CN106373023A - Batching optimization method based on new multi-objective artificial bee colony algorithm - Google Patents

Abstract

The invention relates to a batching optimization method based on a new multi-objective artificial bee colony algorithm. The method comprises steps: 1, batching parameters are initialized; 2, a batching optimization multi-objective model is built, and a fitness value is evaluated; 3, data are updated; 4, after preset algebra optimization, each colony selects partial individuals with excellent information for information exchange, the selected individuals form one list, the list is transmitted to another colony, each colony needs to prepare one repalcement list, and individuals in the list are replaced by individuals from other colonies; and 5, if a preset ending condition is not achieved, the second step is returned, and if an iterative termination condition is achieved, calculation is stopped, and a result is outputted finally. The method of the invention has the advantages that the global search capability is strong; the convergence rate is quick; the solution convergence precision is high; the solution distribution is uniform; the solution comprehensive performance is excellent; various feasible batching schemes can be provided with only one-time operation; and control and management on the batching process by a batching person are facilitated.

Description

A kind of blending optimization method based on new multi objective artificial bee colony algorithm

Technical field

The present invention relates to a kind of blending optimization method based on new multi objective artificial bee colony algorithm, belong to coloured Metal manufactures field, is related to Swarm Intelligence Algorithm field simultaneously.

Background technology

For the blending process of high accuracy copper plate/strip production line, influence factor is many and interrelated, except joining Than will in processing range with ensure product quality outside, it is also contemplated that cost of material, feeding sequence, stock, Play the factors such as the maximization recycling of the melt and raw material scaling loss in fusion process and waste material.To having The blending process of These characteristics, comprises substantial amounts of uncertain information, using traditional linear model, it is modeled, It is difficult to describe the relatedness of influence factor during it, and cannot rationally process multiple constraintss.Right In multivariate, the non-linear and solution of multi-objective Model, existing blending optimization computational methods are not appropriate for, Because traditional optimal solution is it is difficult to adapt to complicated and changeable in high-precision copper plate/strip dispensing and fusion process Site environment, and multiple alternative multifarious optimal value that has coordinates the experience of ingredients personnel to have more and can grasp The property made.

For this extensive, the non-linear, complicated optimum problem with multiple targets of high accuracy copper coin dispensing, Traditional Multipurpose Optimal Method has following deficiency: 1) traditional method can only obtain a handkerchief in once running Tired support optimal solution, because traditional method is all multi-objective optimization question to be converted into single-object problem asked Solution；2) when Pareto forward position is recessed, generally, traditional method cannot ensure that finding all of handkerchief tires out Support optimal solution；3) almost all of traditional method is required for certain priori.So, in recent years, extensively The Multiobjective Intelligent optimized algorithm based on bio-inspired computing of general simulation Biology seed coating, has obtained scholars' Extensive concern, uses it to solution multi-objective optimization question and yields good result.But the relatively early increasing proposing Strong Pareto evolution algorithm implementation complexity is higher, quick non-dominant genetic algorithm when population is close to convergence, blind Mesh carries out intersecting, mutation operation can make the population already restrained deviate real Pareto disaggregation, receives to algorithm Hold back and have undesirable effect；Although multi-objective particle swarm algorithm optimal speed is very fast, it is easily trapped into local Excellent, robustness is poor, and the result randomness obtaining is big.These already present algorithms are relative complex in solution During multi-objective optimization question, obtained solution can't reach satisfied convergence and the uniformity of angle distribution will Ask.

Content of the invention

Solving extensive, non-linear, the complicated blending optimization problem with multiple targets for existing calculating When the defect that come out, the present invention proposes and a kind of use for reference multiple Nidus Vespae Apiss cooperations in the Nature and look for food The blending optimization method of the new multi objective artificial bee colony algorithm of behavior.

