Addressing the challenge of predicting beyond-visual-range air confrontation threats in air situational awareness, a two-stage threat prediction method for large-scale beyond-visual-range air confrontation was proposed. Firstly, a Vector Autoregressive Model-based algorithm for short-term flight path prediction was introduced. This algorithm utilizes real-time radar data to predict the multi-variable flight state of both enemy and allied aircraft. Secondly, a general air confrontation threat area simulation model with minimal decision variables was introduced, integrating Multilayer Perceptron Deep Neural Networks to model the beyond-visual-range air confrontation threat area. Our proposed method effectively addresses the real-time threat range of organized enemy aircraft and provides targeted threat warnings for our aircraft. Experimental results demonstrate that the prediction error of the short-term track prediction algorithm based on VAR is significantly smaller than results obtained from multiple deep learning methods. Furthermore, the computational efficiency of the beyond-visual-range air confrontation threat region, based on the MLP-DNN method, is approximately 8000 times greater of traditional threat region simulation.