阅读说明：本技术 一种丙烯聚合质量在线检测系统 (Propylene polymerization quality on-line measuring system ) 是由 陈旭 张红星 慕雪梅 马文辉 张长军 刘小燕 马艳萍 竺栋荣 许云波 吴冬 谢昕 于 2020-06-05 设计创作，主要内容包括：本发明涉及一种丙烯聚合质量在线检测系统,包括：混沌重构模块,将丙烯聚合过程的操作参数依据其混沌特性进行重构得到输入变量；Gabor多尺度分析模块,将所述输入变量以频率为基准分析其多尺度特性,通过Gabor核函数对变量进行多尺度重构,提取输入变量在不同频率下各尺度各方向的局部纹理特征信息,得到输入特征信号；极端随机树测量模型模块,以聚丙烯产品样本集和先验知识作为依据,将所述输入特征信号转化后输出对应的聚丙烯产品的软测量值；以及自适应粒子群模块,用以采用自适应粒子群算法对检测系统的分叉阈值参数进行优化。(The invention relates to an on-line detection system for propylene polymerization quality, which comprises: the chaos reconstruction module reconstructs the operating parameters of the propylene polymerization process according to the chaos characteristics of the operating parameters to obtain input variables; the Gabor multi-scale analysis module analyzes the multi-scale characteristics of the input variable by taking frequency as a reference, performs multi-scale reconstruction on the variable through a Gabor kernel function, and extracts local texture feature information of the input variable in each scale and each direction under different frequencies to obtain an input feature signal; the extreme random tree measurement model module is used for converting the input characteristic signals and outputting corresponding soft measurement values of the polypropylene products by taking a polypropylene product sample set and priori knowledge as basis; and the self-adaptive particle swarm module is used for optimizing the bifurcation threshold parameter of the detection system by adopting a self-adaptive particle swarm algorithm.)

1. An on-line detection system for propylene polymerization quality, which is used for measuring the melt index of quality index of propylene polymerization process, and is characterized by comprising:

the chaos reconstruction module reconstructs the operating parameters of the propylene polymerization process according to the chaos characteristics of the operating parameters to obtain input variables;

the Gabor multi-scale analysis module analyzes the multi-scale characteristics of the input variable by taking frequency as a reference, performs multi-scale reconstruction on the variable through a Gabor kernel function, and extracts local texture feature information of the input variable in each scale and each direction under different frequencies to obtain an input feature signal;

the extreme random tree measurement model module is used for converting the input characteristic signals and outputting corresponding soft measurement values of the polypropylene products by taking a polypropylene product sample set and priori knowledge as basis; and

and the self-adaptive particle swarm module is used for optimizing the bifurcation threshold parameter of the detection system by adopting a self-adaptive particle swarm algorithm.

2. The on-line detection system for propylene polymerization quality as claimed in claim 1, wherein the operation parameters are a first propylene feeding flow rate, a second propylene feeding flow rate, a third propylene feeding flow rate, a main catalyst flow rate, an auxiliary catalyst flow rate, a temperature in the stirred tank, a pressure in the tank, a liquid level in the tank, and a volume concentration of hydrogen in the tank.

3. The system of claim 1, wherein the chaotic system of the operating parameters is expressed as z (n) ═ s (n), s (n + T)₁),s(n+T₂),...,s(n+T_d-1)]Wherein s (n) is the n-th sampling point signal of propylene polymerization process, T₁,T₂,...,T_d-1Respectively the sampling time after the nth sampling point; in the chaotic system, the delay time satisfies T_mCondition m τ where τ is the delay time, T_mAnd (3) representing the m-th sampling moment, reconstructing an input signal of the propylene polymerization process into a dynamic chaotic system signal z (n) ═ s (n), s (n + tau), s (n +2 tau), and]where z (n) is the chaos reconstructed signal at the nth moment, τ is the delay time, and d is the embedding dimension of the input signal.

4. The on-line detection system for propylene polymerization quality according to claim 3, wherein the delay time and the embedding dimension are obtained by mutual information method and pseudo-nearest neighbor method, respectively.

