CN102609593A - Polypropylene melt index predicating method based on multiple priori knowledge mixed model - Google Patents
Polypropylene melt index predicating method based on multiple priori knowledge mixed model Download PDFInfo
- Publication number
- CN102609593A CN102609593A CN2012100550178A CN201210055017A CN102609593A CN 102609593 A CN102609593 A CN 102609593A CN 2012100550178 A CN2012100550178 A CN 2012100550178A CN 201210055017 A CN201210055017 A CN 201210055017A CN 102609593 A CN102609593 A CN 102609593A
- Authority
- CN
- China
- Prior art keywords
- model
- variable
- neural network
- melt index
- priori
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Injection Moulding Of Plastics Or The Like (AREA)
Abstract
The invention discloses a polypropylene melt index predicating method based on a multiple priori knowledge mixed model, which fully explores and utilizes priori knowledge of a polypropylene industrial site, and is used for organically integrating various priori knowledge, embedding the priori knowledge into a multilayer perceptron neural network in a non-linear equality constraint form, and optimizing a network weight number by means of a particle swarm optimization algorithm based on an augmented Lagrange multiplier constraint processing mechanism. Based on the multiple priori knowledge neural network model, the multiple priori knowledge neural network model is organically integrated with a polypropylene melt index simplification mechanism model into a harmonic average mixed soft-measuring model. The multiple priori knowledge mixed soft-measuring modeling method has good fitting prediction ability, and is capable of enhancing model extrapolation capacity and realizing good unity of model extrapolation and prediction accuracy of polypropylene melt indexes. Besides, the method is capable of avoiding zero gain and gain inversion and guaranteeing safety in practical polypropylene melt index quality closed-loop control application.
Description
Technical field
The present invention relates to the soft measurement forecast of polypropylene industrial process field, especially, relate to the soft measurement forecasting procedure of a kind of polypropylene melt index based on multiple priori mixture model.
Background technology
Polypropylene (Polypropylene is abbreviated as PP) is to be the polymkeric substance that monomer polymerization forms with the propylene.Polypropylene is a kind of general-purpose plastics of function admirable, is widely used in fields such as packing, manufacturing, weaving and many civilian consumption.Melting index is one of main quality index of polypropylene product, reflects the resin flows performance, thereby is determining the trade mark of resin.The polyacrylic resin trade mark is various, reaches hundreds of from several kinds.As far as the polypropylene process units,,, stern challenge has been proposed for the modeling and the control of melting index if will produce the polypropylene product of high-quality, many trades mark because working conditions change is frequent in the operation of propylene polymerization.
The modeling difficulty of production quality control mainly contains following several aspect in the polypropylene production:
1) lack the melting index in-line analyzer, melting index can only be through hand sampling, and the off-line assay obtains, and analyzes once in general every 2-4 hour, and time lag is big, is difficult to satisfy the requirement of control in real time, causes the product quality fluctuation big.
2) the propylene polymerization processes principle is complicated, comprises multistep physics, chemical reaction; Influence factor is numerous, and is non-linear strong, the multiple dimensioned characteristic of measurement data; When producing different product, the trade mark switches frequent, and operating conditions changes greatly; For the process of such complicacy, be difficult to adopt the model of a single operating mode to describe.
3) the polypropylene production status changes, and mismatch takes place model easily in the soft instrument long-time running, needs regularly to safeguard and proofread and correct.
Traditional polypropylene soft measurement model mainly contains mechanism model and data-driven model, based on the simple rerum natura mechanism model of correlation experience exploitation, can the course of reaction approximate trend change, and precision of prediction but is difficult to guarantee.In fact; The polyolefin polymerization course of reaction exists severe nonlinear and uncertainty, is an extremely complicated chemical reaction, develop a desirable melting index quality modelling by mechanism and take time and effort; In addition the produced on-site condition is different, not promotion.Though and it is convenient to set up the melting index soft-sensing model based on the clear data driving method, and to a certain degree can satisfy modeling accuracy.But for the occasion that such polymerization process trade mark switches, the working point often changes; Leaving the operating area of modeling data, simple application data driving method not only lacks the extrapolation performance; And when the quality closed-loop control; The blindly extrapolation of training data zone causes maloperation easily, brings tremendous influence to production.
Summary of the invention
The objective of the invention is provides a kind of based on the polypropylene melt index soft measurement forecasting procedure of multiple priori neural network with simple mechanism mixture model in order to overcome the not high deficiency that can't guarantee with clear data driving model extrapolability and security performance of existing pure mechanism model precision.This method is the harmonic average mixture model with multiple priori neural network model and the combination of polypropylene melt index simplification mechanism model; Good match predictive ability is not only arranged; Strengthened the model extrapolability; Implementation model extrapolation and to the good unification of melting index precision of prediction can be avoided zero gain and gain inversion simultaneously, guarantees the security performance of network in practical application.
