The artificial neural network model using PSO-BP and GA-BP hybrid algorithms predicts the residual stress of GH3625 superalloy formed by selective laser melting. A sample set was generated for the experimental design through the response surface method, with laser power, scanning speed and scanning spacing as the input layer of the model, and residual stress as the output layer of the model for prediction and optimization. The prediction model was verified and comparatively analyzed using the correlation coefficient R2 and the average absolute relative error eAARE evaluation index. The results show that: BP, GA-BP and PSO-BP neural network models can all predict the residual stress of GH3625 superalloy under different process parameters, and the BP neural network optimized by the algorithm has higher prediction accuracy. Among them, the GA-BP neural network has the highest prediction accuracy for the residual stress of GH3625 superalloy formed by selective laser melting, and the model performance is superior. Its correlation coefficient R2 and relative average absolute error eAARE are 0.909 and 2.06% respectively.