CN114707424B - Chemical process soft measurement method based on quality-related slow characteristic analysis algorithm - Google Patents
Chemical process soft measurement method based on quality-related slow characteristic analysis algorithm Download PDFInfo
- Publication number
- CN114707424B CN114707424B CN202210456258.7A CN202210456258A CN114707424B CN 114707424 B CN114707424 B CN 114707424B CN 202210456258 A CN202210456258 A CN 202210456258A CN 114707424 B CN114707424 B CN 114707424B
- Authority
- CN
- China
- Prior art keywords
- quality
- slow
- chemical process
- model
- measurement method
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001311 chemical methods and process Methods 0.000 title claims abstract description 43
- 238000004458 analytical method Methods 0.000 title claims abstract description 39
- 238000000691 measurement method Methods 0.000 title claims abstract description 26
- 238000000034 method Methods 0.000 claims abstract description 94
- 230000008569 process Effects 0.000 claims abstract description 72
- 238000012360 testing method Methods 0.000 claims abstract description 43
- 239000013598 vector Substances 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 22
- 238000012824 chemical production Methods 0.000 claims abstract description 8
- QQONPFPTGQHPMA-UHFFFAOYSA-N propylene Natural products CC=C QQONPFPTGQHPMA-UHFFFAOYSA-N 0.000 claims description 35
- 238000006116 polymerization reaction Methods 0.000 claims description 33
- 125000004805 propylene group Chemical group [H]C([H])([H])C([H])([*:1])C([H])([H])[*:2] 0.000 claims description 33
- 238000004519 manufacturing process Methods 0.000 claims description 29
- 239000011159 matrix material Substances 0.000 claims description 24
- 238000000354 decomposition reaction Methods 0.000 claims description 13
- 230000009466 transformation Effects 0.000 claims description 12
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000009826 distribution Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 238000003723 Smelting Methods 0.000 claims description 4
- 229910000831 Steel Inorganic materials 0.000 claims description 4
- 239000010959 steel Substances 0.000 claims description 4
- 238000004065 wastewater treatment Methods 0.000 claims description 4
- 238000013461 design Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 230000002087 whitening effect Effects 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 2
- 238000005259 measurement Methods 0.000 abstract description 13
- 238000012544 monitoring process Methods 0.000 abstract description 5
- 239000000155 melt Substances 0.000 description 9
- 239000007791 liquid phase Substances 0.000 description 6
- 239000004743 Polypropylene Substances 0.000 description 2
- 239000003054 catalyst Substances 0.000 description 2
- -1 polypropylene Polymers 0.000 description 2
- 229920001155 polypropylene Polymers 0.000 description 2
- 239000002994 raw material Substances 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 1
- 238000012773 Laboratory assay Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 229910052739 hydrogen Inorganic materials 0.000 description 1
- 239000001257 hydrogen Substances 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005312 nonlinear dynamic Methods 0.000 description 1
- 239000012071 phase Substances 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
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
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
-
- 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
- G16C20/10—Analysis or design of chemical reactions, syntheses or processes
-
- 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
- G16C20/30—Prediction of properties of chemical compounds, compositions or mixtures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/08—Probabilistic or stochastic CAD
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE 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
The invention relates to a chemical process soft measurement method based on a quality-related slow feature analysis algorithm, which comprises the steps of firstly collecting process variables of a chemical process to be predicted in normal operation to form a training sample set, extracting slow features of the training sample by adopting the quality-related slow feature analysis algorithm, and then establishing a related vector regression model based on the extracted slow features; and secondly, collecting technological parameters in the actual running process of the chemical process on line, obtaining a test sample, extracting slow characteristics of the test sample, and estimating a predicted value of the test sample by using an existing model to obtain an on-line quality predicted result of the chemical process. According to the invention, the quality-related slow feature analysis algorithm and the related vector regression model are combined, an effective chemical process soft measurement model is established, the dynamic characteristics and the nonlinearity problems commonly existing in the process can be treated simultaneously, and the online prediction efficiency and performance of the process are improved, so that the chemical production process is more reliable, and the product quality monitoring is more stable.
Description
Technical Field
The invention relates to a process soft measurement method, in particular to a chemical process soft measurement method based on a quality-related slow characteristic analysis algorithm.
