数据驱动工业异常值检测与容错模型研究  

Outlier detection and fault tolerant model of industrial process control based on data driven

作　　者：钱虹 张超凡 张凯文 QIAN Hong;ZHANG Chaofan;ZHANG Kaiwen(Shanghai University of Electric Power,Shanghai 200090,China;Shanghai Key Laboratory of Power Station Automation Technology,Shanghai 200090,China)

机构地区：[1]上海电力大学,上海200090 [2]上海市电站自动化技术重点实验室,上海200090

出　　处：《中国测试》2021年第3期82-91,共10页China Measurement & Test

基　　金：上海市科委地方能力建设项目(18020500900);上海市自然科学基金(19ZR1420700)。

摘　　要：针对工业过程控制数据异常时,控制系统无法实现闭环稳定可靠工作的问题,提出一种基于KNN算法和Ridge回归算法结合对异常数据高精度恢复,并设计实现控制系统容错运行的数据驱动模型。首先利用基于径向基核函数的支持向量回归机(RBF-SVR)观测器对目标变量状态进行异常检测,其次使用Ridge算法对异常数据点的K个最近邻工况数据进行回归运算,从而恢复异常点数据,最后通过容错切换机制实现控制数据异常时系统容错运行。使用电厂历史数据验证方法的有效性并与其他数据恢复方法进行对比。结果表明,使用该文所提方法对异常数据的恢复值与实际原始值之间的MAPE仅为2.4789%,与RBF-SVR软测量模型相比回归准确度提高6.209%,恢复的数据能够可靠应用于系统容错控制运行中。Interferenced by uncertain factors,industrial process control data becomes too abnormal to be used to maintain closed loop operation of control system reliably.An model for data recovery and fault tolerance control,which is based on KNN algorithm and ridge regression algorithm,is established.First of all,the support vector regression machine observer based on the radial basis kernel function(RBF-SVR)is established to detect target variable status.Then,utilizing the Ridge algorithm to data fitting and recover the fault point data on the K nearest neighbors data which selected from the historical database by KNN.Lastly,Establishing a fault-tolerant switching mechanism to achieve stable operation when data is abnormal.Plant field history data is applied to verify the effectiveness of the method,The mean absolute percentage error between the actual original data and the data which recovered by the method proposed in this article is 2.4789%.Compared with the soft measurement method model based on RBF-SVR,the regression accuracy is improved by 6.209%,so the recovered data can be reliably applied to system fault-tolerant control operation.

关 键 词：异常数据检测 SVR软测量 KNN算法 Ridge回归算法 容错模型 

分 类 号：TK37[动力工程及工程热物理—热能工程]