CN106778053A - A kind of alert correlation based on correlation becomes quantity measuring method and system - Google Patents
A kind of alert correlation based on correlation becomes quantity measuring method and system Download PDFInfo
- Publication number
- CN106778053A CN106778053A CN201710206963.0A CN201710206963A CN106778053A CN 106778053 A CN106778053 A CN 106778053A CN 201710206963 A CN201710206963 A CN 201710206963A CN 106778053 A CN106778053 A CN 106778053A
- Authority
- CN
- China
- Prior art keywords
- correlation
- data
- time series
- variable
- point
- Prior art date
- Legal status (The legal status 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 status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
- G16Z99/00—Subject matter not provided for in other main groups of this subclass
Landscapes
- Testing And Monitoring For Control Systems (AREA)
- Alarm Systems (AREA)
Abstract
Become quantity measuring method and system the invention discloses a kind of alert correlation based on correlation, wherein, the method is by setting up binary time series to associated variable, binding time sequence segment method and coefficient correlation tendency method, obtain abnormal data section exactly from historical data fast automaticly, so as to carry out anomaly data detection, for the dynamic alert threshold design for realizing multivariable warning system provides favourable condition, so as to reduce interference alarm, the efficiency of site operation personnel's treatment alarm is improved, production security has been ensured.
Description
Technical field
The invention belongs to field of signal processing, more particularly to a kind of alert correlation based on correlation become quantity measuring method and
System.
Background technology
Safety in production of the warning system to ensureing Thermal generation unit plays vital effect with Effec-tive Function, by
The influencing each other between associated variable in actual industrial process, traditional single argument alarm threshold value method for designing is there may be big
Amount interference alarm (failing to report alert and false alarm) simultaneously causes the generation of " alarm is excessive " so that the notice of site operation personnel is subject to
Influence, increases the difficulty that correct disposal is made when abnormal production status occur.
Correlation often occurs substantially compared with nominal situation between there is abnormal variations per hour based on equipment in flowchart process or system
A large amount of cases of change are true, find a kind of detection method of the alert correlation variable based on correlation and are in come automatic filtering out
The detection method of normal condition and unusual condition data segment is very necessary.
The content of the invention
In order to solve the above problems, the first object of the present invention there is provided a kind of alert correlation variable based on correlation
Detection method.The method can exactly obtain abnormal data section from historical data fast automaticly, so as to carry out abnormal number
According to detection, for the dynamic alert threshold design for realizing multivariable warning system provides favourable condition, so that interference alarm is reduced,
The efficiency of site operation personnel's treatment alarm is improved, production security has been ensured.
Alert correlation based on correlation of the invention becomes quantity measuring method, and the method is complete in server or processor
Into it is specifically included:
Step 1:The alarm variable and multiple phases associated there of predetermined time period are extracted from history detection data
The data of variable are closed, selection one of which alarm variable and correlated variables are used as detection object;
Step 2:Judge the dynamic deferred relation between the alarm variable and correlated variables of selection, and then set up two elementary times
Sequence T is simultaneously standardized as T ';
Step 3:Under the constraint of the minimum interval caused by noise, binary time series T ' is segmented;
Step 4:Ask for the coefficient correlation and its correlative trend of each segmentation;
Step 5:According to correlative trend and actual trend comparison, abnormal data section and its relevant information are obtained.
Further, in the step 2, if there is dynamic deferred relation between alarm variable and correlated variables, will
The alarm variable of predetermined time period or the time span of correlated variables are translated and are kept alarm variable and correlated variables
Between to there is dynamic deferred relation constant;If not existing dynamic deferred relation between alarm variable and correlated variables, need not
Translation.
The present invention first determines whether the dynamic deferred relation alarmed between variable and correlated variables, and then establishes accurate two
Elementary time sequence T, and then abnormal data section can be exactly obtained, the efficiency of site operation personnel's treatment alarm is improve, protect
Production security is hindered.
Further, also included before the step 3:The minimum interval that calculating is caused by noise, its specific mistake
Journey includes:
(3.1.1) obtains each lower flex point of the alarm variable of current preset time span, and tries to achieve adjacent lower flex point
The distance between, and then constitute array d;
(3.1.2) is ranked up to array d and removes repeat element, obtains array d0;Ask for the slope variation of array d0
The distance between maximum point and its immediate lower flex point dm, dm are the minimum interval in alarm variable;
(3.1.3) obtains the minimum interval of correlated variables to correlated variables repeat step (3.1.1) and (3.1.2)
dh;
(3.1.4) using the higher value in dm and dh as binary time series T ' minimum interval.
Minimum interval is used to reduce influence of the noise to segmentation result, when actual industrial process data are processed, makes an uproar
The interference of sound can cause the time interval between key point too short, therefore, do not closed in the neighborhood region of minimum interval
Key point search.
Further, the process that binary time series T ' is segmented is included in the step 3:
Using binary time series T ' as data segment to be divided;
According to coefficient correlation between data in data segment to be divided, the data segment classification category belonging to data segment to be divided is judged
Property;
Wherein, according to preset correlation coefficient number scope, data segment categorical attribute includes weak related data section, middle related data section
With strong correlation data segment.
Wherein, dividing weak related data section and middle related data section can evade leakage segmentation phenomenon, and strong correlation data segment can be advised
Circumvent fitting phenomenon.
Further, the process being segmented to binary time series T ' in the step 3, also includes:
For data segment to be divided, using the method for linear interpolation, by the data point in binary time series T ' in its institute
Projection on category segmentation head and the tail data point line is used as match point;
Farthest point is found as the crucial turning point being segmented next time by the use of orthogonal distance, then determines to treat divided data section
With the presence or absence of strong correlation data segment, and update crucial turning point;
Repeat the above steps, until no longer there is data segment to be divided untill.
