Forecasting formation density from well logging data based on machine learning model

Formation density can reflect the pressure state and fluid migration of the reservoir, which is crucial for the re-development of depleted reservoirs. Although various prediction models have been developed using density inversion, the Terzaghi correction, and machine learning techniques, these models are difficult to meet the high-precision requirements during the calculation process. This fact limits their effectiveness in oil and gas exploration and development. However, the formation density and the detector counting rate during well logging process exhibit a nonlinear relationship. A system structure integrating Convolutional neural network (CNN) and Transformer is suggested to accomplish the goal of automatic formation density prediction and solve the problem of insufficient model feature extraction ability under multiple logging data conditions. The reason for adopting the integrated structure is to enhance prediction accuracy and robustness through collaborative optimization of multiple models. The CNN mainly extracts feature regions of interest and Transformer encoder is utilized to assign high weights to the regions of interest. The CNN-Transformer model also includes the novel S-shaped Rectified Linear Activation Unit (S-ReLU) function. Based on the counting rates of the detector’s energy windows, the Pearson correlation coefficient method is applied for feature selection. Bayesian optimization combined with K-fold cross validation is used to fine-tune the key model hyperparameters. The proposed CNN-Transformer model is compared with the traditional inversion model, the CNN model and the Transformer model in terms of prediction accuracy. The results demonstrate that the CNN-Transformer model offers greater robustness and smaller prediction deviations than other machine learning models. This study provides a reliable and fast approach for predicting formation density while minimizing exploration cost and improving exploration efficiency for oil and gas.