The invention discloses a fast probabilistic power flow calculation method of an improved extreme learning machine considering power flow characteristics. The method mainly comprises the steps of 1) obtaining basic data of a power network; 2) establishing a mapping relationship f: Pi, Qi to Ui, [theta]i based on the basic data of the power network; 3) decomposing the mapping relationship, and establishing an extreme learning machine neural network; 4) optimizing hidden layer parameters of the extreme learning machine neural network so as to establish an improved extreme learning machine neuralnetwork; 5) inputting the basic data of the power network into the improved extreme learning machine neural network, and calculating to obtain the probabilistic power flow of the power network. According to the invention, a time-consuming solving process for a large-scale high-dimensional complex nonlinear power flow equation in PPF calculation of a high-precision simulation method is replaced, so that engineering actual requirements of PPF calculation on precision and speed are considered.