Abstract: In order to improve the molding accuracy of (double-side incremental sheet forming, DSIF) sheet, the square cone-box parts were used as test parts, the process parameters such as tool diameter, layer spacing, forming Angle, plate thickness and forming depth were used as influencing factors, and minimized the bottom rebound value and side wall bulge value were used as optimization objectives to design orthogonal tests. The test results were calculated by Abaqus finite element simulation. Through the establishment of multi-input and multi-output BP(back propagation) neural network prediction model, combined with the non-dominated sorting genetic algorithm (NAGA-Ⅱ) with elite strategy, multi-objective optimization of DSIF process parameters was carried out. Based on the entropy weight approaching ideal solution sorting method (technique for order preference by similarity to ideal solution, TOPSIS), a set of optimal process parameter combinations were determined from Pareto solution set to improve the accuracy of optimization results. The optimal combination of process parameters obtained by optimization and screening were tested. The results show that the measured springback value of the bottom part is 0.693 mm, the bulge value of the side wall is 0.934 mm, and the errors of the selected target values are 6.31% and 2.09%, respectively. It can be seen that the established multi-objective optimization process is feasible and provides an effective optimization scheme for the springback reduction of DSIF parts.