CN110008565B - Industrial process abnormal working condition prediction method based on operation parameter correlation analysis - Google Patents

Industrial process abnormal working condition prediction method based on operation parameter correlation analysis Download PDF

Info

Publication number
CN110008565B
CN110008565B CN201910244872.5A CN201910244872A CN110008565B CN 110008565 B CN110008565 B CN 110008565B CN 201910244872 A CN201910244872 A CN 201910244872A CN 110008565 B CN110008565 B CN 110008565B
Authority
CN
China
Prior art keywords
relevance
data
parameter
window
prediction
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.)
Active
Application number
CN201910244872.5A
Other languages
Chinese (zh)
Other versions
CN110008565A (en
Inventor
徐正国
王豆
陈积明
程鹏
孙优贤
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN201910244872.5A priority Critical patent/CN110008565B/en
Publication of CN110008565A publication Critical patent/CN110008565A/en
Application granted granted Critical
Publication of CN110008565B publication Critical patent/CN110008565B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/20—Administration of product repair or maintenance

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • Theoretical Computer Science (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Geometry (AREA)
  • Computer Hardware Design (AREA)
  • Evolutionary Computation (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Testing And Monitoring For Control Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an industrial process abnormal working condition prediction method based on operation parameter correlation analysis, which can be applied to fault prediction and health management of an industrial process. The method starts from the relevance among the operation parameters of the industrial process, and carries out the prediction of the abnormal working condition based on the relevance analysis of the operation parameters. In the single-parameter prediction stage, the method predicts each operation parameter through an exponential smoothing method according to the existing sensor data. In the correlation analysis stage, the invention calculates the correlation of the operation parameters by the known values of the operation parameters and the predicted values of the parameters, wherein the correlation of the parameters is represented by the similarity of a series of indexes representing the parameter curves. In the relevance trend prediction stage, the method constructs a multiple autoregressive model to predict the parameter relevance. The method provided by the invention takes the relevance of the operation parameters into consideration, can obtain more complete equipment abnormal information and more advanced prediction results, and has practical significance for the fault prediction of industrial equipment.

