首页|基于SRKDA的系统故障演化过程分解方法研究

基于SRKDA的系统故障演化过程分解方法研究

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
为研究系统故障演化过程中可能蕴含的多种演化特征,对演化过程的分解进行研究,提出基于谱回归核判别分析(SRK-DA)的演化过程分解方法.首先介绍演化过程的特点和分解原理,其次论证对象集合对演化过程的可表示性,给出分解方法流程,最后进行实例分析.研究结果表明:分解演化过程本质上是对象与系统功能状态对应关系的确定,各对象集合都对应了各自的子演化过程;线性和非线性条件下对象可表示各种功能状态;对象标签矩阵须满足标签值的均匀分布特征;使用SRKDA算法可以确定最大准确度和最优对象标签集合,实现演化过程的分解;实例分析得到在20000 次迭代后最大准确度为0.85,3 个子演化过程分别包含41,33,26 个对象.研究结果可为系统故障过程的特征分析提供参考方法.
Research on decomposition method of system fault evolution process based on SRKDA
In order to study the multiple evolution characteristics possibly contained in the system fault evolution process,the decomposition of evolution process was studied,and a decomposition method of evolution process based on the spectral regres-sion kernel discriminant analysis(SRKDA)was proposed.The characteristics and decomposition principle of evolution process were introduced,and the representability of object set to the evolution process was demonstrated.The procedure of de-composition method was given,and finally an example analysis was carried out.The results show that the decomposition evolu-tion process is essentially the determination of the corresponding relationship between object and system function state,and each object set corresponds to its own sub-evolution process.The object can represent various function states under linear and nonlinear conditions.The object label matrix must meet the uniform distribution characteristics of label values.The maximum accuracy and the optimal object label set can be determined by using the SRKDA algorithm,and the decomposition of evolu-tion process can be realized.The example analysis shows that the maximum accuracy is 0.85 after 20000 iterations,and the three sub-evolution processes contain 41,33 and 26 objects,respectively.

safety system engineeringsystem fault evolution processspectral regression kernel discriminant analysis(SRKDA)evolution decomposition methodmaximum accuracyobject label matrix

崔铁军、李莎莎

展开 >

沈阳理工大学 环境与化学工程学院,辽宁 沈阳 110159

安全系统工程 系统故障演化过程 SRKDA 演化分解方法 最大准确度 对象标签矩阵

国家自然科学基金

52004120

2024

10.11731/j.issn.1673-193x.2024.03.027

中国安全生产科学技术
中国安全生产科学研究院

中国安全生产科学技术

CSTPCD北大核心
影响因子：1.119
ISSN：1673-193X
年,卷(期)：2024.20(3)
  • 16