Abstract: For the demand of real-time accurate mapping of digital twin active power grid,a multidimensional prediction index system is constructed,and a multidimensional state prediction algorithm for power grid key indicators based on LSTM is established to realize the change prediction of key indicator attributes of digital twin active power grid.Firstly,the prediction model of multidimensional system is established for the data collection of 4 key indicators such as load characteristics,which can effectively control the state of the active power grid;secondly,the feature learning fitting of multidiniensional data based on LSTM prediction algorithm is proposed,and the prediction of the next phase of multidimensional data is realized to map to the power digital twin to realize the synchronization and intelligent regulation of smart energy system operation planning and implementation;finally,the A simulation test model is established and arithmetic examples show that the deep learning-based digital twin power grid multidimensional situation sense prediction method can better predict and discriminate the power grid situation,so as to realize the accurate planning of future energy system to provide decision guarantee.