Description

Industrial process abnormal working condition prediction method based on operation parameter correlation analysis
Technical Field
The invention belongs to the technical field of reliability engineering, and relates to an industrial process abnormal working condition prediction method based on operation parameter correlation analysis.
Background
With the continuous emergence of complex systems and the increasing demand of real-time monitoring of industrial processes, modern industrial equipment is often equipped with a plurality of sensors to monitor the operation state of the industrial equipment in the operation process. Meanwhile, multiple fault modes may occur in the operation process of the equipment, a certain fault may correspond to a plurality of symptoms, and under the condition, the single sensor information cannot completely reflect the operation state of the equipment, so that fault prediction based on multi-sensor information is generated at the right moment. The failure prediction based on multi-sensor information aims to analyze the operation state of the equipment using comprehensive sensor information, thereby making more reliable equipment diagnosis and prediction. With the continuous development of sensing technology, the use of multiple sensors for condition monitoring, fault diagnosis and prediction of equipment has become a trend.
For a plurality of sensors during the operation of the equipment, the represented operation parameters do not exist independently, and the change of each operation parameter during the operation of the equipment is actually the reaction to the current operation state of the equipment. Under the normal operation state of the industrial equipment, the operation parameters are usually relatively stable and maintained at a relatively stable level, so that the relevance of the operation parameters is relatively stable. However, when an abnormal working condition occurs, the response of each operation parameter to the abnormality is different, so that the relevance of the operation parameters changes, and the relevance and the trend of the relevance change necessarily imply the information of the abnormality and even the fault of the equipment.
Disclosure of Invention
Aiming at the technical situation in the prior art, the invention aims to judge the state of the equipment in the operation process through the relevance of the operation parameters and predict the abnormal working condition of the equipment through predicting the relevance change trend of the operation parameters aiming at the problem that the operation parameters of the equipment have relevance in the operation process.
The concept of the present invention will now be explained as follows:
the invention provides an industrial process abnormal working condition prediction method based on operation parameter relevance change trend. In the single-parameter prediction stage, the method predicts each operation parameter through an exponential smoothing method according to the existing sensor data. In the correlation analysis stage, the invention calculates the correlation of the operation parameters by the known values of the operation parameters and the predicted values of the parameters, wherein the correlation of the parameters is represented by the similarity of a series of indexes representing the parameter curves. In the relevance trend prediction stage, the method constructs a multiple autoregressive model to predict the parameter relevance. The method provided by the invention takes the relevance of the operation parameters into consideration, and can obtain more complete equipment abnormal information and more advanced prediction results.
According to the invention concept, the invention provides an industrial process abnormal working condition prediction method based on operation parameter correlation analysis, which comprises the following steps:
step 1: carrying out fixed step length prediction on a measured value sequence acquired by each sensor in the industrial process by adopting a Holt exponential smoothing model;
step 2: constructing a triple characterized by the variation trend of the operation parameters of the industrial process in a sliding window mode, expressing a measured value sequence in a data window, and calculating the relevance of any two operation parameters in the window through a relevance index based on Euclidean distance;
and step 3: constructing a relevance prediction model according to the calculated relevance data, wherein the relevance prediction model is a multiple autoregressive model, and estimating model parameters through a partial least square algorithm;
and 4, step 4: and for the newly obtained data, predicting the abnormal working condition of the equipment according to the relevance prediction model, and continuously updating the relevance prediction model and the model parameters before predicting the abnormal working condition.
Based on the above scheme, the following implementation manner can be specifically adopted for each step:
preferably, step 1 is specifically as follows:
step 1.1: for industrial equipment with a plurality of sensors, the number of the sensors is recorded as N, when the equipment is in the operation process, operation parameter values, namely sensor data, representing the operation state of the equipment are continuously acquired, and measurement sequences of the sensors are recorded as
Figure GDA0002672550720000021
Wherein K represents the length of the sequence,
Figure GDA0002672550720000022
represents the measured value of the sensor i at the kth sampling time point;
step 1.2: for the sensor i, a Holt exponential smoothing model is adopted to predict the measured value sequence of the sensor i, and the measured value is given
Figure GDA0002672550720000023
Its smoothed value can be calculated according to the following equation:
Figure GDA0002672550720000024
Figure GDA0002672550720000025
wherein the content of the first and second substances,
Figure GDA0002672550720000026
to represent
Figure GDA0002672550720000027
A smoothed value of (d);
Figure GDA0002672550720000028
is a linear growth factor, representing the trend after smoothing; alpha and beta are smooth coefficients, and the value ranges are (0, 1); initial conditions for the Holt model are as follows:
Figure GDA0002672550720000031
Figure GDA0002672550720000032
Figure GDA0002672550720000033
step 1.3: after obtaining the smooth value and the linear growth factor of the measured data, the result is used for prediction and prediction value
Figure GDA0002672550720000034
Comprises the following steps:
Figure GDA0002672550720000035
where l represents the prediction step size and τ is the sampling interval of the device sensor signal.
Preferably, step 2 comprises the following substeps:
