自适应双层果蝇相关向量机起重机当量载荷谱预测方法  

Crane equivalent load spectrum prediction method based on relevance vector machine improved by adaptive double layer fruit fly algorithm

作　　者：董青[1] 徐格宁[1] 

机构地区：[1]太原科技大学机械工程学院,山西太原030024

出　　处：《机械设计》2017年第2期86-93,共8页Journal of Machine Design

基　　金："863"国家高技术研究发展计划资助项目(2013AA040203);"十二五"国家科技支撑计划资助项目(2011BAK06B05-05)

摘　　要：起重机载荷谱是其疲劳剩余寿命计算和安全评估的主要影响因素之一。针对载荷谱的随机性、不确定性及数据样本有限的问题,提出自适应双层果蝇相关向量机(ADRVM)的起重机当量载荷谱预测方法。以混合核函数为基础,将自适应步长的果蝇算法直接应用于相关向量机(RVM)核参数的优化选择中,同时为避免人为因素的影响,提出自适应双层果蝇算法对自适应步长的附加参数进行选取,从而克服了单一核函数下RVM的局限性,解决了自适应步长引起的附加参数的选取没有理论公式和有效依据的问题,提高了预测精度与鲁棒性。用ZLJ5551JQZ110V汽车起重机的小样本载荷谱对提出的方法进行验证。结果表明,在训练特性及预测精度方面,所提方法均优于粒子群优化算法的相关向量机、v-SVRM和LMBP神经网络。Crane load spectrum is one of the most important factors on the fatigue residual life calculation and safety assess- ment. Aiming at the randomness, uncertainty and finiteness of load spectrum samples, an equivalent load spectrum prediction method based on relevance vector machine improved by the adaptive double layer fruit fly algorithm was proposed. Based on the mixed kernel function, the adaptive step-size fruit fly algorithm was applied to the optimization of kernel parameters of relevance vector machine（RVM）. Meanwhile, in order to void the influence of human factors, the adaptive double layer fruit fly algorithm was proposed to select adaptive step-size additional parameters. Therefore, the limitation of RVM under single kernel function was eliminated. The problem that no theoretical formula and effective basis for the selection of additional parameters in the adaptive step-size was solved. The prediction accuracy and robust character were improved, The proposed method was validated using ZLJ5551JQZ110V truck crane load spectrum. The results demonstrated that the proposed method was superior to other methods, including relevance vector machine based on particle swarm optimization algorithm（PSORVM）, v-SVRM and LMBP.

关 键 词：混合核函数 自适应双层果蝇算法 相关向量机 当量载荷谱 

分 类 号：TH213[机械工程—机械制造及自动化]