Sand Screenout Early Warning Models Based on Combinatorial Neural Network and Physical Models

Sand screenout is a critical challenge in hydraulic fracturing, affecting both the construction process and operational safety. This paper proposes a sand screenout warning model that integrates a combinatorial neural network and physical approaches to enhance both the speed and accuracy of sand screenout warnings. Firstly, the combined neural network uses a Transformer to capture key features during fracturing construction from historical data, and the extracted features are input to the Gated Recurrent Unit (GRU) for temporal prediction and the Crested Porcupine Optimizer (CPO) to further optimise the GRU-Transformer hyperparameters of the model. Additionally, the physical model improves the conventional inverse slope method by incorporating a threshold and sliding module, which enhances slope calculation and warning accuracy. The results showed that for fracturing pressure prediction, the proposed CPO-GRU-Transformer model obtained an RMSE value of 0.842 MPa, MAE of 0.613 Mpa, and $R^2$ of 0.971, a smaller RMSE and MAE and a larger $R^2$ than the three pressure prediction models, namely LSTM, GRU, and CPO-GRU. The proposed sand screenout warning model has been applied in the field construction of the U shale gas area in the Sichuan Basin. The warning points of the model proposed in this study were advanced by 73.5 s on average compared with the manual warning points in the three validated fracturing segments, with a successful warning rate of 85.71%, which greatly avoids the possibility of sand screenout and provides a method of fast calculation speed and high prediction accuracy, providing an early warning of sand screenout.