CN102117365A - On-line modeling and optimizing method suitable for recovering coking coarse benzene - Google Patents

Abstract

The invention discloses an on-line modeling and optimizing method suitable for recovering coking coarse benzene, and belongs to the field of recovery of coking chemical products. The on-line modeling and optimizing method is characterized by providing a method for one-line modeling and multi-object optimization on coarse benzene recovery. In the invention, a coarse benzene recovery process flow model is established by combining mechanism analysis, multiple regression and system identification modeling, so that the behavior of the system can be accurately described and the requirement for a detection device is reduced; and the coarse benzene recovery multi-object function as an object is provided and the solving is carried out by adopting a particle swarm optimization algorithm so as to obtain the optimized variable of the operation process. The on-line modeling and optimizing method has the advantages of high speed and high efficiency. In addition, by setting an adjustable object weight, the requirement for people on manual adjustment on the specific gravity of single object can be met. The excellent performances of the on-line modeling and optimizing method can be proved through actual production data.

Description

A kind of line modeling and optimization method that is applicable to that coking crude benzene reclaims

Technical field

The present invention relates to crude benzol recovery line modeling and optimization method that a kind of coking chemistry product reclaims the field, specifically is that a kind of crude benzol that is applicable to reclaims the method and the software realization thereof of line modeling and multiple-objection optimization.

Background technology

Along with national energy-saving and emission-reduction work is deeply implemented, resource and environmental pressure that coking industry faces are more and more outstanding.Industry restructuring eliminates the backward production facilities, and improves pollution control, and the road of walking sustainable development is the direction that coking industry must be adhered to for a long time and make great efforts.Under this background, based on current coking production technology and equipment, discharging, cut down the consumption of energy with green production, minimizing is target, in conjunction with practical condition, optimizes process control procedure model and parameter, will be a significant practice.Coking crude benzene is an important chemical material, and it is the important production link in coke-oven plant that crude benzol reclaims, and optimizes the crude benzol reclaimer operation, has key effect to enhancing productivity, save energy and reduce the cost and improving the cleaner production level.

Optimization research for coking crude benzene recovery technology at present also rests in the summary of experience of production management and plant maintenance basically, does not have the comparatively theoretical analysis and the system summary of system.The quality of equipment operation situation is bigger to every index influence that crude benzol reclaims in the production run, and operator's experience also plays certain positive role in addition, but this can only solve subproblem.Realize that it is the important means that promotes recovery levels that crude benzol reclaims the continuous on-line optimization of technological process.

Wu Xue etc. delivered the article of a piece " improving the modification measures of crude benzol output " by name in 2008 at " fuel and chemical industry " periodical.Article does not reach the design production problems at changing the product recovery, has analyzed concrete reason, and has tested.Article is according to actual production situation, and it is on the low side and ature of coal is relatively poor that the influence factor that sums up includes stove coal volatile matter; It is bigger to go into the fluctuation of tubular furnace vapor pressure; It is on the low side to go into to wash the oil-poor flow of benzene tower, about 60m ³/ h; Condensate cooler outlet benzene temperature degree is higher, about 38 ℃; Gas temperature is higher behind the final cooling tower, about 31 ℃; Oil-poor cooler outlet temperature drift, about 35 ℃; It is on the low side to go into crude benzol workshop section water at low temperature flow, about 60m ³/ h, pressure is on the low side, about 0.2MPa.Taked corresponding modification measures at above reason, comprised and formulate reasonable blending ratio; Transform oil-poor pipeline, improve oil-poor flow to 80m ³More than/the h; Transform condensate cooler and low temperature water system.By above transformation, this factory's crude benzol reclaims operations and reaches technological requirement, and the crude benzol recovery is increased to 1%, has brought favorable economic benefit.This is foundation to reclaim technology, is to optimize the main mode that crude benzol reclaims in conjunction with actual production situation to equipment and operating conditions transformation, this mode does not relate to process mechanism, do not need to set up crude benzol removal process mathematical model, only need to understand the parameter regulation and the equipment operation situation of institute's adopting process, technical less demanding.

He Xuejun etc. 2006 have delivered the article of a piece " utilizing CASE(Computer Aided Software Engineering) to optimize crude benzol production " by name at " fuel and chemical industry " periodical, have pointed out the directive significance of assistant software to production operation.Crude benzol recovery computer assisted procedure software not only comprises analog computation, performance analysis, operating mode prediction, the calculating of control setting value and the assessment of tower condition of unit and total system, fault diagnosis and analysis, equipment control, shutting down guiding, technical examination training, the benzene of equipment belongs to physicochemical property inquiry and every systemic-functions such as technical literature and data check of hydro carbons and coal gas, and crude benzol reclaimer operation economic evaluation system is provided.But generally have only working condition, status of equipment or the market price that bigger variation takes place, when process economy is produced obviously influence, just be necessary to be optimized adjustment.Because when adjusting, system condition will be from a state variation to another state, and whole process of production all needs to experience transition period, makes to produce to play pendulum.And the non-steady state of producing tends to lost units, and easily has an accident.

Optimizing research work mainly is to operate to adjust and demarcate.Seek optimum cycle washing oil amount (decision variable), system condition should be stepping, and operation adjustment needs proceed step by step, to make every effort to safe reliable, avoids accident, and the harmful effect of producing is reduced to minimum level.Optimizing regulating step mainly comprises:

1). the adjustment of circulating washing oil amount.Divide two stages to carry out, the phase one eases down to 160 by existing 180, the about 1.7L/m of oil-gas ratio ³Subordinate phase eases down to 135m ³/ h, the about 1.4L/m of oil-gas ratio ³

2). the adjustment of tubular furnace.When the circulating washing oil amount was adjusted, not only the rich oil outlet temperature did not change, and steam exit temperature also should remain unchanged.When the circulating washing oil amount reduces,, for guaranteeing the quality of washing oil, need appropriate air inlet or the flue turnover panel of adjusting tubular furnace, to elongate flame if steam exit temperature reduces.

3). the adjustment of oil-poor refrigeratory chilled water.Oil-poor to go into tower temperature-resistant in order to keep, and suitably turns down cooling water valve, reduces cooling water amount.

