proposed, integrating 2D convolution and bidirectional LSTM to effectively capture both spatial and temporal features of LWD data. The model demonstrates high stability and performance across various metrics in extensive cross-validation tests. Lastly, an Adaptive Incremental Learning Algorithm (AILA) is introduced, incorporating online learning mechanisms and dynamic adaptation strategies to enhance the model's real-time adaptability. These advancements contribute to more accurate and adaptive lithology identification systems, which potentially improve real-time decision-making in drilling operations and lead to more efficient and cost-effective exploration and production processes.