CN103631145B - Multiple operating modes process monitoring method and system based on monitor control index switching - Google Patents
Multiple operating modes process monitoring method and system based on monitor control index switching Download PDFInfo
- Publication number
- CN103631145B CN103631145B CN201310675045.4A CN201310675045A CN103631145B CN 103631145 B CN103631145 B CN 103631145B CN 201310675045 A CN201310675045 A CN 201310675045A CN 103631145 B CN103631145 B CN 103631145B
- Authority
- CN
- China
- Prior art keywords
- operating mode
- monitor control
- control index
- vector
- model
- 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
Links
Abstract
The invention discloses a kind of multiple operating modes process monitoring method and system based on monitor control index switching, including: gather the normal data under different operating mode as training sample set;Obtain HMM based on this training sample set, and obtain the limit of the control corresponding to monitor control index of HMM;Training sample based on each operating mode is set up the statistical model of corresponding operating mode respectively and is analyzed model, and obtains the limit of the control corresponding to monitor control index of each statistical model analysis model;Based on the process data design condition vector obtained in real time, and then calculate difference operating mode vector;Norm according to difference operating mode vector, calculate corresponding monitor control index in real time, and it is limited comparison with the control corresponding to the monitor control index of corresponding model, monitor the operation conditions of operating mode, the method real-time acquisition process data ensure the reliability of monitoring, and need not the data Gaussian distributed under each operating mode, there is the higher suitability.
Description
Technical field
The present invention relates to process monitoring field, particularly relate to a kind of based on monitor control index switching multiple operating modes process monitoring method and
System.
Background technology
For process monitoring and troubleshooting issue, traditional method uses multivariatestatistical process control technology mostly
(Multivariable Statistical Process Control, MSPC), wherein with pivot analysis (Principal Component
Analysis, PCA) and method that offset minimum binary (Partial Least Squares, PLS) is representative supervise at industrial process
Control is successfully applied.Traditional MSPC method all assume process operation under single operation operating mode, but
Actually often switch frequently in multiple operating modes due to reason processes such as product change, production capacity adjustment.
The data clothes of each operating mode are assumed that based on the multiple operating modes process such as pivot analysis and Support Vector data description monitoring method
From Gauss distribution, but this might not set up in practice.Although and multiple operating modes set up unified model comparing foundation
The method of multiple models is simple, but lacks the real-time identification to operating mode, this work shape that can cause monitoring current device
Condition.Although and multiple operating modes process based on rarefaction representation monitoring method is it is not assumed that data Gaussian, but the method does not accounts for
Dynamic characteristic in chemical process.It addition, single current data is judged that belonging to it, operating mode may not by effect of noise
Accurately.
Summary of the invention
One of the technical problem to be solved is to need to provide a kind of multiple operating modes process prison based on monitor control index switching
Prosecutor method, whether it belongs to same operating according to the process data obtained in real time judges whether this process event occurs accordingly
Barrier.Additionally, additionally provide a kind of multiple operating modes process monitoring system based on monitor control index switching.
In order to solve above-mentioned technical problem, the invention provides a kind of multiple operating modes process monitoring side based on monitor control index switching
Method, including: acquisition step, gather the normal data under different operating mode as training sample set;First obtaining step, based on
Described training sample set obtains HMM, and obtains the control corresponding to monitor control index of described HMM
System limit;Second obtaining step, training sample based on each operating mode sets up the statistical model of corresponding operating mode respectively and analyzes model,
And obtain the limit of the control corresponding to monitor control index of each statistical model analysis model;Calculation procedure, based on the mistake obtained in real time
Number of passes is according to design condition vector, and calculates difference operating mode vector based on described operating mode vector;Monitoring step, according to described difference
The norm of operating mode vector, chooses described HMM or statistical model corresponding to each operating mode is analyzed model and supervised accordingly
Control index, and selected monitor control index is calculated in real time, by the prison of calculated real-time monitor control index Yu this model
Control control limit comparison corresponding to index, monitors the operation conditions of this operating mode.
In one embodiment, the monitor control index of described statistical model analysis model farther includes the first monitor control index and second
Monitor control index, the control that described first monitor control index is corresponding is limited to the first control limit, the control that described second monitor control index is corresponding
It is limited to the second control limit;By following steps, described monitoring step judges that operating mode process is the most normal further: if described difference
The norm of operating mode vector is zero, then the statistical model calculating operating mode vector corresponding to this difference operating mode vector analyzes the real-time of model
First monitor control index and real-time second monitor control index, described real-time first monitor control index more than first control limit or described in real time
Second monitor control index more than the second control in limited time, then judges that operating mode process occurs abnormal;If the model of described difference operating mode vector
Number is not zero, then calculate the real-time NLLP monitor control index of described HMM, be more than at this NLLP monitor control index
The control of the monitor control index of described HMM in limited time, then judges that operating mode process occurs abnormal, wherein, and NLLP table
Show the negative log-likelihood probability of described HMM.
In one embodiment, in described monitoring step, if norm | | I | |=0 of difference operating mode vector, then according to as follows
Statistical model corresponding to expression formula design condition q analyzes the real-time first monitor control index D of modelr(s) and real-time second prison
Control index Dp(s):
Dp(s)=sTP(q)[Λ(q)]-1P(q)TS,
Wherein, r is residual error subspace, and p is principal component subspace, and s is the statistical model vector of the process data obtained in real time,Statistical model for setting up under operating mode q analyzes the projection matrix in residual error space, P in model(q)For set up under operating mode q
Statistical model analyzes the load matrix in model, Λ(q)Corresponding to pivot during the statistical model of foundation analyzes model under operating mode q
The diagonal matrix that is combined into of covariance matrix eigenvalue.
