Abstract: In order to better mine the spatiotemporal characteristics of electric vehicle (EV) charging load and improve the accuracy of charging load prediction, this paper combines clustering algorithms and deep learning algorithms to propose a multi-scenario short-term forecasting method for EV charging load. Initially, based on residents' travel habits, the intrinsic distribution characteristics of EV charging data are analyzed through clustering algorithms, and different charging scenarios are constructed in conjunction with EV user behavior. Subsequently, a correlation analysis of multi-dimensional factors affecting the load is conducted to determine the optimal input combination for the prediction model. To fully capture the temporal associations of these input features, the Transformer algorithm is enhanced with long short-term memory (LSTM), and utilized to establish a charging load prediction model for each scenario. Finally, the prediction results of each scenario are integrated to obtain the overall forecast of the charging load. Experiments with actual EV charging data from Shijiazhuang city validate the effectiveness of the proposed method and confirm its superiority under different time scales and input combinations.