基于动态核独立分量分析的高含硫天然气净化过程异常检测与诊断

目前高含硫天然气净化过程存在多参数动态相关的特性,导致基于静态多元统计过程监控方法对于异常状态检测效果较差。提出一种考虑参数时序自相关性的动态核独立分量分析(DKICA)异常检测与诊断方法。首先,引入自回归(AR)模型,通过参数辨识确定模型阶次,描述监控过程的时序自相关性;然后,将原始变量投影到核独立元空间,通过监控独立元对应的T²和SPE统计量是否超出正常状态设定的控制限,实现异常检测;最后计算所述T²统计量对原始变量的一阶偏导数,绘制贡献图实现异常诊断。以某高含硫天然气净化厂采集的数据进行分析,结果表明基于DKICA高含硫天然气净化过程异常检测精度要优于静态独立分量分析所得的检测精度。

Anomaly detection and diagnosis of high sulfur natural gas purification process based on dynamic kernel independent component analysis

At present, the parameters of high sulfur gas purification process present timing autocorrelation characteristics, resulting in poor static multivariate statistical process monitoring for abnormal condition. An anomaly detection and diagnosis method called Dynamic Kernel Independent Component Analysis (DKICA) was proposed, which considered the timing autocorrelation of parameters. Firstly, Auto-Regression (AR) model was introduced. The model order was determined by the parameter identification to describe the timing of autocorrelation in the monitoring process. Secondly, original variables were projected to a kernel independent space, their T² and SPE statistics were monitored to realize anomaly detection by judging whether they exceeded control limit of normal condition. Finally, the first order partial derivative of the T² statistic to original variable was calculated, and the contribution plot was given to achieve abnormality diagnosis. The data collected from a high sulfur gas purification plant was analyzed, and the results showed the detection accuracy of DKICA was prior to that of Kernel Independent Component Analysis (KICA).