Background
Real-time on-line monitoring and control of industrial processes plays a significant role in improving enterprise production efficiency and guaranteeing production safety, and is greatly dependent on measurement of process key product quality indexes. In the actual process, due to factors such as a severe measuring environment, an expensive measuring instrument, measurement hysteresis and the like, the timely acquisition of key quality indexes is quite difficult. The soft measurement technique uses the intrinsic information between the process data to build a mathematical model between the key variable and the auxiliary variable to make a predictive estimate of the key variable. The principal component analysis and partial least squares estimation and the extension method thereof are representative models of the process soft measurement technology. The method is based on static assumption of the process when extracting hidden variables, the samples used for modeling are irrelevant to time, the product quality of the process under the unsteady state working condition cannot be effectively predicted, and the method has low precision and is difficult to use for a long time in practical industrial application.
However, due to the characteristics of the process itself, the application of feedback control, and the dynamic characteristics of system noise, the process is often accompanied by the dynamic characteristics, and for soft measurement of quality indexes, the dynamic data characteristics play an important role in regression analysis, and the regression results have a great influence on the performance of the quality control system. Therefore, a data modeling and soft measurement method aiming at dynamic characteristics of a chemical production process needs to be provided. Meanwhile, due to the reasons of complex process mechanism, multiple production stages, complex operation conditions and the like, the process variable and the quality index of the chemical production process often have strong nonlinearity, the complex data characteristics are fully considered in the soft measurement method in the prior art, and the online prediction efficiency and the online prediction performance of the chemical process are further required to be improved.
Disclosure of Invention
In order to overcome the defects and shortcomings of the prior art, the invention provides a chemical process soft measurement method based on a quality-related slow characteristic analysis algorithm. The method can be used in the propylene polymerization process, takes the process variable and the quality variable of the propylene polymerization production process as modeling samples, adopts a quality-related slow feature analysis algorithm to extract slow features of the samples, and establishes a related vector regression soft measurement method on the basis so as to realize the prediction of the quality variable in the propylene polymerization production process.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
a chemical process soft measurement method based on a quality-related slow feature analysis algorithm comprises the following steps:
(1) Collecting process variables of chemical process to be predicted in normal operation as training sample set for sample data composition modelingWherein->Representing a process variable, input from a DCS database; y represents a quality variable, obtained by laboratory testing;
(2) Preprocessing the process variables in the training sample set;
(3) Extracting slow features of a training sample by adopting a quality-related slow feature analysis algorithm;
(4) Establishing a related vector regression model according to the extracted slow features, and estimating an updated value of the model parameter by using a recursive maximum expectation method in combination with an initial value of the current model parameter;
(5) When the updated values of all the model parameters meet the convergence requirement, obtaining the final value of the model parameters, further obtaining a correlation vector regression model based on quality correlation slow feature analysis, and entering the step (6); otherwise, taking the updated value of the model parameter as an initial value of the model parameter, and returning to the step (4);
(6) And (3) collecting technological parameters in the actual running process of the chemical process to be predicted on line to obtain a test sample set, preprocessing the test sample in the step (2), extracting slow characteristics of the test sample, and calculating a predicted value of the test sample according to the obtained correlation vector regression model based on quality correlation slow characteristic analysis to obtain an on-line predicted result of the chemical process.
Preferably, the chemical process includes a papermaking wastewater treatment process, a steel smelting process and a propylene polymerization production process.
Preferably, the chemical process is a propylene polymerization production process.
By carrying out slow feature analysis on a data set acquired in the propylene polymerization production process, slow features related to the quality index are extracted, and the slow features reflect slow change features of the process and have correlation with the quality index, so that dynamic information of the propylene polymerization process can be effectively captured. Based on Bayesian theory, a correlation vector regression model between slow features and quality indexes is established, and nonlinear relations between process variables and quality indexes can be effectively described. Compared with the existing soft measurement methods of other propylene polymerization processes, the method can greatly improve the nonlinear dynamic modeling and quality prediction effects of the propylene polymerization process, greatly improve the prediction capacity of the model, and realize more effective product quality monitoring based on the model.
Preferably, in step (1), the process variables include temperature, pressure, concentration, weight, pH and flow; the mass variable includes melt index.