5. The on-line detection system for polymerization quality of propylene according to claim 1, wherein the Gabor kernel function is defined as follows:

where z denotes coordinate information of a reconstruction variable, u denotes a direction of the Gabor filter, v denotes a scale of the Gabor filter, i is a complex symbol, exp (ik)_u,vz) an oscillation function in the form of a complex exponential, σ²Is kernel function width, k_u,vRepresenting the response of the Gabor filter in each direction of the respective scale.

6. The system for on-line detection of propylene polymerization quality according to claim 1, wherein the input characteristic signal is obtained by convolution of Gabor kernel function, and the expression is as follows:

G_u，v(z)＝f(z)*ψ_u，v(z) (2)

wherein G is_u,v(z) represents a convolution function of the corresponding dimension v and the direction u near the coordinate z, psi is a Gabor kernel function, and the input variable is analyzed by the Gabor kernel function to obtain a complex input characteristic signal:

G_u，v(z)＝Re(G_u，v(z))+jIm(G_u，v(z)) (3)

the amplitude and phase of the Gabor characteristic signal are respectively:

7. the system of claim 1, wherein the extreme random tree measurement model module employs an extreme random tree to perform input-to-output mapping modeling based on an ensemble learning framework.

8. The system for the on-line detection of the polymerization quality of the propylene according to claim 1, wherein the extreme random tree ensures the sparsity of the model by introducing a Gaussian prior distribution with zero mean value of the weight vector given by the hyperparameter, and the hyperparameter can be estimated by adopting a method of maximizing an edge likelihood function.

9. The system for on-line detection of the polymerization quality of propylene according to claim 1, wherein the adaptive particle swarm module is completed by the following process:

(5.1) randomly generating initial particle group velocity and position;

(5.2) calculating the population diversity index D (t):

wherein Gbest (t) is a global optimal solution reached by the whole particle swarm in the t-th iteration, Fit (Gbest (t)) represents a corresponding fitness value of Gbest (t), m is the particle swarm size, s_i(t) is the position of the ith particle at the tth iteration, Fit(s)_i(t)) represents s_i(t) corresponding fitness value;

(5.3) updating the learning rate parameter Φ (t):

(5.4) updating the speed and position of the particles to generate new populations;

s_in(t+1)＝s_in(t)+q_in(t+1) (9)

wherein alpha is₁Is an individual acceleration parameter, α₂Is the global acceleration parameter that is,andis a random number between 0 and 1, t is the iteration number, and p is the particle swarm size; q. q.s_in(t +1) is the speed of the nth component of the ith particle at the t +1 th iteration, q_in(t) is the velocity of the nth component of the ith particle at the t-th iteration, s_in(t +1) is the position of the nth component of the ith particle at the t +1 th iteration, s_in(t) is the position of the nth component of the ith particle at the tth iteration, Lbest_inThe optimal solution is achieved by the nth component of the ith particle, wherein n is 1, and 2 is an optimization parameter of the detection system;

(5.5) judging whether the algorithm termination condition is met, if so, outputting the global optimal particles and the optimal solution represented by the global optimal particles, and ending iteration; otherwise, returning to (5.2) and continuing the iteration.

10. The system for on-line detection of propylene polymerization quality according to claim 1, further comprising a system update module for on-line updating the detection system, periodically inputting off-line experimental data into the training set, and updating the adaptive extreme stochastic tree measurement model.

Technical Field

The invention relates to the field of polymerization process measuring instruments, the field of machine learning and the field of intelligent optimization, in particular to a chaotic multi-scale self-adaptive propylene polymerization quality online detection system.

Background

Polypropylene is a thermoplastic resin prepared by polymerizing propylene, the most important downstream product of propylene is 50% of propylene in the world, 65% of propylene in China is used for preparing polypropylene, and the polypropylene is one of five common plastics and is closely related to daily life of people. Polypropylene is the fastest growing commodity thermoplastic resin in the world, second only to polyethylene and polyvinyl chloride. In order to make the polypropylene products in China have market competitiveness, the development of impact-resistant copolymerization products, random copolymerization products, BOPP and CPP film materials, fibers and non-woven fabrics with good balance of rigidity, toughness and fluidity and the application of polypropylene in the fields of automobiles and household appliances are important research subjects in the future.

