In view of the problems such as the continuous arrival of new orders in the rocket shell processing workshop, and the frequent reconstruction of the production scheduling scheme which leads to the high cost of reconstruction time and low efficiency of the production scheduling scheme, this paper establishes the mathematical model of multi-objective constraint virtual unit production. It takes the total production time of multiple orders, the total transportation distance of transportation equipment and load balancing factors of workshop production equipments as objective functions, workshop equipment processing capacity and rocket shell processing technology as constraint conditions. Then, an improved NSGA-II algorithm is proposed to solve the mathematical model. In order to improve the NSGA-II algorithm's local search ability, both the “first improvement” strategy and the “random improvement” strategy which improve the local search ability of the algorithm are introduced in interlace operation. Finally, the improved NSGA-II algorithm is compared with the NSGA-II algorithm and the NSGA-III algorithm on the hypervolume and the spacing in a instance. As a result, the improved NSGA-II algorithm performs more efficiently on virtual cellular reconstruction of rocket shell processing.