In the invention, the chemical process can be various chemical processes, in particular to chemical processes needing multi-parameter control, such as papermaking wastewater treatment process, steel smelting process, propylene polymerization production process and the like. The process variable may be temperature, pressure, concentration, weight, pH, flow, etc. In the propylene polymerization production process, the melt index is one of important quality variables (quality indexes) for determining the product brand of the polypropylene product, but the online analysis and measurement of the melt index are difficult to achieve at present, and the melt index is required to be obtained through sampling offline analysis and analysis, and generally takes 2-4 hours as a measurement unit. The process variables such as temperature, pressure, flow and the like are collected and recorded by the DCS in a large quantity by taking minutes as a measurement unit.
Preferably, in step (2), the data set isPreprocessing, and normalizing according to the following formula:
where u represents the mean value of each process variable, σ represents the standard deviation of each process variable, and X represents the normalized process variable; the preprocessed training sample set is denoted as { X, y }.
Preferably, the step (3) specifically includes the following steps:
(a) The optimization problem of the quality-related slow feature analysis algorithm is as follows:
wherein J (w) represents an objective function, w represents a weight vector,a first order derivative representing the normalized process variable; superscript T denotes matrix transposition;
(b) The Lagrangian multipliers lambda and delta are introduced to convert the optimization problem into a generalized eigenvalue decomposition problem:
wherein I represents an identity matrix;
(c) For covariance matrixAnd (3) performing eigenvalue decomposition:
wherein U is an orthogonal matrix, and Λ is a diagonal matrix; obtaining a whitening matrix
(d) Converting the generalized eigenvalue decomposition problem in step (b) into an eigenvalue decomposition problem as follows:
wherein,representing the feature vector;
(e) Calculating weight vectorsThereby obtaining slow features s of the training sample:
preferably, in step (4), the training sample set of the relevance vector regression model is { s, y }, where s represents a slow feature set extracted from N process variable samples, denoted as s= { s 1 ,s 2 ,…,s N -a }; y represents a sample set corresponding to N output variables, denoted y= { y 1 ,y 2 ,…,y N }. The step (4) comprises the following 3 steps:
(i) Model parameters { alpha, sigma } for a correlation vector regression model 2 Random initialization, where α= [ α ] 0 ,α 1 ,…,α N ] T Is an N+1-dimensional super-parameter vector, N is the number of training samples, sigma 2 Representing the variance;
(ii) According to the sample set { s, y } and the current model parameters, calculating the update values of the posterior mean mu and the covariance sigma of the model, wherein the specific formula is as follows:
wherein A represents a parameter { alpha } by N+1 dimensions 0 ,α 1 ,…,α N Diagonal matrix of } denoted a=diag (α 0 ,α 1 ,…,α N ) The method comprises the steps of carrying out a first treatment on the surface of the Phi represents the transformation phi(s) by the kernel n ) The N× (n+1) -dimensional design matrix is constructed and is denoted as Φ= [ Φ(s) 1 ),φ(s 2 ),…,φ(s N )] T ;φ(s n ) Representation of slow features s n Performing kernel transformation, wherein the kernel transformation formula is phi(s) n )=[1,K(s n ,s 1 ),K(s n ,s 2 ),…,K(s n ,s N )] T The method comprises the steps of carrying out a first treatment on the surface of the Subscript N is a variable, and represents the nth value from 1 to N; s is(s) n As a variable, the slow characteristic of the nth sample is represented, and the value is s 1 To s N The method comprises the steps of carrying out a first treatment on the surface of the The kernel function K (·) in the kernel transformation takes the RBF kernel function of the formula:
wherein sigma 1 As a core parameter, subscript j is a variable, and the value is from 1 to N;
(iii) Calculating model parameters { alpha, sigma } according to the current posterior probability and covariance 2 The specific formula of the updated value of } is:
wherein Σ is ii Is the i-th diagonal element in Σ.
Preferably, in step (5), the updated value Θ of the model parameter is used new With its original model parameters theta old If, for all model parameters, satisfy theta new -Θ old || 2 And (3) if epsilon is less than epsilon, the step (6) is carried out, otherwise, the step (4) is carried out, wherein epsilon is a model convergence threshold value, and the convergence threshold values of a plurality of model parameters can be the same or different.
