CN103631145A - Monitoring index switching based multi-operating-mode process monitoring method and system - Google Patents
Monitoring index switching based multi-operating-mode process monitoring method and system Download PDFInfo
- Publication number
- CN103631145A CN103631145A CN201310675045.4A CN201310675045A CN103631145A CN 103631145 A CN103631145 A CN 103631145A CN 201310675045 A CN201310675045 A CN 201310675045A CN 103631145 A CN103631145 A CN 103631145A
- Authority
- CN
- China
- Prior art keywords
- operating mode
- monitor control
- control index
- real
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Testing And Monitoring For Control Systems (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a monitoring index switching based multi-operating-mode process monitoring method and system. The method includes: acquiring normal data in different operating modes to serve as a training sample set; obtaining a hidden Markov model on the basis of the training sample set, and acquiring control limits corresponding to monitoring indexes of the hidden Markov model; respectively establishing statistical pattern analysis models of corresponding operating modes on the basis of training samples of each operating mode, and acquiring control limits corresponding to monitoring indexes of each statistical pattern analysis model; computing operating mode vectors on the basis of process data acquired in real time, and further computing differential operating mode vectors; computing corresponding real-time monitoring indexes according to norms of the differential operating mode vectors, and comparing the real-time monitoring indexes with the control limits corresponding to the monitoring indexes of the corresponding models so as to monitor running states of the operating modes. The method has the advantages that the process data are acquired in real time, reliability in monitoring is guaranteed, data in each operating mode need not to obey Gaussian distribution, and applicability is higher.
Description
Technical field
The present invention relates to process monitoring field, relate in particular to a kind of multiple operating modes process method for supervising and system of switching based on monitor control index.
Background technology
For process monitoring and troubleshooting issue, traditional method adopts multivariate statistics process control technology (Multivariable Statistical Process Control mostly, MSPC), wherein with pivot analysis (Principal Component Analysis, PCA) and offset minimum binary (Partial Least Squares, PLS) in industrial process monitoring, be successfully applied for the method for representative.Traditional MSPC method all supposes that process operation is under single operation operating mode, but in fact due to the switching frequently in a plurality of operating modes of being everlasting of the reason processes such as product changes, production capacity adjustment.
Based on multiple operating modes process method for supervising such as pivot analysis and Support Vector data descriptions, all suppose the data Gaussian distributed of each operating mode, but this might not set up in practice.Although and a plurality of operating modes are set up to unified model and compare that to set up the method for a plurality of models simple, yet lacking the identification to operating mode in real time, this can cause monitoring the working condition of current device.Although and the Gauss of non-tentation data of multiple operating modes process method for supervising based on rarefaction representation, the method does not have to consider the dynamic perfromance in chemical process.In addition, single current data being judged to operating mode under it is subject to the impact of noise may be inaccurate.
Summary of the invention
One of technical matters to be solved by this invention is that a kind of multiple operating modes process method for supervising switching based on monitor control index need to be provided, and whether it belongs to same operating according to the process data of Real-time Obtaining judges whether this process breaks down accordingly.In addition, also provide a kind of multiple operating modes process supervisory system of switching based on monitor control index.
In order to solve the problems of the technologies described above, the invention provides a kind of multiple operating modes process method for supervising switching based on monitor control index, comprising: acquisition step, gathers normal data under different operating modes as training sample set; The first obtaining step, obtains Hidden Markov Model (HMM) based on described training sample set, and obtains the corresponding control limit of monitor control index of described Hidden Markov Model (HMM); The second obtaining step, the training sample based on each operating mode is set up respectively the statistical model analytical model of corresponding operating mode, and obtains the corresponding control limit of monitor control index of each statistical model analytical model; Calculation procedure, the process data design condition vector based on Real-time Obtaining, and based on described operating mode vector calculation difference operating mode vector; Monitoring step, according to the norm of described difference operating mode vector, choose the corresponding monitor control index of statistical model analytical model corresponding to described Hidden Markov Model (HMM) or each operating mode, and selected monitor control index is calculated in real time, the corresponding control limit of the monitor control index of the real-time monitor control index calculating and this model is compared, monitored the operation conditions of this operating mode.
In one embodiment, the monitor control index of described statistical model analytical model further comprises the first monitor control index and the second monitor control index, control corresponding to described the first monitor control index is limited to the first control limit, and control corresponding to described the second monitor control index is limited to the second control limit; Described monitoring step further judges that by following steps whether operating mode process is normal: if the norm of described difference operating mode vector is zero, calculate real-time the first monitor control index and real-time second monitor control index of the statistical model analytical model of the operating mode vector that this difference operating mode vector is corresponding, at described real-time the first monitor control index, be greater than the first control limit or described real-time the second monitor control index is greater than the second control in limited time, judge operating mode process and occur abnormal; If the norm of described difference operating mode vector is non-vanishing, calculate the real-time NLLP monitor control index of described Hidden Markov Model (HMM), the control that is greater than the monitor control index of described Hidden Markov Model (HMM) at this NLLP monitor control index is prescribed a time limit, judging operating mode process occurs abnormal, wherein, NLLP represents the negative log-likelihood probability of described Hidden Markov Model (HMM).