4). add the demarcation of heating gas, chilled water, open steam, pump electric current and oil washing consumption.Coal gas and steam have the measured value of recorder, and each pump electric current can draw according to washing oil storage in the system and the variation of washing oil tank liquor position with the card table value of reading, washing oil, and cooling water inflow is come out by heat Balance Calculation by software.

According to above regulating step, find to keep under the constant substantially situation of crude benzol output through test, the circulation oil absorption, add heating gas and cooling water inflow obviously reduces, obtained favorable economic benefit.

The characteristics of this optimization method are to use crude benzol to reclaim computer assisted procedure software, according to working condition, status of equipment or the market price etc. crude benzol are reclaimed to be optimized, and adjust relevant device and operation.Owing to be to be optimized according to computer assisted procedure software, its result has reduced artificial factor.But finish in case regulate, in longer a period of time, can not carry out adjustment once more, can not reach crude benzol and produce total optimization.A kind of line modeling and optimization method that is applicable to that coking crude benzene reclaims that the present invention proposes, set up the modeling of crude benzol removal process according to real-time field data and historical data, and with energy-saving and cost-reducing, cleaner production with to enhance productivity be integration objective, the timing optimization performance variable, obtain the Optimizing operation variable information, for operating personnel with reference to utilization.

Summary of the invention:

Purpose of the present invention and the technical matters that will solve are with energy-saving and cost-reducing, cleaner production and to enhance productivity be integration objective, and the timing optimization coking crude benzene reclaims the technological process performance variable.This method is set up crude benzol according to on-the-spot real time data and historical data and is reclaimed the technological process model, and utilizes particle cluster algorithm to find the solution multiple objective function, obtains the Optimizing operation variable information, for operating personnel with reference to utilization.

Technical scheme of the present invention comprises that mainly coking crude benzene reclaims online process model building of technology and online multiple-objection optimization.

Reclaim the technology online process model building stage at crude benzol, reclaim production run real time data and historical data according to the crude benzol of gathering, adopt the method for Analysis on Mechanism, multiple regression and System Discrimination that crude benzol recovery each several part is carried out modeling respectively, with math equation accurate description crude benzol removal process.In application, along with the continuous adding of new data,, old historical data is upgraded according to the thought of rectangular window, it is up-to-date to guarantee to rebulid the used data of model per half an hour, and the used data total amount of modeling remains unchanged.

In the multiple-objection optimization stage, according to the updated model of having set up, with energy-saving and cost-reducing, cleaner production with to enhance productivity be integration objective, adopt particle cluster algorithm solving-optimizing target function, obtain the Optimizing operation variable information, comprising: go out tubular furnace rich oil pipe temperature, debenzolizing tower top temperature, debenzolizing tower bottom pressure, hot oil-poor flow and temperature.Experiment showed, that crude benzol reclaims process optimization and can obtain desirable effect.

Technical scheme of the present invention is: a kind of line modeling and optimization method that is applicable to that coking crude benzene reclaims specifically may further comprise the steps:

At first, according to washing the modeling of benzene process mechanism, tubular furnace heating rich oil process multiple regression modeling and taking off benzene procedures system identification modeling, the regularly online crude benzol that acquires reclaims technological process high precision integrated model;

Inferior, is target according to the multiple-objection optimization part with production efficiency, energy-saving consumption-reducing and cleaner production, with debenzolizing tower top temperature, debenzolizing tower bottom pressure, to go out tubular furnace rich oil temperature, hot oil-poor flow and temperature be optimization variable, adopt particle swarm optimization algorithm regularly to find the solution objective function, at the set-point of each optimization variable of line computation, thus the optimization of realization modeling process.

Further, the described benzene process mechanism modeling process of washing is: the crude benzol recycle section adopts whole cold wash benzene process.Whole cold wash benzene process process is that the coal gas from ammonium sulfate workshop section comes enters final cooling tower, is cooled to 25～28 ℃ at this; Then by washing the benzene tower by two successively at the bottom of the tower.Oil-poorly extract out from oil-poor groove with pump, deliver to and wash the benzene cat head and spray, contact with the coal gas that comes up at the bottom of the tower is reverse at Ta Nei, the benzene hydrocarbon in the coal gas is absorbed by washing oil.Washing benzene is the mass transport process that absorbs crude benzol in the coal gas according to the similar compatibility principle with the tar washing oil.

Crude benzol reclaims and to belong to mass transfer and detachment process, defers to law of conservation of mass, heat conservation law and vapour-liquid basic theories such as theory mutually.Washing benzene is the mass transport process that absorbs crude benzol in the coal gas according to the similar compatibility principle with the tar washing oil, obey Henry's law and Raoult's law, the dividing potential drop of crude benzol is obeyed Dalton's law in the coal gas, when the lean solution of washing oil absorption crude benzol gained can be considered ideal solution, the equilibrium vapour pressure of its crude benzol was obeyed Raoult's law.When the dividing potential drop of crude benzol in the coal gas greater than the washing oil liquid level on during the crude benzol equilibrium vapour pressure, the crude benzol in the coal gas is promptly absorbed by washing oil, both differ bigger, absorption process heal easily, absorption rate also heals soon.The limit that washing oil absorbs crude benzol is that both equate, this moment, gas-liquid two-phase reached equilibrium state, and absorbing expulsive force is zero.

Go out to wash in the benzene tower coal gas that the balanced type between the benzene hydrocarbon content c (mass percent) is expressed as among the benzene hydrocarbon content a and washing oil:

$c=\frac{2.24*a*p}{1.25*\mathrm{Mm}*{\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000048963620000022.png"\; state="\{\{state\}\}">p</figure-callout>}}^0}$

P wherein ⁰Saturated vapour pressure for crude benzol under the recovered temperature.P is for going out the tower gas pressure.Mm is the relative molecular mass of tar washing oil.Wherein absorption being influenced bigger factor mainly contains: wash medial temperature, the washing oil of gas-liquid two-phase surface of contact in the benzene tower receptivity and circulating washing oil amount, circulating washing oil benzene content, wash benzene inner-tower filling material surface area, coal gas general pressure and temperature and flow velocity etc.Medial temperature during absorption can be represented by the formula:

$T=\frac{a1*(\mathrm{FT}\_305)*(\mathrm{TI}\_305)+a2*(\mathrm{FT}\_301)*(\mathrm{TI}\_311)}{a1*(\mathrm{FT}\_305)+a2*(\mathrm{FT}\_301)}$

In the formula, FT_305 is hot oil-poor flow, and YI_305 is for going out the oil-poor temperature of refrigeratory, and FT_301 is for going out to wash benzene tower gas flow, and TI_311 is for going into to wash benzene tower gas temperature, and a1 is oil-poor with heat with a2 and goes out to wash the relevant coefficient of benzene tower coal gas rerum natura.

