John Chen

University of Calgary

Reservoir simulation based on physics-informed spatial-temporal machine learning

Surrogate models play a vital role in reducing computational complexity and time costs for reservoir simulations. However, traditional surrogate models suffer from limitations in autonomous temporal information learning and restrictions in generalization potential, which is due to a lack of integration with physical knowledge. In response to these challenges, a Physics-Informed Spatial-Temporal Neural Network (PI-STNN) is proposed in this work, which incorporates flow theory into a loss function and uniquely integrates a Deep Convolutional Encoder-Decoder (DCED) with a Convolutional Long Short-Term Memory Network (ConvLSTM). In pursuit of evaluating the PI-STNN model's robustness and generalization capabilities, a detailed analysis is conducted, wherein its performance is contrasted with that of a purely data-driven model sharing the same neural network architecture. Furthermore, a transfer learning strategy is applied to adapt a trained PI-STNN model to fractured flow fields, exploring the impact of natural fractures on prediction accuracy.