Small disturbance signals similar to noise always exist in the daily operation of power systems, and their use for load parameter identification can solve a problem of time-varying and distributed loads that cannot be handled by traditional overall measurement and identification methods. This paper proposes a parameter identification method of a "Z+IM" comprehensive load model based on a small perturbation signal measured by PMU. This method adopts the rolling identification framework of PMU measurement data, and divides rolling identification into two steps: data processing and load parameter identification. First, from the characteristics of small disturbance signals measured by PMUs, relatively high quality small disturbance data sets of PMUs are obtained through the steps of initial screening, preprocessing, coarse screening of identifiable sets and denoising. Then, the time-varying, electromagnetic and mechanical parameters of the load are identified by a two-stage identification strategy based on the idea of prediction error. The load parameters obtained by the proposed method can be directly applied to domestic mainstream simulation programs such as PSASP and BPA without conversion. This has practical engineering application value. Finally, the effectiveness and robustness of the proposed method are verified by the simulation of a 3-machine 9-bus system and the actual grid in Hunan province.