Cumulative PSO-Kriging model for slope reliability analysis

Abstract

The particle swarm optimization (PSO) algorithm is introduced in the Kriging modeling process to overcome the limits of pattern search method's single-point search scheme as well as its heavy dependence on the initial guess solution when obtaining the optimal correlation parameters. PSO-Kriging is proved to give better simulation by interpolating and extrapolating the unobserved points. In reliability analysis, cumulative formation of repeatedly using sampling points in previous iterations is introduced into PSO-Kriging and the classic response surface method (RSM). Cumulative formation can take full advantage of available sampling information and avoid reciprocating oscillation in the iterative process. One explicit nonlinear limit state function example demonstrated that cumulative scheme can make both PSO-Kriging and RSM much more effective, no matter latin hypercube sampling (LHS) or iteratively interpolating sampling (IIS) approach is utilized. Cumulative PSO-Kriging seems to be even more stable and efficient. Two slope reliability analysis examples including a practical nuclear plant breakwater's reliability analysis problem proved that the proposed cumulative PSO-Kriging is very suitable for the reliability analysis of real engineering structures.