Accurate prediction of shield machine attitude is essential for safe and efficient tunnel construction. This study proposes a novel approach termed the Downtime-Excavation Attitude Model (DEAM), which integrates a downtime attitude model, an excavation model, and gated attention to capture key features under different operating states. The method was validated using data collected from the super-large-diameter shield tunnel project in Haizhu Bay, Guangzhou. For pitch angle attitude (PAA) prediction, DEAM achieved an R2 value of 0.92, with a root mean square error (RMSE) of 0.69° and a mean absolute error (MAE) of 0.39°, outperforming existing approaches. Ablation experiments showed that geological parameters are critical, while SHAP-based feature importance analysis identified chamber pressure, hydraulic cylinder parameters, burial depth, and return slurry flow rate as dominant factors. These findings support robust attitude prediction and targeted optimization of shield attitude control.