针对复杂空战环境下无人机航迹预测精度不足的问题,提出基于深度学习中长短期记忆(Long Short Term Memory,LSTM)网络的一种航迹预测方法.采用Afsim兵棋推演软件构建1V2空战对抗场景,通过清洗输出样本,获得了空战无人机航迹、姿态及角速度多种特征数据集;搭建航迹预测LSTM网络模型,采用训练集对隐藏单元与时间步长敏感性分析;将测试集输入LSTM模型进行预测精度与时效性验证.LSTM网络模型综合长期记忆无人机状态信息的变化与上一时刻记忆得到预测值,解决了传统预测模型不能结合长期信息综合预测的弊端;并且采用大量学习的方式缩短了预测时间.结果表明:LSTM网络模型航迹预测绝对误差较小且时效性较强,可为空战航迹预测提供一定的参考.
Real-time track prediction of air combat UAV based on LSTM
Aiming at the problem of insufficient accuracy of UAV track prediction in complex air combat environments,a track prediction method based on deep learning Long Short Term Memory(LSTM)network is proposed.Afsim War game simulation software was used for constructing the 1V2 air combat scenario,and multiple characteristic data sets of the attitude angular velocity of air combat UAV were obtained by cleaning the output samples.The LSTM network model for track prediction is built using the training set hidden unit and time step sensitivity analysis.The test set was input into the LSTM model to verify the prediction accuracy and timeliness.The LSTM network model integrateds long-term memory of UAV state information changes and the memory of the last moment to obtain the predicted value,which solves the problem that the traditional prediction model can not combine long-term information integrated prediction.Also,the prediction time is shortened by using a lot of learning.The results show that the LSTM network model has small absolute errors and strong timeliness,which can provide a certain reference for air combat track prediction.
deep learningneural networklong short-term memory networktrack predictionWar gameair combat UAV
NETL
NSTL
万方数据
