WO2019214367A1 - 一种动态报警阈值设计和报警消除的方法与系统 - Google Patents

一种动态报警阈值设计和报警消除的方法与系统 Download PDF

Info

Publication number
WO2019214367A1
WO2019214367A1 PCT/CN2019/080501 CN2019080501W WO2019214367A1 WO 2019214367 A1 WO2019214367 A1 WO 2019214367A1 CN 2019080501 W CN2019080501 W CN 2019080501W WO 2019214367 A1 WO2019214367 A1 WO 2019214367A1
Authority
WO
WIPO (PCT)
Prior art keywords
alarm
steady
variables
variable
working area
Prior art date
Application number
PCT/CN2019/080501
Other languages
English (en)
French (fr)
Inventor
王建东
余彦
杨子江
王振
张超
周东华
Original Assignee
山东科技大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 山东科技大学 filed Critical 山东科技大学
Priority to KR1020197033568A priority Critical patent/KR102160202B1/ko
Publication of WO2019214367A1 publication Critical patent/WO2019214367A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G08SIGNALLING
    • G08BSIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B29/00Checking or monitoring of signalling or alarm systems; Prevention or correction of operating errors, e.g. preventing unauthorised operation
    • G08B29/18Prevention or correction of operating errors
    • G08B29/185Signal analysis techniques for reducing or preventing false alarms or for enhancing the reliability of the system
    • GPHYSICS
    • G08SIGNALLING
    • G08BSIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B29/00Checking or monitoring of signalling or alarm systems; Prevention or correction of operating errors, e.g. preventing unauthorised operation
    • G08B29/18Prevention or correction of operating errors
    • G08B29/20Calibration, including self-calibrating arrangements
    • G08B29/24Self-calibration, e.g. compensating for environmental drift or ageing of components
    • G08B29/26Self-calibration, e.g. compensating for environmental drift or ageing of components by updating and storing reference thresholds