The saturated vapor pressure size of crude benzol can obtain by the size inquiry of contact temperature.

Because washing oil is constantly circulation in recycle section, has continuity.Therefore in continuous half-hour period, can think not variation of crude benzol content in the washing oil.

Further, described tubular furnace multiple regression modeling:

Tubular furnace heating process influence factor is less relatively, adopts the multiple regression modeling to meet the demands, and Model Calculation speed is fast, helps being optimized finding the solution.Regression modeling is under the condition by methods such as sampling or experiment acquisition data, asks for one according to these data and can represent the mathematical expression form that concerns between them.If the output of model is subjected to the influence of a plurality of variablees simultaneously, then be called multivariate regression model.The essence of multiple regression modeling is used mathematical statistic method exactly, the measurement data mirror image is analyzed and handled, thereby obtain the experimental formula of mutual relationship between the reflection variable, its subject matter that will solve is how to go to estimate each parameter in the regression model according to the sample observations of variable.

To given data point (X _i, Y _i) _{I=1,2, Λ m}, the regression function that set up with sample information can go to determine to set up the criterion of sample regression function from different perspectives as much as possible near estimating PRF.When residual sum of squares (RSS) minimum with output true value and best estimate, when promptly least square is principle, establish φ for all number of times are no more than the function class of the polynomial expression formation of n, be exactly therefrom definite one:

$p_n\left(X\right)=\underset{k=0}{\overset{n}{\mathrm{\&Sigma;}}}{a}_k{X}^k\&Element;\mathrm{\&phi;}$

Make:

$I=\min \left\{\underset{i=0}{\overset{m}{\mathrm{\&Sigma;}}}{\left[p_n(X_i-Y_i)\right]}^2\right\}$

The input and output system of selection of regression model has two kinds: experience back-and-forth method and Model Selection method.On experience and basis of mechanism analysis, adopt the way of Model Selection to carry out the final definite of model input variable type and order.Because when being mixed with unnecessary influence factor in the equation, will cause precision of prediction to reduce.Each explanatory variable satisfies absence of multicollinearity supposition, i.e. linear independence between the observed reading of each explanatory variable.

Lot of experiment results shows: the multilinear fitting model can reflect field condition well, accurately simulates whole process of production.The matrix form of polynary model of fit can be expressed as:

$\left[\begin{array}{c}{Y}_1\\ Y_2\\ M\\ Y_n\end{array}\right]=\left[\begin{array}{ccccc}1& X_{21}& X_{31}& \mathrm{\&Lambda;}& X_{k1}\\ 1& X_{22}& X_{32}& \mathrm{\&Lambda;}& X_{k2}\\ M& M& M& & M\\ 1& X_{2n}& X_{3n}& \mathrm{\&Lambda;}& X_{\mathrm{kn}}\end{array}\right]\left[\begin{array}{c}{\mathrm{\&beta;}}_1\\ {\mathrm{\&beta;}}_2\\ M\\ {\mathrm{\&beta;}}_k\end{array}\right]$

N group observations (X in the formula _i, X _2i, X _3i, Λ, X _Ki) (i=1,2, Λ is explained variable Y and a plurality of explanatory variable to be done n observation station get n).β _i(i=1,2 ..., k) being the parameter of model, k-1 is the number of explanatory variable.In multiple linear regression model, regression coefficient β _i(i=1,2 ..., k) expression is to work as under the constant condition of other explanatory variable of control, the unit change of i explanation is to the influence of explained variable mean value.

In the overall linear regression function, each regression coefficient is unknown, can only utilize sample observations that it is estimated.Usually adopt least square method to seek the estimator of β

Promptly seek the estimation of β

Meet the following conditions:

$\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{(Y_i-\underset{j=0}{\overset{m}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ij}}{\widehat{\mathrm{\&beta;}}}_j)}^2=\min \underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{(Y_i-\underset{j=0}{\overset{m}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ij}}{\widehat{\mathrm{\&beta;}}}_j)}^2$

Here x _I0=0 (i=1,2, Λ n), or is write as matrix form

${\left|\right|Y-X\widehat{\mathrm{\&beta;}}\left|\right|}^2=\min {\left|\right|Y-\mathrm{X\&beta;}\left|\right|}^2$

The general available differential method is found the solution

$\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{(Y_i-\underset{j=0}{\overset{m}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ij}}{\widehat{\mathrm{\&beta;}}}_j)}^2{x}_{\mathrm{ik}}=0(k=\mathrm{0,1},L,m)$

Following formula can be written as:

$\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{Y}_i{x}_{\mathrm{ik}}=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{m}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ij}}{x}_{\mathrm{ik}}{\widehat{\mathrm{\&beta;}}}_j=\underset{j=0}{\overset{m}{\mathrm{\&Sigma;}}}\left(\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ij}}{x}_{\mathrm{ik}}\right){\widehat{\mathrm{\&beta;}}}_j(k=\mathrm{0,1},L,m)$

Use matrix representation, following formula can be written as:

$X^{T}Y=\left(X^{T}X\right)\widehat{\mathrm{\&beta;}}$

This is a normal equations, can get:

$\widehat{\mathrm{\&beta;}}={\left(X^{T}X\right)}^{-1}{X}^{T}Y$

Will

Bring equation of linear regression into, can get:

