Multi-population genetic algorithm-based optimized response surface reliability analysis method

The Response Surface Method (RSM), a widely employed surrogate modeling technique in structural reliability analysis, often suffers from oversimplification when cross-terms are omitted. Conversely, the arbitrary inclusion of cross-terms can result in elevated computational demands and introduce distortions in the outcomes. To mitigate these limitations, this study introduces an optimized RSM framework based on a Multi-Population Genetic Algorithm (MPGA). The proposed methodology incorporates the Weighted Response Surface Method (WRSM) as a dynamic optimization structure, employing the test set prediction error rate as a constraint threshold. A composite objective function is formulated by integrating the computational cost function with the test set prediction error rate. The MPGA is utilized to determine the optimal configuration of the response surface model, with the approach validated through a computational case study. The findings confirm that the proposed method effectively identifies optimal combinations of cross-terms that maintain a balance between computational efficiency and modeling accuracy, thereby improving the precision of failure probability confidence interval estimations.