Abstract: The screenout severely restricts the safety and efficiency of fracturing treatment in Jimsar Shale Oil. The change in treatment pressure is the most direct response signal of the screenout. Accurate prediction of treatment pressure is the key to achieving early warning against screenout. Based on three machine learning methods, namely LSTM, BiLSTM, and GRU, a multivariate time-series treatment pressure prediction model was established. Grid search algorithms and Bayesian optimization algorithms were used to optimize the model’s hyperparameters, while five-fold cross-validation was employed to prevent overfitting. The performance of the prediction model was evaluated using mean squared error, root mean squared error, and mean absolute error as metrics, and the model was applied to actual fracturing sections. The research results indicate that the BiLSTM model with five-fold cross-validation and dropout rate set to 0.3 outperforms the LSTM and GRU models under the same constraints, with mean squared error reductions of 67.6% and 89.9%, mean squared root error reductions of 43.1% and 68.3%, and mean absolute error reductions of 28.6% and 67.6%, respectively. Therefore, the BiLSTM model with five-fold cross-validation and dropout rate set to 0.3 demonstrates superior generalization capability and robustness, making it more reliable for predicting treatment pressure. The research findings provide a model and method for the prediction of treatment pressure in Jimsar Shale Oil fracturing treatments.