In one embodiment, by the statistical model vector s of following formula calculating process data:
S=col [μ, Σ, Ξ], wherein, μ represents the mean vector in the w duration of acquisition in real time, and Σ represents second moment, Ξ table
Showing High Order Moment, col [] represents the form that matrix arrangement becomes column vector, described second moment include variance, covariance and from
Correlation coefficient.
In one embodiment, in described monitoring step, if norm | | I | | ≠ 0 of difference operating mode vector, then according to as follows
The real-time NLLP monitor control index of the formula described HMM of calculating:
NLLP=-logPr (Onew|s*),
Wherein, Pr () represents the probability that event occurs, OnewFor the up-to-date process data obtained in real time, s*Represent OnewInstitute
The operating mode at place.
In one embodiment, in described calculation procedure, Viterbi algorithm is used to obtain based on the process data obtained in real time
Take described operating mode vector I=[i1,i2,…,iw]T, wherein, w represents the acquisition duration of process data, ij(j=1 ..., w) represent
Operating mode sequence number residing for w duration internal procedure data.
In one embodiment, utilize following formula to calculate described difference operating mode vector:
I=[i1,▽i2,…,▽iw-1]T, wherein, ij=1-ψ (ij+1-ij), function ψ () value at 0 is 1, remaining
Point value is 0.
According to a further aspect in the invention, a kind of multiple operating modes process monitoring system based on monitor control index switching, bag are additionally provided
Including: acquisition module, it is for gathering the normal data under different operating mode as training sample set;First acquisition module, its base
Obtain HMM in described training sample set, and obtain corresponding to the monitor control index of described HMM
Control limit;Second acquisition module, its training sample based on each operating mode is set up the statistical model of corresponding operating mode respectively and is analyzed mould
Type, and obtain the limit of the control corresponding to monitor control index of each statistical model analysis model;Computing module, it is based on obtaining in real time
The process data design condition vector taken, and calculate difference operating mode vector based on described operating mode vector;Monitoring module, its basis
The norm of described difference operating mode vector, chooses described HMM or statistical model corresponding to each operating mode analyzes model
Corresponding monitor control index, and selected monitor control index is calculated in real time, will calculated real-time monitor control index and this
Control limit comparison corresponding to the monitor control index of model, monitors the operation conditions of this operating mode.
In one embodiment, the monitor control index of described statistical model analysis model farther includes the first monitor control index and second
Monitor control index, the control that described first monitor control index is corresponding is limited to the first control limit, the control that described second monitor control index is corresponding
It is limited to the second control limit;In described monitoring module, judge that operating mode process is the most normal by following steps further: if described
The norm of difference operating mode vector is zero, then the statistical model calculating operating mode vector corresponding to this difference operating mode vector analyzes model
Real-time first monitor control index and real-time second monitor control index, at described real-time first monitor control index more than the first control limit or described
Real-time second monitor control index more than the second control in limited time, then judges that operating mode process occurs abnormal;If described difference operating mode is vectorial
Norm be not zero, then calculate the real-time NLLP monitor control index of described HMM, at this NLLP monitor control index
Prescribe a time limit more than the control of the monitor control index of described HMM, then judge that operating mode process occurs abnormal, its
In, NLLP represents the negative log-likelihood probability of described HMM.
In one embodiment, described computing module uses Viterbi algorithm acquisition described based on the process data obtained in real time
Operating mode vector I=[i1,i2,…,iw]T, wherein, w represents the acquisition duration of process data, ij(j=1 ..., w) represent w duration
Operating mode sequence number residing for internal procedure data.
Compared with prior art, one or more embodiments of the invention can have the advantage that
The present invention calculates different controls according to all normal data under different operating modes respectively from the normal data of same operating
System limit, and acquisition process data calculate difference operating mode vector in real time, and the norm being finally based on difference operating mode vector selects switching
To suitable monitor control index, judge that this process is the most normal by the control limit contrasting this monitor control index corresponding, the party
Method real-time acquisition process data ensure that the reliability of judgement, and need not the data Gaussian distributed under each operating mode,
There is the higher suitability.
Other features and advantages of the present invention will illustrate in the following description, and, partly become aobvious from description
And be clear to, or understand by implementing the present invention.The purpose of the present invention and other advantages can be by wanting in description, right
The structure asking specifically noted in book and accompanying drawing realizes and obtains.
Accompanying drawing explanation
Accompanying drawing is for providing a further understanding of the present invention, and constitutes a part for description, with embodiments of the invention
It is provided commonly for explaining the present invention, is not intended that limitation of the present invention.In the accompanying drawings:
Fig. 1 is the flow chart of based on monitor control index switching according to an embodiment of the invention multiple operating modes process monitoring method;
Fig. 2 is the schematic diagram that the real process data according to the present invention one example gathers;
Fig. 3 is the block diagram of based on monitor control index switching according to an embodiment of the invention multiple operating modes process monitoring system;
Fig. 4 is the curve chart of the testing result of finite mixtures Gauss model method in the test case 1 according to the present invention one example;
Fig. 5 is the curve chart of the testing result of SPA method in the test case 1 according to the present invention one example;
Fig. 6 is the inspection of multiple operating modes process monitoring method based on monitor control index switching in the test case 1 according to the present invention one example
Survey the curve chart of result;
Fig. 7 is the curve chart of the testing result of finite mixtures Gauss model method in the test case 2 according to the present invention one example;
Fig. 8 is the curve chart of the testing result of SPA method in the test case 2 according to the present invention one example;
Fig. 9 is the inspection of multiple operating modes process monitoring method based on monitor control index switching in the test case 2 according to the present invention one example
Survey the curve chart of result;
Figure 10 is the structural representation of the continuous stirring heating tank of another example according to the present invention;
Figure 11 is the curve chart of the testing result of the finite mixtures Gauss model method of another example according to the present invention;
Figure 12 is the curve chart of the testing result of the SPA method of another example according to the present invention;
Figure 13 is the detection knot of the multiple operating modes process monitoring method based on monitor control index switching of another example according to the present invention
The curve chart of fruit.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing, the present invention is made the most detailed
Describe in detail bright.
