A novel hybrid PIPSO–WSVR method for structural reliability analysis

A novel hybrid PSO–SVR method is proposed that combines support vector regression (SVR) with the particle swarm optimization (PSO) algorithm to improve the computational efficiency of predicting structural failure. To express the influence of each parameter on the results of the approximate model, weights are introduced into the kernel function, and the method is called weighted SVR (WSVR). To select the optimal SVR parameters to improve the fitting results of the WSVR response surface model, the PSO algorithm is used. To improve the PSO's convergence rate and its ability of jump out of local extrema, a parasitic immune particle swarm optimization (PIPSO) algorithm is proposed. The basic idea of PIPSO is that an exploration strategy and a high frequency of immune system mutations are used for the particles of the host population to expand the search space of the algorithm and suppress premature convergence. Results from computational tests show that PIPSO has a faster convergence speed and a better search accuracy than PSO, linearly decreasing inertia weight PSO (LDIWPSO), and e2-PSO. More importantly, the PIPSO–WSVR method has a higher accuracy and efficiency than the SPSO–SVR, LDIWPSO–SVR, and e2-PSO–SVR methods with the same sample size. Therefore, the proposed PIPSO-WSVR shows promising results for practical structural reliability analysis.