CN108804784A - A kind of instant learning soft-measuring modeling method based on Bayes's gauss hybrid models - Google Patents
A kind of instant learning soft-measuring modeling method based on Bayes's gauss hybrid models Download PDFInfo
- Publication number
- CN108804784A CN108804784A CN201810516991.7A CN201810516991A CN108804784A CN 108804784 A CN108804784 A CN 108804784A CN 201810516991 A CN201810516991 A CN 201810516991A CN 108804784 A CN108804784 A CN 108804784A
- Authority
- CN
- China
- Prior art keywords
- parameter
- data
- indicate
- gauss
- gauss component
- 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.)
- Pending
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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Probability & Statistics with Applications (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of instant learning soft-measuring modeling methods based on Bayes's gauss hybrid models, belong to complex industrial process modeling and hard measurement field.The present invention is used to have non-linear, non-Gaussian system time-varying industrial process, pass through a kind of strategy of online real-time update part, optimal gauss component number is determined using bayesian information criterion, when new test data arrives, it calculates it and is under the jurisdiction of the posterior probability of each gauss component, and its mahalanobis distance between training data is found out, it regard the two fusion as index of similarity；Finally, the maximum one group of data of similarity are chosen from original training sample to establish current GPR models, and carry out model output prediction, have been reached raising product quality, have been reduced the effect of production cost.
Description
Technical field
The present invention relates to a kind of instant learning soft-measuring modeling methods based on Bayes's gauss hybrid models, belong to complicated
Industrial process modeling and hard measurement field.
Background technology
For some, there are the industrial process of non-linear, time-varying and non-Gaussian system, and to the requirement of product quality in the process
It is continuously improved, needs directly to determine some the process variable of product quality carries out stringent monitoring and control.But due to certain
A little measuring instruments are expensive or the restriction of technical conditions so that these variables are not used in line apparatus measures and obtain.
In order to solve these problems, it can be estimated and predicted by the method for establishing soft-sensing model, it is common soft
Measurement method has Partial Least Squares (partial least squares, PLS), artificial neural network (artificial
Neural networks, ANN), support vector machines (support vector machine, SVM) etc..PLS can be very good to locate
The linear problem of reason process, however actual industrial process usually present it is non-linear, therefore linear method no longer be applicable in.It is non-linear to build
Mould method such as ANN, SVM etc., although can preferably processing procedure it is non-linear, it is more there are Optimal Parameters the problems such as.
In recent years, Gaussian process returns (Gaussian process regression, GPR) and receives more and more
Concern, as a kind of nonparametric density estimation, it can not only obtain the predicted value of model, moreover it is possible to obtain predicted value to model
Trust value.Compared with the methods of ANN, SVM, GPR needs the parameter optimized less, has in solving small sample, nonlinear problem
There is unique advantage.Therefore selection GPR modelings.
After offline established model puts into operation, some due to production environment change and are wanted to product quality
The reasons such as the continuous change asked, established model is it is possible that the case where being no longer desirable for current working before, prediction
As a result it cannot meet required precision.For this problem, common solution have based on sliding window (Moving window,
MW) and the method based on instant learning (Just-in-time learning, JITL), but the length of window of MW is difficult to determine
And it is not suitable for the mutation of process, JITL methods generate the principle of similar output according to similar input, and selection and test sample are most
Similar one group of training sample is predicted to establish partial model, can preferably solve process mutation problems.
However, for the time-varying industrial process of some presentation non-Gaussian systems, traditional JITL methods are to be based on Euclidean distance
Or the similarity criteria that is combined with angle selects set of metadata of similar data, is unable to fully the non-Gaussian system in view of process data.
