Abstract: With the objective of minimizing the total construction duration and the total cost, a multi-objective optimization model of railway projects is established based on RSM method and an improved NSGA-II algorithm is proposed for the railway construction scenarios including linear, strip, and block activities, where linear activities can be reverse constructed and the construction rates can be variable. The algorithm designs a uniform evolutionary elite selection strategy for hierarchical selection of individuals to improve the population diversity and convergence. Also, the mutation and crossover operators of differential evolution algorithm are introduced to construct hierarchical multi-strategy adaptive mutation and crossover operators to balance the local and global search abilities of the entire population. Results show that the consideration of special activities and construction directions can enhance the applicability of the model to railway projects; the modified algorithm has fast convergence speed, stable operation, and better results, which can satisfy the optimization of large-scale railway project schedules.

Key words: repetitive project scheduling, NSGA-II (non-dominated sorting genetic algorithm-II), duration-cost, multi-objective optimization