Since the probabilistic pwer flow is vulnerable to the scale of the power grid

the conventional probabilistic power flow algorithm based on cumulant method is improved. When the amount of random variables is small

the computational complexity is reduced by calculating the partial sensitivity matrix. When the amount of random variables is big

the computational scale of power grid is reduced by the network equivalent and decomposition methods. These modification can improve the computational speed of cumulant method of the probabilistic power flow. The accuracy and practicality are both testified by the IEEE systems and several actual large-scale power grid cases.