CN101251747B - Modelling method for industrial device model for dimethylbenzene oxidation reaction - Google Patents

Abstract

The present invention discloses a modeling method for paraxylene (PX) oxidation reaction industrial device models. Through multivariate linear regression technology, a rate constant correlation model for all reaction factors and all steps of consecutive reaction is built, and optimization is carried out directly on the basis of the production data of an industrial device to look for the regression coefficients of the rate constant correlation model, so as to build a model capable of well describing characteristics of the industrial device. The modeling method avoids insufficient understanding on reaction mechanism and the characteristics of the industrial device, can describe the complex influence of each reaction factor to the process of PX oxidation reaction through higher order terms, builds the industrial device model directly on the basis of the production data of the industrial device, avoids that the model of PX oxidation reaction is built through the experimental device tests which consume time and have high cost. Experimental results show that: the built industrial device model has good prediction accuracy, can completely meet application demands of the industrial device, and has very high industrial application value.

Description

The modeling method of p xylene oxidation reaction industrial device model

Technical field

The invention belongs to petrochemical complex Chemical Reaction Engineering field; Relate to the pure terephthalic acid (hereinafter to be referred as PTA; Be Pure Terephthalic Acid) modeling method of P-xylene in the production technology (hereinafter to be referred as PX, i.e. P-xylene) oxidation reaction industrial device model.

Background technology

PTA is the important source material of synthetic polyester fibers and plastics, mainly is used for the intermedium phthalic acid glycol ester (PET) of synthesizing polyester, and the PTA process units is the faucet device of whole chemical fibre industry.Whole PTA production technology comprises oxidation unit and refined unit, and its flow process is as shown in Figure 1.With the P-xylene is raw material, and acetic acid is solvent, at cobalt acetate, under the effect of manganese acetate catalyst, is promoter and airborne oxygen reaction with hydrogen bromide (or tetrabromoethane), generates terephthalic acid (TPA).A large amount of reaction heat of emitting in the reaction are taken away through the evaporation of solvent, and reclaim this part heat through byproduct steam.Oxidation liquid is through the crystallizer decrease temperature and pressure of series connection, and through filtering, drying obtains the intermediate product crude terephthalic acid again.Solvent etc. reclaim and recycle through solvent recovery unit.Crude terephthalic acid is made into certain density slurry with deionized water, is heated to require to deliver to hydrogenation reactor after the solution temperature.Through catalytic hydrogenation reaction, make that impurities is converted into water-soluble substances in the crude terephthalic acid.Hydrogenation reaction solution is sent to the hydro-extractor separation after the decrease temperature and pressure step by step in the crystallizer of series connection, the filter cake that obtains through filtration and dry, makes the fibre-grade pure terephthalic acid then again with the deionized water making beating.Wherein the PX oxidation reaction is the core of whole PTA production technology, and whether its operating conditions suitable, stable quality, output and the material consumption that is directly connected to final products and energy consumption etc.

Along with the enhancing of the development of technology, the market demand and in the face of intense market competition, in order to obtain bigger scale and benefit, PTA manufacturing enterprise has carried out capacity expansion revamping to the process units of tradition (or even newly-built) one after another; And because the frequent variation of raw material sources etc., make the original design conditions of optimum operation condition substantial deviation of oxidation reaction apparatus.How to confirm the operative technique level of operating conditions optimum under the present production status, raising oxidation reaction apparatus; Really accomplishing to make process units peace, steady, long, full, excellent the operation, is one of major issue of business decision layer and production technology personnel institute common concern, urgent solution.Utilize PX oxidation reaction industrial device model to instruct and produce adjustment, optimization means operating conditions, eliminate production bottleneck, implementation procedure monitoring, fault diagnosis are one of valid approach of raising device economic benefit.In addition,, also can be advanced control the important real-time calculated value of controlled variable is provided, play the effect of " soft instrument " for device through online the putting into operation of PX oxidation reaction industrial device model.

PX oxidation reaction industrial device model quantitatively, has synthetically been described the influence of each reaction factor to course of reaction with the form of mathematics.If predicted results can reflect actual conditions well, will play important effect to the optimization production operation.But the PX oxidation reaction is a liquid-phase catalytic oxidation under the HTHP; The coexistence of course of reaction gas-liquid-solid three-phase; The a lot of subsidiary reactions of simultaneous; Course of reaction relates to heat transfer, mass transfer, solid crystal and the slurry suspension etc. of gas-liquid, and each reaction factor (mainly containing catalyst concn and composition thereof, temperature of reaction, ratio of solvent, reactant concentration, the residence time etc.) is complicated to the course of reaction influence.Because the restriction of its complicacy and human present cognitive level, real industrial PX oxidation reaction process mechanism model can't be set up; The production data that the accumulation of industrial reaction process is a large amount of; These data are containing the characteristic information of commercial plant course of reaction; But always limited from the sample data that actual production process is gathered, comprise the different redundant information of a large amount of repetition degree in the sample data, reasons such as the extensive existence of man-made noise; Make simple mathematical model be difficult to entirely accurate ground and embody each effects of operation conditions based on sample data; Even appearance is violated the situation of already known processes mechanism and experimental knowledge, its restricted application, the demand that is difficult to satisfied operation optimization on a large scale and control.Therefore, the foundation of the good PX oxidation reaction industrial device model of estimated performance is a difficult point of present PTA production field.

