CN103838958B - Vague intelligent optimal soft measuring instrument and method used in propylene polymerization production process - Google Patents

Abstract

The invention discloses a vague intelligent optimal soft measuring instrument used in the propylene polymerization production process. The vague intelligent optimal soft measuring instrument comprises the propylene polymerization production process, an intelligent site instrument, a control station, a DCS database storing data, a vague intelligent optimal soft measuring instrument body and a fusion index soft measuring value display instrument. The intelligent site instrument and the control station are connected with the propylene polymerization production process and the DCS database, and the optimal soft measuring instrument body is connected with the DCS database and the soft measuring value display instrument. The vague intelligent optimal soft measuring instrument body comprises a model updating module, a data preprocessing module, a PCA module, a neural network model module and a vague intelligent optimal module. The invention further provides a soft measuring method achieved through the soft measuring instrument. According to the vague intelligent optimal soft measuring instrument used in the propylene polymerization production process and the soft measuring method, on-line measurement is achieved, parameters are optimized on line, the speed of soft measuring is high, a model is automatically updated, and the antijamming capability and precision are high.

Description

Fuzzy intelligence optimum propylene polymerization production process optimal soft survey instrument and method

Technical field

The present invention relates to a kind of optimal soft survey instrument and method, specifically a kind of fuzzy intelligence optimum propylene polymerization production Process optimum soft measuring instrument and method.

Background technology

Polypropylene is a kind of thermoplastic resin prepared by propylene polymerization, the most important downstream product of propylene, the world third The 50% of alkene, the 65% of China's propylene is all used to polypropylene processed, is one of five big general-purpose plastics, close with our daily life Related.Polypropylene is fastest-rising general thermoplastic resin in the world, and total amount is only only second to polyethylene and polrvinyl chloride.For making China's polypropylene product has the market competitiveness, develops rigidity, toughness, the good crushing-resistant copolymerization product of sexual balance that flows, is randomly total to Dimerization product, BOPP and CPP film material, fiber, nonwoven cloth, and develop polypropylene in the application of automobile and field of household appliances, it is all Research topic important from now on.

Melt index is that polypropylene product determines one of important quality index of product grade, and it determines the difference of product Purposes, is an important step of control of product quality during polypropylene produces to the measurement of melt index, to producing and scientific research, all There are very important effect and directive significance.

However, the on-line analyses measurement of melt index is difficult to accomplish at present, on the one hand it is online melt index analyser Lack, to be on the other hand existing in-line analyzer by often blocking measure inaccurate even cannot be normally using being led to Using upper difficulty.Therefore, in current commercial production MI measurement, mainly obtained by manual sampling, offline assay , and typically every 2-4 hour can only be analyzed once, and time lag is big, and the quality control producing to propylene polymerization brings tired Difficulty, becomes a bottleneck problem being badly in need of solving in production.The online soft sensor instrument of polypropylene melt index and technique study, Thus becoming a forward position and the focus of academia and industrial quarters.

Content of the invention

In order to overcome that the certainty of measurement of existing propylene polymerization production process at present is not high, easily affected by anthropic factor Deficiency, it is an object of the invention to provide a kind of on-line measurement, on-line parameter optimization, hard measurement speed is fast, model automatically updates, Strong antijamming capability, the fuzzy intelligence optimum propylene polymerization production process melt index optimal soft survey instrument of high precision and side Method.

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

A kind of fuzzy intelligence optimum propylene polymerization production process optimal soft survey instrument, including propylene polymerization production process, For measure the easy field intelligent instrument surveying variable, the control station for measuring performance variable, the DCS database depositing data, Fuzzy intelligence optimum hard measurement instrument and melt index hard measurement display instrument, described field intelligent instrument, control station and propylene gather Close production process to connect, described field intelligent instrument, control station be connected with DCS database, described DCS database and fuzzy intelligence The input of optimum hard measurement instrument connects, and the outfan of described fuzzy intelligence optimum hard measurement instrument is shown with melt index hard measurement Instrument connect it is characterised in that：Described fuzzy intelligence optimum hard measurement instrument includes：

（1）, data preprocessing module, the mode input variable for inputting from DCS database carries out pretreatment, to defeated Enter variable centralization, that is, deduct the meansigma methodss of variable；It is normalized, that is, divided by the constant interval of variate-value again；

（2）, PCA principal component analysiss module, for by input variable pre -whitening processing and variable decorrelation, by input Variable applies a linear transformation and realizes, i.e. main constituent is obtained by C=MU, and wherein M is input variable, and C is principal component scores square Battle array, U is loading matrix.If being reconstructed to initial data, can be by M=CU^TCalculate, the wherein transposition of subscript T representing matrix.When When the main constituent number chosen is less than the variable number of input variable, M=CU^T+ E, wherein E are residual matrix；

