一类解决变应力加速寿命试验参数估计的方法.pdf

DOI :10.13700/j.bh.1001-5965.2005.11.004 2005 11 November　2005 　31 11 Journal of Beijing University of eronautics and stronautics Vol.31　No.11 苑中魁　　王少萍 ( , 100083) 　　　　　:变应力加速寿命试验的极大似然函数是高维非线性复杂目 函 数, 其待估参数多, 采用梯度下降优化方法进行参数估计容易陷入局部极值, 而采用 全局优化方法又存在寻优效率低的问题.为了解决复杂多维目 函数优化的瓶颈问 题, 设计了 一种基于实数编码遗传算法和Powell 法的遗传加速方法.利用适应度函数 获得两种优化方法的最佳切换点, 最大程度发挥遗传算法和Powell 算法的优点, 既提 高了多维非线性目 函数寻优效率又保证了参数估计的全局最优.液压泵加速寿命 试验实例分析结果表明, 遗传加速方法可以在寻优前期利用遗传算法保证待估参数 的全局最优估计, 在寻优后期快速逼近最优值, 使寻优成功率达到85%. 　　:加速寿命试验;极大似然估计;遗传算法;Powell 法;遗传加速方 法 :V 241.01 : 　　　　 :1001-5965(2005)11-1172-05 One method solving parameter estimation of varying stress accelerated life test Yuan Zhongkui　Wang Shaoping (School of utomation Science and Electrical Engineering, Beijing University of eronautics and stronautics, Beijing 100083, China) Abstract :The maximum likelihood function of varying stress accelerated life is a nonlinear multi-dimension complex object functionwith five evaluated parameters, whose parameter estimation is easy to get into local optimiza- tion with grads descending algorithmwhile has low searching efficiency with global optimal algorithm.To solve the bottleneck between direct and intellective optimization of multi-dimension complex object function, the genetic ac- celerated algorithm was presented based on real code genetic algorithm and Powell method.Through designing per- fect switch with adaptive function, the genetic accelerated algorithm takes the advantages both genetic algorithm and Powell method at furthest that ensure global optimization and keep rapid searching velocity to multi-dimension com- plex object function. ppl