1.School of Electric and Electronic Engineering, North China Electric Power University, Beijing 102206, China;2.State Grid Economic and Technological Research Institute Co., Ltd., Beijing 102209, China;3.State Grid Corporation of China, Beijing 100031, China;4.Economic and Technical Research Institute of State Grid Zhejiang Electric Power Co., Ltd., Hangzhou 310007, China

Electric vehicle (EV) is the flexible load that can provide flexibility to the power system. Most of the existing studies modeling the flexibility of EVs only consider the uncertainty of charging behavior and the impact of time-of-use tariffs. The deviation between the day-ahead tariff and real-time tariff is ignored, and the modeling of real-time tariff and charging load multi-timescale time-series characteristics is neglected. Aiming at this problem, this paper summarizes the manifestations and influencing factors of the flexibility of EVs, and proposes a probabilistic modeling method of the flexibility of EVs based on the temporal attention mechanism by considering the uncertainty of tariff-oriented response and the uncertainty of charging behavior. The different timescale weights are extracted by the time-series attention mechanism. A multi-timescale feature extraction network based on the temporal convolutional network is designed to learn the uncertainty of charging behavior and electricity price, and extract multi-timescale flexibility fluctuation features. The cases show that the proposed model can effectively learn charging behavior uncertainty and tariff-oriented response uncertainty, and its probabilistic modeling effect has higher reliability and accuracy.