Military aircraft state monitoring data have the characteristics of multidimensionality, nonlinearity and strong coupling, and monitoring data samples are of large capacity with high updating rate. So, existing state prediction models are difficult to meet the engineering requirements of high precision and realtime performance. How to predict military aircraft status informations accurately and fast is a difficult problem which needs to be settled. The paper uses the deep belief learning theory and simulated annealing algorithm to present a state prediction model for military aircraft key components. The research process consists of three stages: choosing monitoring parameters of the aircraft key components under different mission profile in order to design network samples; developing a deep learning network and prediction model; validating the model with testing date. Lastly, an aircraft engine was employed to demonstrate the efficacy of the proposed approach. The predicted results were compared with those by the SVM and ELMAN. It is found that the average error between the predicted and the measured values based on the proposed approach is less than 0.04, much smaller than that of the SVM and ELMAN method, and its running time is shorter than the above two methods, which proves the feasibility and effectiveness of the proposed approach.