CN106778053B - Correlation-based alarm correlation variable detection method and system - Google Patents

Abstract

The invention discloses a correlation-based alarm correlation variable detection method and system, wherein, the method establishes a binary time series for correlation variables, combines the time series segmentation method and the correlation coefficient trend method, and quickly and automatically detects historical data from historical data. Accurately obtain abnormal data segments in the system, so as to detect abnormal data, and provide favorable conditions for the realization of dynamic alarm threshold design of multivariable alarm system, thereby reducing interference alarms, improving the efficiency of on-site operators to handle alarms, and ensuring production safety.

Description

Correlation-based alarm correlation variable detection method and system

technical field

The invention belongs to the field of signal processing, and in particular relates to a correlation-based alarm correlation variable detection method and system.

Background technique

The alarm system plays a vital role in ensuring the safe production and efficient operation of coal-fired generating units. Due to the interaction between the associated variables in the actual industrial process, the traditional single-variable alarm threshold design method may generate a large number of interference alarms (leakage and leakage). alarms and false alarms) and lead to the occurrence of "too many alarms", which affects the attention of on-site operators and increases the difficulty of correct handling when abnormal production conditions occur.

Based on a large number of case facts that the correlation between variables in the process of equipment or system abnormality often changes significantly compared with normal operating conditions, a correlation-based detection method for alarm-related variables is found to automatically filter out normal and abnormal conditions. The detection method of the data segment is very necessary.

SUMMARY OF THE INVENTION

In order to solve the above problems, the first object of the present invention is to provide a correlation-based method for detecting alarm-related variables. The method can quickly and automatically obtain abnormal data segments from historical data accurately, so as to detect abnormal data, and provide favorable conditions for realizing the design of dynamic alarm thresholds of multivariable alarm systems, thereby reducing interference alarms and improving on-site operators to handle alarms. efficiency and ensure the safety of production.

The correlation-based alarm correlation variable detection method of the present invention, which is completed in a server or a processor, specifically includes:

Step 1: extract the alarm variable of preset time length and the data of a plurality of related variables associated with it from the historical detection data, and select a group of alarm variables and related variables as the detection object;

Step 2: Determine the dynamic delay relationship between the selected alarm variables and related variables, and then establish a binary time series T and standardize it to T';

Step 3: Segment the binary time series T' under the constraint of the minimum time interval caused by noise;

Step 4: Obtain the correlation coefficient of each segment and its correlation trend;

Step 5: According to the correlation trend and the actual trend comparison, the abnormal data segment and its related information are obtained.

Further, in the step 2, if there is a dynamic delay relationship between the alarm variable and the related variable, then the preset time length of the alarm variable or the time length of the related variable is shifted and the existence of the alarm variable and the related variable is maintained. The dynamic delay relationship is unchanged; if there is no dynamic delay relationship between the alarm variable and the related variable, no translation is required.

The invention firstly judges the dynamic delay relationship between the alarm variable and the related variable, and then establishes an accurate binary time series T, and then can accurately acquire abnormal data segments, improves the efficiency of the on-site operators in handling the alarm, and ensures the production safety. sex.

Further, before the step 3, it also includes: calculating the minimum time interval caused by noise, and the specific process includes:

(3.1.1) Obtain each lower inflection point of the alarm variable of the current preset time length, and obtain the distance between adjacent lower inflection points, and then form an array d;

(3.1.2) Sort the array d and remove the duplicate elements to get the array d0; find the distance dm between the point with the largest slope change of the array d0 and the closest lower inflection point, dm is the minimum time in the alarm variable interval;

(3.1.3) Repeat steps (3.1.1) and (3.1.2) for the relevant variables to obtain the minimum time interval dh of the relevant variables;

(3.1.4) Take the larger value of dm and dh as the minimum time interval of the binary time series T'.

The minimum time interval is used to reduce the influence of noise on the segmentation results. When dealing with actual industrial process data, the interference of noise will cause the time interval between key points to be too short. Click Search.

Further, the process of segmenting the binary time series T' in the step 3 includes:

Take the binary time series T' as the data segment to be divided;

According to the correlation coefficient between the data in the data segment to be divided, determine the classification attribute of the data segment to which the data segment to be divided belongs;

Wherein, according to the preset correlation coefficient range, the data segment classification attributes include weakly correlated data segments, moderately correlated data segments, and strongly correlated data segments.

Among them, dividing weakly correlated data segments and medium correlated data segments can avoid the phenomenon of missing segmentation, and strongly correlated data segments can avoid the phenomenon of overfitting.