Technical solution of the present invention is: fully discover and use the on-the-spot priori of polypropylene industrial; Various prioris are organically blended; And be embedded in the multilayer perceptron neural network with the form of non-linear equality constraint; Based on augmentation method of Lagrange multipliers Constraints Processing mechanism, optimize network weight simultaneously with particle swarm optimization algorithm.On the basis of multiple priori neural network model, itself and the combination of polypropylene melt index simplification mechanism model are harmonic average mixing soft-sensing model.Concrete steps are following:
The data of 1) from on-the-spot DCS station and laboratory assay platform, gathering crucial auxiliary variable of polypropylene production process and leading variable are as training sample;
2) application data pre-processing module comprises that wavelet threshold denoising, robust yardstick leave detection of value point and rejecting and minimax method for normalizing training sample is carried out pre-service, thereby obtains smooth and reliable modeling data, and step is:
(1) wavelet threshold denoising: with original signal through after flexible and the translation; Be decomposed into a series of subband signals with different spatial resolutions, different frequency characteristic and directivity characteristics; These subband signals have local features such as good time domain, frequency domain; These characteristics can be used to represent the local feature of original signal, and then realize the localization analysis to signal time, frequency, extract the waveform character of signal effectively.
(2) the robust yardstick detects and rejects from the value point:
; According to
decision rule; If the situation that
appears in certain sample; Then this time sampled value promptly from the value point, is rejected.Wherein,
is that intermediate value and
of data sequence
is the intermediate value index of intermediate value absolute value deviation
(3) minimax normalization:
Wherein,
is the sample data of not nominal; And minimax sample value;
is the sample data of nominal, and the minimax sample value of nominal;
3) set up normalized mutual information time delay recognition module, utilize this recognition module to obtain time delay, the auxiliary variable of polypropylene melt index forecast model and leading variable are carried out the sequential coupling;
4) fully excavate and utilize the on-the-spot priori of polypropylene industrial, various prioris are organically blended, and be embedded in the multilayer perceptron neural network, make up multiple priori neural network model with the form of nonlinear inequalities constraint;
5) based on augmentation method of Lagrange multipliers Constraints Processing mechanism, be built with trapped particle crowd optimal module multiple priori neural network weight is carried out on-line optimization;
6) on the basis that makes up multiple priori neural network model, this model and the combination of polypropylene melt index simplification mechanism model for mixing soft measurement forecasting model, are forecast polypropylene melt index in harmonic average.
The said normalized mutual information time delay recognition module of setting up, the concrete steps that obtain time delay
are:
(1) sampling period
of setting leading variable polypropylene melt index
; The sampling period of auxiliary variable
is
; Confirm that maximum hysteresis step-length
is
, wherein
;
(2) arrange again according to the sample data of each auxiliary variable of big young pathbreaker
of hysteresis step-length;
,
is time delay to be determined.Again the arrangement of process data is equivalent to and each variable is converted into
individual sub-variable, and wherein the sample number of each variable equals leading variable sample number
in
individual sub-variable;
(3) utilize following formula normalized mutual information method to obtain the normalized mutual information of
individual sub- variable and leading variable
Can get the corresponding time delay step-length of auxiliary variable
does
Then obtain time delay
;
The step of the multiple priori neural network model of described structure module:
(1) makes up multiple priori neural network
In the formula, non-linear latent node adopts non-linear Sigmoid activation function,
Linear latent node adopts linear activation function
This network structure is three layers of feedforward network with single latent layer; Direct annexation has been set up in input and output; More can reflect the I/O relation of process, zero gain occur for fear of real process simultaneously, latent layer is made up of non-linear and linear latent node two parts.
(2) excavate the on-the-spot priori of polypropylene, comprise process gain and characteristic thereof, the monotonicity of process, and process characteristic curve concavity and convexity are concrete by obtaining as follows:
1) process gain and characteristic thereof: for process gain
Given constraint
Mi, MiMinimum and maximal value for gain provide according to scene or practical experience.
In addition; Can guarantee during as
monotonic decay that gains can guarantee the level and smooth increase that gains during as
.
2) monotonicity of process: work as monotone increasing;
is described; When monotone decreasing then,
3) process characteristic curve concavity and convexity: the family curve between process variable, generally can find out the concavity and convexity of curve.For
; During as
; The process characteristic curve shows concavity; Otherwise during as
, be convexity.