First embodiment
Fig. 1 is the flow chart of based on monitor control index switching according to an embodiment of the invention multiple operating modes process monitoring method.Under
Face combines Fig. 1 and is described in detail the method.
Step S110, the normal data gathered under different operating mode (i.e. power-equipment duty under certain condition) is made
For training sample set.
Specifically, from chemical process data base, the normal data under different operating mode is obtained as training sample set:Wherein,(i=1 ... M) it is the data sample of i-th operating mode,
Represent the real number matrix of N row m row, NiRepresenting the number of samples of i-th operating mode, N represents total number of samples, m table
Show the number of sensor.
Step S120, obtains HMM based on training sample set, and obtains the monitoring of this HMM
Control limit corresponding to index;Training sample based on each operating mode is set up the statistical model of corresponding operating mode respectively and is analyzed model
(hereinafter referred to as SPA model), and the different control limit corresponding to the monitor control index of the SPA model obtaining each.
In detail, the control of the NLLP monitor control index that the present embodiment obtains HMM limits δNLLP, NLLP table
Show the negative log-likelihood probability of HMM.Specifically, utilize training sample set B, use EM algorithm (
Big Expectation Algorithm) the parameter set λ obtaining HMM the control limit obtaining NLLP monitor control index can be trained
δNLLP.Generally, utilizing normal training sample set to calculate NLLP monitor control index, the confidence level according to choosing (such as exists
Choose in concrete instance is the confidence level of 98%), i.e. can obtain controlling limit δNLLP。
For SPA model, utilize the training sample B of each operating modeiSet up the SPA model under corresponding operating mode, SPA mould
The monitor control index of type farther includes the first monitor control index Dr(i.e. residual error subspace monitor control index) and the second monitor control index Dp
(i.e. principal component subspace monitor control index), the control limit corresponding to the monitor control index of SPA model farther includes the first control
Limit δrLimit δ is controlled with secondp.With δNLLPAcquisition similar, according to normal training sample calculate DrAnd DpMonitor control index,
Further according to choose confidence level (such as choose in concrete instance be 98% confidence level), i.e. can be controlled
Limit δrAnd δp。
Additionally, SPA method is based on central limit theorem, and the thought of central limit theorem be " no matter the distribution of stochastic variable
How, the statistic of stochastic variable Gaussian distributed progressively ", so the method for the present embodiment is not required to each operating mode
Under data Gaussian distributed.
Step S130, based on the process data design condition vector obtained in real time, and calculates difference work based on this operating mode vector
Condition vector.
In the present embodiment, the process data obtained in real time is visualized in one slidably window, calculates the most here
Process data is the data in window, by moving forward the number of passes excessively of renewable window interior along sampling time axle sliding window
According to, the duration obtaining data in real time is then the length of sliding window.Specifically, use based on the process data obtained in real time
Viterbi algorithm obtains operating mode vector Ik=[i1k,i2k,…,iwk]T, wherein, w represents the acquisition duration of process data,
ijk(j=1 ..., w) represent the operating mode sequence number residing for kth sliding window (the most current sliding window) interior data.
Fig. 2 is the schematic diagram that the real process data according to the present invention one example gathers.Two cunnings easy to understand, in Fig. 2
Dynamic window exists the most simultaneously, but sliding window I comprises the process data in sampling time section [10,50], sliding window II
Comprise the process data in sampling time section [70,110].
Further, the expression formula of difference operating mode vector is Ik=[i1k,▽i2k,…,▽i(w-1)k]T, wherein,
▽ijk=1-ψ (i(j+1)k-ijk), ijk(j=1 ..., w) represent that current sliding window is (i.e. in sliding window k) residing for data
Operating mode sequence number, function ψ () value at 0 is 1, remaining some value be 0.
Step S140, norm based on above-mentioned difference operating mode vector, choose HMM or each operating mode is corresponding
The corresponding monitor control index of SPA model, and selected monitor control index is calculated in real time, by calculated real-time prison
Control index and the limit comparison of the control corresponding to monitor control index of this model, monitor the operation conditions of this operating mode.
First, it is determined that whether norm | | I | | of difference operating mode vector is zero.If norm | | I | | of difference operating mode vector is zero,
Illustrate that current acquired process data belongs to same operating, i.e. operating mode does not change, such as sliding window I in Fig. 2
Shown in.If norm | | I | | of difference operating mode vector is not zero, then the current acquired process data of explanation belongs to different work
Condition, as shown in sliding window II in Fig. 2.
