Abstract: The artificial neural network is a massively parallel distribution information processing system consisting of simple processing units. It has the ability to learn and build nonlinear complex relational models. Once trained by data set samples, it can predict the output. In recent years, it has been used in various extrapolations of nuclear physics and solved many complex problems. All the results point to the fact that the neural network approach is a very useful tool to further improve the accuracy of nuclea mass models, and can predict those unknown but critical nuclear data. Based on two representative macroscopic microscopic mass models: the finite-range droplet model (FRDM) and the Weizs-cker-Skyrme model (WS4), the results of the model were optimized using the artificial neural network approach. After adding the neural network approach, the root mean square deviations of the binding energy of the two models are improved in varying degrees. The root mean square deviation between the binding energy data given by FRDM and 2 095 experimental data decreases from 0.579 MeV to 0.354 MeV, and the root mean square deviation between the binding energy data given by WS4 and 2 095 experimental data decreases from 0.292 MeV to 0.210 MeV. The one-neutron separation energy can reflect the accuracy of the nuclear mass model to a certain extent. Thus, based on the optimized data, the oneneutron separation energy of Z=82 isotope chain was calculated. The root mean square deviations of the oneneutron separation energy of FRDM and WS4 are 39.9 keV and 40.8 keV, respectively. In addition, accurate binding energy data of superheavy nuclei is very important for studying the synthesis of superheavy nuclei. In recent years, to find the exact position of the island of stability and extend the periodic table, many theoretical models for studying the synthesizing mechanism of superheavy nuclei are being developed. To accurately reproduce the experimental data of superheavy nuclear synthesis and reasonably predict the evaporation residual cross section of unknown superheavy nuclear synthesis, all models have encountered the same problem: basic nuclear data, such as neutron separation energy, must be reliable to some extent. In this work, combined with the binding energy data calculated by the original model, the binding energy data were extrapolated, and the ability of the extrapolation data of the two models to describe the one-neutron separation energy of superheavy nuclei was compared. The root mean square deviations of the one-neutron separation energy of FRDM and WS4 in the superheavy nuclear region are 40.1 keV and 188.1 keV, respectively. The results show that the FRDM has better performance in the superheavy nuclear region than the WS4. In order to synthesize new elements, it is theoretically necessary to study the synthesis mechanism of new elements Z=119 and Z=120. Thus, the oneneutron separation energy of the new elements Z=119 and Z=120 isotope chains were predicted. The predicted results show that the one-neutron separation energy of Z=119 and Z=120 isotope chains has an obvious odd-even effect, and the difference of one-neutron separation energy given by the two optimization models is about 300 keV.