(3) design object function.Objective function imposed restriction to be embedded into priori in the neural network, and on the basis of error term, various prioris are directly retrained with nonlinear inequalities represent as objective function,
Described augmentation Lagrange multiplier particle group optimizing module step:
(1) sets initial value: the maximum update times of Lagrange multiplier and penalty factor, the maximum iteration time of particle cluster algorithm, population population, population dimension, inertia weight
W, cognitive coefficient c
1With the c of coefficient of association
2,Initial Lagrangian, initial penalty factor.In independent variable space initialization particulate random site and speed.
(2) according to the adaptive value of the Augmented Lagrangian Functions of each particle of computes; To each particulate; If adaptive value is greater than its desired positions, then with it as current desired positions
; If adaptive value is then reset overall optimum position
greater than full crowd's desired positions
Wherein,
;
is target function value;
is variable (following is population);
inequality constrain for violating;
is Lagrange multiplier, and
is penalty factor.
(3) judge whether to satisfy the population maximum iteration time; As do not reach; Then press following formula and upgrade
individual particle's velocity and position, return (2); Otherwise, forward next step to
Wherein: subscript "
" the expression particulate the
Dimension,
tExpression the
tGeneration,
Be particulate
iCurrent location,
Be particulate
Current flight speed,
Be particulate
The individual desired positions that is experienced,
Be overall desired positions.
is two separate random functions.In order to reduce during evolution; Particulate leaves the possibility of search volume;
is defined in the certain limit usually; If in promptly the search volume of problem is limited to
, then can set
.
(4) judge whether to satisfy stopping criterion for iteration, if satisfy, stop calculating the output optimal value; Otherwise the following formula of pressing of Lagrange multiplier and penalty factor upgrades, and returns (2)
Described harmonic average mixes the step of soft measurement forecasting model:
(1), optimizes the simple mechanism model parameter of polypropylene melt index with unconstrained optimization algorithm (genetic algorithm or method of steepest descent algorithm) by the on-the-spot leading variable and the auxiliary variable data of gathering;
(2) by the on-the-spot leading variable and the auxiliary variable data of gathering, according to the objective function of setting up, optimize multiple priori neural network weight parameter with augmentation Lagrange multiplier particle group optimizing module, obtain multiple priori neural network model;
(3) make up harmonic average mixing soft-sensing model
Wherein,
;
;
is respectively multiple priori neural network model, the simple mechanism model of melting index and harmonic average mixing soft-sensing model;
is the weight of multiple priori neural network model, the simple mechanism model of melting index; Adopt the harmonic average method to obtain, promptly
Two submodel variances
wherein,
Can guarantee that like this mixture model
does not have when estimating partially its variance
.
In this harmonic average mixing soft-sensing model, melting index is simplified the general trend that mechanism model is held process, and multiple priori neural network model guarantees the overall performance of prediction through local accuracy's compensation.This method simple, intuitive when guaranteeing the effective accuracy of prediction, strengthens the model extrapolability, with the implementation model extrapolation with to the good unification of melting index precision of prediction.
Beneficial effect of the present invention mainly shows: 1, use normalized mutual information time delay recognition module is carried out the sequential coupling to the auxiliary variable and the leading variable of soft-sensing model, has guaranteed the precision of model prediction to a great extent; 2, fully excavate and utilize the on-the-spot priori of polypropylene industrial; Various prioris are organically blended; And be embedded in the multilayer perceptron neural network with the form of nonlinear inequalities constraint, optimize network weight based on augmentation Lagrange multiplier particle swarm optimization algorithm module simultaneously.Good match predictive ability is not only arranged, can avoid zero gain and gain inversion simultaneously, guarantee the security performance of model in practical application; 3, on the basis of multiple priori neural network model; Itself and the combination of polypropylene melt index simplification mechanism model are multiple priori harmonic average mixing soft-sensing model; Strengthened the model extrapolability, implementation model extrapolation and to the good unification of melting index precision of prediction.
Description of drawings
Fig. 1 dicyclo plumber liquid propylene bulk polymerization schematic diagram of device of planting
Fig. 2 is a theory diagram proposed by the invention
The multiple priori of Fig. 3 is mixed the online application synoptic diagram of soft measurement forecasting model
Fig. 4 is conventional softer measuring method and the inventive method value of forecasting comparison diagram
Fig. 5 is that traditional data drives flexible measurement method (not being with the priori constraint) and the inventive method (constraint of band priori) gain security comparison diagram.
Embodiment
Below in conjunction with accompanying drawing the present invention is further described.
