CNN-LSTM Model Optimized by Bayesian Optimization for Predicting Single-Well Production in Water Flooding Reservoir

Abstract

Geared toward the problems of predicting the unsteadily changing single oil well production in water flooding reservoir, a machine learning model based on CNN (convolutional neural network) and LSTM (long short-term memory) is established which realizes precise predictions of monthly single-well production. This study is considering more than 60 dynamic and static factors that affect the changes of oil well production, introduce water injection parameters into data set, select 11 main control factors, and then, build a CNN-LSTM model optimized by Bayesian optimization. The effectiveness of the proposed model is verified in a realistic reservoir. The experiment results show that the prediction accuracy of the proposed model is over 90%, which suggests a penitential application in an extensive range of applications. Production forecasting by the developed model is simple, efficient, and accurate, which can provide a guidance for the dynamic analysis of a water flooding reservoir, and work as a good reference of the development and production of other types of reservoirs.