Chinese patent (ZL02148477.5) discloses a kind of modeling method of P-xylene liquid-phase catalytic oxidation kinetic model.At first, set up neural network rate constant model, the i.e. correlation model of each reaction factor and rate constant through semicontinuous gas-liquid reactor data; Then, through neural network rate constant model is added correction factor, gather the data of industrial reactor; Adopt optimized Algorithm; Match total departure minimum with model is a target, confirms correction factor, sets up commercial plant P-xylene liquid-phase catalytic oxidation kinetic model.The model of setting up can pass through each reaction factor value, i.e. temperature of reaction (x ₁, ℃), ratio of solvent (x ₂, mol/Kg), Co catalysts concentration (x ₂, ppmw), manganese catalyst concentration (x ₄, ppmw), and bromine promoter concentration (x ₅, ppmw),, try to achieve that the consecutive reaction network respectively goes on foot reaction rate constants k in the industrialization P-xylene liquid-phase catalytic oxidation by the rate constant neural network model ₁, i=1,2,3,4; And then try to achieve PX in the reactor discharging, p-tolyl aldehyde (p-tolualdehyde; Abbreviation TALD), p-methylbenzoic acid (p-toluic acid; Abbreviation PT acid), to carboxyl benzaldehyde (4-carboxybenzaldehyde; Abbreviation 4-CBA), the content of terephthalic acid (TPA) intermediate products such as (terephthalic acid are called for short TA) and final product.

There are some following shortcomings in this patent (ZL02148477.5):

1) must there be the experimental data of semicontinuous gas-liquid reactor to set up the neural network correlation model of each reaction factor and rate constant.Because PX oxidation reaction time consumption of experimental process and spend hugely, so people hold in close confidence its experimental data, are difficult to obtain.

2) neural network modeling approach need have a large amount of training samples, could set up reliable model, and PX oxidation reaction experimentation experimental data is always very limited.

3) neural network structure is complicated, and the model of foundation is easy to over-fitting, influences the precision of prediction of model.

4) because there is greatest differences in lab scale, semicontinuous gas-liquid reactor and industrial processes large-scale (or huge) the continuous oxidation reaction device under the laboratory condition; Therefore; Only add correction term, might be difficult to describe the characteristic of commercial plant PX oxidation reaction process through the neural network rate constant model that experimental data is set up.

Simultaneously; Though just have the people that the PX oxidation reaction is studied from the twenties in 20th century; And obtaining many achievements in research aspect the mechanism of catalytic reaction research of PX oxidation reaction mechanism and cobalt manganese bromine; But publish document mostly to PX low-temperature atmosphere-pressure course of reaction, the HTHP oxidation technology of reaction conditions and industrial manufacture process differs greatly.A spot of document (the PX oxidation reaction being studied disclosed document like units such as the Dalian synthesising fibre institute of China, high mountain head factory, University Of Tianjin, Zhejiang University, East China University of Science) is though be the oxidation technology to HTHP; But relevant kinetic model is set up in the experiment of all under lab scale, semi-continuous gas-liquid reactor condition, carrying out.Since under the amplification of production run, the laboratory condition lab scale, there is huge difference in semi-continuous gas-liquid reactor with industrial processes large-scale (or huge) continuous oxidation reaction device, is difficult to describe the characteristic of commercial plant through the kinetic model of laboratory data foundation.

Summary of the invention

The object of the invention provides a kind of modeling method of PX oxidation reaction industrial device model.Employing is set up the industrial device model framework from PX, TALD, PT acid, 4-CBA, to the consecutive type reaction network (as shown in Figure 2) of final products TA.Set up each key reaction factor, i.e. temperature of reaction (x through multicomponent linear regressioning technology ₁, ℃), ratio of solvent (x ₂, mol/Kg), concentration of cobalt ions (x ₃, ppmw), manganese ion concentration (x ₄, ppmw), bromide ion concentration (x ₅, ppmw), and high-order term (x ₁ ², x ₂ ²..., x ₅ ², x ₁ ³, x ₂ ³..., x ₅ ³..., x ₁ ^m, x ₂ ^m..., x ₅ ^m), with the correlation model of each step consecutive reaction rate constant.Directly based on the commercial plant production data, excavate the regression coefficient of rate constant correlation model, the PX oxidation reaction industrial device model of commercial plant characteristic can be well described in final foundation.Based on PX oxidation reaction industrial device model, can calculate under the differential responses factor content of intermediate product and final products TA; In the examination commercial plant, each reaction factor is to the influence of course of reaction; Adjust or the like for optimization, process control, fault diagnosis, the production load of production operating conditions, basis and foundation are provided.

Main contents of the present invention are following:

A kind of modeling method of p xylene oxidation reaction industrial device model is characterized in that, this modeling method comprises following steps:

(1) adopt from P-xylene, p-tolyl aldehyde, p-methylbenzoic acid, to carboxyl benzaldehyde, to the consecutive type reaction network of final products terephthalic acid (TPA), set up following p xylene oxidation reaction industrial device model framework:

$\left\{\begin{array}{c}{r}_1=\frac{d{C}_1}{\mathrm{dt}}=-k_1{C}_1^{n_1}{C}_{O_2}^{m_1}\\ r_2=\frac{d{C}_2}{\mathrm{dt}}=k_1{C}_1^{n_1}{C}_{O_2}^{m_1}-k_2{C}_2^{n_2}{C}_{O_2}^{m_2}\\ r_3=\frac{d{C}_3}{\mathrm{dt}}=k_2{C}_2^{n_2}{C}_{O_2}^{m_2}-k_3{C}_3^{n_3}{C}_{O_2}^{m_3}\\ {\mathrm{<figure-callout\; id="4"\; label="r"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">r</figure-callout>}}_4=\frac{{\mathrm{<figure-callout\; id="4"\; label="dC"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">dC</figure-callout>}}_4}{\mathrm{dt}}=k_3{C}_3^{n_3}{C}_{O_2}^{m_3}-k_4{C}_4^{n_4}{C}_{O_2}^{m_4}\\ r_5=\frac{{\mathrm{dC}}_5}{\mathrm{dt}}{k}_4{C}_4^{n_4}{C}_{O_2}^{{\mathrm{<figure-callout\; id="4"\; label="m"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">m</figure-callout>}}_4}\end{array}\right.$