step 2.1: the operation parameter correlation analysis comprises three stages of time sequence segmentation fitting, triple representation and correlation calculation; in the stage of time series piecewise fitting, for a data window with a fixed length L, it is recorded as a window WjOperating parameter XiN, L data in the window are 1, 2
Figure GDA0002672550720000036
Where j is the start of the window and the corresponding sampling time is tj(ii) a If it is assumed that m data in one of the time slots in the window
Figure GDA0002672550720000037
Can be fitted by a line segment for the measured values therein
Figure GDA0002672550720000038
Which corresponds to a value in the fitted line segment of
Figure GDA0002672550720000039
The fitting error ERR of the line segment is calculated as:
Figure GDA00026725507200000310
when the window data with j as the starting point is subjected to piecewise linearization, the method is characterized in that
Figure GDA00026725507200000311
Starting to perform line segment fitting on the obtained object; the process of data segmentation fitting is carried out according to the following steps, wherein the set fitting error threshold is recorded as omegaE：
Step (1): set fitting starting point as
Figure GDA00026725507200000312
Fitted endpoint of
Figure GDA00026725507200000313
Wherein, for the window W with j as the starting pointjThe initial fitting starting point is
Figure GDA00026725507200000314
Step (2): for data
Figure GDA00026725507200000315
Adopting a linear regression mode to carry out line segment fitting, thereby obtaining corresponding line segment data
Figure GDA00026725507200000316
Calculating the fitting error ERR according to the fitting error ERR calculation formula;
and (3): if ERR is less than or equal to omegaEIf so, changing h to h +1, and repeating the step (2); if ERR > omegaEIf yes, saving the current fitting end point (namely the data segmentation point), resetting h to be 2, returning to the step 1, and fitting the next part of data by taking the current fitting end point as a new fitting starting point;
repeating the steps (1) to (3) until the window WjAll the data in the linear block are segmented, namely the segmented and linearized data is obtained
Figure GDA0002672550720000041
Step 2.2: in the segment triplet representation stage, a segment s is described in the form of the following tripletj，
Figure GDA0002672550720000042
Wherein k isjWhich represents the slope of the line segment,
Figure GDA0002672550720000043
indicates the length of the line segment on the time axis, rjIndicating the rate of increase of the value of the line, i.e. for line data
Figure GDA0002672550720000044
For window WjNew sequence after data line segmentation
Figure GDA0002672550720000045
Get its three-tuple sequence representation as s1，s2，...，snN represents the number of line segments after data segmentation in the window;
step 2.3: in the correlation calculation stage, two operation parameters V in the equipment operation processAAnd an operating parameter VBFirstly, segmenting data after line segmentation: at the window WjIn, record parameter VAIs a segmentation point of
Figure GDA0002672550720000046
Parameter VBIs a segmentation point of
Figure GDA0002672550720000047
Wherein n isAAnd nBThe number of line segments after the measured data of the operating parameters A and B are segmented in the window is respectively represented; for parameter VAAnd VBThe segmentation points are merged, repeated items are removed, the segmentation points are arranged from small to large, and the segmentation point sequence of the two parameters is obtained as
Figure GDA0002672550720000048
Subsequently, a parameter V is obtained from the sequence of segmentation points and the representation of tripletsAAnd VBThe new triplet sequence is
Figure GDA0002672550720000049
And
Figure GDA00026725507200000410
obtaining a window WjTwo internal parameters VAAnd VBAfter the triple sequence, the relevance index d based on Euclidean distance is usedABTo calculate two parameters V in the windowAAnd VBThe relevance of (A):
Figure GDA00026725507200000411
Figure GDA00026725507200000412
in the formula:
Figure GDA00026725507200000413
as a parameter VAThe slope of the ith line segment,
Figure GDA00026725507200000414
as a parameter VBThe slope of the ith line segment,
Figure GDA00026725507200000415
as a parameter VAThe numerical growth rate of the ith line segment,
Figure GDA00026725507200000416
as a parameter VBThe numerical growth rate of the ith line segment.
Preferably, step 3 comprises the following substeps:
step 3.1: constructing a relevance prediction model: setting the prediction step length as f, determining model design parameters U and M according to the known data length to enable U + f + M-1 to be equal to a window WjAnd constructing the following matrix:
Figure GDA0002672550720000051
Figure GDA0002672550720000052
wherein, { d1，d2，...，dU+f+M-1Denotes a parameter VAAnd VBA correlation sequence of djPresentation Window WjTwo internal parameters VAAnd VBCorrelation index d ofAB；
Step 3.2: for each relevance sequence, using the relevance values of U relevance values to predict the relevance value after f steps, constructing a relevance prediction model:
Fp＝Dpθ
wherein the parameter θ ═ θ1，θ2，...，θU]TAnd can be obtained by a partial least squares algorithm.
Preferably, step 4 comprises the following substeps:
in prediction, a new matrix is constructed using the newly acquired data from the sensors:
Figure GDA0002672550720000053
subsequently, the relevance values are f-step predicted using the constructed relevance prediction model, i.e.
Figure GDA0002672550720000054
When in use
Figure GDA0002672550720000055
Judging the abnormality of the equipment, wherein dnormalIs a parameter V when the equipment is in a normal operation state at the initial operation stageAAnd VBCorrelation value of ωpA drift amount threshold for the correlation value relative to the initial normal; if it is
Figure GDA0002672550720000056
Reconstructing the model by using new data obtained by the sensor to update the model parameter theta;
and continuously predicting the relevance value of a certain prediction step length along with the updating of the data, thereby predicting the occurrence time of the abnormal working condition of the equipment.