Under the constraint of minimum interval, just for weak related data section, time span is long and correlation for the present invention
Not visible data section, and time span is long and correlation significantly, but subdata section after dividing still is the number of strong correlation
Time series division is carried out according to section, the crucial turning point of data segment to be divided is obtained, piecewise linearity is carried out to original time series
Represent, so as to avoid overfitting and careless omission from being segmented.
Further, in the step 4, the coefficient correlation of each segmentation and its specific mistake of correlative trend are asked for
Journey, including:
(4.1):Time series is divided according to the final crucial turning point for obtaining, the time is calculated using formula of correlation coefficient
The coefficient correlation of each segmentation in sequence;
(4.2):Single side hypothsis inspection is carried out to correlation of variables, significance is set, checked according to single side hypothsis and tied
Correlation between fruit and significance confirmation variable, determines coefficient correlation trend.
The present invention calculates the coefficient correlation of each segmentation in time series using formula of correlation coefficient, then related to variable
Property carry out single side hypothsis inspection, significance is set, between confirming variable according to single side hypothsis assay and significance
Correlation, it is determined that coefficient correlation trend, for accurately obtain abnormal data section and its relevant information provide precise information, enter
And improve the efficiency of alarm.
The second object of the present invention is to provide a kind of alert correlation based on correlation and becomes amount detection systems.
A kind of alert correlation based on correlation of the invention becomes amount detection systems, including:
Data extraction module, its be used for from history detection data extract predetermined time period alarm variable and with its phase
The data of multiple correlated variables of association, selection one of which alarm variable and correlated variables are used as detection object;
Time series sets up module, and it is used to judge the dynamic deferred pass between the alarm variable of selection and correlated variables
System, and then set up binary time series T and be standardized as T ';
Time series segmentation module, it is used under the constraint of the minimum interval caused by noise, to two elementary times
Sequence T ' is segmented;
Coefficient correlation asks for module, its coefficient correlation and its correlative trend for being used to ask for each segmentation;
Abnormal data acquisition module, it is used for according to correlative trend and actual trend comparison, obtain abnormal data section and
Its relevant information.
In the time series sets up module, if there is dynamic deferred relation between alarm variable and correlated variables,
The time span of the alarm variable of predetermined time period or correlated variables is translated and is kept alarm variable to become to related
There is dynamic deferred relation between amount constant;If not existing dynamic deferred relation, nothing between alarm variable and correlated variables
Need translation.
The present invention first determines whether the dynamic deferred relation alarmed between variable and correlated variables, and then establishes accurate two
Elementary time sequence T, and then abnormal data section can be exactly obtained, the efficiency of site operation personnel's treatment alarm is improve, protect
Production security is hindered.
Further, the system also includes:Minimum interval computing module, it is used to obtain current preset time span
Alarm variable each lower flex point, and try to achieve the distance between adjacent lower flex point, and then constitute array d;
Array d is ranked up and removes repeat element, obtain array d0;Ask for the maximum point of the slope variation of array d0
And its distance between immediate adjacent lower flex point dm, dm are the minimum interval alarmed in variable;
The distance between lower flex point and adjacent lower flex point of correlated variables are obtained, and then obtains the minimum time of correlated variables
Interval dh;
Using the higher value in dm and dh as binary time series minimum interval;
Minimum interval is used to reduce influence of the noise to segmentation result, when actual industrial process data are processed, makes an uproar
The interference of sound can cause the time interval between key point too short, therefore, do not closed in the neighborhood region of minimum interval
Key point search.
Further, the time series segmentation module, including:Divided data section acquisition module is treated, when it is used for binary
Between sequence T ' as data segment to be divided;
According to coefficient correlation between data in data segment to be divided, the data segment classification category belonging to data segment to be divided is judged
Property;
Wherein, according to preset correlation coefficient number scope, data segment categorical attribute includes weak related data section, middle related data section
With strong correlation data segment.
Further, the time series segmentation module, also includes:
Match point asks for module, and it is used for for data segment to be divided, using the method for linear interpolation, after standardization
Data point in binary time series T ' is in the projection being segmented belonging to it on head and the tail data point line as match point;
Crucial turning point calculates update module, and it is used to find farthest point as being segmented next time by the use of orthogonal distance
Crucial turning point, then determine to treat to treat divided data section with the presence or absence of the third in divided data section, and crucial turning point is updated, until not
Untill there is data segment to be divided again.
Under the constraint of minimum interval, just for weak related data section, time span is long and correlation for the present invention
Not visible data section, and time span is long and correlation significantly, but subdata section after dividing still is the number of strong correlation
Time series division is carried out according to section, the crucial turning point of data segment to be divided is obtained, piecewise linearity is carried out to original time series
Represent, so as to avoid overfitting and careless omission from being segmented.
Further, the coefficient correlation asks for module, including:
Segmentation coefficient correlation computing module, it is used to divide time series according to the final crucial turning point for obtaining, utilizes
Formula of correlation coefficient calculates the coefficient correlation of each segmentation in time series;
Coefficient correlation trend determining module, it is used to carry out correlation of variables single side hypothsis inspection, sets conspicuousness water
Flat, the correlation between confirming variable according to single side hypothsis assay and significance determines coefficient correlation trend.
The present invention calculates the coefficient correlation of each segmentation in time series using formula of correlation coefficient, then related to variable
Property carry out single side hypothsis inspection, significance is set, between confirming variable according to single side hypothsis assay and significance
Correlation, it is determined that coefficient correlation trend, for accurately obtain abnormal data section and its relevant information provide precise information, enter
And improve the efficiency of alarm.
