Abstract: Operational aggregation behavior prediction has encountered the challenges of large feature space, dynamic changes of related combat units and large behavior noise, etc. To address these issues, a operational aggregation behavior prediction method based on a 3D convolution neural network is proposed. In this method, a three-dimension convolution neural network is constructed by introducing the time dimension into the two-dimension convolution so as to recognize the operational aggregation behaviors. After that, a reconfigurable hierarchical long short-term memory (LSTM) network is adopted to analyze the temporal aggregation behavior data of related combat units, with which the key factors of aggregation behaviors such as time, location could be calculated. Experiment results suggest that the proposed method could predict operational aggregation behaviors accurately. Meanwhile, the method will perform much better when introducing the man-in-the-loop mechanism.

Key words: operational aggregation behavior, 3D convolution neural network, spatio-temporal feature, situation comprehension