Shale Gas Production Prediction Based on the TCN-FFT Model Using Multiwell Production Data

Abstract

Abstract. Accurate prediction of shale gas production is crucial for optimizing shale gas well production strategies. Current research utilizing single-well historical data necessitates extensive long-term production datasets as training inputs to achieve accurate predictive outcomes. In this regard, this article proposes a temporal convolutional network–fast Fourier transform (TCN-FFT) model with an attention mechanism to predict well production only using short-term wellhead pressure data. The model combines TCN and FFT modules to extract local and global features, respectively. This allows the model to learn flow patterns from the production data of other wells and improves prediction accuracy for a well with shorter production histories. Experimental results demonstrate that, for wells with short production histories, the proposed method outperforms single-well-based prediction models in terms of both prediction accuracy and trend capture. Compared with the worst-performing single-well model, it achieves improvements of approximately 77%, 84%, and 65% in root mean squared error (RMSE), mean absolute error (MAE), and symmetric mean absolute percentage error (SMAPE), respectively, highlighting its practical value for engineering applications.