Invention content
In order to solve the problems, such as that presently, there are the present invention proposes a kind of instant learning based on Bayes's gauss hybrid models
Soft-measuring modeling method, it has considered not only the time variation of process, and establishes part GPR models in selection set of metadata of similar data
When, the non-Gaussian feature of data is fully taken into account, more rational selection set of metadata of similar data.The method includes：
Step 1：Collect input, output data obtains historical training dataset；
Step 2：X is known training sample, and optimal gauss component number K is determined using bayesian information criterion BIC,
The description of BIC such as formula (1)：
BIC=-2logp (X | Θ)+dlog (N) (1)
Logp in formula (1) (X | Θ) indicates that the log-likelihood function of training sample, d indicate possessed by K gauss component
The number of free parameter, N indicate the number of training sample
Step 3：According to after optimal gauss component number K and the initial parameter of given gauss hybrid models GMM, formula is utilized
(5), (6), (7) continuous iteration obtain the parameter of final GMM until the difference of front and back two subparameter is less than the threshold value that sets
Θ, GMM's is described in detail as follows：
Include the data set X { x of N number of training samplei∈Rm, i=1,2 ... N }, m indicates the dimension of input data, the data
The probability density function of collection is expressed as：
Wherein, Θ=[α1,μ1,Σ1；α2,μ2,Σ2；……；αk,μk,Σk] be GMM parameter, K be gauss component
Number, θkFor the parameter of k-th of gauss component, θk=(μk,Σk), μkAnd ΣkThe mean value of respectively k-th gauss component and association side
Poor matrix, αkFor the ratio shared by k-th of gauss component, and0<αk<1, wherein the probability of k-th of gauss component is close
Spending function is：
The unknown parameter in GMM methods is solved by expectation-maximization algorithm, specific solution procedure be divided into E step and
M is walked, and is described as follows：
E is walked：According to current the l times newer parameterWithI-th of training sample is calculated by Bayesian formula to belong to
The probability of k-th of gauss componentWherein CkIndicate k-th of gauss component
M is walked：Update algorithm parameter
Step 4：When coming a new input data xq, concentrated and selected from historical data using instant learning JITL algorithms
Most like one group of data establish the Gaussian process of part and return GPR models therewith, JITL algorithms and GPR modeling methods it is detailed
Description is as follows respectively：
JITL algorithms：JITL methods are that the thought of similar output is generated according to similar input, from training sample selection with
One group of most like training sample of the test sample that currently arrives models, and the core of JITL is the selection of similarity criteria, base
In the similarity criteria of Euclidean distance and angle be a kind of common method, i.e.,：
Wherein, distance d indicates that 2 norms between the test sample currently to arrive and training sample, θ indicate the two samples
Between angle, γ be a coefficient, value is between 0 to 1；
However, for some non-gaussian industrial process, GMM can preferably be described the non-Gaussian system of process, phase
Than in traditional similarity criteria, the similarity criteria based on Bayes's gauss hybrid models BGMM can preferably select similar
Sample establishes GPR models, by the parameter Θ of the optimal gauss component number K and each ingredient that step 2 and 3 respectively obtain,
Corresponding similarity criteria can be expressed as：
Wherein xqIndicate the sample newly to arrive, xiIndicate i-th of training sample, p (Ck|xq) indicate the sample x newly to arriveqBelong to
In the posterior probability of k-th of gauss component,For the mahalanobis distance between two samples, for currently arriving
The x comeq, using above-mentioned similarity criteria, selection and xqOne group of most like data establish current GPR models
GPR modeling methods：Known training sample set X { xi∈Rm, i=1,2 ... N } and Y { yi∈ R, i=1,2 ... N } respectively
M dimension input datas and 1 dimension output data are represented, the relationship between outputting and inputting can be expressed as：
yi=f (xi)+ε (10)
Wherein f indicates that a kind of unknown functional form, ε indicate that mean value is 0, and variance isWhite noise
For new test sample xq, then its output predicted value yqAlso meet Gaussian Profile, mean value and variance distinguish table
It is shown as：
yq(xq)=cT(xq)C-1Y (11)
Wherein, c (xq)=[c (xq,x1),…,c(xq,xN)]TIt is the covariance for testing input data and training input data
Matrix,For the covariance matrix between training input data, c (xq,xq) indicate test input data with itself
Covariance value；
The radial base covariance function of GPR selections, function are described as follows：
Wherein, v indicates the overall measurement of priori, ωtIt indicates per the corresponding weight of dimension data, δijFor
Kronecher operators indicate the relative importance of each auxiliary variable；
Parameter in formula (13) is obtained using Maximum-likelihood estimationIts log-likelihood function is：
Parameter θ examination is gathered out using training set and verification collection, the parameter then optimized with conjugate gradient method, parameter
After determination, for new test data, soft-sensing model output can be obtained by formula (11)；
Step 5：The sample point x that will newly arriveqIt brings the established part GPR models of step 4 into, obtains final estimated value
yq。
Optionally, the method be applied in complex industrial process to can not variable measured directly prediction technique.
Optionally, the complex industrial process includes chemical industry, metallurgy and fermentation process.
Optionally, the method is applied to the prediction technique for butane concentration during debutanizing tower.
