Combining Machine Learning and Reservoir Simulation for Robust Optimization of Completion Design and Well Location of Unconventional Wells

Predictive modeling has played a crucial role in optimizing unconventional development plans, but existing models either oversimplify reservoir physics or are too computationally intensive for large-scale optimization. This study introduces a novel approach that combines reservoir physics with machine learning to create a more efficient and reliable predictive model. By integrating physics-based constraints with data-driven insights, the model ensures that predictions remain realistic while leveraging diverse data sources. Additionally, a probabilistic approach is used to generate P10-P50-P90 production curves, providing a comprehensive understanding of production uncertainty.

The model is trained and tested on data from over 1,800 unconventional wells, incorporating key completion design parameters such as lateral length, proppant concentration, and well spacing. The machine learning component enhances the model's adaptability, allowing it to process complex datasets that would be difficult to integrate into traditional physics-based models. The results demonstrate strong predictive accuracy, with correlation coefficients exceeding 0.75 in the test set. Sensitivity analysis reveals that injection fluid volume, lateral length, and proppant concentration are among the most influential factors in production forecasting for the analyzed field.

This approach addresses a key challenge faced by operators in unconventional reservoirs—maximizing both production and profitability. By enabling rapid and accurate predictions of well performance, the model helps optimize well placement and completion designs, ultimately improving estimated ultimate recovery (EUR) and net present value. The fusion of machine learning and reservoir physics offers a powerful tool for decision-making in unconventional development, ensuring that operators can efficiently evaluate thousands of scenarios and generate optimal strategies for field development.