小样本下基于CNN-GRU网络的弹丸落点预测  

Projectile Impact Point Prediction Based on CNN-GRU Network under Small Samples

作　　者：王现磊 陈铎 薛景元 王义江 WANG Xianlei;CHEN Duo;XUE Jingyuan;WANG Yijiang(Unit 63861 of PLA,Baicheng 137001,China)

机构地区：[1]解放军63861部队,吉林白城137001

出　　处：《火力与指挥控制》2024年第7期64-69,共6页Fire Control & Command Control

摘　　要：为充分挖掘弹丸径向速度在时空上的规律,提高弹丸落点预测精度,提出了一种基于卷积神经网络结合门控循环单元的弹丸落点预测方法。分别利用卷积神经网络(CNN)和门控循环单元(GRU)网络对弹丸的径向速度在时间和空间上的强相关性特征进行提取,学习弹丸高度复杂的非线性飞行轨迹,构建弹丸落点预测模型。通过某型炮弹径向速度数据作为训练集和测试集进行弹丸落点预测,并与MLP、LSTM和CNN-LSTM时序预测方法进行比较。实验结果表明,CNN-GRU预测模型能够有效提取径向速度序列中的时空间信息,学习出弹丸相对雷达的位置,相比其他模型具有预测精度高、收敛速度快且稳定性好的优势。In order to fully explore the law of projectile radial velocity in time and space,and improve the accuracy of projectile impact-point prediction,a method of projectile impact point prediction based on CNN-GRU is proposed.CNN and GRU networks are used respectively to extract the strong correlation characteristics of the projectile radial velocity in time and space,to learn the highly complex nonlinear flight trajectory,and to build the prediction model of projectile impact points.The radial velocity data of a certain type of projectile is used as the training set and test set to predict the impact points,and compared with the time series prediction methods of MLP,LSTM and CNN-LSTM.The experimental results show that the CNN-GRU prediction model can effectively extract the spatiotemporal information in the projectile radial velocity sequence,and learn the position of the projectile relative to the radar.The comparisons with other models show that the predication model has higher prediction accuracy,faster convergence speed and better stability.

关 键 词：径向速度 落点预测 卷积神经网络 门控循环单元 

分 类 号：TJ06[兵器科学与技术—兵器发射理论与技术]