Over the past fifty years, the design point based methods such as first and second order reliability method (FORM and SORM) for structural reliability analysis and optimization design obtained full development because of their simplicity and effectiveness.But this kind of moment method is approximate, and usually it is not applicable to the complex problems with disjoint failure domains and multiple design points [1-7].The method of numerical simulation such as Monte Carlo simulation, line sampling and subset simulation can guarantee enough accuracy.However, it is seldom applied in engineering practice due to the expensive computational cost.Moreover, the simulation based method is difficult to conduct the reliability based design optimization, since it produces the inconsistent reliability results derived from the stochastic convergence.Currently, it is urgent to look for a method which possesses enough accuracy and high efficiency for uncertainty propagation analysis of structures, especially subjected to random dynamic excitations.