利用跟随周期均值显著化序列异常数据的学习算法
|
摘要
针对工业控制中硬件存储空间、计算性能极为有限,无法应用计算机可运行的异常序列数据检测算法,以及常见可用算法比较复杂的问题,提出了一种利用跟随周期均值显著化序列异常数据的学习算法。首先,对序列数据预处理,利用有异常位置标记的序列集,可以得到最优周期T和周期均值差值阈值Dmax;其次,按照周期T分组检测序列,求出最近两组均值的差值,差值超过阈值时可判断出现异常。算法在学到参数后,判断异常过程所需存储空间和运算量很少,实验结果表明此算法对序列异常数据有显著化分离作用,在实际工程中抗干扰能力好,可有效减少异常点的误判率。
For industrial control,the hardware storage space and computing performance are very limited,it is unable to run the abnormal series data detection algorithms and its common algorithms are morecomplex.The paper proposes the learning algorithm based on following period mean value to outstand a-nomaly series data.Firstly,after series data preprocessing,abnormal locations of the series set is marked,so the algorithm can get the best period T and the threshold Dmax of period mean D-value;Secondly,thedetected series is divided into groups according to T and compute the D-value of the adjacent two periodmean values.If the D-value exceeds Dmax,an anomaly appears.After the algorithm has learned the pa-rameters,the less storage space and computation are needed during the detection process.The theoreticalanalysis and simulation results show that the algorithm can effectively outstand the anomaly series dataand efficiently reduce the anomaly error rate by the good anti-interference ability in practical engineer-ing.
For industrial control,the hardware storage space and computing performance are very limited,it is unable to run the abnormal series data detection algorithms and its common algorithms are morecomplex.The paper proposes the learning algorithm based on following period mean value to outstand a-nomaly series data.Firstly,after series data preprocessing,abnormal locations of the series set is marked,so the algorithm can get the best period T and the threshold Dmax of period mean D-value;Secondly,thedetected series is divided into groups according to T and compute the D-value of the adjacent two periodmean values.If the D-value exceeds Dmax,an anomaly appears.After the algorithm has learned the pa-rameters,the less storage space and computation are needed during the detection process.The theoreticalanalysis and simulation results show that the algorithm can effectively outstand the anomaly series dataand efficiently reduce the anomaly error rate by the good anti-interference ability in practical engineer-ing.
引文
[1]T J LEE,J GOTTSCHLICH,N TATBUL,et al.Greenhouse:a zero-positive machine learning system for time-series anomaly detection[EB/OL].https://arxiv.org/pdf/1801.03168.pdf,2018-12-11.
[2]程艳云,张守超,杨杨.基于大数据的时间序列异常点检测研究[J].计算机技术与发展,2016,26(5):139-144.
[3]孙梅玉.基于距离和密度的时间序列异常检测方法研究[J].计算机工程与应用,2012,48(20):11-17.
[4]苏卫星,朱云龙,胡琨元,等.基于模型的过程工业时间序列异常值检测方法[J].仪器仪表学报,2012,33(9):2 080-2 087.
[5]张琪,胡宇鹏,嵇存,等.边缘计算应用:传感数据异常实时检测算法[J].计算机研究与发展,2018,55(3):524-536.
[6]杨建中,马驰飞,陈吉红,等.一种数控系统指令域序列异常数据检测方法:CN 105160181A[P].2015.
[7]戴国骏,张翔,张怀相,等.电动车窗防夹系统的建模与实现[J].汽车工程,2008,30(6):539-541.
[8]W S RA,H J LEE,B P JIN,et al.Practical pinch detection algorithm for smart automotive power window control systems[J].IEEE Transactions on Industrial Electronics,2008,55(3):1 376-1 384.
[9]刘建国,付恒,饶政玉,等.基于高斯滤波和近似积分的电动车窗防夹算法[J].汽车工程,2017(12):1 464-1 471.
[10]马伟泽,张申科,汪宏杰.采用霍尔传感器的汽车电动车窗防夹设计[J].汽车工程,2008,30(12):1 122-1 124.
[2]程艳云,张守超,杨杨.基于大数据的时间序列异常点检测研究[J].计算机技术与发展,2016,26(5):139-144.
[3]孙梅玉.基于距离和密度的时间序列异常检测方法研究[J].计算机工程与应用,2012,48(20):11-17.
[4]苏卫星,朱云龙,胡琨元,等.基于模型的过程工业时间序列异常值检测方法[J].仪器仪表学报,2012,33(9):2 080-2 087.
[5]张琪,胡宇鹏,嵇存,等.边缘计算应用:传感数据异常实时检测算法[J].计算机研究与发展,2018,55(3):524-536.
[6]杨建中,马驰飞,陈吉红,等.一种数控系统指令域序列异常数据检测方法:CN 105160181A[P].2015.
[7]戴国骏,张翔,张怀相,等.电动车窗防夹系统的建模与实现[J].汽车工程,2008,30(6):539-541.
[8]W S RA,H J LEE,B P JIN,et al.Practical pinch detection algorithm for smart automotive power window control systems[J].IEEE Transactions on Industrial Electronics,2008,55(3):1 376-1 384.
[9]刘建国,付恒,饶政玉,等.基于高斯滤波和近似积分的电动车窗防夹算法[J].汽车工程,2017(12):1 464-1 471.
[10]马伟泽,张申科,汪宏杰.采用霍尔传感器的汽车电动车窗防夹设计[J].汽车工程,2008,30(12):1 122-1 124.