New energy generation exhibits intermittent /volatility, and the theoretical study of power system planning, operation and control based on uncertainties using point estimation method is the current research hotspot. This paper focuses on the three-point estimaten method, which is commonly used in the calculation of probabilistic load flow in power systems in recent years, to carry out an improved study, and proposes an improved three-point estimate method for probabilistic load flow calculation. In order to overcome the problem of insufficient tail accuracy of the three-point estimation method, the method is calculated by adding a combination of normally distributed tail points and combining with the three-point estimation method, and the correlation input variables are processed by Nataf transformation and Cholesky decomposition. The Cornish-Fisher series is fitted to obtain the probability distribution function. The IEEE14 node and IEEE33 node systems were tested, and the results verify the correctness of the improved three-point estimate probabilistic load flow algorithm.