$\hat{Y}={\widehat{\mathrm{\&beta;}}}_0+{\widehat{\mathrm{\&beta;}}}_1{x}_1+{\widehat{\mathrm{\&beta;}}}_2{x}_2+L+{\widehat{\mathrm{\&beta;}}}_m{x}_m$

The sample conditional mean of explained variable

With actual observed value Y _iBetween deviation, promptly residual term is e _i, multivariate sample linear regression function can be expressed as:

$Y_i={\hat{Y}}_i+e_i$

According to electronic data and analysis data analysis to collecting, crude benzol removal process multiple regression modeling.The input/output variable of model is determined in conjunction with the collection in worksite data analysis that by Analysis on Mechanism the tubular furnace input includes stove purified gas pipe pressure, goes into tubular furnace vapor pressure and rich oil pipe flow, and output includes the regenerator vapor (steam) temperature and goes out tubular furnace rich oil temperature.Model form is expressed as with math equation:

y ₁＝b ₁₀+b ₁₁x ₁+b ₁₂x ₂+b ₁₃x ₃

y ₂＝b ₂₀+b ₂₁x ₁+b ₂₂x ₂+b ₂₃x ₃

Y wherein ₁For going into the regenerator vapor (steam) temperature; y ₂Be the rich oil temperature of coming out of the stove.x ₁For going into stove purified gas pipe pressure; x ₂For going into the tubular furnace vapor pressure; x ₃Be rich oil pipe flow; b _IjBe corresponding regression coefficient.

The described benzene procedures system identification modeling of taking off:

It is more to take off benzene process influence factor, and model internal mechanism complexity adopts the method for on-line identification could obtain the high precision model, and satisfies the system real time requirement.Resulting input and output data are tested in the system identifying method utilization, according to selected principle, determine a model that meets preferably with the institute examining system from model I.It is more to take off benzene process influence factor, model internal mechanism complexity, so we adopt the method for System Discrimination to carry out modeling to taking off the benzene process, simultaneously in order to satisfy the system real time requirement, this project is utilized the method for on-line identification.The main contents of System Discrimination have the following aspects.

1), identification purpose

According to difference, also different to modeling demand to the system model application scenario.For example, the model that the check and the fault detection and diagnosis of theoretical model parameter are used then requires to build more accurately.Using system discrimination method of the present invention is set up the model of energy simulated real system, behavior, and then as the basis of optimization system performance variable setting value, therefore model need truly reflect the characteristic of system, and input and output must be able to reflect the influence of system variable setting value to process.

2), priori

Priori is the general designation of before carrying out identification model system's mechanism and operating conditions, modeling purpose etc. being understood.Here comprise that crude benzol reclaims technology, device therefor major parameter, historical data and expertise etc.Wherein crude benzol recovery technology and device therefor situation have been done introduction in Analysis on Mechanism, and historical data and expertise are mainly obtained from existing record of production and field staff.

3), experimental design

Identification is a process of extracting relevant system information from experimental data, the basis of identification is the input and output data, and data from experiment and detection to system, therefore the data of identification experiment and observation should have more sample number, the behavioral trait that can comprise system reflects the useful information of system nature's characteristic.According to the analysis to data with existing, it all can satisfy modeling demand on quality and quantity, also can draw same conclusion on model result.

The model of System Discrimination not only should the current model running situation of accurate response, and good estimated performance will be arranged.The quantity of appropriate selection training sample is crucial to reaching good estimated performance, because less lack of training samples is with the characteristic of complete reaction system model, and because the native system model is a time-varying model, too much training sample can incorporate too much historical information, can not appropriately react current system model characteristic.

4), model class other determine

Be to determine the model classification, need be testing before do necessary supposition on the basis of knowledge, i.e. the form that embodies of definite system mathematic model.According to analyzing recovery system characteristics and data with existing, the crude benzol recovery system belongs to the steady chemical process that time lag is arranged, and adopts the form of nonlinear difference equation to carry out model representation, the deterministic permanent parameter model of thing in a time period, experiment simultaneously shows that third-order model is comparatively suitable.

5), parameter estimation

After the model classification is determined, need to utilize the parameter of recursive least-squares rectangular window method criterion estimation model according to inputoutput data.Experiment shows that the recursive least-squares criterion has degree of precision, and algorithm robustness height, and complexity is lower, can reflect the intrinsic propesties of system.

6), modelling verification

The model that identification is come out will be tested, and the actual measurement output that is about to output of gained model and system compares, if both differ bigger, then needs to revise the model structure supposition till model meets the demands.Experiment shows that when the crude benzol removal process was taken off benzene identification model employing single order or second order, model accuracy was lower, and the higher complexity of third-order model precision is not high relatively.

In the present invention, use the Recursive Least Squares of rectangular window method to carry out system's time-varying parameter identification.The rectangular window method is a kind of effective time-varying parameter discrimination method.The essence of this method is only limited group of past data of foundation of certain estimation constantly, and in each time period, a new data point increase is come in, and Geju City data point is then disallowable goes out, and has kept the number of data point always to equal N like this.Least square method of recursion apply-official formula is as follows:

Wherein

Expression be carved into during based on i+1 i+N constantly between N observed reading y Λ, y _I+2, Λ, y _NResulting θ estimated value, the optimal function of seeking data by the quadratic sum of minimum error is mated.Existing correlation theory proves that least-squares estimation has consistance and validity.

