Deep learning-based analysis of the main controlling factors of different gas-fields recovery rate

Abstract

Due to the importance of the main controlling factors for oil and gas field development, numerical simulation methods, physical experimental methods and other methods have been used to study the problem. However, it is difficult to find the main controlling factors of a certain type of gas field using these methods. Therefore, a two-fold three-network model is proposed to solve the difficulties by coupling dynamic production data and static geological engineering data in this paper. First fold is consisted of 1D convolution network and Long Short-Term Memory neural network (LSTM), can perform good feature extraction and learn long time sequence dependence for dynamic production sequence data. Second fold made of BP neural network, is mainly dealing with static geological engineering data. By combining the two folds, the model can couple dynamic production data and static geological engineering data at the same time. Finally, the Garson feature selection are used to obtain the main controlling factors of gas field recovery rate based on trained network model. The experimentally obtained trained model can fit the recovery rate of gas field well. This shows that the proposed method can effectively discover the main controlling factors for gas field for different types, which has wide application for gas development.