Reservoir Porosity Prediction Based on Deep Learning and Attention Mechanism

Porosity is an essential parameter in reservoir characterization and evaluation. At the same time, conventional porosity prediction methods exhibit several limitations, such as the high cost of drilling and coring, and inadequate exploration of logging curve patterns. In this study, a novel approach based on the 1DCNN-BiLSTM network with an attention mechanism was presented to facilitate the solution of these problems. First, the network extracts the local features of the logging curves using 1D Convolutional Neural Network, and then extracts the long-term sequence features of the logging curves using the Bidirectional Long Short-Term Memory network, while adding an attention layer behind the Bidirectional Long Short-Term Memory network to further highlight the features that are more significant for porosity prediction and minimize the influence of other features to improve the accuracy of porosity prediction. We applied the established network to real field area data; the prediction results show that our network has better prediction performance and generalization ability compared to the LSTM network and BPNN.