The melt index is one of important quality indexes for determining the grade of a product of polypropylene, determines different purposes of the product, is an important link for controlling the product quality in the production of polypropylene in the measurement of the melt index, and has very important function and guiding significance for production and scientific research.

However, the online analysis and measurement of the melt index are difficult to achieve at present, on one hand, the lack of the online melt index analyzer is caused, and on the other hand, the existing online analyzer is difficult to use due to the fact that the online melt index analyzer is often blocked and inaccurate in measurement or even cannot be used normally. Therefore, currently, MI measurement in industrial production is mainly obtained by manual sampling and off-line assay analysis, and generally, MI can only be analyzed once every 2-4 hours, so that the time delay is large, which brings difficulty to quality control of propylene polymerization production and becomes a bottleneck problem to be solved urgently in production.

Disclosure of Invention

In order to overcome the defects that the measurement precision of the existing propylene polymerization production process is not high and is easily influenced by human factors, the invention aims to provide the chaotic multi-scale self-adaptive propylene polymerization quality on-line detection system which realizes self-adaptive on-line measurement, gives consideration to the multi-scale characteristic and the chaotic characteristic of the polymerization process, and has high confidence level and strong robustness.

Therefore, the invention provides an online detection system for propylene polymerization quality, which is used for measuring the melt index of the quality index of the propylene polymerization process and comprises the following steps:

the chaos reconstruction module reconstructs the operating parameters of the propylene polymerization process according to the chaos characteristics of the operating parameters to obtain input variables;

the Gabor multi-scale analysis module analyzes the multi-scale characteristics of the input variable by taking frequency as a reference, performs multi-scale reconstruction on the variable through a Gabor kernel function, and extracts local texture feature information of the input variable in each scale and each direction under different frequencies to obtain an input feature signal;

the extreme random tree measurement model module is used for converting the input characteristic signals and outputting corresponding soft measurement values of the polypropylene products by taking a polypropylene product sample set and priori knowledge as basis; and

and the self-adaptive particle swarm module is used for optimizing the bifurcation threshold parameter of the detection system by adopting a self-adaptive particle swarm algorithm.

In the system for online detection of propylene polymerization quality, the operation parameters are preferably a first propylene feeding flow rate, a second propylene feeding flow rate, a third propylene feeding flow rate, a main catalyst flow rate, an auxiliary catalyst flow rate, a temperature in the stirred tank, a pressure in the stirred tank, a liquid level in the stirred tank, and a volume concentration of hydrogen in the stirred tank.

In the system for on-line detection of propylene polymerization quality according to the present invention, preferably, the chaotic system expression of the operation parameter is z (n) ═ s (n), s (n + T)₁),s(n+T₂),...,s(n+T_d-1)]Wherein s (n) is the n-th sampling point signal of propylene polymerization process, T₁,T₂,...,T_d-1Respectively the sampling time after the nth sampling point; in the chaotic system, the delay time satisfies T_mCondition m τ where τ is the delay time, T_mAnd (3) representing the m-th sampling moment, reconstructing an input signal of the propylene polymerization process into a dynamic chaotic system signal z (n) ═ s (n), s (n + tau), s (n +2 tau), and]where z (n) is the chaos reconstructed signal at the nth moment, τ is the delay time, and d is the embedding dimension of the input signal.

In the system for on-line detection of propylene polymerization quality according to the present invention, preferably, the delay time and the embedding dimension are obtained by a mutual information method and a pseudo-nearest neighbor method, respectively.

The system for on-line detecting the polymerization quality of propylene of the present invention preferably comprises the Gabor kernel function defined as follows:

where z denotes coordinate information of a reconstruction variable, u denotes a direction of a Gabor filter, v denotes a scale of the Gabor filter, and i is a complex numberSymbol, exp (ik)_u，vz) an oscillation function in the form of a complex exponential, σ²Is kernel function width, k_u，vRepresenting the response of the Gabor filter in each direction of the respective scale.

