Abstract: With the rapid popularization of electric vehicles, the coupling between the traffic network and power grid is further deepened, and the travel mode of the traffic network will have a significant impact on the charging load of electric vehicles. The traditional charging load simulation method relies on the individual modeling of the traffic network and electric vehicles, and has strong assumptions. A method of electric vehicle charging load scenario generation based on data-driven convolutional autoencoder and conditional generative adversarial network（CGAN） is proposed. Firstly, the convolutional autoencoder based on unsupervised learning is used to reduce the dimension of travel prediction data in traffic network and adaptively extract the feature information.Secondly, a CGAN suitable for the generation of the day-ahead current traffic network charging load scenario is designed, and the feature information extracted from the convolutional autoencoder is used to implicitly learn the conditional probability distribution of electric vehicle charging load corresponding to different travel modes of traffic network. Thus, the generation of the day-ahead electric vehicle charging load scenario is realized, which provides the support for the operation of the power grid and charging station. Finally, taking an actual city traffic network as an example, the necessity of the proposed convolutional autoencoder and the effectiveness of the CGAN are verified.