CN102880151B  Doublelayer data modeldriven plantlevel chemical process monitoring method  Google Patents
Doublelayer data modeldriven plantlevel chemical process monitoring method Download PDFInfo
 Publication number
 CN102880151B CN102880151B CN201210387023.3A CN201210387023A CN102880151B CN 102880151 B CN102880151 B CN 102880151B CN 201210387023 A CN201210387023 A CN 201210387023A CN 102880151 B CN102880151 B CN 102880151B
 Authority
 CN
 China
 Prior art keywords
 amp
 data
 new
 submodule
 monitoring
 Prior art date
Links
Classifications

 Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSSSECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSSREFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
 Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
 Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
 Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
 Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Abstract
Description
Technical field
The invention belongs to safety monitoring and the field of quality control of chemical engineering industry production run, particularly a kind of level of factory Monitoring of Chemical method of duallayer data modeldriven.
Background technology
Along with the productionscale expanding day of modern chemical industry process, a kind of novel level of factory process more and more causes the concern of people.Also obtain the extensive concern of academia and industry member for the modeling of level of factory process, monitor and forecast thereupon.A typical level of factory process comprises multiple operating unit, multiple flow path device usually, even includes multiple operation workshop.The largesize chemical process of complexity like this, adopts traditional monitoring method based on mechanism model will be very difficult.By contrast, owing to have accumulated a large amount of data in process, be more applicable to and complex largesized level of factory process based on the modeling of data and monitoring method.Level of factory process monitoring method conventional is at present all be based upon on the basis of datadriven model mostly, in these large class methods, usually first whole level of factory process is divided into multiple submodule according to operating unit and function, the monitoring model then set up based on data for different submodules respectively carries out piecemeal monitoring to process.But these class methods often have ignored the mutual relationship between submodule, and this relation likely has influence on the Monitoring Performance of whole level of factory process.Therefore, if on the basis of submodule data modeling, then a model can be set up to the mutual relationship between them, the capturing ability of data model to level of factory procedural information will greatly be improved.Meanwhile, based on the monitoring system of a duallayer data model, will make RealTime Monitoring to the overall situation of level of factory process and local mode of operation, the robotization being very beneficial for level of factory process is implemented simultaneously.
Summary of the invention
The object of the invention is to the deficiency for existing level of factory process monitoring method, a kind of level of factory Monitoring of Chemical method of duallayer data modeldriven is provided.
The object of the invention is to be achieved through the following technical solutions: a kind of level of factory Monitoring of Chemical method of duallayer data modeldriven, comprises the following steps:
(1) Distributed Control System (DCS) is utilized to collect the training data sample set of the data composition modeling of level of factory chemical process: X ∈ R ^{n × m}.Wherein, n is the number of sample data collection, and m is the variable number of sample data collection.
(2) by whole level of factory process data collection X ∈ R ^{n × m}be divided into different submodules wherein b=1,2 ..., B is the sequence number of submodule, and B is total submodule number.By these submodule data sets stored in database backup.
(3) for the data set that each submodule is corresponding, it is normalized, sets up the data statistics monitoring model based on pivot analysis, extract pivot information, set up monitoring and statistics amount and the monitoring limit thereof of submodule.
(4) after obtaining the pivot information of each submodule, arrangement being restarted to it, being combined into new data set for building second layer data statistic analysis model.
(5) for the new data matrix of mixing pivot information, adopt oneclass support vector machines to carry out modeling to it, set up the relational model between each submodule, complete Holistic modeling and the monitoring of level of factory process.
(6) collect new process data, and preservice and normalization are carried out to it.
(7) new process data is divided into multiple submodule, is input in submodule principal component model, calculate corresponding pivot composition.
(8) reconfigure the pivot information of different submodule, new data matrix is input in oneclass support vector machines model, calculate corresponding monitoring and statistics value, judge the running status of active procedure.
The invention has the beneficial effects as follows: the present invention is by building duallayer data Statistic analysis models to level of factory process, overcome the defect of traditional piecemeal modeling method, establish the relational model between each submodule in level of factory process, and introduce oneclass support vector machines and build a new distance statistics amount for process monitoring.Compare other current level of factory process monitoring method, the present invention not only can monitor the unit of process in each submodule, and can effectively in conjunction with the relation information between each submodule of level of factory process, the overall situation of second layer data model to whole level of factory process is utilized to monitor, substantially increase the Monitoring Performance of level of factory chemical process, be very beneficial for expansion and the enforcement of the robotization of level of factory process industrial.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of level of factory chemical process;
Fig. 2 is the monitoring result to normal data set in level of factory process in the inventive method;
