复合主元分析方法在电站冗余传感器故障检测中的应用

Application of Hybrid Principal Component Analysis to Fault Detection of Redundant Sensors in Power Plants

提出一种应用于电站冗余传感器故障检测的复合主元分析方法:首先故障检测系统连接实时/历史数据库获取冗余传感器的实时数据,并判断系统的工况是否稳定.若工况稳定,则采用主元分析方法(PCA)估计参数值,否则采用仪表监测算法(ICMP)对非稳态信号进行估计.再将估计值与每个测量信号形成残差,采用改进的序贯概率比方法评估残差大小,并进行异常预警.该方法具备了PCA算法在稳定工况下的可靠性和ICMP算法在非稳态工况下的有效性.选用某600 MW机组热力系统的一组实际监测参数作为算例,验证了该方法的先进性.最后给出检测系统的工程实施架构和技术方案.

A hybrid principal component analysis method for fault detection of power plant redundant sensors is being proposed based on principal component analysis（PCA） and instrument calibration monitoring program（ICMP）.First,real-time data of relevant redundant sensors are acquired by fault detection system from relevant real-time/history database so as to estimate the stability of the system.If it is stable,PCA shall be used to perform estimation for steady-state signals, and if not,ICMP shall be adopted for unsteady-state signals.Then residuals are achieved from the difference between estimated data and measurement signals for each sensor.Finally,an adapted sequential probability ratio test（ASPRT） method is used to evaluate the residuals,and subsequently an alarm will be given if necessary.The hybrid method combines the advantages of high stability of PCA algorithm under steady-state conditions and high effectiveness of ICMP algorithm under unsteady-state conditions,of which the advancement has been verified by calculation with monitoring parameters of an actual 600 MW power unit＇s thermal system.Moreover,implementation scheme and technical plan for application of the system to actual engineering projects are also put forward.