With reference to Fig. 2, Fig. 3, a kind of polypropylene melt index forecast method based on multiple priori mixture model, the practical implementation method is following:
(1) off-line modeling
The polypropylene melt index forecast model based on multiple priori mixture model is set up in elder generation's off-line initialization, and detailed process is following:
1) from the annular tube type liquid propylene bulk polymerization field device (Fig. 1) of certain petro-chemical corporation of China, sets up the needed data of soft-sensing model through two approach collections.According to dicyclo pipe liquid propylene bulk polymerization production technology and reaction mechanism; The auxiliary variable of confirming soft-sensing model is density of hydrogen
, hydrogen feed amount
, propylene monomer inlet amount
, jacket water (J.W.) temperature
and the catalyzer flow rate
etc. of two endless tubes, and leading variable is the melting index of R202 endless tube.Gather device process data (6 minutes sampling times of DCS record in real time from the data server that utilizes manufacturer's information station; Several 4600 groups of original sample) with the laboratory assay data (per 4 hours by laboratory personnel's chemical examination once, totally 115 groups) of the hand-kept R202 endless tube melting index of time period as training sample;
2) application data pre-processing module comprises that wavelet threshold denoising, robust yardstick leave detection of value point and rejecting and minimax method for normalizing training sample is carried out pre-service, thereby obtains smooth and reliable modeling data, and step is:
(1) wavelet threshold denoising: original training sample through after flexible and the translation, is decomposed into a series of tools
The subband signal that different spatial resolutions, different frequency characteristic and directivity characteristics are arranged, and then realize localization analysis to signal time, frequency, extract the waveform character of signal effectively.
(2) the robust yardstick detects and rejects from the value point:
; According to
decision rule; If the situation that
appears in certain sample; Then this time sampled value promptly from the value point, is rejected.
(3) minimax normalization:
Wherein,
is the sample data of not nominal; And minimax sample value;
is the sample data of nominal, and the minimax sample value of nominal;
3) set up normalized mutual information time delay recognition module, utilize this recognition module to obtain time delay, the auxiliary variable of polypropylene melt index forecast model and leading variable are carried out the sequential coupling, step is following:
(1) sampling period of setting leading variable endless tube R202 melting index is
; Auxiliary variable (density of hydrogen, hydrogen feed amount, propylene monomer inlet amount, jacket water (J.W.) temperature and the catalyzer flow rate that comprise two endless tubes) sampling period is
, then maximum hysteresis step-length
is confirmed as
.Leading variable is
with the ratio of auxiliary variable sample number;
(2) arrange again according to the sample data of big each auxiliary variable of young pathbreaker of hysteresis step-length;
,
is time delay to be determined.Again the arrangement of process data is equivalent to and each variable is converted into
individual sub-variable, and wherein the sample number of each variable equals leading variable sample number
in
individual sub-variable;
(3) utilize following formula normalized mutual information method to obtain the normalized mutual information of
individual variable and leading variable
Then can get the corresponding time delay step-length
of each auxiliary variable does
Each variable time delay
then;
4) fully excavate and utilize the on-the-spot priori of polypropylene industrial, various prioris are organically blended, and be embedded in the multilayer perceptron neural network, make up multiple priori neural network model with the form of nonlinear inequalities constraint;
Concrete performing step is following:
(1) makes up multiple priori neural network
In the formula, 12 non-linear latent nodes adopt non-linear Sigmoid activation function,
1 latent node of linearity adopts linear activation function
(2) excavate the on-the-spot priori of polypropylene, comprise process gain and characteristic thereof, the monotonicity of process, and process characteristic curve concavity and convexity are concrete by obtaining as follows:
1) process gain and characteristic thereof: given gain constraint
Wherein
Mi, MiMinimum and maximal value for gain are respectively 0.1 and 1 according to the scene value of providing.
In addition, set
and can guarantee the monotonic decay that gains.
2) monotonicity of process: by the relation of density of hydrogen and polypropylene melt index; Can this process be monotone increasing, i.e.
.
3) process characteristic curve concavity and convexity: by the family curve between the process variable; Can find out that curve is a convexity, i.e.
.
(3) design object function.Objective function imposed restriction to be embedded into priori in the neural network, and on the basis of error term, various prioris are directly retrained with nonlinear inequalities represent as objective function,
5) based on augmentation Lagrange multiplier particle group optimizing module network weight is optimized, concrete steps are:
(1) set initial value: the maximum update times of Lagrange multiplier and penalty factor is 10, and the maximum iteration time 100 of particle cluster algorithm, population population are 35, population dimension 93, inertia weight
WBe 0.9, cognitive coefficient c
1Be 2.3 with the c of coefficient of association
2.Be 2.0.Initialization Lagrangian and penalty factor are in independent variable space initialization particulate random site and speed.
