Physics-informed machine learning model based on Kolmogorov–Arnold networks for shale oil reservoirs: Robust production forecasting and flow mechanism insights

The complex, multi-scale mass transfer processes and mechanisms in shale oil reservoirs present significant challenges for the rapid and accurate prediction of production rates. To address these challenges, this study proposes a physics-informed machine learning hybrid model designed to overcome the high computational cost of numerical simulations and the limited applicability of semi-analytical methods. The model integrates the Dynamic Drainage Area concept to achieve efficient and highly convergent solutions, which are then hybridized with a Kolmogorov–Arnold Network (KAN), utilizing field data as labels. The performance of the developed model was benchmarked against Transformer-based time-series forecasting frameworks and a Multi-layer Perceptron integrated hybrid model. Results from comprehensive experiments demonstrate that the proposed methodology achieves superior predictive accuracy and significantly enhanced computational efficiency, being five times faster than the conventional numerical model. This approach leverages the computational efficiency and convergence guarantees inherent to semi-analytical models, while enhancing predictive accuracy and generalizability through the regression capabilities of KAN, providing valuable insights into complex flow mechanisms. The model was successfully applied in a shale oil well, demonstrating high-fidelity production forecasting capabilities with matching accuracy exceeding 95%, an improvement of 10.4%–18.7% over the semi-analytical model, and providing valuable insights into subsurface flow mechanisms. By integrating production forecasting with insights into shale reservoir flow mechanisms, the model facilitates rapid evaluation of reservoir performance and exploration of flow dynamics, thereby laying the foundation for optimizing production strategies and advancing the study of mass transfer mechanisms in shale reservoirs.