Further, the process of segmenting the binary time series T' in the step 3 also includes:

For the data segment to be divided, the linear interpolation method is used, and the projection of the data point in the binary time series T' on the line connecting the data points of the beginning and the end of the segment to which it belongs is used as the fitting point;

Use the orthogonal distance to find the farthest point as the key turning point of the next segment, and then determine whether there is a strongly correlated data segment in the data segment to be segmented, and update the key turning point;

The above steps are repeated until there are no more data segments to be divided.

Under the constraint of the minimum time interval, the present invention is only for weakly correlated data segments, the time length is too long and the correlation is not significant, and the time length is too long and the correlation is significant, but the sub-data segments after division are still strongly correlated The data segment is divided into time series, the key turning points of the data segment to be divided are obtained, and the piecewise linear representation of the original time series is performed to avoid overfitting and omission of segmentation.

Further, in the step 4, the specific process of obtaining the correlation coefficient of each segment and its correlation trend includes:

(4.1): Divide the time series according to the finally obtained key turning points, and use the correlation coefficient formula to calculate the correlation coefficient of each segment in the time series;

(4.2): Carry out unilateral hypothesis test on the correlation of variables, set the significance level, confirm the correlation between variables according to the unilateral hypothesis test result and the significance level, and determine the trend of the correlation coefficient.

The invention uses the correlation coefficient formula to calculate the correlation coefficient of each segment in the time series, and then performs unilateral hypothesis test on the correlation of variables, sets the significance level, and confirms the correlation between variables according to the unilateral hypothesis test result and the significance level. The correlation coefficient trend is determined, and accurate data is provided for accurately obtaining abnormal data segments and their related information, thereby improving the efficiency of alarming.

The second object of the present invention is to provide a correlation-based alarm-related variable detection system.

A correlation-based alarm correlation variable detection system of the present invention includes:

a data extraction module, which is used for extracting data of an alarm variable of a preset time length and a plurality of related variables associated with it from the historical detection data, and selects a group of alarm variables and related variables as the detection object;

A time series establishment module, which is used to judge the dynamic delay relationship between the selected alarm variables and related variables, and then establish a binary time series T and standardize it to T';

a time series segmentation module for segmenting a binary time series T' under the constraint of a minimum time interval caused by noise;

Correlation coefficient obtaining module, which is used to obtain the correlation coefficient of each segment and its correlation trend;

The abnormal data acquisition module is used to acquire abnormal data segments and related information according to the correlation trend and the actual trend comparison.

In the time series establishment module, if there is a dynamic delay relationship between the alarm variable and the related variable, the preset time length of the alarm variable or the time length of the related variable is shifted and the dynamic delay between the alarm variable and the related variable is maintained. The delay relationship does not change; if there is no dynamic delay relationship between the alarm variable and the related variable, no translation is required.

The invention firstly judges the dynamic delay relationship between the alarm variable and the related variable, and then establishes an accurate binary time series T, and then can accurately acquire abnormal data segments, improves the efficiency of the on-site operators in handling the alarm, and ensures the production safety. sex.

Further, the system also includes: a minimum time interval calculation module, which is used to obtain each lower inflection point of the alarm variable of the current preset time length, and obtains the distance between adjacent lower inflection points, and then forms an array d;

Sort the array d and remove the duplicate elements to obtain the array d0; find the distance dm between the point with the largest slope change in the array d0 and its closest adjacent lower inflection point, dm is the minimum time interval in the alarm variable;

Obtain the lower inflection point of the relevant variable and the distance between the adjacent lower inflection points, and then obtain the minimum time interval dh of the relevant variable;

Take the larger of dm and dh as the minimum time interval of the binary time series;

The minimum time interval is used to reduce the influence of noise on the segmentation results. When dealing with actual industrial process data, the interference of noise will cause the time interval between key points to be too short. Click Search.

Further, the time series segmentation module includes: a data segment acquisition module to be divided, which is used to use the binary time series T' as the data segment to be divided;

According to the correlation coefficient between the data in the data segment to be divided, determine the classification attribute of the data segment to which the data segment to be divided belongs;

Wherein, according to the preset correlation coefficient range, the data segment classification attributes include weakly correlated data segments, moderately correlated data segments, and strongly correlated data segments.

Further, the time series segmentation module also includes:

The fitting point obtaining module is used to use the linear interpolation method for the data segment to be divided, and the projection of the data points in the standardized binary time series T' on the line connecting the head and tail data points of the segment to which it belongs is taken as fitting point;

The key turning point calculation and updating module is used to use the orthogonal distance to find the farthest point as the key turning point of the next segment, and then determine whether there is a third data segment to be divided in the data segment to be divided, and update the key turning point until Until there are no more data segments to be divided.