Debenzolizing tower is carried out the System Discrimination modeling, with eight important process variable as input, comprise debenzolizing tower top temperature, go into the regenerator vapor (steam) temperature, go into debenzolizing tower rich oil flow, crude benzol capacity of returns, rich oil benzene content, go out tubular furnace rich oil temperature, debenzolizing tower bottom pressure and steam in the boiler pressure, crude benzol output is as output.The model input and output are expressed as with math equation:

${\left(\begin{array}{c}1\\ a_1\\ a_2\end{array}\right)}^T\left(\begin{array}{c}y\left(k\right)\\ y(k-1)\\ y(k-2)\end{array}\begin{array}{}\end{array}\right)={\left(\begin{array}{c}{B}_{11}\\ B_{12}\\ B_{13}\\ B_{14}\\ B_{15}\\ B_{16}\\ B_{17}\\ B_{18}\end{array}\right)}^T\left(\begin{array}{c}{U}_1\left(k\right)\\ U_2\left(k\right)\\ U_3\left(k\right)\\ U_3\left(k\right)\\ U_5\left(k\right)\\ U_6\left(k\right)\\ U_7\left(k\right)\\ U_8\left(k\right)\end{array}\right)+{\left(\begin{array}{c}{B}_{21}\\ B_{22}\\ B_{23}\\ B_{24}\\ B_{25}\\ B_{26}\\ B_{27}\\ B_{28}\end{array}\right)}^T\left(\begin{array}{c}{U}_1(k-1)\\ U_2(k-1)\\ U_3(k-1)\\ U_4(k-1)\\ U_5(k-1)\\ U_6(k-1)\\ U_7(k-1)\\ U_8(k-1)\end{array}\right)+{\left(\begin{array}{c}{B}_{31}\\ B_{32}\\ B_{33}\\ B_{34}\\ B_{35}\\ B_{36}\\ B_{37}\\ B_{38}\end{array}\right)}^T\left(\begin{array}{c}{U}_1(k-2)\\ U_2(k-2)\\ U_3(k-2)\\ U_4(k-2)\\ U_5(k-2)\\ U_6(k-2)\\ U_7(k-2)\\ U_8(k-2)\end{array}\right)$

In the formula, y (k) and u (k) are output of current system and input, and y (k-1), y (k-2), u (k-1) and u (k-2) are historical output of system and input, a _iAnd B _IjBe coefficient to be identified, i=1,2,3, j=1...8.

Described multiple-objection optimization step is as follows:

Choose optimization variable

The optimization of crude benzol removal process control is a multiple goal, non-linear, constrained optimization problem.This optimization problem can be expressed as follows:

$\underset{X\&Element;R}{\min }F\left(X\right)$

F(X)＝(f ₁(X)，f ₂(X)，f ₃(X)) ^T

R＝{X|g(X)≤0}，g(X)＝(g ₁(X)，g ₂(X)) ^T

$X={(x_1,x_2,x_3,x_4,x_5)}^T,X\&Element;R\&Subset;E^n$

Wherein F (X) is the optimization aim vector, and g (X) is a constrained vector, and X is decision variable, i.e. optimization variable.The optimization variable of choosing must be able to embody has certain influence to optimization aim, reclaims process characteristic and model according to crude benzol, and determining tubular furnace rich oil pipe temperature, debenzolizing tower top temperature, debenzolizing tower bottom pressure, hot oil-poor flow and temperature is optimization variable.Reclaim at crude benzol on the basis of technological process integrated model, with suitable mathematical form production efficiency, energy-saving consumption-reducing and three targets of cleaner production are described, adopt particle swarm optimization algorithm regularly to find the solution the set-point of objective function in each optimization variable of line computation, formula is as follows:

v _ij(t+1)＝w×v _ji(t)+c ₁×rand()×(p _ij(t)-x _ij(t))+c ₂×rand()×(p _ij(t)-x _ij(t))

x _ij(t+1)＝x _ij(t)+v _ij(t+1)

Wherein, subscript " i " expression particulate i, i=1,2...s, s are the sums of particle in this colony; The j dimension of " j " expression particulate, t represents t generation, c ₁, c ₂Be the study factor, rand () represents random function, and value is between 0 and 1, and w represents inertia weight; v _IjExpression particulate i j dimension variable change speed, x _IjExpression particulate i j dimension variable present position; The desired positions that all particulates lived through in the colony is p, is called overall desired positions, finds the solution to draw value after the variable optimization, target information.

Utilize particle cluster algorithm to find the solution objective function

Particle swarm optimization algorithm is based on one of algorithm of swarm intelligence.Be the effective technology that solves complicated optimum problem, have stronger global convergence ability and robustness.

It is as follows that the application particle swarm optimization algorithm is found the solution crude benzol recovery technological process multiple objective function step:

1), the encoding scheme of problem identificatioin

Separating from solution space of problem is mapped to the representation space with certain structure, promptly shows that with specific sign indicating number string list separating of problem is called encoding scheme.Select suitable coding method to produce directly influence to the performance and the solving result of algorithm.Crude benzol reclaims the technological process model and has continuous fundamental function, adopts real number vector coding scheme, need not special conversion, can directly find the solution on Problem Areas.

2), determine the evaluation function of optimization problem

In solution procedure, the quality of separating by means of adaptive value evaluation.Therefore must select suitable objective function to calculate fitness according to the concrete feature of problem.According to the characteristics of crude benzol removal process model, each single goal represents to adopt different forms.

The target of cleaner production is mainly reflected in: identically go into to wash benzene tower coal gas and contain under the benzene condition that coal gas benzene hydrocarbon content reduces behind the tower, promptly coal gas contains benzene reduction (f1) and obtains maximal value behind the tower, and mathematical description is:

$\max f1=\frac{a-a^{\&prime;}}{a}$

In the formula, a represents to wash coal gas benzene hydrocarbon content behind the benzene tower; After optimizing, a ' correspondence washes coal gas benzene content value behind the benzene tower.This objective function has reflected reducing pollutant emission, reaching the demand of cleaner production.

The tubular furnace heating needed to consume purified gas during crude benzol reclaimed and produces, and was the important embodiment of system's energy resource consumption.Reduce under the condition that the energy resource consumption objective body satisfies the production demand now and reduce purified gas consumption.Directly gas flow detects and does not still satisfy the requirements, therefore estimate according to the variation of purified gas pressure that according to analyzing as can be known, it is less relatively to go into big more its consumption of tubular furnace purified gas pressure, be that purified gas pressure recruitment (f2) obtains maximal value, mathematical description is:

$\max f2=\frac{{p^{\&prime;}}_1-p_1}{p_1}$

P in the formula ₁Tubular furnace purified gas pressure, p ' are gone in expression ₁The corresponding back variable-value of optimizing.This objective function has reflected reducing the demand of energy resource consumption.

