Abstract: The establishment of an accurate constitutive model was the basis for further research on metal powder forming. Based on the modified Drucker-Prager Cap model and a BP neural network optimized by genetic algorithm, the numerical simulation of aluminum alloy powder forming process was carried out, and the constitutive model parameters were predicted. For the research of aluminum alloy powder forming, firstly, the range of the Drucker-Prager Cap model parameters were determined, and the numerical simulation of the forming process was realized based on the finite element analysis platform of ABAQUS and its subroutine USDFLD. Then, a BP neural network model optimized by genetic algorithm was established by taking the suppression force data of numerical simulation as input and the Drucker-Prager Cap model parameters as output. Finally, the constitutive model parameters of the powder die compaction tests were predicted by inversion. The results show that the average absolute percentage error（MAPE） between the numerical simulation results and the die compaction tests data after parameters inversion is only 5.10%. The BP neural network model can predict the parameters of the constitutive model quickly, effectively and accurately.