In one embodiment, in described monitoring step, if the norm of difference operating mode vector || ▽ I||=0, according to the real-time first monitor control index D of the corresponding statistical model analytical model of following expression design condition q
rand real-time the second monitor control index D (s)
p(s):
D
p(s)＝s
TP
(q)[Λ
(q)]
-1P
(q)Ts，
Wherein, r is residual error subspace, and p is pivot subspace, the statistical model vector of the process data that s is Real-time Obtaining,
for the projection matrix in residual error space in the statistical model analytical model of setting up under operating mode q, P
(q)for the load matrix in the statistical model analytical model of setting up under operating mode q, Λ
(q)for the diagonal matrix that in the statistical model analytical model of setting up under operating mode q, the corresponding covariance matrix eigenwert of pivot is combined into.
In one embodiment, by the statistical model vector s:s=col[μ of following formula computation process data, Σ, Ξ], wherein, μ represents the mean vector in the w duration of Real-time Obtaining, Σ represents second moment, Ξ represents High Order Moment, col[] represent matrix to be arranged in the form of column vector, described second moment comprises variance, covariance and coefficient of autocorrelation.
In one embodiment, in described monitoring step, if the norm of difference operating mode vector || ▽ I|| ≠ 0, the following formula of basis calculates the real-time NLLP monitor control index of described Hidden Markov Model (HMM):
NLLP＝-logPr(O
new|s
*)，
Wherein, the probability that Pr () presentation of events occurs, O
newfor the up-to-date process data of Real-time Obtaining, s
*represent O
newresiding operating mode.
In one embodiment, in described calculation procedure, the process data based on Real-time Obtaining is used Viterbi algorithm to obtain described operating mode vector I=[i
1, i
2..., i
w]
t, wherein, w represents the duration that obtains of process data, i
j(j=1 ..., w) represent the residing operating mode sequence number of w duration internal procedure data.
In one embodiment, utilize following formula to calculate described difference operating mode vector:
▽ I=[▽ i
1, ▽ i
2..., ▽ i
w-1]
t, wherein, ▽ i
j=1-ψ (i
j+1-i
j), function ψ () is 1 in 0 place's value, all the other some values are 0.
According to a further aspect in the invention, also provide a kind of multiple operating modes process supervisory system of switching based on monitor control index, having comprised: acquisition module, it is for gathering normal data under different operating modes as training sample set; The first acquisition module, it obtains Hidden Markov Model (HMM) based on described training sample set, and obtains the corresponding control limit of monitor control index of described Hidden Markov Model (HMM); The second acquisition module, its training sample based on each operating mode is set up respectively the statistical model analytical model of corresponding operating mode, and obtains the corresponding control limit of monitor control index of each statistical model analytical model; Computing module, its process data design condition vector based on Real-time Obtaining, and based on described operating mode vector calculation difference operating mode vector; Monitoring module, it is according to the norm of described difference operating mode vector, choose the corresponding monitor control index of statistical model analytical model corresponding to described Hidden Markov Model (HMM) or each operating mode, and selected monitor control index is calculated in real time, the corresponding control limit of the monitor control index of the real-time monitor control index calculating and this model is compared, monitored the operation conditions of this operating mode.
In one embodiment, the monitor control index of described statistical model analytical model further comprises the first monitor control index and the second monitor control index, control corresponding to described the first monitor control index is limited to the first control limit, and control corresponding to described the second monitor control index is limited to the second control limit; In described monitoring module, further by following steps, judge that whether operating mode process is normal: if the norm of described difference operating mode vector is zero, calculate real-time the first monitor control index and real-time second monitor control index of the statistical model analytical model of the operating mode vector that this difference operating mode vector is corresponding, at described real-time the first monitor control index, be greater than the first control limit or described real-time the second monitor control index is greater than the second control in limited time, judge operating mode process and occur abnormal; If the norm of described difference operating mode vector is non-vanishing, calculate the real-time NLLP monitor control index of described Hidden Markov Model (HMM), the control that is greater than the monitor control index of described Hidden Markov Model (HMM) at this NLLP monitor control index is prescribed a time limit, judging operating mode process occurs abnormal, wherein, NLLP represents the negative log-likelihood probability of described Hidden Markov Model (HMM).
In one embodiment, the process data of described computing module based on Real-time Obtaining used Viterbi algorithm to obtain described operating mode vector I=[i
1, i
2..., i
w]
t, wherein, w represents the duration that obtains of process data, i
j(j=1 ..., w) represent the residing operating mode sequence number of w duration internal procedure data.
Compared with prior art, one or more embodiment of the present invention can have the following advantages by tool:
The present invention calculates respectively different control limits according to all normal data under different operating modes and the normal data of same operating, and Real-time Obtaining process data is calculated difference operating mode vector, the last norm based on difference operating mode vector selects to switch to suitable monitor control index, by contrasting the control that this monitor control index is corresponding with it, limit to judge that whether this process is normal, the method Real-time Obtaining process data has guaranteed the reliability of judgement, and do not need the data Gaussian distributed under each operating mode, there is higher applicability.
Other features and advantages of the present invention will be set forth in the following description, and, partly from instructions, become apparent, or understand by implementing the present invention.Object of the present invention and other advantages can be realized and be obtained by specifically noted structure in instructions, claims and accompanying drawing.
