Abstract: In view of the challenges caused by the large-scale access of electric vehicles (EV) and wind power to the system dispatching and operation, this paper proposed a regional power grid dispatching strategy based on long-term and short-term memory network (LSTM), taking into account the difference between electric vehicle demand and time-of-use electricity price. First of all, according to the difference of demand, the grid-connected EV was divided into three types: rigid EV, fast charging flexible EV, and slow charging flexible EV; and the load models were established respectively. Secondly, considering the difference of fast/slow charge flexible EV response speed and time-sharing electricity price, as well as the power characteristics of conventional generator sets and fast response units, the strategy was divided into three stages: day-ahead, model training, and intra-day. In the day-ahead stage, considering the unit operation cost of the regional power grid and the fees paid by electric vehicle owners, a multi-objective optimal scheduling model was established. In the model training phase, the intra-day scheduling model was obtained through a large amount of data training for LSTM network; in the intra-day stage, the day-ahead scheduling results and intra-day ultra-short-term prediction data were input into the intra-day scheduling model to get the intra-day controllable unit scheduling plan. Finally, the effectiveness and economy of the strategy are verified by a review in the future.