（3）, neural network model module, using one four layers of fuzzy neural network, by error function minimize come Complete to be input to a kind of nonlinear of output, in mapping, keep topological invariance；

（4）, fuzzy intelligence optimization module, for being optimized to neutral net using fuzzy intelligence optimization module, including：

（4.1）Algorithm initialization, goes out initial particle colony X according to structure of fuzzy neural network parametric configuration to be optimized =(x¹,x²,···,x^N), initial translational speed V=(v¹,v²,···,v^N), initial each particle successive dynasties optimal value OP= (p¹,p²,···,p^N) and global optimum p^g；

（4.2）Particle swarm optimization algorithm is executed by following formula, allows population to restrain：

In formula, x is the position vector of particle, and i is the sequence number of particle, and k is algorithm iteration algebraically, and v is particle rapidity, and p represents Initial each particle successive dynasties optimal value OP=(p¹,p²,···,p^N) and global optimum p^gOptimum value set.For kth time The speed of i-th particle in iterative algebra；The position vector of i-th particle in kth time iterative algebra；For kth time repeatedly The successive dynasties optimal location of i-th particle, p in counting from generation to generationⁱFor the successive dynasties optimal solution of i-th particle, w is speed weight coefficient, c₁、 c₂It is respectively the attraction coefficient of particle successive dynasties optimal solution and group optimal solution, r₁、r₂It is respectively random number.

（4.3）Here it is to be estimated based on a kind of Evolving State that intelligent granule group's algorithm is embodied in speed weighted value w Meter；The Evolving State of population includes probe phase, development stage, polymerizer and jumps out the phase；Quantificational expression is come by f；

d_iIt is the average distance of particle i other particles in population；d_gBe in population optimal solution in population other The average distance of particle；d_maxAnd d_minIt is { d respectively_iIn maximum, minima；d_iCalculate according to the following formula：

x_iFor the position of i-th particle, x_jFor the position of j-th particle, ‖ ‖ is norm expression formula.P is grain in population The number of son；

By the calculating formula that f value to obtain particle group velocity weight it is：

As preferably a kind of scheme, described fuzzy intelligence optimal soft measurement model also includes：Model modification module, is used for The online updating of model, periodically offline analysis data is input in training set, updates neural network model.

As preferred yet another approach：Transmission letter as the basic fuzzy neural network input layer of modeling to the second layer Number uses Gaussian function：

U in formula⁽¹⁾、u⁽²⁾It is respectively input and the output vector of the second layer, vector centered on m, σ is width parameter, and q, j divide Wei not dimension and node ID.

The nonlinear fitting ability of this fuzzy neural network can be strengthened using Gauss nonlinear function.

As preferred yet another approach：In PCA principal component analysiss module, PCA method realizes the pre- white of input variable Change is processed, and can simplify the input variable of neural network model, and then improve the performance of model.

The flexible measurement method that a kind of fuzzy intelligence optimum polypropylene production process optimal soft survey instrument is realized, described soft survey It is as follows that amount method implements step：

（1）To propylene polymerization production process object, according to industrial analyses and operation analysis, selection operation variable and easy survey become As the input of model, performance variable and easy variable of surveying are obtained amount by DCS database；

（2）Pretreatment is carried out to sample data, to input variable centralization, that is, deducts the meansigma methodss of variable；Returned again One change is processed, that is, divided by the constant interval of variate-value；

（3）PCA principal component analysiss module, for by input variable pre -whitening processing and variable decorrelation, by input Variable applies a linear transformation and realizes, i.e. main constituent is obtained by C=MU, and wherein M is input variable, and C is principal component scores square Battle array, U is loading matrix.If being reconstructed to initial data, can be by M=CU^TCalculate, the wherein transposition of subscript T representing matrix.When When the main constituent number chosen is less than the variable number of input variable, M=CU^T+ E, wherein E are residual matrix；

（4）Initial neural network model is set up based on mode input, output data, using one four layers of fuzznet Network, completes by error minimize to be input to a kind of nonlinear of output, keeps topological invariance in mapping；

（5）Structure optimization is carried out to fuzzy neural network using fuzzy intelligence optimization module, including：

（5.1）Algorithm initialization, goes out initial particle colony X according to structure of fuzzy neural network parametric configuration to be optimized =(x¹,x²,···,x^N), initial translational speed V=(v¹,v²,···,v^N), initial each particle successive dynasties optimal value OP= (p¹,p²,···,p^N) and global optimum p^g；

（5.2）Particle swarm optimization algorithm is executed by following formula, allows population to restrain：

In formula, x is the position vector of particle, and i is the sequence number of particle, and k is algorithm iteration algebraically, and v is particle rapidity, and p represents Initial each particle successive dynasties optimal value OP=(p¹,p²,···,p^N) and global optimum p^gOptimum value set.For kth time The speed of i-th particle in iterative algebra；The position vector of i-th particle in kth time iterative algebra；For kth time repeatedly The successive dynasties optimal location of i-th particle, p in counting from generation to generationⁱFor the successive dynasties optimal solution of i-th particle, w is speed weight coefficient, c₁、 c₂It is respectively the attraction coefficient of particle successive dynasties optimal solution and group optimal solution, r₁、r₂It is respectively random number.