Definitions

  • the invention relates to the field of automation technologies such as industrial alarm systems, and in particular to a method and system for dynamic alarm threshold design and alarm elimination.
  • the alarm system is an important part of modern industrial systems and plays a vital role in the safe and efficient operation of the production process. Therefore, the research of industrial alarm systems has received more and more attention from industry and academia.
  • industrial alarm systems due to the influence of external disturbances such as noise and the lack of consideration of the correlation between variables, industrial alarm systems generate a large number of false alarms and leak alarms. False alarms and leak alarms can seriously affect the performance of industrial alarm systems, which not only leads to useful The alarm information was overwhelmed, and the "wolf came" effect caused the operator to ignore the alarm information for a long time, and did not trust or even shut down the industrial alarm system.
  • the present invention proposes a method and system for dynamic alarm threshold design and alarm elimination.
  • the present invention provides a method for dynamic alarm threshold design and alarm elimination, which can effectively reduce the false alarm rate and false negative rate of an industrial alarm system, and at the same time guide the operator to react to eliminate the alarm that has occurred, so as to effectively Improve the accuracy of the alarm and prevent the alarm information from being ignored.
  • the present invention provides a system for dynamic alarm threshold design and alarm cancellation based on the above method to provide the achievability of the above method from a system or software aspect.
  • a method for dynamic alarm threshold design and alarm elimination includes:
  • the steady-state working area is established by using the normalized normal historical data, and the points within the steady-state working area are regarded as normal points, otherwise they are regarded as abnormal points;
  • the steady-state working area is represented as a series of hyperbola in the parallel coordinate system.
  • the abnormal point is represented as a polyline in the parallel coordinate system.
  • the positional relationship between the polyline and the hyperbola is used to determine the adjustment variable of the abnormal point, and the hyperbolic curve is utilized. Tangent to determine the adjustment range of the adjustment variable;
  • the adjustment information is continuously received until the variables of the abnormal points are adjusted to the corresponding adjustment range, and the alarm is eliminated.
  • the maximum value and the minimum value of each variable are calculated, the difference between the maximum value and the minimum value of each variable is calculated, and the value of each variable of each sample point is subtracted from the minimum value of each variable, and then divided by The difference between the corresponding maximum and minimum values, thereby normalizing the values of the individual variables of the sample points to between 0 and 1.
  • a clustering algorithm is used to select r samples from m normal historical data to construct a steady-state working area, and the ratio of the normal historical data to the sample is set according to the set alarm false alarm rate.
  • the clustering algorithm includes but is not limited to a K-means clustering algorithm, a BIRCH algorithm, a CURE algorithm, an MDS_CLUSTER, a DBSCAN algorithm, an OPTICS algorithm, a DENCLUE algorithm, and the like.
  • a k-means clustering algorithm may be selected, and the specific processes include:
  • X(t) is the point on the surface of the super ellipsoid
  • C is the central vector of the super ellipsoid, ie the mean of each sample
  • P is the characteristic matrix of the super ellipsoid, ie the covariance matrix of each sample.
  • the sample point is regarded as a normal sample point, otherwise it is regarded as an abnormal sample point.
  • the projection on the surface of the hyperellipsoid steady-state working area is defined as the dynamic alarm threshold of the sample point.
  • each variable of the normal sample point when the value of each variable of the normal sample point is far from the corresponding dynamic alarm threshold, the sample point has a large adjustment margin, and the production process is relatively safe; otherwise, if some or some variable of the sample point The value is close to the corresponding dynamic alarm threshold, and an alarm may be generated next.
  • a dynamic alarm threshold corresponding to the intersection of the center of the steady-state working area and the intersection of the steady-state working area is defined as a dynamic alarm threshold of the abnormal sample point.
  • an abnormal sample point it is only necessary to adjust it to the intersection of the abnormal sample point and the center of the steady-state working area and the steady-state working area, so that the abnormal point can be returned to the steady state. Eliminate the alarm in the work area.
  • all variables are divided into two categories, an unmanipulatable variable and a manipulatable variable, and for a new abnormal sample point, the value of its uncontrollable variable is substituted.
  • the mathematical model of the steady-state working area the mathematical model of the dimensionality reduction space of the steady-state working area is obtained.
  • the mathematical model of the reduced-dimensional reduced space is represented in the parallel coordinate system, and the new abnormal sample points are parallelized.
  • the adjustment variable is determined by the representation in the coordinate system and the relative position of the dimensionality reduction space represented in the parallel coordinate system.
  • the adjustment variables are called adjustment variables, and these ranges are called adjustment adjustments corresponding to the adjustment variables.
  • Range the adjustment variable is judged by the representation of the corresponding sample point in the parallel coordinate system and the representation of the ng-dimensional hyperellipsoid in a parallel coordinate system, that is, the relative position of ng-1 to the hyperbola, wherein, for n For a variable alarm system, the first g variables are unmanipulatable variables, and the last ng variables are manipulated variables.
  • the invention also provides a system for dynamic alarm threshold design and alarm elimination, comprising:
  • a data standardization module for normalizing normal historical data such that values of respective variables are between 0 and 1;
  • a steady-state workspace building module for constructing a steady-state working area using normalized normal historical data to determine whether a new sample point is normal, and for calculating a dynamic alarm threshold and eliminating an alarm;
  • a dynamic alarm threshold calculation module for dynamically setting an alarm threshold for each new sample point using a projection along a surface of the steady-state working area along its respective coordinate axis direction;
  • the variable adjustment module is configured to determine an adjustment variable of the abnormal point and an adjustment range corresponding to the adjustment variable, and is used to guide the operator to perform a reaction cancellation alarm;
  • the alarm elimination module is configured to adjust the value of the adjustment variable of each abnormal point to the corresponding adjustment range, thereby eliminating the alarm.
  • the invention establishes a steady-state working area by using normal historical data, and regards a point within the steady-state working area as a normal sample point, otherwise it is regarded as an abnormal sample point.
  • a steady-state workspace is used to design a dynamic alarm threshold that reflects the operating state of the system and can help the operator react to an alarm that occurs.
  • the present invention can determine the adjustment range corresponding to the adjustment variable and the adjustment variable by visualizing the steady-state working area in a parallel coordinate system. By adjusting the value of the adjustment variable to their corresponding adjustment range, the alarm can be eliminated.