Preferably, in the step (6), new technological parameters of the chemical production process to be predicted are collected online to obtain a test sample set X test And performing normalization pretreatment, and then extracting slow features s of the test sample test :
s test =X test w (10)
Will s test As input to the model, its predicted value y test The distribution of (2) is:
wherein the variance isThe average value of the distribution is the online prediction result of the chemical production process to be predicted.
Compared with the prior art, the invention has the beneficial effects that:
according to the invention, the quality-related slow feature analysis algorithm and the related vector regression model are combined, an effective chemical process soft measurement model is established, the dynamic characteristics and the nonlinearity problems commonly existing in the process can be treated simultaneously, and the online prediction efficiency and performance of the process are improved, so that the chemical production process is more reliable, and the product quality monitoring is more stable.
Drawings
FIG. 1 is a flow chart of the present invention as applied to a propylene polymerization production process.
Detailed Description
The present invention will be further illustrated with reference to examples, but the scope of the present invention is not limited thereto.
Taking propylene polymerization production process as an example, the invention is further described:
a chemical process soft measurement method based on quality-related slow feature analysis algorithm, the chemical process is propylene polymerization production process, the method is aimed at the quality prediction problem of propylene polymerization process, firstly process variable data in normal working state is collected by a distributed control system, and slow features of a sample are extracted by adopting the quality-related slow feature analysis algorithm; and then, a related vector regression model is established according to the extracted slow features, and the structural parameters of the model are estimated by a recursive maximum expectation method. And collecting technological parameters of the propylene polymerization production process on line to obtain a test sample, extracting slow characteristics of the test sample, and estimating a predicted value of the test sample by using an existing correlation vector regression model on the basis to obtain a final quality predicted result.
Referring to fig. 1, the chemical process soft measurement method based on the quality-related slow feature analysis algorithm of the invention comprises the following steps:
(1) Historical data collected by a distributed control system and used for normal operation of propylene polymerization production process is used as a training sample setWherein->Representing a process variable, input from a DCS database; y represents a quality variable, obtained from laboratory assays.
The process variable may be temperature, pressure, concentration, weight, pH, flow, etc. In the propylene polymerization production process, the melt index is one of important quality variables (quality indexes) for determining the product brand of the polypropylene product, but the online analysis and measurement of the melt index are difficult to achieve at present, and the melt index is required to be obtained through sampling offline analysis and analysis, and generally takes 2-4 hours as a measurement unit. The process variables such as temperature, pressure, flow and the like are collected and recorded by the DCS in a large quantity by taking minutes as a measurement unit.
(2) For data setsPreprocessing, specifically normalizing according to the following formula:
where u represents the mean value of each process variable, σ represents the standard deviation of each process variable, and X represents the normalized process variable; the preprocessed training sample set is denoted as { X, y }.
(3) And extracting slow features of the training sample by adopting a quality-related slow feature analysis algorithm, wherein the third step is divided into the following 5 steps:
(a) The optimization problem of the quality-related slow feature analysis algorithm is as follows:
wherein J (w) represents an objective function, w represents a weight vector,a first order derivative representing the normalized process variable; all superscripts T represent matrix transposition;
(b) The Lagrangian multipliers lambda and delta are introduced to convert the optimization problem into a generalized eigenvalue decomposition problem:
wherein I represents an identity matrix;
(c) For covariance matrixAnd (3) performing eigenvalue decomposition:
where U is an orthogonal matrix and Λ is a diagonal matrix. Obtaining a whitening matrix
(d) Converting the generalized eigenvalue decomposition problem in step (b) into an eigenvalue decomposition problem as follows:
wherein,representing eigenvectors, solving the eigenvalue decomposition problem to obtain +.>
(e) Calculating weight vectorsThereby obtaining slow features s of the training sample:
(4) Building a correlation vector regression model according to the extracted slow features, wherein a training sample set of the model is { s, y }, wherein s represents a slow feature set extracted from N process variable samples and is recorded as s = { s 1 ,s 2 ,…,s N -a }; y represents a sample set corresponding to N output variables, denoted y= { y 1 ,y 2 ,…,y N }. Combining the initial value of the current model parameter, and estimating to obtain the updated value of the model parameter by using a recursive maximum expectation method, wherein the step four is divided into the following 3 steps:
