Optimized deep learning model assisted pressure transient analysis for automatic reservoir characterization

An optimized deep learning model consisting of long short-term memory and fully connected neural networks has been proposed for the automatic interpretation of constant rate pressure drawdown tests conducted in infinite acting homogeneous reservoirs. The pressure change and pressure derivative data along with their corresponding log (CDe2S) value has been used for training purpose. The hyper-parameter tuning has been conducted to obtain 80:10:10 data split ratio, batch size of 64, Adam optimization algorithm, learning rate of 0.01, and 100000 dataset size as the most suitable choices for training the model. The mean relative errors of 0.0034, 0.0042, and 0.0046 and mean absolute errors of 0.0438, 0.0556, and 0.0585 have been obtained for the train, validation, and test data, respectively, during training. The performance of the trained model has been validated using simulated data for six pressure drawdown test cases. The minimum and maximum absolute errors of 0.0110 and 0.0813, respectively, have been obtained for the test cases. The proposed model provides high accuracy in predicting log (CDe2S) from pressure change and pressure derivative data input for noisy data as well, with minimized manual intervention.