Abstract: With the rapid development of electric vehicles, the huge random load brought by electric vehicles has brought great challenges to the safe and stable operation of distribution networks. The electric vehicle load data are nonlinear, and the characteristics are not obvious, so a fusion model electric vehicle load forecasting method based on similar cycle and mode decomposition is proposed to solve the above problems. Firstly, the similar weeks and features screened by the Pearson correlation coefficient and dynamic time warping (DTW) were used to compose the similar weekly load sequence. Then, a complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is used to decompose the similar weekly load sequences into high-frequency and low-frequency components, and the high-frequency components are input into the convolutional neural network (CNN)-gate recurrent unit (GRU). The low-frequency components are fed into the kernel extreme learning machine (KELM) model, and the hyperparameters of the network model are optimized using an improved sparrow search algorithm. Finally, the forecast results of different components are summed to output the final load forecast sequence. Experimental results show that the proposed method has a faster prediction speed and higher prediction accuracy.