Compared with prior art, the beneficial effects of the invention are as follows:
The present invention choose industry variants between correlation as judge work dotted state whether exception feature, by right
Associated variable sets up multivariate time series, binding time sequence segment method and coefficient correlation tendency method, farthest reduces
Fitting phenomenon and leakage segmentation phenomenon, obtain abnormal data section, so as to carry out exception exactly from historical data fast automaticly
Data Detection provides favourable condition to realize the dynamic alert threshold design of multivariable warning system, so as to reduce interference report
It is alert, the efficiency of site operation personnel's treatment alarm is improved, ensure production security.
Brief description of the drawings
The Figure of description for constituting the part of the application is used for providing further understanding of the present application, and the application's shows
Meaning property embodiment and its illustrated for explaining the application, does not constitute the improper restriction to the application.
Fig. 1 is that the alert correlation based on correlation of the invention becomes quantity measuring method flow chart;
Fig. 2 (a) is the time series variation and segmentation result figure of air preheater inlet flue gas temperature of the invention;
Fig. 2 (b) is the time series variation and segmentation result figure that air preheater of the invention exports cigarette temperature;
Fig. 3 is each segmentation coefficient correlation and its confidential interval in the specific embodiment of the invention;
Fig. 4 is correlative trend of the variable in each segmentation in the specific embodiment of the invention;
Fig. 5 (a) is first group of multivariate time series scatter diagram of the invention and fitting a straight line;
Fig. 5 (b) is second group of multivariate time series scatter diagram of the invention and fitting a straight line;
Fig. 5 (c) is the 3rd group of multivariate time series scatter diagram and fitting a straight line of the invention;
Fig. 5 (d) is the 4th group of multivariate time series scatter diagram and fitting a straight line of the invention;
Fig. 6 is that the alert correlation based on correlation of the invention becomes amount detection systems structural representation;
Fig. 7 asks for modular structure schematic diagram for coefficient correlation of the invention.
Specific embodiment
It is noted that described further below is all exemplary, it is intended to provide further instruction to the application.Unless another
Indicate, all technologies used herein and scientific terminology are with usual with the application person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singulative
Be also intended to include plural form, additionally, it should be understood that, when in this manual use term "comprising" and/or " bag
Include " when, it indicates existing characteristics, step, operation, device, component and/or combinations thereof.
Fig. 1 is that the alert correlation based on correlation of the invention becomes quantity measuring method flow chart.
Alert correlation based on correlation as shown in Figure 1 becomes quantity measuring method, and the method is in server or processor
Complete, it is specifically included:
Step 1:The alarm variable and multiple phases associated there of predetermined time period are extracted from history detection data
Close the data of variable and select one group of correlated variables and alarm variable as detection object.
Specifically, time span is the initial data of multiple correlated variables of N before extracting present operating point, selects one group
Alarm variable and one group of correlated variables are detected.
As a example by concrete application scene with the method for the invention in specific example is as power plant:
Air preheater inlet flue gas temperature in selected power plant, used as alarm variable, air preheater outlet wind-warm syndrome is correlation
Variable.In a stopping accident in power plant, the sampling period is 1 second before shutdown is chosen from historical data, and sample size is N
=3600 data, one group of air preheater inlet flue gas temperature of selection and air preheater outlet wind-warm syndrome are used for anomaly data detection.
Step 2:Judge the dynamic deferred relation between alarm variable and correlated variables, and then set up binary time series T
And it is standardized as T ';
Specifically, dynamic deferred relation is whether there is between judgment variable, if it is present needing to obtain the two variables
One section of primary data ask for time delay h, then by existing length be the alarm variable or the time span of correlated variables of N
Translation h.If not existing dynamic deferred relation between variable, without translation;H is positive number;
Binary time series T is set up to having completed two groups of variables that are translation or being not required to be translated, and during by binary
Between sequence criteria turn to T '.
In specific implementation process, judge do not exist between air preheater inlet flue gas temperature and air preheater outlet wind-warm syndrome
Dynamic deferred relation, therefore time delay is 0.Then binary time series is built to two groups of data, and is standardized, standardized
Binary time series afterwards is designated as T '=[T1(t), T2(t)], wherein t=1 ..., N, N are positive integer;T1Represent air preheat
Device inlet flue gas temperature, T2Represent air preheater outlet wind-warm syndrome.
Step 3:Under the constraint of the minimum interval caused by noise, the binary time series T after standardization is entered
Row segmentation.
Wherein, also include before step 3:The minimum interval that calculating is caused by noise, its detailed process includes:
(3.1.1) obtains each lower flex point of the alarm variable of current preset time span, and tries to achieve adjacent lower flex point
The distance between, and then constitute array d={ d1, d2, d3 ... ..., dx };
(3.1.2) is ranked up to array d and removes repeat element, obtains array d0={ d1, d2, d3 ... ..., dx0 },
Wherein x0 < x, wherein, x0 and x are positive integer;Ask for array d0 the maximum point m of slope variation and its it is corresponding it is adjacent under
The distance between flex point dm, dm are the minimum interval in alarm variable;
(3.1.3) obtains the minimum interval of correlated variables to correlated variables repeat step (3.1.1) and (3.1.2)
dh;
(3.1.4) using the higher value in dm and dh as binary time series T ' minimum interval δ, 0 < δ < n, N
It is positive integer.
Wherein, in step (3.1.2), the method for asking for the point m of the slope variation maximum of array d0 can be used:Draw
With d0 as abscissa, x0 is the broken line graph of ordinate, and the maximum point m of slope variation is found in broken line graph.
It should be noted that the method for the maximum point m of the above-mentioned slope variation for asking for array d0 is only a kind of embodiment, also
Can be realized using other existing methods.
It is 40s to calculate and obtain the minimum interval δ caused by noise, and minimum interval is used for into crucial point search
In, by the calculating of coefficient correlation and the screening of data segment to be divided after, it is final to determine to obtain 5 key points, i.e., 4 points
Section.Therefore shown in time series segmentation figure such as Fig. 2 (a) and Fig. 2 (b).
