Improving the accuracy of short-term multiphase production forecasts in unconventional tight oil reservoirs using contextual Bi-directional long short-term memory

To improve the accuracy of short-term multiphase production forecasts with one-step-ahead predictions, a Contextual Bi-directional Long Short-Term Memory (C-Bi-LSTM) was developed. C-Bi-LSTM contains two neural networks: the Feed Forward Neural Network (FFNN) and the Bi-directional Long Short-Term Memory (Bi-LSTM). Both networks complement each other to learn the forecasting task in sync. The FFNN learns to extract useful features from static parameters (e.g., initial completion parameters) to modify future production forecasts. Meanwhile, Bi-LSTM learns the effect of dynamic parameters such as oil rate, gas rate, water rate, and future control policy (shut-in) on production. Bayesian optimization was utilized to fine-tune the hyperparameters of the models and C-Bi-LSTM was compared with models such as Linear Regression (LR), Artificial Neural Network (ANN), Long short-term memory (LSTM), Gated Recurrent Unit (GRU), and Bi-LSTM. The model evaluation was performed with two cases, namely Case I (833 wells) and Case II (54 wells). The wells in both cases produce from the unconventional Middle Bakken and the Three Forks formations in Mountrail County, ND. In Case I, the production history of each well was split into 60 %-20 %-20 % for training, validation, and testing, respectively, and the models were evaluated based on the test set using the hindcasting approach. In Case II, the models were evaluated on wells that were not included during training to assess their generalization performance in space. The results indicate that C-Bi-LSTM outperforms the other models in both cases, maintaining a more stable accuracy throughout training, validation, and testing. Benefits of the proposed model include (1) improved shortterm production forecasting accuracy, (2) integration of static parameters to directly influence production time series, (3) flexibility of its architecture for future research, and (4) potential cost savings for operators in shortterm production optimization tasks.