Fig. 3 is the monitoring result to abnormal data set in level of factory process in the inventive method;
Fig. 4 is the monitoring result to abnormal data set in level of factory process in traditional Principal Component Analysis Model.
Embodiment
The present invention is directed to the monitoring problem of level of factory chemical process, first utilize Distributed Control System (DCS) to collect the data of this process, necessary preservice and normalization are carried out to it, then whole process data collection is divided into different submodules.For the data set that each submodule is corresponding, set up the Principal Component Analysis Model of respectively, and set up the control limit of monitoring and statistics amount.After obtaining pivot information corresponding to each submodule data set, be reassembled into new data set, then adopt oneclass support vector machines to carry out modeling to it.Time new process data is monitored, equally these data are divided into different submodules, the average of each submodule modeling data and standard deviation is utilized to be normalized it, after obtaining normal data, the Principal Component Analysis Model recycling each submodule calculates the pivot information of new data in different submodule.The all pivot information obtained in submodule are reconfigured, is input in the oneclass support vector machines model of the second layer, calculate the value of distance monitoring and statistics amount, realize monitoring the overall situation of level of factory process.
The key step of the technical solution used in the present invention is as follows respectively:
The first step utilizes Distributed Control System (DCS) to collect the twodimentional training sample set of the data composition modeling of level of factory chemical process: X ∈ R ^{n × m}.Wherein, n is the number of sample data collection, and m is the variable number of sample data collection, and R is real number.
Second step according to the operating unit type of process and relevant functional attributes, the twodimentional training sample set X ∈ R that step 1 is obtained ^{n × m}be divided into different submodules: wherein b=1,2 ..., B is the sequence number of submodule, and B is total submodule number.By these submodule data sets stored in database backup.
3rd step, for data set corresponding to each submodule, is normalized it, sets up the data statistics monitoring model based on pivot analysis, extracts pivot information, sets up monitoring and statistics amount and the monitoring limit thereof of submodule.
The data set corresponding to each submodule b=1,2 ..., B carries out preservice and normalization, and namely make the average of each process variable be zero, variance is 1, obtains new data matrix collection.The fundamental purpose of this step is the result in order to make the yardstick of process data can not have influence on monitoring.After normalization, the data of various process variable are just under identical yardstick, can not have influence on followup monitoring effect afterwards.Then, pivot analysis is carried out to this data set, former space is divided into principal component space and residual error space, choose suitable pivot number, the loading matrix P obtained _{b}with score matrix T _{b}, and obtain the residual error of modeling, that is:
On the basis of Principal Component Analysis Model, by constructing following two statistic T ^{2}the monitoring to each submodule inside is realized with SPE:
Wherein, t _{b}for submodule score matrix T _{b}in vector, corresponding to the pivot variable of a submodule data, for the estimation residual error of these data.In order to monitor the state of data, we need to set up and SPE _{b}the monitoring and statistics limit of statistic and SPE _{b, lim}, that is:
Wherein, F represents the statistical distribution of F form, x ^{2}for x ^{2}the statistical distribution of form, α is statistical confidence, g _{b}=var (SPE _{b})/[2mean (SPE _{b})], h _{b}=2 [mean (SPE _{b})] ^{2}/ var (SPE _{b}), wherein mean (SPE _{b}) and var (SPE _{b}) be respectively SPE _{b}the average of statistic and variance.Respectively to b=1,2 ..., the different submodules of B carry out modeling, complete the structure of ground floor data model.
4th step obtains the pivot information of each submodule b=1,2 ..., after B, arrangement being restarted to it, being combined into new data set for building second layer data statistic analysis model, that is:
T _{com}=[T _{1}T _{2}…T _{b}…T _{B}]。
5th step, for the new data matrix of mixing pivot information, adopts oneclass support vector machines to carry out modeling to it, sets up the relational model between each submodule, complete Holistic modeling and the monitoring of level of factory process.
For two dimension mixing main metadata matrix T _{com}, set up oneclass support vector machines Data Analysis Model.First nonlinear function is utilized to project in highdimensional feature space by process data, namely oneclass support vector machines is by solving optimal problem Modling model below:
Wherein, R and a is radius and the centre of sphere of suprasphere in highdimensional feature space respectively, Φ (t _{i}) be nonlinear projection function, C is oneclass support vector machines parameter, and by this parameter, oneclass support vector machines can average out between the volume of suprasphere and the mistake point rate of sample, and ξ is the slack variable of each sample.In the solution procedure of reality, following dual propositions is usually adopted to construct oneclass support vector machines, that is:
Wherein, K (t _{i}, t _{j})=< Φ (t _{i}), Φ (t _{j}) > is kernel function, be usually chosen for the form of gaussian kernel function, α is the coefficient that each sample is corresponding, i.e. Lagrange multiplier.The modeling result of oneclass support vector machines is: the α value that most of sample is corresponding is zero, and only have the α value that the crucial sample of fraction is corresponding nonvanishing, these samples are called as support vector.
In highdimensional feature space, ask for suprasphere the centre of sphere and radius as follows:
Wherein, t _{0}for one of them support vector in single class vector machine model.
6th step collects new process data, and carries out preservice and normalization to it.
