基于遗传BP算法预测贮存寿命
Prediction of Storage Life Based on Genetic BP Algorithm
郭俊伶 1彭志凌 1班伟2
作者信息
- 1. 中北大学 机电工程学院,太原 030051
- 2. 宜昌测试技术研究所,湖北 宜昌 443003
- 折叠
摘要
目的 针对引信自然贮存试验数据统计方法计算量大且试验时间开展较长的问题,提出 BP 和遗传算法相结合的方法(遗传BP算法),通过步加试验解决寿命预测问题.方法 通过步加试验数据求其试验各级应力下的环境因子,由环境因子将各级应力试验时间折合成实际贮存时间,根据Arrhenius 模型求出可靠度函数.其次,采用遗传算法优化BP神经网络,避免陷入BP局部最优问题,将步加试验数据代入遗传 BP算法进行训练,提高预测的精度和准确度.将正常应力下的数据代入遗传BP算法进行测试,求出可靠度预测值.最终对比实际、Arrhenius模型、遗传BP算法的贮存可靠度预测值.结果 实际、Arrhenius模型、遗传BP算法的贮存可靠度预测值相近,证明遗传BP算法可以满足引信贮存可靠度的预测.结论 采用遗传BP算法对步加试验进行寿命预测,可以有效减少试验时长和降低试验成本.
Abstract
In order to solve the problem of large amount of calculation and long test time in the statistical method of fuze natural storage test data,the work aims to propose a method combining BP and genetic algorithm(genetic BP algorithm),so as to solve the life prediction problem through step test.Firstly,the environmental factors under various levels of stress were cal-culated through step test data.The environmental factors were used to convert the stress test time at each level into the actual storage time,and the reliability function was calculated based on the model.Secondly,genetic algorithm was used to optimize the BP neural network to avoid the local optimal problem of BP.The step test data were substituted into the genetic BP algo-rithm for training,to improve the accuracy and precision of prediction.The data under normal stress were substituted into the genetic BP algorithm for testing,and the predicted reliability value was calculated.Finally,the actual storage reliability value and the predicted storage reliability values of model,and genetic BP algorithm were compared,which were similar,proving that the genetic BP algorithm could meet the prediction of fuze storage reliability.The genetic BP algorithm for predicting the life-span of step test can effectively reduce the test duration and lower the test cost.
关键词
步加试验/BP神经网络/遗传算法/恒湿步温/环境因子/Arrhenius模型Key words
step test/BP neural network/genetic algorithm/constant humidity step temperature/environmental factor/Ar-rhenius model引用本文复制引用
出版年
2024
NETL
NSTL
万方数据