摘要
由于传感装置失效导致的工业过程数据错误会严重威胁系统的安全平稳运行。虚拟传感技术能够推断、解释和预测系统的真实行为,实现过程变量观测值的冗余输出。为了提高工业系统的安全性,提出了基于虚拟传感技术的工业过程数据错误诊断方法。首先针对系统中的各个观测变量,构建基于高斯过程自回归（GPAR）模型的虚拟传感器,通过监测各虚拟传感器预测残差的波动变化,辨识、区分由于传感器故障导致的观测值错误和由于过程扰动引起的数据异常。一旦检测到过程数据出现错误,隔离发生数据错误的过程变量,并构造基于高斯过程多变量回归（GPMR）模型的虚拟传感器,实现过程数据错误值的修复。最后通过连续搅拌釜式加热器（CSTH）仿真实验与某炼厂闪蒸塔现场应用案例验证了本文所提方法的有效性。
The faulty measurements result from failure of the sensing device will seriously threaten the safety and normal operation of system for industrial process.The virtual sensing technology is commonly used to provide the redundant or virtual instruments for online real-time monitoring of process parameters.Thus,the real behaviors of the industrial system can be inferred,explained and predicted by the virtual sensing technology.To reduce the risk of production process and improve the safety of plant operations,a diagnosis method of faulty measurements for industrial process based on virtual sensing technology is proposed.First,virtual sensors are constructed based on Gauss process auto-regressive（GPAR） model for each process variable in the industrial system.Then,the anomalies of process data that result from process disturbances or faulty measurements can be identified through monitoring the prediction residuals of GPAR virtual sensors.Once the presence of faulty measurements is detected,the process variable whose sensor is faulty can be isolated.Furthermore,the virtual sensor based on Gauss process multivariate regression（GPMR） model is built to reconstruct true measurements.The results of applications for both continuous stirred tank heater（CSTH） simulation platform and the flash tower in the real industrial field demonstrate the effectiveness of this proposed method.
作者
胡瑾秋
郝笑笑
张来斌
Hu Jinqiu;Hao Xiaoxiao;Zhang Laibin(College of Mechanical and Transportation Engineering, China University of Petroleum, Beijing 102249, China)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2018年第3期29-36,共8页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(51574263)
北京市科技新星计划(Z181100006218048)
国家重点研发计划(2017YFC0805801)
中国石油大学(北京)青年创新团队C计划(C201602)项目资助
关键词
工业过程
错误数据诊断
虚拟传感
传感器故障
过程扰动
industrial process
fauhy measurements diagnosis
virtual sensing
sensor fault
process disturbances
