Abstract: Nuclear mass plays an important role on nuclear structure and nuclear astrophysics. For the exotic nuclei far away from the stability valley, at least three orders of the magnitude for the half lives of emitters might be arisen by hundreds of keV difference of the separation energy given by the nuclear mass. At present, ^149\rmLu is the proton emitter with shortest lives discovered in experiments. The separation energies of the neighbor undetected nuclei will influence the nucleosynthesis path. To this end, we apply Artificial Neural Network (ANN) model and Bayesian Neural Network (BNN) model to constrain the proton separation energy, and the practice indicates that the two approaches have the overfitting and precision problems respectively. In order to improve the precision and mitigate the overfitting problem, based on BNN model, and take the uncertainties of the labels and the ensemble uncertainty of the models into account simultaneously, we develop the BNN-Beihang (BH) model. The uncertainties of ^148\rmLu and ^160-161\rmRe predicted by the model decreased to less than 100 keV for the first time. These results will effectively improve the precision of the theoretical predictions to the Lu and Re chain proton emitters' lives.