Minimum interval is used to reduce influence of the noise to segmentation result, when actual industrial process data are processed, makes an uproar
The interference of sound can cause the time interval between key point too short, therefore, do not closed in the neighborhood region of minimum interval
Key point search.
Specifically, the process that binary time series T ' is segmented is included in step 3:
Using binary time series T ' as data segment to be divided;
According to coefficient correlation between data in data segment to be divided, the data segment classification category belonging to data segment to be divided is judged
Property;
Wherein, according to preset correlation coefficient number scope, data segment categorical attribute includes weak related data section, middle related data section
With strong correlation data segment.
For example:Under current dividing condition, the coefficient correlation between each segmentation is calculated, and then divided data section is treated in acquisition;
Wherein, treat divided data section have weak related data section, middle related data section and strong correlation data segment these three:
The first, meets coefficient correlation for 0.5 >=ρs>=0.3 weak related data section;
Second, meetAnd 0.9 > ρsThe < ρ of > 0.5 or 0.1sThe data segment of < 0.3;
The third, meets zs> N/3 and ρs>=0.9 or ρsAfter being divided, its subdata section is≤0.1 data segment
Strong correlation;Wherein, ρs≤ 0.1 is the coefficient correlation between s-th segmentation internal variable of the binary time series T ' after standardization;
zsIt is number of samples in s-th segmentation;N is the predetermined time period of binary time series T ';N and s is positive integer.
It should be noted that the present invention can also set other preset correlation coefficient number scopes, weak correlation is divided into data segment
Data segment, middle related data section and strong correlation data segment these three data segments.
S-th segmentation internal variable X of time series T 'iAnd XjBetween Spearman sample correlation coefficients be:
Wherein, dividing the first and second data segment can evade leakage segmentation phenomenon, and dividing the third data segment can evade
Over-fitting.
Further, the process being segmented to the binary time series T ' after standardization in step 3, also includes:
For data segment to be divided, using the method for linear interpolation, by the number in the binary time series T ' after standardization
Projection of the strong point on segmentation head and the tail data point line belonging to it is used as match point;
Farthest point is found as the crucial turning point being segmented next time by the use of orthogonal distance, then determines to treat divided data section
Divided data section is treated with the presence or absence of the third in the step (3.2.1), and updates crucial turning point;
Repeat the above steps, until no longer there is data segment to be divided untill.
Wherein, the parametric equation of space cathetus AB is represented by:
The coordinate of any point P0 is represented by straight line AB:
[(XiB-XiA)β+XiA, (tB-tA)β+tA]。
Wherein, X represents variable, and t represents time variable, i=1, and 2, alarm variable and correlated variables, A and B points are represented respectively
The two ends of straight line AB are not represented, and β represents a definite value.Therefore, the distance of point P to straight line AB can be defined as:
WhereinRefer to P points to the distance of straight line AB
Wherein,WhenTake pole
Corresponding parameter during small valueThe minimum of head and the tail line AB is then segmented belonging to data point P to its
Distance is that orthogonal distance isThe maximum of the D in each segmentation is the key of the segmentation
Turning point.
The present invention by taking Fig. 5 (a)-Fig. 5 (d) this four groups of multivariate time serieses as an example, using the method for linear interpolation, is incited somebody to action respectively
The data point in binary time series T after standardization is in the projection being segmented belonging to it on head and the tail data point line as fitting
Point, shown in the fitting a straight line for respectively obtaining, such as Fig. 5 (a)-Fig. 5 (d).
Under the constraint of minimum interval, just for weak related data section, time span is long and correlation for the present invention
Not visible data section, and time span is long and correlation significantly, but subdata section after dividing still is the number of strong correlation
Time series division is carried out according to section, the crucial turning point of data segment to be divided is obtained, piecewise linearity is carried out to original time series
Represent, so as to avoid overfitting and careless omission from being segmented.
Step 4:Ask for the coefficient correlation and its correlative trend of each segmentation.
Specifically, the coefficient correlation of each segmentation and its detailed process of correlative trend are asked for, including:
(4.1):Time series is divided according to the final crucial turning point for obtaining, the time is calculated using formula of correlation coefficient
The coefficient correlation of each segmentation in sequence;
According to the final crucial turnover point set P={ P for obtaining1, P2..., PkDivide variable time series, it is now many
The Piecewise Linear Representation of elementary time sequence T ' is:
TPLR=<f1[(Xi(p1), p1), (Xi(p2), p2)] ..., fK[(Xi(pK-1), pK-1), (Xi(pK), pK)]>。
Wherein f1[(Xi(p1), p1), (Xi(p2), p2)] represent in segmentation [pj, pj+1] in linear fit function.
(4.2):Single side hypothsis inspection is carried out to correlation of variables, significance is set, checked according to single side hypothsis and tied
Correlation between fruit and significance confirmation variable, determines coefficient correlation trend.
In step (4.2), single side hypothsis inspection:H0:ρs[Xi, Xj]=0vs H1:ρs[Xi, Xj] > 0;
H0:ρs[Xi, Xj]=0 vs H2:ρs[Xi, Xj] < 0;
When the number of samples n > 10 of hypothesis testing are participated in, stochastic variable Us is defined as:
Given level of significance α, if Us> tα(zs- 2), then H0 relative with H1 is rejected, if Us<-tα(zs- 2), then with H2
Relative H0 is rejected, wherein tα(zs- 2) quantile of statistic Us is represented, now, X in s-th segmentationiAnd XjCorrelation
It is considered as significant, symbol direction signs(Xi, Xj) respectively value be 1 or -1, if | Us| < tα(zs- 2), no matter for
H1 or H2, H0 can not be rejected, without significant correlation between this variations per hour, symbol direction signs(Xi, Xj) value be 0.
