Production Forecasting of Shale-Oil Horizontal Wells Using a Duong-Constrained Transformer with Singular Spectrum Analysis

Summary

Accurate production forecasting of shale-oil horizontal wells is essential for optimizing reservoir-development strategies and supporting informed decision-making. However, shale-oil production exhibits strong nonlinearity, significant noise, and long-term temporal dependencies, which limit the prediction accuracy, stability, and interpretability of conventional forecasting methods and purely data-driven models. Therefore, we propose a hybrid data-physics forecasting framework that combines singular spectrum analysis (SSA), the Duong decline model, and a Transformer architecture. SSA in the proposed framework is first used to decompose production time-series data and enhance feature representation, mitigating signal noise. The Duong decline model is then incorporated as a physics-informed constraint to guide the model training process and improve prediction stability and interpretability. Then, a Transformer-based model with a self-attention mechanism is utilized to capture complex long-term dependencies in production sequences and achieve high-precision forecasting. Experiments are conducted using field production data from shale-oil horizontal wells in the Ordos Basin. Comparative results demonstrate that the proposed SSA-Transformer-Duong model outperforms several conventional machine learning and deep learning models in both single-step and multistep forecasting tasks, achieving lower root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) values and higher coefficient of determination (R2) scores. Ablation studies confirm the effectiveness of SSA-based feature enhancement and Duong-based physics-informed constraints. In addition, Shapley additive explanations (SHAP)-based interpretability analysis reveals that historical production sequences dominate model predictions, while geological and engineering parameters provide essential contextual constraints. The proposed framework significantly improves the accuracy and robustness of shale-oil production forecasting, providing a promising approach for data-driven prediction of complex oil and gas production dynamics.