Mainly be in order to obtain the crude benzol product during crude benzol reclaim to be produced, thus production efficiency to improve objective body present: produce more crude benzol product under identical consumption of raw materials, promptly crude benzol output (f3) obtains maximal value, and its mathematical description is:

max?f3＝f(x ₁，x ₂，x ₃，x ₄，x ₅，x ₆，x ₇，x ₈)

In the formula, f3 represents to take off benzene part identification model crude benzol output, x _i(i=1,2, L 8) be the model input variable.This objective function has reflected improving the demand of crude benzol recycle section production efficiency.

The final objective function is the linear weighted function combination of above single goal function, and its basic thought is that a plurality of single goals are converted into single goal problem solving of equal value.Constraint condition is determined according to crude benzol reclaimer operation rules, is comprised performance variable bound and model constrained.The function representation of comprehensive adjustable weight as shown in the formula:

$\max f4=\frac{q_1}{\max f1}f1+\frac{q_2}{\max f2}f2+\frac{q_3}{\max f3}f3$

In the formula, q ₁, q ₂, q ₃The adjustable weight of three single goals of expression correspondence respectively.Experiment finds that the adjusting of weight is bigger to the target influence, and weight has reflected the degree of emphasizing to each single goal.

3), choose controlled variable

The controlled variable of PSO algorithm generally includes maximum algebraically that the scale, algorithm of particulate carry out, inertial coefficient, cognitive parameter, and some other auxiliary controlled variable etc., and suitable controlled variable helps to reduce computing time and improves and optimize performance.

4), the flight model of design particulate

In particle cluster algorithm, the operation of most critical is a speed how to determine particulate.Because particulate is described by multi-C vector, corresponding particulate flying speed also is a multi-C vector.In flight course, particulate is dynamically adjusted flying speed and the direction of oneself by means of self memory and social sharing information along each component direction.Particulate flying speed and position are determined by following formula:

v _ij(t+1)＝w×v _ji(t)+c ₁×rand()×(p _ij(t)-x _ij(t))+c ₂×rand()×(p _ij(t)-x _ij(t))

x _ij(t+1)＝x _ij(t)+v _ij(t+1)

Wherein, subscript " i " expression particulate i, i=1,2...s, s are the sums of particle in this colony.J " represent that the j of particulate ties up, t represents t generation, c ₁, c ₂Be the study factor, rand () represents random function, and value is between 0 and 1, and w represents inertia weight.v _IjExpression particulate i j dimension variable change speed, x _IjExpression particulate i j dimension variable present position.The desired positions that all particulates lived through in the colony is p, is called overall desired positions.

5), determine the stop criterion of algorithm

The most frequently used stop criterion of PSO algorithm is the flight algebraically that preestablishes a maximum, perhaps when the fitness of separating in the search procedure no longer takes place obviously to improve in generation, stops algorithm after continuous what.

6), programing operation is found the solution

Program according to designed algorithm structure, and carry out finding the solution of concrete optimization problem, the accurately fixed and reliability of the validity of the quality verification algorithm by the acquisition problem.

The pseudo-code of PSO algorithm is described as:

1, initialization population scale and dimension, the desired positions that the flight of initialization particulate population position and speed, optimum individual is experienced and, the desired positions of all particulate processes and adaptive value, iterations;

2, calculate the adaptive value of each particulate according to objective function;

3,, upgrade individual optimal location and population optimal location according to adaptive value;

4, upgrade the position and the speed of each particulate;

5, judge whether the particulate fitness satisfies termination condition, is then to withdraw from, otherwise change step 2.

Innovation of the present invention and advantage thereof mainly contain:

(1) at the characteristics of crude benzol recovery technological process different piece, adopt the method for Analysis on Mechanism, multiple regression and System Discrimination to carry out process model building respectively, institute's established model has higher precision, has reached the engineering application requirements.

(2) crude benzol recovery technological process model regularly upgrades, and implementation procedure is followed the tracks of modeling, guarantees the validity of model.

(3) adopt particle swarm optimization algorithm to find the solution crude benzol and reclaim multi-objective problem, obtain the optimization solution of one group of correspondence, for operating personnel's reference, the speed of finding the solution satisfies application requirements.

Description of drawings:

Fig. 1 crude benzol reclaims optimizes the flowage structure block diagram.

Fig. 2 tubular furnace input/output relation and structured flowchart.

Fig. 3 debenzolizing tower System identification model input/output relation synoptic diagram.

The curve of output synoptic diagram of the preceding 120 groups of data institute established models of Fig. 4.

The logic diagram of Fig. 5 particle cluster algorithm flow process.

Embodiment:

Below in conjunction with accompanying drawing the inventive method and system are described in detail.The present invention proposes is applicable to that process flow diagram that coking crude benzene reclaims online modeling optimization method as shown in Figure 1.Concrete implementation step is as follows:

1. modeling part

1.1 wash benzene process mechanism modeling process

Go out to wash benzene hydrocarbon content a (kg/m in the benzene tower coal gas ³) and washing oil in balanced type between the benzene hydrocarbon content c (mass percent) be expressed as:

$c=\frac{2.24*a*p}{1.25*\mathrm{Mm}*{\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000048963620000022.png"\; state="\{\{state\}\}">p</figure-callout>}}^0}$

In the formula: p ⁰Saturated vapour pressure for crude benzol under the recovered temperature.P is for going out the tower gas pressure.Mm is the relative molecular mass of tar washing oil.Wherein absorption being influenced bigger factor mainly contains: wash medial temperature, the washing oil of gas-liquid two-phase surface of contact in the benzene tower receptivity and circulating washing oil amount, circulating washing oil benzene content, wash benzene inner-tower filling material surface area, coal gas general pressure and temperature and flow velocity etc.Medial temperature during absorption can be represented by the formula:

$T=\frac{1872*(\mathrm{FT}\_305)*(\mathrm{TI}\_305)+0.05616*(\mathrm{FT}\_301)*(\mathrm{TI}\_311)}{1872*(\mathrm{FT}\_305)+0.05616*(\mathrm{FT}\_301)}$

In the formula, FT_305 is hot oil-poor flow, and TI_305 is for going out the oil-poor temperature of refrigeratory, and FT_301 is for going out to wash benzene tower gas flow, and TI_311 is for going into to wash benzene tower gas temperature.