The present invention be the technical scheme is that one kind is based on new multi objective people worker bee for achieving the above object The blending optimization method of group's algorithm, comprises the following steps:

Step 1: dispensing parameter initialization；

Step 2: set up blending optimization multi-objective Model, and assess fitness value；

Step 3: data updates；

Step 4: after the optimization presetting algebraically, each population selected section carries excellent Information Individuality be used for information exchange；Selected one list of individual composition, this list will be sent to another In individual population；Each population will prepare a replacement list, and the individuality in this list will be by from other kinds The individual replacement that group comes；

Step 5: if end condition not up to set in advance, return to step 2, if reach iteration ends Condition, then stop calculating, last output result.

Described dispensing parameter initialization, comprises the following steps:

Carry out real coding using 2 × (n+1) matrix forms, n is raw material type number；

Determine initial population number, each population scale, each body position of random initializtion, determine maximum Cycle-index mnc；

Determine variable dimension d, i.e. the number of decision variable；

Initialization important parameter " limit ":

Limit=mnc × d/2

Wherein, d is the number of decision variable, and mnc is maximum cycle.

Described 2 × (n+1) matrix forms are [(x, q), (t₀,t_q)]^t, matrix column number is the length of chromosome, the A line chromosome represents inventory x of raw material, and the second row chromosome represents raw material smelting time t₀With a melt Smelting time t_q, (n+1)th integer of inventory row chromosome is corresponding to have used melt constant q.

Described blending optimization multi-objective Model particularly as follows:

Set up following two object functions, if c_iFor the price of raw material in i-th, x_iInput for i-th kind of raw material Amount, c has been melt price, then cost of material:

$f_1\left(x\right)=\underset{i=1}{\overset{n}{\σ}}{c}_i{x}_i$

Compensate and expect cost:

$f_2(x,t,t_q)=\underset{i=1}{\overset{n}{\σ}}\[c_i{x}_i{\mathrm{\η}}_i(t_q-t_i)\]+{\mathrm{cq\η}}_q\left(t_q\right)$

Wherein, t_qFor playing the smelting time of melt, t_iMaking time for i-th kind of raw material；

Feed intake totle drilling cost:

$\begin{array}{c}{f}_1(x,t,t_q)=f_1\left(x\right)+f_2(x,t,t_q)\\ =\underset{i=1}{\overset{n}{\σ}}{c}_i{x}_i\[1+{\mathrm{\η}}_i(t_q-t_i)\]+{\mathrm{cq\η}}_q\left(t_q\right)\end{array}$

Bring old material amount into:

$f_2\left(x^{\′}\right)=f_2(x_{k+1},x_{k+2}\mathrm{...}{x}_n)=\underset{i=k+1}{\overset{m}{\σ}}{\mathrm{\α}}_i{x}_i+\underset{i=m+1}{\overset{p}{\σ}}{\mathrm{\β}}_i{x}_i+\underset{i=p+1}{\overset{n}{\σ}}{\mathrm{\δ}}_i{x}_i$

Wherein, η_i(t_q-t_i)=g_iln(h_it_i+ 1) the scaling loss time function for raw material j；η_q(t_q)=g_qln(h_qt_q+1) For playing function heat time heating time of melt burn out rate；α is the penalty factor that this trade mark old material brings deficiency into；β is it The penalty factor of deficiency brought into by his board for old material；δ brings the penalty factor of deficiency into for chemical waste；

In conjunction with totle drilling cost f that feeds intake₁With bring old material amount f into₂, for representing blending optimization multi-objective Model minimum shape The multi-goal optimizing function of formula is written as:

min{f₁(x,t,t_q),-f₂(x')}

Described data updates and comprises the following steps:

Randomly generate n people worker bee in search space first, n represents the size of each population, for feasible Solution x_ij, i=1,2 ..., n, j are vector dimension, and its maximum is set to d；After completing initial phase, will Execute following three iterative process:

3.4) employ the neighbour in nectar source corresponding to the random prospecting of honeybee, and nectar source position in its memory is repaiied Change, a new candidate solution is found according to an existing solution, this candidate solution is solved near Neighbour, new nectar source v_ijProduce according to formula below:

v_ij=x_ij+δ_ij(x_ij-x_kj)

Wherein, k=1,2 ..., n；J=1,2 ..., d are the indexes randomly selecting, and k is random determination, but can not Identical with i；δ_ijRandomly generate in the range of [- 1,1], this state modulator in x_ijThe product in the new nectar source of surrounding Raw.