Accompanying drawing explanation
Accompanying drawing is used to provide a further understanding of the present invention, and forms a part for instructions,, jointly for explaining the present invention, is not construed as limiting the invention with embodiments of the invention.In the accompanying drawings:
Fig. 1 is the process flow diagram of the multiple operating modes process method for supervising that switches based on monitor control index according to an embodiment of the invention;
Fig. 2 is the schematic diagram of the real-time process data acquisition of one example according to the present invention;
Fig. 3 is the block diagram of the multiple operating modes process supervisory system switched based on monitor control index according to an embodiment of the invention;
Fig. 4 is the curve map of the testing result of finite mixtures Gauss model method in the test case 1 of one example according to the present invention;
Fig. 5 is the curve map of the testing result of SPA method in the test case 1 of one example according to the present invention;
Fig. 6 is the curve map of the testing result of the multiple operating modes process method for supervising that switches based on monitor control index in the test case 1 of one example according to the present invention;
Fig. 7 is the curve map of the testing result of finite mixtures Gauss model method in the test case 2 of one example according to the present invention;
Fig. 8 is the curve map of the testing result of SPA method in the test case 2 of one example according to the present invention;
Fig. 9 is the curve map of the testing result of the multiple operating modes process method for supervising that switches based on monitor control index in the test case 2 of one example according to the present invention;
Figure 10 is according to the structural representation of the continuous stirring heating tank of another example of the present invention;
Figure 11 is according to the curve map of the testing result of the finite mixtures Gauss model method of another example of the present invention;
Figure 12 is according to the curve map of the testing result of the SPA method of another example of the present invention;
Figure 13 is according to the curve map of the testing result of the multiple operating modes process method for supervising switching based on monitor control index of another example of the present invention.
Embodiment
For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing, the present invention is described in further detail.
the first embodiment
Fig. 1 is the process flow diagram of the multiple operating modes process method for supervising that switches based on monitor control index according to an embodiment of the invention.Below in conjunction with Fig. 1, the method is elaborated.
Step S110, gathers normal data under different operating modes (being power-equipment duty under certain condition) as training sample set.
Specifically, from chemical process database, obtain normal data under different operating modes as training sample set:
wherein,
(i=1 ... M) be the data sample of i operating mode,
the real number matrix that represents the capable m row of N, N
ithe number of samples that represents i operating mode, N represents total number of samples, m represents the number of sensor.
Step S120, obtains Hidden Markov Model (HMM) based on training sample set, and obtains the corresponding control limit of monitor control index of this Hidden Markov Model (HMM); Training sample based on each operating mode is set up respectively the statistical model analytical model (being designated hereinafter simply as SPA model) of corresponding operating mode, and obtains the corresponding different control limit of monitor control index of each SPA model.
In detail, the present embodiment obtains the control limit δ of the NLLP monitor control index of Hidden Markov Model (HMM)
nLLP, NLLP represents the negative log-likelihood probability of Hidden Markov Model (HMM).Particularly, utilize training sample set B, use EM algorithm (being greatest hope algorithm) can be trained the control limit δ that obtains the parameter set λ of Hidden Markov Model (HMM) and obtain NLLP monitor control index
nLLP.Conventionally, utilize normal training sample set to calculate NLLP monitor control index, according to the confidence level of choosing (being 98% confidence level such as what choose in concrete instance), can controlled limit δ
nLLP.
For SPA model, utilize the training sample B of each operating mode
iset up the SPA model under corresponding operating mode, the monitor control index of SPA model further comprises the first monitor control index D
r(being residual error subspace monitor control index) and the second monitor control index D
p(being pivot subspace monitor control index), the corresponding control limit of monitor control index of SPA model further comprises the first control limit δ
rwith the second control limit δ
p.With δ
nLLPobtain similarly, according to normal training sample, calculate D
rand D
pmonitor control index, then according to the confidence level chosen (being 98% confidence level such as what choose in concrete instance), can controlled limit δ
rand δ
p.
In addition, SPA method is based on central limit theorem, and the thought of central limit theorem is " no matter the distribution of stochastic variable how, the statistic of stochastic variable is Gaussian distributed progressively ", so the method for the present embodiment does not need the data Gaussian distributed under each operating mode.
Step S130, the process data design condition vector based on Real-time Obtaining, and based on this operating mode vector calculation difference operating mode vector.
In the present embodiment, by the process data imagery of Real-time Obtaining at one slidably in window, the process data of calculating is here the data in window, by move forward the process data of renewable window interior along sampling time axle moving window, the duration of Real-time Obtaining data is the length of moving window.Particularly, the process data based on Real-time Obtaining is used Viterbi algorithm to obtain operating mode vector I
k=[i
1k, i
2k..., i
wk]
t, wherein, w represents the duration that obtains of process data, i
jk(j=1 ..., w) represent the residing operating mode sequence number of k moving window (being current moving window) interior data.
Fig. 2 is the schematic diagram of the real-time process data acquisition of one example according to the present invention.Easily understand, two moving windows in Fig. 2 do not exist simultaneously, but moving window I comprises the process data in sampling time section [10,50], and moving window II comprises the process data in sampling time section [70,110].
Further, the expression formula of difference operating mode vector is ▽ I
k=[▽ i
1k, ▽ i
2k..., ▽ i
(w-1) k]
t, wherein, ▽ i
jk=1-ψ (i
(j+1) k-i
jk), i
jk(j=1 ..., w) representing the residing operating mode sequence number of current moving window (being moving window k) interior data, function ψ () is 1 in 0 place's value, all the other some values are 0.