Wherein,

r ₁, r ₂, r ₃, r ₄, r ₅Be respectively P-xylene, p-tolyl aldehyde, p-methylbenzoic acid, to the reaction rate of carboxyl benzaldehyde, terephthalic acid (TPA);

T is reaction time (min);

n ₁, n ₂, n ₃, n ₄For P-xylene, p-tolyl aldehyde, p-methylbenzoic acid, to the order of reaction of carboxyl benzaldehyde;

m ₁, m ₂, m ₃, m ₄Be oxygen respectively go on foot consecutive reaction progression;

k ₁, k ₂, k ₃, k ₄For consecutive reaction respectively goes on foot pairing rate constant;

C ₁, C ₂, C ₃, C ₄, C ₅,

For P-xylene, p-tolyl aldehyde, p-methylbenzoic acid, to the concentration (mol/Kg) of carboxyl benzaldehyde, terephthalic acid (TPA), oxygen relative unit weight solvent acetic acid;

(2) set up each key reaction factor, i.e. temperature of reaction (x through multicomponent linear regressioning technology ₁, ℃), ratio of solvent (x ₂, mol/Kg), concentration of cobalt ions (x ₃, ppmw), manganese ion concentration (x ₄, ppmw), bromide ion concentration (x ₅, ppmw) and the high-order term (x of reaction factor ₁ ², x ₂ ²..., x ₅ ², x ₁ ³, x ₂ ³..., x ₅ ³..., x ₁ ^m, x ₂ ^m..., x ₅ ^m) go on foot the correlation model of consecutive reaction rate constant with each, promptly

$\left\{\begin{array}{c}{k}_i={\mathrm{\&beta;}}^{\left(i\right)}{x}^T\\ {\mathrm{\&beta;}}^{\left(i\right)}=[{\mathrm{\&beta;}}_0^{\left(i\right)}{\mathrm{\&beta;}}_1^{\left(i\right)}{\mathrm{\&beta;}}_2^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_5^{\left(i\right)}{\mathrm{\&beta;}}_6^{\left(i\right)}{\mathrm{\&beta;}}_7^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_{10}^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_{5m-4}^{\left(i\right)}{\mathrm{\&beta;}}_{5m-3}^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_{5m}^{\left(i\right)}]\\ x=[1x_1{x}_2\&CenterDot;\&CenterDot;\&CenterDot;x_5{x}_1^2{x}_2^2\&CenterDot;\&CenterDot;\&CenterDot;x_5^2\&CenterDot;\&CenterDot;\&CenterDot;x_1^m{x}_2^m\&CenterDot;\&CenterDot;\&CenterDot;x_5^m]\end{array}\right.$

Wherein,

β ⁽ⁱ⁾Be k _iThe regression coefficient vector;

X by 1, the vector formed of the high-order term of above-mentioned reaction factor and above-mentioned reaction factor;

(3) directly based on the commercial plant production data, try to achieve above-mentioned β through the intelligent optimization algorithm optimizing ⁽ⁱ⁾, try to achieve above-mentioned k through above-mentioned steps (2) then _i, the final p xylene oxidation reaction industrial device model of describing the commercial plant characteristic of setting up, the content of intermediate product and final product in the monitoring reaction course in real time.

The said order of reaction adopts the reaction order numerical value of generally acknowledging, i.e. n ₁=0.65, n ₂=1, n ₃=1, n ₄=1, m ₁=0, m ₂=0, m ₃=0, m ₄=0.

Comprise a following determining step (2#) in the said step (2):

Judge that whether total error E that computes obtains is than preset value ε _All(common 0.05≤ε _All≤0.15) little, if then identification finishes, and tries to achieve said β ⁽ⁱ⁾Otherwise, adjust search through intelligent optimization algorithm, recomputate.

$E_j=\left\{\begin{array}{c}\frac{1}{5}\underset{i=1}{\overset{5}{\mathrm{\&Sigma;}}}\left|\frac{{\tilde{C}}_i^{\left(j\right)}-C_i^{\left(j\right)}}{C_i^{\left(j\right)}}\right|,\mathrm{if}\underset{i=1}{\overset{5}{\mathrm{\&Sigma;}}}\left|\frac{{\tilde{C}}_i^{\left(j\right)}-C_i^{\left(j\right)}}{C_i^{\left(j\right)}}\right|\&GreaterEqual;\mathrm{\&epsiv;}\\ 0,\mathrm{if}\underset{i=1}{\overset{5}{\mathrm{\&Sigma;}}}\left|\frac{{\tilde{C}}_i^{\left(j\right)}-C_i^{\left(j\right)}}{C_i^{\left(j\right)}}\right|<\mathrm{\&epsiv;}\end{array}\right.,j=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,n$

Wherein, 1＞ε＞0; More commonly usedly, 0.01≤ε≤0.05

$E=\frac{1}{n}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{E}_j$

Wherein, j representes j sample point; C _i ^(j)The assay value of representing i intermediate product of j sample point; The Model Calculation value of representing i intermediate product of j sample point; E _jIt is the Model Calculation error of j sample point; ε is each sample point error of calculation limit, if the Model Calculation error is worth less than this, then is not counted in total error.

The aforementioned calculation process can realize through the automatic arithmetic facility of routine, for example mcu programming.