The method for predicting the abnormal working condition of the industrial process based on the correlation analysis of the operation parameters can be used for a complex industrial system with a plurality of sensors. The method starts from the relevance among the operation parameters of the industrial process, and carries out the prediction of the abnormal working condition based on the relevance analysis of the operation parameters. In the single-parameter prediction stage, the method predicts each operation parameter through an exponential smoothing method according to the existing sensor data. In the correlation analysis stage, the invention calculates the correlation of the operation parameters by the known values of the operation parameters and the predicted values of the parameters, wherein the correlation of the parameters is represented by the similarity of a series of indexes representing the parameter curves. In the relevance trend prediction stage, the method constructs a multiple autoregressive model to predict the parameter relevance. The method provided by the invention takes the relevance of the operation parameters into consideration, and can obtain more complete equipment abnormal information and more advanced prediction results. The method provides powerful data support for subsequent equipment health management, is particularly valuable for high-reliability equipment maintenance management, and has wide prospects in the aspect of practical engineering application.
Drawings
FIG. 1 shows measured values and predicted results of turbine operating parameters;
FIG. 2 is a comparison of the predicted trend of correlation between turbine vacuum A and other parameters with the true value;
FIG. 3 shows the prediction result of the time of occurrence of an abnormality in a steam turbine.
Detailed Description
The present invention will now be further described with reference to the accompanying drawings, and some of the principles have been described in detail above, and will not be described again here. The following example illustrates the specific operation steps and verifies the effectiveness of the proposed method using a real case based on the turbine rough protection trip data.
The milling machine data records the operational degradation process of cutting metal material with the milling cutter. The initial working condition of the operation of the steam turbine is load 250MW and condenser vacuum 93kPa, the vacuum value of the condenser is used as an indication parameter, vacuum A starts to indicate abnormity from the 762 th sampling point, and when the vacuum value is reduced to 81kPa, the steam turbine trips. The method for predicting the abnormal working condition of the industrial process comprises the following steps:
step 1: and (3) predicting the measured value sequence acquired by each sensor in the industrial process by using a Holt exponential smoothing model in a fixed step length mode. The method specifically comprises the following substeps:
step 1.1: for industrial equipment with a plurality of sensors, the number of the sensors is recorded as N, when the equipment is in the operation process, operation parameter values, namely sensor data, representing the operation state of the equipment are continuously acquired, and measurement sequences of the sensors are recorded as
Figure GDA0002672550720000061
Wherein K represents the length of the sequence,
Figure GDA0002672550720000062
represents the measured value of the sensor i at the kth sampling time point;
step 1.2: for the sensor i, a Holt exponential smoothing model is adopted to predict the measured value sequence of the sensor i, and the measured value is given
Figure GDA0002672550720000063
Its smoothed value can be calculated according to the following equation:
Figure GDA0002672550720000064
Figure GDA0002672550720000065
wherein the content of the first and second substances,
Figure GDA0002672550720000071
to represent
Figure GDA0002672550720000072
A smoothed value of (d);
Figure GDA0002672550720000073
is a linear growth factor, representing the trend after smoothing; alpha and beta are smooth coefficients, and the value ranges are (0, 1); initial conditions for the Holt model are as follows:
Figure GDA0002672550720000074
Figure GDA0002672550720000075
Figure GDA0002672550720000076
step 1.3: after obtaining the smooth value and the linear growth factor of the measured data, the result is used for prediction and prediction value
Figure GDA0002672550720000077
Comprises the following steps:
Figure GDA0002672550720000078
where l represents the prediction step size and τ is the sampling interval of the device sensor signal. In this example, the preset prediction step size is l-15.
According to step 1, for each operational parameter measurement sequence, a fixed-step prediction is performed, the result is given in fig. 1, and at the same time, an actual state monitoring measurement sequence is also given.
Step 2: and constructing a triple group characterized by the variation trend of the operation parameters of the industrial process in a sliding window mode, expressing a measured value sequence in a data window, and calculating the relevance of any two operation parameters in the window through a relevance index based on the Euclidean distance. The method specifically comprises the following substeps:
step 2.1: correlation of operating parametersThe analysis comprises three stages of time sequence segmentation fitting, triple representation and relevance calculation; in the stage of time series piecewise fitting, for a data window with a fixed length L, it is recorded as a window WjIn this example, the sliding window has a length of 100. Operating parameter XiN, L data in the window are 1, 2
Figure GDA0002672550720000079
Where j is the start of the window and the corresponding sampling time is tj(ii) a If it is assumed that m data in one of the time slots in the window
Figure GDA00026725507200000710
It is possible to fit exactly one line segment for the measured values therein
Figure GDA00026725507200000711
Which corresponds to a value in the fitted line segment of
Figure GDA00026725507200000712
The fitting error ERR of the line segment is calculated as:
Figure GDA00026725507200000713
when the window data with j as the starting point is subjected to piecewise linearization, the method is characterized in that
Figure GDA00026725507200000714
Starting to perform line segment fitting on the obtained object; the process of data segmentation fitting is carried out according to the following steps (wherein the set fitting error threshold is recorded as omega)EIn this example ωE＝0.025)：
Step (1): set fitting starting point as
Figure GDA0002672550720000081
Fitted endpoint of
Figure GDA0002672550720000082
Wherein, for the window W with j as the starting pointjThe initial fitting starting point is
Figure GDA0002672550720000083
Step (2): for data
Figure GDA0002672550720000084
Adopting a linear regression mode to carry out line segment fitting, thereby obtaining corresponding line segment data
Figure GDA0002672550720000085
Calculating the fitting error ERR according to the fitting error ERR calculation formula;
and (3): if ERR is less than or equal to omegaEIf so, changing h to h +1, and repeating the step (2); if ERR > omegaEIf yes, saving the current fitting end point (namely the data segmentation point), resetting h to be 2, returning to the step 1, and fitting the next part of data by taking the current fitting end point as a new fitting starting point;