  • the present invention is beneficial for reducing false alarms and leak alarms in an industrial alarm system, so that the operator is not drowned in a large number of interference alarms, so that the operator can find a truly useful alarm in time.
  • the present invention is useful for helping production operators to properly respond to alarms that have occurred and to guide them in taking effective measures to eliminate the alarm.
  • 1 is a system flowchart of designing a dynamic alarm threshold and eliminating an alarm according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of calculation of a dynamic alarm threshold at a normal point
  • FIG. 3 is a schematic diagram of calculation of a dynamic alarm threshold of an abnormal point
  • Figure 4 is a schematic diagram of determining the adjustment variable
  • Figure 5 is a schematic diagram of calculation of the adjustment range
  • FIG. 6 is a time sequence diagram of a process variable according to an embodiment of the present invention.
  • FIG. 7 is a schematic diagram of a dynamic alarm threshold according to an embodiment of the present invention.
  • FIG. 8 is a schematic diagram of adjustment of a variable according to an embodiment of the present invention.
  • FIG. 9 is a system block diagram of designing a dynamic alarm threshold and eliminating an alarm according to an embodiment of the present invention.
  • orientation or positional relationship is based on the orientation or positional relationship shown in the drawings, and is merely a relative relationship for the purpose of describing the structural relationship of the components or components of the present invention, and is not specifically referring to any component or component of the present invention, and may not be construed as a Limitations of the invention.
  • an exemplary embodiment of the present invention is:
  • a method of designing dynamic alarm thresholds and eliminating alarms including the following steps:
  • Step S1 normalizing the normal historical data, so that the value of each variable falls between 0 and 1;
  • Step S2 using the normalized normal historical data to establish a steady-state working area, the point within the steady-state working area is regarded as a normal point, otherwise it is regarded as an abnormal point;
  • Step S3 designing a dynamic alarm threshold for each new sample point
  • Step S4 using a parallel coordinate system to determine an adjustment variable, and using a tangent to determine an adjustment range of the adjustment variable;
  • step S5 each adjustment variable is adjusted to their corresponding adjustment range, thereby eliminating the alarm.
  • step S2 r data is selected from m normal historical data to construct a steady-state working area, where m is the total number of samples.
  • the r-samples from the m normal historical data to construct the steady-state working area are realized by the k-means clustering algorithm.
  • the k-means clustering algorithm includes the following steps:
  • a super ellipsoid is used as a mathematical model of the steady-state working area, and the equation of the super ellipsoid is:
  • step S2 if the new sample point is in the super ellipsoid, the sample point is normal, otherwise it is abnormal. That is, for the new sample point X(t), if:
  • C and P represent the center vector and the feature matrix of the super ellipsoid, respectively.
  • the dynamic alarm threshold corresponding to the normal point X(t) is Where l i (t) represents the dynamic low alarm threshold and h i (t) represents the dynamic high alarm threshold.
  • the dynamic alarm threshold corresponding to the intersection of the steady-state workspace center and the steady-state working area X'(t) is defined as the dynamic alarm threshold of X(t).
  • the blue ellipse represents the boundary of the steady-state working area
  • the red asterisk indicates the abnormal point X(t)
  • the green dotted line L indicates the center of the steady-state working area and the sample point X(t).
  • the red five-pointed star X'(t) represents the intersection of the line segment L and the steady-state working area.
  • the dynamic alarm threshold for point X'(t) is defined as the dynamic alarm threshold for the abnormal sample point X(t).
  • Point X'(t) is located on the steady-state working area, which is a normal sample point. Equation (1) can be used to calculate its dynamic alarm threshold.
  • Equation (6) represents a ng-dimensional hyperellipsoid, which we project onto ng-1 two-dimensional planes, ie x g+1 x g+2 planes, x g+2 x g+3 planes,..., x n-1 x n plane, get ng-1 ellipse. These ng-1 ellipses are then mapped into a parallel coordinate system to obtain a ng-1 pair hyperbola. The ng-1 pairs of hyperbolas are sequentially connected in the order of the coordinate axes, and the representation of the ng-dimensional hyperellipsoid shown in the formula (6) in the parallel coordinate system is obtained.
  • the projection of the ng-dimensional hyperellipsoid in the x i x i+1 plane is an ellipse
  • the mathematical model of the projected ellipse is:
  • the ellipse in the x i x i+1 plane shown in the formula (8) corresponds to the hyperbola in the parallel coordinate system, and the mathematical model of the hyperbola in the xy plane is:
  • the adjustment variable is judged by the representation of the sample point X(t) in the parallel coordinate system and the representation of the n-g dimension hyperellipsoid in the parallel coordinate system, that is, the relative position of the n-g-1 to the hyperbola. As shown in Figure 4(a), if
  • variable x i and the variable x i+1 are adjustment variables; as shown in Figure 4(b), if
  • variable x i is the adjustment variable. As shown in Figure 4(c), if
  • variable x i+1 is the adjustment variable. As shown in Figure 4(d), if
  • variable x i is an adjustment variable
  • variable x i+1 is an adjustment variable, and which one is an adjustment variable is determined by the operator.
  • the adjustment range of the adjustment variable can be judged by the tangent of the hyperbola. As shown in FIG. 5, if the variable x i+1 is an adjustment variable, and the variable x i is not an adjustment variable, then the adjustment range of the adjustment variable x i+1 is among them:
  • variable x i+1 is an adjustment variable and the variable x i is also an adjustment variable
  • the adjustment range of the adjustment variable x i+1 is among them:
  • x' i (t) [r i l (t) + r i h (t)] /2, where r i l (t) and r i h (t) are the adjustment variables, respectively The upper and lower limits of the adjustment range of x i .
  • another embodiment of the present invention is a system for designing a dynamic alarm threshold and eliminating an alarm, including:
  • a data standardization module for normalizing normal historical data such that values of respective variables are between 0 and 1;
  • a steady-state workspace building module for constructing a steady-state working area using normalized normal historical data to determine whether a new sample point is normal, and for calculating a dynamic alarm threshold and eliminating an alarm;
  • a dynamic alarm threshold calculation module for calculating a dynamic alarm threshold of a new sample point
  • variable adjustment module is used to determine the adjustment range corresponding to the adjustment variable and the adjustment variable, and is useful for guiding the operator to respond to the alarm;
  • the alarm elimination module is used to adjust the value of each adjustment variable to their corresponding adjustment range, thereby eliminating the alarm.
  • the alarm system consists of six process variables, namely the actual load x 1 of the unit, the main steam pressure x 2 , the main steam flow x 3 , the main feed water flow x 4 , the total air volume x 5 and the total coal supply x 6 .
  • Figure 6 is a time series diagram of these six process variables.
  • X(t) [0.60, 0.30, 0.27, 0.72, 0.90, 0.40]
  • its dynamic alarm threshold is shown in Figure 7, which is an anomaly.
  • the variables x 4 , x 5 and x 6 are adjustment variables whose adjustment ranges are [0.58, 0.66], [0.45, 0.85] and [0.50, 0.90], respectively.
  • the value of the adjustment variable is adjusted to the middle of their adjustment range by default. Adjusting the values of the adjustment variables x 4 , x 5 and x 6 to 0.62, 0.65 and 0.70 allows the anomaly point X(t) to return to the steady-state working area, thereby eliminating the alarm.

