Saturation and Pressure Prediction for Multi-Layer Irregular Reservoirs with Variable Well Patterns

The well pattern and boundary shape of reservoirs determine the distribution remaining oil to a large extent, especially for small-scale reservoir blocks. However, it is difficult replicate experiences from other directly predict because variety irregular shapes corresponding patterns. Meanwhile, regular can hardly suit shapes. In this paper, we propose placement method undeveloped multi-step prediction framework both saturation pressure fields any pattern. To boost physical information input characteristics, feature amplification approach based on formulae initially presented. Then, 3D convolution technology employed first time in increase spatial vertical direction input. Moreover, complete two-field prediction, concept multi-task learning adopted time, improving rationality forecast. Through loss-based ablation test, found that operation adopt will accuracy some extent. By testing manually designed real irregular-shape reservoirs, our proven be an accurate fast with its loss less than 0.01 calculation 10 s future one year.