（5.3）Here it is to be estimated based on a kind of Evolving State that intelligent granule group's algorithm is embodied in speed weighted value w Meter；The Evolving State of population includes probe phase, development stage, polymerizer and jumps out the phase；Quantificational expression is come by f；

d_iIt is the average distance of particle i other particles in population；d_gBe in population optimal solution in population other The average distance of particle；d_maxAnd d_minIt is { d respectively_iIn maximum, minima；d_iCalculate according to the following formula：

x_iFor the position of i-th particle, x_jFor the position of j-th particle, ‖ ‖ is norm expression formula.P is grain in population The number of son；

By the calculating formula that f value to obtain particle group velocity weight it is：

As preferably a kind of scheme, described flexible measurement method also includes：Periodically offline analysis data is input to training Concentrate, update neural network model.

As preferred yet another approach：Transmission letter as the basic fuzzy neural network input layer of modeling to the second layer Number uses Gaussian function：

U in formula⁽¹⁾、u⁽²⁾It is respectively input and the output vector of the second layer, vector centered on m, σ is width parameter, and q, j divide Wei not dimension and node ID.

The nonlinear fitting ability of this fuzzy neural network can be strengthened using Gauss nonlinear function.

Further, in described step（3）Middle realized at the prewhitening of input variable using PCA principal component analytical method Reason, can simplify the input variable of neural network model, and then improve the performance of model.

The technology design of the present invention is：The important quality index melt index of propylene polymerization production process is carried out online Excellent hard measurement, overcomes that existing polypropylene melt index measuring instrumentss certainty of measurement is not high, is easily affected not by anthropic factor Foot, introduces the optimization module based on intelligent granule colony optimization algorithm and carries out Automatic Optimal to fuzzy neural network parameter and structure, It is not required to very important person to adjust neutral net for experience or repeatedly test, thus obtaining that there is optimum melt index hard measurement function Online optimal soft survey instrument.

Beneficial effects of the present invention are mainly manifested in：1st, on-line measurement；2nd, on-line parameter Automatic Optimal；3rd, hard measurement speed Hurry up；4th, model automatically updates；5th, strong antijamming capability；6th, high precision.

Brief description

Fig. 1 is the basic structure signal of fuzzy intelligence optimum propylene polymerization production process optimal soft survey instrument and method Figure；

Fig. 2 is fuzzy intelligence optimum hard measurement instrument structural representation；

Fig. 3 is propylene polymerization production process Hypol technique productions flow chart.

Specific embodiment

Below in conjunction with the accompanying drawings the present invention is described further.The embodiment of the present invention is used for illustrating the present invention, and not It is to limit the invention, in the protection domain of spirit and claims of the present invention, what the present invention was made any repaiies Change and change, both fall within protection scope of the present invention.

Embodiment 1

1. with reference to Fig. 1, Fig. 2 and Fig. 3, a kind of fuzzy intelligence optimum propylene polymerization production process optimal soft survey instrument, bag Include propylene polymerization production process 1, field intelligent instrument 2, the control station for measuring performance variable for measuring easy survey variable 3rd, DCS database 4, fuzzy intelligence optimal soft measurement model 5 and the melt index flexible measured value display instrument 6 of data, institute are deposited State field intelligent instrument 2, control station 3 is connected with propylene polymerization production process 1, described field intelligent instrument 2, control station 3 and DCS Data base 4 connects, and described DCS data base 4 is connected with the input of fuzzy intelligence optimum hard measurement instrument 5, and described fuzzy intelligence is The outfan of excellent hard measurement instrument 5 be connected with melt index flexible measured value display instrument 6 it is characterised in that：Described fuzzy intelligence is optimum Hard measurement instrument includes：

（1）, data preprocessing module, the mode input variable for inputting from DCS database carries out pretreatment, to defeated Enter variable centralization, that is, deduct the meansigma methodss of variable；It is normalized, that is, divided by the constant interval of variate-value again；

（2）, PCA principal component analysiss module, for by input variable pre -whitening processing and variable decorrelation, by input Variable applies a linear transformation and realizes, i.e. main constituent is obtained by C=MU, and wherein M is input variable, and C is principal component scores square Battle array, U is loading matrix.If being reconstructed to initial data, can be by M=CU^TCalculate, the wherein transposition of subscript T representing matrix.When When the main constituent number chosen is less than the variable number of input variable, M=CU^T+ E, wherein E are residual matrix；