Abstract

一种动态报警阈值设计和报警消除的方法与系统,通过根据各个样本点的不同动态设置相应的报警阈值,并确定相应异常点的调整变量,能够减少工业报警系统中的误报警和漏报警,使得操作人员不至于淹没于大量的干扰报警中,方便操作人员及时发现真正有用的报警,并且有益于帮助操作人员对已经发生的报警做出反应,指导他们如何消除报警。

Description

一种动态报警阈值设计和报警消除的方法与系统 技术领域
本发明涉及工业报警系统等自动化技术领域,尤其涉及一种动态报警阈值设计和报警消除的方法与系统。
背景技术
报警系统是现代工业系统的重要组成部分,对生产过程的安全高效运行有着至关重要的作用。因此,工业报警系统的研究受到了工业界和学术界越来越广泛的关注。然而由于噪声等外部干扰的影响,以及没有考虑变量之间的相关性,工业报警系统会产生大量的误报警和漏报警,误报警和漏报警会严重影响工业报警系统的性能表现,不仅导致有用的报警信息被淹没,更因“狼来了”效应造成操作人员长期忽视报警信息,不信任甚至关闭工业报警系统。
因此,如何提高报警的准确度和防止报警信息被忽视、被遗漏,是目前急需解决的问题。
发明内容
本发明为了解决上述问题,提出了一种动态报警阈值设计和报警消除的方法与系统。
首先,本发明提供一种动态报警阈值设计和报警消除的方法,该方法可以有效降低工业报警系统的误报率和漏报率,同时指导操作人员做出反应来消除已经发生的报警,以有效提高报警的准确度,并防止报警信息被忽视。
其次,本发明基于上述方法提供一种动态报警阈值设计和报警消除的系统, 以从系统或软件方面提供上述方法的可实现性。
为了实现上述目的,本发明采用如下技术方案:
一种动态报警阈值设计和报警消除的方法,包括:
将正常历史数据标准化,使得各个变量的取值落在0到1之间;
利用标准化后的正常历史数据建立稳态工作区,稳态工作区之内的点被视为正常点,否则被视为异常点;
为每个新来的样本点,利用沿着其各个坐标轴方向在稳态工作区表面上的投影动态设置报警阈值;
稳态工作区在平行坐标系中被表示为一系列的双曲线,异常点在平行坐标系中被表示为折线,利用折线与双曲线的位置关系来确定异常点的调整变量,利用双曲线的切线来确定调整变量的调整范围;
不断接收调整信息,直至各个异常点的变量调整至对应的调整范围内,消除报警。
进一步的,计算各个变量的最大值和最小值,计算每个变量最大值与最小值之间的差值,将每个样本点的各个变量的取值减去各个变量的最小值,然后除以对应的最大值和最小值的差值,从而将样本点的各个变量的取值标准化到0到1之间。
进一步的,利用聚类算法从m个正常历史数据中选取r个样本来构建稳态工作区,且正常历史数据与样本的比值根据设定的报警误报率来设置。
进一步的,所述聚类算法包括但不限于K均值聚类算法、BIRCH算法、CURE算法、MDS_CLUSTER、DBSCAN算法、OPTICS算法、DENCLUE算法等。
作为一种优选方式,可以选择k均值聚类算法,具体过程包括:
(1)从m个正常历史数据中随机选取r个样本,计算这r个样本的均值μ r和协方差矩阵W r
(2)计算每个样本点X(t)到样本中心的马氏距离d(t),其中d(t)的计算公式为:
Figure PCTCN2019080501-appb-000001
(3)将每个样本点的马氏距离d(t)按照从小到大的顺序排列,选出距离较小的r个样本,然后计算这r个样本的均值μ' r和协方差矩阵W' r
(4)如果协方差矩阵W r的行列式等于协方差矩阵W' r的行列式,聚类算法就终止。否则将μ' r和W' r的值分别赋给μ r和W r,聚类算法跳转到步骤(2)。
进一步的,利用超椭球体作为稳态工作区的数学模型,超椭球体的方程为
[X(t)-C] TP[X(t)-C]=1
其中,X(t)为超椭球体表面上的点,C为超椭球体的中心向量,即各个样本的均值,P为超椭球体的特征矩阵,即各个样本的协方差矩阵。
进一步的,如果新来的样本点在超椭球体内,该样本点就被视为正常的样本点,否则就被视为异常的样本点。
进一步的,对于一个正常的样本点,沿着各个坐标轴方向,在超椭球体稳态工作区表面上的投影被定义为该样本点的动态报警阈值。
进一步的,当正常的样本点的各个变量的取值远离对应的动态报警阈值,该样本点就有较大的调整裕度,生产过程较为安全;否则,如果样本点的某个或者某些变量的取值靠近对应的动态报警阈值,接下来可能会有报警产生。
进一步的,对于异常的样本点,其与稳态工作区的中心的连线与稳态工作区的交点所对应的动态报警阈值被定义为该异常样本点的动态报警阈值。
进一步的,对于异常的样本点,只需要将其调整到所述异常的样本点和稳态工作区的中心的连线与稳态工作区的交点处,就能让该异常点回到稳态工作区内,从而消除报警。
进一步的,利用平行坐标系来确定调整变量的过程中,将所有变量分为两大类,不可操纵变量和可操纵变量,对于一个新来的异常样本点,将它的不可操纵变量的值代入到稳态工作区的数学模型中,得到稳态工作区的降维缩减空间的数学模型;同时将降维缩减空间的数学模型在平行坐标系中表示出来,利用新来的异常样本点在平行坐标系中的表示与降维缩减空间在平行坐标系中表示的相对位置来确定调整变量。
进一步的,如果调整某些变量的值到某个范围内可以使得异常的样本点回到稳态工作区内,这些变量就被称为调整变量,并且这些范围就被称为调整变量对应的调整范围;调整变量通过对应的样本点在平行坐标系中的表示与n-g维的超椭球体在平行坐标系中的表示,也就是n-g-1对双曲线的相对位置来判断,其中,对于含有n个变量的报警系统,前g个变量是不可操纵变量,后n-g个变量是可操纵变量。
本发明还提供一种动态报警阈值设计和报警消除的系统,包括:
数据标准化模块,用于将正常历史数据标准化,使得各个变量的取值位于0到1之间;
稳态工作区构建模块,用于利用标准化后的正常历史数据构建稳态工作区,从而用来判断新来的样本点是否是正常的,以及用来计算动态报警阈值和消除报警;
动态报警阈值计算模块,用于为每个新来的样本点,利用沿着其各个坐标轴方向在稳态工作区表面上的投影动态设置报警阈值;
变量调整模块,用来确定异常点的调整变量和调整变量对应的调整范围,用来指导操作人员做出反应消除报警;
报警消除模块,用于将各个异常点的调整变量的值调整到对应的调整范围内,从而消除报警。
与现有技术相比,本发明的有益效果为:
本发明利用正常的历史数据建立稳态工作区,将稳态工作区之内的点视为正常的样本点,否则视为异常的样本点。对于每个新来的样本点,利用稳态工作区来设计动态报警阈值,动态报警阈值可以反映系统的运行状态并且可以帮助操作人员对发生的报警做出反应。
本发明通过将稳态工作区可视化在平行坐标系中,可以确定调整变量和调整变量对应的调整范围。将调整变量的值调整到他们对应的调整范围内,就可以消除报警。
综上,本发明有益于减少工业报警系统中的误报警和漏报警,使得操作人员不至于淹没于大量的干扰报警中,方便操作人员及时发现真正有用的报警。并且本发明有益于帮助生产操作人员对已经发生的报警做出正确反应,指导他们如采取有效措施消除报警。
附图说明