When temperature in 25～32 scopes, the saturated vapor pressure of crude benzol can approximate representation be:

P＝0.6605*T-3.77056

1.2 multiple regression modeling

According to electronic data and the analysis data analysis to collecting, multiple regression procedure is adopted in the modeling of crude benzol removal process tubular furnace.The input/output variable of model determines that in conjunction with the collection in worksite data analysis tubular furnace input/output relation as shown in Figure 2 by Analysis on Mechanism.Model form is expressed as with math equation:

y ₁＝b ₁₀+b ₁₁x ₁+b ₁₂x ₂+b ₁₃x ₃

y ₂＝b ₂₀+b ₂₁x ₁+b ₂₂x ₂+b ₂₃x ₃

In the formula: y ₁For going into the regenerator vapor (steam) temperature; y ₂Be the rich oil temperature of coming out of the stove.x ₁For going into stove purified gas pipe pressure; x ₂For going into the tubular furnace vapor pressure; x ₃Be rich oil pipe flow; b _IjBe corresponding regression coefficient.

To be example through tubular furnace and debenzolizing tower model that the pretreated 09 year third quarter, data obtained, all in 1%, the model accuracy height has reached fine modeling effect to its error under 3 σ standards.

1.3 System Discrimination modeling

Debenzolizing tower is carried out the System Discrimination modeling, according to the field data of gathering, with eight important process variable as input, comprise debenzolizing tower top temperature, go into the regenerator vapor (steam) temperature, go into debenzolizing tower rich oil flow, crude benzol capacity of returns, rich oil benzene content, go out tubular furnace rich oil temperature, debenzolizing tower bottom pressure and steam in the boiler pressure, crude benzol output is as output, the model input and output as shown in Figure 3, the type input and output are expressed as with math equation:

${\left(\begin{array}{c}1\\ a_1\\ a_2\end{array}\right)}^T\left(\begin{array}{c}y\left(k\right)\\ y(k-1)\\ y(k-2)\end{array}\begin{array}{}\end{array}\right)={\left(\begin{array}{c}{B}_{11}\\ B_{12}\\ B_{13}\\ B_{14}\\ B_{15}\\ B_{16}\\ B_{17}\\ B_{18}\end{array}\right)}^T\left(\begin{array}{c}{U}_1\left(k\right)\\ U_2\left(k\right)\\ U_3\left(k\right)\\ U_3\left(k\right)\\ U_5\left(k\right)\\ U_6\left(k\right)\\ U_7\left(k\right)\\ U_8\left(k\right)\end{array}\right)+{\left(\begin{array}{c}{B}_{21}\\ B_{22}\\ B_{23}\\ B_{24}\\ B_{25}\\ B_{26}\\ B_{27}\\ B_{28}\end{array}\right)}^T\left(\begin{array}{c}{U}_1(k-1)\\ U_2(k-1)\\ U_3(k-1)\\ U_4(k-1)\\ U_5(k-1)\\ U_6(k-1)\\ U_7(k-1)\\ U_8(k-1)\end{array}\right)+{\left(\begin{array}{c}{B}_{31}\\ B_{32}\\ B_{33}\\ B_{34}\\ B_{35}\\ B_{36}\\ B_{37}\\ B_{38}\end{array}\right)}^T\left(\begin{array}{c}{U}_1(k-2)\\ U_2(k-2)\\ U_3(k-2)\\ U_4(k-2)\\ U_5(k-2)\\ U_6(k-2)\\ U_7(k-2)\\ U_8(k-2)\end{array}\right)$

In the formula, y (k) and u (k) are output of current system and input, and y (k-1), y (k-2), u (k-1) and u (k-2) are historical output of system and input, a _iAnd B _IjBe coefficient to be identified, i=1,2,3, j=1...8.

To the debenzolizing tower modeling, all down all in 5%, the model accuracy height has reached fine modeling effect to the gained model error in 3 σ standards according to above analysis.Accompanying drawing 4 is the curve of output of preceding 120 groups of data institute established models in the test.

2 particle group optimizing parts

The particle swarm optimization algorithm flow process as shown in Figure 5.Particulate population s is set to 20, and j is 5, and maximum iteration time is 5000.c ₁, c ₂Be 0.5, inertia weight w gets 0.9.Weight is regulated q ₁, q ₂, q ₃Difference value 0.1,0.1,0.8.Use the c# programming language to realize optimized Algorithm, find the solution and draw value after the variable optimization, target information.This moment, particulate flying speed and position were shown below:

v _ij(t+1)＝0.9×v _ji(t)+0.5×rand()×(p _ij(t)-x _ij(t))+0.5×rand()×(p _ij(t)-x _ij(t))

x _ij(t+1)＝x _ij(t)+v _ij(t+1)

For example on-the-spot actual operating mode situation: crude benzol period output: 368.638kg; Heat oil-poor flow: 64.48m3/h; The oil-poor temperature of heat: 33.55 ℃; Go out tubular furnace rich oil temperature: 171.95 ℃; Debenzolizing tower bottom pressure: 33.04KPa; Debenzolizing tower top temperature: 95.9 ℃; Go into tubular furnace vapor pressure: 653.48KPa.Parameter is set through after optimizing more than the use, output can reach about 386.5kg, and it is about 5.2% that energy consumption can reduce, and coal gas contains benzene and can reduce about 4.6% behind the tower.

Claims (5)

1. a line modeling and an optimization method that is applicable to that coking crude benzene reclaims is characterized in that,

At first, according to washing the modeling of benzene process mechanism, tubular furnace heating rich oil process multiple regression modeling and taking off benzene procedures system identification modeling, the regularly online crude benzol that acquires reclaims technological process high precision integrated model;

Secondly, is target according to the multiple-objection optimization part with production efficiency, energy-saving consumption-reducing and cleaner production, with debenzolizing tower top temperature, debenzolizing tower bottom pressure, to go out tubular furnace rich oil temperature, hot oil-poor flow and temperature be optimization variable, adopt particle swarm optimization algorithm regularly to find the solution objective function, at the set-point of each optimization variable of line computation, thus the optimization of realization modeling process.

