In order to reduce the impact of the uncertainty of photovoltaic output prediction on the economy and safety of cross-provincial optimal operation

a two-level optimal operation strategy is proposed

which takes into account the uncertainty of photovoltaic prediction. First

the long short-term memory（LSTM） neural network is used as the photovoltaic output prediction model. By using the k-means method

the output fluctuation of several photovoltaic power stations is clustered

and the clustered photovoltaic output data set is brought into the LSTM model to generate large-scale photovoltaic output prediction results of day-ahead and intra-day considering spatial correlation. Secondly

based on the day-ahead photovoltaic output prediction curve and load demand curve

a day-ahead dispatching model combining thermal power and pumped storage is established. The daily scheduling results are taken as input variables to carry out probabilistic power flow calculation to complete the first level scheduling. Based on the NSGA-II algorithm

the optimal power flow of multi-objective AC/DC hybrid system is calculated to complete the two-level scheduling. Finally

the actual data of two provinces and the modified IEEE 39 node are taken as the system example

and the two-level optimal scheduling plan under the environment of uncertain photovoltaic output is analyzed. The results of the calculation example show that the proposed scheduling strategy can coordinate and optimize the schedulable resources of the two provinces

reduce the risk of network operation overrun

and improve the economy and stability of the power system.