CN103839103B - Propylene polymerization production process BP Optimal predictor system and method - Google Patents

Abstract

The invention discloses a kind of propylene polymerization production process BP Optimal predictor system, comprise propylene polymerization production process, field intelligent instrument, control station, store data DCS database, based on the Optimal predictor system of continuous space ant group algorithm training multimode BP neural network and melt index forecast value display instrument.Field intelligent instrument and control station are connected with propylene polymerization production process, are connected with DCS database; Optimal predictor system is connected with DCS database and predicted value display instrument.The described Optimal predictor system based on continuous space ant group algorithm training multimode BP neural network comprises model modification module, data preprocessing module, PCA principal component analysis (PCA) module, neural network model module and neural network multimode and optimizes module.And provide the forecasting procedure of a kind of forecast system realization.The present invention realizes on-line measurement, on-line parameter optimization, forecast speed is fast, model upgrades automatically, antijamming capability is strong, precision is high.

Description

Propylene polymerization production process BP Optimal predictor system and method

Technical field

The present invention relates to a kind of Optimal predictor system and method, specifically a kind of propylene polymerization production process BP Optimal predictor system and method.

Background technology

Polypropylene is a kind of thermoplastic resin obtained by propylene polymerization, the most important downstream product of propylene, 50% of World Propylene, and 65% of China's propylene is all used to polypropylene processed, is one of five large general-purpose plastics, closely related with our daily life.Polypropylene is fastest-rising interchangeable heat plastic resin in the world, and total amount is only only second to tygon and Polyvinylchloride.For making China's polypropylene product, there is the market competitiveness, crushing-resistant copolymerization product, random copolymerization product, BOPP and CPP film material, fiber, nonwoven cloth that exploitation rigidity, toughness, mobility balance, and exploitation polypropylene is in the application of automobile and field of household appliances, is all research topic important from now on.

Melting index is one of important quality index of polypropylene product determination product grade, which determine the different purposes of product, be an important step of production quality control during polypropylene is produced to the measurement of melting index, to production and scientific research, have very important effect and directive significance.

But; the on-line analysis of melting index is measured and is difficult at present accomplish; being the shortage of online melting index analyser on the one hand, is that the inaccurate difficulty that even cannot normally use in caused use measured by existing in-line analyzer owing to often can block on the other hand.Therefore, the measurement of MI in current commercial production, is mainly obtained by hand sampling, off-line assay, and general every 2-4 hour can only analyze once, time lag is large, and the quality control of producing to propylene polymerization brings difficulty, becomes in production the bottleneck problem being badly in need of solving.The online forecasting system and method research of polypropylene melt index, thus become a forward position and the focus of academia and industry member.

Summary of the invention

In order to the deficiency of the impact that the measuring accuracy overcoming current existing propylene polymerization production process is not high, be subject to human factor, the object of the present invention is to provide a kind of on-line measurement, on-line parameter optimization, forecast speed is fast, model upgrades automatically, antijamming capability is strong, precision is high based on the propylene polymerization production process melting index Optimal predictor system and method for continuous space ant group algorithm training multimode BP neural network.

The technical solution adopted for the present invention to solve the technical problems is:

A kind of propylene polymerization production process BP Optimal predictor system, comprise propylene polymerization production process, for measuring the field intelligent instrument easily surveying variable, for measuring the control station of performance variable, the DCS database of store data, based on Optimal predictor system and the melt index forecast display instrument of continuous space ant group algorithm training multimode BP neural network, described field intelligent instrument, control station is connected with propylene polymerization production process, described field intelligent instrument, control station is connected with DCS database, described DCS database is connected with training the input end of the Optimal predictor system of multimode BP neural network based on continuous space ant group algorithm, the output terminal of the described Optimal predictor system based on continuous space ant group algorithm training multimode BP neural network is connected with melt index forecast display instrument, it is characterized in that: the described Optimal predictor system based on continuous space ant group algorithm training multimode BP neural network comprises:

(1), data preprocessing module, carry out pre-service for the mode input variable will inputted from DCS database, to input variable centralization, namely deduct the mean value of variable; Be normalized again, namely divided by the constant interval of variate-value;

(2), PCA principal component analysis (PCA) module, for by input variable pre-whitening processing and variable decorrelation, realize by applying a linear transformation to input variable, namely major component is obtained by C=xU, wherein x is input variable, and C is principal component scores matrix, and U is loading matrix.If be reconstructed raw data, can by x=CU ^tcalculate, the wherein transposition of subscript T representing matrix.When the major component number chosen is less than the variable number of input variable, x=CU ^t+ E, wherein E is residual matrix;

(3), neural network model module, for adopting BP neural network, having been minimized a kind of nonlinear being input to output by error function, in mapping, keep topological invariance; Need to set up some sub neural networks, the training objective of first sub-BP network is forecast result and actual result gap J ₁minimum;

$J_1=\frac{1}{N}\underset{l=1}{\overset{N}{\mathrm{\Σ}}}{(F_1\left(x_l\right)-d\left(x_l\right))}^2---\left(1\right)$

N is number of samples, and x is input variable, and l is sample point sequence number, F ₁() is sub-network forecast result, and d () is actual result.

From second sub-network, training objective becomes and makes the prediction error of network little as far as possible, the difference that the forecast result of network is large as far as possible again with network forecast result before simultaneously, and objective function is as follows:

$J_i=\frac{1}{N}\underset{l=1}{\overset{N}{\mathrm{\Σ}}}{(F_i\left(x_l\right)-d\left(x_l\right))}^2-\frac{\mathrm{\λ}}{N}\underset{l=1}{\overset{N}{\mathrm{\Σ}}}{(F_i\left(x_l\right)-F\left(x_l\right))}^2---\left(2\right)$

Ji is the training objective of a front i sub-network, F _i() is the forecast result of i-th network; D () is actual result; F () is the synthesis result of a front i-1 sub-network; λ is regulating parameter, and N is number of samples.

The end condition of training is that after the new sub-network obtained is added multimode neural network, the prediction error of network group no longer reduces.

Adopt a kind of continuous space ant group algorithm to train and optimization each BP network, concrete steps are:

A () algorithm initialization, constructs initial disaggregation S=(s according to RBF neural structure to be optimized ₁, s ₂..., s _n), n is the number of initial solution, and sn is the n-th initial solution, determines the size M of ant group, arranges the threshold value MaxGen of ant optimization algorithm iteration number of times and the iterations sequence number gen=0 of initialization ant optimization;

B () calculates fitness value G corresponding to disaggregation S _i(i=1,2 ..., n), fitness value is larger represents Xie Yuehao; Determining to separate according to following formula again concentrates each solution by the probability P of initial solution got as ant optimizing _i(i=1,2 ..., n)

$P_a\left(k\right)=\frac{G_a}{{\mathrm{\Σ}}_{a=1}^n{G}_i}(a=\mathrm{1,2},\·\·\·,n)---\left(3\right)$

N is the number of initial solution, and sn is the n-th initial solution, and k is iterations.Initialization performs the ant numbering a=0 of optimizing algorithm;

C () ant a chooses the initial solution of a solution in S as optimizing, selection rule does wheel disc choosing according to P;

D () ant a carries out optimizing on the basis of the initial solution chosen, find and better separate s _a';

If e () a<M, then a=a+1, return step c; Otherwise continue to perform step f downwards;

F if, () gen<MaxGen, then gen=gen+1, the better homographic solution separated in replacement S using all ants in steps d to obtain, returns step b; Otherwise perform step g downwards;

G () calculates fitness value G corresponding to disaggregation S _a(a=1,2 ..., n), choose the optimum solution of the maximum solution of fitness value as algorithm, terminate algorithm and return.

Each ant can be circulated during optimizing fixing number of times on the basis of its selected initial solution, searches the better probability separated with what improve algorithm.

(4), neural network multimode optimizes module, for building each sub-network in step (5.4)

$O\left(x\right)=\frac{1}{i}\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{F}_j\left(x\right)---\left(4\right)$

I is total sub-network number, and x is input variable, and O () is model output, F _j() is the output of a jth sub-network; Namely the forecast result of final multimode BP neural network is the mean value of each sub-network forecast result.

As preferred a kind of scheme, the described Optimal predictor model based on continuous space ant group algorithm training multimode BP neural network also comprises: model modification module, for the online updating of model, will regularly off-line analysis data be input in training set, upgrade neural network model.

As another scheme preferred: in described continuous space ant group algorithm training multimode BP neural network model, train sub-BP neural network, then built and form neural network inverse system; Selection standard due to sub-network is that prediction error is little, large with other sub-network difference, so the comprehensive forecasting effect of good, the different again sub neural network of these values of forecasting can have better forecast precision and stability.

As another scheme preferred: in PCA principal component analysis (PCA) module, PCA method realizes the pre-whitening processing of input variable, can simplify the input variable of neural network model, and then improve the performance of model.

The forecasting procedure that propylene polymerization production process BP Optimal predictor system realizes, described forecasting procedure specific implementation step is as follows:

(1) to propylene polymerization production process object, according to industrial analysis and Operations Analyst, select performance variable and easily survey the input of variable as model, performance variable and easily survey variable are obtained by DCS database;

(2) pre-service is carried out to sample data, to input variable centralization, namely deduct the mean value of variable; Be normalized again, namely divided by the constant interval of variate-value;

(3) PCA principal component analysis (PCA) module, for by input variable pre-whitening processing and variable decorrelation, realizes by applying a linear transformation to input variable, namely major component is obtained by C=xU, wherein x is input variable, and C is principal component scores matrix, and U is loading matrix.If be reconstructed raw data, can by x=CU ^tcalculate, the wherein transposition of subscript T representing matrix.When the major component number chosen is less than the variable number of input variable, x=CU ^t+ E, wherein E is residual matrix;

(4) set up several initial sub neural network models based on mode input, output data, adopt BP neural network, be input to a kind of nonlinear of output by error minimize, in mapping, keep topological invariance; The training objective of first sub-BP network is forecast result and actual result gap J ₁minimum;

$J_1=\frac{1}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_1\left(x_l\right)-d\left(x_l\right))}^2---\left(1\right)$

N is number of samples, and x is input variable, and l is sample point sequence number, F ₁() is sub-network forecast result, and d () is actual result.

From second sub-network, training objective becomes and makes the prediction error of network little as far as possible, the difference that the forecast result of network is large as far as possible again with network forecast result before simultaneously, and objective function is as follows:

$J_i=\frac{1}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_i\left(x_l\right)-d\left(x_l\right))}^2-\frac{\mathrm{\&lambda;}}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_i\left(x_l\right)-F\left(x_l\right))}^2---\left(2\right)$

Ji is the training objective of a front i sub-network, F _i() is the forecast result of i-th network; D () is actual result; F () is the synthesis result of a front i-1 sub-network; λ is regulating parameter, and N is number of samples.

The end condition of training is that after the new sub-network obtained is added multimode neural network, the prediction error of network group no longer reduces.

Adopt a kind of continuous space ant group algorithm to train and optimization each BP network, concrete steps are:

A () algorithm initialization, constructs initial disaggregation S=(s according to RBF neural structure to be optimized ₁, s ₂..., s _n), n is the number of initial solution, and sn is the n-th initial solution, determines the size M of ant group, arranges the threshold value MaxGen of ant optimization algorithm iteration number of times and the iterations sequence number gen=0 of initialization ant optimization;

B () calculates fitness value G corresponding to disaggregation S _i(i=1,2 ..., n), fitness value is larger represents Xie Yuehao; Determining to separate according to following formula again concentrates each solution by the probability P of initial solution got as ant optimizing _i(i=1,2 ..., n)

$P_a\left(k\right)=\frac{G_a}{{\mathrm{\&Sigma;}}_{a=1}^n{G}_i}(a=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,n)---\left(3\right)$

N is the number of initial solution, and sn is the n-th initial solution, and k is iterations.Initialization performs the ant numbering a=0 of optimizing algorithm;

C () ant a chooses the initial solution of a solution in S as optimizing, selection rule does wheel disc choosing according to P;

D () ant a carries out optimizing on the basis of the initial solution chosen, find and better separate s _a';

If e () a<M, then a=a+1, return step c; Otherwise continue to perform step f downwards;

F if, () gen<MaxGen, then gen=gen+1, the better homographic solution separated in replacement S using all ants in steps d to obtain, returns step b; Otherwise perform step g downwards;

G () calculates fitness value G corresponding to disaggregation S _a(a=1,2 ..., n), choose the optimum solution of the maximum solution of fitness value as algorithm, terminate algorithm and return.

Each ant can be circulated during optimizing fixing number of times on the basis of its selected initial solution, searches the better probability separated with what improve algorithm.

(5), neural network multimode optimizes module, for building each sub-network in step (4)

$O\left(x\right)=\frac{1}{i}\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{F}_j\left(x\right)---\left(4\right)$

I is total sub-network number, and x is input variable, and O () is model output, F _j() is the output of a jth sub-network; Namely the forecast result of final multimode BP neural network is the mean value of each sub-network forecast result.

Technical conceive of the present invention is:

Online Optimal predictor is carried out to the important quality index melting index of propylene polymerization production process, overcome that existing polypropylene melt index measurement instrument measuring accuracy is not high, the deficiency of the impact that is subject to human factor, set up by the method for continuous space ant group algorithm training multimode BP neural network that forecast precision is high, the forecasting model of good stability obtains optimum forecast result.

Beneficial effect of the present invention is mainly manifested in: 1, on-line measurement; 2, on-line parameter Automatic Optimal; 3, forecast that speed is fast; 4, model upgrades automatically; 5, antijamming capability is strong; 6, precision is high.

Accompanying drawing explanation

Fig. 1 is the basic structure schematic diagram of propylene polymerization production process BP Optimal predictor system and method;

Fig. 2 is the Optimal predictor system architecture schematic diagram based on continuous space ant group algorithm training multimode BP neural network;

Fig. 3 is propylene polymerization production process Hypol explained hereafter process flow diagram.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described further.The embodiment of the present invention is used for explaining and the present invention is described, instead of limits the invention, and in the protection domain of spirit of the present invention and claim, any amendment make the present invention and change, all fall into protection scope of the present invention.

Embodiment 1

1. with reference to Fig. 1, Fig. 2 and Fig. 3, a kind of propylene polymerization production process BP Optimal predictor system, comprise propylene polymerization production process 1, for measuring the field intelligent instrument 2 easily surveying variable, for measuring the control station 3 of performance variable, the DCS database 4 of store data, based on Optimal predictor system 5 and the melt index forecast value display instrument 6 of continuous space ant group algorithm training multimode BP neural network, described field intelligent instrument 2, control station 3 is connected with propylene polymerization production process 1, described field intelligent instrument 2, control station 3 is connected with DCS database 4, described DCS database 4 is connected with training the input end of the Optimal predictor system 5 of multimode BP neural network based on continuous space ant group algorithm, the output terminal of the described Optimal predictor system 5 based on continuous space ant group algorithm training multimode BP neural network is connected with melt index forecast value display instrument 6, the described Optimal predictor system based on continuous space ant group algorithm training multimode BP neural network comprises:

(1), data preprocessing module, carry out pre-service for the mode input variable will inputted from DCS database, to input variable centralization, namely deduct the mean value of variable; Be normalized again, namely divided by the constant interval of variate-value;

(2), PCA principal component analysis (PCA) module, for by input variable pre-whitening processing and variable decorrelation, realize by applying a linear transformation to input variable, namely major component is obtained by C=xU, wherein x is input variable, and C is principal component scores matrix, and U is loading matrix.If be reconstructed raw data, can by x=CU ^tcalculate, the wherein transposition of subscript T representing matrix.When the major component number chosen is less than the variable number of input variable, x=CU ^t+ E, wherein E is residual matrix;

(3), neural network model module, for adopting BP neural network, having been minimized a kind of nonlinear being input to output by error function, in mapping, keep topological invariance; Need to set up some sub neural networks, the training objective of first sub-BP network is forecast result and actual result gap J ₁minimum;

$J_1=\frac{1}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_1\left(x_l\right)-d\left(x_l\right))}^2---\left(1\right)$

N is number of samples, and x is input variable, and l is sample point sequence number, F ₁() is sub-network forecast result, and d () is actual result.

From second sub-network, training objective becomes and makes the prediction error of network little as far as possible, the difference that the forecast result of network is large as far as possible again with network forecast result before simultaneously, and objective function is as follows:

$J_i=\frac{1}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_i\left(x_l\right)-d\left(x_l\right))}^2-\frac{\mathrm{\&lambda;}}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_i\left(x_l\right)-F\left(x_l\right))}^2---\left(2\right)$

Ji is the training objective of a front i sub-network, F _i() is the forecast result of i-th network; D () is actual result; F () is the synthesis result of a front i-1 sub-network; λ is regulating parameter, and N is number of samples.

The end condition of training is that after the new sub-network obtained is added multimode neural network, the prediction error of network group no longer reduces.

Adopt a kind of continuous space ant group algorithm to train and optimization each BP network, concrete steps are:

A () algorithm initialization, constructs initial disaggregation S=(s according to RBF neural structure to be optimized ₁, s ₂..., s _n), n is the number of initial solution, and sn is the n-th initial solution, determines the size M of ant group, arranges the threshold value MaxGen of ant optimization algorithm iteration number of times and the iterations sequence number gen=0 of initialization ant optimization;

B () calculates fitness value G corresponding to disaggregation S _i(i=1,2 ..., n), fitness value is larger represents Xie Yuehao; Determining to separate according to following formula again concentrates each solution by the probability P of initial solution got as ant optimizing _i(i=1,2 ..., n)

$P_a\left(k\right)=\frac{G_a}{{\mathrm{\&Sigma;}}_{a=1}^n{G}_i}(a=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,n)---\left(3\right)$

N is the number of initial solution, and sn is the n-th initial solution, and k is iterations.Initialization performs the ant numbering a=0 of optimizing algorithm;

C () ant a chooses the initial solution of a solution in S as optimizing, selection rule does wheel disc choosing according to P;

D () ant a carries out optimizing on the basis of the initial solution chosen, find and better separate s _a';

If e () a<M, then a=a+1, return step c; Otherwise continue to perform step f downwards;

F if, () gen<MaxGen, then gen=gen+1, the better homographic solution separated in replacement S using all ants in steps d to obtain, returns step b; Otherwise perform step g downwards;

G () calculates fitness value G corresponding to disaggregation S _a(a=1,2 ..., n), choose the optimum solution of the maximum solution of fitness value as algorithm, terminate algorithm and return.

Each ant can be circulated during optimizing fixing number of times on the basis of its selected initial solution, searches the better probability separated with what improve algorithm.

(4), neural network multimode optimizes module, for building each sub-network in step (3)

$O\left(x\right)=\frac{1}{i}\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{F}_j\left(x\right)---\left(4\right)$

I is total sub-network number, and x is input variable, and O () is model output, F _j() is the output of a jth sub-network; Namely the forecast result of final multimode BP neural network is the mean value of each sub-network forecast result.

In PCA principal component analysis (PCA) module, PCA method realizes the pre-whitening processing of input variable, can simplify the input variable of neural network model, and then improves the performance of model.

2. propylene polymerization production process process flow diagram as shown in Figure 3, according to reaction mechanism and flow process analysis, consider in polypropylene production process each factor that melting index has an impact, get nine performance variables conventional in actual production process and easily survey variable as mode input variable, have: three bursts of propylene feed flow rates, major catalyst flow rate, cocatalyst flow rate, temperature in the kettle, pressure, liquid level, hydrogen volume concentration in still.

Table 1 lists 9 the mode input variablees inputted as the Optimal predictor system 5 based on continuous space ant group algorithm training multimode BP neural network, to be respectively in temperature in the kettle (T), still in pressure (p), still hydrogen volume concentration (X in liquid level (L), still _v), 3 bursts of propylene feed flow rates (first gang of propylene feed flow rate f1, second gang of propylene feed flow rate f2, the 3rd gang of propylene feed flow rate f3), 2 bursts of catalyst charge flow rates (major catalyst flow rate f4, cocatalyst flow rate f5).Polyreaction in reactor is that reaction mass mixes rear participation reaction repeatedly, and therefore mode input variable relates to the mean value in process variable employing front some moment of material.The data acquisition mean value of last hour in this example.Melting index off-line laboratory values is as the output variable of the Optimal predictor system 5 based on continuous space ant group algorithm training multimode BP neural network.Obtained by hand sampling, off-line assay, within every 4 hours, analyze and gather once.

Table 1 trains mode input variable needed for the Optimal predictor system of multimode BP neural network based on continuous space ant group algorithm

Field intelligent instrument 2 and control station 3 are connected with propylene polymerization production process 1, are connected with DCS database 4; Optimal predictor system 5 is connected with DCS database 4 and predicted value display instrument 6.Field intelligent instrument 2 measures the easy survey variable that propylene polymerization produces object, is transferred to DCS database 4 by easily surveying variable; Control station 3 controls the performance variable that propylene polymerization produces object, performance variable is transferred to DCS database 4.The variable data recorded in DCS database 4 is as the input of the Optimal predictor system 5 based on continuous space ant group algorithm training multimode BP neural network, predicted value display instrument 6 is for showing the output of the Optimal predictor system 5 based on continuous space ant group algorithm training multimode BP neural network, i.e. predicted value.

Based on the Optimal predictor system 5 of continuous space ant group algorithm training multimode BP neural network, comprising:

(1) data preprocessing module 7, for carrying out pre-service to mode input, i.e. centralization and normalization.To input variable centralization, deduct the mean value of variable exactly, make variable be the variable of zero-mean, thus shortcut calculation; To input variable normalization, be exactly the constant interval divided by input variable value, be that the value of variable is fallen within-0.5 ~ 0.5, simplify further.

(2) PCA principal component analysis (PCA) module 8, for to input variable pre-whitening processing and variable decorrelation, a linear transformation is applied to input variable, make between each component of variable after converting uncorrelated mutually, its covariance matrix is unit battle array simultaneously, and namely major component is obtained by C=xU, and wherein x is input variable, C is principal component scores matrix, and U is loading matrix.If be reconstructed raw data, can by x=CU ^tcalculate, the wherein transposition of subscript T representing matrix.When the major component number chosen is less than the variable number of input variable, x=CU ^t+ E, wherein E is residual matrix.

(3) neural network model module 9, adopt BP neural network, BP neural network can learn and store a large amount of input-output mode map relations, and without the need to disclosing the math equation describing this mapping relations in advance.Its learning rules use method of steepest descent, constantly adjusted the weights and threshold of network, make the error sum of squares of network minimum by backpropagation.BP neural network model has the hidden layer of an input layer, an output layer and indefinite number.Can prove in theory, BP neural network can approach nonlinear system arbitrarily.BP neural network BP training algorithm has minimized a kind of nonlinear being input to output by error function, keep topological invariance in mapping; Need to set up some sub neural networks, the training objective of first sub-BP network is forecast result and actual result gap J ₁minimum;

$J_1=\frac{1}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_1\left(x_l\right)-d\left(x_l\right))}^2---\left(1\right)$

N is number of samples, and x is input variable, and l is sample point sequence number, F ₁() is sub-network forecast result, and d () is actual result.

From second sub-network, training objective becomes and makes the prediction error of network little as far as possible, the difference that the forecast result of network is large as far as possible again with network forecast result before simultaneously, and objective function is as follows:

$J_i=\frac{1}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_i\left(x_l\right)-d\left(x_l\right))}^2-\frac{\mathrm{\&lambda;}}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_i\left(x_l\right)-F\left(x_l\right))}^2---\left(2\right)$

Ji is the training objective of a front i sub-network, F _i() is the forecast result of i-th network; D () is actual result; F () is the synthesis result of a front i-1 sub-network; λ is regulating parameter, and N is number of samples.

The end condition of training is that after the new sub-network obtained is added multimode neural network, the prediction error of network group no longer reduces.

Adopt a kind of continuous space ant group algorithm to train and optimization each BP network, concrete steps are:

A () algorithm initialization, constructs initial disaggregation S=(s according to RBF neural structure to be optimized ₁, s ₂..., s _n), n is the number of initial solution, and sn is the n-th initial solution, determines the size M of ant group, arranges the threshold value MaxGen of ant optimization algorithm iteration number of times and the iterations sequence number gen=0 of initialization ant optimization;

B () calculates fitness value G corresponding to disaggregation S _i(i=1,2 ..., n), fitness value is larger represents Xie Yuehao; Determining to separate according to following formula again concentrates each solution by the probability P of initial solution got as ant optimizing _i(i=1,2 ..., n)

$P_a\left(k\right)=\frac{G_a}{{\mathrm{\&Sigma;}}_{a=1}^n{G}_i}(a=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,n)---\left(3\right)$

N is the number of initial solution, and sn is the n-th initial solution, and k is iterations.Initialization performs the ant numbering a=0 of optimizing algorithm;

C () ant a chooses the initial solution of a solution in S as optimizing, selection rule does wheel disc choosing according to P;

D () ant a carries out optimizing on the basis of the initial solution chosen, find and better separate s _a';

If e () a<M, then a=a+1, return step c; Otherwise continue to perform step f downwards;

F if, () gen<MaxGen, then gen=gen+1, the better homographic solution separated in replacement S using all ants in steps d to obtain, returns step b; Otherwise perform step g downwards;

G () calculates fitness value G corresponding to disaggregation S _a(a=1,2 ..., n), choose the optimum solution of the maximum solution of fitness value as algorithm, terminate algorithm and return.

Each ant can be circulated during optimizing fixing number of times on the basis of its selected initial solution, searches the better probability separated with what improve algorithm.

(4), neural network multimode optimizes module 10, for building each sub-network in step (3)

$O\left(x\right)=\frac{1}{i}\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{F}_j\left(x\right)---\left(4\right)$

I is total sub-network number, and x is input variable, and O () is model output, F _j() is the output of a jth sub-network; Namely the forecast result of final multimode BP neural network is the mean value of each sub-network forecast result.

In PCA principal component analysis (PCA) module, PCA method realizes the pre-whitening processing of input variable, can simplify the input variable of neural network model, and then improves the performance of model.

(5) model modification module 11, for the online updating of model, is regularly input in training set by off-line analysis data, upgrades neural network model.

Embodiment 2

1., with reference to Fig. 1, Fig. 2 and Fig. 3, a kind of propylene polymerization production process optimal forecasting procedure based on continuous space ant group algorithm training multimode BP neural network comprises the following steps:

(1) to propylene polymerization production process object, according to industrial analysis and Operations Analyst, select performance variable and easily survey the input of variable as model, performance variable and easily survey variable are obtained by DCS database;

(2) pre-service is carried out to sample data, to input variable centralization, namely deduct the mean value of variable; Be normalized again, namely divided by the constant interval of variate-value;

(3) PCA principal component analysis (PCA) module, for by input variable pre-whitening processing and variable decorrelation, realizes by applying a linear transformation to input variable, namely major component is obtained by C=xU, wherein x is input variable, and C is principal component scores matrix, and U is loading matrix.If be reconstructed raw data, can by x=CU ^tcalculate, the wherein transposition of subscript T representing matrix.When the major component number chosen is less than the variable number of input variable, x=CU ^t+ E, wherein E is residual matrix;

(4) set up initial neural network model based on mode input, output data, adopt BP neural network, be input to a kind of nonlinear of output by error minimize, in mapping, keep topological invariance; The training objective of first sub-BP network is forecast result and actual result gap J ₁minimum;

$J_1=\frac{1}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_1\left(x_l\right)-d\left(x_l\right))}^2---\left(1\right)$

N is number of samples, and x is input variable, and l is sample point sequence number, F ₁() is sub-network forecast result, and d () is actual result.

From second sub-network, training objective becomes and makes the prediction error of network little as far as possible, the difference that the forecast result of network is large as far as possible again with network forecast result before simultaneously, and objective function is as follows:

$J_i=\frac{1}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_i\left(x_l\right)-d\left(x_l\right))}^2-\frac{\mathrm{\&lambda;}}{N}\underset{l=1}{\overset{N}{\mathrm{\&Sigma;}}}{(F_i\left(x_l\right)-F\left(x_l\right))}^2---\left(2\right)$

Ji is the training objective of a front i sub-network, F _i() is the forecast result of i-th network; D () is actual result; F () is the synthesis result of a front i-1 sub-network; λ is regulating parameter, and N is number of samples.

The end condition of training is that after the new sub-network obtained is added multimode neural network, the prediction error of network group no longer reduces.

Adopt a kind of continuous space ant group algorithm to train and optimization each BP network, concrete steps are:

A () algorithm initialization, constructs initial disaggregation S=(s according to RBF neural structure to be optimized ₁, s ₂..., s _n), n is the number of initial solution, and sn is the n-th initial solution, determines the size M of ant group, arranges the threshold value MaxGen of ant optimization algorithm iteration number of times and the iterations sequence number gen=0 of initialization ant optimization;

B () calculates fitness value G corresponding to disaggregation S _i(i=1,2 ..., n), fitness value is larger represents Xie Yuehao; Determining to separate according to following formula again concentrates each solution by the probability P of initial solution got as ant optimizing _i(i=1,2 ..., n)

$P_a\left(k\right)=\frac{G_a}{{\mathrm{\&Sigma;}}_{a=1}^n{G}_i}(a=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,n)---\left(3\right)$

N is the number of initial solution, and sn is the n-th initial solution, and k is iterations.Initialization performs the ant numbering a=0 of optimizing algorithm;

C () ant a chooses the initial solution of a solution in S as optimizing, selection rule does wheel disc choosing according to P;

D () ant a carries out optimizing on the basis of the initial solution chosen, find and better separate s _a';

If e () a<M, then a=a+1, return step c; Otherwise continue to perform step f downwards;

F if, () gen<MaxGen, then gen=gen+1, the better homographic solution separated in replacement S using all ants in steps d to obtain, returns step b; Otherwise perform step g downwards;

G () calculates fitness value G corresponding to disaggregation S _a(a=1,2 ..., n), choose the optimum solution of the maximum solution of fitness value as algorithm, terminate algorithm and return.

Each ant can be circulated during optimizing fixing number of times on the basis of its selected initial solution, searches the better probability separated with what improve algorithm.

(5), all sub neural networks are built, for building each sub-network in step (4)

$O\left(x\right)=\frac{1}{i}\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{F}_j\left(x\right)---\left(4\right)$

I is total sub-network number, and x is input variable, and O () is model output, F _j() is the output of a jth sub-network; Namely the forecast result of final multimode BP neural network is the mean value of each sub-network forecast result.

Further, in described step (3), adopt PCA principal component analytical method to realize the pre-whitening processing of input variable, the input variable of neural network model can be simplified, and then improve the performance of model.

2. the concrete implementation step of the method for the present embodiment is as follows:

Step 1: to propylene polymerization production process object 1, according to industrial analysis and Operations Analyst, selects performance variable and easily surveys the input of variable as model.

Step 2: carry out pre-service to sample data, is completed by data preprocessing module 7.

Step 3: carry out principal component analysis (PCA) to through pretreated data, is completed by PCA principal component analysis (PCA) module 8.

Step 4: module 9 sets up some initial neural network models based on mode input, output integrating step (4).Input data obtain as described in step 1, export data and chemically examine acquisition by off-line.

Step 5: module 10 integrating step (5) is got up according to the structure of sub-network prediction error by all sub neural networks;

Step 6: off-line analysis data is regularly input in training set by model modification module 11, upgrades neural network model, and the Optimal predictor system 5 based on continuous space ant group algorithm training multimode BP neural network has been set up.

Step 7: the real-time model input variable data that the Optimal predictor system 5 based on continuous space ant group algorithm training multimode BP neural network established transmits based on DCS database 4 carry out the Optimal predictor based on continuous space ant group algorithm training multimode BP neural network to the melting index of propylene polymerization production process 1.

Step 8: melt index forecast display instrument 6 shows the output of the Optimal predictor system 5 based on continuous space ant group algorithm training multimode BP neural network, completes the display of the Optimal predictor to propylene polymerization production process melting index.

Claims (2)

1. a propylene polymerization production process BP Optimal predictor system, comprise propylene polymerization production process, for measuring the field intelligent instrument easily surveying variable, for measuring the control station of performance variable, the DCS database of store data, based on Optimal predictor system and the melt index forecast display instrument of continuous space ant group algorithm training multimode BP neural network, described field intelligent instrument, control station is connected with propylene polymerization production process, described field intelligent instrument, control station is connected with DCS database, described DCS database is connected with training the input end of the Optimal predictor system of multimode BP neural network based on continuous space ant group algorithm, the output terminal of the described Optimal predictor system based on continuous space ant group algorithm training multimode BP neural network is connected with melt index forecast display instrument, it is characterized in that: the described Optimal predictor system based on continuous space ant group algorithm training multimode BP neural network comprises:

(1), data preprocessing module, carry out pre-service for the mode input variable will inputted from DCS database, to input variable centralization, namely deduct the mean value of variable; Be normalized again, namely divided by the constant interval of variate-value;

(2), PCA principal component analysis (PCA) module, for by input variable pre-whitening processing and variable decorrelation, realize by applying a linear transformation to input variable, namely major component is obtained by C=xU, wherein x is input variable, and C is principal component scores matrix, and U is loading matrix; If be reconstructed raw data, can by x=CU ^tcalculate, the wherein transposition of subscript T representing matrix; When the major component number chosen is less than the variable number of input variable, x=CU ^t+ E, wherein E is residual matrix;

(3), neural network model module, for adopting BP neural network, having been minimized a kind of nonlinear being input to output by error function, in mapping, keep topological invariance; Need to set up some sub neural networks, the training objective of first sub-network is forecast result and actual result gap J ₁minimum;

$J_1=\frac{1}{N}\underset{l=1}{\overset{N}{\&Sigma;}}{\left(F_1\right(x_l)-d(x_l\left)\right)}^2---\left(1\right)$

N is number of samples, and x is input variable, and l is sample point sequence number, F ₁() is sub-network forecast result, and d () is actual result;

From second sub-network, training objective becomes and makes the prediction error of network little as far as possible, the difference that the forecast result of network is large as far as possible again with network forecast result before simultaneously, and objective function is as follows:

$J_i=\frac{1}{N}\underset{l=1}{\overset{N}{\&Sigma;}}{\left(F_i\right(x_l)-d(x_l\left)\right)}^2-\frac{\mathrm{\&lambda;}}{N}\underset{l=1}{\overset{N}{\&Sigma;}}{\left(F_i\right(x_l)-F(x_l\left)\right)}^2---\left(2\right)$

J _ifor the training objective of a front i sub-network, F _i() is the forecast result of i-th network; D () is actual result; F () is the synthesis result of a front i-1 sub-network; λ is regulating parameter, and N is number of samples;

The end condition of training is that after the new sub-network obtained is added multimode neural network, the prediction error of network group no longer reduces;

Adopt a kind of continuous space ant group algorithm to train and optimization each BP neural network, concrete steps are:

A () algorithm initialization, goes out initial disaggregation S=(s according to BP neural network configuration to be optimized ₁, s ₂..., s _n), n is the number of initial solution, s _nbe the n-th initial solution, determine the size M of ant group, the threshold value MaxGen of ant optimization algorithm iteration number of times is set and the iterations sequence number gen=0 of initialization ant optimization;

B () calculates fitness value G corresponding to disaggregation S _i(i=1,2 ..., n), fitness value is larger represents Xie Yuehao; Determining to separate according to following formula again concentrates each solution by the probability P of initial solution got as ant optimizing _i(i=1,2 ..., n)

$P_a\left(k\right)=\frac{G_a}{{\mathrm{\&Sigma;}}_{a=1}^n{G}_i},(a=1,2,...,n)---\left(3\right)$

N is the number of initial solution, s _nbe the n-th initial solution, k is iterations; Initialization performs the ant numbering a=0 of optimizing algorithm;

C () ant a chooses the initial solution of a solution in S as optimizing, selection rule does wheel disc choosing according to P;

D () ant a carries out optimizing on the basis of the initial solution chosen, find and better separate s _a';

If e () a<M, then a=a+1, return step c; Otherwise continue to perform step f downwards;

F if, () gen<MaxGen, then gen=gen+1, the better homographic solution separated in replacement S using all ants in steps d to obtain, returns step b; Otherwise perform step g downwards;

G () calculates fitness value G corresponding to disaggregation S _a(a=1,2 ..., n), choose the optimum solution of the maximum solution of fitness value as algorithm, terminate algorithm and return;

Each ant can be circulated during optimizing fixing number of times on the basis of its selected initial solution, searches the better probability separated with what improve algorithm;

(4), neural network multimode optimizes module, for building each sub-network in step (3)

$O\left(x\right)=\frac{1}{i}\underset{j=1}{\overset{i}{\&Sigma;}}{F}_j\left(x\right)---\left(4\right)$

I is total sub-network number, and x is input variable, and O () is model output, F _j() is the output of a jth sub-network; Namely the forecast result of final multimode BP neural network is the mean value of each sub-network forecast result;

The described Optimal predictor system based on continuous space ant group algorithm training multimode BP neural network also comprises: model modification module, for the online updating of model, is regularly input in training set by off-line analysis data, upgrades neural network model;

The described sub-BP neural network of Optimal predictor systematic training based on continuous space ant group algorithm training multimode BP neural network, is then built and is formed neural network inverse system; Selection standard due to sub-network is that prediction error is little, large with other sub-network difference, so the comprehensive forecasting effect of good, the different again sub neural network of these values of forecasting can have better forecast precision and stability; In PCA principal component analysis (PCA) module, PCA method realizes the pre-whitening processing of input variable, can simplify the input variable of neural network model, and then improves the performance of model.

2., by the forecasting procedure that propylene polymerization production process BP Optimal predictor system as claimed in claim 1 realizes, it is characterized in that: described forecasting procedure specific implementation step is as follows:

(5.1) to propylene polymerization production process object, according to industrial analysis and Operations Analyst, select performance variable and easily survey the input of variable as model, performance variable and easily survey variable get temperature, pressure, liquid level, hydrogen gas phase percentage, 3 strands of propylene feed flow velocitys and 2 strands of these variablees of catalyst charge flow velocity, are obtained by DCS database;

(5.2) pre-service is carried out to sample data, to input variable centralization, namely deduct the mean value of variable; Be normalized again, namely divided by the constant interval of variate-value;

(5.3) PCA principal component analysis (PCA) module, for by input variable pre-whitening processing and variable decorrelation, realizes by applying a linear transformation to input variable, namely major component is obtained by C=xU, wherein x is input variable, and C is principal component scores matrix, and U is loading matrix; If be reconstructed raw data, can by x=CU ^tcalculate, the wherein transposition of subscript T representing matrix; When the major component number chosen is less than the variable number of input variable, x=CU ^t+ E, wherein E is residual matrix;

(5.4) set up several initial sub neural network models based on mode input, output data, adopt BP neural network, be input to a kind of nonlinear of output by error minimize, in mapping, keep topological invariance; The training objective of first sub-BP network is forecast result and actual result gap J ₁minimum;

$J_1=\frac{1}{N}\underset{l=1}{\overset{N}{\&Sigma;}}{\left(F_1\right(x_l)-d(x_l\left)\right)}^2---\left(1\right)$

N is number of samples, and x is input variable, and l is sample point sequence number, F ₁() is sub-network forecast result, and d () is actual result;

From second sub-network, training objective becomes and makes the prediction error of network little as far as possible, the difference that the forecast result of network is large as far as possible again with network forecast result before simultaneously, and objective function is as follows:

$J_i=\frac{1}{N}\underset{l=1}{\overset{N}{\&Sigma;}}{\left(F_i\right(x_l)-d(x_l\left)\right)}^2-\frac{\mathrm{\&lambda;}}{N}\underset{l=1}{\overset{N}{\&Sigma;}}{\left(F_i\right(x_l)-F(x_l\left)\right)}^2---\left(2\right)$

J _ifor the training objective of a front i sub-network, F _i() is the forecast result of i-th network; D () is actual result; F () is the synthesis result of a front i-1 sub-network; λ is regulating parameter, and N is number of samples;

The end condition of training is that after the new sub-network obtained is added multimode neural network, the prediction error of network group no longer reduces;

Adopt a kind of continuous space ant group algorithm to train and optimization each BP neural network, concrete steps are:

A () algorithm initialization, goes out initial disaggregation S=(s according to BP neural network configuration to be optimized ₁, s ₂..., s _n), n is the number of initial solution, s _nbe the n-th initial solution, determine the size M of ant group, the threshold value MaxGen of ant optimization algorithm iteration number of times is set and the iterations sequence number gen=0 of initialization ant optimization;

B () calculates fitness value G corresponding to disaggregation S _i(i=1,2 ..., n), fitness value is larger represents Xie Yuehao; Determining to separate according to following formula again concentrates each solution by the probability P of initial solution got as ant optimizing _i(i=1,2 ..., n)

$P_a\left(k\right)=\frac{G_a}{{\mathrm{\&Sigma;}}_{a=1}^n{G}_i},(a=1,2,...,n)---\left(3\right)$

N is the number of initial solution, s _nbe the n-th initial solution, k is iterations; Initialization performs the ant numbering a=0 of optimizing algorithm;

C () ant a chooses the initial solution of a solution in S as optimizing, selection rule does wheel disc choosing according to P;

D () ant a carries out optimizing on the basis of the initial solution chosen, find and better separate s _a';

If e () a<M, then a=a+1, return step c; Otherwise continue to perform step f downwards;

F if, () gen<MaxGen, then gen=gen+1, the better homographic solution separated in replacement S using all ants in steps d to obtain, returns step b; Otherwise perform step g downwards;

G () calculates fitness value G corresponding to disaggregation S _a(a=1,2 ..., n), choose the optimum solution of the maximum solution of fitness value as algorithm, terminate algorithm and return;

Each ant can be circulated during optimizing fixing number of times on the basis of its selected initial solution, searches the better probability separated with what improve algorithm;

(5.5) neural network multimode optimizes module, for building each sub-network in step (5.4)

$O\left(x\right)=\frac{1}{i}\underset{j=1}{\overset{i}{\&Sigma;}}{F}_j\left(x\right)---\left(4\right)$

I is total sub-network number, and x is input variable, and O () is model output, F _j() is the output of a jth sub-network; Namely the forecast result of final multimode BP neural network is the mean value of each sub-network forecast result;

Described forecasting procedure also comprises: be regularly input in training set by off-line analysis data, upgrades neural network model;

Described forecasting procedure trains sub-BP neural network, is then built and forms neural network inverse system; Selection standard due to sub-network is that prediction error is little, large with other sub-network difference, so the comprehensive forecasting effect of good, the different again sub neural network of these values of forecasting can have better forecast precision and stability; In described step (5.3), adopt PCA principal component analytical method to realize the pre-whitening processing of input variable, the input variable of neural network model can be simplified, and then improve the performance of model.