李小波, 汪翔, 王睿轶,等.考虑变点和测量误差的剩余寿命预测方法研究[J].河南理工大学学报（自然科学版）,2026,45(5):20-29.

LI X B, WANG X, WANG R Y,et al.Research on remaining life estimation method considering change points and measurement errors[J].Journal of Henan Polytechnic University(Natural Science) ,2026,45(5):20-29.

考虑变点和测量误差的剩余寿命预测方法研究

李小波1, 汪翔1, 王睿轶2, 沈青2

1.上海工程技术大学 城市轨道交通学院，上海  201620；2.上海地铁电子科技有限公司，上海  200233

摘要: 目的为进一步提高产品剩余寿命的预测精度，提出一种同时考虑变点和测量误差的两阶段非线性Wiener过程退化模型。  方法 首先，根据同类型产品退化数据的统计特性，利用变点监测算法离线估计产品的变点，并通过极大似然估计法求解模型初始参数；其次，基于同类型产品的变点位置确定退化角准则中的阈值，实现在线监测目标产品的变点，在此基础上，利用Kalman滤波算法估计目标产品的隐状态，在线更新模型参数，并推导考虑模型参数在线更新的剩余寿命概率密度函数表达式；最后，分别采用仿真数据和NASA电解电容加速退化的试验数据与本文模型进行对比分析，验证所提方法的有效性和合理性。  结果仿真数据分析结果表明，与单阶段非线性Wiener过程相比，所提方法的绝对平均误差减小1.558 9；加速老化试验数据验证结果表明，与传统的2种Wiener过程模型相比，本文模型对电容的剩余寿命预测总均方误差分别减小524.473 3，112.759 1。  结论研究结果对于不同使用时间段退化规律不同的产品剩余寿命研究具有重要参考价值，尤其适用于产品性能退化初期与退化末期特征差异明显的情形。

关键词:剩余寿命预测;两阶段非线性Wiener过程;变点估计;测量误差

doi:10.16186/j.cnki.1673-9787.2025050045

基金项目:国家自然科学基金资助项目（51907117）

收稿日期:2025/05/23

修回日期:2025/08/20

出版日期:2026-07-24

Research on remaining life estimation method considering change points and measurement errors

Li Xiaobo1, Wang Xiang1, Wang Ruiyi2, Shen Qing2

1.College of Urban Rail Transit， Shanghai University of Engineering Science， Shanghai  201620， China；2.Shanghai Metro Electronic Technology Co.， Ltd.， Shanghai  200233， China

Abstract: Objectives In order to improve the prediction accuracy of the remaining service life of the product， a two-stage nonlinear Wiener process degradation model in which both change points and measurement errors were considered was proposed.  Methods Firstly， based on the statistical characteristics of degradation data for products of the same type，a change point detection algorithm was used to estimate the change points of the product offline based on statistical characteristics of degradation data for products of the same type. The maximum likelihood estimation method was used to solve the initial model parameters. Then，based on the change point position of the same type products， the threshold value of degradation angle criterion was determined to realize monitoring the change point of target products online. On this basis， the Kalman filtering algorithm was used to estimate the hidden state of the target product， and the model parameters were updated online. The expressions for the RUL probability density function and cumulative distribution function considering online updating of the model parameters were derived. Finally， the effectiveness and rationality of the proposed method were verified by comparing and analyzing the simulated degradation data and the experimental data on NASA electrolytic capacitor accelerated degradation， respectively.  Results Compared with the single-stage nonlinear Wiener process， the simulation data analysis results showed that the mean absolute error（MAE） of the proposed algorithm decreased by 1.558 9； Comparinged with the traditional two Wiener process models， the verification results of accelerated aging test data indicated that the mean square error of capacitor remaining life prediction using the proposed method have been improved by 524.473 3 and 112.759 1， respectively. Conclusions The research results have important reference values for remaining life study of products with different degradation patterns during different usage periods， especially suitable for situations where there was a significant difference in degradation characteristics between the early and late stages of product performance degradation.

Key words:remaining useful life prediction;two-stage nonlinear Wiener process;change point detection;measurement error