Abstract: In the dynamic management of complex engineering, the traditional critical path method (CPM) is difficult to deal with the construction period risk caused by the high uncertainty of the new process due to the static time assumption and resource limitation. In this paper, a dynamic duration probability prediction model based on PERT is constructed. The uncertainty of task time is quantified by three-point estimation method. Combined with critical path identification and Monte Carlo simulation, a risk adaptive management framework is formed. Taking an underground project as an example, the expected value of the total duration of the critical path is determined by estimating the process time parameters for 59 days, and the analytical method predicts an overdue probability of 41.27%. Combined with Matlab 50, 000 Monte Carlo simulations, the expected duration of 58 days and the probability of overdue 28.9% under asymmetric distribution are revealed, and the accurate prediction ability of the model to extreme risks is verified. The research shows that PERT combined with Monte Carlo simulation can significantly reduce the prediction error of construction period. Its dynamic path drift analysis and risk contribution ranking provide scientific basis for resource optimization and theoretical and technical support for risk control of complex projects.