Advantage of the present invention: multicomponent linear regressioning technology is adopted directly through each key reaction factor and high-order term in (1), sets up the correlation model of each key reaction factor and rate constant.This correlation model not only can be described the complex effects rule of each reaction factor to the PX oxidation reaction, and model structure is simple.(2) directly based on the commercial plant production data, excavate the characteristic information of commercial plant, set up PX oxidation reaction industrial device model, avoided obtaining data through experimental provision test consuming time, that cost is huge.Simultaneously, avoided greatest differences, can't describe the characteristic of commercial plant through the PX oxidation reaction model of experimental provision test data foundation owing to experimental provision and commercial plant.

1.PX oxidation reaction network and model framework

The PX oxidation reaction is complicated free radical reaction; Many intermediate products and accessory substance are arranged in the reaction system; Component to all takes in, from analytic angle still be the Model Calculation angle all be impossible, and also there is no need from the commercial Application angle.The present invention adopts lumped reaction kinetic model framework, promptly only considers important intermediate component and final reacting product, proposes simple relatively reaction kinetics network.Different researchers has proposed different reaction networks, but thinks that all the PX oxidation reaction is a consecutive type reaction.Consider that each concentration of intermediate products situation of change goes up interested component with industry in the course of reaction, adopt reaction network as shown in Figure 2, wherein k ₁, k ₂, k ₃, k ₄For consecutive reaction respectively goes on foot pairing rate constant.According to the consecutive reaction step, make the concentration (mol/Kg) of PX, TALD, PT acid, 4-CBA, TA, oxygen relative unit weight solvent acetic acid be respectively respectively: C ₁, C ₂, C ₃, C ₄, C ₅,

Then according to the mass action law, the reaction velocity equation of each component can be expressed as:

$\left\{\begin{array}{c}{r}_1=\frac{d{C}_1}{\mathrm{dt}}=-k_1{C}_1^{n_1}{C}_{O_2}^{m_1}\\ r_2=\frac{d{C}_2}{\mathrm{dt}}=k_1{C}_1^{n_1}{C}_{O_2}^{m_1}-k_2{C}_2^{n_2}{C}_{O_2}^{m_2}\\ r_3=\frac{d{C}_3}{\mathrm{dt}}=k_2{C}_2^{n_2}{C}_{O_2}^{m_2}-k_3{C}_3^{n_3}{C}_{O_2}^{m_3}\\ {\mathrm{<figure-callout\; id="4"\; label="r"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">r</figure-callout>}}_4=\frac{{\mathrm{<figure-callout\; id="4"\; label="dC"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">dC</figure-callout>}}_4}{\mathrm{dt}}=k_3{C}_3^{n_3}{C}_{O_2}^{m_3}-k_4{C}_4^{n_4}{C}_{O_2}^{m_4}\\ r_5=\frac{{\mathrm{dC}}_5}{\mathrm{dt}}{k}_4{C}_4^{n_4}{C}_{O_2}^{m_4}\end{array}\right.---\left(1\right)$

R wherein ₁, r ₂, r ₃, r ₄, r ₅Be respectively the reaction rate of PX, TALD, PT acid, 4-CBA, TA, t is reaction time (min), n ₁, n ₂, n ₃, n ₄Be the order of reaction of PX, TALD, PT acid, 4-CBA, m ₁, m ₂, m ₃, m ₄Be oxygen respectively go on foot consecutive reaction progression.System of equations (1) is and comprises undetermined parameter k ₁, k ₂, k ₃, k ₄, n ₁, n ₂, n ₃, n ₄And m ₁, m ₂, m ₃, m ₄PX oxidation reaction kinetics model framework, order of reaction n wherein ₁, n ₂, n ₃, n ₄And m ₁, m ₂, m ₃, m ₄Can obtain through experiment, or adopt universally recognized value, i.e. n ₁=0.65, n ₂=1, n ₃=1, n ₄=1, m ₁=0, m ₂=0, m ₃=0, m ₄=0.

2. rate constant correlation model

In the PX oxidation reaction process, the main factor that influences reaction rate has: temperature of reaction (x ₁, ℃), ratio of solvent (x ₂, mol/Kg), concentration of cobalt ions (x ₃, ppmw), manganese ion concentration (x ₄, ppmw), bromide ion concentration (x ₅, ppmw), and these 5 reaction factors are complicated to the reaction rate influence.In the above-mentioned kinetic model framework, rate constants k _i, i=1,2,3,4 have contained these influence factors.

Because there are complicated nonlinear relationship in each reaction factor and rate constant, therefore each reaction factor is added high-order term, set up the non-linear correlation model.If, add 2,3 of each reaction factor ..., m time, promptly

x ₁ ²，x ₂ ²，…，x ₅ ²，…，x ₁ ^m，x ₂ ^m，…，x ₅ ^m

After adding high-order term, the nonlinear relationship between each reaction factor and the rate constant can be described; It is big more to add high-order term number of times m, and it is high more to describe nonlinear degree, but model stability also reduces.Therefore in practical application, be benchmark, choose and add minimum high-order term number of times to reach model prediction precision or fitting precision.Then to k _i, i=1,2,3,4 multiple linear regression model is:

$k_i={\mathrm{\&beta;}}_0^{\left(i\right)}+{\mathrm{\&beta;}}_1^{\left(i\right)}{x}_1+{\mathrm{\&beta;}}_2^{\left(i\right)}{x}_2+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&beta;}}_5^{\left(i\right)}{x}_5+{\mathrm{\&beta;}}_6^{\left(i\right)}{x}_1^2+{\mathrm{\&beta;}}_7^{\left(i\right)}{x}_2^2+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&beta;}}_{10}^{\left(i\right)}{x}_5^2+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&beta;}}_{5m-4}^{\left(i\right)}{x}_1^m+{\mathrm{\&beta;}}_{5m-3}^{\left(i\right)}{x}_2^m+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&beta;}}_{5m}^{\left(i\right)}{x}_5^m$