New process data is divided into multiple submodule by the 7th step, is input in submodule principal component model, calculates corresponding pivot composition.
First, by new process data x ^{new}be divided into multiple submodule, namely
Different submodule data vectors is input in corresponding submodule principal component model, calculates corresponding pivot composition as follows
8th step reconfigures the pivot information of different submodule, is input in oneclass support vector machines model by new data matrix, calculates corresponding monitoring and statistics value, judges the running status of active procedure.
By as follows for pivot information combination corresponding for each submodule data vector:
On this basis, the new pivot blended data vector will obtained be input in oneclass support vector machines model, in higher dimensional space, calculate the distance of this data vector to the suprasphere centre of sphere, that is:
Work as D _{new}≤ D _{lim}during=R, illustrate that the running status of level of factory process is good, current do not have fault to occur, on the contrary, if D _{new}>D _{lim}during=R, illustrate that this monitoring and statistics amount has surmounted its Statistical control limit, level of factory process likely there occurs certain fault, and needing provides warning message to process.
The validity of the inventive method is described below in conjunction with a concrete level of factory chemical process example.As shown in Figure 1, whole process is made up of 5 different operating units the process flow diagram of this process.Choose the monitoring of 33 primary process variable for this level of factory process, as shown in table 1, according to the corresponding informance of process operation Elementary Function and variable, monitored parameters is divided into 3 different submodules, this is because wherein there is the variable of two operating units very few, and and adjacent operating unit coupling is comparatively strong, therefore they is put under in other operating unit.Collection 960 data points are used for modeling altogether during the course, next set forth in detail implementation step of the present invention in conjunction with this detailed process:
1. collection process data, data prediction, level of factory process piecemeal, submodule Principal Component Analysis Model builds (ground floor data model structure)
Preservice is carried out to 960 that collect effective normal data samples, removes average and variance, and be divided into three different submodules, respectively corresponding 16,10 and 7 process variable.Like this, the submodule data set obtained is respectively X _{1}∈ R ^{960 × 16}, X _{2}∈ R ^{960 × 10}and X _{3}∈ R ^{960 × 7}.Then, respectively for each submodule data set, set up its Principal Component Statistics analytical model, utilize cumulative proportion in ANOVA principle, namely the cumulative data of pivot explains that degree is more than 85%, the pivot number chosen is respectively 11,5 and 3, and cumulative proportion in ANOVA is respectively 88.72%, 91.64% and 90.54%.In addition, the T that obtains of each Principal Component Analysis Model ^{2}with SPE add up quantitative statistics limit be respectively 25.1908 and 8.3297(first submodule), 15.2452 and 4.9734(second submodule), 11.4300 and 5.6033(the 3rd submodule).
2. based on the level of factory process global modeling (second layer data model structure) of oneclass support vector machines
The pivot information obtained by each submodule restarts combination, obtains following new main metadata matrix, utilizes oneclass support vector machines to carry out modeling to it, determine the position of the centre of sphere of suprasphere and the size of radius in higher dimensional space.When Selecting All Parameters, make a wrong point rate control about 1%, the monitoring and statistics amount of such gained just represents the confidence limit of 99%.In highdimensional feature space, the radius size of the suprasphere finally obtained is 0.7059.
3. obtain current Monitoring Data information, and preservice and normalization are carried out to it
In order to test the validity of new method, respectively normal data set and abnormal data set are tested, wherein, normal data set comprises 500 data points, abnormal data set comprises 960 data points, and (wherein, 160 is normal data, 800 is abnormal data afterwards), be first utilize the average in model bank and variance to be normalized it, make this test data and modeling data have identical yardstick.
4. the online monitoring of level of factory chemical process
First each test data normal data concentrated is divided into three different submodule data vectors, utilizes the Principal Component Analysis Model that each submodule is corresponding to carry out pivot information extraction to it.After obtaining the pivot information of each submodule, combination is restarted to it, be input in oneclass support vector machines model, in highdimensional feature space, calculate the distance of itself and the suprasphere centre of sphere, i.e. the value of distance statistics amount, and compared with radius, obtain RealTime Monitoring result, as shown in Figure 2.As can be seen from this figure, duallayer data model is to the monitoring result of normal data or satisfied, and the data volume of wrong report, substantially about 1%, meets the hypothesis of modeling.Next monitor abnormal data set, as shown in Figure 3, as can be seen from Figure, after 160 sampled points, duallayer data model has successfully monitored the exception of level of factory process to result, and continues to provide warning message afterwards.By contrast, if we adopt traditional Principal Component Analysis Model to monitor, Monitoring Performance will reduce greatly, as shown in Figure 4.This also illustrates us, to introduce oneclass support vector machines when second layer data modeling be suitable, is also effective.
Table 1: level of factory process monitoring variable list
Abovedescribed embodiment is used for explaining and the present invention is described, instead of limits the invention, and in the protection domain of spirit of the present invention and claim, any amendment make the present invention and change, all fall into protection scope of the present invention.
Claims (6)
Priority Applications (1)
Application Number  Priority Date  Filing Date  Title 