Step S140, norm based on above-mentioned difference operating mode vector, choose the corresponding monitor control index of SPA model corresponding to Hidden Markov Model (HMM) or each operating mode, and selected monitor control index is calculated in real time, the corresponding control limit of the monitor control index of the real-time monitor control index calculating and this model is compared, monitored the operation conditions of this operating mode.
First, judge the norm of difference operating mode vector || whether ▽ I|| is zero.If the norm of difference operating mode vector || ▽ I|| is zero, illustrates that current obtained process data belongs to same operating, and operating mode does not change, as shown in moving window I in Fig. 2.If the norm of difference operating mode vector || ▽ I|| is non-vanishing, illustrates that current obtained process data belongs to different operating modes, as shown in moving window II in Fig. 2.
Then, in the norm of difference operating mode vector || when ▽ I|| is zero, the real-time monitor control index that calculates the operating mode that this difference operating mode vector is corresponding (is the real-time monitor control index D of SPA model
rand D (s)
p(s)), at real-time the first monitor control index D
r(s) be greater than its corresponding control limit δ
rtime or real-time the second monitor control index D
p(s) be greater than its corresponding control limit δ
p, judge operating mode process and occur abnormal.By following expression, carry out the monitor control index of the SPA model of design condition q:
D
p(s)＝s
TP
(q)[Λ
(q)]
-1P
(q)Ts，
Wherein, r is residual error subspace, and p is pivot subspace, the statistical model vector of the process data that s is Real-time Obtaining,
for the projection matrix in residual error space in the SPA model of setting up under operating mode q, P
(q)for the load matrix in the SPA model of setting up under operating mode q, Λ
(q)for the diagonal matrix that in the SPA model of setting up under operating mode q, the corresponding covariance matrix eigenwert of pivot is combined into.Particularly, by the statistical model vector s of following formula computation process data:
s＝col[μ,Σ,Ξ]
Wherein, μ represents the mean vector in the w duration of Real-time Obtaining, and Σ represents second moment, and Ξ represents High Order Moment, symbol col[] represent matrix to be arranged in the form of column vector, above-mentioned second moment comprises variance, covariance and coefficient of autocorrelation.
Norm at difference operating mode vector || when ▽ I|| is non-vanishing, calculate the real-time NLLP monitor control index of Hidden Markov Model (HMM), the corresponding control limit of the monitor control index δ that is greater than Hidden Markov Model (HMM) at this NLLP monitor control index
nLLPtime, judge operating mode process and occur abnormal.Specifically by following formula, obtain real-time NLLP monitor control index:
NLLP＝-logPr(O
new|s
*)，
Wherein, the probability that Pr () presentation of events occurs, O
newfor the up-to-date process data of Real-time Obtaining, s
*represent O
newresiding operating mode.In fact, because the calculating of NLLP monitor control index depends on operating mode, so the also difference to some extent of the control of the NLLP index under different operating modes limit.
In one example, also can pass through formula NLLP=-logPr (O
new| λ) calculate NLLP monitor control index, the complete parameter set that wherein λ is Hidden Markov Model (HMM).
It should be noted that, due to according to the difference of difference operating mode vector, different monitoring index is different at the appropriateness of current time, the present embodiment carries out index switching intelligently between NLLP monitor control index and SPA Model Monitoring index, rather than uses on earth as a monitor control index in existing method for supervising always.As mentioned above, when the divided working status vector of being on duty is non-vanishing, this stylish floor data quantity not sufficient that enters moving window, cannot " fill " window, the statistic of obtaining is inaccurate, causes the rate of false alarm of fault to increase, and therefore selects the NLLP monitor control index of Hidden Markov Model (HMM).When the divided working status vector of being on duty is zero, illustrate that data are in being same as operating mode, the statistic of obtaining is accurate, monitor more comprehensive, so two indexs in selection SPA model.
Finally, output procedure monitored results, to facilitate the operating mode to breaking down to safeguard.
example
In order further to understand the present invention, below two examples are described.
Example one: numerical simulation
By linear system below, produce data:
X=[x wherein
1x
2x
3]
trepresent the measurement of three sensors, s=[s
1s
2]
trepresent two independently data sources, e=[e
1e
2e
3]
trepresent independently the Gaussian noise that three averages are 0, standard deviation is 0.1.The system of listing below when three different operating modes, the probability distribution that data source is obeyed:
Operating mode 1:s
1: N (10,0.8); s
2: N (12,1.3)
Operating mode 2:s
1: N (5,0.6); s
2: N (20,0.7)
Operating mode 3:s
1: N (16,1.5); s
2: N (30,2.5)
N (μ, σ wherein
2) expression average is μ, variance is σ
2gaussian distribution.Description to the test data of three kinds of different operating modes is as shown in table 1.Wherein, skew fault is the steady state value deviation on sensor, and to be amplitude on sensor taper to large deviation from little to drifting fault.
The description of test data in table 1. numerical simulation
In order to verify the validity of this example and in the advantage aspect fault detect rate and rate of false alarm, the method (not having index to switch) of having chosen finite mixtures Gauss model (FGMM) method and statistical model analysis (SPA) is method as a comparison.Fig. 4-Fig. 9 has shown the schematic diagram of the whole bag of tricks for the fault detect result of test case 1 and test case 2.Wherein, the BIP index in Fig. 4, Fig. 7 is the Bayesian inference probability level in limited gauss hybrid models, the D in Fig. 5, Fig. 8
pindex is the monitor control index of principal component space in SPA model, D
rindex is the monitor control index in residual error space in SPA model, the D in Fig. 6, Fig. 9
p-NLLP index is D
pindex and NLLP index are switched, D
r-NLLP index is D
rindex and NLLP index are switched.