（3）, neural network model module, using one four layers of fuzzy neural network, by error function minimize come Complete to be input to a kind of nonlinear of output, in mapping, keep topological invariance；

（4）, fuzzy intelligence optimization module, for being optimized to neutral net using fuzzy intelligence optimization module, including：

（4.1）Algorithm initialization, goes out initial particle colony X according to structure of fuzzy neural network parametric configuration to be optimized =(x¹,x²,···,x^N), initial translational speed V=(v¹,v²,···,v^N), initial each particle successive dynasties optimal value OP= (p¹,p²,···,p^N) and global optimum p^g；

（4.2）Particle swarm optimization algorithm is executed by following formula, allows population to restrain：

In formula, x is the position vector of particle, and i is the sequence number of particle, and k is algorithm iteration algebraically, and v is particle rapidity, and p represents Initial each particle successive dynasties optimal value OP=(p¹,p²,···,p^N) and global optimum p^gOptimum value set.For kth time The speed of i-th particle in iterative algebra；The position vector of i-th particle in kth time iterative algebra；For kth time repeatedly The successive dynasties optimal location of i-th particle, p in counting from generation to generationⁱFor the successive dynasties optimal solution of i-th particle, w is speed weight coefficient, c₁、 c₂It is respectively the attraction coefficient of particle successive dynasties optimal solution and group optimal solution, r₁、r₂It is respectively random number.

（4.3）Here it is to be estimated based on a kind of Evolving State that intelligent granule group's algorithm is embodied in speed weighted value w Meter；The Evolving State of population includes probe phase, development stage, polymerizer and jumps out the phase；Quantificational expression is come by f；

d_iIt is the average distance of particle i other particles in population；d_gBe in population optimal solution in population its The average distance of his particle；d_maxAnd d_minIt is { d respectively_iIn maximum, minima；d_iCalculate according to the following formula：

x_iFor the position of i-th particle, x_jFor the position of j-th particle, ‖ ‖ is norm expression formula.P is grain in population The number of son；

By the calculating formula that f value to obtain particle group velocity weight it is：

Described fuzzy intelligence optimal soft measurement model also includes：Model modification module, for the online updating of model, will determine Phase, offline analysis data was input in training set, updated neural network model.

Transmission function as fuzzy neural network input layer to the second layer on modeling basis uses Gaussian function：

U in formula⁽¹⁾、u⁽²⁾It is respectively input and the output vector of the second layer, vector centered on m, σ is width parameter, and q, j divide Wei not dimension and node ID.

The nonlinear fitting ability of this fuzzy neural network can be strengthened using Gauss nonlinear function.

In PCA principal component analysiss module, PCA method realizes the pre -whitening processing of input variable, can simplify nerve net The input variable of network model, and then improve the performance of model.

2. propylene polymerization production process flow chart as shown in figure 3, analyzed according to reaction mechanism and flow process it is contemplated that In polypropylene production process, melt index is produced with each factor of impact, takes conventional nine operations in actual production process to become Amount and easy variable of surveying, as mode input variable, have：Three bursts of propylene feed flow rates, major catalyst flow rate, cocatalyst flow rate, kettle Interior temperature, pressure, liquid level, hydrogen volume concentration in kettle.

Mode input variable needed for table 1 fuzzy intelligence optimum hard measurement instrument

Table 1 lists 9 mode input variables as fuzzy intelligence optimum hard measurement instrument 5 input, respectively kettle interior temperature Degree（T）, pressure in kettle（p）, liquid level in kettle（L）, hydrogen volume concentration in kettle（X_v）, 3 bursts of propylene feed flow rates（First strand third Alkene feed flow rates f1, second burst of propylene feed flow rate f2, the 3rd burst of propylene feed flow rate f3）, 2 bursts of catalyst charge flow rates（Main Catalyst flow rate f4, cocatalyst flow rate f5）.Polyreaction in reactor is to participate in reaction after reaction mass mixes repeatedly , therefore mode input variable is related to the meansigma methodss using front some moment for the process variable of material.In this example, data is using front The meansigma methodss of one hour.The offline laboratory values of melt index are as the output variable of fuzzy intelligence optimum hard measurement instrument 5.By artificial Sampling, offline assay obtain, and analysis collection in every 4 hours is once.

Field intelligent instrument 2 and control station 3 are connected with propylene polymerization production process 1, are connected with DCS database 4；Optimum soft Measuring instrument 5 is connected with DCS database 4 and hard measurement value display instrument 6.Field intelligent instrument 2 measures propylene polymerization and produces object Easily survey variable, easy variable of surveying is transferred to DCS database 4；Control station 3 controls propylene polymerization to produce the performance variable of object, will Performance variable is transferred to DCS database 4.In DCS database 4, the variable data of record is as fuzzy intelligence optimum hard measurement instrument 5 Input, hard measurement value display instrument 6 be used for show fuzzy intelligence optimum hard measurement instrument 5 output, i.e. hard measurement value.

