摘要
针对设备剩余寿命预测误差较大的问题,提出一种基于高阶隐半马尔科夫模型(HOHSMM)的剩余寿命预测模型。首先基于隐半马尔科夫模型,建立了HOHSMM,提出一种基于排列的HOHSMM降阶方法和复合节点机制,并相应地改进状态转移矩阵和观测矩阵,使得高阶模型转化为对应的一阶模型,将更多的节点依赖关系信息储存在待估计参数组中。其次,采用智能优化算法群代替EM算法,对模型进行参数估计以及结构优化,实现了智能优化算法对高阶模型拓扑结构的简化。再次,定义并推导了高阶模型中的状态驻留变量,运用基于多项式拟合的预测方法实现了在先验分布未知情况下的设备剩余寿命预测。最后,通过美国卡特彼勒公司液压泵数据集对所提框架进行了验证,结果表明,基于高阶隐半马尔科夫模型的设备剩余寿命预测方法是更加有效的。
To solve the problem of large residual life prediction error,a residual life prediction model based on Higher-order Hidden Semi Markov Model(HOHSMM)was proposed.Based on Hidden Semi Markov Model,HOHSMM was established,and an order reduction method was proposed based on permutation and composite nodes mechanism,accordingly improve the state transition matrix and the observation matrix,making the higher order model be converted into the corresponding first-order model.Thus there would be more nodes dependency information storing in the parameters to be estimated.Estimation Maximal algorithm was replaced by intelligent optimization algorithm group to estimate the parameters and optimize the structure of the model,which simplified the topology of the high-order model.T he state lingering variable in the higher-order model were defined and derived,and the prediction method based on polynomial fitting was used to realize the prediction of equipment residual life in the case of unknown prior distribution.The framework was validated by the hydraulic pump data set of Caterpillar Inc.The results showed that the residual life prediction method based on the High-order Hidden Semi Markov Model was more effective.
作者
刘文溢
刘勤明
周林森
LIU Wenyi;LIU Qinming;ZHOU Linshen(School of Business,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《计算机集成制造系统》
EI
CSCD
北大核心
2022年第8期2387-2398,共12页
Computer Integrated Manufacturing Systems
基金
国家自然科学基金资助项目(71840003)
上海市自然科学基金资助项目(19ZR1435600)
教育部人文社会科学研究规划基金资助项目(20YJAZH068)
上海理工大学科学发展项目(2020KJFZ038)。
关键词
高阶隐半马尔科夫模型
复合节点
模型降阶
状态驻留
多项式拟合
剩余寿命预测
higher-order hidden semi Markov model
composite nodes
model order reduction
state lingering
polynomial fitting
residual useful life
