Application of deep learning in predicting fracture porosity

Tóm tắt

Deep learning (DL) neural network analysis is the latest development from the Artifi cial Neural Network (ANN) and it is being used more and more in petroleum engineering. In this study, the way to develop a new DL model for well log analysis was attempted and successfully implemented using well log data from a location in the Cuu Long basin, Vietnam. Three sets of analyses were conducted,ie., the first analysis set with a single hidden layer ANN model, the second analysis set with multiple hidden layer ANN model and the third with a DL neural network model. The DL-predicted porosity for a fractured granite basement reservoir of an oil field in the Cuu Long basin was found in the range from 0.0 to 0.082, showing a good match with the conventionally-calculated values. The final deep learning model consists of 5-input layers of gamma ray (GR), deep resistivity (LLD), sonic (DT), density (RHOB) and neutron porosity (NPHI), having 5 hidden neuron layers with 14 neurons per layer. It is worth noting that the transfer function of the rectified linear unit (ReLU), typical for a deep learning analysis, was implemented to replace the common sigmoidal transfer function, ensuring the successful application of DL model. Last but not least, the problem of vanishing gradient specific for a DL neural network model was also explained in details in this paper.