Fuzzy intelligence optimum hard measurement instrument 5, including：

（1）Data preprocessing module 7, for pretreatment is carried out to mode input, i.e. centralization and normalization.Input is become Amount centralization, it is simply that deducting the meansigma methodss of variable, makes the variable that variable is zero-mean, thus simplifying algorithm；Input variable is returned One changes the constant interval it is simply that divided by input variable value, is that the value of variable is fallen within -0.5～0.5, simplifies further.

（2）PCA principal component analysiss module 8, for being variable decorrelation to input variable pre -whitening processing, to input variable Apply a linear transformation so that conversion after each component of variable between orthogonal, simultaneously its covariance matrix be unit battle array, I.e. main constituent is obtained by C=MU, and wherein M is input variable, and C is principal component scores matrix, and U is loading matrix.If to initial data It is reconstructed, can be by M=CU^TCalculate, the wherein transposition of subscript T representing matrix.When the main constituent number chosen is less than input variable Variable number when, M=CU^T+ E, wherein E are residual matrix.

（3）Neural network model module 9, using four layers of fuzzy neural network, minimizes by error function Become to be input to a kind of nonlinear of output, in mapping, keep topological invariance.

（4）Fuzzy intelligence optimization module 10：Using the optimization module based on intelligent granule colony optimization algorithm to fuzzy neural Network carries out structure optimization, by the powerful global optimizing ability of intelligent granule colony optimization algorithm come each layer of optimization neural network it Between structural parameters, and neural network learning is carried out with this, thus setting up intelligent granule group's algorithm of propylene polymerization melt index The optimum hard measurement instrument of Optimization of Fuzzy neutral net.Implement step as follows：

（4.1）Algorithm initialization, goes out initial particle colony X according to structure of fuzzy neural network parametric configuration to be optimized =(x¹,x²,···,x^N), initial translational speed V=(v¹,v²,···,v^N), initial each particle successive dynasties optimal value OP= (p¹,p²,···,p^N) and global optimum p^g；

（4.2）Particle swarm optimization algorithm is executed by following formula, allows population to restrain：

In formula, x is the position vector of particle, and i is the sequence number of particle, and k is algorithm iteration algebraically, and v is particle rapidity, and p represents Initial each particle successive dynasties optimal value OP=(p¹,p²,···,p^N) and global optimum p^gOptimum value set.For kth time The speed of i-th particle in iterative algebra；The position vector of i-th particle in kth time iterative algebra；For kth time iteration The successive dynasties optimal location of i-th particle, p in algebraicallyⁱFor the successive dynasties optimal solution of i-th particle, w is speed weight coefficient, c₁、c₂ It is respectively the attraction coefficient of particle successive dynasties optimal solution and group optimal solution, r₁、r₂It is respectively random number.

（4.3）Here it is to be estimated based on a kind of Evolving State that intelligent granule group's algorithm is embodied in speed weighted value w Meter；The Evolving State of population includes probe phase, development stage, polymerizer and jumps out the phase；Quantificational expression is come by f；

d_iIt is the average distance of particle i other particles in population；d_gBe in population optimal solution in population other The average distance of particle；d_maxAnd d_minIt is { d respectively_iIn maximum, minima；d_iCalculate according to the following formula：

x_iFor the position of i-th particle, x_jFor the position of j-th particle, ‖ ‖ is norm expression formula.P is grain in population The number of son；

By the calculating formula that f value to obtain particle group velocity weight it is：

Transmission function as fuzzy neural network input layer to the second layer on modeling basis uses Gaussian function：

U in formula⁽¹⁾、u⁽²⁾It is respectively input and the output vector of the second layer, vector centered on m, σ is width parameter, and q, j divide Wei not dimension and node ID.

The nonlinear fitting ability of this fuzzy neural network can be strengthened using Gauss nonlinear function.

（5）Offline analysis data, for the online updating of model, is periodically input to training set by model modification module 11 In, update neural network model.

Embodiment 2

1., with reference to Fig. 1, Fig. 2 and Fig. 3, a kind of fuzzy intelligence optimum propylene polymerization production process optimal soft measuring method includes Following steps：

（1）To propylene polymerization production process object, according to industrial analyses and operation analysis, selection operation variable and easy survey become As the input of model, performance variable and easy variable of surveying are obtained amount by DCS database；

（2）Pretreatment is carried out to sample data, to input variable centralization, that is, deducts the meansigma methodss of variable；Returned again One change is processed, that is, divided by the constant interval of variate-value；