Present invention has the advantages that：
By a kind of strategy of online real-time update part, optimal gauss component is determined using bayesian information criterion
Number calculates it and is under the jurisdiction of the posterior probability of each gauss component, and find out itself and training data when new test data arrives
Between mahalanobis distance, by the two fusion be used as index of similarity；Finally, it is maximum that similarity is chosen from original training sample
One group of data establish current GPR models, and carry out model output prediction, reached raising product quality, reduced production
The effect of cost.
Description of the drawings
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those of ordinary skill in the art, without creative efforts, other are can also be obtained according to these attached drawings
Attached drawing.
Fig. 1 is the JITL soft sensor modeling flow charts based on BGMM；
Fig. 2 is the BIC values corresponding to different gauss component numbers；
Fig. 3 is the RMSE for the different proportion modeling for choosing training sample.
Specific implementation mode
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached drawing to embodiment party of the present invention
Formula is described in further detail.
Embodiment：
The present embodiment provides a kind of instant learning soft-measuring modeling method in Bayes's gauss hybrid models, the present embodiment
By common chemical process --- for debutanizing tower process.Experimental data comes from debutanizing tower process, to butane concentration into
Row prediction, referring to Fig. 1, the method includes：
Step 1：Collect input, output data obtains historical training dataset.
Step 2：Known training sample X determines optimal gauss component number K using BIC.The description of BIC such as formula
(1)：
BIC=-2logp (X | Θ)+dlog (N) (1)
Logp in formula (1) (X | Θ) indicates that the log-likelihood function of training sample, d indicate possessed by K gauss component
The number of free parameter, N indicate the number of training sample
Step 3：After obtaining optimal gauss component number K, gauss hybrid models (Gaussian mixture are given
Model, GMM) initial parameter, GMM algorithms are described in detail as follows：
The known data set X { x comprising N number of training samplei∈Rm, i=1,2 ... N }, m indicates the dimension of input data, it
Probability density function can be expressed as：
Wherein Θ=[α1,μ1,Σ1；α2,μ2,Σ2；……；αk,μk,Σk] be GMM parameter, K be gauss component
Number, θkFor the parameter of k-th of gauss component, θk=(μk,Σk), μkAnd ΣkThe mean value of respectively k-th gauss component and association side
Poor matrix, αkFor the ratio shared by k-th of gauss component, and0<αk<1, wherein the probability of k-th of gauss component is close
Spending function is：
The unknown parameter in GMM methods is solved by expectation-maximization algorithm.Specific solution procedure be divided into E steps and
M is walked, and is described as follows：
E is walked：With current the l times newer parameterWithI-th of training sample, which is calculated, by Bayesian formula belongs to the
The probability of k gauss component, as shown in formula (4)：Wherein CkIndicate k-th of gauss component.
M is walked：Algorithm parameter is updated using such as formula (5), (6), (7).
After obtaining the initial parameter of GMM, using formula (5), (6), (7) continuous iteration, until the difference of front and back two subparameter
Less than the threshold value set, the parameter Θ of final GMM is obtained.
Step 4：When coming a new input data xq, using instant learning (Just-in-time learning,
JITL) algorithm concentrates the Gaussian process for selecting one group of most like therewith data to establish part to return from historical data
(Gaussian process regression, GPR) model.The detailed description of JITL algorithms and GPR modeling methods is respectively such as
Under：
JITL algorithms：JITL methods are that the thought of similar output is generated according to similar input, from training sample selection with
One group of most like training sample of the test sample that currently arrives models；The core of JITL is the selection of similarity criteria, base
In the similarity criteria of Euclidean distance and angle be a kind of common method, i.e.,：
Wherein, distance d indicates that 2 norms between the test sample currently to arrive and training sample, θ indicate the two samples
Between angle, γ be a coefficient, value is between 0 to 1；
However, for some non-gaussian industrial process, GMM can preferably be described the non-Gaussian system of process, phase
Than in traditional similarity criteria, the similarity criteria based on Bayes's gauss hybrid models BGMM can preferably select similar
Sample establishes GPR models, by the parameter Θ of the optimal gauss component number K and each ingredient that step 2 and 3 respectively obtain,
Corresponding similarity criteria can be expressed as：
Wherein xqIndicate the sample newly to arrive, xiIndicate i-th of training sample, p (Ck|xq) indicate the sample x newly to arriveqBelong to
In the posterior probability of k-th of gauss component,For the mahalanobis distance between two samples, for currently arriving
The x comeq, using above-mentioned similarity criteria, selection and xqOne group of most like data establish current GPR models.