构成本申请的一部分的说明书附图用来提供对本申请的进一步理解,本申请的示意性实施例及其说明用于解释本申请,并不构成对本申请的不当限定。
图1为本发明实施例所述设计动态报警阈值和消除报警的系统流程图;
图2为正常点的动态报警阈值的计算示意图;
图3为异常点的动态报警阈值的计算示意图;
图4为调整变量的确定示意图;
图5为调整范围的计算示意图;
图6为本发明实施例的过程变量的时间序列图;
图7为本发明实施例的动态报警阈值示意图;
图8为本发明实施例的变量的调整示意图;
图9为本发明实施例所述设计动态报警阈值和消除报警的系统框图。
具体实施方式:
下面结合附图与实施例对本发明作进一步说明。
应该指出,以下详细说明都是例示性的,旨在对本申请提供进一步的说明。除非另有指明,本文使用的所有技术和科学术语具有与本申请所属技术领域的普通技术人员通常理解的相同含义。
需要注意的是,这里所使用的术语仅是为了描述具体实施方式,而非意图限制根据本申请的示例性实施方式。如在这里所使用的,除非上下文另外明确指出,否则单数形式也意图包括复数形式,此外,还应当理解的是,当在本说明书中使用术语“包含”和/或“包括”时,其指明存在特征、步骤、操作、器件、组件和/或它们的组合。
在本发明中,术语如“上”、“下”、“左”、“右”、“前”、“后”、 “竖直”、“水平”、“侧”、“底”等指示的方位或位置关系为基于附图所示的方位或位置关系,只是为了便于叙述本发明各部件或元件结构关系而确定的关系词,并非特指本发明中任一部件或元件,不能理解为对本发明的限制。
本发明中,术语如“固接”、“相连”、“连接”等应做广义理解,表示可以是固定连接,也可以是一体地连接或可拆卸连接;可以是直接相连,也可以通过中间媒介间接相连。对于本领域的相关科研或技术人员,可以根据具体情况确定上述术语在本发明中的具体含义,不能理解为对本发明的限制。
如图1所示,本发明的一种典型实施例是:
一种设计动态报警阈值和消除报警的方法,包括以下步骤:
步骤S1,将正常历史数据标准化,使得各个变量的取值落在0到1之间;
步骤S2,利用标准化后的正常历史数据建立稳态工作区,稳态工作区之内的点被视为正常点,否则被视为异常点;
步骤S3,为每个新来的样本点设计动态报警阈值;
步骤S4,利用平行坐标系来确定调整变量,利用切线来确定调整变量的调整范围;
步骤S5,将各个调整变量调整到他们对应的调整范围内,从而消除报警。
步骤S1中:计算n维的正常历史数据的每一维的最小值
Figure PCTCN2019080501-appb-000002
和最大值
Figure PCTCN2019080501-appb-000003
其中i=1,2,…,n。对于每一个样本点
Figure PCTCN2019080501-appb-000004
利用公式:
Figure PCTCN2019080501-appb-000005
将该样本点标准化,使它的各个变量的取值位于0到1之间。
步骤S2中:从m个正常历史数据中选取r个数据来构建稳态工作区,其中m为样本总数。稳态工作区包含的样本数r根据误报率来选取,如果报警系统的误报率为σ,那么r=m(1-σ)。
这是因为远离样本中心的样本点分布的很稀疏,如果稳态工作区要包含所有样本的话,稳态工作区就会过大,从而导致大量的漏报警。反之,如果稳态工作区包含的样本很少的话,稳态工作区就会过小,从而导致大量的误报警。
从m个正常历史数据中选取r个样本来构建稳态工作区是通过k均值聚类算法来实现。
进一步的,所述k均值聚类算法包括以下步骤:
(1)从m个正常历史数据中随机选取r个样本,计算这r个样本的均值μ r和协方差矩阵W r
(2)计算每个样本点X(t)到样本中心的马氏距离d(t),其中d(t)的计算公式为:
Figure PCTCN2019080501-appb-000006
(3)将每个样本点的马氏距离d(t)按照从小到大的顺序排列,选出距离较小的r个样本,然后计算这r个样本的均值μ' r和协方差矩阵W' r
(4)如果协方差矩阵W r的行列式等于协方差矩阵W' r的行列式,聚类算法就终止。否则将μ' r和W' r的值分别赋给μ r和W r,聚类算法跳转到步骤(2)。
进一步的,所述步骤S2中用超椭球体作为稳态工作区的数学模型,超椭球体的方程为:
[X(t)-C] TP[X(t)-C]=1
其中,X(t)为超椭球体表面上的点,C为超椭球体的中心向量,P为超椭球体的特征矩阵。在实际应用过程中,中心向量C和特征矩阵P用样本均值和协方差矩阵来代替。也就是说,C=μ r和P=W r
进一步的,步骤S2中,如果新来的样本点在超椭球体内,该样本点就是正常的,否则就是异常的。也就是说,对于新来的样本点X(t),如果:
[X(t)-C] TP[X(t)-C]≤1。样本点X(t)就是正常的,否则就是异常的。
S3中:对于一个正常的样本点X(t),它沿着各个坐标轴方向,在稳态工作区表面上的投影被定义为它的动态报警阈值。如图2所示,蓝色的椭圆表示稳态工作区的边界,红色的星号表示一个正常的样本点X(t),红色的圆圈表示样本点在稳态工作区上的投影。那么变量x 1的动态报警阈值为-0.8和0.8,变量x 2的动态报警阈值为-0.56和0.56。
对于正常的样本点X(t)=[x 1(t),x 2(t),…,x n(t)],它沿着x i轴方向在稳态工作区表面上的投影分别为:
Figure PCTCN2019080501-appb-000007
其中,
Figure PCTCN2019080501-appb-000008
公式(2)中,U(t)和V(t)的计算公式为:
Figure PCTCN2019080501-appb-000009
Figure PCTCN2019080501-appb-000010
Figure PCTCN2019080501-appb-000011
的计算公式为:
Figure PCTCN2019080501-appb-000012
在公式(3)中,
Figure PCTCN2019080501-appb-000013
表示由矩阵
Figure PCTCN2019080501-appb-000014
的第i 1,i 1+1,…,i 2行,第j 1,j 1+1,…,j 2列的元素组成的矩阵。其中矩阵
Figure PCTCN2019080501-appb-000015
的计算公式为:
Figure PCTCN2019080501-appb-000016
公式(4)中,C和P分别表示超椭球体的中心向量和特征矩阵。
因此,正常点X(t)对应的动态报警阈值为
Figure PCTCN2019080501-appb-000017
其中l i(t)表示动态低报警阈值,h i(t)表示动态高报警阈值。
对于一个异常的样本点X(t),它和稳态工作区中心的连线与稳态工作区的交点X'(t)对应的动态报警阈值被定义为X(t)的动态报警阈值。如图3所示,蓝色的椭圆表示稳态工作区的边界,红色的星号表示异常点X(t),绿色的点画线L表示稳态工作区中心和样本点X(t)的连线,红色的五角星X'(t)表示线段L和稳态工作区的交点。点X'(t)的动态报警阈值被定义为异常的样本点X(t)的动态报警阈值。点X'(t)位于稳态工作区上,它是一个正常的样本点,可以用公式(1)来计算它的动态报警阈值。
对于一个异常点X(t),它对应的X'(t)的计算公式为:
X'(t)=θX(t)+(1-θ)C,0≤θ≤1        (5)
在公式(4)中,θ的计算公式为:
Figure PCTCN2019080501-appb-000018
S4中:将所有的变量分为两大类,不可操纵变量和可操纵变量。对于一个新来的异常样本点X(t),将它的不可操纵变量的值代入到稳态工作区的数学模型中,得到降维缩减空间的数学模型。如果样本点在稳态工作区内,它也应该在降维缩减空间内。因此,将异常的样本点调整到稳态工作区内等价于调整可操纵变量的值将该样本点调整到降维缩减空间内。在本发明中,平行坐标系被用来将异常点调整到它的降维缩减空间内。
进一步的,对于含有n个变量的报警系统,如果它的前g个变量是不可操纵变量,后n-g个变量是可操纵变量,对于异常点X(t),它的降维缩减空间的数学模型为:
Figure PCTCN2019080501-appb-000019
在公式(6)中,
Figure PCTCN2019080501-appb-000020
以及
Figure PCTCN2019080501-appb-000021
在公式(7)中,
Figure PCTCN2019080501-appb-000022
以及
W(t)=[x 1(t),x 2(t),…,x g(t)]
公式(6)表示一个n-g维的超椭球体,我们将其投影到n-g-1个二维平面,也就是x g+1x g+2平面,x g+2x g+3平面,…,x n-1x n平面,得到n-g-1个椭圆。然后将这n-g-1个椭圆映射到平行坐标系中,得到n-g-1对双曲线。将这n-g-1对双曲线按照坐标轴顺序依次连接起来,就得到了公式(6)中所示的n-g维的超椭球体在平行坐标系中的表示。
进一步的,n-g维超椭球体在x ix i+1平面内的投影为椭圆,并且投影得到的椭圆的数学模型为:
Figure PCTCN2019080501-appb-000023
其中
Figure PCTCN2019080501-appb-000024
以及
Figure PCTCN2019080501-appb-000025
进一步的,公式(8)中所示的x ix i+1平面内的椭圆对应平行坐标系中的双曲线,该双曲线在xy平面内的数学模型为:
Figure PCTCN2019080501-appb-000026
其中
Figure PCTCN2019080501-appb-000027
以及
Figure PCTCN2019080501-appb-000028
进一步的,调整变量通过样本点X(t)在平行坐标系中的表示与n-g维的超椭球体在平行坐标系中的表示,也就是n-g-1对双曲线的相对位置来判断。如图4(a)所示,如果
Figure PCTCN2019080501-appb-000029
那么变量x i和变量x i+1都是调整变量;如图4(b)所示,如果
Figure PCTCN2019080501-appb-000030
那么变量x i是调整变量。如图4(c)所示,如果
Figure PCTCN2019080501-appb-000031
那么变量x i+1是调整变量。如图4(d)所示,如果
Figure PCTCN2019080501-appb-000032
那么要么变量x i是调整变量,要么变量x i+1是调整变量,具体哪一个变量是调整变量由操作人员来确定。在公式(10),(11),(12)和(13)中
Figure PCTCN2019080501-appb-000033
进一步的,调整变量的调整范围可以通过双曲线的切线来判断。如图5所示,如果变量x i+1是调整变量,并且变量x i不是调整变量,那么调整变量x i+1的调整范围为
Figure PCTCN2019080501-appb-000034
其中:
Figure PCTCN2019080501-appb-000035
在公式(14)中
Figure PCTCN2019080501-appb-000036
以及
Figure PCTCN2019080501-appb-000037
如果变量x i+1是调整变量,并且变量x i也是调整变量,那么调整变量x i+1的调整范围为
Figure PCTCN2019080501-appb-000038
其中:
Figure PCTCN2019080501-appb-000039
在公式(15)中,x' i(t)=[r i l(t)+r i h(t)]/2,其中r i l(t)和r i h(t)分别为调整变量x i的调整范围的上下限。
S5中:利用公式(10),(11),(12)和(13)来确定调整变量,然后利用公式(14)和(15)计算各个调整变量的调整范围,将各个调整变量的值调整到他们对应的调整范围内,就可以将异常的样本点调整到稳态工作区内,从而消除报警。
如图9所示:本发明的再一实施例是一种设计动态报警阈值和消除报警的系统,包括:
数据标准化模块,用于将正常历史数据标准化,使得各个变量的取值位于0到1之间;
稳态工作区构建模块,用于利用标准化后的正常历史数据构建稳态工作区,从而用来判断新来的样本点是否是正常的,以及用来计算动态报警阈值和消除报警;
动态报警阈值计算模块,用于计算新来的样本点的动态报警阈值;
变量调整模块,用来确定调整变量和调整变量对应的调整范围,有益于指导操作人员做出反应消除报警;
报警消除模块,用于将各个调整变量的值调整到他们对应的调整范围内,从而消除报警。
一个实例被用来具体说明如何设计动态报警阈值和消除报警。报警系统包含6个过程变量,分别是机组实际负荷x 1,主蒸汽压力x 2,主蒸汽流量x 3,主给水流量x 4,总风量x 5和总给煤量x 6。图6是这6个过程变量的时间序列图。对于一个新来的样本点X(t)=[0.60,0.30,0.27,0.72,0.90,0.40],它的动态报警阈值如图7所示,它是一个异常点。如图8所示,变量x 4,x 5和x 6是调整变量,它们的调整范围分别是[0.58,0.66],[0.45,0.85]和[0.50,0.90]。
消除报警时,默认将调整变量的值调整到它们的调整范围的中间值。将调整变量x 4,x 5和x 6的值调整到0.62,0.65和0.70,可以使异常点X(t)回到稳态工作区内,从而消除报警。
以上所述仅为本申请的优选实施例而已,并不用于限制本申请,对于本领域的技术人员来说,本申请可以有各种更改和变化。凡在本申请的精神和原则之内,所作的任何修改、等同替换、改进等,均应包含在本申请的保护范围之内。
上述虽然结合附图对本发明的具体实施方式进行了描述,但并非对本发明保护范围的限制,所属领域技术人员应该明白,在本发明的技术方案的基础上, 本领域技术人员不需要付出创造性劳动即可做出的各种修改或变形仍在本发明的保护范围以内。