(i) Model parameters { alpha, sigma } for a correlation vector regression model 2 Random initialization, where α= [ α ] 0 ,α 1 ,…,α N ] T Is an N+1-dimensional super-parameter vector, N is the number of sample points, sigma 2 Representing the variance;
(ii) According to the sample set { s, y } and the current model parameters, calculating the update values of the posterior mean mu and the covariance sigma of the model, wherein the specific formula is as follows:
wherein A represents a parameter { alpha } by N+1 dimensions 0 ,α 1 ,…,α N Diagonal matrix of } denoted a=diag (α 0 ,α 1 ,…,α N ) The method comprises the steps of carrying out a first treatment on the surface of the Phi represents the transformation phi(s) by the kernel n ) The N× (n+1) -dimensional design matrix is constructed and is denoted as Φ= [ Φ(s) 1 ),φ(s 2 ),…,φ(s N )] T ;φ(s n ) Representation of slow features s n Performing kernel transformation, wherein the kernel transformation formula is phi(s) n )=[1,K(s n ,s 1 ),K(s n ,s 2 ),…,K(s n ,s N )] T The method comprises the steps of carrying out a first treatment on the surface of the Subscript N is a variable, and represents the nth value from 1 to N; s is(s) n As a variable, the slow characteristic of the nth sample is represented, and the value is s 1 To s N The method comprises the steps of carrying out a first treatment on the surface of the The kernel function K (·) in the kernel transformation takes the RBF kernel function of the formula:
wherein sigma 1 As a core parameter, subscript j is a variable, and the value is from 1 to N;
(iii) Calculating model parameters { alpha, sigma } according to the current posterior probability and covariance 2 The specific formula of the updated value of } is:
wherein Σ is ii Is the i-th diagonal element in Σ.
(5) Using updated values theta of model parameters new With its original model parameters theta old If, for all model parameters, satisfy theta new -Θ old || 2 And (3) if epsilon is less than epsilon, the step (6) is carried out, otherwise, the step (4) is returned, wherein epsilon is a threshold value of model convergence.
(6) Collecting new technological parameters of propylene polymerization production process on line to obtain a test sample set X test And performing normalization pretreatment, and then extracting slow features s of the test sample test :
s test =X test w (10)
Will s test As input to the model, its predicted value y test The distribution of (2) is:
wherein the variance isThe average value of the distribution is the online prediction result of the propylene polymerization process.
The method is particularly applied to a propylene polymerization production plant, and the key quality index (melt index) in the propylene polymerization production process is predicted. First, 9 process variables related to the melt index are selected as input variables of the model based on a priori knowledge, and specifically, as shown in table 1, the melt index is taken as an output variable of the model. Sample data of 9 process variables were obtained from the DCS system of propylene polymerization plant, and melt index was analyzed in the laboratory and collected every 4 hours. The results show that the predicted result of the method is basically consistent with the actual melt index, and the root mean square error between the predicted value and the actual value is less than 0.2.
TABLE 1 propylene polymerization Process variable description
Label (Label) | Variable name |
1 | Temperature of No. 1 liquid phase continuous stirred tank reactor |
2 | Pressure of No. 1 liquid phase continuous stirred tank reactor |
3 | Liquid level of No. 1 liquid phase continuous stirred tank reactor |
4 | Propylene flow rate of No. 1 liquid phase continuous stirred tank reactor |
5 | Raw material flow rate of No. 2 liquid phase continuous stirred tank reactor |
6 | Raw material flow rate of No. 1 gas-phase fluidized bed reactor |
7 | Hydrogen concentration in No. 1 liquid phase continuous stirred tank reactor |
8 | Flow rate of the main catalyst |
9 | Flow rate of auxiliary catalyst |
Firstly, collecting process variables of a propylene polymerization production process in normal operation to form a training sample set, extracting slow features of the training sample by adopting a quality-related slow feature analysis algorithm, and then establishing a related vector regression model based on the extracted slow features; and secondly, collecting technological parameters in the actual running process of the propylene polymerization production process on line, obtaining a test sample, extracting slow characteristics of the test sample, and estimating a predicted value of the test sample by using an existing model to obtain an on-line predicted result of the melt index in the propylene polymerization production process. According to the invention, the quality-related slow feature analysis algorithm and the related vector regression model are combined, an effective soft measurement model of the propylene polymerization production process is established, the dynamic characteristics and the nonlinearity problems commonly existing in the process can be treated simultaneously, and the on-line prediction efficiency and performance of the propylene polymerization production process are improved, so that the propylene polymerization production process is more reliable, and the product quality monitoring is more stable.