repeating the steps (1) to (3) until the window WjAll the data in the linear block are segmented, namely the segmented and linearized data is obtained
Figure GDA0002672550720000086
Step 2.2: in the segment triplet representation stage, a segment s is described in the form of the following tripletj，
Figure GDA0002672550720000087
Wherein k isjWhich represents the slope of the line segment,
Figure GDA0002672550720000088
indicates the length of the line segment on the time axis, rjIndicating the rate of increase of the value of the line, i.e. for line data
Figure GDA0002672550720000089
For window WjNew sequence after data line segmentation
Figure GDA00026725507200000810
Get its three-tuple sequence representation as s1，s2，...，snN represents the number of line segments after data segmentation in the window;
step 2.3: in the correlation calculation stage, two operation parameters V in the equipment operation processAAnd an operating parameter VBFirstly, segmenting data after line segmentation: with an operating parameter VAAnd an operating parameter VBFor example, in the window WjIn, record parameter VAIs a segmentation point of
Figure GDA00026725507200000811
Parameter VBIs a segmentation point of
Figure GDA00026725507200000812
Wherein n isAAnd nBThe number of line segments after the measured data of the operating parameters A and B are segmented in the window is respectively represented; for parameter VAAnd VBThe segmentation points are merged, repeated items are removed, the segmentation points are arranged from small to large, and the segmentation point sequence of the two parameters is obtained as
Figure GDA00026725507200000813
Subsequently, a parameter V is obtained from the sequence of segmentation points and the representation of tripletsAAnd VBThe new triplet sequence is
Figure GDA00026725507200000814
And
Figure GDA00026725507200000815
obtaining a window WjTwo internal parameters VAAnd VBAfter the triple sequence, the relevance index d based on Euclidean distance is usedABTo calculate two parameters V in the windowAAnd VBThe relevance of (A):
Figure GDA0002672550720000091
Figure GDA0002672550720000092
in the formula:
Figure GDA0002672550720000093
as a parameter VAThe slope of the ith line segment,
Figure GDA0002672550720000094
as a parameter VBThe slope of the ith line segment,
Figure GDA0002672550720000095
as a parameter VAThe numerical growth rate of the ith line segment,
Figure GDA0002672550720000096
as a parameter VBThe numerical growth rate of the ith line segment.
And step 3: and constructing a relevance prediction model according to the calculated relevance data, wherein the relevance prediction model is a multiple autoregressive model, and estimating model parameters by a partial least square algorithm. The method specifically comprises the following substeps:
step 3.1: constructing a relevance prediction model: setting the prediction step length as f, determining model design parameters U and M according to the known data length to enable U + f + M-1 to be equal to a window WjAnd constructing the following matrix:
Figure GDA0002672550720000097
Figure GDA0002672550720000098
wherein, { d1，d2，...，dU+f+M-1Denotes a parameter VAAnd VBA correlation sequence of djPresentation Window WjTwo internal parameters VAAnd VBCorrelation index d ofAB；
Step 3.2: for each relevance sequence, using the relevance values of U relevance values to predict the relevance value after f steps, constructing a relevance prediction model:
Fp＝Dpθ
wherein the parameter θ ═ θ1，θ2，...，θU]TAnd can be obtained by a partial least squares algorithm.
And 4, step 4: and for the newly obtained data, predicting the abnormal working condition of the equipment according to the relevance prediction model, and continuously updating the relevance prediction model and the model parameters before predicting the abnormal working condition. The method specifically comprises the following substeps:
in prediction, a new matrix is constructed using the newly acquired data from the sensors:
Figure GDA0002672550720000101
subsequently, the relevance values are f-step predicted using the constructed relevance prediction model, i.e.
Figure GDA0002672550720000102
When in use
Figure GDA0002672550720000103
Judging the abnormality of the equipment, wherein dnormalIs a parameter V when the equipment is in a normal operation state at the initial operation stageAAnd VBCorrelation value of ωpA drift amount threshold for the correlation value relative to the initial normal; if it is
Figure GDA0002672550720000104
Reconstructing the model by using new data obtained by the sensor to update the model parameter theta;
and continuously predicting the relevance value of a certain prediction step length along with the updating of the data, thereby predicting the occurrence time of the abnormal working condition of the equipment. In this example, the prediction step f is 10, and the threshold ω is setp＝0.1。
According to the steps 2 to 4, relevance calculation and relevance change trend prediction are carried out, and the result is shown in fig. 2. Subsequently, according to the set failure threshold ωpThe abnormal condition prediction is performed, and the result is shown in fig. 3, and meanwhile, the failure time obtained by detecting the relevance sequence obtained by using the actual data according to the set threshold is also given.
FIG. 1 shows measured values and predicted results of turbine operating parameters. As can be seen from the figure, the single parameter can be predicted well by using exponential smooth prediction. FIG. 2 shows the predicted trend of the correlation between turbine vacuum A and other parameters compared with the true value. As can be seen, a good prediction effect is obtained. Fig. 3 shows the prediction result of the turbine abnormality occurrence time. Fig. 3 shows the correlation between vacuum a and the remaining 6 parameters on the abscissa 1-6, respectively, and the ordinate indicates the predicted occurrence time of the abnormal condition. As can be seen from FIG. 3, a relatively accurate prediction result is obtained by using the prediction of the relevance of several sets of parameters, i.e., the occurrence of abnormal conditions can be predicted in advance by a certain step length. In addition, through the analysis of the raw data, the vacuum a is known to indicate an abnormality from the 762 th sampling point, and the decrease is accelerated, and through the analysis of the correlation change trend, as can be seen from fig. 3, the time for predicting the occurrence of the abnormal condition is the 590 th sampling point (ID6) at the earliest, that is, the occurrence of the abnormality is captured earlier. More importantly, through the relevance change trend, the parameter corresponding to the earliest abnormity generation is found, the judgment of the position of the abnormity generation is facilitated, and the important function of eliminating the abnormity is achieved.