Claims (10)

  1. 一种动态报警阈值设计和报警消除的方法,其特征是:包括:
    将正常历史数据标准化,使得各个变量的取值落在0到1之间;
    利用标准化后的正常历史数据建立稳态工作区,稳态工作区之内的点被视为正常点,否则被视为异常点;
    为每个新来的样本点,利用沿着其各个坐标轴方向在稳态工作区表面上的投影动态设置报警阈值;
    稳态工作区在平行坐标系中被表示为一系列的双曲线,异常点在平行坐标系中被表示为折线,利用折线与双曲线的位置关系来确定异常点的调整变量,利用双曲线的切线来确定调整变量的调整范围;
    不断接收调整信息,直至各个异常点的变量调整至对应的调整范围内,消除报警。
  2. 如权利要求1所述的一种动态报警阈值设计和报警消除的方法,其特征是:计算各个变量的最大值和最小值,计算每个变量最大值与最小值之间的差值,将每个样本点的各个变量的取值减去各个变量的最小值,然后除以对应的最大值和最小值的差值,从而将样本点的各个变量的取值标准化到0到1之间。
  3. 如权利要求1所述的一种动态报警阈值设计和报警消除的方法,其特征是:利用聚类算法从m个正常历史数据中选取r个样本来构建稳态工作区,且正常历史数据与样本的比值根据设定的报警误报率来设置。
  4. 如权利要求1所述的一种动态报警阈值设计和报警消除的方法,其特征是:利用超椭球体作为稳态工作区的数学模型,超椭球体的方程为:
    [X(t)-C] TP[X(t)-C]=1
    其中,X(t)为超椭球体表面上的点,C为超椭球体的中心向量,即各个样本的均值,P为超椭球体的特征矩阵,即各个样本的协方差矩阵;
    如果新来的样本点在超椭球体内,该样本点就被视为正常的样本点,否则就被视为异常的样本点。
  5. 如权利要求1所述的一种动态报警阈值设计和报警消除的方法,其特征是:对于一个正常的样本点,沿着各个坐标轴方向,在超椭球体稳态工作区表面上的投影被定义为该样本点的动态报警阈值。
  6. 如权利要求1所述的一种动态报警阈值设计和报警消除的方法,其特征是:对于异常的样本点,其与稳态工作区的中心的连线与稳态工作区的交点所对应的动态报警阈值被定义为该异常样本点的动态报警阈值。
  7. 如权利要求1所述的一种动态报警阈值设计和报警消除的方法,其特征是:对于异常的样本点,只需要将其调整到所述异常的样本点和稳态工作区的中心的连线与稳态工作区的交点处,就能让该异常点回到稳态工作区内,从而消除报警。
  8. 如权利要求1或7所述的一种动态报警阈值设计和报警消除的方法,其特征是:利用平行坐标系来确定调整变量的过程中,将所有变量分为两大类,不可操纵变量和可操纵变量,对于一个新来的异常样本点,将它的不可操纵变量的值代入到稳态工作区的数学模型中,得到稳态工作区的降维缩减空间的数学模型;同时将降维缩减空间的数学模型在平行坐标系中表示出来,利用新来的异常样本点在平行坐标系中的表示与降维缩减空间在平行坐标系中表示的相对位置来确定调整变量。
  9. 如权利要求1或7所述的一种动态报警阈值设计和报警消除的方法,其特征是:如果调整某些变量的值到某个范围内可以使得异常的样本点回到稳态工作区内,这些变量就被称为调整变量,并且这些范围就被称为调整变量对应的调整范围;调整变量通过对应的样本点在平行坐标系中的表示与n-g维的超椭球体在平行坐标系中的表示,也就是n-g-1对双曲线的相对位置来判断,其中,对于含有n个变量的报警系统,前g个变量是不可操纵变量,后n-g个变量是可操纵变量。
  10. 一种动态报警阈值设计和报警消除的系统,其特征是:包括:
    数据标准化模块,用于将正常历史数据标准化,使得各个变量的取值位于0到1之间;
    稳态工作区构建模块,用于利用标准化后的正常历史数据构建稳态工作区,从而用来判断新来的样本点是否是正常的,以及用来计算动态报警阈值和消除报警;
    动态报警阈值计算模块,用于对每个新来的样本点,利用沿着其各个坐标轴方向在稳态工作区表面上的投影动态设置报警阈值;
    变量调整模块,用来确定调整变量和调整变量对应的调整范围,用来指导操作人员做出反应消除报警;
    报警消除模块,用于将各个异常点的调整变量的值调整到对应的调整范围内,从而消除报警。
PCT/CN2019/080501 2018-05-07 2019-03-29 一种动态报警阈值设计和报警消除的方法与系统 WO2019214367A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
KR1020197033568A KR102160202B1 (ko) 2018-05-07 2019-03-29 동적 경보 임계치 설계와 경보 제거의 방법

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810427685.6A CN108615340B (zh) 2018-05-07 2018-05-07 一种动态报警阈值设计和报警消除的方法与系统
CN201810427685.6 2018-05-07

Publications (1)

Publication Number Publication Date
WO2019214367A1 true WO2019214367A1 (zh) 2019-11-14

Family

ID=63662330

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/080501 WO2019214367A1 (zh) 2018-05-07 2019-03-29 一种动态报警阈值设计和报警消除的方法与系统

Country Status (3)

Country Link
KR (1) KR102160202B1 (zh)
CN (1) CN108615340B (zh)
WO (1) WO2019214367A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113781758A (zh) * 2021-09-07 2021-12-10 浙江大学 面向高端燃煤发电装备的变量协同动态报警阈值优化方法
CN116206427A (zh) * 2023-05-06 2023-06-02 安徽智寰科技有限公司 一种基于通用指标自适应阈值的分级报警方法

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108615340B (zh) * 2018-05-07 2019-04-02 山东科技大学 一种动态报警阈值设计和报警消除的方法与系统
EP3874478A1 (en) * 2018-10-31 2021-09-08 Assa Abloy Ab Controlling operational state of a sensor device for break-in detection
CN109738368A (zh) * 2019-02-25 2019-05-10 浙江省检验检疫科学技术研究院 一种毒蛾危害检测的特征光谱指数及检测系统及方法
CN111024120B (zh) * 2019-11-25 2021-06-08 国网上海市电力公司 一种智能电力井盖倾角监测装置的检验方法
CN110703721B (zh) * 2019-11-27 2021-10-29 英特尔产品(成都)有限公司 用于监控生产过程的方法和装置