Under the constraint of the minimum time interval, the present invention is only for weakly correlated data segments, the time length is too long and the correlation is not significant, and the time length is too long and the correlation is significant, but the sub-data segments after division are still strongly correlated The data segment is divided into time series, the key turning points of the data segment to be divided are obtained, and the piecewise linear representation of the original time series is performed to avoid overfitting and omission of segmentation.

Further, the correlation coefficient obtaining module includes:

a segment correlation coefficient calculation module, which is used to divide the time series according to the finally obtained key turning points, and use the correlation coefficient formula to calculate the correlation coefficient of each segment in the time series;

Correlation coefficient trend determination module, which is used to perform unilateral hypothesis test on the correlation of variables, set the significance level, confirm the correlation between variables according to the unilateral hypothesis test result and the significance level, and determine the correlation coefficient trend.

The invention uses the correlation coefficient formula to calculate the correlation coefficient of each segment in the time series, and then performs unilateral hypothesis test on the correlation of variables, sets the significance level, and confirms the correlation between variables according to the unilateral hypothesis test result and the significance level. The correlation coefficient trend is determined, and accurate data is provided for accurately obtaining abnormal data segments and their related information, thereby improving the efficiency of alarming.

Compared with the prior art, the beneficial effects of the present invention are:

The present invention selects the correlation between industrial variables as a feature for judging whether the state of a work point is abnormal, establishes a multivariate time series for the correlated variables, combines the time series segmentation method and the correlation coefficient trend method, and minimizes the over-fitting phenomenon and the The phenomenon of missing segmentation can quickly and automatically obtain abnormal data segments from historical data, so as to detect abnormal data and provide favorable conditions for the realization of dynamic alarm threshold design of multi-variable alarm system, thereby reducing interference alarms and improving on-site operators. The efficiency of the alarm ensures the safety of production.

Description of drawings

The accompanying drawings that form a part of the present application are used to provide further understanding of the present application, and the schematic embodiments and descriptions of the present application are used to explain the present application and do not constitute improper limitations on the present application.

Fig. 1 is the flow chart of the alarm correlation variable detection method based on correlation of the present invention;

Fig. 2 (a) is the variable time series and segmentation result diagram of the inlet smoke temperature of the air preheater of the present invention;

Fig. 2 (b) is the variable time series and segmentation result diagram of the smoke temperature at the outlet of the air preheater of the present invention;

Fig. 3 is each subsection correlation coefficient and its confidence interval in the specific embodiment of the present invention;

Fig. 4 is the correlation trend of the variable in each subsection in the specific embodiment of the present invention;

Figure 5(a) is the first group of multivariate time series scatter plots and fitted straight lines of the present invention;

Figure 5(b) is a second group of multivariate time series scatter plots and fitted straight lines of the present invention;

Figure 5(c) is the third group of multivariate time series scatter plots and fitted straight lines of the present invention;

Figure 5(d) is the fourth group of multivariate time series scatter plots and fitted straight lines of the present invention;

6 is a schematic structural diagram of a correlation-based alarm associated variable detection system of the present invention;

FIG. 7 is a schematic structural diagram of a correlation coefficient obtaining module of the present invention.

Detailed ways

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

It should be noted that the terminology used herein is for the purpose of describing specific embodiments only, and is not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural as well, furthermore, it is to be understood that when the terms "comprising" and/or "including" are used in this specification, it indicates that There are features, steps, operations, devices, components, and/or combinations thereof.

FIG. 1 is a flow chart of the method for detecting an alarm-related variable based on a correlation of the present invention.

As shown in Figure 1, the method for detecting alarm-related variables based on correlation is completed in a server or processor, and specifically includes:

Step 1: Extracting data of an alarm variable of a preset time length and a plurality of related variables associated therewith from the historical detection data, and selecting a set of related variables and alarm variables as detection objects.

Specifically, the raw data of multiple related variables with a time length of N before the current working point are extracted, and a set of alarm variables and a set of related variables are selected for detection.

Taking the specific application scenario of the method of the present invention in a specific example as a power plant as an example:

The flue gas temperature at the inlet of the air preheater in the power plant is selected as the alarm variable, and the air temperature at the outlet of the air preheater is the relevant variable. In a shutdown accident in a power plant, the sampling period before shutdown is selected from historical data as 1 second, the sample size is N=3600 data, and a set of air preheater inlet flue gas temperature and air preheater outlet air temperature is selected for abnormal Data detection.

Step 2: Determine the dynamic delay relationship between the alarm variable and related variables, and then establish a binary time series T and standardize it to T';

Specifically, it is judged whether there is a dynamic delay relationship between the variables. If there is, it is necessary to obtain a piece of initial data of these two variables to obtain the delay time h, and then calculate the time length of the existing alarm variable with length N or related variable. Translate h. If there is no dynamic delay relationship between variables, no translation is required; h is a positive number;

Establish a binary time series T for the two groups of variables that have been shifted or do not need to be shifted, and normalize the binary time series to T'.