Then, when norm | | I | | of difference operating mode vector is zero, the real-time of operating mode corresponding to this difference operating mode vector is calculated
Monitor control index (the i.e. real-time monitor control index D of SPA modelr(s) and Dp(s)), at real-time first monitor control index DrS () is big
In the control limit δ that it is correspondingrTime or real-time second monitor control index DpS () is more than its corresponding control limit δp, then judge
Go out operating mode process and occur abnormal.The monitor control index of the SPA model of design condition q is carried out by following expression:
Dp(s)=sTP(q)[Λ(q)]-1P(q)TS,
Wherein, r is residual error subspace, and p is principal component subspace, and s is the statistical model vector of the process data obtained in real time,For the projection matrix in residual error space, P in the SPA model of foundation under operating mode q(q)For the SPA mould set up under operating mode q
Load matrix in type, Λ(q)Covariance matrix eigenvalue corresponding to pivot in the SPA model of foundation under operating mode q is spelled
The diagonal matrix become.Specifically, by the statistical model vector s of following formula calculating process data:
S=col [μ, Σ, Ξ]
Wherein, μ represents the mean vector in the w duration of acquisition in real time, and Σ represents that second moment, Ξ represent High Order Moment, symbol
Number col [] represents the form that matrix arrangement becomes column vector, and above-mentioned second moment includes variance, covariance and autocorrelation coefficient.
At norm | | I | | when being not zero of difference operating mode vector, then the real-time NLLP monitoring calculating HMM refers to
Mark, at this NLLP monitor control index more than the control limit δ corresponding to the monitor control index of HMMNLLPTime, then sentence
Breaking, it is abnormal the generation of operating mode process.Especially by the following formula real-time NLLP monitor control index of acquisition:
NLLP=-logPr (Onew|s*),
Wherein, Pr () represents the probability that event occurs, OnewFor the up-to-date process data obtained in real time, s*Represent OnewInstitute
The operating mode at place.It practice, owing to the calculating of NLLP monitor control index depends on operating mode, so the NLLP under different operating modes
The control limit of index is the most otherwise varied.
In one example, it is also possible to by formula NLLP=-logPr (Onew| λ) calculate NLLP monitor control index, its
Middle λ is the complete parameter collection of HMM.
Significantly, since according to the difference of difference operating mode vector, different monitoring index is at the appropriateness of current time
Difference, the present embodiment carry out between NLLP monitor control index and SPA Model Monitoring index intelligently index switching rather than
As a monitor control index in existing monitoring method always with on earth.When as it has been described above, be on duty, divided working status vector is not zero, this
Time newly entering sliding window floor data amount not enough, a window i.e. " cannot be filled ", the statistic obtained is inaccurate,
The rate of false alarm of causing trouble increases, and therefore selects the NLLP monitor control index of HMM.Be on duty divided working status to
When amount is zero, illustrating that data are in and be same as operating mode, the statistic obtained is accurate, and monitoring is more comprehensive, so selecting SPA
Two indices in model.
Finally, output procedure monitored results, to facilitate the operating mode to breaking down to safeguard.
Example
In order to be further appreciated by the present invention, below two examples are illustrated.
Example one: numerical simulation
Linear system generation data with following:
Wherein x=[x1 x2 x3]TRepresent the measurement of three sensors, s=[s1 s2]TRepresent two independent data sources,
E=[e1 e2 e3]TThree averages representing independent are 0, standard deviation is the Gaussian noise of 0.1.The system of being listed below is in three
During different operating mode, the probability distribution that data source is obeyed:
Operating mode 1:s1:N(10,0.8);s2:N(12,1.3)
Operating mode 2:s1:N(5,0.6);s2:N(20,0.7)
Operating mode 3:s1:N(16,1.5);s2:N(30,2.5)
Wherein N (μ, σ2) expression average is μ, variance is σ2Gauss distribution.Three kinds of different operating modes are tested data
Describe as shown in table 1.Wherein, shift fault is the steady state value deviation on sensor, and drifting fault is the amplitude on sensor
Big deviation is tapered to from little.
Table 1. numerical simulation is tested the description of data
In order to verify the effectiveness of this example and the advantage in terms of fault detect rate and rate of false alarm, have chosen finite mixtures high
Method (the not having index to switch) method as a comparison of this model (FGMM) method and statistical model analysis (SPA).
Fig. 4-Fig. 9 shows the various method schematic diagram for test case 1 with the failure detection result of test case 2.Wherein, Fig. 4,
BIP index in Fig. 7 is the Bayesian inference probability level in limited gauss hybrid models, the D in Fig. 5, Fig. 8pRefer to
It is designated as the monitor control index of principal component space, D in SPA modelrIndex is the monitor control index in residual error space in SPA model, Fig. 6,
D in Fig. 9p-NLLP index is DpIndex switches with NLLP index, Dr-NLLP index is DrIndex refers to NLLP
Mark switching.
Table 2 show fault diagnosis rate and the summary of rate of false alarm of various method.
Distinct methods fault detect rate (FDR) and the comparison of rate of false alarm (FAR) in table 2. example one
Indicate: in table, data represent percent (%).
It is not difficult to find out that the method for the present invention is compared additive method and had higher fault detect rate from upper table, and it is low to have rate of false alarm
Feature.
Example two: continuous stirring heating tank (CSTH)
Shown in Figure 10 is the structural representation of continuous stirring heating tank.In figure, TC is temperature controller, and FT is that flow becomes
Sending device, FC is flow controller, and TT is temperature transmitter, and LC is liquid level controller, and LT is fluid level transmitter.