Claims (1)

1. An industrial process abnormal working condition prediction method based on operation parameter correlation analysis is characterized by comprising the following steps:
step 1: carrying out fixed step length prediction on a measured value sequence acquired by each sensor in the industrial process by adopting a Holt exponential smoothing model;
step 2: constructing a triple characterized by the variation trend of the operation parameters of the industrial process in a sliding window mode, expressing a measured value sequence in a data window, and calculating the relevance of any two operation parameters in the window through a relevance index based on Euclidean distance;
and step 3: constructing a relevance prediction model according to the calculated relevance data, wherein the relevance prediction model is a multiple autoregressive model, and estimating model parameters through a partial least square algorithm;
and 4, step 4: for newly obtained data, predicting the abnormal working condition of the equipment according to the relevance prediction model, and continuously updating the relevance prediction model and the model parameters before predicting the abnormal working condition;
the step 1 comprises the following substeps:
step 1.1: for industrial equipment with a plurality of sensors, the number of the sensors is recorded as N, when the equipment is in the operation process, operation parameter values, namely sensor data, representing the operation state of the equipment are continuously acquired, and measurement sequences of the sensors are recorded as
Figure FDA0002594709670000011
Wherein K represents the length of the sequence,
Figure FDA0002594709670000012
represents the measured value of the sensor i at the kth sampling time point;
step 1.2: for the sensor i, a Holt exponential smoothing model is adopted to predict the measured value sequence of the sensor i, and the measured value is given
Figure FDA0002594709670000013
Its smoothed value can be calculated according to the following equation:
Figure FDA0002594709670000014
Figure FDA0002594709670000015
wherein the content of the first and second substances,
Figure FDA0002594709670000016
to represent
Figure FDA0002594709670000017
A smoothed value of (d);
Figure FDA0002594709670000018
is a linear growth factor, representing the trend after smoothing; alpha and beta are smooth coefficients, and the value ranges are (0, 1); initial conditions for the Holt model are as follows:
Figure FDA0002594709670000019
Figure FDA00025947096700000110
Figure FDA00025947096700000111
step 1.3: after obtaining the smooth value and the linear growth factor of the measured data, the result is used for prediction and prediction value
Figure FDA00025947096700000112
Comprises the following steps:
Figure FDA0002594709670000021
wherein l represents the predicted step length, and τ is the sampling interval of the device sensor signal;
step 2 comprises the following substeps:
step 2.1: the operation parameter correlation analysis comprises three stages of time sequence segmentation fitting, triple representation and correlation calculation; in the stage of time series piecewise fitting, for a data window with a fixed length L, it is recorded as a window WjOperating parameter XiN, L data in the window are 1, 2
Figure FDA0002594709670000022
Where j is the start of the window and the corresponding sampling time is tj(ii) a If it is assumed that m data in one of the time slots in the window
Figure FDA0002594709670000023
Can be fitted by a line segment for the measured values therein
Figure FDA0002594709670000024
Which corresponds to a value in the fitted line segment of
Figure FDA0002594709670000025
The fitting error ERR of the line segment is calculated as:
Figure FDA0002594709670000026
when the window data with j as the starting point is subjected to piecewise linearization, the method is characterized in that
Figure FDA0002594709670000027
Starting to perform line segment fitting on the obtained object; the process of data segmentation fitting is carried out according to the following steps, wherein the set fitting error threshold is recorded as omegaE：
Step (1): set fitting starting point as
Figure FDA0002594709670000028
Fitted endpoint of
Figure FDA0002594709670000029
h is 2; wherein, for the window W with j as the starting pointjThe initial fitting starting point is
Figure FDA00025947096700000210
Step (2): for data
Figure FDA00025947096700000211
Adopting a linear regression mode to carry out line segment fitting, thereby obtaining corresponding line segment data
Figure FDA00025947096700000212
Calculating the fitting error ERR according to the fitting error ERR calculation formula;
