Research on Deep Spatio-Temporal Model and Its Application in Situation Prediction

Abstract

Situation prediction refers to predicting the state information of things in the future on the basis of existing information, and situational information contains complex laws of time and space. Traditional methods only consider a single factor or separate time and space. At the same time, due to the limitations of traditional algorithms, it is not possible to accurately predict air combat events with long interval dependencies. In order to solve these problems, we propose a deep spatio-temporal model based on the dynamic graph convolution and attention mechanisms. The model extracts and analyzes the features of space and time respectively. Experimental results show that the model proposed in this paper has more stable training process and higher prediction accuracy.