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160116343A1 (en) * 2014-10-24 2016-04-28 Google Inc. Adaptive threshold manipulation for movement detecting sensors
CN105718754A (zh) * 2016-03-09 2016-06-29 中国石油大学(北京) 一种炼化过程参数动态报警阈值的生成方法及装置
CN106368813A (zh) * 2016-08-30 2017-02-01 北京协同创新智能电网技术有限公司 一种基于多元时间序列的异常报警数据检测方法
CN106407082A (zh) * 2016-09-30 2017-02-15 国家电网公司 一种信息系统告警方法和装置
CN106778053A (zh) * 2017-03-31 2017-05-31 国网山东省电力公司电力科学研究院 一种基于相关性的报警关联变量检测方法及系统
CN107784799A (zh) * 2016-08-26 2018-03-09 北京协同创新智能电网技术有限公司 一种基于在线变化方向的多变量报警方法及系统
CN108615340A (zh) * 2018-05-07 2018-10-02 山东科技大学 一种动态报警阈值设计和报警消除的方法与系统

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN100340207C (zh) * 2006-01-20 2007-10-03 华南理工大学 一种智能化监护系统报警方法
JP4638939B2 (ja) * 2006-03-31 2011-02-23 富士通株式会社 閾値補正回路,回路および閾値補正機能付き回路基板
CN201438379U (zh) * 2009-09-01 2010-04-14 中广核工程有限公司 一种远程报警阈值修改系统
KR100982034B1 (ko) * 2010-05-11 2010-09-13 주식회사 이피아이솔루션즈 데이터베이스 성능 모니터링 방법 및 시스템
CN105933681B (zh) * 2016-07-08 2018-08-21 中国人民解放军国防科学技术大学 面向行为识别的灵敏度自适应调整方法
CN206725647U (zh) * 2017-04-12 2017-12-08 北京海博智恒电气防火科技有限公司 随动报警阈值剩余电流探测装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160116343A1 (en) * 2014-10-24 2016-04-28 Google Inc. Adaptive threshold manipulation for movement detecting sensors
CN105718754A (zh) * 2016-03-09 2016-06-29 中国石油大学(北京) 一种炼化过程参数动态报警阈值的生成方法及装置
CN107784799A (zh) * 2016-08-26 2018-03-09 北京协同创新智能电网技术有限公司 一种基于在线变化方向的多变量报警方法及系统
CN106368813A (zh) * 2016-08-30 2017-02-01 北京协同创新智能电网技术有限公司 一种基于多元时间序列的异常报警数据检测方法
CN106407082A (zh) * 2016-09-30 2017-02-15 国家电网公司 一种信息系统告警方法和装置
CN106778053A (zh) * 2017-03-31 2017-05-31 国网山东省电力公司电力科学研究院 一种基于相关性的报警关联变量检测方法及系统
CN108615340A (zh) * 2018-05-07 2018-10-02 山东科技大学 一种动态报警阈值设计和报警消除的方法与系统

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113781758A (zh) * 2021-09-07 2021-12-10 浙江大学 面向高端燃煤发电装备的变量协同动态报警阈值优化方法
CN116206427A (zh) * 2023-05-06 2023-06-02 安徽智寰科技有限公司 一种基于通用指标自适应阈值的分级报警方法
CN116206427B (zh) * 2023-05-06 2023-06-30 安徽智寰科技有限公司 一种基于通用指标自适应阈值的分级报警方法

Also Published As

Publication number Publication date
CN108615340B (zh) 2019-04-02
CN108615340A (zh) 2018-10-02
KR102160202B1 (ko) 2020-09-25
KR20200011416A (ko) 2020-02-03

Similar Documents

Publication Publication Date Title
WO2019214367A1 (zh) 一种动态报警阈值设计和报警消除的方法与系统
CN106368813B (zh) 一种基于多元时间序列的异常报警数据检测方法
WO2019001220A1 (zh) 一种基于决策树的数控车床车削过程能耗预测系统及方法
Zhou et al. Distributed data stream clustering: A fast em-based approach
Zhao et al. Bearing-based formation maneuvering
Wang et al. Tighten after relax: Minimax-optimal sparse PCA in polynomial time
Arif et al. A data mining approach for developing quality prediction model in multi-stage manufacturing
CN108153267B (zh) 一种基于误差主元分析模型的工业过程监测方法
Yu et al. Designing dynamic alarm limits and adjusting manipulated variables for multivariate systems
CN111522232B (zh) Mimo异因子全格式无模型控制方法
Gan et al. Abnormal network traffic detection based on improved LOF algorithm
WO2018077152A1 (zh) 一种主机变量异常检测方法及系统
Appa et al. On setting scale efficient targets in DEA
CN110045714B (zh) 一种基于权重近邻规则的工业过程监控方法
WO2019143889A8 (en) System and method for optimizing energy use of a structure using a clustering-based rule-mining approach
TW202016560A (zh) 擾動源定位方法
Prakash et al. MSC-clustering and forward stepwise regression for virtual metrology in highly correlated input spaces
CN117278425B (zh) 一种信息技术运维管理方法及系统
Killian et al. Pre-processing of Partition Data for Enhancement of LOLIMOT
KR102608018B1 (ko) 가상물리시스템의 이상치 탐지 장치 및 방법
KR102250062B1 (ko) 반도체 소자의 제조 방법 및 장치
CN115253356B (zh) 基于边缘计算技术的蒸馏塔控制系统、方法及蒸馏塔系统
Tandon et al. Nonlinear disturbance observer based robust control for continuous stirred tank reactor
CN108984863B (zh) 一种基于方向距离与超效率模型的图纸设计效率评估方法
Du et al. Matching method based on similarity of working trajectories

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 20197033568

Country of ref document: KR

Kind code of ref document: A

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19800807

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19800807

Country of ref document: EP

Kind code of ref document: A1