Table 2 is depicted as the fault diagnosis rate of the whole bag of tricks and the summary of rate of false alarm.
The comparison of distinct methods fault detect rate (FDR) and rate of false alarm (FAR) in table 2. example one
Indicate: data representation percentage (%) in table.
From upper table, be not difficult to find out that method of the present invention compares additive method and have higher fault detect rate, and have the advantages that rate of false alarm is low.
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 flow transmitter, and FC is flow controller, and TT is temperature transmitter, and LC is fluid level controller, and LT is fluid level transmitter.
In continuous stirring heating tank, hot water and cold water fully mix and by steam, it are heated in tank.In system, have a plurality of control loops, thereby guarantee liquid level, flow and temperature are operated in the working point of setting.Table 3 has reacted two corresponding parameters of standard condition of continuous stirring heating tank, and the unit of each physical measurement here (electric signal) is milliampere (mA).
Two corresponding parameters of standard condition of table 3. continuous stirring heating tank
Table 4 is the descriptions to test data in continuous stirring heating tank.For level sensor fault, compare the detection effect of distinct methods below.
The description of test data in table 4. continuous stirring heating tank
Figure 11-13 have shown fault diagnosis rate and the rate of false alarm of the whole bag of tricks.A same example, the BIP index in Figure 11 is the Bayesian inference probability level in limited gauss hybrid models, the D in Figure 12
pindex is the monitor control index of principal component space in SPA model, D
rindex is the monitor control index in residual error space in SPA model, the D in Figure 13
p-NLLP index is D
pindex and NLLP index are switched, D
r-NLLP index is D
rindex and NLLP index are switched.
The summary of concrete fault diagnosis rate and rate of false alarm is as shown in table 5.
The comparison of distinct methods fault detect rate (FDR) and rate of false alarm (FAR) in table 5. example two
Indicate: data representation percentage (%) in table.
From upper table, be not difficult to find out that method of the present invention compares additive method and have higher fault detect rate, and have the advantages that rate of false alarm is low.
In sum, the present invention obtains respectively different control limits according to all normal data under different operating modes and the normal data of same operating, and calculate difference operating mode vector according to the process data of Real-time Obtaining, the last norm based on difference operating mode vector is switched to more applicable monitor control index, select to calculate the more excellent monitor control index switching to, by contrasting the control that this monitor control index is corresponding with it, limit to judge that whether this process is normal, the method Real-time Obtaining process data has guaranteed the reliability of judgement, and do not need the data Gaussian distributed under each operating mode, there is higher applicability.
the second embodiment
Fig. 3 is the block diagram of the multiple operating modes process supervisory system switched based on monitor control index according to an embodiment of the invention.Below in conjunction with Fig. 3, the supervisory system of the present embodiment is elaborated.
As shown in Figure 3, native system comprises collection model 30, the first acquisition module 31, the second acquisition module 32, computing module 33 and monitoring module 34.Acquisition module 30, the first acquisition module 31, the second acquisition module 32, computing module 33 and the monitoring module 34 of the present embodiment carried out respectively the operation of the step S110 to S140 of the first embodiment.
In detail, acquisition module 30 is for gathering normal data under different operating modes as training sample set.Particularly, acquisition module 30 obtains normal data under different operating modes as training sample set from chemical process database:
wherein,
(i=1 ... M) be the data sample of i operating mode,
the real number matrix that represents the capable m row of N, N
ithe number of samples that represents i operating mode, N represents total number of samples, m represents the number of sensor.
Training sample set the first acquisition module 31 gathering based on acquisition module 30 is set up Hidden Markov Model (HMM), and obtains the corresponding control limit of monitor control index of this Hidden Markov Model (HMM).
Conventionally, utilize normal training sample set to calculate the NLLP monitor control index of Hidden Markov Model (HMM), according to the confidence level of choosing (being 98% confidence level such as what choose in concrete instance), can controlled limit δ
nLLP.Particularly, utilize training sample set B, use EM algorithm (being greatest hope algorithm) can be trained the control limit δ that obtains the parameter set λ of Hidden Markov Model (HMM) and obtain NLLP monitor control index
nLLP.
In addition, the training sample of each operating mode that the second acquisition module 32 gathers based on acquisition module 30 is set up respectively the statistical model analytical model (being SPA model) of corresponding operating mode, and obtains the corresponding control limit of monitor control index of each SPA model.
The monitor control index of SPA model further comprises the first monitor control index D
r(being residual error subspace monitor control index) and the second monitor control index D
p(being pivot subspace monitor control index), the corresponding control limit of monitor control index of SPA model further comprises the first control limit δ
rwith the second control limit δ
p.With δ
nLLPobtain similarly, according to normal training sample, calculate D
rand D
pmonitor control index, then according to the confidence level chosen (being 98% confidence level such as what choose in concrete instance), can controlled limit δ
rand δ
p.
Its process data based on Real-time Obtaining of 33 of computing modules is carried out design condition vector, and based on this operating mode vector calculation difference operating mode vector.
