| Summary: | This study combined a neural network and Latin hypercube sampling (LHS) to calibrate soil parameters. The Monte Carlo parameters were calibrated by generating different numbers of training samples for pressuremeter tests and excavations. The results showed that when the number of samples was 25 or 50, the parameter calibration accuracy was very high. However, the improvement in accuracy did not increase significantly with a further increase in the number of samples, but tended to be stable. The number of training samples was set at 50 to strike a balance between the calibration accuracy and efficiency for four parameters. For 25 groups of samples, the calibration results using LHS were better than those using orthogonal sampling. Compared to stochastic optimization algorithms, a neural network combined with LHS could significantly reduce the calibration time. This method was applied to actual foundation pit engineering in China. The results showed that using the proposed calibration method clearly improved the accuracy when predicting the deformation induced by the excavation.
|
|---|