Abstract:In regard to the contraction problem in the process of selective laser sintering, orthogonal experiment and measuring methods were applied to acquire the training samples, then back propagation neural network or support vector regression optimized based on genetic algorithm (GA-SVR) was used to establish a quantifiable model between technical parameters and contraction rate, and to go much further to make parameter optimization for the quantifiable model by using self-adaptive variant particle swarm optimization. Based on same training sample, the result shows that GA-SVR perform forecasts better than does back propagation neural network, thereby by using particle swarm optimization to optimize parameter we get a best parameter combination by preheating temperature 85 ℃, laser power 19.8 W, scanning velocity 2590 mm/s, powder coating thickness 0.1 mm, wall thickness 1 mm. The model can be more accurate to control the contraction phenomena during practical manufacturing, and provides a new idea for optimizing control in the process of selective laser sintering.