Prediction of coalbed methane well productivity based on attention mechanism of CNN-GRU

Abstract

In order to optimize the current prediction method of coalbed methane production and solve the problem of losing key information due to the influence of historical state of coalbed methane production data and too long sequence, a coalbed methane production prediction model based on attention mechanism of convolutional neural network was proposed. Firstly, the five drainage parameters with high correlation with daily gas production are selected, and the CBM productivity prediction model is constructed and trained based on the field drainage data of the well in Hancheng Block. Finally, the production capacity of the well area in the next 160 days is predicted by the model. The research results show that combined model CNN-GRUA with attention mechanism overcomes the problems that traditional prediction methods cannot deal with, including nonlinearity between data, time sequence and information loss, by extracting high-level features of data and learning the correlation of time series. Compared with BP neural network, convolutional neural network (CNN), gated recurrent unit(GRU) and CNN-GRU without attention mechanism model, the CNN-GRUA combined model has higher prediction accuracy, and the mean absolute percentage error is 1.72%.

Keywords

• coalbed methane productivity prediction
• deep learning
• convolutional neural networks
• gated recurrent units
• attention mechanism