In detail, the process data of computing module 33 based on Real-time Obtaining used Viterbi algorithm to obtain operating mode vector I=[i
1, i
2..., i
w]
t, wherein, w represents the duration that obtains of process data, i
j(j=1 ..., w) represent the residing operating mode sequence number of w duration internal procedure data.
After obtaining this operating mode vector, computing module 33 further utilizes following formula to calculate difference operating mode vector: ▽ I=[▽ i
1, ▽ i
2..., ▽ i
w-1]
t, wherein, ▽ i
j=1-ψ (i
j+1-i
j), function ψ () is 1 in 0 place's value, all the other some values are 0.
Finally, monitoring module 34 is according to the norm of above-mentioned difference operating mode vector, choose the corresponding monitor control index of SPA model corresponding to Hidden Markov Model (HMM) or each operating mode, and selected monitor control index is calculated in real time, the corresponding control limit of the monitor control index of the real-time monitor control index calculating and this model is compared, monitored the operation conditions of this operating mode.
It should be noted that, according to the difference of difference operating mode vector, different monitoring index is different at the appropriateness of current time, the present embodiment carries out index switching intelligently between NLLP monitor control index and SPA Model Monitoring index, rather than uses on earth as a monitor control index in existing method for supervising always.
Specifically, if the norm of difference operating mode vector is zero, real-time the first monitor control index and real-time the second monitor control index that calculate the SPA model of the operating mode vector that this difference operating mode vector is corresponding (are the real-time monitor control index D of SPA model
rand D (s)
p(s)), at real-time the first monitor control index D
r(s) be greater than its corresponding control limit δ
rtime or real-time the second monitor control index D
p(s) be greater than its corresponding control limit δ
p, judge operating mode process and occur abnormal.By following expression, carry out the monitor control index of the SPA model of design condition q:
D
p(s)＝s
TP
(q)[Λ
(q)]
-1P
(q)Ts，
Wherein, r is residual error subspace, and p is pivot subspace, the statistical model vector of the process data that s is Real-time Obtaining,
for the projection matrix in residual error space in the SPA model of setting up under operating mode q, P
(q)for the load matrix in the SPA model of setting up under operating mode q, Λ
(q)the diagonal matrix being combined into for the corresponding covariance matrix eigenwert of the pivot in the SPA model of setting up under operating mode q.Particularly, by the statistical model vector s of following formula computation process data:
s＝col[μ,Σ,Ξ]
Wherein, μ represents the mean vector in the w duration of Real-time Obtaining, and Σ represents second moment, and Ξ represents High Order Moment, symbol col[] represent matrix to be arranged in the form of column vector, above-mentioned second moment comprises variance, covariance and coefficient of autocorrelation.
If the norm of difference operating mode vector is non-vanishing, calculate the real-time NLLP monitor control index of Hidden Markov Model (HMM), the control that is greater than the monitor control index of Hidden Markov Model (HMM) at this NLLP monitor control index is prescribed a time limit, judging operating mode process occurs abnormal, wherein, NLLP represents the negative log-likelihood probability of Hidden Markov Model (HMM).Specifically by following formula, obtain real-time NLLP monitor control index:
NLLP＝-logPr(O
new|s
*)，
Wherein, the probability that Pr () presentation of events occurs, O
newfor the up-to-date process data of Real-time Obtaining, s
*represent O
newresiding operating mode.
In sum, the present invention obtains respectively different control limits according to all normal data under different operating modes and the normal data of same operating, and calculate difference operating mode vector according to the process data of Real-time Obtaining, the last norm based on difference operating mode vector is switched to more applicable monitor control index, select to calculate the more excellent monitor control index switching to, by contrasting the control that this monitor control index is corresponding with it, limit to judge that whether this process is normal, the method Real-time Obtaining process data has guaranteed the reliability of judgement, and do not need the data Gaussian distributed under each operating mode, there is higher applicability.
The above; be only specific embodiment of the invention case, protection scope of the present invention is not limited to this, is anyly familiar with those skilled in the art in technical manual of the present invention; to modification of the present invention or replacement, all should be within protection scope of the present invention.
Claims (10)
1. the multiple operating modes process method for supervising switching based on monitor control index, comprising:
Acquisition step, gathers normal data under different operating modes as training sample set;
The first obtaining step, obtains Hidden Markov Model (HMM) based on described training sample set, and obtains the corresponding control limit of monitor control index of described Hidden Markov Model (HMM);
The second obtaining step, the training sample based on each operating mode is set up respectively the statistical model analytical model of corresponding operating mode, and obtains the corresponding control limit of monitor control index of each statistical model analytical model;
Calculation procedure, the process data design condition vector based on Real-time Obtaining, and based on described operating mode vector calculation difference operating mode vector;
Monitoring step, according to the norm of described difference operating mode vector, choose the corresponding monitor control index of statistical model analytical model corresponding to described Hidden Markov Model (HMM) or each operating mode, and selected monitor control index is calculated in real time, the corresponding control limit of the monitor control index of the real-time monitor control index calculating and this model is compared, monitored the operation conditions of this operating mode.