GPR modeling methods：Known training sample set X { xi∈Rm, i=1,2 ... N } and Y { yi∈ R, i=1,2 ... N } respectively
Represent m dimension input datas and 1 dimension output data.Relationship between outputting and inputting can be indicated such as formula (10)：
yi=f (xi)+ε (10)
Wherein f indicates that a kind of unknown functional form, ε indicate that mean value is 0, and variance isWhite noise.
For new test sample xq, then its output predicted value yqAlso meet Gaussian Profile, mean value and variance can divide
It is not expressed as formula (11), (12)：
yq(xq)=cT(xq)C-1Y (11)
Wherein c (xq)=[c (xq,x1),…,c(xq,xN)]TIt is the covariance for testing input data and training input data
Matrix,For the covariance matrix between training input data, c (xq,xq) indicate test input data with itself
Covariance value.
GPR can select different covariance functions, select radial base covariance function, function description such as formula herein
(12)：Wherein v indicates the overall measurement of priori, ωtIt indicates per the corresponding weight of dimension data, δijIt is calculated for Kronecher
Son indicates the relative importance of each auxiliary variable.
Generally the parameter in formula (13) is obtained with Maximum-likelihood estimationIts log-likelihood function
For：
First set parameter θ to a rational initial value, the parameter then optimized with conjugate gradient method generally uses
Parameter θ examination is gathered out by training set and verification collection, makes it in a zone of reasonableness；After parameter determines, for new test number
According to, can be obtained by formula (11) soft-sensing model output.
Step 5：The sample point x that will newly arriveqIt brings the established part GPR models of step 4 into, obtains final estimated value
yq。
Fig. 3 is to choose different ratio data to establish root-mean-square error corresponding to partial model, and using being based on Europe
The JITL methods of formula distance and angle are compared with institute's extracting method of the present invention.As seen from the figure, based on Bayes's gauss hybrid models
Instant learning soft-measuring modeling method can preferably estimate butane concentration.
The present invention determines optimal Gauss using bayesian information criterion by a kind of strategy of online real-time update part
Ingredient number calculates it and is under the jurisdiction of the posterior probability of each gauss component, and find out itself and instruction when new test data arrives
Practice the mahalanobis distance between data, regard the two fusion as index of similarity；Finally, it is chosen from original training sample similar
Maximum one group of data are spent to establish current GPR models, and carry out model output prediction, have been reached raising product quality, have been dropped
The effect of low production cost.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all the present invention spirit and
Within principle, any modification, equivalent replacement, improvement and so on should all be included in the protection scope of the present invention.
Claims (4)
1. a kind of instant learning soft-measuring modeling method based on Bayes's gauss hybrid models, which is characterized in that the method
Including：
Step 1：Collect input, output data obtains historical training dataset；
Step 2：X is known training sample, and optimal gauss component number K is determined using bayesian information criterion BIC, BIC's
Description such as formula (1)：
BIC=-2logp (X | Θ)+dlog (N) (1)
Logp in formula (1) (X | Θ) indicates that the log-likelihood function of training sample, d indicate freedom possessed by K gauss component
The number of parameter, N indicate the number of training sample；
Step 3：According to after optimal gauss component number K and the initial parameter of given gauss hybrid models GMM, using formula (5),
(6), (7) continuous iteration obtains the parameter Θ of final GMM until the difference of front and back two subparameter is less than the threshold value that sets,
GMM's is described in detail as follows：
Include the data set X { x of N number of training samplei∈Rm, i=1,2 ... N }, m indicates the dimension of input data, the data set
Probability density function is expressed as：
Wherein, Θ=[α1,μ1,Σ1；α2,μ2,Σ2；……；αk,μk,Σk] be GMM parameter, K is the number of gauss component, θk
