Stratified yield prediction in fractured wells based on a multitasking bidirectional recurrent network

Abstract

Prediction of fluid production from hydraulically fractured wells is often difficult due to the incomplete understanding of the production mechanism and the limited availability of data. In this paper, we propose a prediction model (MTL-Bi-LSTM-CNN) to predict production in all layers of fractured wells based on convolutional neural network (CNN), bidirectional long and short-term memory network (Bi-LSTM) and multi-task learning (MTL). The model integrates the fracturing parameters with data from various layers of each well. It harnesses both forward and backward bi-directional information from multiple layered fractured wells to capture the intrinsic dependencies of data features, enhancing the accuracy of fluid production prediction. The results demonstrate that the prediction of stratified data using the network model proposed in this paper has higher accuracy and smaller error than single-well unstratified prediction. Consequently, this approach significantly improves the accuracy, robustness, and versatility of predicting fluid production and moisture content in fractured wells.