In continuous stirring heating tank, hot water and cold water are sufficiently mixed in tank and are heated it by steam.System is deposited
In multiple control loops, thus ensureing liquid level, flow and temperature are operated in the operating point of setting.Table 3 has reacted continuous stirring
The parameter corresponding to two standard conditions of heating tank, the unit of each physical measurement here (signal of telecommunication) is milliampere (mA).
The parameter corresponding to two standard conditions of table 3. continuous stirring heating tank
Table 4 is to the description testing data in continuous stirring heating tank.Below for level sensor fault, compare not Tongfang
The Detection results of method.
Table 4. continuous stirring heating tank is tested the description of data
Figure 11-13 shows fault diagnosis rate and the rate of false alarm of various method.An ibid example, the BIP index in Figure 11
For the Bayesian inference probability level in limited gauss hybrid models, the D in Figure 12pIndex is that in SPA model, pivot is empty
Between monitor control index, DrIndex is the monitor control index in residual error space in SPA model, the D in Figure 13p-NLLP index is Dp
Index switches with NLLP index, Dr-NLLP index is DrIndex switches with NLLP index.
The summary of concrete fault diagnosis rate and rate of false alarm is as shown in table 5.
Distinct methods fault detect rate (FDR) and the comparison of rate of false alarm (FAR) in table 5. example two
Indicate: in table, data represent percent (%).
It is not difficult to find out that the method for the present invention is compared additive method and had higher fault detect rate from upper table, and it is low to have rate of false alarm
Feature.
In sum, the present invention obtains respectively according to the normal data of all normal data under different operating modes with same operating
Different control limits, and calculate difference operating mode vector according to the process data obtained in real time, it is finally based on difference operating mode vector
Norm be switched to the monitor control index that is more suitable for, select to calculate the more excellent monitor control index switched to, refer to by contrasting this monitoring
Marking corresponding control limit and judge that this process is the most normal, the method real-time acquisition process data ensure that the reliable of judgement
Property, and need not the data Gaussian distributed under each operating mode, there is the higher suitability.
Second embodiment
Fig. 3 is the block diagram of based on monitor control index switching according to an embodiment of the invention multiple operating modes process monitoring system.Below
In conjunction with Fig. 3, the monitoring system of the present embodiment is described in detail.
As it is shown on figure 3, native system includes collection model the 30, first acquisition module the 31, second acquisition module 32, calculates mould
Block 33 and monitoring module 34.Acquisition module the 30, first acquisition module 31, second acquisition module 32 of the present embodiment,
Computing module 33 and monitoring module 34 perform the operation of the step S110 to S140 of first embodiment respectively.
In detail, acquisition module 30 is for gathering the normal data under different operating mode as training sample set.Specifically,
Acquisition module 30 obtains the normal data under different operating mode as training sample set from chemical process data base:Wherein,(i=1 ... M) it is the data sample of i-th operating mode,
Represent the real number matrix of N row m row, NiRepresenting the number of samples of i-th operating mode, N represents total number of samples, m table
Show the number of sensor.
Training sample set the first acquisition module 31 gathered based on acquisition module 30 sets up HMM, and obtains
Control limit corresponding to the monitor control index of this HMM.
Generally, normal training sample set is utilized to calculate the NLLP monitor control index of HMM, according to the confidence chosen
Level (such as choose in concrete instance be 98% confidence level), i.e. can obtain control limit δNLLP.Specifically,
Utilize training sample set B, use EM algorithm (i.e. EM algorithm) that the parameter obtaining HMM can be trained
Collect λ and obtain the control limit δ of NLLP monitor control indexNLLP。
Additionally, the training sample of each operating mode that the second acquisition module 32 gathers based on acquisition module 30 sets up corresponding work respectively
The statistical model of condition analyzes model (i.e. SPA model), and obtains the control corresponding to monitor control index of each SPA model
Limit.
The monitor control index of SPA model farther includes the first monitor control index Dr(i.e. residual error subspace monitor control index) and second
Monitor control index Dp(i.e. principal component subspace monitor control index), the control limit bag further corresponding to the monitor control index of SPA model
Include the first control limit δrLimit δ is controlled with secondp.With δNLLPAcquisition similar, according to normal training sample calculate DrAnd DpPrison
Control index, further according to choose confidence level (such as choose in concrete instance be 98% confidence level),
Obtain controlling limit δrAnd δp。
Then it carrys out design condition vector based on the process data obtained in real time to computing module 33, and calculates based on this operating mode vector
Difference operating mode vector.
In detail, computing module 33 uses Viterbi algorithm to obtain operating mode vector based on the process data obtained in real time
I=[i1,i2,…,iw]T, wherein, w represents the acquisition duration of process data, ij(j=1 ..., w) represent w duration internal procedure number
According to residing operating mode sequence number.
After obtaining this operating mode vector, it is vectorial that computing module 33 calculates difference operating mode further with following formula:
I=[i1,▽i2,…,▽iw-1]T, wherein, ij=1-ψ (ij+1-ij), function ψ () value at 0 is 1, and remaining point takes
Value is 0.
Finally, monitoring module 34, according to the norm of above-mentioned difference operating mode vector, chooses HMM or each operating mode
The corresponding corresponding monitor control index of SPA model, and selected monitor control index is calculated in real time, by calculated
Monitor control index and the limit comparison of the control corresponding to monitor control index of this model, monitor the operation conditions of this operating mode in real time.
It should be noted that the difference according to difference operating mode vector, different monitoring index is different at the appropriateness of current time,
The present embodiment carries out index switching rather than intelligently as existing between NLLP monitor control index and SPA Model Monitoring index
A monitor control index in monitoring method is always with on earth.
