Due to the hysteresis of the shield machine, it is difficult to adjust the shield attitude in time. In order to solve the problem that the shield pitch angle is difficult to be adjusted, a combined prediction model of shield vertical attitude based on Long Short-term memory (LSTM) and Support Vector Regression (SVR) algorithm in gray theory is proposed. The original shield data were analyzed by the gray theory method to determine the gray correlation between shield tunneling parameters and stratigraphic parameters on the shield pitch angle, and the redundant parameters with the gray correlation degree lower than 0.8 were removed and subjected to wavelet transform (WT) denoising. The prediction model is compared with the combined prediction model with all parameters as input layers (AWT-LSTM-SVR) and two single prediction models, and its accuracy is verified by the example of the water transmission tunnel of the Pearl River Delta Water Resources Allocation Project. The results show that the combined WT-LSTM-SVR model has the best prediction effect and the WT-LSTM model has the worst prediction effect; on the other hand, the prediction accuracy of the WT-LSTM-SVR model is greater than that of the AWT-LSTM-SVR model, and the average absolute error percentage of the combined WT-LSTM-SVR prediction model is 1.28%, with an error reduction of 2.9% The proposed combined WT-LSTM-SVR prediction model can effectively combine two single models, and the gray theory analysis can improve the prediction accuracy of the model, which further strengthens the generalization ability and accuracy of the combined model and provides some reference value for other shield vertical attitude adjustment in advance.