In general, it is inappropriate to directly apply the traditional univariate approach on power analysis to omics datasets which are characterized by high dimensions with 100s to 1000s of features and a relative small number of samples.
The power analysis in MetaboAnalyst is based on the R package SSPA that was originally developed for power calculation for gene expression datasets. For a given pilot data, the method estimates the average power for a given false discovery rate. For more details, please refer to the original publication by M van Iterson et al