The existing Probabilistic Load Flow Calculation based on Point Estimate Method (PLF-PEM) does not consider the case where the correlation matrix of random input variables is non-positive definite. To overcome the deficiency and describe the statistical characteristics of the outputs more accurately, a PLF-PEM computing method based on Principal Component Analysis (PCA) and Cornish-Fisher series expansion is suggested. The proposed method uses the PCA theory to deal with the correlated random inputs and applies the PEM to obtain the moments of the outputs. Then, based on the cumulant theory and the Cornish-Fisher expansion and by using the discrete sample data of the input variables, the digital characteristics and probability statistical properties of the outputs are obtained. Test examples show that the method is appropriate for fast PLF calculation of power systems with high penetrations of new energy power generations. This work is supported by Natural Science Foundation of Guangdong Province (No. 2014A030313509).