In the specific implementation process, it is determined that there is no dynamic delay relationship between the air temperature at the inlet of the air preheater and the air temperature at the outlet of the air preheater, so the delay time is 0. Then construct a binary time series for the two sets of data and standardize them. The binary time series after standardization is denoted as T′=[T ₁ (t), T ₂ (t)], where t=1,...,N , N is a positive integer; T ₁ represents the inlet smoke temperature of the air preheater, and T ₂ represents the air temperature at the outlet of the air preheater.

Step 3: Segment the normalized binary time series T under the constraint of the minimum time interval caused by noise.

Wherein, before step 3, it also includes: calculating the minimum time interval caused by noise, and the specific process includes:

(3.1.1) Obtain each lower inflection point of the alarm variable of the current preset time length, and obtain the distance between adjacent lower inflection points, and then form an array d={d1,d2,d3,...,dx};

(3.1.2) Sort the array d and remove the duplicate elements to get the array d0={d1,d2,d3,...,dx0}, where x0<x, where x0 and x are both positive integers; find the array The distance dm between the point m where the slope of d0 changes the most and its corresponding adjacent lower inflection point, dm is the minimum time interval in the alarm variable;

(3.1.3) Repeat steps (3.1.1) and (3.1.2) for the relevant variables to obtain the minimum time interval dh of the relevant variables;

(3.1.4) Take the larger value of dm and dh as the minimum time interval δ of the binary time series T', 0<δ<n, and N is a positive integer.

Among them, in step (3.1.2), the method of obtaining the point m with the largest slope change of the array d0 can be adopted: draw a line graph with d0 as the abscissa and x0 as the ordinate, and find the largest slope change in the line graph. the point m.

It should be noted that, the above-mentioned method for obtaining the point m where the slope of the array d0 changes the most is only an embodiment, and other existing methods may also be used for implementation.

The minimum time interval δ caused by noise is calculated to be 40s, and the minimum time interval is used in the key point search. After the calculation of the correlation coefficient and the screening of the data segment to be divided, it is finally determined to obtain 5 key points, that is, 4 points. part. Therefore, the time series segmentation diagrams are shown in Figure 2(a) and Figure 2(b).

The minimum time interval is used to reduce the influence of noise on the segmentation results. When dealing with actual industrial process data, the interference of noise will cause the time interval between key points to be too short. Click Search.

Specifically, the process of segmenting the binary time series T' in step 3 includes:

Take the binary time series T' as the data segment to be divided;

According to the correlation coefficient between the data in the data segment to be divided, determine the classification attribute of the data segment to which the data segment to be divided belongs;

Wherein, according to the preset correlation coefficient range, the data segment classification attributes include weakly correlated data segments, moderately correlated data segments, and strongly correlated data segments.

For example, in the current division situation, the correlation coefficient between each segment is calculated, and then the data segment to be divided is obtained; wherein, the data segment to be divided includes three types: weakly correlated data segment, medium correlated data segment and strongly correlated data segment:

The first is a weakly correlated data segment with a correlation coefficient of 0.5≥ρ _s ≥0.3;

Second, satisfy and 0.9> _ρs >0.5 or 0.1< _ρs <0.3;

The third type, after the data segment satisfying z _s >N/3 and ρ _s ≥ 0.9 or ρ _s ≤ 0.1 is divided, its sub-data segments are strongly correlated; where ρ _s ≤ 0.1 is the normalized binary time Correlation coefficient between variables in the s-th subsection of the sequence T'; z _s is the number of samples in the s-th subsection; N is the preset time length of the binary time series T'; N and s are positive integers .

It should be noted that other preset correlation coefficient ranges can also be set in the present invention, and the data segments are divided into three types of data segments: weakly correlated data segments, medium correlated data segments, and strongly correlated data segments.

The Spearman sample correlation coefficient between variables X _i and X _j within the s-th subsection of the time series T' is:

Among them, dividing the first and second data segments can avoid the phenomenon of missing segmentation, and dividing the third data segment can avoid the phenomenon of overfitting.

Further, the process of segmenting the standardized binary time series T' in step 3 also includes:

For the data segment to be divided, the linear interpolation method is used, and the projection of the data points in the standardized binary time series T' on the line connecting the data points at the beginning and end of the segment to which it belongs is used as the fitting point;

Use the orthogonal distance to find the farthest point as the key turning point of the next segmentation, then determine whether the third data segment to be divided in the step (3.2.1) exists in the data segment to be divided, and update the key turning point;

The above steps are repeated until there are no more data segments to be divided.