(2)

Also can be expressed as:

$\left\{\begin{array}{c}{k}_i={\mathrm{\&beta;}}^{\left(i\right)}{x}^T\\ {\mathrm{\&beta;}}^{\left(i\right)}=[{\mathrm{\&beta;}}_0^{\left(i\right)}{\mathrm{\&beta;}}_1^{\left(i\right)}{\mathrm{\&beta;}}_2^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_5^{\left(i\right)}{\mathrm{\&beta;}}_6^{\left(i\right)}{\mathrm{\&beta;}}_7^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_{10}^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_{5m-4}^{\left(i\right)}{\mathrm{\&beta;}}_{5m-3}^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_{5m}^{\left(i\right)}]\\ x=[1x_1{x}_2\&CenterDot;\&CenterDot;\&CenterDot;x_5{x}_1^2{x}_2^2\&CenterDot;\&CenterDot;\&CenterDot;x_5^2\&CenterDot;\&CenterDot;\&CenterDot;x_1^m{x}_2^m\&CenterDot;\&CenterDot;\&CenterDot;x_5^m]\end{array}\right.---\left(3\right)$

Wherein, β ⁽ⁱ⁾Be k _iThe regression coefficient vector, x by 1, the vector formed of reaction factor and high-order term.

The regression coefficient vector beta of then working as formula (3) ⁽ⁱ⁾Confirm then each reaction factor and rate constant modelling; And then PX oxidation reaction industrial device model (being system of equations (1)) is set up.

3.PX oxidation reaction industrial device model

Because a large amount of production data of industrial PX oxidation reaction process accumulation, these data are containing the characteristic information of commercial plant course of reaction, therefore if can from the commercial plant production data, draw the regression coefficient vector beta of formula (3) ⁽ⁱ⁾Value, can avoid obtaining data through consuming time, the test of the huge experimental provision of cost, try to achieve the regression coefficient vector beta ⁽ⁱ⁾Simultaneously, directly excavate the regression coefficient vector beta from the commercial plant production data ⁽ⁱ⁾Value, extract the characteristic information of commercial plant, make the PX oxidation reaction industrial device model of foundation well describe the characteristic of commercial plant; Avoided because the greatest differences of experimental provision and commercial plant through the experimental provision test data, is tried to achieve the regression coefficient vector beta ⁽ⁱ⁾, the rate constant correlation model of foundation can't each reaction factor of accurate description to the influence of commercial plant course of reaction.

Directly, draw the regression coefficient vector beta of formula (3) based on the commercial plant production process data ⁽ⁱ⁾Value, promptly PX oxidation reaction industrial device model regression parameter debate know as shown in Figure 3.Concrete steps are following:

(1) gathers the commercial plant production process data, form sample data.If gathered n group data, every group of data comprise [x ₁, x ₂, x ₃, x ₄, x ₅, t, C ₁, C ₂, C ₃, C ₄, C ₅].

(2) confirm to add the value of high-order term number of times m, confirm the fitting precision ε of model _All, 0＜ε _All＜1 (in the industry practice application, gets 0.05≤ε usually _All≤0.15).

(3) confirm the regression coefficient vector beta ⁽ⁱ⁾Initial value, set up initial rate constant correlation model; Adopt the reaction order numerical value of generally acknowledging, i.e. n ₁=0.65, n ₂=1, n ₃=1, n ₄=1, m ₁=0, m ₂=0, m ₃=0, m ₄=0; Set up initial p X oxidation reaction industrial device model (being system of equations (1)).

(4) based on system of equations (1), through the independent variable [x of sample data ₁, x ₂, x ₃, x ₄, x ₅, t], the result of calculation of acquisition model, after promptly obtaining to react completely, the content of each intermediate product and final product, promptly

(5) whether judgment models reaches accuracy requirement.Concrete steps are following: calculate total error E

$E_j=\left\{\begin{array}{c}\frac{1}{5}\underset{i=1}{\overset{5}{\mathrm{\&Sigma;}}}\left|\frac{{\tilde{C}}_i^{\left(j\right)}-C_i^{\left(j\right)}}{C_i^{\left(j\right)}}\right|,\mathrm{If}\underset{i=1}{\overset{5}{\mathrm{\&Sigma;}}}\left|\frac{{\tilde{C}}_i^{\left(j\right)}-C_i^{\left(j\right)}}{C_i^{\left(j\right)}}\right|\&GreaterEqual;\mathrm{\&epsiv;}\\ 0,\mathrm{If}\underset{i=1}{\overset{5}{\mathrm{\&Sigma;}}}\left|\frac{{\tilde{C}}_i^{\left(j\right)}-C_i^{\left(j\right)}}{C_i^{\left(j\right)}}\right|<\mathrm{\&epsiv;}\end{array}\right.$ , j=1,2 ..., n, wherein, 1＞ε＞0.

$E=\frac{1}{n}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{E}_j---\left(4\right)$

Wherein, j representes j sample point; C _i ^(j)The assay value of representing i intermediate product of j sample point; The Model Calculation value of representing i intermediate product of j sample point; E _jIt is the Model Calculation error of j sample point; ε is each sample point error of calculation limit, if the Model Calculation error is worth less than this, then is not counted in total error, common 0.01≤ε≤0.05.