2. method for supervising according to claim 1, is characterized in that,
The monitor control index of described statistical model analytical model further comprises the first monitor control index and the second monitor control index, and control corresponding to described the first monitor control index is limited to the first control limit, and control corresponding to described the second monitor control index is limited to the second control limit;
Described monitoring step further judges that by following steps whether operating mode process is normal:
If the norm of described difference operating mode vector is zero, calculate real-time the first monitor control index and real-time second monitor control index of the statistical model analytical model of the operating mode vector that this difference operating mode vector is corresponding, at described real-time the first monitor control index, be greater than the first control limit or described real-time the second monitor control index is greater than the second control in limited time, judge operating mode process and occur abnormal;
If the norm of described difference operating mode vector is non-vanishing, calculate the real-time NLLP monitor control index of described Hidden Markov Model (HMM), the control that is greater than the monitor control index of described Hidden Markov Model (HMM) at this NLLP monitor control index is prescribed a time limit, judging operating mode process occurs abnormal, wherein, NLLP represents the negative log-likelihood probability of described Hidden Markov Model (HMM).
3. method for supervising according to claim 2, is characterized in that,
In described monitoring step, if the norm of difference operating mode vector || ▽ I||=0, according to the real-time first monitor control index D of the corresponding statistical model analytical model of following expression design condition q
rand real-time the second monitor control index D (s)
p(s):
D
p(s)＝s
TP
(q)[Λ
(q)]
-1P
(q)Ts，
Wherein, r is residual error subspace, and p is pivot subspace, the statistical model vector of the process data that s is Real-time Obtaining,
for the projection matrix in residual error space in the statistical model analytical model of setting up under operating mode q, P
(q)for the load matrix in the statistical model analytical model of setting up under operating mode q, Λ
(q)for the diagonal matrix that in the statistical model analytical model of setting up under operating mode q, the corresponding covariance matrix eigenwert of pivot is combined into.
4. method for supervising according to claim 3, is characterized in that, the statistical model vector s by following formula computation process data:
s＝col[μ,Σ,Ξ]，
Wherein, μ represents the mean vector in the w duration of Real-time Obtaining, and Σ represents second moment, and Ξ represents High Order Moment, col[] represent matrix to be arranged in the form of column vector, described second moment comprises variance, covariance and coefficient of autocorrelation.
5. method for supervising according to claim 2, is characterized in that,
In described monitoring step, if the norm of difference operating mode vector || ▽ I|| ≠ 0, the following formula of basis calculates the real-time NLLP monitor control index of described Hidden Markov Model (HMM):
NLLP＝-logPr(O
new|s
*)，
Wherein, the probability that Pr () presentation of events occurs, O
newfor the up-to-date process data of Real-time Obtaining, s
*represent O
newresiding operating mode.
6. method for supervising according to claim 1, is characterized in that, in described calculation procedure, the process data based on Real-time Obtaining is used Viterbi algorithm to obtain described operating mode vector I=[i
1, i
2..., i
w]
t, wherein, w represents the duration that obtains of process data, i
j(j=1 ..., w) represent the residing operating mode sequence number of w duration internal procedure data.
7. method for supervising according to claim 6, is characterized in that, utilizes following formula to calculate described difference operating mode vector:
▽ I=[▽ i
1, ▽ i
2..., ▽ i
w-1]
t, wherein, ▽ i
j=1-ψ (i
j+1-i
j), function ψ () is 1 in 0 place's value, all the other some values are 0.
8. a multiple operating modes process supervisory system of switching based on monitor control index, comprising:
Acquisition module, it is for gathering normal data under different operating modes as training sample set;
The first acquisition module, it obtains Hidden Markov Model (HMM) based on described training sample set, and obtains the corresponding control limit of monitor control index of described Hidden Markov Model (HMM);
The second acquisition module, its training sample based on each operating mode is set up respectively the statistical model analytical model of corresponding operating mode, and obtains the corresponding control limit of monitor control index of each statistical model analytical model;
Computing module, its process data design condition vector based on Real-time Obtaining, and based on described operating mode vector calculation difference operating mode vector;
Monitoring module, it is according to the norm of described difference operating mode vector, choose the corresponding monitor control index of statistical model analytical model corresponding to described Hidden Markov Model (HMM) or each operating mode, and selected monitor control index is calculated in real time, the corresponding control limit of the monitor control index of the real-time monitor control index calculating and this model is compared, monitored the operation conditions of this operating mode.
9. supervisory system according to claim 8, it is characterized in that, the monitor control index of described statistical model analytical model further comprises the first monitor control index and the second monitor control index, control corresponding to described the first monitor control index is limited to the first control limit, and control corresponding to described the second monitor control index is limited to the second control limit;
In described monitoring module, further by following steps, judge that whether operating mode process is normal:
If the norm of described difference operating mode vector is zero, calculate real-time the first monitor control index and real-time second monitor control index of the statistical model analytical model of the operating mode vector that this difference operating mode vector is corresponding, at described real-time the first monitor control index, be greater than the first control limit or described real-time the second monitor control index is greater than the second control in limited time, judge operating mode process and occur abnormal;
If the norm of described difference operating mode vector is non-vanishing, calculate the real-time NLLP monitor control index of described Hidden Markov Model (HMM), the control that is greater than the monitor control index of described Hidden Markov Model (HMM) at this NLLP monitor control index is prescribed a time limit, judging operating mode process occurs abnormal, wherein, NLLP represents the negative log-likelihood probability of described Hidden Markov Model (HMM).