For the parameter of k-th of gauss component, θk=(μk,Σk), μkAnd ΣkThe mean value and covariance square of respectively k-th gauss component
Battle array, αkFor the ratio shared by k-th of gauss component, and0<αk<1, wherein the probability density letter of k-th of gauss component
Number is：
The unknown parameter in GMM methods is solved by expectation-maximization algorithm, specific solution procedure is divided into E steps and M steps,
It is described as follows：
E is walked：According to current the l times newer parameterWithI-th of training sample is calculated by Bayesian formula to belong to k-th
The probability of gauss componentWherein CkIndicate k-th of gauss component
M is walked：Update algorithm parameter
Step 4：When coming a new input data xq, selection is concentrated therewith from historical data using instant learning JITL algorithms
The Gaussian process that one group of most like data establish part returns GPR models, the detailed description of JITL algorithms and GPR modeling methods
It is as follows respectively：
JITL algorithms：JITL methods are that the thought of similar output is generated according to similar input, are selected from training sample and current
One group of most like training sample of the test sample of arrival models, and the core of JITL is the selection of similarity criteria, be based on Europe
The similarity criteria of formula distance and angle is a kind of common method, i.e.,：
Wherein, distance d indicates that 2 norms between the test sample currently to arrive and training sample, θ indicate between the two samples
Angle, γ be a coefficient, value is between 0 to 1；
However, for some non-gaussian industrial process, GMM can preferably be described the non-Gaussian system of process, compared to
Traditional similarity criteria, the similarity criteria based on Bayes's gauss hybrid models BGMM can preferably select similar sample
GPR models are established, it is corresponding by the parameter Θ of the optimal gauss component number K and each ingredient that step 2 and 3 respectively obtain
Similarity criteria can be expressed as：
Wherein xqIndicate the sample newly to arrive, xiIndicate i-th of training sample, p (Ck|xq) indicate the sample x newly to arriveqBelong to
The posterior probability of k gauss component,For the mahalanobis distance between two samples, for what is currently arrived
xq, using above-mentioned similarity criteria, selection and xqOne group of most like data establish current GPR models
GPR modeling methods：Known training sample set X { xi∈Rm, i=1,2 ... N } and Y { yi∈ R, i=1,2 ... N } respectively represent m
Input data and 1 dimension output data are tieed up, the relationship between outputting and inputting can be expressed as：
yi=f (xi)+ε (10)
Wherein f indicates that a kind of unknown functional form, ε indicate that mean value is 0, and variance isWhite noise
For new test sample xq, then its output predicted value yqAlso meet Gaussian Profile, mean value and variance indicate respectively
For：
yq(xq)=cT(xq)C-1Y (11)
Wherein, c (xq)=[c (xq,x1),…,c(xq,xN)]TIt is the covariance square for testing input data and training input data
Battle array,For the covariance matrix between training input data, c (xq,xq) indicate to test the association of input data and itself
Variance yields；
The radial base covariance function of GPR selections, function are described as follows：
Wherein, v indicates the overall measurement of priori, ωtIt indicates per the corresponding weight of dimension data, δijIt is calculated for Kronecher
Son indicates the relative importance of each auxiliary variable；
Parameter in formula (13) is obtained using Maximum-likelihood estimationIts log-likelihood function is：
Parameter θ examination is gathered out using training set and verification collection, the parameter then optimized with conjugate gradient method, parameter determines
Afterwards, for new test data, soft-sensing model output can be obtained by formula (11)；
Step 5：The sample point x that will newly arriveqIt brings the established part GPR models of step 4 into, obtains final estimated value yq。
2. according to the method described in claim 1, it is characterized in that, the method be applied in complex industrial process to can not
The prediction technique of variable measured directly.
3. according to the method described in claim 2, it is characterized in that, the complex industrial process includes chemical industry, metallurgy and fermentation
Process.