(6) judge total error E≤ε _All, or stopped convergence (being that total error has reached minimum), if, then debate to know and finish, try to achieve the regression coefficient vector beta ⁽ⁱ⁾, set up PX oxidation reaction industrial device model; Otherwise, forward (7) step to.

(7) through intelligent optimization algorithm (like genetic algorithm etc.) to the regression coefficient vector beta ⁽ⁱ⁾Adjust search, obtain new regression coefficient vector beta ⁽ⁱ⁾, forward (4) step to.

The idiographic flow block diagram of whole algorithm is as shown in Figure 3.

Description of drawings

Fig. 1 is the PTA production process route;

Fig. 2 is a PX oxidation reaction network;

Fig. 3 is that PX oxidation reaction industrial device model regression parameter is debated the knowledge block diagram.

Embodiment

Explanation through following examples will help to understand the present invention, but not limit content of the present invention.

Below through embodiment the present invention is described further:

Like Fig. 1, the PX oxidation reaction is to be raw material with the P-xylene, and acetic acid is solvent, at cobalt acetate, under the effect of manganese acetate catalyst, is promoter and airborne oxygen reaction with the hydrogen bromide, generates terephthalic acid (TPA).Consider important intermediate component and final reacting product, adopt reaction network as shown in Figure 2, set up the reaction velocity equation (1) of each component.The foundation of PX oxidation reaction industrial device model, promptly debating of regression models parameter known as shown in Figure 3ly, and the practical implementation step is following:

If, to the rate constant correlation model, add m=2 item of each reaction factor, promptly the rate constant correlation model does

$k_i={\mathrm{\&beta;}}_0^{\left(i\right)}+{\mathrm{\&beta;}}_1^{\left(i\right)}{x}_1+{\mathrm{\&beta;}}_2^{\left(i\right)}{x}_2+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&beta;}}_5^{\left(i\right)}{x}_5+{\mathrm{\&beta;}}_6^{\left(i\right)}{x}_1^2+{\mathrm{\&beta;}}_7^{\left(i\right)}{x}_2^2+\&CenterDot;\&CenterDot;\&CenterDot;+{\mathrm{\&beta;}}_{10}^{\left(i\right)}{x}_5^2,i=\mathrm{1,2,3,4}---\left(5\right)$

Adopt the reaction order numerical value of generally acknowledging, i.e. n ₁=0.65, n ₂=1, n ₃=1, n ₄=1, m ₁=0, m ₂=0, m ₃=0, m ₄=0.Then PX oxidation reaction industrial device model framework is:

$\left\{\begin{array}{c}{r}_1=\frac{d{C}_1}{\mathrm{dt}}=-k_1{C}_1^{0.65}\\ r_2=\frac{d{C}_2}{\mathrm{dt}}=k_1{C}_1^{0.65}-k_2{C}_2\\ r_3=\frac{d{C}_3}{\mathrm{dt}}=k_2{C_2-k}_3{C}_3\\ {\mathrm{<figure-callout\; id="4"\; label="r"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">r</figure-callout>}}_4=\frac{{\mathrm{<figure-callout\; id="4"\; label="dC"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">dC</figure-callout>}}_4}{\mathrm{dt}}=k_3{C}_3-k_4{C}_4\\ r_5=\frac{{\mathrm{dC}}_5}{\mathrm{dt}}{k}_4{C}_4\end{array}\right.---\left(6\right)$

Gather the sample data under the different representative operating modes in 200 groups of production runes, every group of sample data comprises [x ₁, x ₂, x ₃, x ₄, x ₅, t, C ₁, C ₂, C ₃, C ₄, C ₅], then adopt intelligent optimization algorithm, be target (or also can adopt error of fitting ε, be target) with formula (4) total error minimum like ε=0.1 less than set-point, try to achieve optimum regression coefficient value ${\mathrm{\&beta;}}^{\left(i\right)}=[{\mathrm{\&beta;}}_0^{\left(i\right)}{\mathrm{\&beta;}}_1^{\left(i\right)}{\mathrm{\&beta;}}_2^{\left(i\right)}...{\mathrm{\&beta;}}_5^{\left(i\right)}{\mathrm{\&beta;}}_6^{\left(i\right)}{\mathrm{\&beta;}}_7^{\left(i\right)}...{\mathrm{\&beta;}}_{10}^{\left(i\right)}],$ I=1,2,3,4 is following:

$\left\{\begin{array}{c}{\mathrm{\&beta;}}^{\left(1\right)}=\left[\begin{array}{ccccccccccc}0.0007& 0.00038& -0.23& 0.31& 0.36& 0.48& 1.06E-05& 0.035& 0.026& 0.028& 0.046\end{array}\right]\\ {\mathrm{\&beta;}}^{\left(2\right)}=\left[\begin{array}{ccccccccccc}-0.0012& 0.0052& -0.35& 0.35& 0.3& 0.37& 6.81E-06& 0.042& 0.026& 0.021& 0.033\end{array}\right]\\ {\mathrm{\&beta;}}^{\left(3\right)}=\left[\begin{array}{ccccccccccc}0.00021& 0.0013& -0.39& 0.037& 0.033& 0.048& -2.92E-06& 0.0068& 0.0035& 0.003& 0.006\end{array}\right]\\ {\mathrm{\&beta;}}^{\left(4\right)}=\left[\begin{array}{ccccccccccc}-0.00064& 0.0036& -0.32& 0.83& 0.79& 1.34& -6.47E-06& 0.054& 0.083& 0.074& 0.17\end{array}\right]\end{array}\right.---\left(7\right)$

Then, with the regression coefficient value of optimum, i.e. formula (7), substitution rate constant correlation model, i.e. formula (5), the rate constant correlation model is just set up; With rate constant correlation model substitution PX oxidation reaction industrial device model framework, i.e. formula (6), PX oxidation reaction industrial device model is just set up, as follows:

