Reliability analysis considering spatial variability by combining spectral representation method and support vector machine

Spatial variability is an essential characteristic of soil properties, particularly in performing reliability analysis of tunnels, as these properties exhibit marked differences depending on the soil surrounding. This set of parameters can be a multidimensional-multivariate random field with various distributions. However, the disadvantage of spectral representation method (SRM) is that it is less accurate when simulating cross-correlation of small samples. Thus, a new approach that combines SRM and support vector machine (SVM) is proposed to accurately describe uncertain soil parameters. The proposed method considers both the auto-correlation and the cross-correlation of a multidimensional-multivariate random field under a small size of samples. In addition, SVM can facilitate the effective identification of small samples. Numerical cases demonstrate that spatial variability significantly influences the reliability of tunnels, and that ignoring the spatial variability of soil properties overestimates the probability of failure. These findings indicate that the combined method is an effective approach for random field simulation and reliability analysis in tunnel engineering.