The access of a large number of nonlinear loads and distributed energy sources has brought the problem of multiple composite disturbances and strong feature correlation of power quality to the new energy grid-connected system. In order to improve the accuracy of disturbance analysis of new energy power system, a power quality disturbance identification method based on improved Kaiser window S-transform and GRU neural network is proposed. Firstly, according to the IEEE standard, the mathematical model of power quality disturbance is established, and seven kinds of most commonly used single disturbance signals and sixteen kinds of composite disturbance signals are simulated. Then, according to the characteristic difference of various disturbances, the frequency range characteristics of all disturbances are determined and obtained by using the frequency interval method of S-transform. Finally, the GRU neural network is used to construct the multi-disturbance feature identification model, and the power quality disturbance identification classification verification is carried out, and the satisfactory results are obtained.