Finite element modeling and simulation are considered one of the effective methods for optimizing additive manufacturing process windows, which can better capture metallurgical features under the interaction of multiple physical fields. However, correctly selecting or obtaining heat source parameters, boundary conditions, and thermophysical parameters is the biggest challenge of finite element modeling. In this work, the laser semi-ellipsoid heat source parameters and powder thermophysical parameters of the laser powder bed fusion process was effectively determined by machine learning method, and an accurate laser powder bed fusion process model of IN738 alloy was established. The simulation error of the size of single-track molten pool under different process conditions is less than 2.0%. The simulation results of stress field by multi-layer and multi-track under different process conditions are consistent with the micro crack and density test results of deposited samples. The relevant research results provide a new idea for correctly establishing the model of additive manufacturing process and realizing reliable finite element simulation.