Missing well logs prediction using deep learning integrated neural network with the self-attention mechanism

Abstract

Well logs are employed for analyzing lithology, determining formation parameters, and evaluating oil and gas reservoirs. However, in practice, well logs are often incomplete or distorted. Due to the complexity of underground structures and media heterogeneity, obtaining accurate results by using existing prediction methods is challenging. Thus, a reliable missing well logs prediction method must be developed. In this study, to estimate the missing well logs, we developed a deep learning model that combines the convolutional neural network (CNN) and bidirectional gated recurrent unit (BGRU) network with the self-attention mechanism. The proposed model comprises two modules. One module uses a CNN to extract the local morphological features of logging data, and the other one uses a BGRU to mine the variation trend and context information of logging data with depth from the output features of the CNN module. Next, the self-attention mechanism enables the network to allocate weights to highlight relevant information, thus improving the prediction accuracy. The application results on actual field data in two different areas demonstrate that the proposed model yields accurate and reliable prediction results and has feasibility and practicability.