(2) according to the adaptive value of the Augmented Lagrangian Functions of each particle of computes; To each particulate; If adaptive value is greater than its desired positions, then with it as current desired positions
; If adaptive value is then reset overall optimum position
greater than full crowd's desired positions
Wherein,
;
is target function value;
is variable (following is population);
inequality constrain for violating;
is Lagrange multiplier, and
is penalty factor.
(3) judge whether to satisfy the population maximum iteration time; As do not reach; Then press following formula and upgrade
individual particle's velocity and position, return (2); Otherwise, forward next step to
(4) judge whether to satisfy stopping criterion for iteration, if satisfy, stop calculating the output optimal value; Otherwise the following formula of pressing of Lagrange multiplier and penalty factor upgrades, and returns (2)
6) on the basis that makes up multiple priori neural network model module, this model and the combination of polypropylene melt index simplification mechanism model module are mixed soft measurement forecasting model for harmonic average.
(1), optimizes the simple mechanism model parameter of following formula polypropylene R202 endless tube melting index with unconstrained optimization algorithm (genetic algorithm or method of steepest descent algorithm) by the on-the-spot leading variable and the auxiliary variable data of gathering;
(2) by on-the-spot leading variable melting index of the polypropylene of gathering and auxiliary variable data; According to the objective function of setting up; Optimize multiple priori neural network model weighting parameter with augmentation Lagrange multiplier particle group optimizing module, obtain multiple priori neural network model;
(3) make up harmonic average mixing soft-sensing model
Wherein,
;
;
is respectively multiple priori neural network model, the simple mechanism model of melting index and harmonic average mixing soft-sensing model;
is the weight of multiple priori neural network model, the simple mechanism model of melting index, adopts the harmonic average method to obtain.
(2) online forecasting
Above-mentioned steps is for mixing soft measurement off-line modeling process.After the modelling, obtain each weights and weight coefficient after, can realize online forecasting, may further comprise the steps:
8) read auxiliary variable data up-to-date in the production run;
9) online forecasting;
10) regularly the normal point of process status is added in the training set, repeats 4)～6) training process so that upgrade model in time.
For the polypropylene melt index forecast method based on multiple priori mixture model that the present invention proposes better is described; Through to the on-the-spot real data modeling of endless tube technology polypropylene, itself and traditional clear data driving model and pure mechanism model are compared.Here set up the multiple priori harmonic average mixing soft-sensing model (MPKNN-SFPM) that least square method supporting vector machine (LSSVM) soft-sensing model, melting index simplification mechanism model (SFPM) and the present invention propose respectively; And adopt performance evaluation index (comprising average square error function, normalization square error function) that each soft-sensing model is made synthetic performance evaluation
Table one, Fig. 4 and Fig. 5 have provided both forecast results.The result shows that the method that the present invention proposes has better prediction ability, and is no matter still regional in extrapolation at total test zone, all measures forecasting procedure than conventional softer and has better precision and tracking trend.The generation that can avoid simultaneously zero gain or gain to reverse guarantees reliability and security in the actual application.
。
Claims (4)
1. based on the polypropylene melt index forecast method of multiple priori mixture model, it is characterized in that may further comprise the steps:
The data of 1) from on-the-spot DCS station and laboratory assay platform, gathering crucial auxiliary variable of polypropylene production process and leading variable are as training sample;
2) application data pre-processing module comprises that wavelet threshold denoising, robust yardstick leave detection of value point and rejecting and minimax method for normalizing training sample is carried out pre-service, thereby obtains smooth and reliable modeling data;
3) set up normalized mutual information time delay recognition module, utilize this recognition module to obtain time delay, the auxiliary variable of polypropylene melt index forecast model and leading variable are carried out the sequential coupling;
4) fully excavate and utilize the on-the-spot priori of polypropylene industrial, various prioris are organically blended, and be embedded in the multilayer perceptron neural network, make up multiple priori neural network model module with the form of nonlinear inequalities constraint;
5) based on augmentation method of Lagrange multipliers Constraints Processing mechanism, be built with trapped particle crowd optimal module the priori neural network weight is carried out on-line optimization;
6) on the basis that makes up multiple priori neural network model, this model and the combination of polypropylene melt index simplification mechanism model module for mixing soft measurement forecasting model, are forecast polypropylene melt index in harmonic average.
2. the polypropylene melt index forecast method based on multiple priori mixture model as claimed in claim 1; It is characterized in that the described normalized mutual information time delay recognition module of setting up, the concrete steps that obtain time delay
are:
(1) sampling period of setting leading variable polypropylene melt index
is
; The sampling period of auxiliary variable
is
; Confirm that maximum hysteresis step-length
is
, wherein
;
(2) arrange again according to the sample data of each auxiliary variable of big young pathbreaker
of hysteresis step-length;
,
is time delay to be determined; Again the arrangement of process data is equivalent to and each variable is converted into
individual sub-variable, and wherein the sample number of each variable equals leading variable sample number
in
individual sub-variable;
(3) utilize following formula normalized mutual information method to obtain the normalized mutual information of
individual sub- variable and leading variable
Wherein
and
is respectively the probability of sub-variable and leading variable melting index, and
is joint probability;
Can get the corresponding time delay step-length of auxiliary variable
does
Then obtain time delay
.
3. the polypropylene melt index forecast method based on multiple priori mixture model as claimed in claim 1 is characterized in that the step of the multiple priori neural network model of described structure module:
(1) makes up multiple priori neural network
In the formula,
Ii,
Jj, kBe respectively input node, latent node, output node number;
,
Corresponding model input and output respectively;
,
,
With
Be respectively input layer to latent layer, latent layer is to output layer, and input layer is to the weights and the output layer threshold value of output layer; Non-linear latent node adopts non-linear Sigmoid activation function,
Linear latent node adopts linear activation function
(2) excavate the on-the-spot priori of polypropylene, comprise process gain and characteristic thereof, the monotonicity of process, and process characteristic curve concavity and convexity, specific as follows:
1) process gain and characteristic thereof:
, and given gain constraint
With
, wherein,
Be hidden node,
Mi, MiBe respectively gain minimum and maximal value, provide according to scene or practical experience; When
The time guarantee the gain monotonic decay, when
The time guarantee that gain-smoothing increases;
2) monotonicity of process: when the process monotone increasing; Then
; Work as monotone decreasing, then
;
3) process characteristic curve concavity and convexity:
; During promptly as
; The process characteristic curve shows concavity; Otherwise during as
, be convexity;
(3) design object function; On the basis of error term, various prioris are directly retrained with nonlinear inequalities represent as objective function:
。
4. the polypropylene melt index forecast method based on multiple priori mixture model as claimed in claim 1 is characterized in that described harmonic average mixes the step of soft measurement forecasting model:
(1) by on-the-spot leading variable of the polypropylene of gathering and auxiliary variable data, with the simple mechanism model parameter of unconstrained optimization algorithm optimization polypropylene melt index;
(2) by the on-the-spot leading variable and the auxiliary variable data of gathering, according to the objective function of setting up, optimize multiple priori neural network weight parameter with augmentation Lagrange multiplier particle group optimizing module, obtain multiple priori neural network model;
(3) make up harmonic average mixing soft-sensing model
Wherein,
;
;
is respectively multiple priori neural network model, the simple mechanism model of melting index and harmonic average mixing soft-sensing model;
is the weight of multiple priori neural network model, the simple mechanism model of melting index, adopts the harmonic average method to obtain.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210055017.8A CN102609593B (en) | 2012-03-05 | 2012-03-05 | Polypropylene melt index predicating method based on multiple priori knowledge mixed model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210055017.8A CN102609593B (en) | 2012-03-05 | 2012-03-05 | Polypropylene melt index predicating method based on multiple priori knowledge mixed model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102609593A true CN102609593A (en) | 2012-07-25 |
| CN102609593B CN102609593B (en) | 2014-01-08 |
Family
ID=46526961
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210055017.8A Active CN102609593B (en) | 2012-03-05 | 2012-03-05 | Polypropylene melt index predicating method based on multiple priori knowledge mixed model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102609593B (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103823369A (en) * | 2013-12-09 | 2014-05-28 | 浙江大学 | Propylene polymerization production process prediction system based on BP multimode network and method |
| CN103824121A (en) * | 2013-12-09 | 2014-05-28 | 浙江大学 | Propylene polymerization production process optimal prediction system based on multimode crowd-sourcing and method |
| CN103823430A (en) * | 2013-12-09 | 2014-05-28 | 浙江大学 | Intelligent weighing propylene polymerization production process optimal soft measurement system and method |
| CN103838206A (en) * | 2013-12-09 | 2014-06-04 | 浙江大学 | Optimal soft measurement meter and method in optimal BP multi-mode propylene polymerization production process |
| CN104361393A (en) * | 2014-09-06 | 2015-02-18 | 华北电力大学 | Method for using improved neural network model based on particle swarm optimization for data prediction |
| CN106127326A (en) * | 2016-05-16 | 2016-11-16 | 葛龙 | A kind of chemical materials processing melt index forecast method |
| CN107015541A (en) * | 2017-04-26 | 2017-08-04 | 燕山大学 | The flexible measurement method being combined based on mutual information and least square method supporting vector machine |
| CN107168255A (en) * | 2017-05-16 | 2017-09-15 | 浙江工业大学 | Polypropylene melt index mixed modeling method based on integrated neural network |
| CN108647485A (en) * | 2018-05-18 | 2018-10-12 | 上海交通大学 | Catalyst carbon deposition measurement method, system, medium and equipment in fluid catalytic cracking |
| CN110135121A (en) * | 2019-06-13 | 2019-08-16 | 中国人民解放军海军航空大学 | A kind of method for diagnosing faults based on Lagrange-population more new algorithm |
| CN111829638A (en) * | 2020-07-24 | 2020-10-27 | 福建世新机器人科技有限公司 | Real-time weight measuring method |
| CN113030156A (en) * | 2021-03-13 | 2021-06-25 | 宁波大学科学技术学院 | Polypropylene melt index soft measurement method based on nonlinear slow characteristic regression model |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5650722A (en) * | 1991-11-20 | 1997-07-22 | Auburn International, Inc. | Using resin age factor to obtain measurements of improved accuracy of one or more polymer properties with an on-line NMR system |
| CN1916791A (en) * | 2006-09-12 | 2007-02-21 | 浙江大学 | Method of soft measuring fusion index of producing propylene through polymerization in industrialization |
| CN101458730A (en) * | 2008-12-11 | 2009-06-17 | 浙江工业大学 | Polypropylene melt index forecast based on non-linear least square improved method |
| CN101458506A (en) * | 2009-01-08 | 2009-06-17 | 浙江工业大学 | Industrial polypropylene producing melt index flexible measurement method based on combination neural net |
| CN102243672A (en) * | 2011-06-22 | 2011-11-16 | 浙江大学 | Gushing operation condition soft sensing modeling method based on hybrid multiple models in shield tunneling process |
-
2012
- 2012-03-05 CN CN201210055017.8A patent/CN102609593B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5650722A (en) * | 1991-11-20 | 1997-07-22 | Auburn International, Inc. | Using resin age factor to obtain measurements of improved accuracy of one or more polymer properties with an on-line NMR system |
| CN1916791A (en) * | 2006-09-12 | 2007-02-21 | 浙江大学 | Method of soft measuring fusion index of producing propylene through polymerization in industrialization |
| CN101458730A (en) * | 2008-12-11 | 2009-06-17 | 浙江工业大学 | Polypropylene melt index forecast based on non-linear least square improved method |
| CN101458506A (en) * | 2009-01-08 | 2009-06-17 | 浙江工业大学 | Industrial polypropylene producing melt index flexible measurement method based on combination neural net |
| CN102243672A (en) * | 2011-06-22 | 2011-11-16 | 浙江大学 | Gushing operation condition soft sensing modeling method based on hybrid multiple models in shield tunneling process |
Non-Patent Citations (3)
| Title |
|---|
| 戴文战,等.: "非线性***神经网络预测控制研究进展", 《控制理论与应用》 * |
| 田华阁: "聚丙烯装置产品质量软测量技术研究", 《中国博士学位论文全文数据库》 * |
| 阮泉,吴铁军: "结合先验知识的神经网络在生化***建模中的应用", 《无锡轻工大学学报》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824121A (en) * | 2013-12-09 | 2014-05-28 | 浙江大学 | Propylene polymerization production process optimal prediction system based on multimode crowd-sourcing and method |
| CN103823430A (en) * | 2013-12-09 | 2014-05-28 | 浙江大学 | Intelligent weighing propylene polymerization production process optimal soft measurement system and method |
| CN103838206A (en) * | 2013-12-09 | 2014-06-04 | 浙江大学 | Optimal soft measurement meter and method in optimal BP multi-mode propylene polymerization production process |
| CN103823369B (en) * | 2013-12-09 | 2016-02-24 | 浙江大学 | Based on propylene polymerization production process forecast system and the method for BP multimode network |
| CN103823430B (en) * | 2013-12-09 | 2016-05-18 | 浙江大学 | Intelligence weighting propylene polymerization production process optimal soft measuring system and method |
| CN103838206B (en) * | 2013-12-09 | 2016-08-17 | 浙江大学 | Optimum BP multimode propylene polymerization production process optimal soft survey instrument and method |
| CN103823369A (en) * | 2013-12-09 | 2014-05-28 | 浙江大学 | Propylene polymerization production process prediction system based on BP multimode network and method |
| CN104361393A (en) * | 2014-09-06 | 2015-02-18 | 华北电力大学 | Method for using improved neural network model based on particle swarm optimization for data prediction |
