作者：钱虹^1,2, 张超凡¹, 张凯文¹

作者单位：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

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

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。

Outlier detection and fault tolerant model of industrial process control based on data driven
QIAN Hong^1,2, ZHANG Chaofan¹, ZHANG Kaiwen¹
1. Shanghai University of Electric Power, Shanghai 200090, China;
2. Shanghai Key Laboratory of Power Station Automation Technology, Shanghai 200090, China
Abstract: 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.
Keywords: abnormal data detect;soft measurement base on SVR;KNN algorithm;Ridge regression algorithm;fault tolerant model
2021, 47(3):82-91  收稿日期: 2020-06-04;收到修改稿日期: 2020-07-23
基金项目: 上海市科委地方能力建设项目(18020500900)；上海市自然科学基金(19ZR1420700)
作者简介: 钱虹(1967-),女,上海市人,教授,博士,研究方向为火电站和核电站的运行智能优化控制、智能诊断、大数据挖掘等
参考文献
[1] 胡向东, 唐贤伦, 胡蓉. 现代检测技术与系统[M]. 北京: 机械工业出版社，2015.
[2] 周东华, DING X. 容错控制理论及其应用[J]. 自动化学报, 2000, 26(6): 788-797
[3] KAMESWARI U S, BABU I R. Sensor data analysis and anomaly detection using predictive analytics for industrial process[C]// IEEE Workshop on computer intelligence: Theories, Applications and Future Directions, 2016: 18.
[4] 沈飞,王辉,张广华, 等. 脉冲X光摄影法测量柱壳内爆膨胀速度的研究[J]. 中国测试, 2020, 46(1): 29-33
[5] CHEN Y, LAN L. Fault detection diagnosis and data recovery for a real building heating/cooling billing system[J]. Energy Conversion and Management, 2010, 51(5): 1015-1024
[6] 赵磊,张永祥,朱丹宸.复杂装备滚动轴承的故障诊断与预测方法研究综述[J].中国测试,2020,46(3):17-25.
[7] 赵璐,许昕,潘宏侠, 等. 基于信息熵与信息融合的供输弹系统故障诊断研究[J]. 机械设计与研究, 2020, 36(1): 169-172
[8] ABDELGHAFAR S, DARWISH A, HASSANIEN A E. Cube satellite failure detection and recovery using optimized support vector machine[J]. Springer,2018：664-674.
[9] 王昕,赵飞,蒋佐富, 等. 迁移学习和卷积神经网络电力设备图像识别方法[J]. 中国测试, 2020, 46(5): 108-113
[10] 李欢欢, 司风琪, 徐治皋. 一种基于鲁棒自联想神经网络的传感器故障诊断方法[J]. 中国电机工程学报, 2012, 32(14): 116-121
[11] 秦一菲, 马明辉, 王岩松. 基于改进KNN算法的交通流异常数据修复方法[J]. 计算机测量与控制, 2018, 26(12): 180-184
[12] AHN J J, BYUN H W, OH K J, et al. Using ridge regression with genetic algorithm to enhance real estate appraisal forecasting[J]. Expert Systems with Applications, 2012, 39(9): 8369-8379
[13] 黄宴委. 基于核岭回归的非线性内模控制[J]. 控制与决策, 2009, 24(7): 1100-1103
[14] 赵磊,张永祥,朱丹宸.复杂装备滚动轴承的故障诊断与预测方法研究综述[J].中国测试,2020,46(3):17-25.
[15] 周彬.多视图视觉检测关键技术及其应用研究[D].杭州：浙江大学，2019.
[16] 任子强,司小胜,胡昌华, 等.融合多源数据的非线性退化建模与剩余寿命预测[J].中国测试,2020,46(2):1-8.
[17] 杨金芳, 翟永杰, 王东风, 等. 基于支持向量回归的时间序列预测[J]. 中国电机工程学报, 2005(17): 110-114
[18] CORTES C, VAPNIK V. Support vector machines[J]. Machine Learning, 1995, 11(20): 1-25
[19] CRISTIANINI N, SHAWETAYLOR J. An introduction to support vector machines[C]. China Machine Press, 2000: 1-28.
[20] 李航. 统计学习方法[M]. 北京: 清华大学出版社, 2012.