10. supervisory system according to claim 8, is characterized in that,
The process data of described computing module based on Real-time Obtaining used Viterbi algorithm to obtain described operating mode vector I=[i
1, i
2..., i
w]
t, wherein, w represents the duration that obtains of process data, i
j(j=1 ..., w) represent the residing operating mode sequence number of w duration internal procedure data;
After obtaining described operating mode vector, described computing module further utilizes following formula to calculate described difference operating mode vector: ▽ I=[▽ i
1, ▽ i
2..., ▽ i
w-1]
t, wherein, ▽ i
j=1-ψ (i
j+1-i
j), function ψ () is 1 in 0 place's value, all the other some values are 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 true CN103631145A (en) | 2014-03-12 |
| CN103631145B 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) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105487524A (en) * | 2015-12-29 | 2016-04-13 | 浙江中烟工业有限责任公司 | Method for monitoring and diagnosing super-high-speed small box packaging machine having multiple condition characteristic |
| CN105573290A (en) * | 2015-12-16 | 2016-05-11 | 浙江中烟工业有限责任公司 | Cigarette factory superspeed carton packaging machine multi-condition process online monitoring and fault diagnosis method |
| CN105718742A (en) * | 2016-01-22 | 2016-06-29 | 张燕 | Dynamic hidden danger early warning method based on statistics |
| 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 |
| 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 |
| CN109491338A (en) * | 2018-11-09 | 2019-03-19 | 南通大学 | A kind of relevant method for diagnosing faults of multimode procedure quality based on sparse GMM |
| CN109634240A (en) * | 2018-12-13 | 2019-04-16 | 宁波大学 | A kind of dynamic process monitoring method based on novel dynamic principal component analysis |
Family Cites Families (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 |
| CN101615186B (en) * | 2009-07-28 | 2012-07-04 | 东北大学 | BBS user abnormal behavior auditing method based on Hidden Markov theory |
| US8260493B2 (en) * | 2010-02-17 | 2012-09-04 | GM Global Technology Operations LLC | Health prognosis for complex system using fault modeling |
| CN103235933B (en) * | 2013-04-15 | 2016-08-03 | 东南大学 | A kind of vehicle abnormality behavioral value method based on HMM |
-
2013
- 2013-12-11 CN CN201310675045.4A patent/CN103631145B/en active Active
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105573290A (en) * | 2015-12-16 | 2016-05-11 | 浙江中烟工业有限责任公司 | Cigarette factory superspeed carton packaging machine multi-condition process online monitoring and fault diagnosis method |
| 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 |
| CN105487524A (en) * | 2015-12-29 | 2016-04-13 | 浙江中烟工业有限责任公司 | Method for monitoring and diagnosing super-high-speed small box packaging machine having multiple condition characteristic |
| 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 warning 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 |
| CN109491338A (en) * | 2018-11-09 | 2019-03-19 | 南通大学 | A kind of relevant method for diagnosing faults of multimode procedure quality based on sparse GMM |
| CN109634240A (en) * | 2018-12-13 | 2019-04-16 | 宁波大学 | A kind of dynamic process monitoring method based on novel dynamic principal component analysis |
| CN109634240B (en) * | 2018-12-13 | 2021-03-09 | 宁波大学 | Dynamic process monitoring method based on novel dynamic principal component analysis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103631145B (en) | 2016-08-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103631145A (en) | Monitoring index switching based multi-operating-mode process monitoring method and system | |
| 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 | |
| CN105259895A (en) | Method and monitoring system for detecting and separating micro fault in industrial process | |
| CN105242534B (en) | Based on telemetry parameter and it is associated with the satellitosis monitoring method to satellite controlling behavior | |
| CN109459993A (en) | A kind of process flow industry process online adaptive Fault monitoring and diagnosis method | |
| CN105955219A (en) | Distributed dynamic process fault detection method based on mutual information | |
| CN102062832B (en) | Power system online prewarning method based on microdisturbance signal low-frequency oscillation mode identification | |
| CN100517141C (en) | System and method for detecting date and diagnosing failure of propylene polymerisation production | |
| CN110008565A (en) | A kind of industrial process unusual service condition prediction technique based on operating parameter association analysis | |
| CN103389701B (en) | Based on the level of factory procedure fault Detection and diagnosis method of distributed data model | |
| CN102789676B (en) | Method for designing industrial alarm on basis of alarm evidence fusion | |
| CN106483847B (en) | A kind of water cooler fault detection method based on adaptive ICA | |
| CN112187528B (en) | Industrial control system communication flow online monitoring method based on SARIMA | |
| CN103488091A (en) | Data-driving control process monitoring method based on dynamic component analysis | |
| CN105404280A (en) | Industrial process fault detection method based on autoregression dynamic hidden variable model | |
| CN104615122B (en) | A kind of industry control signal detection system and detection method | |
| CN104615121B (en) | A kind of train fault diagnostic method and system | |
| CN107862324B (en) | MWSPCA-based CBR prediction model intelligent early warning method | |
| CN108257365B (en) | Industrial alarm design method based on global uncertainty evidence dynamic fusion | |
| CN103033309B (en) | Pressure transmitter with diagnosis | |
| CN103970124A (en) | On-line detection method for industrial control loop multi-period oscillation | |
| CN103617105B (en) | A kind of adaptive multi-layer flow model facility diagnosis method based on data-driven | |
| CN111368428A (en) | Sensor precision reduction fault detection method based on monitoring second-order statistics |
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 |