Production Prediction for Shale Oil Horizontal Wells Using Convolutional Neural Network-Bidirectional Long Short-Term...

Area of Science:

• Petroleum Engineering
• Artificial Intelligence in Energy
• Unconventional Resources

Background:

• Shale oil is crucial for global energy security but challenging to predict due to reservoir complexities.
• Conventional methods struggle with low porosity/permeability and complex fracture networks in shale reservoirs.
• Accurate production forecasting is vital for optimizing development of unconventional oil and gas.

Purpose of the Study:

• To develop a novel and accurate production prediction model for shale oil horizontal wells.
• To integrate Convolutional Neural Networks (CNNs) for spatial feature extraction and Bidirectional Long Short-Term Memory (BiLSTM) networks for temporal dependency.
• To validate the proposed CNN-BiLSTM model using real-world production data.

Main Methods:

• Developed a hybrid CNN-BiLSTM model for shale oil production forecasting.
• Utilized production data from horizontal wells in the Ordos Basin's Chang 7 Member for validation.
• Benchmarked the CNN-BiLSTM model against Support Vector Regression (SVR), Random Forest (RF), XGBoost, LSTM, and BiLSTM.

Main Results:

• The CNN-BiLSTM model demonstrated superior performance over all benchmarked algorithms.
• Achieved a Root-Mean-Squared Error (RMSE) of 0.3103 and Mean Absolute Error (MAE) of 0.1731.
• Obtained a Mean Absolute Percentage Error (MAPE) of 3.2333% and a coefficient of determination (R²) of 0.8856.

Conclusions:

• The CNN-BiLSTM model offers an effective and robust solution for shale oil horizontal well production forecasting.
• The model's synergistic approach enhances prediction accuracy for complex unconventional reservoirs.
• Findings provide valuable insights for developing unconventional oil and gas resources.