基于神经网络的弹道修正弹落点预测方法  

Study on Projectiles Impact Point Prediction Method for Trajectory Correction Based on Neural Network

作　　者：曹红锦[1] 常思江[2] 

机构地区：[1]中国兵器工业第五九研究所,重庆400039 [2]南京理工大学动力工程学院,南京210094

出　　处：《四川兵工学报》2015年第1期17-20,27,共5页Journal of Sichuan Ordnance

摘　　要：为了分析神经网络运用于弹道预测的可行性,构建了实用的弹道预测工具,建立了基于神经网络理论的弹道预测模型。利用二自由度质点弹道模型,选取BP网络和Elman网络进行神经网络弹道预测仿真。基于误差反向传播理论,比较了带动量项算法与自适应学习率算法这2种网络权值训练速度。对2种网络不同隐层节点数的学习误差和预测误差进行了对比分析。数值仿真计算结果表明,神经网络具有较高的预测精度,36.7 km射程仅有不足100m的射程误差,12.3 km射高仅有不足70 m的高度误差,预测结果满足要求,利用神经网络进行弹道预测是合理可行的。In order to discuss the feasibility of neural network uses for trajectory prediction,and to create a practical prediction trajectory tool,we built a trajectory predictive model based on neural network. It used two degrees of freedom particle trajectory model to do neural network trajectory prediction simulation by selected BP network and Elman network. Based error back propagation theory,it compared two types of weight training velocities from momentum back propagation and adaptive learning rate back propagation in the network structures,and the comparative analysis of the learning error and prediction error of the two networks in different amount of hidden nodes were discussed. Numerical results show that it is less than100 m range error of 36. 7 km and less than 70 m height error of 12. 3 km shot high. The neural network has high prediction accuracy,and its trajectory prediction results meet the requirements. The ballistic prediction using neural networks is reasonably practicable.

关 键 词：弹道预测 神经网络 误差反传 数值仿真 

分 类 号：TJ303.4[兵器科学与技术—火炮、自动武器与弹药工程]