Due to the existence of uncertainty analysis and multiple execution models in the real construction process, this study establishes a construction scheduling model with uncertain duration under multi-objective and multi-model resource constraints on the basis of the previous single execution model and determined activity time. The improved non-dominated sequential genetic algorithm (INSGA-II) is used for solving. And the set is divided based on the activity sequential relationship to generate chromosomes, which reduces the generation of a large number of infeasible solutions. A hybrid population initialisation strategy based on elite reverse learning is introduced in the population initialisation to ensure the diversity and convergence of the algorithm. Finally, it is proved through practical cases that both the model and algorithm built in this study can effectively solve the construction scheduling model with uncertain schedule under multi-objective and multi-mode resource constraints.