3.5) after finding new nectar source, x_ijNeed and v_ijBe compared, two nectar sources only preferably that Individual can remain, one follow honeybee select which nectar source depend on be associated with nectar source probit p_i, its calculation is as follows:

$p_i=\frac{{\mathrm{fit}}_i}{\underset{i=1}{\overset{n}{\σ}}{\mathrm{fit}}_i}$

Wherein, fit_iIt is solution x_iFitness value, this fitness value is proportional to the number of Mel in the nectar source of position i Amount, its numerical expression is as follows:

${\mathrm{fit}}_i=\left\{\begin{array}{cc}{v}_i,& if\left(f\right(x_i)\&greaterequal;f(v_i\left)\right)\\ x_i,& if\left(f\right(x_i)<f(v_i\left)\right)\end{array}\right.$

3.6) if in a nectar source Mel quantity do not have in certain iterationses any improve if, this Individual nectar source will be employed honeybee to abandon by corresponding, after nectar source is abandoned, employ honeybee will become scouting Honeybee, and search bee to replace that abandoned nectar source by finding a new nectar source, the row of these people workers bee It is to be described with formula below:

$x_i^j=x_{\min }^j+\mathrm{\&sigma;}(x_{max}^j-x_{\min }^j)$

Wherein, σ is the random number in the range of [- 1,1], when search bee finds the abundanter honey of honey content Behind source, this search bee will become one and employ honeybee.

Described list will be sent in another population, comprise the following steps:

4.3) scale of transmission list is fixed value set in advance, and each population is according to lower column major order It is ready for sending list:

P1: select individual First Principles to be non-dominant level, the individuality in low level will preferably select；

P2: if the individuality in the first level is bigger than the preset value of transmission list size, need to calculate In population, the individual Hamming distance of each pair and crowding distance, calculate and have maximum Hamming distance apart from other individualities Individuality, then select from it nearest l individual enter transmission list, finally select to have larger crowded The individuality of distance；The size of l depends on size k of transmission list and exchange factor δ (0 < δ < 1):

$l=\mathrm{\&delta;}\&times;\frac{k}{3}-1$

P3: if the individuality in the first level is less than the preset value of transmission list size, in ground floor Individuality in secondary will be first placed into transmission list, and remaining individuality will put into transmission row according to domination hierarchical sequence Table；The individuality of same level, preferably selects the individuality with larger crowding distance；This process is until sending Position is filled up in list；

4.4) each population is based on non-dominant level and crowding distance prepares to replace list, according to following preferential Order preparation replacement list:

P1: will be the first to be replaced in the individuality of last level, then replace individual in layer second from the bottom in order Body, by that analogy it is known that replace in list individual all use till；

P2: the individuality of same level, the individuality positioned at less congested area is replaced first.

The present invention has advantages below and a beneficial effect:

1. realize easily, there is stronger ability of searching optimum, fast convergence rate, the convergence high precision of solution Uniformly, it is more excellent to try to achieve solution combination property for the distributivity conciliate, and only need to once run just can provide multiple can The proportion scheme of row, is conducive to ingredients personnel to the control of blending process and management.

2. achieve multiple populations parallel alternatively to multi-objective problem optimization, there is ability of searching optimum by force, Fast convergence rate, the advantages of the distributivity that the convergence high precision of solution is conciliate is uniform, is solving high-precision copper plate/strip Blending optimization this need multiple representative problems of more excellent solution on show apparent advantage, It is not available for single goal intelligent optimization algorithm.

Brief description

Fig. 1 is the coevolution model of many Nidus Vespaes multiple target artificial bee colony algorithm；

Fig. 2 is the execution flow process of many Nidus Vespaes multiple target artificial bee colony algorithm.

Specific embodiment

Below in conjunction with the accompanying drawings and embodiment the present invention is described in further detail.

The present invention proposes a kind of new multi objective using for reference multiple Nidus Vespae Apiss cooperation foraging behaviors in the Nature The blending optimization method of artificial bee colony algorithm, the thought of its derivation algorithm is as shown in Figure 1.