In the invention, the chemical process can be various chemical processes, in particular to chemical processes needing multi-parameter control, such as papermaking wastewater treatment process, steel smelting process, propylene polymerization production process and the like.
The invention has been described in detail with reference to the examples, but the description is only specific embodiments of the invention and should not be construed as limiting the scope of the invention. It should be noted that, for those skilled in the art, it is intended that all changes and modifications made in the present invention shall fall within the scope of the patent coverage of this invention without departing from the spirit of the present invention.
Claims (7)
1. A chemical process soft measurement method based on a quality-related slow characteristic analysis algorithm is characterized by comprising the following steps:
(1) Collecting process variables of chemical process to be predicted in normal operation as training sample set for sample data composition modelingWherein->Representing a process variable, input from a DCS database; y represents a quality variable, obtained by laboratory testing;
(2) Preprocessing the process variables in the training sample set;
(3) Extracting slow features of a training sample by adopting a quality-related slow feature analysis algorithm; the method specifically comprises the following steps:
(a) The optimization problem of the quality-related slow feature analysis algorithm is as follows:
maxJ(w)=(Xw) T yy T (Xw)
w T w=1
wherein J (w) represents an objective function, w represents a weight vector,a first order derivative representing the normalized process variable; superscript T denotes matrix transposition;
(b) The Lagrangian multipliers lambda and delta are introduced to convert the optimization problem into a generalized eigenvalue decomposition problem:
wherein I represents an identity matrix;
(c) For covariance matrixAnd (3) performing eigenvalue decomposition:
wherein U is an orthogonal matrix, and Λ is a diagonal matrix; obtaining a whitening matrix
(d) Converting the generalized eigenvalue decomposition problem in step (b) into an eigenvalue decomposition problem as follows:
wherein,representing the feature vector;
(e) Calculating weight vectorsThereby obtaining slow features s of the training sample:
(4) Establishing a related vector regression model according to the extracted slow features, and estimating an updated value of the model parameter by using a recursive maximum expectation method in combination with an initial value of the current model parameter;
in step (4), the training sample set of the correlation vector regression model is { s, y }, where s represents a slow feature set extracted from N process variable samples, denoted as s= { s 1 ,s 2 ,…,s N -a }; y represents a sample set corresponding to N output variables, denoted y= { y 1 ,y 2 ,…,y N -a }; the step (4) comprises the following 3 steps:
(i) Model parameters { alpha, sigma } for a correlation vector regression model 2 Random initialization, where α= [ α ] 0 ,α 1 ,…,α N ] T Is an N+1-dimensional super-parameter vector, N is the number of sample points, sigma 2 Representing the variance;
(ii) According to the sample set { s, y } and the current model parameters, calculating the update values of the posterior mean mu and the covariance sigma of the model, wherein the specific formula is as follows:
μ=σ -2 ΣΦ T y
Σ=(σ -2 Φ T Φ+A) -1
wherein A represents a parameter { alpha } by N+1 dimensions 0 ,α 1 ,…,α N Diagonal matrix of } denoted a=diag (α 0 ,α 1 ,…,α N ) The method comprises the steps of carrying out a first treatment on the surface of the Phi represents the transformation phi(s) by the kernel n ) The N× (n+1) -dimensional design matrix is constructed and is denoted as Φ= [ Φ(s) 1 ),φ(s 2 ),…,φ(s N )] T ;φ(s n ) Representation of slow features s n Performing kernel transformation, wherein the kernel transformation formula is phi(s) n )=[1,K(s n ,s 1 ),K(s n ,s 2 ),…,K(s n ,s N )] T The method comprises the steps of carrying out a first treatment on the surface of the Subscript N is a variable, and represents the nth value from 1 to N; s is(s) n As a variable, the slow characteristic of the nth sample is represented, and the value is s 1 To s N The method comprises the steps of carrying out a first treatment on the surface of the The kernel function K (·) in the kernel transformation takes the RBF kernel function of the formula:
wherein sigma 1 As a core parameter, subscript j is a variable, and the value is from 1 to N;
(iii) Calculating model parameters { alpha, sigma } according to the current posterior probability and covariance 2 The specific formula of the updated value of } is:
wherein Σ is ii Is the i-th diagonal element in Σ;
(5) When the updated values of all the model parameters meet the convergence requirement, obtaining the final value of the model parameters, further obtaining a correlation vector regression model based on quality correlation slow feature analysis, and entering the step (6); otherwise, taking the updated value of the model parameter as an initial value of the model parameter, and returning to the step (4);
(6) And (3) collecting technological parameters in the actual running process of the chemical process to be predicted on line to obtain a test sample set, preprocessing the test sample in the step (2), extracting slow characteristics of the test sample, and calculating a predicted value of the test sample according to the obtained correlation vector regression model based on quality correlation slow characteristic analysis to obtain an on-line predicted result of the chemical process.