As number of samples n < 10, facing for the Spearman rank correlation coefficients for small samples method hypothesis testing is inquired about
Dividing value, will be corresponding to given zsCritical correlation coefficients with α are expressed as ρα(zs), if | ρs[Xi, Xj] | > ρα(zs), H0 quilts
Refusal, signs(Xi, Xj) value is 1 or -1 respectively, otherwise H0 can not be rejected, symbol direction signs(Xi, Xj) value be 0.
According to the segments for obtaining, the coefficient correlation of each segmentation is calculated, draw coefficient correlation confidential interval, such as Fig. 3
Shown, wherein L represents hop count, and L is positive integer.
Given α=0.05, correlation test is carried out to each segmentation, so that it is determined that the correlative trend between variable,
As shown in Figure 4.
The present invention calculates the coefficient correlation of each segmentation in time series using formula of correlation coefficient, then related to variable
Property carry out single side hypothsis inspection, significance is set, between confirming variable according to single side hypothsis assay and significance
Correlation, it is determined that coefficient correlation trend, for accurately obtain abnormal data section and its relevant information provide precise information, enter
And improve the efficiency of alarm.
Step 5:According to correlative trend and actual trend comparison, abnormal data section and its relevant information are obtained.
In steps of 5, also include:Obtain time series segmentation figure, scatter diagram and the fitting of alarm variable and correlated variables
Rectilinear, coefficient correlation confidential interval, if there is dynamic deferred relation, the time series that also obtain before translating is former
Figure.
It can be seen from correlative trend analysis result, air preheater inlet flue gas temperature and air preheater export wind-warm syndrome just
Positively related relation when in the case of often, but in the t=2142-2522s time periods it is the now section in uncorrelated in two variables
Belong to abnormal data section.Analysis understands, in the data segment, it may be possible to which the failure of air-introduced machine causes air preheater inlet velocity
Reduce, air quantity is reduced, so that in the case where air preheater inlet flue gas temperature is constant, air preheater outlet wind-warm syndrome slightly has
Raise, after air-introduced machine is adjusted, the relation between two variables recovers normal.
The present invention choose industry variants between correlation as judge work dotted state whether exception feature, by right
Associated variable sets up binary time series, binding time sequence segment method and coefficient correlation tendency method, farthest reduces
Fitting phenomenon and leakage segmentation phenomenon, obtain abnormal data section, so as to carry out exception exactly from historical data fast automaticly
Data Detection provides favourable condition to realize the dynamic alert threshold design of multivariable warning system, so as to reduce interference report
It is alert, the efficiency of site operation personnel's treatment alarm is improved, ensure production security.
Fig. 6 is the structural representation that the alert correlation based on correlation of the invention becomes amount detection systems.
As shown in Figure 6, a kind of alert correlation based on correlation of the invention becomes amount detection systems, including:
(1) data extraction module, its be used for from history detection data extract predetermined time period alarm variable and with
Data of its associated multiple correlated variables and as detection object.
(2) time series sets up module, and it is used to judge dynamic deferred between the alarm variable of selection and correlated variables
Relation, and then set up binary time series T and be standardized as T '.
Further, in the time series sets up module, if there is dynamic between alarm variable and correlated variables prolonging
Slow relation, then obtain one section of primary data of the two variables to ask for time delay h, then the alarm of predetermined time period is become
The time span translation h of amount or correlated variables;If not existing dynamic deferred relation between alarm variable and correlated variables,
Without translation.
The present invention first determines whether the dynamic deferred relation alarmed between variable and correlated variables, and then establishes accurate two
Elementary time sequence T, and then abnormal data section can be exactly obtained, the efficiency of site operation personnel's treatment alarm is improve, protect
Production security is hindered.
(3) time series segmentation module, it is used under the constraint of the minimum interval caused by noise, during to binary
Between sequence T ' be segmented.
Further, the time series segmentation module, including:Divided data section acquisition module is treated, when it is used for binary
Between sequence T ' as data segment to be divided;
According to coefficient correlation between data in data segment to be divided, the data segment classification category belonging to data segment to be divided is judged
Property;
Wherein, according to preset correlation coefficient number scope, data segment categorical attribute includes weak related data section, middle related data section
With strong correlation data segment.
Specifically, preset correlation coefficient number scope, weak related data section, middle related data section are included by data segment categorical attribute
With strong correlation data segment by taking following scope as an example:
Treat divided data Duan Yousan kinds:
The first, meets coefficient correlation for 0.5 >=ρs>=0.3 weak related data section;
Second, meetAnd 0.9 > ρsThe < ρ of > 0.5 or 0.1sThe data segment of < 0.3;
The third, meets zs> n/3 and ρs>=0.9 or ρsAfter being divided, its subdata section is still for≤0.1 data segment
It is strong correlation;Wherein, ρs≤ 0.1 is the phase relation between s-th segmentation internal variable of the binary time series T ' after standardization
Number;zsIt is number of samples in s-th segmentation;N is the predetermined time period of binary time series T ';N and s is positive integer..
Wherein, dividing the first and second data segment can evade leakage segmentation phenomenon, and dividing the third data segment can evade
Over-fitting.
Further, the time series segmentation module, also includes:
Match point asks for module, and it is used for for data segment to be divided, using the method for linear interpolation, after standardization
Data point in binary time series T ' is in the projection being segmented belonging to it on head and the tail data point line as match point;
Crucial turning point calculates update module, and it is used to find farthest point as being segmented next time by the use of orthogonal distance
Crucial turning point, then determine to treat to treat divided data section with the presence or absence of the third in divided data section, and crucial turning point is updated, until not
Untill there is data segment to be divided again.
Under the constraint of minimum interval, just for weak related data section, time span is long and correlation for the present invention
Not visible data section, and time span is long and correlation significantly, but subdata section after dividing still is the number of strong correlation
Time series division is carried out according to section, the crucial turning point of data segment to be divided is obtained, piecewise linearity is carried out to original time series
Represent, so as to avoid overfitting and careless omission from being segmented.
