基于前馈神经网络的外挂物投放分离预测  

External Store Separation Prediction Based on Feedforward Neural Network

作　　者：宋居正 刘思位 吴云山 SONG Juzheng;LIU Siwei;WU Yunshan(AVIC Xi’an Aircraft Industry Group Company Ltd.,Xi’an 710089,China)

机构地区：[1]中航西安飞机工业集团股份有限公司,陕西西安710089

出　　处：《弹道学报》2023年第3期33-38,共6页Journal of Ballistics

摘　　要：为研究外挂物投放分离运动特性,提高外挂物投放分离运动计算效率,基于前馈神经网络建立了外挂物投放分离预测模型,并对该模型的预测性能进行了测试分析。首先通过准定常方法对不同投放状态下外挂物的分离运动进行数值模拟,获取外挂物的分离轨迹和姿态数据,作为神经网络训练所用数据样本。然后建立单隐层神经网络,以外挂物投放状态与分离时间作为神经网络的输入,以外挂物的分离轨迹和姿态作为神经网络的输出,对神经网络进行训练。经过1000次迭代优化后,神经网络训练完成。验证结果表明,训练完成的神经网络可以基于投放状态对外挂物分离轨迹和姿态进行快速、精确预测。神经网络对分离轨迹的预测精度高于其对分离姿态的预测精度。神经网络对单个投放状态下1 s内的外挂物分离运动预测用时为0.06 s,作为对比,采用传统准定常方法数值模拟的平均用时为20 h。该方法提高了外挂物投放分离动力学的计算效率,为飞行器的投放分离特性研究提供了新的思路。In order to study the effect of external store drop condition on its separation characteristics and improve the calculation efficiency of the simulation of the separation,an external store separation prediction model was proposed based on the feedforward neural network.The prediction performance of the model was tested and analyzed.The separation of the external store in different drop conditions was numerically simulated by the quasi-steady method.The separation trajectory and attitude data of the external store were obtained and taken as data samples for neural network training.The neural network was then built and trained with the external store drop condition and separation time as the input,and the separation trajectory and attitude were taken as the output.After 1000 iterations of optimization,the neural network training was completed.The verification results show that the trained neural network can make immediate and accurate predictions of the external store’s separation trajectory and attitude based on the drop condition.The prediction accuracy of trained neural network for the separation trajectory is higher than its prediction accuracy for the separation attitude.In a single drop condition,the neural network prediction time of the separation of external store within 1 s is 0.06 s,and the average time of numerical simulation using the traditional quasi-steady method is 20 h.The proposed method improves the computational efficiency of external store drop separation simulation and offers a new approach for the study of the separation characteristics of aircraft.

关 键 词：外挂物投放 轨迹姿态 神经网络 预测模型 

分 类 号：V211.3[航空宇航科学与技术—航空宇航推进理论与工程]