Speaker
Description
The shale gas revolution has underscored a critical requirement for accurate production forecasting to guide resource management and economic planning. However, the complex physical processes in shale formations render traditional numerical simulations inadequate. Here, we present a hybrid artificial intelligence model that synergizes a BERT-based architecture for capturing nonlinear temporal dependencies with Lasso regression for feature selection. Trained on a comprehensive dataset comprises approximately 100,000 data points collected from 78 wells that integrates static geological parameters with dynamic production profiles, our framework is further constrained by physical laws to ensure predictive robustness and interpretability. The model achieves a predictive Average accuracy of R2 = 0.80, significantly surpassing conventional deep learning benchmarks. By leveraging SHAP value analysis, we decode the model's decision-making process to identify key drivers of production, enabling the data-driven optimization of hydraulic fracturing parameters. A subsequent net present value (NPV) assessment demonstrates that this approach can substantially enhance recovery factors and economic returns during the early design phase of development projects. Our work establishes a generalizable, AI-powered paradigm for optimizing extraction strategies in complex subsurface energy systems.
| Country | China |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
