With the increasingly complex power grid structure, the operation scheduling becomes more difficult, and the risk of large blackout accidents is increasing. Therefore, it is particularly important to be able to timely and effectively observe the security situation of the large power grid. In the extraction stage of situational factors, the security situation evaluation system of large power grid is constructed from two aspects of internal and external factors. The external factors are obtained through statistical analysis of the major power failure accidents of the national power grid from 1981 to 2015. In the stage of situation assessment, the weights of each index are obtained by analytic hierarchy process and the improved entropy weight method, the weighted average is obtained from the safety situation assessment value of the large power grid. In the stage of situation prediction, the data processing is formed. The sample, build a deep neural network model, through the sample data training and learning, to complete the prediction of the security situation of the large grid. In order to further verify the validity of the prediction model, it is compared with BP neural network and RBF neural network to verify that the deep neural network model can effectively predict the security situation of large power grids, and the prediction accuracy is higher than the traditional neural network model.