$\left\{\begin{array}{c}{r}_1=\frac{d{C}_1}{\mathrm{dt}}=-k_1{C}_1^{0.65}\\ r_2=\frac{d{C}_2}{\mathrm{dt}}=k_1{C}_1^{0.65}-k_2{C}_2\\ r_3=\frac{d{C}_3}{\mathrm{dt}}=k_2{C_2-k}_3{C}_3\\ {\mathrm{<figure-callout\; id="4"\; label="r"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">r</figure-callout>}}_4=\frac{{\mathrm{<figure-callout\; id="4"\; label="dC"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">dC</figure-callout>}}_4}{\mathrm{dt}}=k_3{C}_3-k_4{C}_4\\ r_5=\frac{{\mathrm{dC}}_5}{\mathrm{dt}}{k}_4{\mathrm{<figure-callout\; id="4"\; label="C"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">C</figure-callout>}}_4\end{array}\right.$

$\left\{\begin{array}{c}{k}_1=0.0007+0.00038x_1-0.23x_2+0.31x_3+0.36x_4+0.48x_5+(1.06E-05)x_1^2+0.035x_2^2+0.026x_3^2+0.028x_4^2+0.046x_5^2\\ k_2=-0.0012+0.0052x_1-0.35x_2+0.35x_3+0.3x_4+0.37x_5+(6.81E-06)x_1^2+0.042x_2^2+0.026x_3^2+0.021x_4^2+0.33x_5^2\\ k_3=0.00021+0.0013x_1-0.039x_2+0.037x_3+0.033x_4+0.048x_5-(2.92E-06)x_1^2+0.0068x_2^2+0.0035x_3^2+0.003x_4^2+0.006x_5^2\\ {\mathrm{<figure-callout\; id="4"\; label="k"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">k</figure-callout>}}_4=-0.00064+0.0036x_1-0.32x_2+0.83x_3+0.79x_4+1.34x_5-(6.47E-06)x_1^2+0.054x_2^2+0.083x_3^2+0.074x_4^2+0.17x_5^2\end{array}\right.$

Then, at given each key reaction factor value, i.e. temperature of reaction (x ₁, ℃), ratio of solvent (x ₂, mol/Kg), concentration of cobalt ions (x ₃, ppmw), manganese ion concentration (x ₄, ppmw), bromide ion concentration (x ₅, value ppmw) just can be calculated the reaction rate constant that respectively goes on foot consecutive reaction through above-mentioned rate constant correlation model; To obtain reaction rate constant substitution PX oxidation reaction industrial device model, promptly formula (6) calculates through numerical integration, just can be in the hope of under given reaction time, and the intermediate product of reactor outlet and the concentration of final product.

If collect the reaction factor of present commercial plant be:

(1) temperature of reaction x ₁=187.79 ℃;

(2) ratio of solvent x ₂=2.0888mol/Kg;

(3) concentration of cobalt ions x ₃=0.037619ppmw;

(4) manganese ion concentration x ₄=0.01794ppmw;

(5) bromide ion concentration x ₅=0.077788ppmw;

Reaction time is: t=70.56min;

Then, each goes on foot the consecutive reaction rate constant and is:

$\left\{\begin{array}{c}{k}_1=0.1808\\ k_2=0.7084\\ k_3=0.0934\\ {\mathrm{<figure-callout\; id="4"\; label="k"\; filenames="926185DEST\_PATH\_H200710047619820080121E000011.png"\; state="\{\{state\}\}">k</figure-callout>}}_4=0.1660\end{array}\right.$

PX oxidation reaction industrial device model based on above foundation calculates the following result of acquisition:

$\left\{\begin{array}{c}{\tilde{C}}_1=0\mathrm{mol}/\mathrm{Kg}\\ {\tilde{C}}_2=2.4E-18\mathrm{mol}/\mathrm{Kg}\\ {\tilde{C}}_3=0.0059\mathrm{mol}/\mathrm{Kg}\\ {\tilde{C}}_4=0.0075\mathrm{mol}/\mathrm{Kg}\\ {\tilde{C}}_5=2.062\mathrm{mol}/\mathrm{Kg}\end{array}\right.$

This example calculation value is compared with actual industrial analytical test value, and error is less than 5%.In the actual industrial device was used, the model fitting error can be less than 10%, and the model fitting error can reach about 5% usually, and therefore, PX oxidation industrial device model has very strong practical application in industry and is worth.

More than, the modeling method of PX oxidation industrial device model is described through instance.

Claims (4)

1. the modeling method of a p xylene oxidation reaction industrial device model is characterized in that, this modeling method comprises following steps:

(1) adopt from P-xylene, p-tolyl aldehyde, p-methylbenzoic acid, to carboxyl benzaldehyde, to the consecutive type reaction network of final products terephthalic acid (TPA), set up following p xylene oxidation reaction industrial device model framework:

$\left\{\begin{array}{c}{r}_1=\frac{{\mathrm{dC}}_1}{\mathrm{dt}}=-k_1{C}_1^{n_1}{C}_{O_2}^{m_1}\\ r_2=\frac{{\mathrm{dC}}_2}{\mathrm{dt}}=k_1{C}_1^{n_1}{C}_{O_2}^{m_1}-k_2{C}_2^{n_2}{C}_{O_2}^{m_2}\\ r_3=\frac{{\mathrm{dC}}_3}{\mathrm{dt}}=k_2{C}_2^{n_2}{C}_{O_2}^{m_2}-k_3{C}_3^{n_3}{C}_{O_2}^{m_3}\\ r_4=\frac{{\mathrm{dC}}_4}{\mathrm{dt}}{=k}_3{C}_3^{n_3}{C}_{O_2}^{m_3}-k_4{C}_4^{n_4}{C}_{O_2}^{m_4}\\ r_5=\frac{{\mathrm{dC}}_5}{\mathrm{dt}}=k_4{C}_4^{n_4}{C}_{O_2}^{m_4}\end{array}\right.$