（3）PCA principal component analysiss module, for by input variable pre -whitening processing and variable decorrelation, by input Variable applies a linear transformation and realizes, i.e. main constituent is obtained by C=MU, and wherein M is input variable, and C is principal component scores square Battle array, U is loading matrix.If being reconstructed to initial data, can be by M=CU^TCalculate, the wherein transposition of subscript T representing matrix.When When the main constituent number chosen is less than the variable number of input variable, M=CU^T+ E, wherein E are residual matrix；

（4）Initial neural network model is set up based on mode input, output data, using one four layers of fuzznet Network, completes by error minimize to be input to a kind of nonlinear of output, keeps topological invariance in mapping；

（5）Structure optimization is carried out to fuzzy neural network using fuzzy intelligence optimization module, including：

（5.1）Algorithm initialization, goes out initial particle colony X according to structure of fuzzy neural network parametric configuration to be optimized =(x¹,x²,···,x^N), initial translational speed V=(v¹,v²,···,v^N), initial each particle successive dynasties optimal value OP= (p¹,p²,···,p^N) and global optimum p^g；

（5.2）Particle swarm optimization algorithm is executed by following formula, allows population to restrain：

In formula, x is the position vector of particle, and i is the sequence number of particle, and k is algorithm iteration algebraically, and v is particle rapidity, and p represents Initial each particle successive dynasties optimal value OP=(p¹,p²,···,p^N) and global optimum p^gOptimum value set.For kth time The speed of i-th particle in iterative algebra；The position vector of i-th particle in kth time iterative algebra；For kth time iteration The successive dynasties optimal location of i-th particle, p in algebraicallyⁱFor the successive dynasties optimal solution of i-th particle, w is speed weight coefficient, c₁、c₂ It is respectively the attraction coefficient of particle successive dynasties optimal solution and group optimal solution, r₁、r₂It is respectively random number.

（5.3）Here it is to be estimated based on a kind of Evolving State that intelligent granule group's algorithm is embodied in speed weighted value w Meter；The Evolving State of population includes probe phase, development stage, polymerizer and jumps out the phase；Quantificational expression is come by f；

d_iIt is the average distance of particle i other particles in population；d_gBe in population optimal solution in population other The average distance of particle；d_maxAnd d_minIt is { d respectively_iIn maximum, minima；d_iCalculate according to the following formula：

x_iFor the position of i-th particle, x_jFor the position of j-th particle, ‖ ‖ is norm expression formula.P is grain in population The number of son；

By the calculating formula that f value to obtain particle group velocity weight it is：

Described flexible measurement method also includes：Periodically offline analysis data is input in training set, updates neutral net mould Type.

Transmission function as fuzzy neural network input layer to the second layer on modeling basis uses Gaussian function：

U in formula⁽¹⁾、u⁽²⁾It is respectively input and the output vector of the second layer, vector centered on m, σ is width parameter, and q, j divide Wei not dimension and node ID.

The nonlinear fitting ability of this fuzzy neural network can be strengthened using Gauss nonlinear function.

Further, in described step（3）Middle realized at the prewhitening of input variable using PCA principal component analytical method Reason, can simplify the input variable of neural network model, and then improve the performance of model.

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

Step 1：To propylene polymerization production process object 1, according to industrial analyses and operation analysis, selection operation variable and easily Survey variable as the input of model.

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

Step 3：Principal component analysiss are carried out to the data through pretreatment, is completed by PCA principal component analysiss module 8.

Step 4：Initial neural network model 9 is set up based on mode input, output.Input data obtains as described in step 1, Output data is obtained by offline chemical examination.

Step 5：The input of initial neutral net 9 and hiding Rotating fields are optimized by fuzzy intelligence optimization module 10.

Step 6：Model modification module 11 periodically offline analysis data is input in training set, updates neutral net mould Type, the soft-sensing model of fuzzy intelligence optimum hard measurement instrument 5 is set up and is completed.

Step 7：The fuzzy intelligence optimum hard measurement instrument 5 establishing is become based on the real-time model input that DCS database 4 transmits Amount data carries out fuzzy intelligence optimum hard measurement to the melt index of propylene polymerization production process 1.

Step 8：Melt index hard measurement display instrument 6 shows the output of fuzzy intelligence optimum hard measurement instrument 5, completes to propylene The display of the optimum hard measurement of polymerization production process melt index.

