Real-Time Prediction Method of Shale Gas Fracturing Construction Pressure Based on MS-1DCNN-TFT Neural Network

To realize the accurate and real-time prediction of shale gas fracturing construction pressure, this study proposes a collaborative innovation method for its pressure signal characteristics. An architecture combining multiscale one-dimensional convolutional neural network (MS-1DCNN) and temporal fusion transformer (TFT) is constructed. The former extracts instantaneous jump to long-term trend features synchronously, and the latter models complex timing dependence. To solve the problem of model failure in long-term prediction, a mild online learning strategy combining FIFO buffer and EMA weight update is designed to realize the smooth adaptation of the model to the change of working conditions. In the actual data experiment, the static model of this method (MAE = 3.21 MPa) has been significantly better than the baseline, and online learning has further reduced its MAE to 1.98 MPa (38% increase), effectively suppressing error drift. It provides high reliability algorithm support for intelligent monitoring of shale gas fracturing and has significant application value for achieving cost reduction and efficiency increase.