Sepsis is a highly lethal disease in intensive care units, and patient indicators are constantly changing, making accurate prediction of patient mortality crucial for doctors to develop appropriate treatment plans. While machine learning and deep learning have been applied to sepsis research, model generalization performance can suffer from underfitting or gradient issues. To address these challenges, we propose using a deep residual network and a deep residual network incorporating Bayesian data prior to predict patient mortality using the MIMIC-III dataset. The accuracy of the two models on the validation set reached 0.9392 and 0.9329, respectively.