2. line modeling and the optimization method that is applicable to that coking crude benzene reclaims according to claim 1, it is characterized in that, described wash benzene process mechanism modeling process for according to go out to wash in the benzene tower coal gas benzene hydrocarbon quality percentage composition c among the benzene hydrocarbon content a and washing oil than between balanced type be expressed as:

$c=\frac{2.24*a*p}{1.25*\mathrm{Mm}*p^0}$

In the formula: p ⁰Saturated vapour pressure for crude benzol under the recovered temperature; P is for going out to wash benzene tower gas pressure; Mm is the relative molecular mass of tar washing oil;

According to receptivity and circulating washing oil amount, the circulating washing oil benzene content of the medial temperature of washing gas-liquid two-phase surface of contact in the benzene tower, washing oil, wash benzene inner-tower filling material surface area, coal gas general pressure and temperature and flow velocity; Medial temperature during absorption can be represented by the formula:

$T=\frac{1872*(\mathrm{FT}\_305)*(\mathrm{TI}\_305)+0.05616*(\mathrm{FT}\_301)*(\mathrm{TI}\_311)}{1872*(\mathrm{FT}\_305)+0.05616*(\mathrm{FT}\_301)}$

In the formula, FT_305 is hot oil-poor flow, and TI_305 is for going out the oil-poor temperature of refrigeratory, and FT_301 is for going out to wash benzene tower gas flow, and TI_311 is for going into to wash benzene tower gas temperature.

3. line modeling and the optimization method that is used for the coking crude benzene recovery according to claim 1, it is characterized in that, described tubular furnace heating rich oil process multiple regression modeling process is for input includes stove purified gas pipe pressure, goes into tubular furnace vapor pressure and rich oil pipe flow according to tubular furnace, output includes the regenerator vapor (steam) temperature and goes out tubular furnace rich oil temperature, is expressed as with math equation:

y ₁＝b ₁₀+b ₁₁x ₁+b ₁₂x ₂+b ₁₃x ₃

y ₂＝b ₂₀+b ₂₁x ₁+b ₂₂x ₂+b ₂₃x ₃

Y wherein ₁For going into the regenerator vapor (steam) temperature; y ₂Be the rich oil temperature of coming out of the stove; x ₁For going into stove purified gas pipe pressure; x ₂For going into the tubular furnace vapor pressure; x ₃Be rich oil pipe flow; b _IjBe corresponding regression coefficient.

4. line modeling and the optimization method that is used for the coking crude benzene recovery according to claim 1, it is characterized in that, the described benzene procedures system identification modeling of taking off is crossed appellation according to the field data of gathering, with eight important process variable as input, comprise debenzolizing tower top temperature, go into the regenerator vapor (steam) temperature, go into debenzolizing tower rich oil flow, crude benzol capacity of returns, rich oil benzene content, go out tubular furnace rich oil temperature, debenzolizing tower bottom pressure and steam in the boiler pressure, crude benzol output is as output, and input and output are expressed as with math equation:

${\left(\begin{array}{c}1\\ a_1\\ a_2\end{array}\right)}^T\left(\begin{array}{c}y\left(k\right)\\ y(k-1)\\ y(k-2)\end{array}\begin{array}{}\end{array}\right)={\left(\begin{array}{c}{B}_{11}\\ B_{12}\\ B_{13}\\ B_{14}\\ B_{15}\\ B_{16}\\ B_{17}\\ B_{18}\end{array}\right)}^T\left(\begin{array}{c}{U}_1\left(k\right)\\ U_2\left(k\right)\\ U_3\left(k\right)\\ U_3\left(k\right)\\ U_5\left(k\right)\\ U_6\left(k\right)\\ U_7\left(k\right)\\ U_8\left(k\right)\end{array}\right)+{\left(\begin{array}{c}{B}_{21}\\ B_{22}\\ B_{23}\\ B_{24}\\ B_{25}\\ B_{26}\\ B_{27}\\ B_{28}\end{array}\right)}^T\left(\begin{array}{c}{U}_1(k-1)\\ U_2(k-1)\\ U_3(k-1)\\ U_4(k-1)\\ U_5(k-1)\\ U_6(k-1)\\ U_7(k-1)\\ U_8(k-1)\end{array}\right)+{\left(\begin{array}{c}{B}_{31}\\ B_{32}\\ B_{33}\\ B_{34}\\ B_{35}\\ B_{36}\\ B_{37}\\ B_{38}\end{array}\right)}^T\left(\begin{array}{c}{U}_1(k-2)\\ U_2(k-2)\\ U_3(k-2)\\ U_4(k-2)\\ U_5(k-2)\\ U_6(k-2)\\ U_7(k-2)\\ U_8(k-2)\end{array}\right)$

In the formula, y (k) and u (k) are output of current system and input, and y (k-1), y (k-2), u (k-1) and u (k-2) are historical output of system and input, a _iAnd B _IjBe coefficient to be identified, i=1,2,3, j=1...8.

5. line modeling and the optimization method that is used for the coking crude benzene recovery according to claim 1, it is characterized in that, the multiple-objection optimization step is as follows: reclaim at crude benzol on the basis of technological process integrated model, with suitable mathematical form production efficiency, energy-saving consumption-reducing and three targets of cleaner production are described, adopt particle swarm optimization algorithm regularly to find the solution the set-point of objective function in each optimization variable of line computation, formula is as follows:

v _ij(t+1)＝w×v _ji(t)+c ₁×rand()×(p _ij(t)-x _ij(t))+c ₂×rand()×(p _ij(t)-x _ij(t))

x _ij(t+1)＝x _ij(t)+v _ij(t+1)

Wherein, subscript " i " expression particulate i, i=1,2...s, s are the sums of particle in this colony; The j dimension of " j " expression particulate, t represents t generation, c ₁, c ₂Be the study factor, rand () represents random function, and value is between 0 and 1, and w represents inertia weight; v _IjExpression particulate i j dimension variable change speed, x _IjExpression particulate i j dimension variable present position; The desired positions that all particulates lived through in the colony is p, is called overall desired positions, finds the solution to draw value after the variable optimization, target information.

Images