In the system for online detection of propylene polymerization quality according to the present invention, preferably, the input characteristic signal is obtained by convolution of Gabor kernel function, and the expression is as follows:

G_u，v(z)＝f(z)*ψ_u，y(z) (2)

wherein G is_u，v(z) represents a convolution function of the corresponding dimension v and the direction u near the coordinate z, psi is a Gabor kernel function, and the input variable is analyzed by the Gabor kernel function to obtain a complex input characteristic signal:

G_u，v(Z)＝Re(G_u，v(Z))+jIm(G_u，v(Z)) (3)

the amplitude and phase of the Gabor characteristic signal are respectively:

in the system for online detection of propylene polymerization quality according to the present invention, preferably, the extreme random tree measurement model module uses an extreme random tree and is based on an integrated learning framework to complete input-to-output mapping modeling.

In the system for online detection of propylene polymerization quality, preferably, the extreme random tree ensures model sparsity by introducing gaussian prior distribution with zero mean value of weight vectors given by hyper-parameters, and the hyper-parameters can be estimated by adopting a method of maximizing an edge likelihood function.

The system for online detection of propylene polymerization quality, provided by the invention, preferably comprises the following steps:

(5.1) randomly generating initial particle group velocity and position;

(5.2) calculating the population diversity index D (t):

wherein Gbest (t) is a global optimal solution reached by the whole particle swarm in the t-th iteration, Fit (Gbest (t)) represents a corresponding fitness value of Gbest (t), m is the particle swarm size, s_i(t) is the position of the ith particle at the tth iteration, Fit(s)_i(t)) represents s_i(t) corresponding fitness value;

(5.3) updating the learning rate parameter Φ (t):

(5.4) updating the speed and position of the particles to generate new populations;

s_in(t+1)=s_in(t)+q_in(t+1) (9)

wherein alpha is₁Is an individual acceleration parameter, α₂Is the global acceleration parameter that is,andis a random number between 0 and 1, t is the iteration number, and p is the particle swarm size; q. q.s_in(t +1) is the speed of the nth component of the ith particle at the t +1 th iteration, q_in(t) is the velocity of the nth component of the ith particle at the t-th iteration, s_in(t +1) is the position of the nth component of the ith particle at the t +1 th iteration, s_in(t) is the position of the nth component of the ith particle at the tth iteration, Lbest_inThe optimal solution is achieved by the nth component of the ith particle, wherein n is 1, and 2 is an optimization parameter of the detection system;

(5.5) judging whether the algorithm termination condition is met, if so, outputting the global optimal particles and the optimal solution represented by the global optimal particles, and ending iteration; otherwise, returning to (5.2) and continuing the iteration.

The system for online detection of propylene polymerization quality preferably further comprises a system updating module, wherein the system updating module is used for online updating of the detection system, inputting offline experimental data into a training set regularly, and updating the self-adaptive extreme random tree measurement model.

Specifically, the technical scheme of the invention is as follows:

the chaos multiscale self-adaptive propylene polymerization quality on-line detection system is used for measuring the quality index melt index of a propylene polymerization process, and comprises a chaos reconstruction module, a Gabor multiscale analysis module, a self-adaptive extreme random tree measurement model module, a system updating module and a self-adaptive particle swarm module. Wherein:

(1) a chaotic reconstruction module: used for reconstructing the model input variable input from the DCS database according to the chaotic characteristic thereof,

the input signal of the propylene polymerization process measuring instrument is 9 operation variables of the industrial propylene polymerization process, namely a first propylene feeding flow rate, a second propylene feeding flow rate, a third propylene feeding flow rate, a main catalyst flow rate, an auxiliary catalyst flow rate, the temperature in the stirring kettle, the pressure in the kettle, the liquid level in the kettle and the volume concentration of hydrogen in the kettle. The chaotic system of the input signal is expressed as z (n) ═ s (n), s (n + T)₁),s(n+T₂),...,s(n+T_d-1)]Wherein s (n) is the n-th sampling point signal of propylene polymerization process, T₁,T₂,...,T_d-1Respectively the sampling time after the nth sampling point. In the chaotic system, the delay time satisfies T_mCondition m τ where τ is the delay time, T_mRepresents the m-th sampling moment, therefore, the propylene process input signal can be reconstructed into a dynamic chaotic system signal z (from the embedding dimension and the delay timen)＝[s(n),s(n+τ),s(n+2τ),...,s(n+(d-1)τ)]Where z (n) is the chaos reconstructed signal at the nth moment, τ is the delay time, and d is the embedding dimension of the input signal. The delay time and the embedding dimension of the chaotic reconstruction are respectively obtained by a mutual information method and a pseudo-nearest neighbor method.