Claims (2)

1. a kind of fuzzy intelligence optimum propylene polymerization production process optimal soft survey instrument, including propylene polymerization production process, use Easily survey the field intelligent instrument of variable, be used for the control station measuring performance variable, the DCS database depositing data, mould in measurement Paste intelligence optimum hard measurement instrument and melt index hard measurement display instrument, described field intelligent instrument, control station and propylene polymerization Production process connects, and described field intelligent instrument, control station are connected with DCS database, and described DCS database is with fuzzy intelligence The input of excellent soft-sensing model connects, and the outfan of described fuzzy intelligence optimal soft measurement model is shown with melt index hard measurement Show instrument connect it is characterised in that：Described fuzzy intelligence optimum hard measurement instrument includes：

(1), data preprocessing module, the mode input variable for inputting from DCS database carries out pretreatment, and input is become Amount centralization, that is, deduct the meansigma methodss of variable；It is normalized, that is, divided by the constant interval of variate-value again；

(2), PCA principal component analysiss module, for by input variable pre -whitening processing and variable decorrelation, by input variable Apply a linear transformation to realize, i.e. main constituent is obtained by C=MU, wherein M is input variable, C is principal component scores matrix, U For loading matrix；If being reconstructed to initial data, can be by M=CU^TCalculate, the wherein transposition of subscript T representing matrix；Work as selection Main constituent number be less than input variable variable number when, M=CU^T+ E, wherein E are residual matrix；

(3), neural network model module, using one four layers of fuzzy neural network, is minimized by error function and to complete It is input to a kind of nonlinear of output, in mapping, keep topological invariance；

(4), fuzzy intelligence optimization module, for being optimized to neutral net using fuzzy intelligence optimization module, including：

(4.1) algorithm initialization, goes out initial particle colony X=according to structure of fuzzy neural network parametric configuration to be optimized (x¹,x²,…,x^N), initial translational speed V=(v¹,v²,…,v^N), initial each particle successive dynasties optimal value OP=(p¹,p²,…, p^N) and global optimum p^g；

(4.2) particle swarm optimization algorithm is executed by following formula, allow population to restrain：

$x_{k+1}^i=x_k^i+v_{k+1}^i---\left(1\right)$

$v_{k+1}^i={\mathrm{wv}}_k^i+c_1{r}_1(p_k^i-x_k^i)+c_2{r}_2(p_k^g-x_k^i)---\left(2\right)$

In formula, x is the position vector of particle, and i is the sequence number of particle, and k is algorithm iteration algebraically, and v is particle rapidity, and p represents initial Each particle successive dynasties optimal value OP=(p¹,p²,…,p^N) and global optimum p^gOptimum value set；v_k ⁱFor in kth time iteration The speed of i-th particle；x_k ⁱPosition vector for i-th particle in kth time iteration；p_k ⁱFor i-th particle in kth time iteration Successive dynasties optimal location, pⁱFor the successive dynasties optimal solution of i-th particle, w is speed weight coefficient, c₁、c₂It is respectively the particle successive dynasties Excellent solution and the attraction coefficient of group optimal solution, r₁、r₂It is respectively random number；

(4.3) described particle swarm optimization algorithm is embodied in speed weighted value w and is estimated based on a kind of Evolving State；Grain The Evolving State of subgroup includes probe phase, development stage, polymerization phase and the phase of jumping out；Quantificational expression is come by f；

$f=\frac{d_g-d_{\min }}{d_{max}-d_{\min }}\∈\[0,1\]---\left(3\right)$

Wherein, f represents the Evolving State of population, d_iIt is the average distance of particle i other particles in population；d_gIt is particle The average distance of optimal solution other particles in population in group；d_maxAnd d_minIt is { d respectively_iIn maximum, minima；d_iPress Calculate according to following formula：

$d_i=\frac{1}{P-1}\underset{j=1,j\≠i}{\overset{P}{\Σ}}\left|\right|x_i-x_j|{|}_2---\left(4\right)$

x_iFor the position of i-th particle, x_jFor the position of j-th particle, ‖ ‖ is norm expression formula；P is particle in population Number；

By the calculating formula that f value to obtain particle group velocity weight it is：

$w=\frac{1}{1+1.5e^{-2.6f}}\∈\[0.4,0.9\]---\left(5\right)$

Described fuzzy intelligence optimum hard measurement instrument also includes：Model modification module, for the online updating of model, periodically will be offline Analysis data is input in training set, updates neural network model；

The transmission function of described fuzzy neural network input layer to the second layer uses Gaussian function：

$\begin{array}{ccc}{u}_{qj}^{\left(2\right)}=\exp (-\frac{{\[u_q^{\left(1\right)}-m_{qj}\]}^2}{{\mathrm{\σ}}_{qj}^2}),& q=1,2,...,n,& j=1,2,...,N\end{array}---\left(6\right)$

U in formula⁽¹⁾、u⁽²⁾It is respectively input and the output vector of the second layer, vector centered on m, σ is width parameter, and q, j are respectively Dimension and node ID；

The nonlinear fitting ability of this fuzzy neural network can be strengthened using Gauss nonlinear function；

In described PCA principal component analysiss module, PCA method realizes the pre -whitening processing of input variable, can simplify neutral net The input variable of model, and then improve the performance of model.