Specifically, if the norm of difference operating mode vector is zero, then calculate operating mode vector corresponding to this difference operating mode vector
Real-time first monitor control index of SPA model and real-time second monitor control index (the i.e. real-time monitor control index D of SPA modelr(s) and
Dp(s)), at real-time first monitor control index DrS () is more than its corresponding control limit δrTime or real-time second monitor control index
DpS () is more than its corresponding control limit δp, then judge that operating mode process occurs abnormal.Calculated by following expression
The monitor control index of the SPA model of operating mode q:
Dp(s)=sTP(q)[Λ(q)]-1P(q)TS,
Wherein, r is residual error subspace, and p is principal component subspace, and s is the statistical model vector of the process data obtained in real time,For the projection matrix in residual error space, P in the SPA model of foundation under operating mode q(q)For the SPA mould set up under operating mode q
Load matrix in type, Λ(q)For the covariance matrix eigenvalue corresponding to the pivot in the SPA model of foundation under operating mode q
The diagonal matrix being combined into.Specifically, by the statistical model vector s of following formula calculating process data:
S=col [μ, Σ, Ξ]
Wherein, μ represents the mean vector in the w duration of acquisition in real time, and Σ represents that second moment, Ξ represent High Order Moment, symbol
Number col [] represents the form that matrix arrangement becomes column vector, and above-mentioned second moment includes variance, covariance and autocorrelation coefficient.
If the norm of difference operating mode vector is not zero, then calculate the real-time NLLP monitor control index of HMM, at this
NLLP monitor control index is prescribed a time limit more than the control of the monitor control index of HMM, then judge that operating mode process occurs different
Often, wherein, NLLP represents the negative log-likelihood probability of HMM.Obtain in real time especially by following formula
NLLP monitor control index:
NLLP=-logPr (Onew|s*),
Wherein, Pr () represents the probability that event occurs, OnewFor the up-to-date process data obtained in real time, s*Represent OnewInstitute
The operating mode at place.
In sum, the present invention obtains respectively according to the normal data of all normal data under different operating modes with same operating
Different control limits, and calculate difference operating mode vector according to the process data obtained in real time, it is finally based on difference operating mode vector
Norm be switched to the monitor control index that is more suitable for, select to calculate the more excellent monitor control index switched to, refer to by contrasting this monitoring
Marking corresponding control limit and judge that this process is the most normal, the method real-time acquisition process data ensure that the reliable of judgement
Property, and need not the data Gaussian distributed under each operating mode, there is the higher suitability.
The above, only the present invention be embodied as case, protection scope of the present invention is not limited thereto, any be familiar with
Those skilled in the art in technical specification of the present invention, modifications of the present invention or replacement, all should be the present invention's
Within protection domain.
Claims (8)
1. a multiple operating modes process monitoring method based on monitor control index switching, including:
Acquisition step, gathers the normal data under different operating mode as training sample set;
First obtaining step, obtains HMM based on described training sample set, utilizes described training sample set to calculate
NLLP monitor control index, obtains the control limit corresponding to NLLP monitor control index, wherein, NLLP according to the confidence level chosen
Represent the negative log-likelihood probability of described HMM;
Second obtaining step, training sample based on each operating mode sets up the statistical model of corresponding operating mode respectively and analyzes model, and
Calculate each statistical model according to described training sample and analyze the first monitor control index and second monitor control index of model, further according to choosing
The first control limit and second that the confidence level taken obtains each monitor control index the most corresponding controls limit;
Calculation procedure, based on the process data design condition vector obtained in real time, and calculates difference work based on described operating mode vector
Condition vector;
Monitoring step, according to the norm of described difference operating mode vector, chooses described HMM or each operating mode is corresponding
Statistical model analyze the corresponding monitor control index of model, and selected monitor control index is calculated in real time, will be calculated
The limit comparison of the control corresponding to monitor control index of real-time monitor control index and this model, monitor the operation conditions of this operating mode, institute
State monitoring step and judge that operating mode process is the most normal by following steps further:
If the norm of described difference operating mode vector is zero, then calculate the statistical model of operating mode vector corresponding to this difference operating mode vector
Analyze real-time first monitor control index of model and real-time second monitor control index, at described real-time first monitor control index more than the first control
System limit or described real-time second monitor control index more than the second control in limited time, then judge that operating mode process occurs abnormal;
If the norm of described difference operating mode vector is not zero, then the real-time NLLP monitoring calculating described HMM refers to
Mark, prescribes a time limit more than the control of the monitor control index of described HMM at this NLLP monitor control index, then judges operating mode
Process occurs abnormal.
Monitoring method the most according to claim 1, it is characterised in that
In described monitoring step, if the norm of difference operating mode vectorThen according to following expression design condition q
Corresponding statistical model analyzes the real-time first monitor control index D of modelr(s) and real-time second monitor control index Dp(s):
Dp(s)=sTP(q)[Λ(q)]-1P(q)TS,
Wherein, r is residual error subspace, and p is principal component subspace, and s is the statistical model vector of the process data obtained in real time,Statistical model for setting up under operating mode q analyzes the projection matrix in residual error space, P in model(q)For set up under operating mode q
Statistical model analyzes the load matrix in model, Λ(q)Corresponding to pivot during the statistical model of foundation analyzes model under operating mode q
The diagonal matrix that is combined into of covariance matrix eigenvalue.