(2) Gabor multiscale analysis module: the method is used for analyzing the multi-scale characteristics of the chaotically reconstructed input variable by taking frequency as a reference, and performing multi-scale reconstruction on the variable through a Gabor kernel function to extract local texture feature information of the input variable in each scale and each direction under different frequencies, wherein the Gabor kernel function is defined as follows:

where z denotes reconstruction variable coordinate information, u denotes a direction of the Gabor filter, v denotes a scale of the Gabor filter, i is a complex symbol, exp (ik)_u,vz) an oscillation function in the form of a complex exponential, σ²Is kernel function width, k_u,vRepresenting the response of the Gabor filter in each direction of the respective scale. The partial function of the Gabor kernel functions is as follows: k is a radical of_u,v ²z²/2σ²Is a Gaussian envelope function, k_u,v ²/σ²To compensate for energy spectrum attenuation, the envelope function can limit the range of the oscillating function, usually by windowing, preserve the locality of the wave, and extract the characteristic information near the coordinates. exp (ik)_u,vz) is an oscillating function whose real part is even symmetric with respect to the cosine function and whose imaginary part is odd symmetric with respect to the sine function. exp (-sigma)²/2) represents the filtered DC component, [ exp (ik)_u，vz)-exp(-σ²/2)]The purpose of the operation is to eliminate the influence of the DC component on the filtering effect, the kernel function width sigma²To determine the bandwidth size of the Gabor filter. k is a radical of_u，vRepresenting the response of the Gabor filter in the respective direction of the respective scale, each k_u，vAll represent a Gabor filter, so that when a plurality of different k's are selected_u，vA plurality of different filter banks can be obtained.

The Gabor features are obtained by convolution of kernel functions, and the expression is as follows:

G_u，v(z)＝f(z)*ψ_u，v(z) (2)

wherein G is_u，v(z) represents the convolution function of the corresponding dimension v and direction u around the coordinate z, and ψ is the Gabor kernel function. Analyzing the input variable by using a Gabor function to obtain a complex input characteristic signal:

G_u，v(z)＝Re(G_u，v(z))+jIm(G_u，v(z)) (3)

the amplitude and phase of the Gabor characteristic signal are respectively:

(3) a self-adaptive pole random tree measurement model module: the method is used for establishing an online detection system for the polymerization quality of propylene, and adopts an extreme random tree and an integrated learning framework to complete input-to-output mapping modeling. The extreme random tree training splitting rule ensures the sparsity of the model by introducing the Gaussian prior distribution of the zero mean value of the weight vector given by the hyper-parameters, and the hyper-parameters can be estimated by adopting a method of maximizing the edge likelihood function. The purpose of the whole model is to design a system according to a sample set and prior knowledge, so that the system can predict the output of the polypropylene melt index for new data.

(4) Self-adaptation particle swarm module: the method is used for optimizing the bifurcation threshold parameter of the detection system by adopting a self-adaptive particle swarm optimization, and is completed by adopting the following processes:

(4.1) randomly generating initial particle group velocity and position;

(4.2) calculating the population diversity index D (t):

wherein Gbest (t) is a global optimal solution reached by the whole particle swarm in the t-th iteration, Fit (Gbest (t)) represents a corresponding fitness value of Gbest (t), m is the particle swarm size, s_i(t) is the position of the ith particle at the tth iteration, Fit(s)_i(t)) represents s_i(t) corresponding fitness value;

(4.3) updating the learning rate parameter Φ (t):

(4.4) updating the velocity and position of the particles to generate new populations;

s_in(t+1)＝s_in(t)+q_in(t+1) (9)

wherein alpha is₁Is an individual acceleration parameter, α₂Is the global acceleration parameter that is,andis a random number between 0 and 1, t is the iteration number, and p is the particle swarm size; q. q.s_in(t +1) is the speed of the nth component of the ith particle at the t +1 th iteration, q_in(t) is the velocity of the nth component of the ith particle at the t-th iteration, s_in(t +1) is the position of the nth component of the ith particle at the t +1 th iteration, s_in(t) is the position of the nth component of the ith particle at the tth iteration, Lbest_inThe optimal solution is achieved by the nth component of the ith particle, wherein n is 1, and 2 is an optimization parameter of the detection system;

