一种基于神经网络的中制导改进算法  

A Modified Algorithm on the Midcourse Guidance Based on BP Neural Network

作　　者：魏倩[1] 蔡远利[1] 

机构地区：[1]西安交通大学电子与信息工程学院,西安710049

出　　处：《西安交通大学学报》2016年第7期125-130,共6页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(61308120;61463029)

摘　　要：针对地球扁率影响下的大气层外导弹中段制导问题,提出了基于BP神经网络的模型预测改进算法,并且创新使用轨道偏差解析解来构造训练样本集。首先,利用极点变换方法把弹体受到的J_2项摄动引力优化分解为与运动轨迹相关的扰动函数;然后,采用偏差状态空间的转移矩阵,建立起导弹在J_2项摄动作用下的轨道偏差公式;最后,利用偏差公式构造取值广泛的训练样本集并训练BP神经网络,从而建立起关于虚拟目标信息的预测模型,计算出中段制导控制所需的增益速度矢量。该模型的优点是利用极点变换和状态转移矩阵直接求解J_2项摄动偏差,避免了进行大规模的数值积分运算;神经网络拥有强大的学习能力,保证了预测模型的全面性及精确性;BP神经网络可以预先离线训练、学习,大大缩短了计算时间。与传统Lambert迭代补偿修正方法相比,改进型BP神经网络补偿算法可以同时满足实时计算速度及计算精度的双重要求,具有较强的实际工程意义。Aiming at solving the midcourse guidance problem of exo-atmosphere missile under the influence of earth＇s oblateness perturbation,aprediction model for missile guidance was proposed based on BP neural network.This method provides a new training sample set constructed by the analytical formula of trajectory deviation.First,by using the pole transform method,the missile＇s J_2 perturbed gravity is decomposed into the disturbing function related to its flight trajectory.Then,with the state space matrix method,the analytic solution of trajectory deviation with the J_2 perturbation is calculated.Finally,using trajectory deviation function to construct a wide range of training sample set,the BP neural network of prediction model is established.The neural network can forecast the virtual target point information,so as to calculate the vector of gained velocity for midcourse guidance control.Using the modified algorithm,the trajectory deviation of J_2 perturbation can be directly solved by pole transform and state transition matrix,avoiding large-scale numerical calculation.The BP neural network has powerful learning and training ability,ensuring the comprehensiveness and accuracy of the prediction model and saving calculation time by the off-line training and learning before the simulation tests.In comparisonwith traditional correction method of Lambert guidance,this modified algorithm can satisfy the requirements on both efficiency and accuracy of real-time computation,being of practical engineering significance.

关 键 词：极点变换 摄动偏差 J2项摄动 BP神经网络 

分 类 号：V448.2[航空宇航科学与技术—飞行器设计]