Development of Attention-Based Proxy Models in the Prediction of Reservoir Production

Feng Zhang


In this work, a Transformer model was built based on Encoder-Decoder architecture to evaluate the performance of reservoirs integrating production data, geological data, and dynamic data. Structure changes are made in the Transformer model to utilize various data types to improve the accuracy of prediction. More data sets (i.e., simulations and field cases) will be applied to develop an accurate and robust prediction model for different applications.