Monitoring method the most according to claim 2, it is characterised in that calculate process data by following formula
Statistical model vector s:
S=col [μ, Σ, Ξ],
Wherein, μ represents the mean vector in the w duration of acquisition in real time, and Σ represents that second moment, Ξ represent High Order Moment, col []
Representing the form that matrix arrangement becomes column vector, described second moment includes variance, covariance and autocorrelation coefficient.
Monitoring method the most according to claim 1, it is characterised in that
In described monitoring step, if the norm of difference operating mode vectorThen calculate described hidden horse according to equation below
The real-time NLLP monitor control index of Er Kefu model:
NLLP=-logPr (Onew|s*),
Wherein, Pr () represents the probability that event occurs, OnewFor the up-to-date process data obtained in real time, s*Represent OnewInstitute
The operating mode at place.
Monitoring method the most according to claim 1, it is characterised in that in described calculation procedure, based on obtaining in real time
The process data taken uses Viterbi algorithm to obtain described operating mode vector I=[i1,i2,…,iw]T, wherein, w represented number of passes
According to acquisition duration, ijExpression operating mode sequence number residing for w duration internal procedure data, wherein j=1 ..., w.
Monitoring method the most according to claim 5, it is characterised in that utilize following formula to calculate described difference
Operating mode vector:
Wherein,Function ψ () value at 0 is 1, remaining
Point value is 0.
7. a multiple operating modes process monitoring system based on monitor control index switching, including:
Acquisition module, it is for gathering the normal data under different operating mode as training sample set;
First acquisition module, it obtains HMM based on described training sample set, utilizes described training sample set meter
Calculate NLLP monitor control index, obtain the control corresponding to NLLP monitor control index according to the confidence level chosen and limit, wherein,
NLLP represents the negative log-likelihood probability of described HMM;
Second acquisition module, its training sample based on each operating mode sets up the statistical model of corresponding operating mode respectively and analyzes model,
And calculate each statistical model according to described training sample and analyze the first monitor control index of model and the second monitor control index, further according to
The first control limit and second that the confidence level chosen obtains each monitor control index the most corresponding controls limit;
Computing module, it is based on the process data design condition vector obtained in real time, and calculates difference based on described operating mode vector
Operating mode vector;
Monitoring module, it, according to the norm of described difference operating mode vector, chooses described HMM or each operating mode pair
The statistical model answered analyzes the corresponding monitor control index of model, and calculates selected monitor control index in real time, will calculate
The real-time monitor control index arrived and the limit comparison of the control corresponding to monitor control index of this model, monitor the operation conditions of this operating mode,
In described monitoring module, judge that operating mode process is the most normal by following steps further:
If the norm of described difference operating mode vector is zero, then calculate the statistical model of operating mode vector corresponding to this difference operating mode vector
Analyze real-time first monitor control index of model and real-time second monitor control index, at described real-time first monitor control index more than the first control
System limit or described real-time second monitor control index more than the second control in limited time, then judge that operating mode process occurs abnormal;
If the norm of described difference operating mode vector is not zero, then the real-time NLLP monitoring calculating described HMM refers to
Mark, prescribes a time limit more than the control of the monitor control index of described HMM at this NLLP monitor control index, then judges operating mode
Process occurs abnormal.
Monitoring system the most according to claim 7, it is characterised in that
Described computing module uses Viterbi algorithm to obtain described operating mode vector based on the process data obtained in real time
I=[i1,i2,…,iw]T, wherein, w represents the acquisition duration of process data, ijRepresent the work residing for w duration internal procedure data
Condition sequence number, wherein j=1 ..., w;
Obtaining after described operating mode vector, described computing module further with following formula calculate described difference operating mode to
Amount:Wherein,Function ψ () value at 0 is 1, remaining
Point value is 0.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310675045.4A CN103631145B (en) | 2013-12-11 | 2013-12-11 | Multiple operating modes process monitoring method and system based on monitor control index switching |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310675045.4A CN103631145B (en) | 2013-12-11 | 2013-12-11 | Multiple operating modes process monitoring method and system based on monitor control index switching |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103631145A CN103631145A (en) | 2014-03-12 |
CN103631145B true CN103631145B (en) | 2016-08-17 |
Family
ID=50212340
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310675045.4A Active CN103631145B (en) | 2013-12-11 | 2013-12-11 | Multiple operating modes process monitoring method and system based on monitor control index switching |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103631145B (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105573290B (en) * | 2015-12-16 | 2017-12-29 | 浙江中烟工业有限责任公司 | Cigarette factory ultrahigh speed case packing machine multiple operating modes process is monitored on-line and method for diagnosing faults |
