Abstract: With the rapid development of electric vehicles, their charging frequency and daily charging volume have increased dramatically, which has a great impact on the stable operation of the power grid, thus it is important to study the charging load prediction for EVs. However, due to the privacy of user's charging behavior data, the machine learning prediction models constructed in the current study lack the consideration of this important factor, resulting in low prediction accuracy. To address this problem, this paper proposes a federated learning (FL) method for short-term EV charging load prediction that considers the user's charging behavior and privacy protection, namely, considering the user's charging start and end times, battery charge status, battery capacity, and user's selected charging power. The method is based on a locally trained and centrally aggregated model. The model training mechanism of local training and central aggregation is used to achieve collaborative short-term EV charging load prediction while ensuring the privacy of user data. Finally, the proposed method is validated by using charging load data from several operators in a city. The results show that the proposed method can be adopted to effectively improve the accuracy of EV short-term charging load prediction while ensuring the security of user privacy data, and has good model generalization capability.