4. according to the method described in claim 3, it is characterized in that, the method be applied to during debutanizing tower for fourth
The prediction technique of alkane concentration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810516991.7A CN108804784A (en) | 2018-05-25 | 2018-05-25 | A kind of instant learning soft-measuring modeling method based on Bayes's gauss hybrid models |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810516991.7A CN108804784A (en) | 2018-05-25 | 2018-05-25 | A kind of instant learning soft-measuring modeling method based on Bayes's gauss hybrid models |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108804784A true CN108804784A (en) | 2018-11-13 |
Family
ID=64089100
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810516991.7A Pending CN108804784A (en) | 2018-05-25 | 2018-05-25 | A kind of instant learning soft-measuring modeling method based on Bayes's gauss hybrid models |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108804784A (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110046378A (en) * | 2019-02-28 | 2019-07-23 | 昆明理工大学 | A kind of integrated Gaussian process recurrence soft-measuring modeling method of the selective layering based on Evolutionary multiobjective optimization |
| CN110046377A (en) * | 2019-02-28 | 2019-07-23 | 昆明理工大学 | A kind of selective ensemble instant learning soft-measuring modeling method based on isomery similarity |
| CN110083065A (en) * | 2019-05-21 | 2019-08-02 | 浙江大学 | A kind of adaptive soft-sensor method having supervision factorial analysis based on streaming variation Bayes |
| CN110197286A (en) * | 2019-05-10 | 2019-09-03 | 武汉理工大学 | A kind of Active Learning classification method based on mixed Gauss model and sparse Bayesian |
| CN110309491A (en) * | 2019-06-26 | 2019-10-08 | 大连海事大学 | A kind of transient phase division methods and system based on local Gaussian mixture model |
| CN110442942A (en) * | 2019-07-26 | 2019-11-12 | 北京科技大学 | A kind of multiechelon system analysis method for reliability based on Bayes's mixing |
| CN110516282A (en) * | 2019-07-03 | 2019-11-29 | 杭州电子科技大学 | A kind of indium phosphide crystal pipe modeling method based on Bayesian statistics |
| CN110717601A (en) * | 2019-10-15 | 2020-01-21 | 厦门铅笔头信息科技有限公司 | Anti-fraud method based on supervised learning and unsupervised learning |
| CN110728024A (en) * | 2019-09-16 | 2020-01-24 | 华东理工大学 | Vine copula-based soft measurement method and system |
| CN110737938A (en) * | 2019-09-28 | 2020-01-31 | 桂林理工大学 | Method and device for predicting shrinkage and creep of recycled concrete based on GPR |
| CN110795841A (en) * | 2019-10-24 | 2020-02-14 | 北京交通大学 | Mathematical modeling method for uncertainty of intermittent energy output |
| CN111222708A (en) * | 2020-01-13 | 2020-06-02 | 浙江大学 | Power plant combustion furnace temperature prediction method based on transfer learning dynamic modeling |
| CN111539444A (en) * | 2020-02-12 | 2020-08-14 | 南阳理工学院 | Gaussian mixture model method for modified mode recognition and statistical modeling |
| CN111612101A (en) * | 2020-06-04 | 2020-09-01 | 华侨大学 | Gene expression data clustering method, device and equipment of nonparametric Watton mixed model |
| CN111797574A (en) * | 2020-07-17 | 2020-10-20 | 浙江工业大学 | Integrated Gaussian process regression model method for polymer molecular weight distribution |
| CN111858991A (en) * | 2020-08-06 | 2020-10-30 | 南京大学 | Small sample learning algorithm based on covariance measurement |
| CN113570070A (en) * | 2021-09-23 | 2021-10-29 | 深圳市信润富联数字科技有限公司 | Streaming data sampling and model updating method, device, system and storage medium |
| CN113792799A (en) * | 2021-09-16 | 2021-12-14 | 平安科技（深圳）有限公司 | Bayesian-based data matching method, device, equipment and readable storage medium |
| CN113962081A (en) * | 2021-10-20 | 2022-01-21 | 江南大学 | Rectifying tower single-ton energy consumption estimation method and system based on auxiliary measurement information |
| CN115096359A (en) * | 2022-06-17 | 2022-09-23 | 北京航空航天大学 | Metal roof health monitoring system and method |
| CN116679026A (en) * | 2023-06-27 | 2023-09-01 | 江南大学 | Self-adaptive unbiased finite impulse response filtering sewage dissolved oxygen concentration estimation method |
| CN117688367A (en) * | 2024-01-25 | 2024-03-12 | 国能日新科技股份有限公司 | Wind power generation ultra-short term power prediction method and device based on instant learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103927412A (en) * | 2014-04-01 | 2014-07-16 | 浙江大学 | Real-time learning debutanizer soft measurement modeling method on basis of Gaussian mixture models |