Among them, the parametric equation of the straight line AB in space can be expressed as:

The coordinates of any point P0 on the straight line AB can be expressed as:

[(X _iB -X _iA )β+X _iA , (t _B -t _A )β+t _A ].

Among them, X represents the variable, t represents the time variable, i=1, 2, respectively represents the alarm variable and related variable, A and B represent the two ends of the straight line AB respectively, and β represents a fixed value. Therefore, the distance from point P to line AB can be defined as:

in Refers to the distance from point P to line AB

in, when The parameter corresponding to the minimum value Then the minimum distance from the data point P to the line AB between the beginning and the end of the segment to which it belongs, that is, the orthogonal distance is The maximum value of D in each segment is the critical turning point of the segment.

The present invention takes the four groups of multivariate time series as shown in Fig. 5(a)-Fig. 5(d) as examples, and uses the method of linear interpolation to place the data points in the standardized binary time series T at the beginning and end of the segment to which they belong. The projections on the line connecting the points are used as fitting points, and the fitted straight lines are obtained respectively, as shown in Fig. 5(a)-Fig. 5(d).

Under the constraint of the minimum time interval, the present invention is only for weakly correlated data segments, the time length is too long and the correlation is not significant, and the time length is too long and the correlation is significant, but the sub-data segments after division are still strongly correlated The data segment is divided into time series, the key turning points of the data segment to be divided are obtained, and the piecewise linear representation of the original time series is performed to avoid overfitting and omission of segmentation.

Step 4: Obtain the correlation coefficient of each segment and its correlation trend.

Specifically, the specific process of obtaining the correlation coefficient of each segment and its correlation trend includes:

(4.1): Divide the time series according to the finally obtained key turning points, and use the correlation coefficient formula to calculate the correlation coefficient of each segment in the time series;

Divide the time series of variables according to the finally obtained set of key turning points P={P ₁ , P ₂ ,..., P _k }. At this time, the piecewise linear representation of the multivariate time series T' is:

T _PLR =< f ₁ [(X _i (p ₁ ), p ₁ ), (X _i (p ₂ ), p ₂ )], . . . , f _K [(X _i (p _K-1 ), p _K-1 ), (X _i (p _K ), p _K )]>.

where f ₁ [(X _i (p ₁ ), p ₁ ), (X _i (p ₂ ), p ₂ )] represents a linear fit function within the piece [p _j , p _j+1 ].

(4.2): Carry out unilateral hypothesis test on the correlation of variables, set the significance level, confirm the correlation between variables according to the unilateral hypothesis test result and the significance level, and determine the trend of the correlation coefficient.

In step (4.2), one-sided hypothesis test: H0: ρ _s [X _i , X _j ]=0 vs H1: ρ _s [X _i , X _j ]>0;

H0: ρ _s [X _i , X _j ]=0 vs H2: ρ _s [X _i , X _j ]<0;

When the number of samples participating in the hypothesis test is n>10, the random variable Us is defined as: Given a significance level α, H0 against H1 is rejected if Us _{> t α} (z _s -2), and H0 against H2 is rejected if Us <-t _α (z _s -2) reject, where t _α (z _s -2) represents the quantile of the statistic Us, at which time the correlation between X _i and X _j in the s-th segment is considered significant, and the sign direction sign _s (X _i , X _j ) take the value of 1 or -1 respectively, if |U _s |<t _α (z _s -2), no matter for H1 or H2, H0 can not be rejected, there is no significant correlation between variables at this time, the sign The direction sign _s (X _i , X _j ) takes a value of 0.

When the number of samples n<10, query the critical value of Spearman's rank correlation coefficient for hypothesis testing of small sample size, and express the critical value of the correlation coefficient corresponding to given z _s and α as ρ _α (z _s ), if |ρ _s [X _i , X _j ]|>ρ _α (z _s ), H0 is rejected, sign _s (X _i , X _j ) is 1 or -1 respectively, otherwise H0 cannot be rejected, the sign direction sign _s (X _i , X _j ) takes a value of 0.

According to the obtained number of segments, the correlation coefficient of each segment is calculated, and the confidence interval of the correlation coefficient is drawn, as shown in Figure 3, where L represents the number of segments, and L is a positive integer.

Given α = 0.05, a correlation test was performed on each segment to determine the correlation trend between variables, as shown in Figure 4.

The invention uses the correlation coefficient formula to calculate the correlation coefficient of each segment in the time series, and then performs unilateral hypothesis test on the correlation of variables, sets the significance level, and confirms the correlation between variables according to the unilateral hypothesis test result and the significance level. The correlation coefficient trend is determined, and accurate data is provided for accurately obtaining abnormal data segments and their related information, thereby improving the efficiency of alarming.

Step 5: According to the correlation trend and the actual trend comparison, the abnormal data segment and its related information are obtained.

In step 5, it also includes: obtaining the time series segmented graph, scatter plot, fitted straight line graph, correlation coefficient confidence interval of the alarm variable and related variables, and if there is a dynamic delay relationship, also obtaining the time series before the translation occurs Original image.

According to the correlation trend analysis results, it can be seen that the air temperature at the inlet of the air preheater is positively correlated with the air temperature at the outlet of the air preheater under normal circumstances, but the two variables are not correlated in the time period of t=2142-2522s. At this time, the segment belongs to the abnormal data segment. The analysis shows that in this data segment, it is possible that the failure of the induced draft fan causes the air speed at the inlet of the air preheater to decrease and the air volume to decrease, so that the air temperature at the outlet of the air preheater remains unchanged under the condition that the smoke temperature at the inlet of the air preheater remains unchanged. Slightly increased, when the induced draft fan was adjusted, the relationship between the two variables returned to normal.

The present invention selects the correlation between industrial variables as the feature for judging whether the state of the work point is abnormal, establishes a binary time series for the correlated variables, and combines the time series segmentation method and the correlation coefficient trend method to minimize the over-fitting phenomenon It can quickly and automatically obtain abnormal data segments from historical data, so as to detect abnormal data, which provides favorable conditions for the realization of dynamic alarm threshold design of multivariable alarm systems, thereby reducing interference alarms and improving field operators. The efficiency of processing alarms ensures the safety of production.

FIG. 6 is a schematic structural diagram of the correlation-based alarm-related variable detection system of the present invention.

As shown in Figure 6, a correlation-based alarm correlation variable detection system of the present invention includes:

(1) A data extraction module, which is used for extracting data of an alarm variable of a preset time length and a plurality of related variables associated therewith from the historical detection data and using it as a detection object.

(2) A time series establishment module, which is used to judge the dynamic delay relationship between the selected alarm variables and related variables, and then establish a binary time series T and normalize it to T'.

Further, in the time series establishment module, if there is a dynamic delay relationship between the alarm variable and the related variable, a piece of initial data of these two variables is obtained to obtain the delay time h, and then the alarm of the preset time length is set. The time length of the variable or related variable is shifted by h; if there is no dynamic delay relationship between the alarm variable and the related variable, no translation is required.

The invention firstly judges the dynamic delay relationship between the alarm variable and the related variable, and then establishes an accurate binary time series T, and then can accurately acquire abnormal data segments, improves the efficiency of the on-site operators in handling the alarm, and ensures the production safety. sex.

(3) A time series segmentation module, which is used to segment the binary time series T' under the constraint of the minimum time interval caused by noise.

Further, the time series segmentation module includes: a data segment acquisition module to be divided, which is used to use the binary time series T' as the data segment to be divided;

According to the correlation coefficient between the data in the data segment to be divided, determine the classification attribute of the data segment to which the data segment to be divided belongs;

Wherein, according to the preset correlation coefficient range, the data segment classification attributes include weakly correlated data segments, moderately correlated data segments, and strongly correlated data segments.

Specifically, a range of correlation coefficients is preset, and the data segment classification attributes include weakly correlated data segments, moderately correlated data segments, and strongly correlated data segments, and the following ranges are taken as an example:

There are three types of data segments to be divided:

The first is a weakly correlated data segment with a correlation coefficient of 0.5≥ρ _s ≥0.3;

Second, satisfy and 0.9> _ρs >0.5 or 0.1< _ρs <0.3;

The third type is that after the data segment satisfying z _s >n/3 and ρ _s ≥ 0.9 or ρ _s ≤ 0.1 is divided, its sub-data segments are still strongly correlated; where ρ _s ≤ 0.1 is the normalized binary The correlation coefficient between variables in the s-th subsection of the time series T'; z _s is the number of samples in the s-th subsection; N is the preset time length of the binary time series T'; N and s are positive Integer. .

Among them, dividing the first and second data segments can avoid the phenomenon of missing segmentation, and dividing the third data segment can avoid the phenomenon of overfitting.

Further, the time series segmentation module also includes:

The fitting point obtaining module is used to use the linear interpolation method for the data segment to be divided, and the projection of the data points in the standardized binary time series T' on the line connecting the head and tail data points of the segment to which it belongs is taken as fitting point;

The key turning point calculation and updating module is used to use the orthogonal distance to find the farthest point as the key turning point of the next segment, and then determine whether there is a third data segment to be divided in the data segment to be divided, and update the key turning point until Until there are no more data segments to be divided.

Under the constraint of the minimum time interval, the present invention is only for weakly correlated data segments, the time length is too long and the correlation is not significant, and the time length is too long and the correlation is significant, but the sub-data segments after division are still strongly correlated The data segment is divided into time series, the key turning points of the data segment to be divided are obtained, and the piecewise linear representation of the original time series is performed to avoid overfitting and omission of segmentation.

(4) Correlation coefficient obtaining module, which is used to obtain the correlation coefficient of each segment and its correlation trend.

As shown in Figure 7, the correlation coefficient calculation module includes:

a segment correlation coefficient calculation module, which is used to divide the time series according to the finally obtained key turning points, and use the correlation coefficient formula to calculate the correlation coefficient of each segment in the time series;

Correlation coefficient trend determination module, which is used to perform unilateral hypothesis test on the correlation of variables, set the significance level, confirm the correlation between variables according to the unilateral hypothesis test result and the significance level, and determine the correlation coefficient trend.

The invention uses the correlation coefficient formula to calculate the correlation coefficient of each segment in the time series, and then performs unilateral hypothesis test on the correlation of variables, sets the significance level, and confirms the correlation between variables according to the unilateral hypothesis test result and the significance level. The correlation coefficient trend is determined, and accurate data is provided for accurately obtaining abnormal data segments and their related information, thereby improving the efficiency of alarming.

(5) Abnormal data acquisition module, which is used for acquiring abnormal data segments and related information according to the correlation trend and the actual trend comparison.

Further, the system also includes: a minimum time interval calculation module, which is used to obtain each lower inflection point of the alarm variable of the current preset time length, and obtains the distance between adjacent lower inflection points, and then forms an array d;

Sort the array d and remove the duplicate elements to obtain the array d0; find the distance dm between the point with the largest slope change of the array d0 and the closest lower inflection point, dm is the minimum time interval in the alarm variable;

Obtain the lower inflection point of the relevant variable and the distance between the adjacent lower inflection points, and then obtain the minimum time interval dh of the relevant variable;

Take the larger of dm and dh as the minimum time interval of the binary time series T'.

The minimum time interval is used to reduce the influence of noise on the segmentation results. When dealing with actual industrial process data, the interference of noise will cause the time interval between key points to be too short. Click Search.

The present invention selects the correlation between industrial variables as the feature for judging whether the state of the work point is abnormal, establishes a binary time series for the correlated variables, and combines the time series segmentation method and the correlation coefficient trend method to minimize the over-fitting phenomenon It can quickly and automatically obtain abnormal data segments from historical data, so as to detect abnormal data, which provides favorable conditions for the realization of dynamic alarm threshold design of multivariable alarm systems, thereby reducing interference alarms and improving field operators. The efficiency of processing alarms ensures the safety of production.

Although the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, they do not limit the scope of protection of the present invention. Those skilled in the art should understand that on the basis of the technical solutions of the present invention, those skilled in the art do not need to pay creative efforts. Various modifications or deformations that can be made are still within the protection scope of the present invention.

Claims (8)

