CN105893700A - Chemical production on-line fault detection and diagnosis technique based on physical-large data hybrid model - Google Patents
Chemical production on-line fault detection and diagnosis technique based on physical-large data hybrid model Download PDFInfo
- Publication number
- CN105893700A CN105893700A CN201610263402.XA CN201610263402A CN105893700A CN 105893700 A CN105893700 A CN 105893700A CN 201610263402 A CN201610263402 A CN 201610263402A CN 105893700 A CN105893700 A CN 105893700A
- Authority
- CN
- China
- Prior art keywords
- model
- data
- variable
- performance
- fault
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/70—Machine learning, data mining or chemometrics
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Chemical & Material Sciences (AREA)
- Crystallography & Structural Chemistry (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a chemical production on-line fault detection and diagnosis technique based on a physical-large data hybrid model. The chemical production on-line fault detection and diagnosis technique is characterized by comprising the following steps: after selecting a target operation unit, scanning all historical data of the unit, and after parameters are verified, establishing an accident knowledge base and a parameter model; in later on-line detection process, directly loading on-line data into the parameter model, scanning so as to obtain fault data in the on-line data, making an alarm, and comparing the fault data with data in the accident knowledge base, thereby obtaining fault reasons. The method disclosed by the invention is applied to confirmed single chemical unit operation, a reliable physical model can be established, a chemometrical method is used in the whole production process, a big-data processing technique is introduced to process real-time data of years, operators focus on faults of performance indexes within a controllable variable range, and the influence of uncontrollable production process variable to fault detection is eliminated.
Description
Technical field
The present invention relates to safety detection and control field in chemical process, particularly relate to a kind of based on physics-several
According to the Chemical Manufacture of mixed model online fault detection and diagnosis technology.
Background technology
Chemical process is extremely complex, along with the physical and chemical reaction much not yet verified.To continuous process work
For industry, maintain a stable operating mode, be possible not only to minimizing accident and secondary disaster, also to stabilized product quality, optimizing
Operating mode under carry out optimal cost operate to obtain maximum economic benefit.Thus fault referred herein, not only include impact dress
Put safe abnormal conditions (being referred to as safety failure), also include deviation (the being referred to as property of product quality and Optimum Operation operating mode
Energy fault).
For complicated chemical process, the most conventional fault detection and diagnosis method is the most all built upon data
On the basis of driving model.In 20 years of past, the professional of chemical industry and message area has carried out a large amount of useful exploration, especially
It is the construction in product practice storehouse before and after 2000 and popularizes, and is deep into chemical production field to informationization technology and brings
Opportunity, expedites the emergence of out a lot of application on site, also makes the application on site of fault detection and diagnosis technology be rapidly developed.The most often
Fault detection and diagnosis method be the most all built upon data-driven model on the basis of.
Statistical method chemometricses based on production history data model, i.e. by dividing a large amount of historical datas
Analysis, obtains the internal association of data, and note abnormalities (the typically destruction of internal association between manufacturing parameter, or whole behaviour
Make operating mode and have deviation relative to normal value) after, navigate on the unit of generation problem, equipment, or parameter accurately, by association
The synchronization of influence factor and analysis result is shown, provide in time for operator and management personnel, fault detect accurately with examine
Disconnected information.
Along with the rise of big data technique, can the statistical processing methods of the biggest data, big data technique compares tradition
Statistical method more highlight and only ask association, pay no attention to cause and effect, focus on the feature of relatedness between card analysis data, big data skill
Art, on solving factory's complexity, implying and be difficult between the variable measured affect relation, has in its distinctive feature, such as production unit A
Production status influences whether the quality (such as purity and sensitive impurity content) of unit A output intermediate products, if intermediate products
Quality cannot on-line testing or testing cost high, these implicit relations also can affect the production feelings of subsequent production unit B
Condition, by big data analysis technique, not only can be monitored the fault of unit A and examine, and in time series, can be right
The fault of production unit B carries out early warning, and reminds operator to make to adjust timely.But for Chemical Manufacture, a lot
Unit operates especially physical process (such as heat exchange, compression, rectification etc.), has the quality and energy balance relations determined, and
Before and after flow process, operating unit is commonly present physical isolation (as isolated two separative elements by relay reservoir), between physical parameter not
There is dependency physically, directly use big data analysis technique, because the noise effect existed between measurement data, or
Being affected by public work (such as power steam temperature, fuel gas pressure forms) impact, the most there are not the data of association in these
The model that big data are set up usually occurs the relatedness of height, allows big data analyst be mistakenly considered to there is association, this
The association planting violation physical law makes the False Rate of fault diagnosis increase, and causes production technology personnel to produce fault detect not
Trust, give and advance industry 4.0 and intelligent plant construction to bring certain resistance.
Summary of the invention
For the problem of above-mentioned existence, present invention aim at providing a kind of change based on physics-big data mixing model
Work produces online fault detection and diagnosis technology.
In order to achieve the above object, the technical solution used in the present invention is as follows: a kind of based on physics-big data mixing model
Chemical Manufacture online fault detection and diagnosis technology, the operational approach of described fault detection and diagnosis technology is as follows: choosing
After taking object run unit, by all historical datas of this unit are scanned, after checking parameter, set up accident knowledge
Storehouse and parameter model；During follow-up on-line checking, online data is importing directly in parameter model, scanned after
Go out the fault data in online data, send alarm, fault data is contrasted with data in accident knowledge base, must be out of order
Reason.
Chemical Manufacture online fault detection and diagnosis technology based on physics-big data mixing model of the present invention,
Its detailed operational approach comprises the following steps:
1) real-time data base (such as IP21, PI etc.) is utilized to collect the training data sample that the data composition of factory level process models
Collection: XRn×m.Wherein, n is the number of sample data set, and m is the variable number of sample data set；
2) for factory level process data collection XRn×m, extract the production unit Q determining physical model, set up this and produce
The physical model of unit, collects this production unit process variable at data subset Xq ∈ Rn×q, q determines physical model for having
Unit operation variable number.
It can also be multiple that production unit Q herein can be one, the most The more the better, more can react chemical industry
The physical laws of production process and make full use of engineering knowledge, rather than depend on merely analysis of statistical data.Factory level crosses number of passes
According to collection XRn×m, having the production unit determining physical model, using physical model to extract the key performance of this production unit
Index, is actually based on physical model and carries out dimensionality reduction, and i.e. according to the chemical composition of this production unit, material, heat are so that changing
Balance each other, chemical reaction equilibrium model, calculates this unit key contributions variable to target capabilities index, plays physics fall
The effect of dimension.
3) give up there is the original variable of the production unit determining physical model, and use the pass extracted according to physical model
Key index k, reduces to k dimension by unit operation subset Xq from q dimension；
The original variable having in the production unit determining physical model is removed in training sample, and instead from physical model
The Key Performance Indicator of middle extraction, by its with have neither part nor lot in physical modeling its under remaining performance variable reconfigure, formed new
Training sample Xnew, Xnew ∈ Rn×p, data set determines the new variables number after production unit physical model dimensionality reduction.
4) Xnew is normalized；Wherein xu_i is the meansigma methods that Xnew respectively arranges, and std_i is the mark that Xnew respectively arranges
Accurate poor, after normalization Xnorm=(Xnew_i-xu_i)/std_i, the absolute value obtaining the Xnorm after conversion is less than
3.5, it is judged that the absolute value Xnorm more than 3.5 is exceptional value.
5) the main metadata direction of factory level process is extracted, the monitoring and statistics amount of the level that sets up a factory and monitoring limit thereof.
With Xnew set up full level of factory pivot analysis or partial least square model, the wherein index for the purpose of stability
Refer to only one group operation operating mode, it is impossible to determine the quantification target of production process in real time；Wherein performance indications refer to produce
Process has the numerical indication that the online or quality of off-line, cost are relevant.
If monitoring is the index for the purpose of stability, then set up pivot procedure fault detection model, form factory
The Fault Model of level scope, calculates at pivot load, SPE, Hotelling T^2 control model under different confidence levels
Enclose.
If monitoring is performance indications, then use concretely comprising the following steps of partial least square model: assume that target capabilities becomes
Amount y and operating condition XnewRn×pRelevant, wherein front h performance variable x1, x2 ..xh are uncontrollable operating conditions, such as
Atmospheric temperature, the client the must being fulfilled for demand to product, namely these performance variables cannot be made to come from by operator
By adjustment；After wherein, (p-h) individual variable is the adjustable performance variables of operator.
First, ground floor simulated target performance variable y_act and uncontrollable operating condition xi (i=1 ... h) set up
A young waiter in a wineshop or an inn takes advantage of model；The predictive value y_cal of computation model, wherein residual error y_res=y_act-y_cal；Y_ used again by second layer model
Res and controllable operating condition xj (j=h+1 ... p) set up partial least square model, the model predication value of y_res y_rcl table
Showing, the final predictive value of final goal performance variable can be expressed: y_t cl=y_cal+y_rcl.
According to the actual controlled scope of operator and the impact on performance indications thereof, it is judged that new performance indications control zone
Between, its concrete operating procedure is: owing to the middle mass data that sample used is directly product practice storehouse directly uses,
Therefore the span of control of monitor control index uses the concept of big data, rather than it is generally to use that employing represents creation data on a small quantity
The confidence level of 95%；Also directly y_cer can be carried out statistical analysis, take its probability of best 25% interval as fault detect,
Its physical significance is: historical normal operating level is 50%, and detects the direction skew 25% that interval tropism energy index is good,
If that is, average is 50 points, then the operation operating mode less than 75 points is accordingly to be regarded as fault.
Meanwhile, the concept of production process sustained improvement, for the statistics of historical operating data, sample space equal are introduced
Value is equivalent to the performance level that the probability of 50% is issued to, and the Fault Control scope of the first step can be energy under 60%-75% probability
The performance level reached, and pass through the detection to fault and diagnosis, basic reason analysis and corrective measure, after 6-12 month,
Again extract historical data, repeat above modeling process, form new performance indications control interval.
6) collect new process data, and it is carried out pretreatment and normalization.
7) by new process data, according to there being the production unit model determining physical model, the key of this unit is calculated
Can index.
8) press the processing method of training sample, form new process data collection, wherein have the production list determining physical model
Unit is expressed by new Key Performance Indicator.
9) if using partial least square method calculation of performance indicators, then, in addition to SPE, Hotelling T^2, also calculate
Current residual error (contribution of controlled variable) and and preferable residual prediction value between whether within span of control.To supervise in real time
The extracting data surveyed goes out critical data, and critical data is applied to step 4) and 5) model that obtains or computational methods are carried out
Calculate, finally judge in the on-line monitoring process data that its calculating the most up to standard is new:
A) judgement of the pivot analysis of corresponding process exception；
B) the latent structure projection model of Key Performance Indicator.
The wherein computational methods of the latent structure projection model of Key Performance Indicator, its detailed operational approach is: according to newly
Production operation data, uncontrollable performance variable currency is input in ground floor model, is calculated current y ' _ cal；
According to current controllable operating variable, in input second layer model, being calculated current y ' _ res, current actual performance refers to
Mark y ' _ act y ' _ cal=y " _ res, it is the actual value of current controllable operating variable, by comparing y " res is controlled
Monitor control index under the influence of performance variable；The desired value of the performance of current production process is that y ' _ tcl, by comparing y ' _ act
And the difference that y ' t_tcl is between this, if difference is beyond the monitoring range of regulation, then shows performance fault occur.
C) judgement whether broken down in controlled range；
10) on the basis of fault detect, find out current failure sensitive variable and insensitive variable respectively, obtain examining of this fault
Disconnected result, instructs operator carry out safety, high-quality and produce control efficiently.
It is an advantage of the current invention that: the method for the present invention is integrated application physics and big data method, in the unit determined
In one chemical engineering unit operation (a certain specific flow process, equipment), set up reliable physical model (typically quality and energy balance mould
Type, and the clearest and the most definite chemical reaction, be not related to the chemical reaction process of complexity), and in whole production process (such as a car
Between multiple equipment) adopt the method (containing pivot analysis, offset minimum binary) making Chemical Measurement, introduce at big data processing technique
The real time data of reason several years, finds unknown fault, in the monitoring to performance fault, uses double-deck partial least square model,
It is conceived to operator's fault of performance indications in controllable operating range of variables, eliminates the shadow of uncontrollable production process variable
Ring.
Accompanying drawing explanation
The operating process sketch of Fig. 1 present invention；
Fig. 2 is the example heating furnace schematic flow sheet of the present invention；
Fig. 3 is the pivot score judgement figure of the fault detection and diagnosis of the present invention；
Fig. 4 is SPE and the T^2 result of calculation figure of the fault detection and diagnosis of the present invention；
Fig. 5 is the key variables result figure occurring stable fault point in the present invention；
Fig. 6-a is detection and the monitored results figure of the performance fault of the present invention；
Fig. 6-b be the present invention Fig. 6-a in performance position of failure point enlarged drawing；
Fig. 7-a is all controlled variables of the present invention contribution plots to performance fault；
Fig. 7-b be the present invention Fig. 7-a in the maximum variable of front 10 performance trouble points impact；
Fig. 8 and Fig. 9 is the main operating parameters table finally given in embodiments of the invention
Figure 10 is embodiments of the invention Zhong Quan factory process units schematic diagram.
Detailed description of the invention
The present invention is described in further detail with detailed description of the invention in explanation below in conjunction with the accompanying drawings.
The effectiveness of the inventive method is described in conjunction with a concrete factory level chemical process example.This process
Flow chart as shown in Fig. 1 and Figure 10, judge through technical staff, whole process is by (the desalination, just of 6 different operating units
Evaporate, atmospheric pressure kiln, atmospheric tower group, heating under reduced pressure stove, vacuum tower group) composition；Therefore the main operating parameters finally given is such as
Shown in the form of Fig. 8 and Fig. 9.
Embodiment 1: detection and the diagnosis to the stable fault of this system.
In order to test the effectiveness of new method, data set is labelled with regular data and abnormal data set is tested, its
Middle normal data comprises 61225 data points, and occurs 66 data points of stable fault (at 55370-55436
Position),
The first step utilizes device real-time data base to collect the training sample set of data composition modeling: XR61291×112.Wherein, one
Table of looking at sees attached list 1.
Second step determines the production unit of physical model for having, and according to physical model extract key performance and refer to
Mark.
As a example by feeding heating furnace in production technology, details sees Fig. 2, in heating furnace unit Q, has about 28 processes to become
Amount, XqR61291×28Some process variable is uncontrollable, and such as heat dissipation capacity is generally and the heat-insulating property of equipment itself and air
Temperature, humidity and wind speed are relevant, and some process variable is controlled, the flow of such as combustion air, material outlet temperature etc..As
These 28 process variables are applied directly in failure monitoring by fruit, not only increase amount of calculation, also bring data to system and make an uproar
Sound interference.For heating furnace, there are two key parameters can express the Key Performance Indicator of this production unit, and have clear and definite
Physical significance.
The wherein flow of fuel: F_fuel,
The efficiency of heating furnace: Ef, with the temperature of heating furnace smoke evacuation, oxygen content, containing CO content, atmospheric condition have clear and definite thing
Reason model, can calculate and obtain.
Thus 28 process variables to heating furnace unit, it now is possible to express with two new variables of F_fule, Ef.
Use same method, by during have and determine other production units of physical model, extract key physical mould
Shape parameter, has a physics dimensionality reduction with entering, and effectively remain relevant material, heat balance closes system information.It is exemplified below:
Each is had the primitive operation parameter removal training sample set of the production unit determining physical model, and generation is with above-mentioned life
Produce the Key Performance Indicator of unit, form the data set Xnew of new factory levelR61291×67。
3rd step is for data set XnewR61291×67It is normalized；Xu_i is the meansigma methods that Xnew respectively arranges, and std_i is
The standard deviation that Xnew respectively arranges, then Xnorm=(the Xnew_i-xu_i)/std_i after normalization, for statistical angle, warp
The value crossing absolutely mostly sample data after conversion should be between [-3.5 ,+3.5], and the most more convenient interpretation one of exceeds the different of scope
Constant value.
After dimensionality reduction and comprise related physical information Xnew then cross the method for polytomy variable pivot analysis carrying out routine,
After standardization, the covariance matrix of sample Xnew is
By pivot, Xnew is decomposed:
Xnew=TPT+E
T=XP
Wherein P is load matrix, is made up of into front q the characteristic vector of S, and T is score matrix.
4th step sets up data statistics monitoring model
By directly observing the two initial row of score matrix T, i.e. observational variable score in two maximum principal component vector, i.e.
Failure condition can be monitored, due to carrying point generally also between [-3.5,3.5] of pivot normal after normalization, deviate significantly from this district
Between i.e. can be considered abnormal, as it is shown on figure 3, in the drawings, when fault occurs, the second pivot score is the most substantially beyond normally controlling model
Enclose, fault detect success.
Realized process by structure monitoring and statistics amount, i.e. square prediction error (SPE) and Hotelling ' s T^2
Stable state has multivariate statistics to monitor.SPE and T^2 all has based on historical data, the upper control limit under different level of confidence.
The aspect that SPE is different from during monitoring.The degree that between SPE principal measure normal processes variable, dependency is changed.T^2 degree
The distance of amount current working distance principal component subspace initial point.
Wherein control limit
Hotelling ' s T^2 calculates as follows:
As shown in Figure 4, its lower control limit when confidence level 95% marks the monitoring figure of SPE and T^2 the most in the drawings, works as statistic
Amount is notable beyond controlling in limited time, i.e. it is believed that break down.During fault condition, the normal relation between usual variable is destroyed,
Or variable is on the whole away from nominal situation, or both occur simultaneously.If confidence level improves (more sure judgement is abnormal), then
Red line in Fig. 4, during wherein red line refers to Fig. 4, position is at the dotted line of the top, by upper shifting.
Use another advantage is that by different variable " contribution " ratios to SPE and T^2, base of SPE and T^2 statistic
May determine that on Ben when appearance is abnormal, which parameter cause.Certain point when being to occur abnormal in upper page in left figure
The variable contribution to SPE and T^2, can be seen that from left figure unusual service condition is mainly what which variable caused.Behaviour can be helped
Make personnel and screen rapidly influence factor, and make correct judgement；Concrete outcome is as shown in Figure 5.
The judgement of the process data that the 5th step is new, as obtained in real time from the real-time data base of production process, then fault inspection
Survey can be with real time implementation.
6th step calculates the physical model of production unit, extracts Key Performance Indicator, and merges other production operation variablees,
The meansigma methods using the sample obtained in the 3rd step is normalized with standard deviation.
7th step: the pivot detection model drawn by the 4th step, calculates current operation variable and at production unit physics mould
The statistical monitoring statistic of the key physical index extracted in type.
8th step calculates each monitoring and statistics value, forms the failure detection result of factory level process, it is judged that active procedure
Running status.
Embodiment 2: detection and the diagnosis to the performance fault of this system:
Next combine this detailed process the enforcement step of the present invention is set forth in:
The first step. with the first step in example 1.
Second step；With second step in example 1.
Performance is criticized mark and is carried out double-deck regression analysis by the 3rd step, obtains the monitoring model of performance fault.
Extracted physical model key index dimensionality reduction obtain containing related physical information Xnew be normalized (same to example
1) and the relation of regression analysis Xnew and factory level Key Performance Indicator, whether performance fault occurs with the process of monitoring.
As a example by production process in accompanying drawing 1, it is (light that the Key Performance Indicator of factory level can be expressed as high value added product
The yield of matter oil product such as gasoline, diesel oil) and the energy consumption of production process.Need both this to consider, simplify herein and be expressed as factory
Performance indicator, i.e. desired value Y is by difference a)-b of following two) form:
A) unit light oil yield incremental income (petrol and diesel oil averagely repaiies price-crude oil price) * (petrol and diesel oil flow/oil flow)
B) processed in units amount energy cost, i.e. (furnace fuel flow * fuel cost+steam flow * steam cost)/crude stream
Amount, owing to water, electricity consumption change less, is not counted in the desired value of energy cost
Double-deck partial least square method is used to draw the regression model of Y ~ Xnew.
First, at XnewRn×pIn, some manufacturing variables is determined by raw material and finished product market, such as different oil varieties,
Itself content containing high added value light oil is different, directly affects Y, and crude charging capacity is from the plan of higher level, and operate
Personnel cannot arbitrarily change, thus first by Y and these uncontrollable factors totally 8 carry out ground floor PLS
The Y_act calculated with creation data and go out and uncontrollable operating condition xi (i=1 ... 8) set up partial least square model；
The predictive value Y_cal of computation model.
Calculate residual error Y_res=Y_act-Y_cal again；Second layer model is again with Y_res and controllable operating condition xj (j=
The model predication value Y_rcl of h+1 ... p) set up partial least square model, Y_res represents, then final goal performance variable
Final predictive value can be expressed: Y_tcl=Y_cal+Y_rcl.
Statistically see, in the case of meeting certain recurrence accuracy, performance indications actual value and error Y_ of predictive value
Err is close to normal distribution, thus it is believed that once difference between Y_err=Y_act-Y_tcl is beyond the scope between, i.e. goes out
Existing performance fault.Control interval can be taken at the control interval under different confidence level such as general statistical Process Control, the most relatively
It is applicable for use with having on a small quantity and represents the confidence level that creation data is the 95% of generally employing.May be used without using the distribution of big data close
The concept of degree,；Namely directly Y_err is carried out statistical analysis, take its probability of best 25% interval as fault detect, its
Physical significance is: historical normal operating level is 50%, and detects the direction skew 25% that interval tropism energy index is good, also
That is, if total score 100 points, average is 50 points, then the operation operating mode less than 75 points is accordingly to be regarded as fault.
In order to test the effectiveness of the method, make the data set of use-case one, by calculated yield-energy consumption and performance index Y.Point
Not being marked in Fig. 6-a and Fig. 6-b by Y < L1 < L2 control interval, wherein in 6-b, the line of band point is current performance, and dotted line is to report to the police
Interval, solid line is the interval that must take corrective action；
Finally show that two control lines the most corresponding L1 and L2, Y that wherein scheme top likely send out beyond i.e. alarm during L1
Raw fault, if Y is beyond L2, then operator must make corresponding adjustment, so that performance recovery is normal.
On trouble point, by each controlled variable contribution degree to the predictive value Y_rcl of residual error Y_res, can directly by
Affect the influence factor of these trouble point performance indications, arrange by the order affecting size, it is simple to operator find rapidly
Cause the controllable operating variable of performance fault, after it is made adjustment, the fall that can disappear performance fault.
5th step: the judgement of new process data, as obtained in real time from the real-time data base of production process, then fault inspection
Survey can be with real time implementation.Obtain current operation operating mode, and calculate the yield-energy consumption index Y ' of equivalent.
6th step: calculate the physical model of production unit, extracts Key Performance Indicator, and merges other production operations change
Amount, uses the meansigma methods of the sample obtained in the 3rd step to be normalized with standard deviation.
7th step: the offset minimum binary bilayer regression model drawn by the 4th step, by current variable input model, counts respectively
Calculate Y_cal, Y_res, Y_tcl.
8th step calculates each monitoring and statistics value, forms the failure detection result of factory level process, it is judged that active procedure
Running status.By Y ', Y_tcl+L1, Y_tcl+L2 get ready on same figure；The Y ' of current point is as beyond Y_tcl+L1 being
The report that is out of order is the most alert, as Y ' then must take action beyond Y_tcl+L2, adjusts controlled variable, and the priority of adjustment is referred to figure
7-a and Fig. 7-b, show that from two figures the size that affects of residual error is sequentially adjusted by controlled variable, until current performance index
In controlled range.
It is only presently preferred embodiments of the present invention it should be noted that above-mentioned, is not used for limiting the protection model of the present invention
Enclosing, combination in any or the equivalents made on the basis of above-described embodiment belong to protection scope of the present invention.
Claims (6)
1. a Chemical Manufacture online fault detection and diagnosis technology based on physics-big data mixing model, it is characterised in that
The operational approach of described fault detection and diagnosis technology is as follows: after choosing object run unit, will own this unit
Historical data is scanned, and after checking parameter, sets up accident knowledge base and parameter model；In follow-up on-line checking process
In, online data is importing directly in parameter model, scanned after draw the fault data in online data, send alarm,
Fault data is contrasted with data in accident knowledge base, draws failure cause；Its detailed operational approach includes following step
Rapid:
1) utilize in factory level real time data historical data composition modeling training data sample set: X ∈ Rn×m, wherein,
N is the number of sample data set, and m is the variable number of sample data set；
2) for factory level process data collection XRn×mIn have determine physical model unit operation Q, by this production unit mistake
Cheng Bianliang collects at data subset XqRn×q, q is the variable number of the unit operation having and determining physical model；
3) extraction has the Key Performance Indicator k in the production unit determining physical model, by unit operation subset Xq from q dimension fall
Tie up for k；
4) in training sample, remove the original variable having in the production unit determining physical model, and instead from physics mould
In type extract Key Performance Indicator, by its with have neither part nor lot in physical modeling its under remaining performance variable reconfigure, formed
New training sample Xnew, Xnew ∈ Rn×p；
5) to XnewRn×pIt is normalized；Wherein xu_i is the meansigma methods that Xnew respectively arranges, and std_i is the mark that Xnew respectively arranges
Accurate poor, after normalization Xnorm=(Xnew_i-xu_i)/std_i, the absolute value obtaining the Xnorm after conversion is less than
3.5, it is judged that the absolute value Xnorm more than 3.5 is exceptional value；
6) the main metadata direction of factory level process is extracted, the monitoring and statistics amount of the level that sets up a factory and monitoring limit model thereof；
If a) monitoring is the index for the purpose of stability, then set up pivot procedure fault detection model, form factory level
The Fault Model of scope, calculates in pivot load, SPE, Hotelling T^2 span of control under different confidence levels；
If b) monitoring be performance indications, then adopt dual residual error return method, the first multiple regression first by performance indications with
Operator's uncontrollable factor impact on performance indications in production process, the second multiple regression is with the residual error of the first multiple regression for should
Variable, the controlled operating condition of operator is that independent variable carries out PLS；
C) according to the actual controlled scope of operator and the impact on performance indications thereof, it is judged that new performance indications control zone
Between, using the gap of current operation performance indications and ideal performance index as failure monitoring index, according to process improvement plan,
Go out improvable span of control in subsequent operation performance indications；
7) collect new on-line monitoring process data, use the average of variable obtained in step 5) training sample and standard deviation its
Carry out pretreatment and normalization；
8) according to there being the on-line monitoring process data that the cell processing determining physical model is new, the key performance calculating this unit refers to
Mark, forms new process data collection, wherein has the production unit determining physical model to be expressed by new Key Performance Indicator；
9) extracting data monitored in real time is gone out critical data, critical data is applied to step 5) and 6) model that obtains or
Computational methods calculate, finally judge in the on-line monitoring process data that its calculating the most up to standard is new:
A) judgement of the pivot analysis of corresponding process exception；
B) the latent structure projection model of Key Performance Indicator；
C) judgement whether broken down in controlled range；
10) calculate monitoring and statistics value, form the failure detection result of factory level process, it is judged that the running status of active procedure；
11) on the basis of fault detect, when finding out fault respectively, each performance variable contribution margin to Monitoring and Controlling amount, obtains
The diagnostic result of this fault.
The online fault detection and diagnosis of Chemical Manufacture based on physics-big data mixing model the most according to claim 1
Technology, it is characterised in that described step 2) and 3) in determine: factory level process data integrates as XRn×m, determine thing having
In the production unit of reason model, physical model is used to extract the Key Performance Indicator of this production unit, i.e. according to this production unit
Chemical composition, material, heat so that chemical phase equilibrium, chemical reaction equilibrium model, calculate this unit and refer to target capabilities
Target key contributions variable, is and carries out dimensionality reduction based on physical model, be expressed as:
XqRn×q, draw from physical model: Xq_new=Function (Xq), XqRn×k, meeting the same of key index
Time, meet the dimensionality reduction requirement that big data process simultaneously.
The online fault detection and diagnosis of Chemical Manufacture based on physics-big data mixing model the most according to claim 1
Technology, it is characterised in that in described step 6) monitoring be performance indications time, use partial least square model concrete steps
For:
Target capabilities variable y and operating condition Xnew ∈ R after physical model dimensionality reductionn×pRelevant, wherein front h performance variable is
Uncontrollable operating condition, these performance variables cannot be made and being adjusted freely by operator, and wherein rear p-h variable is behaviour
Make the adjustable performance variable of personnel；
A) ground floor simulated target performance variable y_act and uncontrollable operating condition xnewi (i=1 ... h) set up regression model；
B) the predictive value y_cal of computation model, wherein residual error y_res=y_act-y_cal；
C) second layer model y_res and controllable operating condition xnewj (j=h+1 ... p) setting up regression model, model is to y_
The predictive value of res is y_rcl；
D) the final predictive value of target capabilities variable is expressed as: y_tcl=y_cal+y_rcl；
E) model prediction deviation can be expressed as: ground floor total deviation y_ter=yact-ytcl, second layer process controllable variables
Deviation: y_cer=y_res-y_rcl.
The online fault detection and diagnosis of Chemical Manufacture based on physics-big data mixing model the most according to claim 3
Technology, it is characterised in that the detection range to performance indications fault is the pre-of the operating condition controlled by operator's reality
Surveying the statistic of deviation y_cer, concrete operating procedure is:
A) by a)-e in claim 3) step, the y_cer, y_cer that calculate training sample are generally near normal distribution.
B) y_cer can be used the standard deviation control interval of routine, such as the span of control in 95% confidence interval, it is possible to directly
Y_cer is carried out statistical analysis, takes its probability of best 25% interval as fault detect；
C) carry out sustained improvement in production process, pass through the detection to fault and diagnosis, basic reason analysis and corrective measure；?
After 6-12 month, again extract historical data, repeat above modeling process, form new performance indications control interval.
The online fault detection and diagnosis of Chemical Manufacture based on physics-big data mixing model the most according to claim 1
Technology, it is characterised in that the computational methods of the latent structure projection model of Key Performance Indicator in described step 9) b, it is detailed
Operational approach is:
A) according to new production operation data, uncontrollable performance variable currency is input in ground floor model, is calculated
Current y ' _ cal；
B) according to current controllable operating variable, in input second layer model, it is calculated current y ' _ res,
C) current actual performance index y ' _ act y ' _ cal=y " _ res, it is the reality of current controllable operating variable
Value, by comparing y " res is the monitor control index under the influence of controllable operating variable；The desired value of the performance of current production process
For, y ' _ tcl, by comparing y ' _ act and y ' t_tcl difference between this, if difference is beyond the monitoring range of regulation, then table
Bright there is performance fault.
The online fault detection and diagnosis of Chemical Manufacture based on physics-big data mixing model the most according to claim 1
Technology, it is characterised in that in step 6), the index for the purpose of stability refers to only one group operation operating mode, it is impossible to determine in real time
The quantification target of production process；Wherein performance indications refer to production process online or the quality of off-line, cost phase
The numerical indication closed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610263402.XA CN105893700B (en) | 2016-04-26 | 2016-04-26 | Based on the online fault detection and diagnosis technology of physics-big data mixed model Chemical Manufacture |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610263402.XA CN105893700B (en) | 2016-04-26 | 2016-04-26 | Based on the online fault detection and diagnosis technology of physics-big data mixed model Chemical Manufacture |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105893700A true CN105893700A (en) | 2016-08-24 |
| CN105893700B CN105893700B (en) | 2019-05-31 |
Family
ID=56704549
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610263402.XA Active CN105893700B (en) | 2016-04-26 | 2016-04-26 | Based on the online fault detection and diagnosis technology of physics-big data mixed model Chemical Manufacture |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105893700B (en) |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106682080A (en) * | 2016-11-21 | 2017-05-17 | 云南电网有限责任公司电力科学研究院 | Transformer fault analysis method and system |
| CN106872672A (en) * | 2016-12-28 | 2017-06-20 | 中南大学 | A kind of method that runnability is hydrocracked for on-line analysis |
| CN106933097A (en) * | 2017-05-15 | 2017-07-07 | 青岛科技大学 | A kind of Fault Diagnosis for Chemical Process method based on multi-level optimization PCC SDG |
| CN107728599A (en) * | 2017-09-01 | 2018-02-23 | 北京中燕信息技术有限公司 | A kind of method and apparatus for determining refinery device valve state |
| CN108664696A (en) * | 2018-04-02 | 2018-10-16 | 国家计算机网络与信息安全管理中心 | A kind of assessment method and device of handpiece Water Chilling Units operating status |
| CN108989083A (en) * | 2018-06-01 | 2018-12-11 | 安徽师范大学 | Fault detection performance optimization method under cloud environment based on mixed strategy |
| CN109213057A (en) * | 2017-06-29 | 2019-01-15 | 上海微电子装备（集团）股份有限公司 | Intelligent diagnostics device and method |
| CN109298993A (en) * | 2017-07-21 | 2019-02-01 | 深圳市中兴微电子技术有限公司 | A kind of method, apparatus and computer readable storage medium detecting failure |
| CN109327330A (en) * | 2018-09-11 | 2019-02-12 | 南京科思倍信息科技有限公司 | Chemical Manufacture Exceptional Slices management method based on data-driven |
| CN109816940A (en) * | 2019-03-21 | 2019-05-28 | 北京天诚同创电气有限公司 | The fault alarm method and device of sewage treatment plant |
| WO2019178948A1 (en) * | 2018-03-21 | 2019-09-26 | 江南大学 | Multi-layer mode monitoring method for flow industrial process |
| CN110912775A (en) * | 2019-11-26 | 2020-03-24 | 中盈优创资讯科技有限公司 | Internet of things enterprise network fault monitoring method and device |
| CN110956308A (en) * | 2019-10-29 | 2020-04-03 | 沈阳天眼智云信息科技有限公司 | Chemical equipment health state prediction method based on time sequence |
| CN112578740A (en) * | 2019-09-30 | 2021-03-30 | 冯恩波 | Fault diagnosis and processing method and system in industrial production process |
| CN112784862A (en) * | 2019-11-07 | 2021-05-11 | 中国石油化工股份有限公司 | Fault diagnosis and identification method for refining process of atmospheric and vacuum distillation unit |
| CN113239187A (en) * | 2021-04-13 | 2021-08-10 | 鹏城实验室 | Monitoring method based on multi-level industrial structure knowledge block division |
| US11120350B2 (en) | 2018-03-21 | 2021-09-14 | Jiangnan University | Multilevel pattern monitoring method for industry processes |
| CN113620137A (en) * | 2021-06-24 | 2021-11-09 | 陆新建 | Emergency manual door opening device of elevator door |
| CN114906817A (en) * | 2021-02-08 | 2022-08-16 | 中国石油化工股份有限公司 | Memory, hydrogen peroxide extraction process control method, device and equipment |
| CN116107280A (en) * | 2023-04-12 | 2023-05-12 | 南京强思数字科技有限公司 | Digital twin plant control system for multidimensional visual management |
| CN116794088A (en) * | 2023-08-03 | 2023-09-22 | 矿冶科技集团有限公司 | Online compensation method for copper flotation foam grade of X fluorescence grade analyzer |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824137A (en) * | 2014-03-20 | 2014-05-28 | 北京信息科技大学 | Multi-condition fault prediction method for complex mechanical equipment |
| CN104062968A (en) * | 2014-06-10 | 2014-09-24 | 华东理工大学 | Continuous chemical process fault detection method |
| CN105004542A (en) * | 2015-07-15 | 2015-10-28 | 浙江中烟工业有限责任公司 | Online monitoring and fault diagnosing method for mixing and flavouring process of cigarette filament production based on principal component analysis |
-
2016
- 2016-04-26 CN CN201610263402.XA patent/CN105893700B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824137A (en) * | 2014-03-20 | 2014-05-28 | 北京信息科技大学 | Multi-condition fault prediction method for complex mechanical equipment |
| CN104062968A (en) * | 2014-06-10 | 2014-09-24 | 华东理工大学 | Continuous chemical process fault detection method |
| CN105004542A (en) * | 2015-07-15 | 2015-10-28 | 浙江中烟工业有限责任公司 | Online monitoring and fault diagnosing method for mixing and flavouring process of cigarette filament production based on principal component analysis |
Non-Patent Citations (1)
| Title |
|---|
| 许洁 等: "基于改进多尺度核主元分析的化工过程故障检测与诊断方法研究", 《仪器仪表学报》 * |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106682080A (en) * | 2016-11-21 | 2017-05-17 | 云南电网有限责任公司电力科学研究院 | Transformer fault analysis method and system |
| CN106872672A (en) * | 2016-12-28 | 2017-06-20 | 中南大学 | A kind of method that runnability is hydrocracked for on-line analysis |
| CN106872672B (en) * | 2016-12-28 | 2018-04-13 | 中南大学 | A kind of method for being hydrocracked runnability for on-line analysis |
| CN106933097A (en) * | 2017-05-15 | 2017-07-07 | 青岛科技大学 | A kind of Fault Diagnosis for Chemical Process method based on multi-level optimization PCC SDG |
| CN109213057A (en) * | 2017-06-29 | 2019-01-15 | 上海微电子装备（集团）股份有限公司 | Intelligent diagnostics device and method |
| CN109213057B (en) * | 2017-06-29 | 2021-10-26 | 上海微电子装备（集团）股份有限公司 | Intelligent diagnosis device and method |
| CN109298993B (en) * | 2017-07-21 | 2022-07-29 | 深圳市中兴微电子技术有限公司 | Method and device for detecting fault and computer readable storage medium |
| CN109298993A (en) * | 2017-07-21 | 2019-02-01 | 深圳市中兴微电子技术有限公司 | A kind of method, apparatus and computer readable storage medium detecting failure |
| CN107728599A (en) * | 2017-09-01 | 2018-02-23 | 北京中燕信息技术有限公司 | A kind of method and apparatus for determining refinery device valve state |
| WO2019178948A1 (en) * | 2018-03-21 | 2019-09-26 | 江南大学 | Multi-layer mode monitoring method for flow industrial process |
| US11120350B2 (en) | 2018-03-21 | 2021-09-14 | Jiangnan University | Multilevel pattern monitoring method for industry processes |
| CN108664696A (en) * | 2018-04-02 | 2018-10-16 | 国家计算机网络与信息安全管理中心 | A kind of assessment method and device of handpiece Water Chilling Units operating status |
| CN108989083A (en) * | 2018-06-01 | 2018-12-11 | 安徽师范大学 | Fault detection performance optimization method under cloud environment based on mixed strategy |
| CN109327330B (en) * | 2018-09-11 | 2022-02-22 | 南京科思倍信息科技有限公司 | Chemical production abnormal slice management method based on data driving |
| CN109327330A (en) * | 2018-09-11 | 2019-02-12 | 南京科思倍信息科技有限公司 | Chemical Manufacture Exceptional Slices management method based on data-driven |
| CN109816940B (en) * | 2019-03-21 | 2023-05-09 | 北京天诚同创电气有限公司 | Fault alarm method and device for sewage treatment plant |
| CN109816940A (en) * | 2019-03-21 | 2019-05-28 | 北京天诚同创电气有限公司 | The fault alarm method and device of sewage treatment plant |
| CN112578740A (en) * | 2019-09-30 | 2021-03-30 | 冯恩波 | Fault diagnosis and processing method and system in industrial production process |
| CN110956308A (en) * | 2019-10-29 | 2020-04-03 | 沈阳天眼智云信息科技有限公司 | Chemical equipment health state prediction method based on time sequence |
| CN112784862A (en) * | 2019-11-07 | 2021-05-11 | 中国石油化工股份有限公司 | Fault diagnosis and identification method for refining process of atmospheric and vacuum distillation unit |
| CN110912775A (en) * | 2019-11-26 | 2020-03-24 | 中盈优创资讯科技有限公司 | Internet of things enterprise network fault monitoring method and device |
| CN114906817A (en) * | 2021-02-08 | 2022-08-16 | 中国石油化工股份有限公司 | Memory, hydrogen peroxide extraction process control method, device and equipment |
| CN114906817B (en) * | 2021-02-08 | 2023-12-05 | 中国石油化工股份有限公司 | Memory, hydrogen peroxide extraction process control method, device and equipment |
| CN113239187A (en) * | 2021-04-13 | 2021-08-10 | 鹏城实验室 | Monitoring method based on multi-level industrial structure knowledge block division |
| CN113239187B (en) * | 2021-04-13 | 2024-05-14 | 鹏城实验室 | Monitoring method based on multi-level industrial structure knowledge block division |
| CN113620137A (en) * | 2021-06-24 | 2021-11-09 | 陆新建 | Emergency manual door opening device of elevator door |
| CN116107280A (en) * | 2023-04-12 | 2023-05-12 | 南京强思数字科技有限公司 | Digital twin plant control system for multidimensional visual management |
| CN116794088A (en) * | 2023-08-03 | 2023-09-22 | 矿冶科技集团有限公司 | Online compensation method for copper flotation foam grade of X fluorescence grade analyzer |
| CN116794088B (en) * | 2023-08-03 | 2024-02-09 | 矿冶科技集团有限公司 | Online compensation method for copper flotation foam grade of X fluorescence grade analyzer |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105893700B (en) | 2019-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105893700A (en) | Chemical production on-line fault detection and diagnosis technique based on physical-large data hybrid model | |
| CN106371427B (en) | Industrial process Fault Classification based on analytic hierarchy process (AHP) and fuzzy Fusion | |
| EP2064626B1 (en) | Kernel-based method for detecting boiler tube leaks | |
| CN104699077B (en) | A kind of failure variable partition method based on nested iterations Fei Sheer discriminant analyses | |
| CN104748807B (en) | A kind of power station main steam flow on-line calculation method based on flux modification | |
| CN106327062A (en) | Method for evaluating state of power distribution network equipment | |
| CN106368813A (en) | Abnormal alarm data detection method based on multivariate time series | |
| CN106447205A (en) | Method for evaluating state of distribution automation terminal based on analytic hierarchy process | |
| CN108133316B (en) | Detection method for familial defects of power equipment of power supply plant | |
| CN112526251B (en) | Transformer substation power equipment fault diagnosis method based on data driving | |
| CN108051709A (en) | Transformer state online evaluation analysis method based on artificial intelligence technology | |
| CN105955219A (en) | Distributed dynamic process fault detection method based on mutual information | |
| CN104536439B (en) | A kind of method for diagnosing faults based on nested iterations Fei Sheer discriminant analysiss | |
| CN110009126B (en) | Online alarm analysis method based on fusion of PLS model and PCA contribution degree | |
| CN110619457A (en) | Monitoring sensor layout method and system based on safety situation assessment | |
| CN108764305A (en) | A kind of improved colony intelligence machine learning fault diagnosis system | |
| CN109144028A (en) | A kind of rectifying column efficiency deterioration detecting | |
| CN112598144A (en) | CNN-LSTM burst fault early warning method based on correlation analysis | |
| CN104615123B (en) | K-nearest neighbor based sensor fault isolation method | |
| CN106599201A (en) | Full life circle management method of gas transmission and distribution equipment | |
| CN106933097A (en) | A kind of Fault Diagnosis for Chemical Process method based on multi-level optimization PCC SDG | |
| CN204808407U (en) | Gas storage storehouse surface equipment's risk assessment system | |
| CN103530502A (en) | Method and device for judging state of negative pressure system | |
| CN108388224A (en) | A kind of Desalting and Dewatering from Crude Oil process operation method for monitoring state based near infrared spectrum | |
| CN115146703A (en) | Multi-model-based on-line monitoring method for wall temperature of furnace chamber under wide load of coal-fired unit |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: Nanjing high tech Zone, Nanjing Software Park (West) unity Road, room 100, No. 227, room 210032 Applicant after: Lu Xinjian Address before: 210036 Gulou District, Jiangsu, Nanjing Blue Sky Park 37-201 Applicant before: Lu Xinjian |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |