CN102622655A - Nonlinear production plan optimization method applied to oil refineries - Google Patents

Abstract

The invention discloses a nonlinear production plan optimization method applied to oil refineries, which includes the steps: firstly, setting up an oil refining production logic network; secondly, setting up process modules of an atmospheric and vacuum CDU (crude distillation unit) and a FCC (fluid catalytic cracking) unit; thirdly, inputting all constraint condition data and determining a target function in two-stage search; and fourthly, using the mixed searching method to optimally determine production plan decision variables. The genetic algorithm and direct search algorithm are combined on the basis of the simulation modeling technique, global profits of an oil refinery production plan can be effectively increased without destroying the constraint conditions, tremendous computation generated during solving of complex production plan problems can be avoided, and optimized solutions can be quickly obtained. The nonlinear production plan optimization method is simple in principle, convenient in implementation, high in transportability and applicable to different production plan environments.

Description

A kind of non-linear production planning optimization method that is applied to refinery

Technical field

The present invention relates to process computer and optimize the field, relate in particular to a kind of non-linear production planning optimization method that is applied to refinery.

Background technology

The production schedule is that oil refining enterprise is organized presetting of activity in production, and it has confirmed what kind of production and operation target enterprise will realize, and how to realize such target.The production schedule is that petroleum chemical enterprise manages and reflection and the refinement of development strategy at the manufacturing side, links the procurement plan at the upper reaches and distribution plan, the plan of needs in downstream, is again that crude oil scheduling, device and pipeline are dispatched, the abstract conclusion of petroleum products blending.The optimization of the production schedule is the core that oil refining enterprise solves production efficiency problem.

The oil refining Production planning model is meant to be concluded the simplification of oil refining enterprise complicated production flow process, and its precision has determined the using value and the directive significance of Optimization result.

The oil refining production run utilizes mathematical modeling to demonstrate more and more important effect gradually; People possibly understand the internal work principle of technology; But when confirming optimal parameter, need a large amount of work; The size of reactor and unit thereof for example, the correct consumption and the proportioning of various materials, perhaps optimum flow rate.Used to pass through repetition test, or the dependence experience solves this type problem, yet if set up and test a large amount of prototype plants, this method can spend great amount of time and energy.Utilize the mathematical modeling instrument can set up virtual prototype and disclose the internal mechanism in the technological process.Can revise arbitrary parameter easily, and see effect immediately.

For example application number is to disclose a kind of non-linear continuous tank type diesel oil blending method in the 200610029415.7 Chinese invention patent applications, and its detailed process comprises:

(1) sets up based on the linearity in the real-time qualitative attribute of the linearity between formation each component real-time traffic of finished diesel fuel and each flow and nonlinear relationship, each component and finished diesel fuel and the nonlinear mathematical model of nonlinear relationship;

(2) in the nonlinear mathematical model of crude oil refinery speed, each side line real-time traffic data and the real-time qualitative data input step of each component (1) being set up, computing obtains constituting each component real-time traffic data of finished diesel fuel;

(3) by the steps such as blending of carrying out finished diesel fuel by each component real-time traffic data of formation finished diesel fuel of step (2) acquisition.

The non-linear empirical model that comprises technological parameter also is integrated in the Production planning model gradually; Current optimization to the oil refining production schedule mainly relies on numerical optimization; But the integrated of nonlinear relationship makes traditional linear programming method powerless, and characteristics such as extensive, the multiple constraint that the production schedule of refining oil simultaneously is intrinsic are also perplexing the use of first heuristics such as evolution algorithm, particle cluster algorithm, ant group algorithm.

Summary of the invention

The invention provides a kind of Hybrid Search method that is applied to the non-linear production planning optimization of refinery; With the Simulation and Modeling Technology is the basis; And innovation ground combines genetic algorithm and direct search algorithm; Can tackle the non-linear of optimization objects, significantly reduce dyscalculia and destroy, promote the overall profit of the production schedule with constraint.

A kind of non-linear production planning optimization method that is applied to refinery may further comprise the steps:

(1) sets up in the said refinery logical network of oil refining process units and material side line, crude oil amount of purchase, Flow of Goods and Materials direction, overall production models and the oil product output data of input correspondence;

(2) comprise atmospheric and vacuum distillation unit and catalytic cracking unit at least in the described oil refining process units, and set up the nonlinear process model of atmospheric and vacuum distillation unit and catalytic cracking unit;

(3) utilize the constraint condition data of intermediate material storage tank, oil quality characteristic, other process units and the demand-supply relation, set up the equation of constraint of two phase targets functions with the profit expectation value of the constraint destruction value of oil refining process and planning cycle;

(4) confirm the decision variable that oil refining is produced in the planning cycle,, utilize the Hybrid Search method to obtain the optimizing decision variable that oil refining is produced in the planning cycle according to the equation of constraint in the step (3).

Flow of Goods and Materials direction described in the step (1) is the flow directions of all materials between the oil refining process units, promptly comprises the flow direction of crude oil, also comprises the flow direction of each intermediate product and final products.

Described crude oil amount of purchase is meant the consumption of each raw materials for production in the refinery.

Described overall production models can utilize prior art, make up in conjunction with concrete oil refining process units, and the overall production models here preferably adopt the linear model of simplification.

Output and index that described oil product output data is final products when for example final products comprise gasoline and diesel oil, comprise index octane number and condensation point of diesel oil commonly used at least.

In step (2); Process characteristic and importance according to atmospheric and vacuum distillation unit and catalytic cracking unit are set up nonlinear process model analysis; In the general refinery referring to equipment also have catalytic reforming unit, delayed coking unit, hydro-refining unit and storage tank etc.; Atmospheric and vacuum distillation unit and catalytic cracking unit are preferably set up nonlinear process model analysis as the visual plant of refinery, productive rate and technological parameter be all the other equipment of nonlinear dependence relation then in overall production models according to the linear model analysis of simplifying.

The nonlinear process model of the atmospheric and vacuum distillation unit described in the step (2) is following:

$H_{u,n,t}=a_{u,0}+a_{u,1}{T}_{u,n,t}+a_{u,2}{T}_{u,n,t}^2+...+a_{u,K}{T}_{u,n,t}^K---\left(1\right)$

Y _u，n，t＝H _u，n，t-H _u，n，t-1 (2)

H wherein _{U, n, t}The n discharging side line of expression atmospheric and vacuum distillation unit u is at the productive rate of t production batch;

U is the atmospheric and vacuum distillation unit sequence number;

N is the discharging side line sequence number of atmospheric and vacuum distillation unit;

T is the production batch sequence number of atmospheric and vacuum distillation unit;

Y _{U, n, t}Expression and H _{U, n, t}Corresponding mass yield;

T _{U, n, t}Be the n discharging side line of expression atmospheric and vacuum distillation unit u cutting temperature at the t production batch;

K is predefined integer, and the order in the representation formula for example generally gets 3, and the big more precision of value is high more, can increase but computing is also corresponding;

a _{U, 0}, a _{U, 1}... a _{U, K}Be in advance according to true boiling point productive rate after the nonlinear process model match of described atmospheric and vacuum distillation unit and cutting temperature relevance parameter, receive the properties influence of different crude oils.For example set two cover atmospheric and vacuum distillation units and process Daqing crude oil (DAQ) and LIAOHE CRUDE (LIH) respectively, their true boiling point parameter is also different.

a _{U, 0}, a _{U, 1}... a _{U, K}Value can be according to the H in the actual production in advance _{U, n, t}And T _{U, n, t}Obtain.

The nonlinear process model of the catalytic cracking unit described in the step (2) is following:

Y _u，n，t＝β _n，0+β _n，1(CONV _u，t-Z)+β _n，2(CONV _u，t-Z) ² (3)

Y _{U, n, t}The n discharging side line of expression catalytic cracking unit u is in the mass yield of t production batch;

CONV _{U, t}Expression catalytic cracking unit u is in the conversion factor of t production batch;

Z is the conversion factor of benchmark;

To catalytic cracking unit, the simplification empirical model that uses document (GARY J I.I-Iandwork, GE, Petroleum Refining, TechnologyandEconomics [M] .MarcelDekkerInc.1984.) to propose.In the practical operation, the primary variables that influences the FCC mass yield is cracking temperature, catalyzer/crude oil ratio, air speed etc.Gary etc. propose an abstract variable with amid all these factors, are referred to as conversion factor.

U is the catalytic cracking unit sequence number;

N is the discharging side line sequence number of catalytic cracking unit;

T is the production batch sequence number of catalytic cracking unit;

β _{N, 0}, β _{N, 1}, β _{N, 2}Be side line mass yield and the conversion factor relevance parameter after a series of matches, β _{N, 0}, β _{N, 1}, β _{N, 2}Value can be according to the Y in the actual production in advance _{U, n, t}And CONV _{U, t}Obtain.

In the step (3), the equation expression formula of the Phase I objective function of setting up with the constraint destruction value of oil refining process is:

$\mathrm{CONB}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{r}_i\×G_i$

CONB shows that the constraint in the whole refinery production run destroys G _iRepresent that single constraint exceeds or not enough absolute value, r _iRepresent predefined penalty factor.

I is the call number of constraint condition, and n is the quantity of constraint condition.

Comprise a large amount of constraint condition in the oil refining process, consider the destruction value that constraint destruction value comprises following constraint among the present invention: storage tank constraint, rerum natura constraint, other process units constraints and market supply and demand constraint.

Wherein the storage tank constraint can be expressed as:

INV _u，t＝INV′ _u，t-1+∑QI _u，n，t (4)

${\mathrm{INV}}_{u,t}<{\mathrm{INV}}_u^{\mathrm{up}}---\left(5\right)$

INV′ _u，t＝INV _u，t-∑QO _u，n，t (6)

${\mathrm{INV}}_{u,t}^{\&prime;}>{\mathrm{INV}}_u^{\mathrm{low}}---\left(7\right)$

INV ' _{U, t-1}Represent the jar storage after t-1 batch of u storage tank paid jar, INV _{U, t}Represent the jar storage after t batch of u storage tank received jar, INV ' _{U, t}Represent the jar storage after t batch of u storage tank paid jar, and

With

The upper and lower bound of representing u storage tank jar storage respectively.

The storage tank operation can be divided into rewinding and pay material two steps.Formula (4) and formula (5) expression rewinding operational constraints, jar must not deposit greater than the storage capacity upper limit after the rewinding; Formula (6) and formula (7) expression pair material operational constraints is paid material back jar and must not deposit less than the storage capacity lower limit.

The rerum natura constraint can be expressed as:

${\mathrm{ON}}_{u,t}=\frac{\mathrm{\&Sigma;}({\mathrm{QI}}_{u,n,t}\&CenterDot;{\mathrm{ON}}_{u,n,t})}{\mathrm{\&Sigma;}{\mathrm{QI}}_{u,n,t}}>93---\left(8\right)$

${\mathrm{PP}}_{u,t}=\mathrm{\&Sigma;}\left[{\left(\frac{{\mathrm{PP}}_{u,n,t}}{\mathrm{\&theta;}}\right)}^{\mathrm{\&lambda;}}{{\mathrm{QI}}^{\mathrm{\&omega;}}}_{u,n,t}\right]\&le;{\left(\mathrm{\&Sigma;}{\mathrm{QI}}_{u,n,t}\right)}^{\mathrm{\&omega;}}---\left(9\right)$

ON _{U, t}And PP _{U, t}Octane value and the condensation point of diesel oil of representing t batch of u blending device, ON _{U, n, t}And PP _{U, n, t}The octane value and the condensation point of diesel oil of expression respective side line material, QI _{U, n, t}The charging value of representing t batch of u blending device n bar side line.

The rerum natura effect of contraction is divided into octane number constraint and condensation point of diesel oil constraint in attemperation apparatus.Formula (8) is the linear restriction condition, and the blended gasoline octane value is the weighted mean value of all charging side line octane values of attemperation apparatus; Formula (9) is the non-linear constrain condition, adopts classical condensation point of diesel oil computing formula, and λ, θ, ω learn from else's experience and test parameter.

The constraint of other process units can be expressed as:

CAP _u，m，t＝∑QI _{u，m，n，t} (10)

∑QI _{u，m，n，t}＝∑QO _{u，m，n，t} (11)

${\mathrm{CAP}}_u^{\mathrm{low}}<{\mathrm{CAP}}_{u,m,t}<{\mathrm{CAP}}_u^{\mathrm{up}}---\left(12\right)$

QO _{u，m，n，t}＝CAP _u，m，tY _{u，m，n，t}(13)

CAP _{U, m, t}The processing capacity of representing u device under the t batch of m scheme, QI _{U, m, n, t}The charging value of representing n bar charging side line under the t batch of m scheme, QO _{U, m, n, t}The discharging value of representing n bar discharging side line under the t batch of m scheme, Y _{U, m, n, t}The productive rate of representing n bar discharging side line under the t batch of m scheme, With The upper and lower bound of representing u device processing capacity respectively.

The processing capacity of formula (10) expression process units under the m scheme equals all feed side line material summations; Formula (11) is the material balance formula, and expression turnover material stream adds and equates; Processing capacity under formula (12) the expression m scheme must satisfy minimum and maximum Capability Requirement; Formula (13) is explained the calculating of process units discharging, tries to achieve through nonlinear model calculating except the productive rate of atmospheric and vacuum distillation unit catalytic cracking unit, and remaining process units productive rate is specified by experience productive rate table.

The market supply and demand constraint can be expressed as:

$C_{i,t}^{\mathrm{low}}\&le;C_{i,t}\&le;C_{i,t}^{\mathrm{up}}---\left(14\right)$

$D_{i,t}^{\mathrm{low}}\&le;D_{i,t}\&le;D_{i,t}^{\mathrm{up}}---\left(15\right)$

C _{I, t}Represent t batch of amount of purchase, D to i crude oil _{I, t}The turnout of t batch of i oil product of expression,

With

Represent the upper and lower bound of i crude oil respectively t batch of supply,

With

Represent the upper and lower bound of i oil product respectively t batch of demand.

In the step (3), the equation expression formula of the Phase objective function of setting up with the profit expectation value of the planning cycle of oil refining process is:

OBJ＝∑P _i·D _i，t-∑P _j·C _j，t-σ·∑S _u，t

P _iAnd P _jThe price of representing oil product and crude oil respectively, D _{I, t}The turnout of t batch of following i oil product of expression, C _J.tRepresent t batch of amount of purchase, S to j crude oil _{U, t}The jar storage of t batch of following u storage tank of expression, σ representation unit jar is deposited cost.

Decision variable comprises in the step (4):

The cutting temperature of atmospheric and vacuum distillation unit;

The conversion factor of catalytic cracking unit;

With corresponding other process units schemes of predetermined productive rate;

The crude oil amount of purchase;

Storage tank is paid the material amount.

Wherein atmospheric and vacuum distillation unit can be some covers, the corresponding some material side lines of every cover, and every corresponding cutting temperature of material side line, and promptly certain material side line can the corresponding cut product of output under the corresponding cutting temperature.For example atmospheric and vacuum distillation unit is 2 covers, and corresponding 6 the material side lines of every cover then amount to 12 cutting temperatures.

The corresponding conversion factor of every cover catalytic cracking unit.

As decision variable; Described other process units schemes are meant productive rate; Promptly specific other process units scheme corresponding a kind of yield of product, for example the final products productive rate has four kinds, is meant four other process units schemes as decision variable so.

Certainly the different productive rates of different product corresponding different parameter in actual production; Found out corresponding relation through simulation or actual production in advance between productive rate and the technological parameter; Utilize the productive rate after the present invention optimizes so, just can utilize corresponding technological parameters to instruct and produce.

The crude oil amount of purchase is meant the consumption of each raw materials for production in the refinery, if three kinds of raw materials are arranged, the crude oil amount of purchase as decision variable then has three so.

Storage tank is paid the load that the material amount is meant each outlet line on all storage tanks; 9 storage tanks are for example arranged; Wherein have 5 storage tanks to have two outlet lines, say 14 outlet lines altogether from the outlet line angle so, promptly paying the material amount as the storage tank of decision variable has 14.

In the step (4), utilize the Hybrid Search method to confirm the optimizing decision variable of oil refining production logical network in the planning cycle, its detailed process is divided into Phase I and Phase;

Phase I is carried out the genetic algorithm of a cover standard floating-point encoding scheme, and step is following:

A, initialization: define the span of each decision variable, use the floating-point encoding scheme to generate population at random, wherein discrete variable is got approximate round values;

B, evaluation: with the input of the decision variable in the population that generates production schedule realistic model; Calculate all individual constraint destruction value CONB (x) in the population; And according to the size of constraint destruction value all individualities are sorted, the constraint destruction value CONB (x) that is about to individuality converts the ordering factor into;

Described production schedule realistic model is the overall production models that are associated through described logical network and the nonlinear process model of atmospheric and vacuum distillation unit and catalytic cracking unit.

C, selection: employing roulette mode is taken out equivalent amount at random in population individuality is as parent;

D, intersection and variation:, use crossover operator and mutation operator and generate new filial generation according to crossover probability that is provided with in advance and variation probability;

Described crossover probability is optional 0.8, promptly selects 80% individuality to participate in crossing operation;

Described variation probability is optional 0.05, and the possibility that promptly new filial generation is morphed is 5%.

E, with the sample loops execution in step a～d of the new filial generation described in the steps d, until reaching predetermined genetic algebra as step a.

Many more results are accurate more for described genetic algebra, for example can be for 500 generations.

F, will accomplish the population that obtains behind the predetermined genetic algebra individual (being new filial generation) and corresponding target function value (among the step b with the corresponding ordering factor of this population individuality) and pass to the Phase algorithm;

The Phase I algorithm is primarily aimed at the population operation to cover whole solution space, and the Phase algorithm is directed against the performance that individual operations is separated with lifting, and concrete steps are following:

G, the population individuality that obtains with the Phase I algorithm are separated x as primary iteration ₀, set primary iteration step-length matrix a ₀＞0, carry out interative computation;

H, the i time iterative solution can be expressed as: x _i=x _I-1+ a _iD;

Wherein i is a number of iterations, and d is the vector of unit length of random direction, and this vector of unit length is a matrix form, and the scale of this vector of unit length is that (1, n), wherein n is the number of all decision variables.The value of each element is only for-1,0 or 1 in the vector of unit length (matrix).Obtain x after d and the step-length matrix dot product ₀Movement matrix;

I, the x that each iteration is obtained _iInput production schedule realistic model carries out query manipulation, and described production schedule realistic model is the described overall production models that are associated through described logical network of Phase I and the nonlinear process model of atmospheric and vacuum distillation unit and catalytic cracking unit;

If satisfy condition: OBJ (x _i)＞OBJ (x _I-1) and CONB (x _i)≤CONB (x _I-1);

Represent the i time iterative query success, be provided with simultaneously: a _I+1=a _i, x _i=x _I-1+ a _iD gets into computing next time again;

Otherwise, if do not satisfy condition: OBJ (x _i)＞OBJ (x _I-1) and CONB (x _i)≤CONB (x _I-1), represent the i time iterative query failure, x is set _i=x _I-1, reduce step-size in search a _I+1=a _i/ 2 recomputate and inquire about x _iIf the continuous successful inquiring not yet that reduces for five times behind the step-size in search then regenerates random direction vector d (value-1,0 or 1 that is direction vector d element changes at random), execution in step i again;

J, when iterations i reaches threshold value, jump out iteration, obtaining the corresponding search result is described optimizing decision variable and the output of corresponding target function value.

High more result is accurate more for the threshold value of iterations i, but calculated amount also increases complicacy, general desirable 100～500 times, for example 200 times thereupon.

Described corresponding target function value is the functional value of two phase targets functions described in the step corresponding with this optimizing decision variable (3).

The Hybrid Search method of Optimal Production plan of the present invention is on the basis of Simulation and Modeling Technology; In conjunction with genetic algorithm and direct search algorithm; Can under the situation of not destroying constraint condition, effectively promote the overall profit of the oil refining production schedule; The huge calculated amount of avoiding complicated production plan problem solving process to produce, and can be optimized rapidly and separate.Principle is simple, and it is convenient to implement, and is portable strong, is applicable to the different production environment of plans.

Description of drawings

Fig. 1 is the synoptic diagram of the production schedule logical network of a certain oil refining enterprise.

Fig. 2 is worth the evolution curve map in Phase I for constraint destruction among the present invention.

Embodiment

In order to describe the present invention more particularly, be elaborated below in conjunction with accompanying drawing and embodiment Hybrid Search method to Optimal Production plan of the present invention.

A kind of non-linear production planning optimization method that is applied to refinery may further comprise the steps:

(1) sets up oil refining production logical network.

As research object, process data is obtained from the scene, workshop with certain refinery.Fig. 1 is the flow sheet of this refinery, and wherein connecting line is represented the route of Flow of Goods and Materials, and arrow is represented the direction of Flow of Goods and Materials.This flow process comprises:

Atmospheric and vacuum distillation unit CDU1, atmospheric and vacuum distillation unit CDU2;

Catalytic cracking unit FCC1, catalytic cracking unit FCC2;

Catalytic reforming unit CRU;

Delayed coking unit DCU;

Hydro-refining unit HT1, hydro-refining unit HT2;

Mediation pond GB, the pond DB that is in harmonious proportion;

Nine logic jars are respectively logic jar TKLSR, logic jar TKHSR, logic jar TKLGO, logic jar TKAGO, logic jar TKVRC, logic jar TKVGO, logic jar TKDIE, logic jar TKCG, logic jar TKCGO; The logic jar is gathering of physics tank field, representes that such as logic jar TKLSR all store the set of the physics jar of material LSR.

Crude oil input DAQ (Daqing crude oil), crude oil input LIH (LIAOHE CRUDE), raw material input methyl tert-butyl ether MTBE;

Product oil output comprises 93#GAS (gasoline), 0#DIE (diesel oil) and kerosene KER.

(2) process model of definition atmospheric and vacuum distillation unit CDU and catalytic cracking unit FCC.

The productive rate of process model and technological parameter are the nonlinear dependence relation, and wherein the mathematic(al) representation of atmospheric and vacuum distillation unit is following:

$H_{u,n,t}=a_{u,0}+a_{u,1}{T}_{u,n,t}+a_{u,2}{T}_{u,n,t}^2+...+a_{u,K}{T}_{u,n,t}^K---\left(1\right)$

Y _u，n，t＝H _u，n，t-H _u，n，t-1 (2)

H wherein _{U, n, t}The n discharging side line of expression atmospheric and vacuum distillation unit u is at the true boiling point productive rate of t production batch, Y _{U, n, t}Represent mass yield on the other side.a _{U, 0}, a _{U, 1}... a _{U, K}Be true boiling point productive rate and the cutting temperature relevance parameter after a series of matches, receive the properties influence of different crude oils.Set two cover atmospheric and vacuum distillation units and process Daqing crude oil (DAQ) and LIAOHE CRUDE (LIH) respectively, their true boiling point parameter is also different.

The mathematic(al) representation of catalytic cracking unit is following:

Y _u，n，t＝β _n，0+β _n，1(CONV _u，t-Z)+β _n，2(CONV _u，t-Z) ² (3)

Y _{U, n, t}The n discharging side line of expression catalytic cracking unit u is at the mass yield of t production batch, CONV _{U, t}Expression catalytic cracking unit u is in the conversion factor of t production batch.β _{N, 0}, β _{N, 1}, β _{N, 2}Be side line mass yield and the conversion factor relevance parameter after a series of matches.

(3) obtain all constraint conditions, define the objective function of two stages search.

Comprise a large amount of constraint condition in the oil refining process, the storage tank constraint can be expressed as:

INV _u，t＝INV′ _u，t-1+∑QI _u，n，t (4)

${\mathrm{INV}}_{u,t}<{\mathrm{INV}}_u^{\mathrm{up}}---\left(5\right)$

INV′ _u，t＝INV _u，t-∑QO _u，n，t (6)

${\mathrm{INV}}_{u,t}^{\&prime;}>{\mathrm{INV}}_u^{\mathrm{low}}---\left(7\right)$

INV ' _{U, t-1}Represent the jar storage after t-1 batch of u storage tank paid jar, INV _{U, t}Represent the jar storage after t batch of u storage tank received jar, INV ' _{U, t}Represent the jar storage after t batch of u storage tank paid jar, and

With The upper and lower bound of representing u storage tank jar storage respectively.

The storage tank operation can be divided into rewinding and pay material two steps.Formula (4) and formula (5) expression rewinding operational constraints, jar must not deposit greater than the storage capacity upper limit after the rewinding; Formula (6) and formula (7) expression pair material operational constraints is paid material back jar and must not deposit less than the storage capacity lower limit.

The rerum natura constraint can be expressed as:

${\mathrm{ON}}_{u,t}=\frac{\mathrm{\&Sigma;}({\mathrm{QI}}_{u,n,t}\&CenterDot;{\mathrm{ON}}_{u,n,t})}{\mathrm{\&Sigma;}{\mathrm{QI}}_{u,n,t}}>93---\left(8\right)$

${\mathrm{PP}}_{u,t}=\mathrm{\&Sigma;}\left[{\left(\frac{{\mathrm{PP}}_{u,n,t}}{\mathrm{\&theta;}}\right)}^{\mathrm{\&lambda;}}{{\mathrm{QI}}^{\mathrm{\&omega;}}}_{u,n,t}\right]\&le;{\left(\mathrm{\&Sigma;}{\mathrm{QI}}_{u,n,t}\right)}^{\mathrm{\&omega;}}---\left(9\right)$

ON _{U, t}And PP _{U, t}Octane value and the condensation point of diesel oil of representing t batch of u blending device, ON _{U, n, t}And PP _{U, n, t}The octane value and the condensation point of diesel oil of expression respective side line material, QI _{U, n, t}The charging value of representing t batch of u blending device n bar side line.

The rerum natura effect of contraction is divided into octane number constraint and condensation point of diesel oil constraint in attemperation apparatus.Formula (8) is the linear restriction condition, and the blended gasoline octane value is the weighted mean value of all charging side line octane values of attemperation apparatus; Formula (9) is the non-linear constrain condition, adopts classical condensation point of diesel oil computing formula, and λ, θ, ω learn from else's experience and test parameter.

The constraint of other process units can be expressed as:

CAP _u，m，t＝∑QI _{u，m，n，t} (10)

∑QI _{u，m，n，t}＝∑QO _{u，m，n，t} (11)

${\mathrm{CAP}}_u^{\mathrm{low}}<{\mathrm{CAP}}_{u,m,t}<{\mathrm{CAP}}_u^{\mathrm{up}}---\left(12\right)$

QO _{u，m，n，t}＝CAP _u，m，tY _{u，m，n，t} (13)

CAP _{U, m, t}The processing capacity of representing u device under the t batch of m scheme, QI _{U, m, n, t}The charging value of representing n bar charging side line under the t batch of m scheme, QO _{U, m, n, t}The discharging value of representing n bar discharging side line under the t batch of m scheme, Y _{U, m, n, t}The productive rate of representing n bar discharging side line under the t batch of m scheme,

With

The upper and lower bound of representing u device processing capacity respectively.

The processing capacity of formula (10) expression process units under the m scheme equals all feed side line material summations; Formula (11) is the material balance formula, and expression turnover material stream adds and equates; Processing capacity under formula (12) the expression m scheme must satisfy minimum and maximum Capability Requirement; Formula (13) is explained the calculating of process units discharging, tries to achieve through nonlinear model calculating except the productive rate of CDU and FCC, and remaining process units productive rate is specified by experience productive rate table.

The market supply and demand constraint can be expressed as:

$C_{i,t}^{\mathrm{low}}\&le;C_{i,t}\&le;C_{i,t}^{\mathrm{up}}---\left(14\right)$

$D_{i,t}^{\mathrm{low}}\&le;D_{i,t}\&le;D_{i,t}^{\mathrm{up}}---\left(15\right)$

C _{I, t}Represent t batch of amount of purchase, D to i crude oil _{I, t}The turnout of t batch of i oil product of expression,

With

Represent the upper and lower bound of i crude oil respectively t batch of supply, With Represent the upper and lower bound of i oil product respectively t batch of demand.

The amount of purchase of formula (14) expression crude oil must satisfy the bound scope with the cycle market supply; The output of formula (15) expression oil product must satisfy the bound scope with the cycle market demand.

The equation expression formula of Phase I objective function is:

$\mathrm{CONB}=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{r}_i\&times;G_i---\left(16\right)$

Formula (16) is the Phase I objective function, shows that the constraint in the whole refinery production run destroys G _iRepresent that single constraint exceeds or not enough absolute value, r _iRepresent predefined penalty factor.

The equation expression formula of Phase objective function is:

OBJ＝∑P _i·D _i，t-∑P _j·C _j，t-σ·∑S _u，t (17)

P _iAnd P _jThe price of representing oil product and crude oil respectively, S _{U, t}The jar storage of t batch of following u storage tank of expression, σ representation unit jar is deposited cost, D _{I, t}The turnout of t batch of following i oil product of expression, C _{J, t}Represent t batch of amount of purchase to j crude oil.

Formula (17) is the objective function of Phase, shows the overall profit of whole refinery in planning cycle, is defined as the poor of income and cost.Income is from the sales revenue of product oil, and cost then is made up of crude oil buying expenses and oil tank inventory cost.For simplifying to analyze cost coefficient has been made reasonable assumption: all turnout all are converted into sales volume; Product oil price and crude oil price remain unchanged in the production schedule cycle; All storage tank unit's inventory costs are identical.

(4) utilize the Hybrid Search method, optimize and confirm production schedule decision variable.

It comprises the decision variable of the production schedule:

CDU cutting temperature (two groups, totally 12)

FCC conversion factor (two groups, totally 2)

Other process units schemes (4)

Crude oil amount of purchase (3)

Storage tank is paid material amount (14)

Being provided with of definition phase I algorithm and Phase algorithm is as shown in table 1.

The setting of table 1 algorithm parameter

Phase I	Value	Phase	Value
				The population size	10	The random direction number	5
Maximum genetic algebra	500	Maximum iteration time (DS)	200
				Crossing-over rate	0.8	The step-length ratio	0.05
Aberration rate	0.2

The Hybrid Search method is divided into Phase I and Phase:

Phase I is carried out the genetic algorithm of a cover standard floating-point encoding scheme, and step is following:

A, initialization: define the span of each decision variable, use the floating-point encoding scheme to generate population at random, wherein discrete variable is got approximate round values;

B, evaluation: with the input of the decision variable in the population that generates production schedule realistic model; Calculate all individual constraint destruction value CONB (x) in the population; And according to the size of constraint destruction value all individualities are sorted, the constraint destruction value CONB (x) that is about to individuality converts the ordering factor into;

C, selection: employing roulette mode is taken out equivalent amount at random in population individuality is as parent;

D, intersection and variation:, use crossover operator and mutation operator and generate new filial generation according to crossover probability that is provided with in advance and variation probability;

E, with the sample loops execution in step a～d of the new filial generation described in the steps d, until reaching predetermined genetic algebra as step a;

F, will accomplish the population that obtains behind the predetermined genetic algebra individual (being new filial generation) and corresponding target function value (among the step b with the corresponding ordering factor of this population individuality) and pass to the Phase algorithm;

The Phase I algorithm is primarily aimed at the population operation to cover whole solution space, and through the performance that the Phase algorithm is separated with lifting to individual operations, concrete steps are following:

G, the population individuality that obtains with the Phase I algorithm are separated x as primary iteration ₀, set primary iteration step-length matrix a ₀＞0, carry out interative computation;

H, the i time iterative solution can be expressed as: x _i=x _I-1+ a _iD;

Wherein i is a number of iterations, and d is the vector of unit length of random direction, and this vector of unit length is a matrix form, and the scale of this vector of unit length is that (1, n), wherein n is the number of all decision variables.The value of each element is only for-1,0 or 1 in the vector of unit length (matrix).Obtain x after d and the step-length matrix dot product ₀Movement matrix;

I, the x that each iteration is obtained _iInput production schedule realistic model carries out query manipulation, and described production schedule realistic model is the described overall production models that are associated through described logical network of Phase I and the nonlinear process model of atmospheric and vacuum distillation unit and catalytic cracking unit;

If satisfy condition: OBJ (x _i)＞OBJ (x _I-1) and CONB (x _i)≤CONB (x _I-1);

Represent the i time iterative query success, be provided with simultaneously: a _I+1=a _i, x _i=x _I-1+ a _iD gets into computing next time again;

Otherwise, if do not satisfy condition: OBJ (x _i)＞OBJ (x _I-1) and CONB (x _i)≤CONB (x _I-1), represent the i time iterative query failure, x is set _i=x _I-1, reduce step-size in search a _I+1=a _i/ 2 recomputate and inquire about x _iIf the continuous successful inquiring not yet that reduces for five times behind the step-size in search then regenerates random direction vector d (value-1,0 or 1 that is direction vector d element changes at random), execution in step (i) again;

J, when iterations i reaches threshold value, jump out iteration, obtaining the corresponding search result is described optimizing decision variable and the output of corresponding target function value.

The Hybrid Search method stops after reaching greatest iteration automatically, exports the optimizing decision variable of whole process, the manufacturing variables after promptly optimizing, and its performance is as shown in table 2:

The performance of table 2 Hybrid Search method

Constraint destroys

Overall situation profit/unit

Computing time

0

14793×10 5

7.15s

The visible Hybrid Search method of table 2 can obtain the profit near global optimum, and does not destroy to appoint and watch constraint.The pairing fitness value of optimizing decision variable is 147930, and Fig. 2 is for the evolution curve map of constraint destruction value in Phase I, and is visible through promptly converging to 0 fast less than 50 generations, proves that this method can find feasible solution rapidly.

Claims (8)

1. a non-linear production planning optimization method that is applied to refinery is characterized in that, may further comprise the steps:

(1) sets up in the said refinery logical network of oil refining process units and material side line, crude oil amount of purchase, Flow of Goods and Materials direction, overall production models and the oil product output data of input correspondence;

(2) comprise atmospheric and vacuum distillation unit and catalytic cracking unit at least in the described oil refining process units, and set up the nonlinear process model of atmospheric and vacuum distillation unit and catalytic cracking unit;

(3) utilize the constraint condition data of intermediate material storage tank, oil quality characteristic, other process units and the demand-supply relation, set up the equation of constraint of two phase targets functions with the profit expectation value of the constraint destruction value of oil refining process and planning cycle;

(4) confirm the decision variable that oil refining is produced in the planning cycle,, utilize the Hybrid Search method to obtain the optimizing decision variable that oil refining is produced in the planning cycle according to the equation of constraint in the step (3).

2. the non-linear production planning optimization method that is applied to refinery as claimed in claim 1 is characterized in that, in the step (3), the equation expression formula of the Phase I objective function of setting up with the constraint destruction value of oil refining process is:

$\mathrm{CONB}=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{r}_i\&times;G_i$

CONB is the constraint destruction value in the whole refinery production run, G _iRepresent that single constraint exceeds or not enough absolute value, r _iRepresent predefined penalty factor;

I is the call number of constraint condition, and n is the quantity of constraint condition.

3. the non-linear production planning optimization method that is applied to refinery as claimed in claim 2 is characterized in that described constraint destruction value comprises the destruction value of following constraint: storage tank constraint, rerum natura constraint, other process units constraints and market supply and demand constraint.

4. the non-linear production planning optimization method that is applied to refinery as claimed in claim 3 is characterized in that, in the step (3), the equation expression formula of the Phase objective function of setting up with the profit expectation value of the planning cycle of oil refining process is:

OBJ＝∑P _i·D _i，t-∑P _j·C _j，t-σ·∑S _u，t

P _iAnd P _jThe price of representing oil product and crude oil respectively, S _{U, t}The jar storage of t batch of following u storage tank of expression, σ representation unit jar is deposited cost, D _{I, t}The turnout of t batch of following i oil product of expression, C _J.tRepresent t batch of amount of purchase to j crude oil.

5. the non-linear production planning optimization method that is applied to refinery as claimed in claim 4 is characterized in that, decision variable comprises in the step (4):

The cutting temperature of atmospheric and vacuum distillation unit;

The conversion factor of catalytic cracking unit;

With corresponding other process units schemes of predetermined productive rate;

The crude oil amount of purchase;

Storage tank is paid the material amount.

6. the non-linear production planning optimization method that is applied to refinery as claimed in claim 5 is characterized in that the nonlinear process model of described atmospheric and vacuum distillation unit is following:

$H_{u,n,t}=a_{u,0}+a_{u,1}{T}_{u,n,t}+a_{u,2}{T}_{u,n,t}^2+...+a_{u,K}{T}_{u,n,t}^K---\left(1\right)$

Y _u，n，t＝H _u，n，t-H _u，n，t-1 (2)

H wherein _{U, n, t}The n discharging side line of expression atmospheric and vacuum distillation unit u is at the productive rate of t production batch;

U is the atmospheric and vacuum distillation unit sequence number;

N is the discharging side line sequence number of atmospheric and vacuum distillation unit;

T is the production batch sequence number of atmospheric and vacuum distillation unit;

Y _{U, n, t}Expression and H _{U, n, t}Corresponding mass yield;

T _{U, n, t}Be the n discharging side line of expression atmospheric and vacuum distillation unit u cutting temperature at the t production batch;

K is predefined integer, the order in the representation formula;

a _{U, 0}, a _{U, 1}... a _{U, K}Be in advance according to true boiling point productive rate after the nonlinear process model match of described atmospheric and vacuum distillation unit and cutting temperature relevance parameter.

7. the non-linear production planning optimization method that is applied to refinery as claimed in claim 6 is characterized in that the nonlinear process model of described catalytic cracking unit is following:

Y _u，n，t＝β _n，0+β _n，1(CONV _u，t-Z)+β _n，2(CONV _u，t-Z) ² (3)

Y _{U, n, t}The n discharging side line of expression catalytic cracking unit u is in the mass yield of t production batch;

CONV _{U, t}Expression catalytic cracking unit u is in the conversion factor of t production batch;

Z is the conversion factor of benchmark;

U is the catalytic cracking unit sequence number;

N is the discharging side line sequence number of catalytic cracking unit;

T is the production batch sequence number of catalytic cracking unit;

β _{N, 0}, β _{N, 1}, β _{N, 2}Be side line mass yield and the conversion factor relevance parameter after a series of matches.

8. the non-linear production planning optimization method that is applied to refinery as claimed in claim 7 is characterized in that, utilizes the Hybrid Search method to confirm the optimizing decision variable of oil refining production logical network in the planning cycle, and its detailed process is divided into Phase I and Phase;

Phase I is carried out the genetic algorithm of a cover standard floating-point encoding scheme, and step is following:

A, initialization: define the span of each decision variable, use the floating-point encoding scheme to generate population at random, wherein discrete variable is got approximate round values;

B, evaluation: with the input of the decision variable in the population that generates production schedule realistic model; Calculate all individual constraint destruction value CONB (x) in the population; And according to the size of constraint destruction value all individualities are sorted, the constraint destruction value CONB (x) that is about to individuality converts the ordering factor into;

Described production schedule realistic model is the overall production models that are associated through described logical network and the nonlinear process model of atmospheric and vacuum distillation unit and catalytic cracking unit;

C, selection: employing roulette mode is taken out equivalent amount at random in population individuality is as parent;

D, intersection and variation:, use crossover operator and mutation operator and generate new filial generation according to crossover probability that is provided with in advance and variation probability;

E, with the sample loops execution in step a～d of the new filial generation described in the steps d, until reaching predetermined genetic algebra as step a;

F, will to accomplish the population that obtains behind the predetermined genetic algebra individual and pass to the Phase algorithm with the individual corresponding ordering factor of this population;

The Phase algorithm, concrete steps are following:

G, the population individuality that obtains with the Phase I algorithm are separated x as primary iteration ₀, set primary iteration step-length matrix a ₀＞0, carry out interative computation;

H, the i time iterative solution can be expressed as: x _i=x _I-1+ a _iD;

Wherein i is a number of iterations, and d is the vector of unit length of random direction, and this vector of unit length is a matrix form, and the scale of this vector of unit length is that (1, n), wherein n is the number of all decision variables, and the value of each element is only for-1,0 or 1 in the vector of unit length;

I, the x that each iteration is obtained _iInput production schedule realistic model carries out query manipulation, and described production schedule realistic model is the described overall production models that are associated through described logical network of Phase I and the nonlinear process model of atmospheric and vacuum distillation unit and catalytic cracking unit;

If satisfy condition: OBJ (x _i)＞OBJ (x _I-1) and CONB (x _i)≤CONB (x _I-1);

Represent the i time iterative query success, be provided with simultaneously: a _I+1=a _i, x _i=x _I-1+ a _iD gets into computing next time again;

Otherwise, if do not satisfy condition: OBJ (x _i)＞OBJ (x _I-1) and CONB (x _i)≤CONB (x _I-1), represent the i time iterative query failure, x is set _i=x _I-1, reduce step-size in search a _I+1=a _i/ 2 recomputate and inquire about x _iIf the continuous successful inquiring not yet that reduces for five times behind the step-size in search then regenerates random direction vector d and execution in step i;

J, when iterations i reaches threshold value, jump out iteration, obtaining the corresponding search result is described optimizing decision variable and the output of corresponding target function value.

Images