Abstract: In order to improve the accuracy and efficiency of the probabilistic power flow algorithm. The paper proposes a method combining the cumulant method with the improved Latin hypercube sampling technique. Firstly,the improved non-parametric kernel density estimation algorithm is proposed,and the probabilistic model of photovoltaic power is established by the method. Secondly,in order to improve the calculation accuracy and efficiency of each order cumulants,according to the distribution of the input random variables,different methods are used to calculate the cumulants of the random variables,and combined with the Gram-Charlier series expansion method obtains the probability distribution of state variables and branch flows. Finally,the proposed probabilistic power flow algorithm is compared with Monte Carlo simulation method,and the accuracy,rapidity and effectiveness of the proposed method are verified by the IEEE-34 node system.