CN105487524B (en) * | 2015-12-29 | 2017-12-08 | 浙江中烟工业有限责任公司 | Ultrahigh speed small box packer condition monitoring and diagnosis method with multi-state characteristic |
CN105718742A (en) * | 2016-01-22 | 2016-06-29 | 张燕 | Dynamic hidden danger early alarm method based on statistics |
CN108015665A (en) * | 2016-10-28 | 2018-05-11 | 中芯国际集成电路制造(上海)有限公司 | Machinery grinding machine abnormity detecting instrument and method for detecting, machinery grinding machine |
CN107147526A (en) * | 2017-05-12 | 2017-09-08 | 上海携程商务有限公司 | Intelligent network fault detection method and system |
CN107707601A (en) * | 2017-06-02 | 2018-02-16 | 语祯物联科技(上海)有限公司 | A kind of method and device of the connection equipment of monitoring in real time |
CN109491338B (en) * | 2018-11-09 | 2021-05-11 | 南通大学 | Multimode process quality-related fault diagnosis method based on sparse GMM |
CN109634240B (en) * | 2018-12-13 | 2021-03-09 | 宁波大学 | Dynamic process monitoring method based on novel dynamic principal component analysis |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5465321A (en) * | 1993-04-07 | 1995-11-07 | The United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration | Hidden markov models for fault detection in dynamic systems |
US6088658A (en) * | 1997-04-11 | 2000-07-11 | General Electric Company | Statistical pattern analysis methods of partial discharge measurements in high voltage insulation |
CN1655082A (en) * | 2005-01-27 | 2005-08-17 | 上海交通大学 | Non-linear fault diagnosis method based on core pivot element analysis |
CN101615186A (en) * | 2009-07-28 | 2009-12-30 | 东北大学 | A kind of BBS user's abnormal behaviour auditing method based on Hidden Markov theory |
CN102163255A (en) * | 2010-02-17 | 2011-08-24 | 通用汽车环球科技运作有限责任公司 | Health prognosis for complex system using fault modeling |
CN103235933A (en) * | 2013-04-15 | 2013-08-07 | 东南大学 | Vehicle abnormal behavior detection method based on Hidden Markov Model |
-
2013
- 2013-12-11 CN CN201310675045.4A patent/CN103631145B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5465321A (en) * | 1993-04-07 | 1995-11-07 | The United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration | Hidden markov models for fault detection in dynamic systems |
US6088658A (en) * | 1997-04-11 | 2000-07-11 | General Electric Company | Statistical pattern analysis methods of partial discharge measurements in high voltage insulation |
CN1655082A (en) * | 2005-01-27 | 2005-08-17 | 上海交通大学 | Non-linear fault diagnosis method based on core pivot element analysis |
CN101615186A (en) * | 2009-07-28 | 2009-12-30 | 东北大学 | A kind of BBS user's abnormal behaviour auditing method based on Hidden Markov theory |
CN102163255A (en) * | 2010-02-17 | 2011-08-24 | 通用汽车环球科技运作有限责任公司 | Health prognosis for complex system using fault modeling |
CN103235933A (en) * | 2013-04-15 | 2013-08-07 | 东南大学 | Vehicle abnormal behavior detection method based on Hidden Markov Model |
Non-Patent Citations (5)
Title |
---|
Statistics Pattern Analysis: A New Process Monitoring Framework and its Application to Semiconductor Batch Processes;Q. Peter He et al.;《AIChE Journal》;20110131;第57卷(第1期);第107-121页 * |
动态系统的故障诊断方法;张萍 等;《控制理论与应用》;20000430;第17卷(第2期);第153-158页 * |
基于HMM的复杂条件故障诊断技术研究;岳夏;《中国博士学位论文全文数据库 工程科技Ⅱ辑》;20121115(第11期);全文 * |
基于统计模量分析间歇过程故障检测方法研究;张成 等;《仪器仪表学报》;20130930;第34卷(第9期);第2103-2110页 * |
门吉芳 等.基于模型的故障诊断方法研究.《机械管理开发》.2010,第25卷(第4期), * |
Also Published As
Publication number | Publication date |
---|---|
CN103631145A (en) | 2014-03-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103631145B (en) | Multiple operating modes process monitoring method and system based on monitor control index switching | |
CN106872657B (en) | A kind of multivariable water quality parameter time series data accident detection method | |
CN107301884B (en) | A kind of hybrid nuclear power station method for diagnosing faults | |
CN110738274A (en) | nuclear power device fault diagnosis method based on data driving | |
CN103713628B (en) | Fault diagnosis method based on signed directed graph and data constitution | |
CN105700518A (en) | Fault diagnosis method during industrial process | |
CN105259895A (en) | Method and monitoring system for detecting and separating micro fault in industrial process | |
CN100517141C (en) | System and method for detecting date and diagnosing failure of propylene polymerisation production | |
CN107357275A (en) | Non-gaussian industrial process fault detection method and system | |
CN105846780A (en) | Decision tree model-based photovoltaic assembly fault diagnosis method | |
CN110977614B (en) | Health diagnosis method for numerical control machine tool | |
CN104360677A (en) | Cigarette processing quality evaluation and diagnosis method | |
CN107065834B (en) | The method for diagnosing faults of concentrator in hydrometallurgy process | |
CN104035431B (en) | The acquisition methods of kernel functional parameter and system for non-linear process monitoring | |
CN102789676B (en) | Method for designing industrial alarm on basis of alarm evidence fusion | |
CN103499921A (en) | Fault diagnosis method for variable structure fuzzy system sensor and application thereof in flight control system | |
CN110083860A (en) | A kind of industrial method for diagnosing faults based on correlated variables selection | |
CN104199441B (en) | Blast furnace multi-state fault separating method based on sparse contribution plot and system | |
CN110348150A (en) | A kind of fault detection method based on dependent probability model | |
CN108536128A (en) | A kind of machine learning fault diagnosis system of parameter optimization | |
CN107862324B (en) | MWSPCA-based CBR prediction model intelligent early warning method | |
CN105974273A (en) | Power distribution network fault positioning system | |
CN105629958A (en) | Intermittence process fault diagnosis method based on sub-period MPCA-SVM | |
CN102495318A (en) | Fault diagnosis method of capacitive equipment | |
CN104142680B (en) | Multiple-sensor fault diagnosis system and method based on robust input training neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |