摘要
由于核动力系统的在线监测参数在获取、传输过程中受到噪声的干扰,导致最终监测信号的随机缺失,对操纵员判断事故种类造成了较大的干扰。为此,提出了滑动时间窗口的动态时间弯曲故障诊断模型:构建在线实时监测参数的待测多元时间序列和已有的事故标准序列,将构建的待测多元时间序列采用滑动窗口去动态寻找标准序列中的最小累积距离,使用动态时间弯曲的算法计算待测序列或标准序列中监测参数缺失导致的序列不等长现象,通过最小累积距离得到待测时间序列的模式类别。结果表明:该方法从核动力系统事故发生的基本原理出发,对诊断结果具有较强的解释性和鲁棒性,同时可引入其他标准事故序列对模型进行拓展,该模型具有较强的拓展性。
The on-line monitoring parameters of nuclear power system are disturbed by noise during the process of acquisition and transmission,which leads to the random deletion of the final monitoring signal.The great disturbance for the operator to judge the type of accident was caused.Therefore,a sliding time window fault diagnosis model based on dynamic time warping(DTW)was proposed.The multivariate time series were built from on-line real-time monitoring and the existing accident standard series.The sliding time window was used to dynamically search the minimum cumulative distance in the standard series.The absence of monitoring parameters in the test series or the standard series leaded to the unequal length sequence,and DTW was used to deal the phenomenon.The pattern category of the time series depended on the minimum cumulative distance.This method is based on the basic principle of the accident in nuclear power system,and has strong explanation and robustness.At the same time,the model can be extended by other standard accident sequences.
作者
赵鑫
蔡琦
王晓龙
ZHAO Xin;CAI Qi;WANG Xiaolong(College of Nuclear Science and Technology,Naval University of Engineering,Wuhan 430033,China)
出处
《原子能科学技术》
EI
CAS
CSCD
北大核心
2019年第6期1070-1077,共8页
Atomic Energy Science and Technology
基金
海军装备军内科研项目资助(HJ20172A06400)
关键词
动态时间弯曲
多元时间序列
随机缺失
滑动窗口
故障诊断
dynamic time warping
multivariate time series
random deletion
sliding time window
fault diagnosis