(4.5) judging whether the algorithm termination condition is met, if so, outputting the global optimal particles and the optimal solution represented by the global optimal particles, and ending iteration; otherwise, returning to (4.2) and continuing iteration;

(5) a system update module: the chaotic multi-scale self-adaptive online detection system for the polymerization quality of propylene further comprises a system updating module, wherein the system updating module is used for updating the detection system online, inputting offline experimental data into a training set periodically and updating a self-adaptive extreme random tree measurement model.

The technical conception of the invention is as follows: the method is characterized in that the melt index of an important quality index in the propylene polymerization process is forecasted on line, in order to overcome the defects of low measurement precision and nonlinear feature extraction of the existing polypropylene melt index measuring instrument, a chaotic phase space reconstruction and a multi-scale analysis method are introduced for feature extraction and sequence reconstruction, and an adaptive method is introduced for analyzing and adjusting a system to adapt to different production working conditions, so that the chaotic multi-scale adaptive detection system in the propylene polymerization production process is obtained. In order to overcome the problem of low measurement precision in the existing propylene polymerization production process, the invention aims to provide an online detection system for propylene polymerization quality of a chaotic multi-scale self-adaptive detection system.

The invention has the following beneficial effects: 1. the chaos reconstruction analysis and the Gabor multi-scale analysis in the propylene polymerization production process effectively represent the dynamic characteristics and nonlinearity under different scales in the actual industrial polymerization process, and the high-precision measurement of the product quality index melt index is realized; 2. the self-adaptive optimization detection system realizes the self-adaptive adjustment of the detection system and the matching of the detection system, and improves the application and popularization capability of the system.

In conclusion, the propylene polymerization quality online detection system provided by the invention can be used for online measurement of important quality index melt index in the propylene polymerization process, overcomes the defects of large time lag and low measurement precision of the traditional detection system, realizes adaptive online measurement, gives consideration to chaotic characteristics and multi-scale characteristics in the polymerization process, and has high confidence level and strong robustness.

Drawings

FIG. 1 is a diagram showing the overall structure of an on-line system for detecting the polymerization quality of propylene according to example 1 of the present invention;

FIG. 2 is a functional block diagram of an on-line detection system for polymerization quality of propylene according to the present invention.

Detailed Description

The following examples illustrate the invention in detail: the present example is carried out on the premise of the technical scheme of the present invention, and detailed embodiments and processes are given, but the scope of the present invention is not limited to the following examples, and the experimental methods without specific conditions noted in the following examples are generally performed according to conventional conditions.

Example 1

Referring to fig. 1 and 2, fig. 1 is an overall architecture diagram of an on-line detection system for propylene polymerization quality according to the present invention; FIG. 2 is a functional block diagram of an on-line detection system for polymerization quality of propylene according to the present invention.

The invention discloses an overall architecture of a chaos multi-scale self-adaptive propylene polymerization quality online detection system, and relates to a propylene polymerization production process 1, a field intelligent instrument 2 for measuring easily-measured variables, a control station 3 for measuring operation variables, a DCS database 4 for storing data, a propylene polymerization quality online detection system 5 and a melt index soft measurement value display instrument 6, wherein the field intelligent instrument 2 and the control station 3 are connected with the propylene polymerization production process 1, the field intelligent instrument 2 and the control station 3 are connected with the DCS database 4, the DCS database 4 is connected with the input end of the propylene polymerization quality online detection system 5, and the output end of the propylene polymerization quality online detection system 5 is connected with the melt index soft measurement value display instrument 6. The above easily measurable variables and the manipulated variables are all operation parameters.