| CN104699894A (en) * | 2015-01-26 | 2015-06-10 | 江南大学 | JITL (just-in-time learning) based multi-model fusion modeling method adopting GPR (Gaussian process regression) |
| CN106056127A (en) * | 2016-04-07 | 2016-10-26 | 江南大学 | GPR (gaussian process regression) online soft measurement method with model updating |
-
2018
- 2018-05-25 CN CN201810516991.7A patent/CN108804784A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103927412A (en) * | 2014-04-01 | 2014-07-16 | 浙江大学 | Real-time learning debutanizer soft measurement modeling method on basis of Gaussian mixture models |
| CN104699894A (en) * | 2015-01-26 | 2015-06-10 | 江南大学 | JITL (just-in-time learning) based multi-model fusion modeling method adopting GPR (Gaussian process regression) |
| CN106056127A (en) * | 2016-04-07 | 2016-10-26 | 江南大学 | GPR (gaussian process regression) online soft measurement method with model updating |
Non-Patent Citations (1)
| Title |
|---|
| 熊伟丽 等: "一种基于EGMM的高斯过程回归软测量建模", 《信息与控制》 * |
Cited By (35)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110046378A (en) * | 2019-02-28 | 2019-07-23 | 昆明理工大学 | A kind of integrated Gaussian process recurrence soft-measuring modeling method of the selective layering based on Evolutionary multiobjective optimization |
| CN110046377A (en) * | 2019-02-28 | 2019-07-23 | 昆明理工大学 | A kind of selective ensemble instant learning soft-measuring modeling method based on isomery similarity |
| CN110046377B (en) * | 2019-02-28 | 2022-06-14 | 昆明理工大学 | Selective integration instant learning soft measurement modeling method based on heterogeneous similarity |
| CN110046378B (en) * | 2019-02-28 | 2022-09-13 | 昆明理工大学 | Selective hierarchical integration Gaussian process regression soft measurement modeling method based on evolutionary multi-objective optimization |
| CN110197286A (en) * | 2019-05-10 | 2019-09-03 | 武汉理工大学 | A kind of Active Learning classification method based on mixed Gauss model and sparse Bayesian |
| CN110197286B (en) * | 2019-05-10 | 2021-03-16 | 武汉理工大学 | Active learning classification method based on Gaussian mixture model and sparse Bayes |
| CN110083065A (en) * | 2019-05-21 | 2019-08-02 | 浙江大学 | A kind of adaptive soft-sensor method having supervision factorial analysis based on streaming variation Bayes |
| CN110083065B (en) * | 2019-05-21 | 2020-07-10 | 浙江大学 | Self-adaptive soft measurement method based on flow type variational Bayesian supervised factor analysis |
| CN110309491B (en) * | 2019-06-26 | 2022-10-14 | 大连海事大学 | Transient phase partitioning method and system based on local Gaussian mixture model |
| CN110309491A (en) * | 2019-06-26 | 2019-10-08 | 大连海事大学 | A kind of transient phase division methods and system based on local Gaussian mixture model |
| CN110516282B (en) * | 2019-07-03 | 2022-11-15 | 杭州电子科技大学 | Indium phosphide transistor modeling method based on Bayesian statistics |
| CN110516282A (en) * | 2019-07-03 | 2019-11-29 | 杭州电子科技大学 | A kind of indium phosphide crystal pipe modeling method based on Bayesian statistics |
| CN110442942A (en) * | 2019-07-26 | 2019-11-12 | 北京科技大学 | A kind of multiechelon system analysis method for reliability based on Bayes's mixing |
| CN110728024A (en) * | 2019-09-16 | 2020-01-24 | 华东理工大学 | Vine copula-based soft measurement method and system |
| CN110737938A (en) * | 2019-09-28 | 2020-01-31 | 桂林理工大学 | Method and device for predicting shrinkage and creep of recycled concrete based on GPR |
| CN110717601B (en) * | 2019-10-15 | 2022-05-03 | 厦门铅笔头信息科技有限公司 | Anti-fraud method based on supervised learning and unsupervised learning |
| CN110717601A (en) * | 2019-10-15 | 2020-01-21 | 厦门铅笔头信息科技有限公司 | Anti-fraud method based on supervised learning and unsupervised learning |
| CN110795841A (en) * | 2019-10-24 | 2020-02-14 | 北京交通大学 | Mathematical modeling method for uncertainty of intermittent energy output |
| CN110795841B (en) * | 2019-10-24 | 2021-10-22 | 北京交通大学 | Mathematical modeling method for uncertainty of intermittent energy output |
| CN111222708A (en) * | 2020-01-13 | 2020-06-02 | 浙江大学 | Power plant combustion furnace temperature prediction method based on transfer learning dynamic modeling |
| CN111222708B (en) * | 2020-01-13 | 2022-09-20 | 浙江大学 | Power plant combustion furnace temperature prediction method based on transfer learning dynamic modeling |
| CN111539444A (en) * | 2020-02-12 | 2020-08-14 | 南阳理工学院 | Gaussian mixture model method for modified mode recognition and statistical modeling |
| CN111539444B (en) * | 2020-02-12 | 2023-10-31 | 湖南理工学院 | Gaussian mixture model method for correction type pattern recognition and statistical modeling |
| CN111612101B (en) * | 2020-06-04 | 2023-02-07 | 华侨大学 | Gene expression data clustering method, device and equipment of nonparametric Watson mixed model |
