Missing Well Log Prediction Using Deep Recurrent Neural Networks
N. Pham and E. Zabihi Naeini
Event name: 81st EAGE Conference and Exhibition 2019
Session: Deep Learning and Data Analytics – Methods and Applications I
Publication date: 03 June 2019
Info: Extended abstract, PDF ( 1.11Mb )
Price: € 20
The integration of multiple well logs together with seismic data is fundamental to construct an accurate hydrocarbon reservoir model. However, not all well log types are available at every single well in an area of interest because of cost limitations or borehole problems. We propose a method to predict missing logs using a deep recurrent neural network. The deep learning architecture consists of bidirectional recurrent network with LSTM cells cascaded with fully-connected neural network. The recurrent elements of the network allow to account for geological depth trends in well logs. The predictions are facilitated by a Monte-Carlo simulation and a dropout layer to quantify the uncertainties. Using wells from the Central North Sea, the results of applying this method are comparable with rock physics based approach but with a faster turnaround. Another advantage is that deep learning allows the integration of multiple well log types which can potentially help to improve the accuracy of the predictions.