According to the reaction mechanism and the process analysis, in consideration of various factors influencing the melt index in the production process of polypropylene, nine common operating parameters in the actual production process are taken as modeling variables, which are respectively as follows: three propylene feed flow rates, main catalyst flow rate, auxiliary catalyst flow rate, temperature, pressure, liquid level in the kettle, and hydrogen volume concentration in the kettle. Table 1 lists 9 modeling variables required by the chaos multi-scale intelligent optimal propylene polymerization process measuring instrument, namely, the temperature in the kettle (T), the pressure in the kettle (p), the liquid level in the kettle (L), the volume concentration of hydrogen in the kettle (XV), the propylene feeding flow rate of 3 strands (the first propylene feeding flow rate f1, the second propylene feeding flow rate f2, the third propylene feeding flow rate f3) and the catalyst feeding flow rate of 2 strandsFlow rates (main catalyst flow rate f4, cocatalyst flow rate f 5). The melt index off-line test value is used as a check value of the polypropylene production quality detection system 5 and is used for calculating a prediction Error to evaluate the prediction performance of the polypropylene production quality detection system 5, wherein the prediction Error is Root Mean Square Error (RMSE), and the calculation formula isWherein the content of the first and second substances,the output value y of the polypropylene production quality detection system 5_iThe melt index off-line test value is obtained by manual sampling and off-line test analysis, and is collected every 4 hours.

TABLE 1 chaos multiscale intelligent optimum propylene polymerization process measuring instrument required modeling variables

Variable sign	Meaning of variables	Variable sign	Meaning of variables
				T	Temperature in the kettle	f1	First propylene feed flow rate
p	Pressure intensity in kettle	f2	The second stream of propyleneFlow rate of feed
				L	Liquid level in the kettle	f3	Third propylene feed flow rate
Xv	Volume concentration of hydrogen in the autoclave	f4	Main catalyst flow rate
						f5	Flow rate of cocatalyst

With continued reference to fig. 2, the functional structure of the propylene polymerization quality on-line detection system 5 comprises:

(1) a chaos reconstruction module 7 for reconstructing the model input variables input from the DCS database according to the chaos characteristics thereof,

the expression of the chaotic system of the input signal in the propylene polymerization production process is z (n) ═ s (n), s (n + T)₁),s(n+T₂),...,s(n+T_d-1)]Wherein s (n) is the n-th sampling point signal of propylene polymerization process, T₁,T₂,...,T_d-1Respectively the sampling time after the nth sampling point. In the chaotic system, the delay time satisfies T_mCondition m τ where τ is the delay time, T_mRepresents the mth sampling instant, therefore, the propylene process input signal can be reconstructed into a dynamic chaotic system signal z (n) (s (n), s (n + τ), s (n +2 τ), and]wherein z (n) is chaotic reconstructed signal at nth time, and tau is delay timeAnd d is the embedding dimension of the input signal. The delay time and the embedding dimension of the chaotic reconstruction are respectively obtained by a mutual information method and a pseudo-nearest neighbor method.

(2) The Gabor multi-scale analysis module 8 is configured to analyze the input variable with frequency as a reference, perform multi-scale reconstruction on the variable through a Gabor kernel function, and represent local texture information of the input variable in each scale and each direction at different frequencies, where the Gabor kernel function is defined as follows:

where z denotes reconstruction variable coordinate information, u denotes a direction of the Gabor filter, v denotes a scale of the Gabor filter, i is a complex symbol, exp (ik)_u,vz) an oscillation function in the form of a complex exponential, σ²Is kernel function width, k_u,vRepresenting the response of the Gabor filter in each direction of the respective scale. The partial function of the Gabor kernel functions is as follows: k is a radical of_u,v ²z²/2σ²Is a Gaussian envelope function, k_u,v ²/σ²To compensate for energy spectrum attenuation, the envelope function can limit the range of the oscillating function, usually by windowing, preserve the locality of the wave, and extract the characteristic information near the coordinates. exp (ik)_u,vz) is an oscillating function whose real part is even symmetric with respect to the cosine function and whose imaginary part is odd symmetric with respect to the sine function. exp (-sigma)²/2) represents the filtered DC component, [ exp (ik)_u,vz)-exp(-σ²/2)]The purpose of the operation is to eliminate the influence of the DC component on the filtering effect, the kernel function width sigma²To determine the bandwidth size of the Gabor filter. k is a radical of_u,vRepresenting the response of the Gabor filter in the respective direction of the respective scale, each k_u,vAll represent a Gabor filter, so that when a plurality of different k's are selected_u,vA plurality of different filter banks can be obtained.

