With the emergence of new types of power system loads such as electric vehicles, the correlation between power sources and loads is becoming increasingly complex. At this time, probabilistic power flow calculations completed using Cholesky decomposition will be affected by the correlation matrix between power system power sources and power system loads. Specifically, as the dimension of the correlation matrix increases, the correlation matrix becomes non positive definite, ultimately leading to the inability to perform power flow calculations. Therefore, the article combines the modified Barzilai-Borwein gradient method with the Nataf transform to propose a novel semi invariant method for handling non positive definite correlation matrices. The impact of various power sources and new load correlations on the calculation results in the IEEE-33 node system is analyzed to verify the accuracy of the proposed method.