Prediction of Low-Permeability Reservoirs Performances Using Long and Short-Term Memory Machine Learning

摘要

In order to overcome the typical limitations of numerical simulation methods used to estimate the production of low-permeability reservoirs,in this study,a new data-driven approach is proposed for the case of water-driven hypo-permeable reservoirs.In particular,given the bottlenecks of traditional recurrent neural networks in handling time series data,a neural network with long and short-term memory is used for such a purpose.This method can reduce the time required to solve a large number of partial differential equations.As such,it can therefore significantly improve the efficiency in predicting the needed production performances.Practical examples about water-driven hypotonic reservoirs are provided to demonstrate the correctness of the method and its ability to meet the requirements for practical reservoir applications.