2. The chemical process soft measurement method based on the quality-related slow feature analysis algorithm according to claim 1, wherein the chemical process soft measurement method is characterized by: the chemical process comprises a papermaking wastewater treatment process, a steel smelting process and a propylene polymerization production process.
3. The chemical process soft measurement method based on the quality-related slow feature analysis algorithm according to claim 2, wherein the chemical process soft measurement method is characterized by: the chemical process is propylene polymerization production process.
4. The chemical process soft measurement method based on the quality-related slow feature analysis algorithm according to claim 1, wherein the chemical process soft measurement method is characterized by: in step (1), the process variables include temperature, pressure, concentration, weight, pH and flow; the mass variable includes melt index.
5. The chemical process soft measurement method based on the quality-related slow feature analysis algorithm according to claim 1 or 4, wherein the chemical process soft measurement method is characterized in that: in step (2), for the datasetPreprocessing, and normalizing according to the following formula:
where u represents the mean value of each process variable, σ represents the standard deviation of each process variable, and X represents the normalized process variable; the preprocessed training sample set is denoted as { X, y }.
6. The mass-based phase of claim 5The chemical process soft measurement method of the slow feature analysis algorithm is characterized by comprising the following steps of: in step (5), the updated value Θ of the model parameter is used new With its original model parameters theta old If, for all model parameters, satisfy theta new -Θ old || 2 And (3) if epsilon is less than epsilon, the step (6) is carried out, otherwise, the step (4) is carried out, wherein epsilon is a model convergence threshold value, and the convergence threshold values of a plurality of model parameters can be the same or different.
7. The chemical process soft measurement method based on the quality-related slow feature analysis algorithm according to claim 6, wherein the chemical process soft measurement method is characterized by: in the step (6), new technological parameters of the chemical production process to be predicted are collected online to obtain a test sample set X test And performing normalization pretreatment, and then extracting slow features s of the test sample test :
s test =X test w
Will s test As input to the model, its predicted value y test The distribution of (2) is:
wherein the variance isThe average value of the distribution is the online prediction result of the chemical production process to be predicted.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210456258.7A CN114707424B (en) | 2022-04-27 | 2022-04-27 | Chemical process soft measurement method based on quality-related slow characteristic analysis algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210456258.7A CN114707424B (en) | 2022-04-27 | 2022-04-27 | Chemical process soft measurement method based on quality-related slow characteristic analysis algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114707424A CN114707424A (en) | 2022-07-05 |
CN114707424B true CN114707424B (en) | 2024-04-09 |
Family
ID=82176415
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210456258.7A Active CN114707424B (en) | 2022-04-27 | 2022-04-27 | Chemical process soft measurement method based on quality-related slow characteristic analysis algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114707424B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116861224B (en) * | 2023-09-04 | 2023-12-01 | 鲁东大学 | Intermittent process soft measurement modeling system based on intermittent process soft measurement modeling method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2011085819A1 (en) * | 2010-01-15 | 2011-07-21 | Zenrobotics Ltd. | A machine-learning system and a method for determining different operating points in such a system |
CN107423503A (en) * | 2017-07-21 | 2017-12-01 | 江南大学 | The GPR modelings estimated based on the slow signature analysis of core and time lag |
CN108549908A (en) * | 2018-04-13 | 2018-09-18 | 浙江科技学院 | Chemical process fault detection method based on more sampled probability core principle component models |
