Prediction on Production of Oil Well with Attention-CNN-LSTM
Shaowei Pan, Jizhe Wang, Wei Zhou
Abstract
The production prediction of an oil well is of great significance for realizing the intelligent monitoring of oil wells and improving oil recovery. In order to overcome the shortcomings of existing methods in the production prediction of oil wells, this study proposes a hybrid deep learning model based on the attention mechanism by combining the convolutional neural networks and the long short-term memory neural networks (Attention-CNN-LSTM). First, Attention-CNN-LSTM extracts the fine-grained features of the input data through CNN. Second, it obtains the coarse-grained features through LSTM and realizes the optimization of the results by the different features of the input data through attention mechanism. The Attention-CNN-LSTM model is trained and tested by using the production datasets of T1 well and T2 well of an oilfield in the southern China, respectively, and the prediction results of Attention-CNN-LSTM on the test set are compared with the prediction results of back propagation neural networks (BP), support vector regression (SVR), LSTM, Attention-LSTM and CNN-LSTM on the test set. The final comparison results show that Attention-CNN-LSTM has the best prediction performance and can achieve more accurate prediction of oil production.