| CN111612101A (en) * | 2020-06-04 | 2020-09-01 | 华侨大学 | Gene expression data clustering method, device and equipment of nonparametric Watton mixed model |
| CN111797574A (en) * | 2020-07-17 | 2020-10-20 | 浙江工业大学 | Integrated Gaussian process regression model method for polymer molecular weight distribution |
| CN111858991A (en) * | 2020-08-06 | 2020-10-30 | 南京大学 | Small sample learning algorithm based on covariance measurement |
| CN113792799A (en) * | 2021-09-16 | 2021-12-14 | 平安科技（深圳）有限公司 | Bayesian-based data matching method, device, equipment and readable storage medium |
| CN113570070A (en) * | 2021-09-23 | 2021-10-29 | 深圳市信润富联数字科技有限公司 | Streaming data sampling and model updating method, device, system and storage medium |
| CN113962081B (en) * | 2021-10-20 | 2022-05-31 | 江南大学 | Rectifying tower single-ton energy consumption estimation method and system based on auxiliary measurement information |
| CN113962081A (en) * | 2021-10-20 | 2022-01-21 | 江南大学 | Rectifying tower single-ton energy consumption estimation method and system based on auxiliary measurement information |
| CN115096359A (en) * | 2022-06-17 | 2022-09-23 | 北京航空航天大学 | Metal roof health monitoring system and method |
| CN116679026A (en) * | 2023-06-27 | 2023-09-01 | 江南大学 | Self-adaptive unbiased finite impulse response filtering sewage dissolved oxygen concentration estimation method |
| CN117688367A (en) * | 2024-01-25 | 2024-03-12 | 国能日新科技股份有限公司 | Wind power generation ultra-short term power prediction method and device based on instant learning |
| CN117688367B (en) * | 2024-01-25 | 2024-05-03 | 国能日新科技股份有限公司 | Wind power generation ultra-short term power prediction method and device based on instant learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108804784A (en) | A kind of instant learning soft-measuring modeling method based on Bayes's gauss hybrid models | |
| CN109816092B (en) | Deep neural network training method and device, electronic equipment and storage medium | |
| CN104699894B (en) | Gaussian process based on real-time learning returns multi-model Fusion Modeling Method | |
| CN109060001B (en) | Multi-working-condition process soft measurement modeling method based on feature transfer learning | |
| CN108647583B (en) | Face recognition algorithm training method based on multi-target learning | |
| JP4681426B2 (en) | Apparatus and method for analyzing relation between operation and quality in manufacturing process, computer program, and computer-readable recording medium | |
| CN109165735B (en) | Method for generating sample picture based on generation of confrontation network and adaptive proportion | |
| CN109360604B (en) | Ovarian cancer molecular typing prediction system | |
| CN109558902A (en) | A kind of fast target detection method | |
| CN105069483B (en) | The method that a kind of pair of categorized data set is tested | |
| CN107168063B (en) | Soft measurement method based on integrated variable selection type partial least square regression | |
| CN109409425B (en) | Fault type identification method based on neighbor component analysis | |
| CN105740984A (en) | Product concept performance evaluation method based on performance prediction | |
| CN104899135A (en) | Software defect prediction method and system | |
| CN109241493A (en) | Key Performance Indicator flexible measurement method based on Markov random field and EM algorithm | |
| CN110309771A (en) | A kind of EAS sound magnetic system tag recognition algorithm based on GBDT-INSGAII | |
| CN105913078A (en) | Multi-mode soft measurement method for improving adaptive affine propagation clustering | |
| CN104634265A (en) | Soft measurement method for thickness of mineral floating foam layer based on multivariate image feature fusion | |
| CN112215412A (en) | Dissolved oxygen prediction method and device | |
| CN110688484B (en) | Microblog sensitive event speech detection method based on unbalanced Bayesian classification | |
| CN110084301B (en) | Hidden Markov model-based multi-working-condition process working condition identification method | |
| CN110779526B (en) | Path planning method, device and storage medium | |
| CN112149896A (en) | Attention mechanism-based mechanical equipment multi-working-condition fault prediction method | |
| CN106569982A (en) | GPR online soft measurement method and system with singular point detection compensation function | |
| CN107222328B (en) | Method and device for predicting saturation inflection point |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181113 |
|
| RJ01 | Rejection of invention patent application after publication |