摘要
以温度为表征参量,基于谱分析方法和时频分布理论,利用温度序列的AR模型的极点参数作为前端处理的特征参数,模拟了智能系统对温度信号的处理过程,来对液化天然气(LNG)分层涡旋事故进行有效的识别与诊断.过程中,为了压缩特征维数并选择可分性最好的特征,采用了基于距离准则的特征选择方法.对温度序列的特征提取表明:从分层第一阶段过渡到第二阶段后,温度信号频率的主峰中心频率部分发生了偏移,而且频率幅度增大;涡旋临界状态的AR谱比较杂乱,既有特征谱峰又有其他谱峰;两种状态的特征参数———中心频率、谱峰幅值、高频能量和高频能量比等区别较大.最后,通过AR模型参数的欧式距离对分层涡旋的发展阶段及状态进行了模式识别.结果表明:相同状态的欧式距离趋向于零,由于信号的随机性而呈现较小的非零量;不同状态之间的欧式距离值差异明显,能够实现分层涡旋的故障状态识别.
By selecting the temperature as a characterization parameter and the pole parameter of AR model of temperature sequence as the feature parameter of front-end processing, the treatment of temperature signals by an intelligent system is simulated based on the spectral analysis and the time - frequency distribution theory , and then the recognition and diagnosis of stratification and rolling are performed for the liquefied natural gas （LNG） in tank. Moreover, the character-selecting method based on distance criterion is adopted to compress the feature dimension and to select an optimal character with good separability. The results of feature extraction for temperature sequence show that （ 1 ） from the stratification state to the rolling state, the center frequency of the main peak deviates from the increasing peak value ; （2） the fuzzy AR spectrum in the critical stratification and rolling states consists of both characteristic peaks and some other peaks ; and （3） the stratification and rolling states are of different characteristic parameters such as the center frequency, the peak value, the high-frequency energy and the high-frequency energy ratio. Moreover, the results of pattern recognition for stratification and rolling based on the Euclidean distance of AR model parameter indicate that the Euclidean distances in similar states are approximate to each other but are not equal to zero due to the randomness of temperature signals, and that the obvious differences in Euclidean distances in different states help to implement the diagnosis and recognition of stratification and rolling for LNG.
出处
《华南理工大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2008年第2期107-111,共5页
Journal of South China University of Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(50474034)
关键词
液化天然气
分层
涡旋
AR谱分析
欧式距离
模式识别
liquefied natural gas
stratification
rolling
AR spectral analysis
Euclidean distance
pattern recognition
作者简介
王海蓉（1974-），女，博士生，主要从事液化天然气储运安全研究．E-mail：wanghairong211@sohu．com
