Combination Model for Sparry Calcite Intelligent Identification in Dongying Depression

In shale oil and gas exploration, rock structure is of paramount importance. The efficient identification of sparry calcite, which serves as an indicator of favorable lithology, is of great significance. But the logging response to sparry calcite is weak and sample sizes are limited. It is therefore difficult to accurately identify sparry calcite by conventional methods. To overcome this problem, we propose a Combination model based on data preprocessing techniques. The model utilizes conventional logging curves as the experimental data. First, the dataset is optimized by SMOTE and wavelet transforms. Subsequently, a CNN-Attention parallel network is trained to extract nonlinear features from the input logging curves. These features are then fed into the BiLSTM to extract potential information regarding the depth domain sequence. Finally, the features are summarized to achieve accurate identification of sparry calcite. The presented model is compared with CNN, BiLSTM, and CNN-BiLSTM models. The CNN-BiLSTM-Attention model provides the most favorable identification results in experiment. The combined model model possesses significant sample optimisation results and feature extraction capabilities, and it can highly efficient identification of sparry calcite in shales.