CN201210387023.3A CN102880151B (en)  20121011  20121011  Doublelayer data modeldriven plantlevel chemical process monitoring method 
Applications Claiming Priority (1)
Application Number  Priority Date  Filing Date  Title 

CN201210387023.3A CN102880151B (en)  20121011  20121011  Doublelayer data modeldriven plantlevel chemical process monitoring method 
Publications (2)
Publication Number  Publication Date 

CN102880151A CN102880151A (en)  20130116 
CN102880151B true CN102880151B (en)  20150211 
Family
ID=47481522
Family Applications (1)
Application Number  Title  Priority Date  Filing Date 

CN201210387023.3A CN102880151B (en)  20121011  20121011  Doublelayer data modeldriven plantlevel chemical process monitoring method 
Country Status (1)
Country  Link 

CN (1)  CN102880151B (en) 
Families Citing this family (1)
Publication number  Priority date  Publication date  Assignee  Title 

CN103376795A (en) *  20130715  20131030  浙江大学  Semiconductor process monitoring method based on integrated leaning modeling technology 
Citations (3)
Publication number  Priority date  Publication date  Assignee  Title 

CN101458522A (en) *  20090108  20090617  浙江大学  Multibehavior process monitoring method based on pivot analysis and vectorial data description support 
CN101907088A (en) *  20100527  20101208  中国人民解放军国防科学技术大学  Fault diagnosis method based on oneclass support vector machines 
CN102566554A (en) *  20120214  20120711  浙江大学  Semiconductor process monitoring method on basis of oneclass support vector machine 
Family Cites Families (2)
Publication number  Priority date  Publication date  Assignee  Title 