and (3): if ERR is less than or equal to omegaEIf so, changing h to h +1, and repeating the step (2); if ERR > omegaEIf yes, saving the current fitting end point, resetting h to be 2, returning to the step (1), and fitting the next part of data by taking the current fitting end point as a new fitting starting point;
repeating the steps (1) to (3) until the window WjAll the data in the linear block are segmented, namely the segmented and linearized data is obtained
Figure FDA00025947096700000213
Step 2.2: in the segment triplet representation stage, a segment s is described in the form of the following tripletj，
Figure FDA00025947096700000214
Wherein k isjWhich represents the slope of the line segment,
Figure FDA0002594709670000031
indicates the length of the line segment on the time axis, rjIndicating the rate of increase of the value of the line, i.e. for line data
Figure FDA0002594709670000032
For window WjNew sequence after data line segmentation
Figure FDA0002594709670000033
Get its three-tuple sequence representation as s1，s2，...，snN represents the number of line segments after data segmentation in the window;
step 2.3: in the correlation calculation stage, two operation parameters V in the equipment operation processAAnd an operating parameter VBFirstly, segmenting data after line segmentation: at the window WjIn, record parameter VAIs a segmentation point of
Figure FDA0002594709670000034
Parameter VBIs a segmentation point of
Figure FDA0002594709670000035
Wherein n isAAnd nBRespectively representing the operating parameters VAAnd VBThe number of line segments of the measurement data after being segmented in the window; for parameter VAAnd VBThe segmentation points are merged, repeated items are removed, the segmentation points are arranged from small to large, and the segmentation point sequence of the two parameters is obtained as
Figure FDA0002594709670000036
Subsequently, a parameter V is obtained from the sequence of segmentation points and the representation of tripletsAAnd VBThe new triplet sequence is
Figure FDA0002594709670000037
And
Figure FDA0002594709670000038
obtaining a window WjTwo internal parameters VAAnd VBAfter the triple sequence, the relevance index d based on Euclidean distance is usedABTo calculate two parameters V in the windowAAnd VBThe relevance of (A):
Figure FDA0002594709670000039
Figure FDA00025947096700000310
in the formula:
Figure FDA00025947096700000311
as a parameter VAThe slope of the ith line segment,
Figure FDA00025947096700000312
as a parameter VBThe slope of the ith line segment,
Figure FDA00025947096700000313
as a parameter VAThe numerical growth rate of the ith line segment,
Figure FDA00025947096700000314
as a parameter VBThe numerical growth rate of the ith line segment;
step 3 comprises the following substeps:
step 3.1: constructing a relevance prediction model: setting the prediction step length as f, determining model design parameters U and M according to the known data length to enable U + f + M-1 to be equal to a window WjAnd constructing the following matrix:
Figure FDA00025947096700000315
Figure FDA00025947096700000316
wherein, { d1，d2，...，dU+f+M-1Denotes a parameter VAAnd VBA correlation sequence of djPresentation Window WjTwo internal parameters VAAnd VBCorrelation index d ofAB；
Step 3.2: for each relevance sequence, using the relevance values of U relevance values to predict the relevance value after f steps, constructing a relevance prediction model:
Fp＝Dpθ
wherein the parameter θ ═ θ1，θ2，...，θU]TIt can be obtained by partial least squares algorithm;
step 4 comprises the following substeps:
in prediction, a new matrix is constructed using the newly acquired data from the sensors:
Figure FDA0002594709670000041
subsequently, the relevance values are f-step predicted using the constructed relevance prediction model, i.e.
Figure FDA0002594709670000042
When in use
Figure FDA0002594709670000043
Judging the abnormality of the equipment, wherein dnormalIs a parameter V when the equipment is in a normal operation state at the initial operation stageAAnd VBCorrelation value of ωpA drift amount threshold for the correlation value relative to the initial normal; if it is
Figure FDA0002594709670000044
Reconstructing the model by using new data obtained by the sensor to update the model parameter theta;
and continuously predicting the relevance value of a certain prediction step length along with the updating of the data, thereby predicting the occurrence time of the abnormal working condition of the equipment.
CN201910244872.5A 2019-03-28 2019-03-28 Industrial process abnormal working condition prediction method based on operation parameter correlation analysis Active CN110008565B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910244872.5A CN110008565B (en) 2019-03-28 2019-03-28 Industrial process abnormal working condition prediction method based on operation parameter correlation analysis