Wherein,

r ₁, r ₂, r ₃, r ₄, r ₅Be respectively P-xylene, p-tolyl aldehyde, p-methylbenzoic acid, to the reaction rate of carboxyl benzaldehyde, terephthalic acid (TPA);

T is a reaction time, and unit is min;

n ₁, n ₂, n ₃, n ₄For P-xylene, p-tolyl aldehyde, p-methylbenzoic acid, to the order of reaction of carboxyl benzaldehyde;

m ₁, m ₂, m ₃, m ₄Be oxygen respectively go on foot consecutive reaction progression;

k ₁, k ₂, k ₃, k ₄For consecutive reaction respectively goes on foot pairing rate constant;

C ₁, C ₂, C ₃, C ₄, C ₅, CO ₂For P-xylene, p-tolyl aldehyde, p-methylbenzoic acid, to the concentration of carboxyl benzaldehyde, terephthalic acid (TPA), oxygen relative unit weight solvent acetic acid, unit is mol/Kg;

(2) set up each key reaction factor, i.e. temperature of reaction x through multicomponent linear regressioning technology ₁, unit is ℃; Ratio of solvent x ₂, unit is mol/Kg; Concentration of cobalt ions x ₃, unit is ppmw; Manganese ion concentration x ₄, unit is ppmw; Bromide ion concentration x ₅, unit is ppmw, and the high-order term of reaction factor

With the correlation model of each step consecutive reaction rate constant, promptly

$\left\{\begin{array}{c}{k}_i={\mathrm{\&beta;}}^{\left(i\right)}{x}^T\\ {\mathrm{\&beta;}}^{\left(i\right)}=[{\mathrm{\&beta;}}_0^{\left(i\right)}{\mathrm{\&beta;}}_1^{\left(i\right)}{\mathrm{\&beta;}}_2^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_5^{\left(i\right)}{\mathrm{\&beta;}}_6^{\left(i\right)}{\mathrm{\&beta;}}_7^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_{10}^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_{5m-4}^{\left(i\right)}{\mathrm{\&beta;}}_{5m-3}^{\left(i\right)}\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&beta;}}_{5m}^{\left(i\right)}]\\ x=[1x_1{x}_2\&CenterDot;\&CenterDot;\&CenterDot;x_5{x}_1^2{x}_2^2\&CenterDot;\&CenterDot;\&CenterDot;x_5^2\&CenterDot;\&CenterDot;\&CenterDot;x_1^m{x}_2^m\&CenterDot;\&CenterDot;\&CenterDot;x_5^m]\end{array}\right.$

Wherein,

β (i) is k _iThe regression coefficient vector;

X by 1, the vector formed of the high-order term of above-mentioned reaction factor and above-mentioned reaction factor;

(3) directly based on the commercial plant production data, try to achieve above-mentioned β through the intelligent optimization algorithm optimizing ⁽ⁱ⁾, try to achieve above-mentioned k through above-mentioned steps (2) then _i, the final p xylene oxidation reaction industrial device model of describing the commercial plant characteristic of setting up, the content of intermediate product and final product in the monitoring reaction course in real time.

2. the modeling method of p xylene oxidation reaction industrial device model according to claim 1 is characterized in that, said order of reaction n ₁=0.65, n ₂=1, n ₃=1, n ₄=1, m ₁=0, m ₂=0, m ₃=0, m ₄=0.

3. modeling method according to claim 1 and 2 is characterized in that, comprises a following determining step (2#) in the said step (2):

Judge that whether total error E that computes obtains is than preset value ε _AllLittle, if then identification finishes, and tries to achieve said β ⁽ⁱ⁾Otherwise, adjust search through intelligent optimization algorithm, recomputate:

$E_j=\left\{\begin{array}{cc}\underset{i=1}{\overset{5}{\mathrm{\&Sigma;}}}\left|\frac{{\tilde{C}}_i^{\left(j\right)}-C_i^{\left(j\right)}}{C_i^{\left(j\right)}}\right|& ,\mathrm{If}\underset{i=1}{\overset{5}{\mathrm{\&Sigma;}}}\left|\frac{{\tilde{C}}_i^{\left(j\right)}-C_i^{\left(j\right)}}{C_i^{\left(j\right)}}\right|\&GreaterEqual;\mathrm{\&epsiv;}\\ 0& ,\mathrm{If}\underset{i=1}{\overset{5}{\mathrm{\&Sigma;}}}\left|\frac{{\tilde{C}}_i^{\left(j\right)}-C_i^{\left(j\right)}}{C_i^{\left(j\right)}}\right|<\mathrm{\&epsiv;}\end{array}\right.,$ J=1,2 ..., n wherein, 1＞ε＞0;

Total error $E=\frac{1}{n}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{E}_j$

Wherein, j representes j sample point;

The assay value of representing i intermediate product of j sample point;

The Model Calculation value of representing i intermediate product of j sample point; E _jIt is the Model Calculation error of j sample point; ε is each sample point error of calculation limit, if the Model Calculation error is worth less than this, then is not counted in total error.

4. modeling method according to claim 3 is characterized in that, said ε scope is 0.01≤ε≤0.05; The preset value ε of said total error E _AllScope is 0.05≤ε all≤0.15.

Images