The application of the Latin hypercube sampling method in the load flow calculation causes some new problems with the large-scale distributed generators connected to the distribution network. One is that the cumulative distribution function of distributed generators cannot be acquired easily; the other is that the correlation coefficient matrix is not positive-definite in correlation control. To solve these two problems

we proposed a new algorithm. The algorithm is based on the improved Latin hypercube sampling with modified correlation matrix. The algorithm stratifies sampling according to the discrete data and modifies the correlation coefficient matrix by positive definite spectral decomposition after which Cholesky decomposition can be applied to all recovered matrix. The speed of the algorithm is fast and the error can be less than 10-4. The accuracy and effectiveness of the proposed algorithm were proven by the comparative tests in the IEEE 33-bus system and PG&E 69-bus system. The simulation results show that this method is faster

and the accuracy and convergence of the sampled and output variables are better than those of Monte Carlo.