2. what a kind of fuzzy intelligence as claimed in claim 1 optimum polypropylene production process optimal soft survey instrument was realized is soft Measuring method it is characterised in that：It is as follows that described flexible measurement method implements step：

(1) to propylene polymerization production process object, according to industrial analyses and operation analysis, selection operation variable and easy variable of surveying are made For the input of model, performance variable and easy survey change and measure temperature, pressure, liquid level, hydrogen gas phase percent, 3 bursts of propylene feed stream Speed and 2 strands of these variables of catalyst charge flow velocity, are obtained by DCS database；

(2) pretreatment is carried out to sample data, to input variable centralization, that is, deduct the meansigma methodss of variable；It is normalized again Process, that is, divided by the constant interval of variate-value；

(3) PCA principal component analysiss module, for by input variable pre -whitening processing and variable decorrelation, by input variable Apply a linear transformation to realize, i.e. main constituent is obtained by C=MU, wherein M is input variable, C is principal component scores matrix, U For loading matrix；If being reconstructed to initial data, can be by M=CU^TCalculate, the wherein transposition of subscript T representing matrix；Work as selection Main constituent number be less than input variable variable number when, M=CU^T+ E, wherein E are residual matrix；

(4) it is based on mode input, output data sets up initial neural network model, using one four layers of fuzzy neural network, Complete by error minimize to be input to a kind of nonlinear of output, in mapping, keep topological invariance；

(5) structure optimization is carried out to fuzzy neural network using fuzzy intelligence optimization module, including：

(5.1) algorithm initialization, goes out initial particle colony X=according to structure of fuzzy neural network parametric configuration to be optimized (x¹,x²,…,x^N), initial translational speed V=(v¹,v²,…,v^N), initial each particle successive dynasties optimal value OP=(p¹,p²,…, p^N) and global optimum p^g；

(5.2) particle swarm optimization algorithm is executed by following formula, allow population to restrain：

$x_{k+1}^i=x_k^i+v_{k+1}^i---\left(1\right)$

$v_{k+1}^i={\mathrm{wv}}_k^i+c_1{r}_1(r_k-x_k^i)+c_2{r}_2(p_k^g-x_k^i)---\left(2\right)$

In formula, x is the position vector of particle, and i is the sequence number of particle, and k is algorithm iteration algebraically, and v is particle rapidity, and p represents initial Each particle successive dynasties optimal value OP=(p¹,p²,…,p^N) and global optimum p^gOptimum value set；v_k ⁱFor in kth time iteration The speed of i-th particle；x_k ⁱPosition vector for i-th particle in kth time iteration；p_k ⁱFor i-th particle in kth time iteration Successive dynasties optimal location, pⁱFor the successive dynasties optimal solution of i-th particle, w is speed weight coefficient, c₁、c₂It is respectively the particle successive dynasties Excellent solution and the attraction coefficient of group optimal solution, r₁、r₂It is respectively random number；

(5.3) described particle swarm optimization algorithm is embodied in speed weighted value w and is estimated based on a kind of Evolving State；Grain The Evolving State of subgroup includes probe phase, development stage, polymerization phase and the phase of jumping out；Quantificational expression is come by f；

$f=\frac{d_g-d_{\min }}{d_{\max }-d_{\min }}\∈\[0,1\]---\left(3\right)$

Wherein, f represents the Evolving State of population, d_iIt is the average distance of particle i other particles in population；d_gIt is particle The average distance of optimal solution other particles in population in group；d_maxAnd d_minIt is { d respectively_iIn maximum, minima；d_iPress Calculate according to following formula：

$d_i=\frac{1}{P-1}\underset{j=1,j\≠i}{\overset{P}{\Σ}}\left|\right|x_i-x_j|{|}_2---\left(4\right)$

x_iFor the position of i-th particle, x_jFor the position of j-th particle, ‖ ‖ is norm expression formula；P is particle in population Number；

By the calculating formula that f value to obtain particle group velocity weight it is：

$w=\frac{1}{1+1.5e^{-2.6f}}\∈\[0.4,0.9\]---\left(5\right)$

Described flexible measurement method also includes：Periodically offline analysis data is input in training set, updates neural network model；

Described flexible measurement method, the transmission function as fuzzy neural network input layer to the second layer on modeling basis uses It is Gaussian function：

$\begin{array}{ccc}{u}_{qj}^{\left(2\right)}=\exp (-\frac{{\[u_q^{\left(1\right)}-m_{qj}\]}^2}{{\mathrm{\σ}}_{qj}^2}),& q=1,2,...,n,& j=1,2,...,N\end{array}---\left(6\right)$

U in formula⁽¹⁾、u⁽²⁾It is respectively input and the output vector of the second layer, vector centered on m, σ is width parameter, and q, j are respectively Dimension and node ID；

The nonlinear fitting ability of this fuzzy neural network can be strengthened using Gauss nonlinear function；

Realize the pre -whitening processing of input variable using PCA principal component analytical method in described step (3), god can be simplified Input variable through network model, and then improve the performance of model.