(4) coefficient correlation asks for module, its coefficient correlation and its correlative trend for being used to ask for each segmentation.
As shown in fig. 7, coefficient correlation asks for module, including:
Segmentation coefficient correlation computing module, it is used to divide time series according to the final crucial turning point for obtaining, utilizes
Formula of correlation coefficient calculates the coefficient correlation of each segmentation in time series;
Coefficient correlation trend determining module, it is used to carry out correlation of variables single side hypothsis inspection, sets conspicuousness water
Flat, the correlation between confirming variable according to single side hypothsis assay and significance determines coefficient correlation trend.
The present invention calculates the coefficient correlation of each segmentation in time series using formula of correlation coefficient, then related to variable
Property carry out single side hypothsis inspection, significance is set, between confirming variable according to single side hypothsis assay and significance
Correlation, it is determined that coefficient correlation trend, for accurately obtain abnormal data section and its relevant information provide precise information, enter
And improve the efficiency of alarm.
(5) abnormal data acquisition module, it is used for according to correlative trend and actual trend comparison, obtains abnormal data section
And its relevant information.
Further, the system also includes:Minimum interval computing module, it is used to obtain current preset time span
Alarm variable each lower flex point, and try to achieve the distance between adjacent lower flex point, and then constitute array d;
Array d is ranked up and removes repeat element, obtain array d0;Ask for the maximum point of the slope variation of array d0
And its distance between immediate lower flex point dm, dm are the minimum interval alarmed in variable;
The distance between lower flex point and adjacent lower flex point of correlated variables are obtained, and then obtains the minimum time of correlated variables
Interval dh;
Using the higher value in dm and dh as binary time series T ' minimum interval.
Minimum interval is used to reduce influence of the noise to segmentation result, when actual industrial process data are processed, makes an uproar
The interference of sound can cause the time interval between key point too short, therefore, do not closed in the neighborhood region of minimum interval
Key point search.
The present invention choose industry variants between correlation as judge work dotted state whether exception feature, by right
Associated variable sets up binary time series, binding time sequence segment method and coefficient correlation tendency method, farthest reduces
Fitting phenomenon and leakage segmentation phenomenon, obtain abnormal data section, so as to carry out exception exactly from historical data fast automaticly
Data Detection provides favourable condition to realize the dynamic alert threshold design of multivariable warning system, so as to reduce interference report
It is alert, the efficiency of site operation personnel's treatment alarm is improved, ensure production security.
Although above-mentioned be described with reference to accompanying drawing to specific embodiment of the invention, not to present invention protection model
The limitation enclosed, one of ordinary skill in the art should be understood that on the basis of technical scheme those skilled in the art are not
Need the various modifications made by paying creative work or deformation still within protection scope of the present invention.
Claims (10)
1. a kind of alert correlation based on correlation becomes quantity measuring method, it is characterised in that the method is in server or processor
Interior completion, it is specifically included:
Step 1:The alarm variable of predetermined time period and multiple related changes associated there are extracted from history detection data
The data of amount, selection one of which alarm variable and correlated variables are used as detection object;
Step 2:Judge the dynamic deferred relation between the alarm variable and correlated variables of selection, and then set up binary time series
T is simultaneously standardized as T ';
Step 3:Under the constraint of the minimum interval caused by noise, binary time series T ' is segmented;
Step 4:Ask for the coefficient correlation and its correlative trend of each segmentation;
Step 5:According to correlative trend and actual trend comparison, abnormal data section and its relevant information are obtained.
2. the alert correlation based on correlation as claimed in claim 1 becomes quantity measuring method, it is characterised in that in the step
In 2, if there is dynamic deferred relation between alarm variable and correlated variables, by the alarm variable or phase of predetermined time period
The time span for closing variable is translated and is kept between alarm variable and correlated variables that to there is dynamic deferred relation constant;If
Do not exist dynamic deferred relation between alarm variable and correlated variables, then without translation.
3. the alert correlation based on correlation as claimed in claim 1 becomes quantity measuring method, it is characterised in that in the step
Also included before 3:The minimum interval that calculating is caused by noise, its detailed process includes:
(3.1.1) obtains each lower flex point of the alarm variable of current preset time span, and tries to achieve between adjacent lower flex point
Distance, and then constitute array d;
(3.1.2) is ranked up to array d and removes repeat element, obtains array d0;The slope variation for asking for array d0 is maximum
Point and its distance between immediate lower flex point dm, dm be the minimum interval alarmed in variable;
(3.1.3) obtains the minimum interval dh of correlated variables to correlated variables repeat step (3.1.1) and (3.1.2);
(3.1.4) using the higher value in dm and dh as binary time series T ' minimum interval.
4. the alert correlation based on correlation as claimed in claim 1 becomes quantity measuring method, it is characterised in that in the step
The process that binary time series T ' is segmented is included in 3:
Using binary time series T ' as data segment to be divided;
According to coefficient correlation between data in data segment to be divided, the data segment categorical attribute belonging to data segment to be divided is judged;
Wherein, according to preset correlation coefficient number scope, data segment categorical attribute includes weak related data section, middle related data section and strong
Related data section.
5. the alert correlation based on correlation as claimed in claim 4 becomes quantity measuring method, it is characterised in that in the step
The process being segmented to binary time series T ' in 3, also includes:
For data segment to be divided, using the method for linear interpolation, by the data point in binary time series T ' at belonging to it points
Projection on section head and the tail data point line is used as match point;
Find farthest point as the crucial turning point being segmented next time by the use of orthogonal distance, then determine to treat in divided data section whether
There is strong correlation data segment, and update crucial turning point;
Repeat the above steps, until no longer there is data segment to be divided untill.
