Abstract: A multi-layer perceptron neural network prediction model between the process parameters of CO₂ gas shielded welding and the weld geometry (melting width and depth) was established, and the mathematical analytic formula of the model was determined based on the welding test data training the model. The virtual simulation model of weld morphology was established by analyzing the characteristics of weld section and surface morphology. The weld morphology prediction and virtualization simulation system was developed by python programming. The results show that the maximum deviation for predicting melting width with the established multi-layer perceptron neural network prediction model was 0.097 mm with a model fitting coefficient of 0.999 269, and that for predicting melting depth was 0.051 mm with a model fitting coefficient of 0.999 567. The mathematical model of weld section morphology with melting depth and melting width as input variables and the mathematical model of surface morphology with melting width as input variables were established.