CN110210687A (en) * | 2019-06-13 | 2019-09-06 | 中南大学 | A kind of Nonlinear Dynamic production process product quality prediction technique returned based on local weighted slow feature |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105205224B (en) * | 2015-08-28 | 2018-10-30 | 江南大学 | Time difference Gaussian process based on fuzzy curve analysis returns soft-measuring modeling method |
-
2022
- 2022-04-27 CN CN202210456258.7A patent/CN114707424B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2011085819A1 (en) * | 2010-01-15 | 2011-07-21 | Zenrobotics Ltd. | A machine-learning system and a method for determining different operating points in such a system |
CN107423503A (en) * | 2017-07-21 | 2017-12-01 | 江南大学 | The GPR modelings estimated based on the slow signature analysis of core and time lag |
CN108549908A (en) * | 2018-04-13 | 2018-09-18 | 浙江科技学院 | Chemical process fault detection method based on more sampled probability core principle component models |
CN110210687A (en) * | 2019-06-13 | 2019-09-06 | 中南大学 | A kind of Nonlinear Dynamic production process product quality prediction technique returned based on local weighted slow feature |
Non-Patent Citations (2)
Title |
---|
基于核慢特征回归与互信息的常压塔软测量建模;蒋昕;杜红彬;李绍军;;化工学报;20151231(05);245-254 * |
基于非线性慢特征回归的动态软测量建模;狄克松;王宇红;尚超;黄德先;;计算机与应用化学;20161128(11);29-33 * |
Also Published As
Publication number | Publication date |
---|---|
CN114707424A (en) | 2022-07-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109060001B (en) | Multi-working-condition process soft measurement modeling method based on feature transfer learning | |
CN108897286B (en) | Fault detection method based on distributed nonlinear dynamic relation model | |
CN108803520B (en) | Dynamic process monitoring method based on variable nonlinear autocorrelation rejection | |
US10919791B2 (en) | Intelligent identification method of sludge bulking based on type-2 fuzzy neural network | |
CN103927412B (en) | Instant learning debutanizing tower soft-measuring modeling method based on gauss hybrid models | |
CN110009020B (en) | Nonlinear process monitoring method based on multi-core principal component analysis model | |
CN109492265B (en) | Wastewater effluent index prediction method based on dynamic nonlinear PLS soft measurement method | |
CN108345284B (en) | Quality-related fault detection method based on two variable blocks | |
CN108549908B (en) | Chemical process fault detection method based on multi-sampling probability kernel principal component model | |
CN112904810B (en) | Process industry nonlinear process monitoring method based on effective feature selection | |
CN109947076B (en) | Industrial process fault diagnosis method based on Bayesian information criterion | |
CN108445867B (en) | non-Gaussian process monitoring method based on distributed ICR model | |
CN114707424B (en) | Chemical process soft measurement method based on quality-related slow characteristic analysis algorithm | |
CN113012766B (en) | Self-adaptive soft measurement modeling method based on online selective integration | |
CN113780420B (en) | GRU-GCN-based method for predicting concentration of dissolved gas in transformer oil | |
CN110942258B (en) | Performance-driven industrial process anomaly monitoring method | |
CN110084301B (en) | Hidden Markov model-based multi-working-condition process working condition identification method | |
CN108427398B (en) | Dynamic process monitoring method based on distributed AR-PLS model | |
CN112149054B (en) | Construction and application of orthogonal neighborhood preserving embedding model based on time sequence expansion | |
CN106599391B (en) | Association vector machine soft measurement modeling method based on dynamic weighting of triangle angle values | |
CN112214006A (en) | Intermittent process fault detection method and system considering two-dimensional dynamic characteristics | |
CN113159088A (en) | Fault monitoring and diagnosis method based on multi-feature fusion and width learning | |
CN116776252A (en) | Industrial process soft measurement method and system for improving Mallow's Cp variable selection | |
CN111188761A (en) | Monitoring method for pump equipment based on Fourier-CVA model | |
CN114861759A (en) | Distributed training method of linear dynamic system model |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |