CN103823430B - Intelligence weighting propylene polymerization production process optimal soft measuring system and method - Google Patents

Abstract

The invention discloses a kind of intelligent weighting propylene polymerization production process optimal soft measuring system, comprise the DCS database of propylene polymerization production process, field intelligent instrument, control station, store data, the soft measuring system of intelligent Weighted optimal and melt index flexible measured value display instrument. Field intelligent instrument and control station are connected with propylene polymerization production process, are connected with DCS database; Optimum soft measuring system is connected with DCS database and soft measured value display instrument. The described soft measuring system of intelligent Weighted optimal comprises model modification module, data preprocessing module, PCA principal component analysis module, neural network model module, intelligent weighted optimization module, model modification module. And the flexible measurement method that provides the soft measuring system of a kind of use to realize. The present invention realizes on-line measurement, on-line parameter optimization, soft measuring speed is fast, model upgrades automatically, antijamming capability is strong, precision is high.

Description

Intelligence weighting propylene polymerization production process optimal soft measuring system and method

Technical field

The present invention relates to a kind of optimum soft measuring system and method, specifically a kind of intelligent weighting propylene polymerization is rawProduce the soft measuring system of process optimum and method.

Background technology

Polypropylene is a kind of thermoplastic resin being made by propylene polymerization, the most important downstream product of propylene,50% of World Propylene, 65% of China's propylene is all for 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, totalAmount is only only second to polyethylene and polyvinyl chloride. For making China's polypropylene product have the market competitiveness, exploitation justProperty, toughness, crushing-resistant copolymerization product, random copolymerization product, BOPP and CPP film that mobility balance is goodMaterial, fiber, nonwoven cloth, and exploitation polypropylene is in the application of automobile and field of household appliances, is all important from now onResearch topic.

Melt index is that polypropylene product is determined one of important quality index of product grade, and it has determined productDifferent purposes, be an important step of control of product quality during polypropylene is produced to the measurement of melt index,To producing and scientific research, there are very important effect and directive significance.

But the on-line analysis of melt index is measured and is difficult at present accomplish, be online melt index analysis on the one handThe shortage of instrument is that existing in-line analyzer is forbidden even cannot owing to often can stopping up to measure on the other handThe normal difficulty using in the use causing. Therefore, the measurement of MI in industrial production at present, is mainly logicalCross manual sampling, the acquisition of off-line assay, and within general every 2-4 hour, can only analyze once time lagGreatly, the quality control of producing to propylene polymerization has brought difficulty, becomes a bottleneck being badly in need of solution in productionProblem. The online soft sensor system and method research of polypropylene melt index, thus academia and industry becomeForward position on boundary and focus.

Summary of the invention

For overcome the certainty of measurement of current existing propylene polymerization production process not high, be subject to human factorThe deficiency of impact, the object of the present invention is to provide a kind of on-line measurement, on-line parameter optimization, soft measurement speedThe intelligent weighting propylene polymerization production process that degree is fast, model upgrades automatically, antijamming capability is strong, precision is high is moltenMelt the optimum soft measuring system of index and method.

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

A kind of intelligent weighting propylene polymerization production process optimal soft measuring system, comprise propylene polymerization production process,For measuring the field intelligent instrument of easy survey variable, for measuring control station, the store data of performance variableDCS database, the soft measuring system of intelligent Weighted optimal and melt index flexible are measured display instrument, described sceneIntelligence instrument, control station are connected with propylene polymerization production process, described field intelligent instrument, control station and DCSDatabase connects, and described DCS database is connected with the input of the soft measuring system of intelligent Weighted optimal, described inThe output of the soft measuring system of intelligence Weighted optimal is measured display instrument with melt index flexible and is connected, and it is characterized in that:The soft measuring system of described intelligent Weighted optimal comprises:

(1), data preprocessing module, for carrying out pretreatment from the mode input variable of DCS database input,To input variable centralization, deduct the mean value of variable; Be normalized again, divided by variate-valueConstant interval;

(2), PCA principal component analysis module, for by input variable prewhitening process and variable decorrelation, pass throughInput variable is applied to a linear transformation and realize, principal component is obtained by C=xU, and wherein x is input variable,C is principal component scores matrix, and U is loading matrix. If initial data is reconstructed, can be by x=CU^TMeterCalculate the wherein transposition of subscript T representing matrix. The variable that is less than input variable when the principal component number of choosing is individualWhen number, x=CU^T+ E, wherein E is residual matrix;

(3), neural network model module, for adopting RBF neutral net, minimizing by error functionA kind of nonlinear that becomes to be input to output, keeps topological invariance in mapping; Need to set up someSub neural network, the training objective of first sub-RBF network is forecast result and actual result gap J₁Minimum;

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

N is number of samples, and x is input variable, and l is sample point sequence number, F₁() is sub-network forecast result, d ()For actual result.

Since second sub-network, training objective becomes and makes the prediction error of network as far as possible little, net simultaneouslyThe forecast result of network and network forecast result before large as far as possible difference again, object 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 network; D () is actual knotReally; F () is the synthesis result of a front i-1 sub-network; λ is for regulating parameter, and N is number of samples.

The end condition of training is that the new sub-network obtaining is added after multimode neutral net, the forecast of network groupError no longer reduces.

Adopt a kind of continuous space ant group algorithm to train and optimization each RBF network, concrete stepsFor:

(a) algorithm initialization, constructs initial disaggregation according to RBF neural network structure to be optimizedS=(s₁,s₂,…,s_n), the number that n is initial solution, sn is n initial solution, determines ant group's big or small M, establishesPut the threshold value MaxGen of ant optimization algorithm iteration number of times and initialize the iterations sequence number of ant optimizationgen=0；

(b) calculate the fitness value G that disaggregation S is corresponding_i(i=1,2 ..., n), the larger Xie Yuehao that represents of fitness value;Determining to separate according to following formula again concentrates each solution to be got the probability P as the initial solution of 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 n initial solution, and k is iterations. Initialize and carry out optimizing algorithmAnt numbering a=0;

(c) ant a chooses a solution in the S initial solution as optimizing, and selection rule is to make wheel disc according to PChoosing;

(d) ant a carries out optimizing on the basis of the initial solution of choosing, and finds better solution s_a'；

(e) if a < M, a=a+1, returns to step c; Otherwise continue execution step f downwards;

(f) if gen < MaxGen, gen=gen+1, uses the better solution that all ants obtain in steps dReplace the homographic solution in S, return to step b; Otherwise perform step g downwards;

(g) calculate the fitness value G that disaggregation S is corresponding_a(a=1,2 ..., n), choose the solution conduct of fitness value maximumThe optimal solution of algorithm, finishes algorithm and returns.

Each ant fixing number of times that can circulate when optimizing on the basis of its selected initial solution, if thisCirculation has obtained better solution, in circulation next time, can and keep the direction of search constant based on this solution; OtherwiseThe solution based on original but can adjust the direction of search still in next time circulation;

Along with the increase of whole ant optimization algebraically, the step-length of Ant Search can reduce intelligently, to be applicable to simultaneouslyThe convergence of whole ant optimization:

del_k=Random·k^r(4)

In formula, del_kFor ant k is for the initial step length of iteration, k is iteration algebraically, Random at random toAmount, r is negative normal real number.

For medium-term and long-term those solutions that are not elected to be optimizing initial solution by ant of disaggregation S, can adopt in genetic algorithmVariation and Crossover Strategy are processed, thereby improve the global optimizing performance of algorithm.

(4), intelligent weighted optimization module, for each sub-network of step (3) is composed to weights; Foundation is everyThe prediction error of individual sub-network, error is less, and weights are larger;

$w_q=(1/e_q)/(\underset{j=1}{\overset{I}{\mathrm{\&Sigma;}}}1/e_j),q=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,I.---\left(5\right)$

$e_j=\frac{1}{N}\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}|F_j\left(x_m\right)-d\left(x_m\right)|---\left(6\right)$

Wq is the weights of q sub-network; Eq is the prediction error of q sub-network; I is total sub-networkNumber; J is sub-network sequence number; N is number of samples, and x is input variable, and m is sample point sequence number, F_j(·)Be j sub-network forecast result, d () is actual result.

The forecast result of final multimode neutral net is the weighted sum of each sub-network forecast result.

$O\left(x\right)=\underset{k=1}{\overset{I}{\mathrm{\&Sigma;}}}(w_k\&CenterDot;F_k\left(x\right))---\left(7\right)$

In formula, x is input variable, and O () is model output, F_k() is k sub-network output, w_kBe kThe weight of individual sub-network, I is sub-network sum.

As preferred a kind of scheme, described intelligent Weighted optimal soft-sensing model also comprises: model modification mouldPiece, for the online updating of model, will regularly be input to off-line analysis data in training set, upgrades neuralNetwork model.

As preferred another scheme: at described intelligent continuous space ant group algorithm training weighting multimode RBFIn neural network model, train sub-RBF neutral net, then its weighted array is got up to form neutral netGroup; Because the selection standard of sub-network is that prediction error is little, large with other sub-network difference, so theseValue of forecasting comprehensive forecasting effect good, different sub neural network again can have better forecast essenceDegree and stability.

As preferred another scheme: in PCA principal component analysis module, PCA method realizes input and becomesThe prewhitening processing of amount, can simplify the input variable of neural network model, and then improves the performance of model.

The flexible measurement method that the optimum soft measuring system of intelligent weighting polypropylene production process realizes is described softMeasuring method specific implementation step is as follows:

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

(2) sample data is carried out to pretreatment, to input variable centralization, deduct the mean value of variable; Enter againRow normalized, divided by the constant interval of variate-value;

(3) PCA principal component analysis module, for input variable prewhitening is processed and variable decorrelation, by rightInput variable applies a linear transformation and realizes, and principal component is obtained by C=xU, and wherein x is input variable,C is principal component scores matrix, and U is loading matrix. If initial data is reconstructed, can be by x=CU^TMeterCalculate the wherein transposition of subscript T representing matrix. The variable that is less than input variable when the principal component number of choosing is individualWhen number, x=CU^T+ E, wherein E is residual matrix;

(4) based on mode input, output data intelligence continuous space ant group algorithm training smart set up several at the beginning ofBeginning sub neural network model, adopt RBF neutral net, completed and be input to output by error minimizeA kind of nonlinear, keeps topological invariance in mapping; The training objective of first sub-RBF networkForecast result and actual result gap J₁Minimum;

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

N is number of samples, and x is input variable, and l is sample point sequence number, F₁() is sub-network forecast result, d ()For actual result.

Since second sub-network, training objective becomes and makes the prediction error of network as far as possible little, net simultaneouslyThe forecast result of network and network forecast result before large as far as possible difference again, object 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 network; D () is actual knotReally; F () is the synthesis result of a front i-1 sub-network; λ is for regulating parameter, and N is number of samples.

The end condition of training is that the new sub-network obtaining is added after multimode neutral net, the forecast of network groupError no longer reduces.

Adopt a kind of continuous space ant group algorithm to train and optimization each RBF network, concrete stepsFor:

(a) algorithm initialization, constructs initial disaggregation according to RBF neural network structure to be optimizedS=(s₁,s₂,…,s_n), the number that n is initial solution, sn is n initial solution, determines ant group's big or small M, establishesPut the threshold value MaxGen of ant optimization algorithm iteration number of times and initialize the iterations sequence number of ant optimizationgen=0；

(b) calculate the fitness value G that disaggregation S is corresponding_i(i=1,2 ..., n), the larger Xie Yuehao that represents of fitness value;Determining to separate according to following formula again concentrates each solution to be got the probability P as the initial solution of 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 n initial solution, and k is iterations. Initialize and carry out optimizing algorithmAnt numbering a=0;

(c) ant a chooses a solution in the S initial solution as optimizing, and selection rule is to make wheel disc according to PChoosing;

(d) ant a carries out optimizing on the basis of the initial solution of choosing, and finds better solution s_a'；

(e) if a < M, a=a+1, returns to step c; Otherwise continue execution step f downwards;

(f) if gen < MaxGen, gen=gen+1, uses the better solution that all ants obtain in steps dReplace the homographic solution in S, return to step b; Otherwise perform step g downwards;

(g) calculate the fitness value G that disaggregation S is corresponding_a(a=1,2 ..., n), choose the solution conduct of fitness value maximumThe optimal solution of algorithm, finishes algorithm and returns.

Each ant fixing number of times that can circulate when optimizing on the basis of its selected initial solution, if thisCirculation has obtained better solution, in circulation next time, can and keep the direction of search constant based on this solution; OtherwiseThe solution based on original but can adjust the direction of search still in next time circulation;

Along with the increase of whole ant optimization algebraically, the step-length of Ant Search can reduce intelligently, to be applicable to simultaneouslyThe convergence of whole ant optimization:

del_k=Random·k^r(4)

In formula, del_kFor ant k is for the initial step length of iteration, k is iteration algebraically, Random at random toAmount, r is negative normal real number.

For medium-term and long-term those solutions that are not elected to be optimizing initial solution by ant of disaggregation S, can adopt in genetic algorithmVariation and Crossover Strategy are processed, thereby improve the global optimizing performance of algorithm.

(5), all sub neural networks of weighted array, for each sub-network of step (5.4) is composed to weights;According to being the prediction error of each sub-network, error is less, and weights are larger;

$w_q=(1/e_q)/(\underset{j=1}{\overset{I}{\mathrm{\&Sigma;}}}1/e_j),q=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,I.---\left(5\right)$

$e_j=\frac{1}{N}\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}|F_j\left(x_m\right)-d\left(x_m\right)|---\left(6\right)$

Wq is the weights of q sub-network; Eq is the prediction error of q sub-network; I is total sub-networkNumber; J is sub-network sequence number; N is number of samples, and x is input variable, and m is sample point sequence number, F_j() isJ sub-network forecast result, d () is actual result.

The forecast result of final multimode neutral net is the weighted sum of each sub-network forecast result.

$O\left(x\right)=\underset{k=1}{\overset{I}{\mathrm{\&Sigma;}}}(w_k\&CenterDot;F_k\left(x\right))---\left(7\right)$

In formula, x is input variable, and O () is model output, F_k() is k sub-network output, w_kBe kThe weight of individual sub-network, I is sub-network sum.

Technical conceive of the present invention is: the important quality index melt index to propylene polymerization production process is carried outOnline optimum soft measurement, overcome existing polypropylene melt index measuring instrument certainty of measurement not high, be subject to peopleFor the deficiency of the impact of factor, by intelligent continuous space ant group algorithm training weighting multimode RBF neutral netMethod set up that forecast precision is high, the forecasting model of good stability obtains optimum soft measurement result.

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

Brief description of the drawings

Fig. 1 is the basic structure schematic diagram of intelligent weighting propylene polymerization production process optimal soft measuring system and method;

Fig. 2 is the soft measuring system structural representation of intelligent Weighted optimal;

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

Detailed description of the invention

Below in conjunction with accompanying drawing, the present invention is described further. The embodiment of the present invention is used for the present invention that explains,Instead of limit the invention, in the protection domain of spirit of the present invention and claim, to thisBright any amendment and the change of making, all falls into protection scope of the present invention.

Embodiment 1

1. with reference to Fig. 1, Fig. 2 and Fig. 3, a kind of intelligent weighting propylene polymerization production process optimal soft measuring system,Comprise propylene polymerization production process 1, for measuring the field intelligent instrument 2 of easy survey variable, for measuring operationThe control station 3 of variable, the DCS database 4 of store data, the soft measuring system 5 of intelligent Weighted optimal andMelt index flexible measured value display instrument 6, described field intelligent instrument 2, control station 3 were produced with propylene polymerizationJourney 1 connects, and described field intelligent instrument 2, control station 3 are connected with DCS database 4, described DCS numberBe connected the soft measurement of described intelligent Weighted optimal with the input of the soft measuring system 5 of intelligent Weighted optimal according to storehouse 4The output of system 5 is connected with melt index flexible measured value display instrument 6, the soft measurement of described intelligent Weighted optimalSystem comprises:

(1), data preprocessing module, for carrying out pretreatment from the mode input variable of DCS database input,To input variable centralization, deduct the mean value of variable; Be normalized again, divided by variate-valueConstant interval;

(2), PCA principal component analysis module, for by input variable prewhitening process and variable decorrelation, pass throughInput variable is applied to a linear transformation and realize, principal component is obtained by C=xU, and wherein x is input variable,C is principal component scores matrix, and U is loading matrix. If initial data is reconstructed, can be by x=CU^TMeterCalculate the wherein transposition of subscript T representing matrix. The variable that is less than input variable when the principal component number of choosing is individualWhen number, x=CU^T+ E, wherein E is residual matrix;

(3), neural network model module, for adopting RBF neutral net, minimizing by error functionA kind of nonlinear that becomes to be input to output, keeps topological invariance in mapping; Need to set up someSub neural network, the training objective of first sub-RBF network is forecast result and actual result gap J1 minimum;

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

N is number of samples, and x is input variable, and l is sample point sequence number, F₁() is sub-network forecast result, d ()For actual result.

Since second sub-network, training objective becomes and makes the prediction error of network as far as possible little, net simultaneouslyThe forecast result of network and network forecast result before large as far as possible difference again, object 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 network; D () is actual knotReally; F () is the synthesis result of a front i-1 sub-network; λ is for regulating parameter, and N is number of samples.

The end condition of training is that the new sub-network obtaining is added after multimode neutral net, the forecast of network groupError no longer reduces.

Adopt a kind of continuous space ant group algorithm to train and optimization each RBF network, concrete stepsFor:

(a) algorithm initialization, constructs initial disaggregation according to RBF neural network structure to be optimizedS=(s₁,s₂,…,s_n), the number that n is initial solution, sn is n initial solution, determines ant group's big or small M, establishesPut the threshold value MaxGen of ant optimization algorithm iteration number of times and initialize the iterations sequence number of ant optimizationgen=0；

(b) calculate the fitness value G that disaggregation S is corresponding_i(i=1,2 ..., n), the larger Xie Yuehao that represents of fitness value;Determining to separate according to following formula again concentrates each solution to be got the probability P as the initial solution of 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 n initial solution, and k is iterations. Initialize and carry out optimizing algorithmAnt numbering a=0;

(c) ant a chooses a solution in the S initial solution as optimizing, and selection rule is to make wheel disc according to PChoosing;

(d) ant a carries out optimizing on the basis of the initial solution of choosing, and finds better solution s_a'；

(e) if a < M, a=a+1, returns to step c; Otherwise continue execution step f downwards;

(f) if gen < MaxGen, gen=gen+1, uses the better solution that all ants obtain in steps dReplace the homographic solution in S, return to step b; Otherwise perform step g downwards;

(g) calculate the fitness value G that disaggregation S is corresponding_a(a=1,2 ..., n), choose the solution conduct of fitness value maximumThe optimal solution of algorithm, finishes algorithm and returns.

Each ant fixing number of times that can circulate when optimizing on the basis of its selected initial solution, if thisCirculation has obtained better solution, in circulation next time, can and keep the direction of search constant based on this solution; OtherwiseThe solution based on original but can adjust the direction of search still in next time circulation;

Along with the increase of whole ant optimization algebraically, the step-length of Ant Search can reduce intelligently, to be applicable to simultaneouslyThe convergence of whole ant optimization:

del_k=Random·k^r(4)

In formula, del_kFor ant k is for the initial step length of iteration, k is iteration algebraically, Random at random toAmount, r is negative normal real number.

For medium-term and long-term those solutions that are not elected to be optimizing initial solution by ant of disaggregation S, can adopt in genetic algorithmVariation and Crossover Strategy are processed, thereby improve the global optimizing performance of algorithm.

(4), intelligent weighted optimization module, for each sub-network of step (3) is composed to weights; Foundation is everyThe prediction error of individual sub-network, error is less, and weights are larger;

$w_q=(1/e_q)/(\underset{j=1}{\overset{I}{\mathrm{\&Sigma;}}}1/e_j),q=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,I.---\left(5\right)$

$e_j=\frac{1}{N}\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}|F_j\left(x_m\right)-d\left(x_m\right)|---\left(6\right)$

Wq is the weights of q sub-network; Eq is the prediction error of q sub-network; I is total sub-networkNumber; J is sub-network sequence number; N is number of samples, and x is input variable, and m is sample point sequence number, F_j(·)Be j sub-network forecast result, d () is actual result.

The forecast result of final multimode neutral net is the weighted sum of each sub-network forecast result.

$O\left(x\right)=\underset{k=1}{\overset{I}{\mathrm{\&Sigma;}}}(w_k\&CenterDot;F_k\left(x\right))---\left(7\right)$

In formula, x is input variable, and O () is model output, F_k() is k sub-network output, w_kBe kThe weight of individual sub-network, I is sub-network sum.

In PCA principal component analysis module, PCA method realizes the prewhitening processing of input variable, Neng GoujianChange the input variable of neural network model, and then improve the performance of model.

Propylene polymerization production process flow chart as shown in Figure 3, according to reaction mechanism and flow process analysis,Consider each factor in polypropylene production process, melt index being exerted an influence, get in actual production processNine conventional performance variables and the easy variable of surveying, 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 has been listed 9 mode input variablees inputting as the soft measuring system 5 of intelligent Weighted optimal, pointWei temperature in the kettle (T), hydrogen volume concentration (X in liquid level (L), still in pressure (p), still in still_v）、3 strands of propylene feed flow rates (first gang of propylene feed flow rate f1, second gang of propylene feed flow rate f2, the 3rd strands thirdAlkene feed flow rates f3), 2 bursts of catalyst charge flow rates (major catalyst flow rate f4, cocatalyst flow rate f5).Polymerisation in reactor is that reaction mass mixes rear participation reaction repeatedly, and therefore mode input variable relates toAnd the process variables of material adopts the mean value in front some moment. Average with last hour of data acquisition in this exampleValue. Melt index off-line laboratory values is as the output variable of the soft measuring system 5 of intelligent Weighted optimal. Pass through peopleWork sampling, off-line assay obtain, and within every 4 hours, analyze and gather once.

The required mode input variable of the soft measuring system of the intelligent Weighted optimal of table 1

Field intelligent instrument 2 and control station 3 are connected with propylene polymerization production process 1, with DCS database 4Be connected; Optimum soft measuring system 5 is connected with DCS database 4 and soft measured value display instrument 6. Site intelligentInstrument 2 is measured the easy survey variable of propylene polymerization production object, will easily survey variable and be transferred to DCS database 4;Control station 3 is controlled the performance variable of propylene polymerization production object, and performance variable is transferred to DCS database 4.In DCS database 4, the variable data of record is as the input of the soft measuring system 5 of intelligent Weighted optimal, soft surveyValue display instrument 6 is for showing the output of the soft measuring system 5 of intelligent Weighted optimal, i.e. soft measured value.

The soft measuring system 5 of intelligence Weighted optimal, comprising:

(1) data preprocessing module 7, for mode input is carried out to pretreatment, i.e. centralization and normalization. RightInput variable centralization, deducts the mean value of variable exactly, and making variable is the variable of zero-mean, thereby simplifiesAlgorithm; To input variable normalization, be exactly the constant interval divided by input variable value, be the value of variable is fallenWithin-0.5～0.5, further simplify.

(2) PCA principal component analysis module 8, for to input variable prewhitening, processing is variable decorrelation, to defeatedEnter variable and apply a linear transformation, make between each component of variable after conversion uncorrelated mutually, simultaneously its associationVariance matrix is unit matrix, and principal component is obtained by C=xU, and wherein x is input variable, and C is that principal component obtainsSub matrix, U is loading matrix. If initial data is reconstructed, can be by x=CU^TCalculate wherein subscriptThe transposition of T representing matrix. In the time that the principal component number of choosing is less than the variable number of input variable,x=CU^T+ E, wherein E is residual matrix.

(3) neural network model module 9, adopts RBF neutral net, and multilayer feedforward neural network is in network structureOn conventionally formed by input layer, hidden layer and output layer. On network characterization, main manifestations is for both without refreshing in layerInterconnected through unit, also without the anti-contact of interlayer. This network is in fact a kind of static network, and its output onlyThe function of existing input, and irrelevant in inputing or outputing of past and future. RBF neural network model hasAn input layer, an output layer and a hidden layer. Can prove in theory, RBF neutral net canApproach arbitrarily nonlinear system. RBF neural network BP training algorithm has minimized input by error functionTo a kind of nonlinear of output, in mapping, keep topological invariance; Need to set up some sons neuralNetwork, the training objective of first sub-RBF network is forecast result and actual result gap J₁Minimum;

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

N is number of samples, and x is input variable, and l is sample point sequence number, F₁() is sub-network forecast result, d ()For actual result.

Since second sub-network, training objective becomes and makes the prediction error of network as far as possible little, net simultaneouslyThe forecast result of network and network forecast result before large as far as possible difference again, object 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 network; D () is actual knotReally; F () is the synthesis result of a front i-1 sub-network; λ is for regulating parameter, and N is number of samples.

The end condition of training is that the new sub-network obtaining is added after multimode neutral net, the forecast of network groupError no longer reduces.

Adopt a kind of continuous space ant group algorithm to train and optimization each RBF network, concrete stepsFor:

(a) algorithm initialization, constructs initial disaggregation according to RBF neural network structure to be optimizedS=(s₁,s₂,…,s_n), the number that n is initial solution, sn is n initial solution, determines ant group's big or small M, establishesPut the threshold value MaxGen of ant optimization algorithm iteration number of times and initialize the iterations sequence number of ant optimizationgen=0；

(b) calculate the fitness value G that disaggregation S is corresponding_i(i=1,2 ..., n), the larger Xie Yuehao that represents of fitness value;Determining to separate according to following formula again concentrates each solution to be got the probability P as the initial solution of 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 n initial solution, and k is iterations. Initialize and carry out optimizing algorithmAnt numbering a=0;

(c) ant a chooses a solution in the S initial solution as optimizing, and selection rule is to make wheel disc according to PChoosing;

(d) ant a carries out optimizing on the basis of the initial solution of choosing, and finds better solution s_a′；

(e) if a < M, a=a+1, returns to step c; Otherwise continue execution step f downwards;

(f) if gen < MaxGen, gen=gen+1, uses the better solution that all ants obtain in steps dReplace the homographic solution in S, return to step b; Otherwise perform step g downwards;

(g) calculate the fitness value G that disaggregation S is corresponding_a(a=1,2 ..., n), choose the solution conduct of fitness value maximumThe optimal solution of algorithm, finishes algorithm and returns.

Each ant fixing number of times that can circulate when optimizing on the basis of its selected initial solution, if thisCirculation has obtained better solution, in circulation next time, can and keep the direction of search constant based on this solution; OtherwiseThe solution based on original but can adjust the direction of search still in next time circulation;

Along with the increase of whole ant optimization algebraically, the step-length of Ant Search can reduce intelligently, to be applicable to simultaneouslyThe convergence of whole ant optimization:

del_k=Random·k^r(4)

In formula, del_kFor ant k is for the initial step length of iteration, k is iteration algebraically, Random at random toAmount, r is negative normal real number.

For medium-term and long-term those solutions that are not elected to be optimizing initial solution by ant of disaggregation S, can adopt in genetic algorithmVariation and Crossover Strategy are processed, thereby improve the global optimizing performance of algorithm.

(4), intelligent weighted optimization module 10, for each sub-network of step (3) is composed to weights; According to beingThe prediction error of each sub-network, error is less, and weights are larger;

$w_q=(1/e_q)/(\underset{j=1}{\overset{I}{\mathrm{\&Sigma;}}}1/e_j),q=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,I.---\left(5\right)$

$e_j=\frac{1}{N}\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}|F_j\left(x_m\right)-d\left(x_m\right)|---\left(6\right)$

Wq is the weights of q sub-network; Eq is the prediction error of q sub-network; I is total sub-networkNumber; J is sub-network sequence number; N is number of samples, and x is input variable, and m is sample point sequence number, F_j(·)Be j sub-network forecast result, d () is actual result.

The forecast result of final multimode neutral net is the weighted sum of each sub-network forecast result.

$O\left(x\right)=\underset{k=1}{\overset{I}{\mathrm{\&Sigma;}}}(w_k\&CenterDot;F_k\left(x\right))---\left(7\right)$

In formula, x is input variable, and O () is model output, F_k() is k sub-network output, w_kBe kThe weight of individual sub-network, I is sub-network sum.

In PCA principal component analysis module, PCA method realizes the prewhitening processing of input variable, Neng GoujianChange the input variable of neural network model, and then improve the performance of model.

(5) model modification module 11, for the online updating of model, is regularly input to training by off-line analysis dataConcentrate, upgrade neural network model.

Embodiment 2

1. with reference to Fig. 1, Fig. 2 and Fig. 3, a kind of intelligent weighting propylene polymerization production process optimal soft measuring method bagDraw together following steps:

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

(2) sample data is carried out to pretreatment, to input variable centralization, deduct the mean value of variable; Enter againRow normalized, divided by the constant interval of variate-value;

(3) PCA principal component analysis module, for input variable prewhitening is processed and variable decorrelation, by rightInput variable applies a linear transformation and realizes, and principal component is obtained by C=xU, and wherein x is input variable,C is principal component scores matrix, and U is loading matrix. If initial data is reconstructed, can be by x=CU^TMeterCalculate the wherein transposition of subscript T representing matrix. The variable that is less than input variable when the principal component number of choosing is individualWhen number, x=CU^T+ E, wherein E is residual matrix;

(4) set up initial neural network model based on mode input, output data, adopt RBF neutral net, logicalCross error minimize and complete a kind of nonlinear that is input to output, in mapping, keep topological novariableProperty; The training objective of first sub-RBF network is forecast result and actual result gap J₁Minimum;

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

N is number of samples, and x is input variable, and l is sample point sequence number, F₁() is sub-network forecast result, d ()For actual result.

Since second sub-network, training objective becomes and makes the prediction error of network as far as possible little, net simultaneouslyThe forecast result of network and network forecast result before large as far as possible difference again, object 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 network; D () is actual knotReally; F () is the synthesis result of a front i-1 sub-network; λ is for regulating parameter, and N is number of samples.

The end condition of training is that the new sub-network obtaining is added after multimode neutral net, the forecast of network groupError no longer reduces.

Adopt a kind of continuous space ant group algorithm to train and optimization each RBF network, concrete stepsFor:

(a) algorithm initialization, constructs initial disaggregation according to RBF neural network structure to be optimizedS=(s₁,s₂,…,s_n), the number that n is initial solution, sn is n initial solution, determines ant group's big or small M, establishesPut the threshold value MaxGen of ant optimization algorithm iteration number of times and initialize the iterations sequence number of ant optimizationgen=0；

(b) calculate the fitness value G that disaggregation S is corresponding_i(i=1,2 ..., n), the larger Xie Yuehao that represents of fitness value;Determining to separate according to following formula again concentrates each solution to be got the probability P as the initial solution of 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 n initial solution, and k is iterations. Initialize and carry out optimizing algorithmAnt numbering a=0;

(c) ant a chooses a solution in the S initial solution as optimizing, and selection rule is to make wheel disc according to PChoosing;

(d) ant a carries out optimizing on the basis of the initial solution of choosing, and finds better solution s_a'；

(e) if a < M, a=a+1, returns to step c; Otherwise continue execution step f downwards;

(f) if gen < MaxGen, gen=gen+1, uses the better solution that all ants obtain in steps dReplace the homographic solution in S, return to step b; Otherwise perform step g downwards;

(g) calculate the fitness value G that disaggregation S is corresponding_a(a=1,2 ..., n), choose the solution conduct of fitness value maximumThe optimal solution of algorithm, finishes algorithm and returns.

Each ant fixing number of times that can circulate when optimizing on the basis of its selected initial solution, if thisCirculation has obtained better solution, in circulation next time, can and keep the direction of search constant based on this solution; OtherwiseThe solution based on original but can adjust the direction of search still in next time circulation;

Along with the increase of whole ant optimization algebraically, the step-length of Ant Search can reduce intelligently, to be applicable to simultaneouslyThe convergence of whole ant optimization:

del_k=Random·k^r(4)

In formula, del_kFor ant k is for the initial step length of iteration, k is iteration algebraically, Random at random toAmount, r is negative normal real number.

For medium-term and long-term those solutions that are not elected to be optimizing initial solution by ant of disaggregation S, can adopt in genetic algorithmVariation and Crossover Strategy are processed, thereby improve the global optimizing performance of algorithm.

(5), all sub neural networks of weighted array, for each sub-network of step (5.4) is composed to weights;According to being the prediction error of each sub-network, error is less, and weights are larger;

$w_q=(1/e_q)/(\underset{j=1}{\overset{I}{\mathrm{\&Sigma;}}}1/e_j),q=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,I.---\left(5\right)$

$e_j=\frac{1}{N}\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}|F_j\left(x_m\right)-d\left(x_m\right)|---\left(6\right)$

Wq is the weights of q sub-network; Eq is the prediction error of q sub-network; I is total sub-networkNumber; J is sub-network sequence number; N is number of samples, and x is input variable, and m is sample point sequence number, F_j() isJ sub-network forecast result, d () is actual result.

The forecast result of final multimode neutral net is the weighted sum of each sub-network forecast result.

$O\left(x\right)=\underset{k=1}{\overset{I}{\mathrm{\&Sigma;}}}(w_k\&CenterDot;F_k\left(x\right))---\left(7\right)$

In formula, x is input variable, and O () is model output, F_k() is k sub-network output, w_kBe kThe weight of individual sub-network, I is sub-network sum.

Further, in described step (3), adopt PCA principal component analytical method to realize input variablePrewhitening processing, can simplify the input variable of neural network model, and then improves 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, select operationVariable and easily survey variable are as the input of model.

Step 2: sample data is carried out to pretreatment, completed by data preprocessing module 7.

Step 3: to carrying out principal component analysis through pretreated data, complete by PCA principal component analysis module 8Become.

Step 4: module 9 is set up some initial neutral net moulds based on mode input, output integrating step (4)Type. Input data obtain as described in step 1, and output data are obtained by off-line chemical examination.

Step 5: module 10 integrating steps (5) according to sub-network prediction error by all sub neural network intelligenceCombining of energy;

Step 6: model modification module 11 is regularly input to off-line analysis data in training set, upgrades nerve netNetwork model, the soft measuring system 5 of intelligent Weighted optimal has been set up.

Step 7: it is real-time that the soft measuring system 5 of intelligent Weighted optimal establishing transmits based on DCS database 4Mode input variable data is carried out the soft measurement of intelligent Weighted optimal to the melt index of propylene polymerization production process 1.

Step 8: melt index flexible is measured the output that display instrument 6 shows the soft measuring system 5 of intelligent Weighted optimal,Complete the demonstration of the soft measurement of optimum to propylene polymerization production process melt index.

Claims (2)

1. an intelligent weighting propylene polymerization production process optimal soft measuring system, comprise propylene polymerization production process,For measuring the field intelligent instrument of easy survey variable, for measuring control station, the store data of performance variableDCS database, the soft measuring system of intelligent Weighted optimal and melt index flexible are measured display instrument, described sceneIntelligence instrument, control station are connected with propylene polymerization production process, described field intelligent instrument, control station and DCSDatabase connects, and described DCS database is connected with the input of the soft measuring system of intelligent Weighted optimal, described inThe output of the soft measuring system of intelligence Weighted optimal is measured display instrument with melt index flexible and is connected, and it is characterized in that:The soft measuring system of described intelligent Weighted optimal comprises:

(1), data preprocessing module, for carrying out pretreatment from the mode input variable of DCS database input,To input variable centralization, deduct the mean value of variable; Be normalized again, divided by variate-valueConstant interval;

(2), PCA principal component analysis module, for by input variable prewhitening process and variable decorrelation, pass throughInput variable is applied to a linear transformation and realize, principal component is obtained by C=xU, and wherein x is input variable,C is principal component scores matrix, and U is loading matrix; If initial data is reconstructed, by x=CU^TCalculate,The wherein transposition of subscript T representing matrix; In the time that the principal component number of choosing 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 RBF neutral net, minimizing by error functionA kind of nonlinear that becomes to be input to output, keeps topological invariance in mapping; Need to set up someSub neural network, the training objective of first sub-RBF network is forecast result and actual result gap J₁Minimum;

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

N is number of samples, and x is input variable, and l is sample point sequence number, F₁() is sub-network forecast result, d ()For actual result;

Since second sub-network, training objective becomes and makes the prediction error of network as far as possible little, net simultaneouslyThe forecast result of network and network forecast result before large as far as possible difference again, object 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 network; D () is actual knotReally; F () is the synthesis result of a front i-1 sub-network; λ is for regulating parameter, and N is number of samples;

The end condition of training is that the new sub-network obtaining is added after multimode neutral net, the forecast of network groupError no longer reduces;

Adopt a kind of continuous space ant group algorithm to train and optimization each RBF network, concrete stepsFor:

(a) algorithm initialization, constructs initial disaggregation according to RBF neural network structure to be optimizedS＝(s₁,s₂,…,s_n), the number that n is initial solution, s_nBe n initial solution, determine ant group's big or small M, establishPut the threshold value MaxGen of ant optimization algorithm iteration number of times and initialize the iterations sequence number of ant optimizationgen＝0；

(b) calculate the fitness value G that disaggregation S is corresponding_i(i=1,2 ..., n), the larger Xie Yuehao that represents of fitness value;Determining to separate according to following formula again concentrates each solution to be got the probability P as the initial solution of 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 n initial solution, k is iterations; Optimizing algorithm is carried out in initializationAnt numbering a=0;

(c) ant a chooses a solution in the S initial solution as optimizing, and selection rule is to make wheel disc according to PChoosing;

(d) ant a carries out optimizing on the basis of the initial solution of choosing, and finds better solution s_a'；

(e) if a < M, a=a+1, returns to step c; Otherwise continue execution step f downwards;

(f) if gen < MaxGen, gen=gen+1, uses the better solution that all ants obtain in steps dReplace the homographic solution in S, return to step b; Otherwise perform step g downwards;

(g) calculate the fitness value G that disaggregation S is corresponding_a(a=1,2 ..., n), choose the solution conduct of fitness value maximumThe optimal solution of algorithm, finishes algorithm and returns;

Each ant fixing number of times that can circulate when optimizing on the basis of its selected initial solution, if thisCirculation has obtained better solution, in circulation next time, can and keep the direction of search constant based on this solution; OtherwiseThe solution based on original but can adjust the direction of search still in next time circulation;

Along with the increase of whole ant optimization algebraically, the step-length of Ant Search can reduce intelligently, to be applicable to simultaneouslyThe convergence of whole ant optimization:

del_k＝Random·k^r(4)

In formula, del_kFor ant k is for the initial step length of iteration, k is iteration algebraically, Random at random toAmount, r is negative normal real number;

For medium-term and long-term those solutions that are not elected to be optimizing initial solution by ant of disaggregation S, can adopt in genetic algorithmVariation and Crossover Strategy are processed, thereby improve the global optimizing performance of algorithm;

(4), intelligent weighted optimization module, for each sub-network of step (3) is composed to weights; Foundation is everyThe prediction error of individual sub-network, error is less, and weights are larger;

$w_q=(1/e_q)/(\underset{j=1}{\overset{I}{\&Sigma;}}1/e_j),q=1,2,...,I.---\left(5\right)$

$e_j=\frac{1}{N}\underset{m=1}{\overset{N}{\&Sigma;}}|F_j\left(x_m\right)-d\left(x_m\right)|---\left(6\right)$

w_qBe the weights of q sub-network; e_qIt is the prediction error of q sub-network; I is total sub-networkNumber; J is sub-network sequence number; N is number of samples, and x is input variable, and m is sample point sequence number, F_j() isJ sub-network forecast result, d () is actual result;

The forecast result of final multimode neutral net is the weighted sum of each sub-network forecast result;

$O\left(x\right)=\underset{k=1}{\overset{I}{\&Sigma;}}(w_k\&CenterDot;F_k(x\left)\right)---\left(7\right)$

In formula, x is input variable, and O () is model output, F_k() is k sub-network output, w_kBe kThe weight of individual sub-network, I is sub-network sum;

The soft measuring system of described intelligent Weighted optimal also comprises: model modification module, for model online moreNewly, regularly off-line analysis data is input in training set, upgrades neural network model;

The described soft measuring system of intelligent Weighted optimal, trains sub-RBF neutral net, then by its set of weightsForm altogether neutral net group; Due to the selection standard of sub-network be prediction error little, with other subnetNetwork difference is large, so these values of forecasting are good, the comprehensive forecasting effect energy of different sub neural network againEnough there is better forecast precision and stability; In PCA principal component analysis module, PCA method realizes inputThe prewhitening processing of variable, can simplify the input variable of neural network model, and then improves the performance of model.

2. realize by the optimum soft measuring system of intelligent weighting polypropylene production process as claimed in claim 1 for one kindFlexible measurement method, is characterized in that: described flexible measurement method 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 easySurvey the input of variable as model, performance variable and easily survey variable are got temperature, pressure, liquid level, hydrogen gas phasePercentage, 3 strands of propylene feed flow velocitys and 2 strands of these variablees of catalyst charge flow velocity, obtained by DCS database;

(5.2) sample data is carried out to pretreatment, to input variable centralization, deduct the mean value of variable; Enter againRow normalized, divided by the constant interval of variate-value;

(5.3) PCA principal component analysis module, for input variable prewhitening is processed and variable decorrelation, by rightInput variable applies a linear transformation and realizes, and principal component is obtained by C=xU, and wherein x is input variable,C is principal component scores matrix, and U is loading matrix; If initial data is reconstructed, by x=CU^TCalculate,The wherein transposition of subscript T representing matrix; In the time that the principal component number of choosing 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 RBF nerveNetwork, completes a kind of nonlinear that is input to output by error minimize, in mapping, keepTopological invariance; The training objective of first sub-RBF network is forecast result and actual result gap J₁Minimum;

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

N is number of samples, and x is input variable, and l is sample point sequence number, F₁() is sub-network forecast result, d ()For actual result;

Since second sub-network, training objective becomes and makes the prediction error of network as far as possible little, net simultaneouslyThe forecast result of network and network forecast result before large as far as possible difference again, object 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 network; D () is actual knotReally; F () is the synthesis result of a front i-1 sub-network; λ is for regulating parameter, and N is number of samples;

The end condition of training is that the new sub-network obtaining is added after multimode neutral net, the forecast of network groupError no longer reduces;

Adopt a kind of continuous space ant group algorithm to train and optimization each RBF network, concrete stepsFor:

(a) algorithm initialization, constructs initial disaggregation according to RBF neural network structure to be optimizedS＝(s₁,s₂,…,s_n), the number that n is initial solution, s_nBe n initial solution, determine ant group's big or small M, establishPut the threshold value MaxGen of ant optimization algorithm iteration number of times and initialize the iterations sequence number of ant optimizationgen＝0；

(b) calculate the fitness value G that disaggregation S is corresponding_i(i=1,2 ..., n), the larger Xie Yuehao that represents of fitness value;Determining to separate according to following formula again concentrates each solution to be got the probability P as the initial solution of 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 n initial solution, k is iterations; Optimizing algorithm is carried out in initializationAnt numbering a=0;

(c) ant a chooses a solution in the S initial solution as optimizing, and selection rule is to make wheel disc according to PChoosing;

(d) ant a carries out optimizing on the basis of the initial solution of choosing, and finds better solution s_a'；

(e) if a < M, a=a+1, returns to step c; Otherwise continue execution step f downwards;

(f) if gen < MaxGen, gen=gen+1, uses the better solution that all ants obtain in steps dReplace the homographic solution in S, return to step b; Otherwise perform step g downwards;

(g) calculate the fitness value G that disaggregation S is corresponding_a(a=1,2 ..., n), choose the solution conduct of fitness value maximumThe optimal solution of algorithm, finishes algorithm and returns;

Each ant fixing number of times that can circulate when optimizing on the basis of its selected initial solution, if thisCirculation has obtained better solution, in circulation next time, can and keep the direction of search constant based on this solution; OtherwiseThe solution based on original but can adjust the direction of search still in next time circulation;

Along with the increase of whole ant optimization algebraically, the step-length of Ant Search can reduce intelligently, to be applicable to simultaneouslyThe convergence of whole ant optimization:

del_k＝Random·k^r(4)

In formula, del_kFor ant k is for the initial step length of iteration, k is iteration algebraically, Random at random toAmount, r is negative normal real number;

For medium-term and long-term those solutions that are not elected to be optimizing initial solution by ant of disaggregation S, can adopt in genetic algorithmVariation and Crossover Strategy are processed, thereby improve the global optimizing performance of algorithm;

(5.5) all sub neural networks of weighted array, compose weights for the each sub-network to step (5.4);According to being the prediction error of each sub-network, error is less, and weights are larger;

$w_q=(1/e_q)/(\underset{j=1}{\overset{I}{\&Sigma;}}1/e_j),q=1,2,...,I.---\left(5\right)$

$e_j=\frac{1}{N}\underset{m=1}{\overset{N}{\&Sigma;}}|F_j\left(x_m\right)-d\left(x_m\right)|---\left(6\right)$

w_qBe the weights of q sub-network; e_qIt is the prediction error of q sub-network; I is total sub-networkNumber; J is sub-network sequence number; N is number of samples, and x is input variable, and m is sample point sequence number, F_j() isJ sub-network forecast result, d () is actual result;

The forecast result of final multimode neutral net is the weighted sum of each sub-network forecast result;

$O\left(x\right)=\underset{k=1}{\overset{I}{\&Sigma;}}(w_k\&CenterDot;F_k(x\left)\right)---\left(7\right)$

In formula, x is input variable, and O () is model output, F_k() is k sub-network output, w_kBe k sonThe weight of network, I is sub-network sum;

Described flexible measurement method, also comprises: regularly off-line analysis data is input in training set, upgrades godThrough network model;

Described flexible measurement method, trains sub-RBF neutral net, then its weighted array is got up to form manyMould neutral net; Because the selection standard of sub-network is that prediction error is little, large with other sub-network difference,So these values of forecasting are good, the comprehensive forecasting effect of different sub neural network can have better againForecast precision and stability; In described step (5.3), adopt PCA principal component analytical method to realize defeatedEnter the prewhitening processing of variable, can simplify the input variable of neural network model, and then improve the property of modelEnergy.