基于实时地质灾害监测数据的预警预报动态阈值分析方法

doi:10.16031/j.cnki.issn.1003-8035.202206009

基于实时地质灾害监测数据的预警预报动态阈值分析方法

doi: 10.16031/j.cnki.issn.1003-8035.202206009

Sichuan Institute of Geological Engineering Investigation Group Co.LTD, Chengdu,　610032

计量
- 文章访问数: 34
- HTML全文浏览量: 30
- PDF下载量: 17
- 被引次数: 0
出版历程
- 收稿日期: 2022-06-15
- 录用日期: 2022-08-17
- 修回日期: 2022-07-15
- 网络出版日期: 2023-05-12

Dynamic threshold analysis method of early warning and forecast based on real-time geological hazards monitoring data

摘要

摘要: 目前地质灾害监测领域对于监测预警信息的发布，主要基于各类监测设备的阈值设定。由于阈值是根据经验值或专家估值设定，缺乏地质灾害不同类型、不同环境的针对性，设定之后较长时间保持不变，如果进行阈值调整，也多是根据经验略微浮动，缺少数据样本分析的科学性。另外监测设备容易受到卫星信号、环境因素等影响，因此在实际运行中可能会出现误报、漏报的情况。为了解决上述问题，于是提出了一种预警阈值自学习自修正从而进行动态调整的方法，引入了两种可变阈值，并提出了一种基于优先级和门以及半马尔可夫过程的VTASs性能评估新方法。半马尔可夫过程的应用使该方法能够考虑具有非高斯分布的工业测量。此外，本文还提出了一种基于遗传算法的优化设计过程，用于优化参数设置，以提高性能指标。通过数值仿真以及与以往研究的比较，说明了该方法的有效性。将该方法在实测点位上进行应用，根据结果可知，相比于使用固定阈值，该方法能有效地减少系统误报漏报，提高地质灾害预警的准确性，从而更好的保护人民生命财产安全。
- 地质灾害 /
- 预警预报 /
- 动态阈值 /
- 自学习自修正
Abstract: Currently, in the field of geological disaster monitoring, monitoring and early warning information is mainly released based on the threshold setting of various types of monitoring equipment. However, since the thresholds are established according to empirical values or expert evaluations, there is a lack of pertinence to different types of geological disasters and different environments. Once set, these thresholds remain unchanged for a long time, and even when adjusted, they only slightly float based on experience, lacking scientific data sample analysis. Additionally, monitoring equipment is susceptible to satellite signals and environmental factors, leading to false alarms and false negatives during operation. To address these issues, a method of dynamic adjustment of self-learning and self-correction early warning thresholds is proposed. This method introduces two variable thresholds and a new performance evaluation method for VTASs based on priority and gate and semi-Markov processes. The application of the semi-Markov process allows the method to consider industrial measurements with non-Gaussian distributions. Moreover, an optimization design process based on genetic algorithms is proposed to improve performance indicators by optimizing parameter settings. Three numerical examples are used to illustrate the effectiveness of this approach and compare it with previous studies. When applied at the measured point, the method effectively reduce false alarms and under-alarms compared to the use of fixed thresholds, improving the accuracy of geological disaster early warning and better protecting the safety of people's lives and property.
- geological hazards /
- early warning and forecasting /
- dynamic threshold /
- Self-learning and self-correction

HTML全文

图 1 过程变量测量值的随机离散信号x(t)（采样数字-假设数据）

Figure 1. Random discrete signal x(t) (sampled digital-hypothesized data) of process variable measured values

下载: 全尺寸图片幻灯片

图 2 x(t)的正常、异常、假报警和错过报警部分分类（采样数字-假设数据）

Figure 2. Classification of normal, abnormal, false alarm and missed alarm parts of x(t) (sampled digital-hypothesized data)

下载: 全尺寸图片幻灯片

图 3 x(t)正常部分和异常部分的分离概率密度函数（q(x)-正常部分/p(x)-异常部分/Threshold-阈值）

Figure 3. Separation probability density function of x(t) normal part and abnormal part (q(x)-normal part/p(x)-abnormal part)

下载: 全尺寸图片幻灯片

图 4 生成的信号具有静态阈值和死区（时间（s）-测量x（t））

Figure 4. The generated signal has a static threshold and a deadband

下载: 全尺寸图片幻灯片

图 5 可变阈值、死区、正常和异常信号的估计概率密度函数（测量范围-概率）

Figure 5. Estimated probability density function for variable thresholds, deadbands, normal and abnormal signals

下载: 全尺寸图片幻灯片

图 6 具有自适应阈值和死区的生成信号（时间（s）-测量x（t））

Figure 6. Generated signal with adaptive threshold and deadband

下载: 全尺寸图片幻灯片

图 7 （a）优先级和门图解，（b）优先级和门FAR计算图解，（c）优先级和门MAR计算图解

Figure 7. (a) Priority and gate diagram, (b) priority and gate False Alarm Rate (FAR) calculation diagram, (c) priority and gate Miss Alarm Rate (MAR)calculation diagram

下载: 全尺寸图片幻灯片

图 8 （a）和（b）分别为带死区的优先级和门FAR和MAR计算图解

Figure 8. （a） and （b） respectively are FAR and MAR calculation diagrams for a priority and gate with deadband

下载: 全尺寸图片幻灯片

图 9 正态（红色）和异常（绿色）测量的概率密度函数（威布尔分布）（测量范围-概率）

Figure 9. Probability density function (Weibull distribution) of normal (red) and abnormal (green) measurements

下载: 全尺寸图片幻灯片

图 10 生成的随机数（时间（采样/秒）-x(t)）

Figure 10. The distribution of the generated random number over time

下载: 全尺寸图片幻灯片

图 11 使用具有自适应阈值的EWMA过滤器（alpha=0.5，窗口大小=100）后生成的随机数（时间（采样/秒）-x(t)）

Figure 11. Generated random numbers after filtering with an EWMA filter （alpha=0.5, window size =100） with adaptive threshold

下载: 全尺寸图片幻灯片

图 12 用自适应阈值（α=20）测量传感器（时间（s）-振动测量x(t) mm/s）

Figure 12. Measuring sensors with adaptive threshold（α= 20）

下载: 全尺寸图片幻灯片

图 13 传感器在正常和异常情况下的概率密度函数（测量范围-概率）

Figure 13. Probability density function of sensor under normal and abnormal conditions

下载: 全尺寸图片幻灯片

图 14 固定阈值预警阈值（数据量-位移监测值（mm））

Figure 14. Fixed threshold alert threshold

下载: 全尺寸图片幻灯片

图 15 滑动窗口长度（数据量-残差平方和）

Figure 15. Sliding window length

下载: 全尺寸图片幻灯片

图 16 动态阈值预警阈值（数据量-位移监测值（mm））

Figure 16. Dynamic alert threshold

下载: 全尺寸图片幻灯片

表 1 可变阈值报警系统的结果

Table 1. Results of variable threshold alarm system

方法	单纯阈	可变阈值	单纯阈和死区	可变阈值和死区
MAR	0.2036	0.1836	0.0452	0.0448
FAR	0.2668	0.2009	0.2668	0.2009

下载: 导出CSV

表 2 蒙特卡罗模拟与半马尔可夫解的比较

Table 2. Comparison between Monte Carlo simulation and semi Markov solution

方法	平均值	方差值	基于半马尔可夫的解
MAR	0.159856	1.34e−04	0.15867
FAR	0.159852	1.34e−04	0.15867

下载: 导出CSV

表 3 所提出的解、蒙特卡罗结果和简单马尔可夫解之间的比较

Table 3. Comparison between the proposed solution, Monte Carlo results and simple Markov solutions

方法	平均值	方差值	所提出的解	蒙特卡罗结果	简单马尔科夫解
MAR	0.0852	7.8019e−4	0.15867	0.1371	0.1070
FAR	0.1478	1.2643e−03	0.15867	0.1371	0.1455

下载: 导出CSV

表 4 可变阈值与其他方法性能指标的比较结果

Table 4. Comparison results of performance metrics of variable threshold method with other methods

方法	ST	VATS	EWMA	VTAS（+EWMA filter）	Evidence-based alarmsystem	3OMAF	3SADT
MAR	0.3519	0.3118	0.0437	0.0518	0.0582	0.2470	0.1511
FAR	0.3806	0.2860	0.0447	0.0241	0.0429	0.2926	0.2527

下载: 导出CSV

表 5 不同方法性能指标的比较结果

Table 5. Comparison results of different methodss’performance metrics

方法	单纯阈	可变阈值	单纯阈（带死区）
MAR	0.460713	0.207569	0.364654
FAR	0.192783	0.289527	0.228910

下载: 导出CSV

表 6 用遗传算法优化VTA

Table 6. Optimizing VTA with genetic algorithm

遗传算法中损失函数的权重			报警系统参数（n, m, α, w）	MAR	FAR	AAD
ω3	ω2	ω1	（1, 1, 25.0249, 82）	0.2109	0.2282	1.2674
1	1	10
RAAD	RFAR	RMAR
1	1e−05	1e−05
ω3	ω2	ω1	（2, 3, 25.039, 82）	0.0556	0.1681	2.8742
100	0.1	0.1
RAAD	RFAR	RMAR
1	1e−03	1e−03
ω3	ω2	ω1	（2, 3, 25.6803, 83）	0.0596	0.1625	2.8979
100	0.5	0.1
RAAD	RFAR	RMAR
1	1e−03	1e−03

下载: 导出CSV

表 7 预警次数对比

Table 7. Comparison of warning times

方法	预警次数
固定阈值	301
滑动窗口算法	62

下载: 导出CSV

表 8 预警性能指标对比

Table 8. Comparison of early warning performance indicators

方法	MAR	FAR
固定阈值	8.3%	13.7%
滑动窗口算法	0.7%	1.3%

下载: 导出CSV

参考文献(22)