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910244872.5A CN110008565B (en) 2019-03-28 2019-03-28 Industrial process abnormal working condition prediction method based on operation parameter correlation analysis

Publications (2)

Publication Number Publication Date
CN110008565A CN110008565A (en) 2019-07-12
CN110008565B true CN110008565B (en) 2020-12-11

Family

ID=67168700

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910244872.5A Active CN110008565B (en) 2019-03-28 2019-03-28 Industrial process abnormal working condition prediction method based on operation parameter correlation analysis

Country Status (1)

Country Link
CN (1) CN110008565B (en)

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111913444B (en) * 2019-09-07 2022-03-18 宁波大学 Chemical process monitoring method based on time sequence multi-block modeling strategy
CN111553606A (en) * 2020-05-06 2020-08-18 上海核工程研究设计院有限公司 Evaluation method for nuclear power equipment component design and equipment index relevance
CN113867262B (en) * 2020-06-30 2023-07-14 华晨宝马汽车有限公司 Device, method and medium for monitoring the operating state of a spindle device
CN111983478A (en) * 2020-07-07 2020-11-24 江苏大学 Electrochemical energy storage power station SOC anomaly detection method based on Holt linear trend model
CN112464146B (en) * 2020-12-03 2022-06-07 北京航空航天大学 Key subsystem based on historical telemetering data and single-machine correlation health baseline construction method
CN112801426B (en) * 2021-04-06 2021-06-22 浙江浙能技术研究院有限公司 Industrial process fault fusion prediction method based on correlation parameter mining
CN113624025B (en) * 2021-08-18 2022-04-22 浙江大学 Condenser vacuum low-jump machine symptom capturing method based on correlation of operation parameters
CN113792988A (en) * 2021-08-24 2021-12-14 河北先河环保科技股份有限公司 Online monitoring data anomaly identification method for enterprise
US20230164035A1 (en) * 2021-11-23 2023-05-25 International Business Machines Corporation Identifying persistent anomalies for failure prediction
CN115034094B (en) * 2022-08-10 2022-10-25 南通恒强轧辊有限公司 Prediction method and system for operation state of metal processing machine tool
CN115509187B (en) * 2022-09-20 2023-04-18 北京中佳瑞通科技有限公司 Industrial big data processing method and system
CN115689071B (en) * 2023-01-03 2023-05-02 南京工大金泓能源科技有限公司 Equipment fault fusion prediction method and system based on associated parameter mining
CN115829422B (en) * 2023-02-21 2024-01-02 北京瀚海蓝山智能科技有限公司 Industrial equipment operation abnormal state identification method based on big data

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109086804A (en) * 2018-07-12 2018-12-25 中石化石油机械股份有限公司 A kind of hydraulic device fault forecast method merged based on multi source status monitoring information and reliability characteristic

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2009260176A (en) * 2008-04-21 2009-11-05 Nec Electronics Corp Method of predicting reliability of semiconductor device, and program therefor
CN108037387B (en) * 2017-12-05 2020-06-19 北京能源集团有限责任公司 Equipment fault analysis method and device based on clustering
CN108873859B (en) * 2018-05-31 2020-07-31 浙江工业大学 Bridge type grab ship unloader fault prediction model method based on improved association rule
CN109110608A (en) * 2018-10-25 2019-01-01 歌拉瑞电梯股份有限公司 A kind of elevator faults prediction technique based on big data study

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109086804A (en) * 2018-07-12 2018-12-25 中石化石油机械股份有限公司 A kind of hydraulic device fault forecast method merged based on multi source status monitoring information and reliability characteristic

Also Published As

Publication number Publication date
CN110008565A (en) 2019-07-12

Similar Documents

Publication Publication Date Title
CN110008565B (en) Industrial process abnormal working condition prediction method based on operation parameter correlation analysis
CN110008253B (en) Industrial data association rule mining and abnormal working condition prediction method
CN110018670B (en) Industrial process abnormal working condition prediction method based on dynamic association rule mining
CA2417074C (en) Method and system for trend detection and analysis
CN110414154B (en) Fan component temperature abnormity detection and alarm method with double measuring points
EP2989705B1 (en) Method and apparatus for defect pre-warning of power device
JP2000259222A (en) Device monitoring and preventive maintenance system
CN108038044A (en) A kind of method for detecting abnormality towards continuous monitored target
CN109960232B (en) Method for selecting leading auxiliary parameter and method for equipment maintenance pre-diagnosis
KR102067344B1 (en) Apparatus and Method for Detecting Abnormal Vibration Data
CN111324105B (en) Small sensor fault detection and separation method based on statistic Mahalanobis distance
JP5827426B1 (en) Predictive diagnosis system and predictive diagnosis method
US7949497B2 (en) Machine condition monitoring using discontinuity detection
CN115689393A (en) Real-time dynamic monitoring system and method for power system based on Internet of things
EP3936957B1 (en) Error-based method for calculating a remaining useful life of an apparatus
CN117150244B (en) Intelligent power distribution cabinet state monitoring method and system based on electrical parameter analysis
CN114326593A (en) Tool life prediction system and method
JP2002090266A (en) Remaining life-predicting device
CN112528227A (en) Sensor abnormal data identification method based on mathematical statistics
EP3869286A1 (en) System and method for detecting turbine underperformance and operation anomaly
CN115034094A (en) Prediction method and system for operation state of metal processing machine tool
JP7020565B2 (en) Process control equipment, process control method and process control program
CN112700039A (en) Steady state detection and extraction method for load operation data of thermal power plant
JP3836741B2 (en) Load characteristic estimating device and system frequency stabilizing device using the same
JP7441811B2 (en) diagnostic equipment

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