6. the alert correlation based on correlation as claimed in claim 5 becomes quantity measuring method, it is characterised in that the step 4
In, the coefficient correlation of each segmentation and its detailed process of correlative trend are asked for, including:
(4.1):Time series is divided according to the final crucial turning point for obtaining, time series is calculated using formula of correlation coefficient
In each segmentation coefficient correlation;
(4.2):Carry out single side hypothsis inspection to correlation of variables, significance be set, according to single side hypothsis assay and
Significance confirms the correlation between variable, determines coefficient correlation trend.
7. a kind of alert correlation based on correlation becomes amount detection systems, it is characterised in that including:
Data extraction module, it is used to being extracted from history detection data the alarm variable of predetermined time period and associated with it
Multiple correlated variables data, selection one of which alarm variable and correlated variables as detection object;
Time series sets up module, and it is used to judge the dynamic deferred relation between the alarm variable of selection and correlated variables, enters
And set up binary time series T and be standardized as T ';
Time series segmentation module, it is used under the constraint of the minimum interval caused by noise, to binary time series
T ' is segmented;
Coefficient correlation asks for module, its coefficient correlation and its correlative trend for being used to ask for each segmentation;
Abnormal data acquisition module, it is used for according to correlative trend and actual trend comparison, obtains abnormal data section and its phase
Pass information.
8. the alert correlation based on correlation as claimed in claim 7 becomes amount detection systems, it is characterised in that in the time
Sequence is set up in module, if there is dynamic deferred relation between alarm variable and correlated variables, by the report of predetermined time period
The time span of alert variable or correlated variables is translated and is kept existing between alarm variable and correlated variables dynamic deferred
Relation is constant;If not existing dynamic deferred relation between alarm variable and correlated variables, without translation;
Further, the system also includes:Minimum interval computing module, its report for being used to obtain current preset time span
The lower flex point of each of alert variable, and the distance between adjacent lower flex point is tried to achieve, and then constitute array d;
Array d is ranked up and removes repeat element, obtain array d0;Ask for array d0 the maximum point of slope variation and its
The distance between immediate adjacent lower flex point dm, dm are the minimum interval in alarm variable;
The distance between lower flex point and adjacent lower flex point of correlated variables are obtained, and then obtains the minimum interval of correlated variables
dh;
Using the higher value in dm and dh as binary time series minimum interval;
Further, the time series segmentation module, including:Divided data section acquisition module is treated, it is used for two elementary time sequences
Row T ' is used as data segment to be divided;
According to coefficient correlation between data in data segment to be divided, the data segment categorical attribute belonging to data segment to be divided is judged;
Wherein, according to preset correlation coefficient number scope, data segment categorical attribute includes weak related data section, middle related data section and strong
Related data section.
9. the alert correlation based on correlation as claimed in claim 8 becomes amount detection systems, it is characterised in that the time sequence
Row segmentation module, also includes:
Match point asks for module, and it is used for for data segment to be divided, using the method for linear interpolation, by the binary after standardization
Data point in time series T ' is in the projection being segmented belonging to it on head and the tail data point line as match point;
Crucial turning point calculates update module, and it is used to find farthest point as the key being segmented next time by the use of orthogonal distance
Turning point, then determine to treat to whether there is strong correlation data segment in divided data section, and crucial turning point is updated, treated until no longer existing
Untill dividing data segment.
10. the alert correlation based on correlation as claimed in claim 9 becomes amount detection systems, it is characterised in that the correlation
Coefficient asks for module, including:
Segmentation coefficient correlation computing module, it is used to divide time series according to the final crucial turning point for obtaining, using correlation
Coefficient formula calculates the coefficient correlation of each segmentation in time series;
Coefficient correlation trend determining module, it is used to carry out correlation of variables single side hypothsis inspection, sets significance, root
Correlation between confirming variable according to single side hypothsis assay and significance, determines coefficient correlation trend.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710206963.0A CN106778053B (en) | 2017-03-31 | 2017-03-31 | A kind of alert correlation variable detection method and system based on correlation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710206963.0A CN106778053B (en) | 2017-03-31 | 2017-03-31 | A kind of alert correlation variable detection method and system based on correlation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106778053A true CN106778053A (en) | 2017-05-31 |
CN106778053B CN106778053B (en) | 2019-04-09 |
Family
ID=58965942
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710206963.0A Expired - Fee Related CN106778053B (en) | 2017-03-31 | 2017-03-31 | A kind of alert correlation variable detection method and system based on correlation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106778053B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108492141A (en) * | 2018-03-28 | 2018-09-04 | 联想(北京)有限公司 | A kind of prediction technique and device of multi-model fusion |
CN108549346A (en) * | 2018-05-14 | 2018-09-18 | 山东科技大学 | A kind of historical data section automatic searching method suitable for System Discrimination |
