<p>Sand screenout is a major challenge in hydraulic fracturing, leading to the blockage of artificial fractures and negatively impacting efficiency and production. To improve early detection accuracy, this paper presents a hybrid neural network model for sand screenout prediction. The model integrates convolutional neural networks (CNN), bidirectional long short-term memory (BiLSTM), and an attention mechanism (Attention) to extract spatio-temporal features from fracturing parameters. Its hyperparameters are optimized using the crested porcupine optimizer (CPO). Furthermore, an enhanced pressure–time double-logarithmic model, incorporating thresholds and inflection points, is used to calculate the slope and improve warning accuracy. Experimental results demonstrate that the proposed model outperforms four comparison models in predicting fracturing pressure, achieving lower mean absolute error (MAE) and root mean squared error (RMSE), and a coefficient of determination (<i>R</i>²) near 0.98. Field tests show the model provides warnings 76 s earlier than manual operations across six fracturing stages in a Sichuan Basin shale gas field. Performance evaluation shows a precision of 84.6%, a recall of 91.7%, and an F1 score of 88.0%. These results validate the model's effectiveness in real-world production, supporting timely interventions to mitigate sand screenout risk and assisting operators in making more informed decisions.</p>