The Gabor features are obtained by convolution of kernel functions, and the expression is as follows:

G_u，v(Z)＝f(z)*ψ_u，v(Z) (2)

wherein G is_u,v(z) represents the convolution function of the corresponding dimension v and direction u around the coordinate z, and ψ is the Gabor kernel function. Analyzing the input variable by using a Gabor function to obtain a complex input characteristic signal:

G_u，v(Z)＝Re(G_u，v(Z))+jIm(G_u，v(Z)) (3)

the amplitude and phase of the Gabor characteristic signal are respectively:

(3) and the self-adaptive extreme random tree module 9 is used for completing input-to-output mapping modeling based on an integrated learning framework by adopting an extreme random tree. The extreme random tree training splitting rule ensures the sparsity of the model by introducing the Gaussian prior distribution of the zero mean value of the weight vector given by the hyper-parameters, and the hyper-parameters can be estimated by adopting a method of maximizing the edge likelihood function. The purpose of the whole model is to design a system according to a sample set and prior knowledge, so that the system can predict the output of the polypropylene melt index for new data.

(4) The adaptive particle swarm optimization module 10 is configured to optimize parameters of the detection system by using an adaptive particle swarm algorithm, and complete the following steps:

(4.1) randomly generating initial particle group velocity and position;

(4.2) calculating the population diversity index D (t):

wherein Gbest (t) is a global optimal solution reached by the whole particle swarm in the t-th iteration, Fit (Gbest (t)) represents a corresponding fitness value of Gbest (t), m is the particle swarm size, s_i(t) is the ithThe position of the particle at the t-th iteration, Fit(s)_i(t)) represents s_i(t) corresponding fitness value;

(4.3) updating the learning rate parameter Φ (t):

(4.4) updating the velocity and position of the particles to generate new populations;

s_in(t+1)＝s_in(t)+q_in(t+1) (9)

wherein alpha is₁Is an individual acceleration parameter, α₂Is the global acceleration parameter that is,andis a random number between 0 and 1, t is the iteration number, and p is the particle swarm size; q. q.s_in(t +1) is the speed of the nth component of the ith particle at the t +1 th iteration, q_in(t) is the velocity of the nth component of the ith particle at the t-th iteration, s_in(t +1) is the position of the nth component of the ith particle at the t +1 th iteration, s_in(t) is the position of the nth component of the ith particle at the tth iteration, Lbest_inThe optimal solution is achieved by the nth component of the ith particle, wherein n is 1, and 2 is an optimization parameter of the detection system;

(4.5) judging whether the algorithm termination condition is met, if so, outputting the global optimal particles and the optimal solution represented by the global optimal particles, and ending iteration; otherwise, returning to (5.2) and continuing the iteration;

(5) the chaotic multi-scale self-adaptive online detection system for the polymerization quality of the propylene further comprises a system updating module 11, wherein the system updating module is used for updating the detection system online, inputting offline experimental data into a training set periodically and updating a self-adaptive extreme random tree measurement model.

Take specific data as an example: in this embodiment, 9 modeling variables required for acquiring in the DCS system are extracted to obtain a variable input matrix:

inputting data into a polypropylene measuring system 5, and obtaining predicted values of the melt index [2.5203,2.5066,2.4838,2.3670 and 2.3759 ] by a chaotic multi-scale self-adaptive measuring module]. Melt index off-line assay values [2.48,2.47,2.49,2.31,2.36]As a check value of the polypropylene measurement system 5, the method is used for calculating a prediction Error to evaluate the prediction precision of the polypropylene production quality measurement system 5, wherein the prediction Error is Root Mean Square Error (RMSE), and the calculation formula isWherein the content of the first and second substances,for the output value, y, of the polypropylene production quality measuring system 5_iThe forecast deviation of the polypropylene production quality detection system 5 is [0.0403,0.0366, -0.0062,0.057,0.0159 ] for the melt index off-line test value]The root mean square error is 0.0403, and the melt index prediction value and the prediction precision of the measurement system are obtained.

The present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof, and it is therefore intended that all such changes and modifications as fall within the true spirit and scope of the invention be considered as within the following claims.