Research on Weak Formation Interval Identification in Oil Drilling Based on CNN-BiLSTM-MHA Model

Abstract

Weak formation intervals in oil drilling are critical sections that may induce complex downhole problems, such as wellbore instability and borehole enlargement. Their accurate identification is of great significance for improving drilling safety and operational efficiency. To address the limitations of traditional weak formation identification methods, including strong dependence on expert experience, high subjectivity, and insufficient utilization of local logging responses and depth-wise sequential dependencies, this paper proposes a weak formation identification method based on a CNN-BiLSTM-MHA model. First, weak formation identification samples are constructed using logging curves and well trajectory parameters, and data preprocessing is performed through missing value handling, outlier correction, standardization, and sliding-window sampling. Then, a convolutional neural network is used to extract local response features, a bidirectional long short-term memory network is employed to model forward and backward dependencies along the depth direction, and a multi-head attention mechanism is introduced to enhance the representation of key intervals and critical features. Finally, comparative experiments and ablation experiments are conducted to verify the effectiveness of the proposed model. Experimental results show that the proposed model achieves an Accuracy, Precision, Recall, and F1-score of 91.52%, 90.28%, 90.15%, and 90.21%, respectively, outperforming comparative models such as SVM, XGBoost, TCN, and Informer. The proposed method can provide methodological support for intelligent identification of weak formation intervals in oil drilling.