[1]	Izadi S L, Shah D S, Shook S R, et al. A Framework for Optimal Design of Alarm Systems [C]. Proceedings of the 7th IFAC SAFEPROCESS, 2009. （in Chinese with English abstract）
[2]	NAGHOOSI E, IZADI I, CHEN T W. A study on the relation between alarm deadbands and optimal alarm limits[C]//Proceedings of the 2011 American Control Conference. June 29 - July 1, 2011, San Francisco, CA, USA. IEEE, 2011: 3627-3632.
[3]	Xu J. Performance Assessment and Design for Univariate Alarm Systems Based on FAR,MAR,and AAD[J]. IEEE Transactions on Automation Science and Engineering,2012,9(02):296 − 307. (in Chinese with English abstract doi: 10.1109/TASE.2011.2176490
[4]	Naghoosi E,Izadi I,Chen T. Estimation of Alarm Chattering[J]. Journal of Process Control,2011,21(09):1243 − 1249. (in Chinese with English abstract doi: 10.1016/j.jprocont.2011.07.015
[5]	Yue C, Izadi I, Tongwen COptimal Alarm Signal Processing: Filter Design and Performance Analysis[J]. J. Automation Science and Engineering, 2013, 10（02）: 446-451. （in Chinese with English abstract）
[6]	Adnan N A,Cheng Y,Izadi I,et al. Study of Generalized Delay-timers in Alarm Configuration[J]. Journal of Process Control,2013,23(03):382 − 395. (in Chinese with English abstract doi: 10.1016/j.jprocont.2012.12.013
[7]	ADNAN N A,IZADI I,CHEN T W. On expected detection delays for alarm systems with deadbands and delay-timers[J]. Journal of Process Control,2011,21(9):1318 − 1331. doi: 10.1016/j.jprocont.2011.06.019
[8]	YANG F, SHAH S L, XIAO D Y. Correlation analysis of alarm data and alarm limit design for industrial processes[C]//Proceedings of the 2010 American Control Conference. June 30 - July 2, 2010, Baltimore, MD, USA. IEEE, 2010: 5850 − 5855.
[9]	Sharma V, et al. A critical review on the application and problems caused by false alarms. Singapore; 2018, s. n. （in Chinese with English abstract）
[10]	AL-DABBAGH A W,HU Wenkai,LAI Shiqi,et al. Toward the advancement of decision support tools for industrial facilities:addressing operation metrics,visualization plots,and alarm floods[J]. IEEE Transactions on Automation Science and Engineering,2018,15:1883 − 1896. doi: 10.1109/TASE.2018.2827309
[11]	Quiñones-Grueiro M, Prieto-Moreno A, Verde C, Llanes-Santiago O. Data-driven monitoring of multimode continuous processes: A review. Chemometr Intell Lab Syst 2019;189（1）: 56–71. （in Chinese with English abstract）
[12]	Hollifield BR, Habibi E, Pinto J. Alarm management: A comprehensive guide. International Society of Automation （ISA）, 2021, s. l. （in Chinese with English abstract）
[13]	KONDAVEETI S R,SHAH S L,IZADI I. Application of multivariate statistics for efficient alarm generation[J]. IFAC Proceedings Volumes,2009,42(8):657 − 662. doi: 10.3182/20090630-4-ES-2003.00109
[14]	CHENG Y,IZADI I,CHEN T W. Optimal alarm signal processing:filter design and performance analysis[J]. IEEE Transactions on Automation Science and Engineering,2013,10(2):446 − 451. doi: 10.1109/TASE.2012.2233472
[15]	ZANG H,LI H G. Optimization of process alarm thresholds:a multidimensional kernel density estimation approach[J]. Process Safety Progress,2014,33(3):292 − 298. doi: 10.1002/prs.11658
[16]	HAN L,GAO H H,XU Y,et al. Combining FAP,MAP and correlation analysis for multivariate alarm thresholds optimization in industrial process[J]. Journal of Loss Prevention in the Process Industries,2016,40:471 − 478. doi: 10.1016/j.jlp.2016.01.022
[17]	Xu X, et al. The optimal design of industrial alarm systems based on evidence theory. Control Eng Pract 2019;46（1）: 142–56. （in Chinese with English abstract）
[18]	GAO Huihui, LIU Feifei, ZHU Qunxiong. A correlation consistency based multivariate alarm thresholds optimization approach[■■]. ●●●●●, 2016
[19]	BAHAR-GOGANI M, ASLANSEFAT K, SHOOREHDELI M A. A novel extended adaptive thresholding for industrial alarm systems[C]//2017 Iranian Conference on Electrical Engineering （ICEE）. May 2-4, 2017. Tehran, Iran. IEEE, 2017.
[20]	Liu X. J.,Xu H. B.,Dong Y. S.,et al. Mining frequent patterns of data stream closure based on sliding windows[J]. Computer Research and Development,2018,43(10):1738 − 1743. (in Chinese with English abstract
[21]	刘学军,徐宏炳,董逸生,等. 基于滑动窗口的数据流闭合频繁模式的挖掘[J]. 计算机研究与发展,2006,43(10):1738 − 1743. [LIU Xuejun,XU Hongbing,DONG Yisheng,et al. Mining frequent closed patterns of data streams based on sliding window[J]. Journal of Computer Research and Development,2006,43(10):1738 − 1743. (in Chinese) doi: 10.1360/crad20061010
[22]	戴慧,阚建飞,李伟仁,等. 多变量时间序列滑动窗口异常点的检测[J]. 南京信息工程大学学报(自然科学版),2014,6(6):515 − 519. [DAI Hui,KAN Jianfei,LI Weiren,etal,School of Computer Engineering,Nanjing Institute of Technology,Department of Electronic Engineering,University of Texas in Arlington],et al. Outlier detection for sliding window of multi-variable time series[J]. Journal of Nanjing University of Information Science & Technology (Natural Science Edition),2014,6(6):515 − 519. (in Chinese with English abstract)