1. a correlation-based alarm correlation variable detection method is characterized in that, the method is completed in a server or a processor, and it specifically comprises: Step 1: extract the alarm variable of preset time length and the data of a plurality of related variables associated with it from the historical detection data, and select a group of alarm variables and related variables as the detection object; Step 2: Determine the dynamic delay relationship between the selected alarm variables and related variables, and then establish a binary time series T and standardize it to T'; Step 3: Segment the binary time series T' under the constraint of the minimum time interval caused by noise; Step 4: Obtain the correlation coefficient of each segment and its correlation trend; Step 5: According to the comparison of the correlation trend and the actual trend, the abnormal data segment and its related information are obtained; The process of segmenting the binary time series T' in the step 3 includes: Take the binary time series T' as the data segment to be divided; According to the correlation coefficient between the data in the data segment to be divided, determine the classification attribute of the data segment to which the data segment to be divided belongs; Wherein, according to the preset correlation coefficient range, the data segment classification attributes include weakly correlated data segments, moderately correlated data segments, and strongly correlated data segments. 2. The correlation-based alarm-related variable detection method of claim 1, wherein in the step 2, if there is a dynamic delay relationship between the alarm variable and the related variable, the preset time length The time length of the alarm variable or the related variable is shifted and the dynamic delay relationship between the alarm variable and the related variable is kept unchanged; if there is no dynamic delay relationship between the alarm variable and the related variable, no translation is required. 3. The correlation-based alarm-related variable detection method according to claim 1, wherein before the step 3, the method further comprises: calculating the minimum time interval caused by noise, and the specific process comprises: (3.1.1) Obtain each lower inflection point of the alarm variable of the current preset time length, and obtain the distance between adjacent lower inflection points, and then form an array d; (3.1.2) Sort the array d and remove the duplicate elements to get the array d0; find the distance dm between the point with the largest slope change of the array d0 and the closest lower inflection point, dm is the minimum time in the alarm variable interval; (3.1.3) Repeat steps (3.1.1) and (3.1.2) for the relevant variables to obtain the minimum time interval dh of the relevant variables; (3.1.4) Take the larger value of dm and dh as the minimum time interval of the binary time series T'. 4. the correlation-based alarm correlation variable detection method as claimed in claim 1, is characterized in that, in described step 3, the process of carrying out segmentation to binary time series T ', also comprises: For the data segment to be divided, the linear interpolation method is used, and the projection of the data point in the binary time series T' on the line connecting the data points of the beginning and the end of the segment to which it belongs is used as the fitting point; Use the orthogonal distance to find the farthest point as the key turning point of the next segment, and then determine whether there is a strongly correlated data segment in the data segment to be segmented, and update the key turning point; The above steps are repeated until there are no more data segments to be divided. 5. the correlation-based alarm correlation variable detection method as claimed in claim 4, is characterized in that, in described step 4, the concrete process that seeks the correlation coefficient of each subsection and its correlation trend, comprises: (4.1): Divide the time series according to the finally obtained key turning points, and use the correlation coefficient formula to calculate the correlation coefficient of each segment in the time series; (4.2): Carry out unilateral hypothesis test on the correlation of variables, set the significance level, confirm the correlation between variables according to the unilateral hypothesis test result and the significance level, and determine the trend of the correlation coefficient. 6. A correlation-based alarm associated variable detection system, characterized in that, comprising: a data extraction module, which is used for extracting data of an alarm variable of a preset time length and a plurality of related variables associated with it from the historical detection data, and selects a group of alarm variables and related variables as the detection object; A time series establishment module, which is used to judge the dynamic delay relationship between the selected alarm variables and related variables, and then establish a binary time series T and standardize it to T'; a time series segmentation module for segmenting a binary time series T' under the constraint of a minimum time interval caused by noise; Correlation coefficient obtaining module, which is used to obtain the correlation coefficient of each segment and its correlation trend; Abnormal data acquisition module, which is used to acquire abnormal data segments and related information according to the correlation trend and actual trend comparison; In the time series establishment module, if there is a dynamic delay relationship between the alarm variable and the related variable, the preset time length of the alarm variable or the time length of the related variable is shifted and the dynamic delay between the alarm variable and the related variable is maintained. The delay relationship remains unchanged; if there is no dynamic delay relationship between the alarm variable and the related variable, no translation is required; Further, the system also includes: a minimum time interval calculation module, which is used to obtain each lower inflection point of the alarm variable of the current preset time length, and obtains the distance between adjacent lower inflection points, and then forms an array d; Sort the array d and remove the duplicate elements to obtain the array d0; find the distance dm between the point with the largest slope change in the array d0 and its closest adjacent lower inflection point, dm is the minimum time interval in the alarm variable; Obtain the lower inflection point of the relevant variable and the distance between the adjacent lower inflection points, and then obtain the minimum time interval dh of the relevant variable; Take the larger of dm and dh as the minimum time interval of the binary time series; Further, the time series segmentation module includes: a data segment acquisition module to be divided, which is used to use the binary time series T' as the data segment to be divided; According to the correlation coefficient between the data in the data segment to be divided, determine the classification attribute of the data segment to which the data segment to be divided belongs; Wherein, according to the preset correlation coefficient range, the data segment classification attributes include weakly correlated data segments, moderately correlated data segments, and strongly correlated data segments. 7. The correlation-based alarm correlation variable detection system according to claim 6, wherein the time series segmentation module further comprises: The fitting point obtaining module is used to use the linear interpolation method for the data segment to be divided, and the projection of the data points in the standardized binary time series T' on the line connecting the head and tail data points of the segment to which it belongs is taken as fitting point; The key turning point calculation and updating module is used to find the farthest point using the orthogonal distance as the key turning point of the next segment, and then determine whether there is a strongly correlated data segment in the data segment to be divided, and update the key turning point until it no longer exists. until the data segment is to be divided. 8. The correlation-based alarm correlation variable detection system according to claim 7, wherein the correlation coefficient obtaining module comprises: a segment correlation coefficient calculation module, which is used to divide the time series according to the finally obtained key turning points, and use the correlation coefficient formula to calculate the correlation coefficient of each segment in the time series; Correlation coefficient trend determination module, which is used to perform unilateral hypothesis test on the correlation of variables, set the significance level, confirm the correlation between variables according to the unilateral hypothesis test result and the significance level, and determine the correlation coefficient trend.