Abstract: Gas metal arc welding（GMAW）, which is widely utilized in the industrial manufacturing field, has a crucial impact on the performance of fillet weld joints through its geometrical morphological parameters. Using orthogonal experimental design method, the data related to GMAW process parameters and the geometrical morphological parameters of fillet weld joints were collected to train and test a Backpropagation（BP） neural network. Additionally, the particle swarm optimization（PSO） was incorporated to construct a PSO-BP neural network predictive model that can describe the relationship between GMAW process parameters and the geometrical morphological parameters of fillet weld joints. The results show that the model exhibits high precision in predicting the geometrical morphological parameters of GMAW fillet weld joints. In terms of the prediction of joint geometrical morphological parameters, after optimization by the PSO algorithm, the PSO-BP model has significant reductions in mean absolute percentage error（MAPE） compared to the original BP model. In conclusion, the PSO-BP model has remarkable advantages in predicting the geometrical morphological parameters of GMAW fillet weld joints and can provide a solid theoretical basis for the automation and intelligentization of the welding process.