| CN104361393B (en) * | 2014-09-06 | 2018-02-27 | 华北电力大学 | Data predication method is used for based on the improved neural network model of particle swarm optimization algorithm |
| CN106127326B (en) * | 2016-05-16 | 2020-08-18 | 苏州京高数字科技有限公司 | Chemical material processing melt index forecasting method |
| CN106127326A (en) * | 2016-05-16 | 2016-11-16 | 葛龙 | A kind of chemical materials processing melt index forecast method |
| CN107015541A (en) * | 2017-04-26 | 2017-08-04 | 燕山大学 | The flexible measurement method being combined based on mutual information and least square method supporting vector machine |
| CN107168255A (en) * | 2017-05-16 | 2017-09-15 | 浙江工业大学 | Polypropylene melt index mixed modeling method based on integrated neural network |
| CN108647485A (en) * | 2018-05-18 | 2018-10-12 | 上海交通大学 | Catalyst carbon deposition measurement method, system, medium and equipment in fluid catalytic cracking |
| CN108647485B (en) * | 2018-05-18 | 2021-04-02 | 上海交通大学 | Method, system, medium and equipment for measuring carbon deposition of catalyst in fluid catalytic cracking |
| CN110135121A (en) * | 2019-06-13 | 2019-08-16 | 中国人民解放军海军航空大学 | A kind of method for diagnosing faults based on Lagrange-population more new algorithm |
| CN110135121B (en) * | 2019-06-13 | 2023-04-18 | 中国人民解放军海军航空大学 | Fault diagnosis method based on Lagrange-particle swarm update algorithm |
| CN111829638A (en) * | 2020-07-24 | 2020-10-27 | 福建世新机器人科技有限公司 | Real-time weight measuring method |
| CN113030156A (en) * | 2021-03-13 | 2021-06-25 | 宁波大学科学技术学院 | Polypropylene melt index soft measurement method based on nonlinear slow characteristic regression model |
| CN113030156B (en) * | 2021-03-13 | 2023-02-24 | 宁波大学科学技术学院 | Polypropylene melt index soft measurement method based on nonlinear slow characteristic model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102609593B (en) | 2014-01-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102609593A (en) | Polypropylene melt index predicating method based on multiple priori knowledge mixed model | |
| Han et al. | Flood forecasting using support vector machines | |
| Raissi et al. | Statistical process optimization through multi-response surface methodology | |
| CN101315557B (en) | Propylene polymerization production process optimal soft survey instrument and method based on genetic algorithm optimization BP neural network | |
| CN101315556B (en) | Propylene polymerization production process optimal soft survey instrument and method based on chaos optimization | |
| CN101382801B (en) | Optimum soft measuring instrument based on EGA-optimized polymerization of propylene production process and method | |
| CN107292344B (en) | Robot real-time control method based on environment interaction | |
| Chitralekha et al. | Application of support vector regression for developing soft sensors for nonlinear processes | |
| Varadharajan et al. | Can machine learning accelerate process understanding and decision‐relevant predictions of river water quality? | |
| CN103823430B (en) | Intelligence weighting propylene polymerization production process optimal soft measuring system and method | |
| CN103413038A (en) | Vector quantization based long-term intuitionistic fuzzy time series prediction method | |
| CN103823964A (en) | Global optimum soft measurement instrument and method applied to propylene polymerization production | |
| Chaturvedi | Rainfall Prediction using Back-Propagation Feed Forward Network | |
| Wang et al. | A compensation method for gyroscope random drift based on unscented Kalman filter and support vector regression optimized by adaptive beetle antennae search algorithm | |
| CN104318072A (en) | QKF-MMF (Quantitative Kalman Filtering-Multi Method Fusion) based multi-sensor quantitative fusion method | |
| Asghari et al. | Spatial rainfall prediction using optimal features selection approaches | |
| CN102663220B (en) | Online dynamic forecast method for polypropylene fusion index | |
| Kala et al. | CEEMDAN hybridized with LSTM model for forecasting monthly rainfall | |
| Yoshida | Probabilistic Inverse Analysis for Geotechnics | |
| CN103838206A (en) | Optimal soft measurement meter and method in optimal BP multi-mode propylene polymerization production process | |
| CN103824121A (en) | Propylene polymerization production process optimal prediction system based on multimode crowd-sourcing and method | |
| CN103955170A (en) | Propylene polymerization production process online forecasting system and method based on group intelligent optimization | |
| Wang et al. | Estimating effluent turbidity in the drinking water flocculation process with an improved random forest model | |
| CN103838205A (en) | Optimum soft measurement instrument and method in BP global optimum propylene polymerization production process | |
| CN103823369B (en) | Based on propylene polymerization production process forecast system and the method for BP multimode 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 |