WO2007119111A2 (en) *  20051110  20071025  Aurelium Biopharma Inc.  Tissue diagnostics for ovarian cancer 
US8457414B2 (en) *  20090803  20130604  National Instruments Corporation  Detection of textural defects using a one class support vector machine 

2012
 20121011 CN CN201210387023.3A patent/CN102880151B/en not_active IP Right Cessation
Patent Citations (3)
Publication number  Priority date  Publication date  Assignee  Title 

CN101458522A (en) *  20090108  20090617  浙江大学  Multibehavior process monitoring method based on pivot analysis and vectorial data description support 
CN101907088A (en) *  20100527  20101208  中国人民解放军国防科学技术大学  Fault diagnosis method based on oneclass support vector machines 
CN102566554A (en) *  20120214  20120711  浙江大学  Semiconductor process monitoring method on basis of oneclass support vector machine 
NonPatent Citations (3)
Title 

基于支持向量机的单类分类方法综述;吴定海等;《计算机工程》;20110330;第37卷(第5期);第187189页 * 
基于迭代多模型ICASVDD的间歇过程故障在线监测;王培良等;《仪器仪表学报》;20090731;第30卷(第7期);第13471352页 * 
超球体多类支持向量机理论;徐图,何大可;《控制理论与应用》;20091130;第26卷(第11期);第12931297页 * 
Also Published As
Publication number  Publication date 

CN102880151A (en)  20130116 
Similar Documents
Publication  Publication Date  Title 

Mkrtchyan et al.  Bayesian belief networks for human reliability analysis: A review of applications and gaps  
CN103576641B (en)  For monitoring the system and method for the assets in operating process unit  
CN102208028B (en)  Fault predicting and diagnosing method suitable for dynamic complex system  
CN103001328B (en)  Fault diagnosis and assessment method of intelligent substation  
Alaeddini et al.  Using Bayesian networks for root cause analysis in statistical process control  
Peitsman et al.  Application of blackbox models to HVAC systems for fault detection  
Kano et al.  Monitoring independent components for fault detection  
CN102735999B (en)  Gas insulated substation (GIS) partial discharge online monitoring system and fault mode identifying method thereof  
CN101978389B (en)  Vao productivity suite  
CN104063586B (en)  Bayesian network failure prediction method based on polymorphic fault tree  
Seem  Using intelligent data analysis to detect abnormal energy consumption in buildings  
CN104573740A (en)  SVM classification modelbased equipment fault diagnosing method  
US20040102924A1 (en)  Decision support for operations and maintenance (DSOM) system  
CN102520697B (en)  Onsite information preprocessing method of remote cooperative diagnosis  
JP6141235B2 (en)  How to detect anomalies in time series data  
CN101945009B (en)  Positioning method and device of power communication network fault based on case and pattern matching  
Perelman et al.  Sensor placement for fault location identification in water networks: A minimum test cover approach  
Ge et al.  Plantwide industrial process monitoring: A distributed modeling framework  
CN103870659B (en)  A kind of fault of numerical control machine tool analyzes method  
Akhavian et al.  An integrated data collection and analysis framework for remote monitoring and planning of construction operations  
US9602345B2 (en)  Performance optimizations for wireless access points  
CN102289590B (en)  Method for estimating operating state of SF6 highvoltage circuit breaker and intelligent system  
CN106443259A (en)  Transformer fault diagnosis new method based on Euclidean clustering and SPOSVM  
CN103198175B (en)  Based on the Diagnosis Method of Transformer Faults of fuzzy clustering  
CN103245861A (en)  Transformer fault diagnosis method based on Bayesian network 
Legal Events
Date  Code  Title  Description 

C06  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  
CF01  Termination of patent right due to nonpayment of annual fee  
CF01  Termination of patent right due to nonpayment of annual fee 
Granted publication date: 20150211 Termination date: 20181011 