摘要
在导弹贮存延寿工程的立项论证阶段,需要先统计分析每个导弹贮存期内环境剖面数据、故障数据等以分析导弹的贮存故障率变化趋势,进而科学确立延寿目标。针对高可靠性导弹故障数据样本少、环境数据的变化存在非线性、传统方法预测贮存可靠度精度不高等问题,提出SelfSVR算法对导弹贮存故障率进行预测。通过贝叶斯优化方法自适应计算出Mixup生成的样本个数、SVR算法的超参数核函数系数和核函数惩罚系数。基于Mixup增强算法构建新的训练样本。使用SVR算法对导弹贮存故障率进行非线性预测。通过示例验证,Self-SVR算法在预测非线性变化的贮存故障率方面具有良好的拟合效果,在R2系数、EVS和MSE三个指标上优于对比算法。在消融实验中证明了贝叶斯优化和Mixup增强算法对Self-SVR算法均具有优化作用。
In the project demonstration stage of missile storage life extension project,the environmental profile data and fault data of each missile storage period are analyzed to understand the change trend of the missile storage failure rate,and the life extension goal is then scientifically established.The problems of a few fault data samples for high reliability missiles,nonlinear changes in environmental data,and the low accuracy of traditional methods to predict storage reliability are addressed by proposing the Self-SVR algorithm to predict missile storage failure rates.The number of samples generated by Mixup algorithm,the coefficient of the hyperparameter kernel function,and the penalty coefficient of the kernel function of the SVR algorithm are adaptively calculated by the Bayesian optimization method.New training samples are then constructed based on the Mixup enhancement algorithm.The SVR algorithm is adopted to nonlinearly predict the missile storage failure rate.The results show that the Self-SVR algorithm has a good fitting effect in predicting the storage failure rate of nonlinear changes and is superior to comparison algorithms in R2 coefficient,EVS and MSE.In the ablation experiment,it is proven that both Bayesian optimization and the Mixup enhancement algorithm can optimize the Self-SVR algorithm.
作者
陈凯诺
张福光
张涵
尹延涛
杜光传
Chen Kainuo;Zhang Fuguang;Zhang Han;Yin Yantao;Du Guangchuan(Naval Aviation University,Yantai 264001,China;Yantai Education Enrollment Examination Institute,Yantai 264003,China)
出处
《战术导弹技术》
北大核心
2024年第5期74-82,98,共10页
Tactical Missile Technology
关键词
支持向量回归
贝叶斯优化
Mixup算法
数据增强
贮存故障率
导弹贮存延寿
延寿目标
Support Vector Regression
Bayesian optimization
Mixup algorithm
data enhancement
storage failure rate
missile storage life extension
life extension goal
作者简介
通讯作者:陈凯诺,博士。