CN108573055A (en) * | 2018-04-24 | 2018-09-25 | 山东科技大学 | A kind of multivariable alarm monitoring method and system based on qualitiative trends analysis |
CN108629335A (en) * | 2018-06-05 | 2018-10-09 | 华东理工大学 | Adaptive face key feature points selection method |
WO2019214367A1 (en) * | 2018-05-07 | 2019-11-14 | 山东科技大学 | Method and system for dynamic alarm threshold design and alarm cancellation |
CN111947903A (en) * | 2020-07-08 | 2020-11-17 | 中核武汉核电运行技术股份有限公司 | Vibration abnormity positioning method and device |
CN113344737A (en) * | 2021-06-04 | 2021-09-03 | 北京国电通网络技术有限公司 | Device control method, device, electronic device and computer readable medium |
CN113515093A (en) * | 2020-04-10 | 2021-10-19 | 阿里巴巴集团控股有限公司 | Data processing method, data processing device, production control method, production control device, equipment and storage medium |
CN113781758A (en) * | 2021-09-07 | 2021-12-10 | 浙江大学 | Variable collaborative dynamic alarm threshold optimization method for high-end coal-fired power generation equipment |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104915568A (en) * | 2015-06-24 | 2015-09-16 | 哈尔滨工业大学 | Satellite telemetry data abnormity detection method based on DTW |
CN106368813A (en) * | 2016-08-30 | 2017-02-01 | 北京协同创新智能电网技术有限公司 | Abnormal alarm data detection method based on multivariate time series |
-
2017
- 2017-03-31 CN CN201710206963.0A patent/CN106778053B/en not_active Expired - Fee Related
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104915568A (en) * | 2015-06-24 | 2015-09-16 | 哈尔滨工业大学 | Satellite telemetry data abnormity detection method based on DTW |
CN106368813A (en) * | 2016-08-30 | 2017-02-01 | 北京协同创新智能电网技术有限公司 | Abnormal alarm data detection method based on multivariate time series |
Non-Patent Citations (2)
Title |
---|
FAN YANG,ET AL.: "Correlation analysis of alarm data and alarm limit design for industrial processes", 《2010 AMERICAN CONTROL CONFERENCE MARRIOTT WATERFRONT,BALTIMORE,MD ,USA》 * |
王娟,等: "异常检测在报警关联分析中的应用", 《解放军理工大学学报(自然科学版)》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108492141A (en) * | 2018-03-28 | 2018-09-04 | 联想(北京)有限公司 | A kind of prediction technique and device of multi-model fusion |
CN108573055A (en) * | 2018-04-24 | 2018-09-25 | 山东科技大学 | A kind of multivariable alarm monitoring method and system based on qualitiative trends analysis |
WO2019214367A1 (en) * | 2018-05-07 | 2019-11-14 | 山东科技大学 | Method and system for dynamic alarm threshold design and alarm cancellation |
KR20200011416A (en) * | 2018-05-07 | 2020-02-03 | 산동 유니버시티 오브 사이언스 앤드 테크놀로지 | Dynamic Alarm Threshold Design and Method and System of Alarm Cancellation |
KR102160202B1 (en) | 2018-05-07 | 2020-09-25 | 산동 유니버시티 오브 사이언스 앤드 테크놀로지 | Dynamic alarm threshold design and method of alarm elimination |
CN108549346A (en) * | 2018-05-14 | 2018-09-18 | 山东科技大学 | A kind of historical data section automatic searching method suitable for System Discrimination |
CN108629335A (en) * | 2018-06-05 | 2018-10-09 | 华东理工大学 | Adaptive face key feature points selection method |
CN113515093A (en) * | 2020-04-10 | 2021-10-19 | 阿里巴巴集团控股有限公司 | Data processing method, data processing device, production control method, production control device, equipment and storage medium |
CN111947903A (en) * | 2020-07-08 | 2020-11-17 | 中核武汉核电运行技术股份有限公司 | Vibration abnormity positioning method and device |
CN113344737A (en) * | 2021-06-04 | 2021-09-03 | 北京国电通网络技术有限公司 | Device control method, device, electronic device and computer readable medium |
CN113344737B (en) * | 2021-06-04 | 2023-11-24 | 北京国电通网络技术有限公司 | Device control method, device, electronic device and computer readable medium |
CN113781758A (en) * | 2021-09-07 | 2021-12-10 | 浙江大学 | Variable collaborative dynamic alarm threshold optimization method for high-end coal-fired power generation equipment |
Also Published As
Publication number | Publication date |
---|---|
CN106778053B (en) | 2019-04-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106778053A (en) | A kind of alert correlation based on correlation becomes quantity measuring method and system | |
CN106368813B (en) | A kind of abnormal alarm data detection method based on multivariate time series | |
Filzmoser | A multivariate outlier detection method | |
US10338551B2 (en) | P and ID and control system synchronization | |
JP6141235B2 (en) | How to detect anomalies in time series data | |
US20150205966A1 (en) | Industrial Control System Emulator for Malware Analysis | |
Dunia et al. | Process monitoring using principal components in parallel coordinates | |
CN110287827B (en) | Bridge strain data outlier identification method based on data correlation | |
CN109933905A (en) | A kind of mechanical equipment health state evaluation method based on multidimensional early warning analysis | |
US11435736B2 (en) | Cause determination of anomalous events | |
CN113654080B (en) | Method and device for determining on-off state of burner in gas boiler | |
Marais et al. | Comparing statistical process control charts for fault detection in wastewater treatment | |
CN110866558A (en) | Multi-source data fusion analysis-based rotating equipment state early warning method | |
CN108508860B (en) | Process industrial production system data monitoring method based on coupling relation | |
Guo et al. | Optimal design of a synthetic chart for monitoring process dispersion with unknown in-control variance | |
CN110298767B (en) | Thermal power plant time sequence variable abnormity monitoring method and system | |
CN104216814B (en) | Log compression monitoring method and log compression monitoring device | |
Livi et al. | On the long-term correlations and multifractal properties of electric arc furnace time series | |
EP3950966A1 (en) | Blast furnace abnormality assessment device, blast furnace abnormality assessment method, and blast furnace operation method | |
Yang et al. | Monitoring and Diagnosing Process Loss Using a Weighted‐Loss Control Chart | |
Sansana et al. | A functional data-driven approach to monitor and analyze equipment degradation in multiproduct batch processes | |
Bilen et al. | Comparison of control charts for autocorrelated process control | |
CN117466153B (en) | Fault detection method, device, computer equipment and readable storage medium | |
CN108549346A (en) | A kind of historical data section automatic searching method suitable for System Discrimination | |
CN109300285B (en) | Early warning method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190409 Termination date: 20200331 |
|
CF01 | Termination of patent right due to non-payment of annual fee |