Step 1: parameter initialization

1.1) carry out real coding using 2 × (n+1) matrix forms, matrix column number is the length of chromosome. The first row chromosome represents inventory x of raw material；Second row chromosome represents raw material smelting time t₀Molten with rising The smelting time t of body_q.(n+1)th integer of inventory row chromosome can correspond to and use melt constant q, is following Ring can be skipped in calculating, then the matrix form of chromosome is [(x, q), (t₀,t_q)]^t.Such as a kind of copper 5 kinds of raw materials of strip product needed, then its chromosome be represented by [(3,0.12,0.03,0.15,0.0153,2.5), (60,35,32,46,45,72)].It is 2 units that this coding stands plays melt weight, and overall smelting time is 72 Individual unit.The first raw material puts into 3 Unit Weights, and the time needing 60 units of melting, (charging time was After 12 unit interval), by that analogy.

1.2) determine initial population number, generally, whole group selects 2～4 populations, specifically Population number can be determined by specific optimization problem；Determine each population scale, generally, each Population comprises 30～80 individualities, and specific population scale can be determined by specific optimization problem；Initial at random Change each body position, determine maximum cycle mnc.

1.3) determine variable dimension d, i.e. the number of decision variable.The copper plate/strip dispensing participating in certain trade mark needs Want n kind raw material, represented with vector x, including new metal x₁, this trade mark old material x₂, other trade mark old materials x₃With Chemical waste x₄, in addition to new metal, be referred to as old material x ＇=x ＇ | x₂∪x₃∪x₄}；The throwing of n kind raw material The material time (i.e. feeding sequence) is represented with vectorial t；Play melt smelting time, i.e. smelting time t_q.If x_iFor The input amount of i-th kind of raw material, t_iFor the making time of i-th kind of raw material, k, m, p, n are positive integer, decision-making Variable-definition is as follows:

X=(x₁,x₂,…,x_n)；T=(t₁,t₂,…,t_n)；

x₁=(x₁,x₂,…,x_k)；x₂=(x_k+1,x_k+2,…,x_m)；

x₃=(x_m+1,x_m+2,…,x_p)；x₄=(x_p+1,x_p+2,…,x_n).

1.4) initialization important parameter " limit ":

Limit=mnc × d/2

Wherein, d is the number of decision variable, and mnc is maximum cycle.

Step 2: set up blending optimization multi-objective Model, and assess fitness value, be to solve for dispensing mould below The method of type, solution is exactly last dispensing result.

2.1) blending optimization multi-objective Model

Various materials price variances needed for dispensing are larger, so in blending process, meeting element On the premise of dividing ratio, need by adjusting input amount come reduces cost.In fusion process, due to chemistry The property of element itself, has melting loss of elements, needs in burdening calculation to consider that scaling loss amount carries out stokehold compensation. The charging time of raw material and smelting time greatly affect its burn out rate, and this results in compensation dosage and differs, and enters one Step impact feeds intake cost.Copper yield is relatively low, and price is higher, so the old material of copper plate/strip puts into smelting furnace again Produce, great to cost-effective, alleviation domestic copper plate/strip question meaning in short supply.Meeting the feelings of ingredients principle Under condition, should try one's best and bring old material into more.Based on above-mentioned consideration, set up following two object functions, if c_iFor i-th The price of middle raw material, c has been melt price, then cost of material:

$f_1\left(x\right)=\underset{i=1}{\overset{n}{\&sigma;}}{c}_i{x}_i$

Compensate and expect cost:

$f_2(x,t,t_q)=\underset{i=1}{\overset{n}{\&sigma;}}\&lsqb;c_i{x}_i{\mathrm{\&eta;}}_i(t_q-t_i)\&rsqb;+{\mathrm{cq\&eta;}}_q\left(t_q\right)$

Feed intake totle drilling cost:

$\begin{array}{c}{f}_1(x,t,t_q)=f_1\left(x\right)+f_2(x,t,t_q)\\ =\underset{i=1}{\overset{n}{\&sigma;}}{c}_i{x}_i\&lsqb;1+{\mathrm{\&eta;}}_i(t_q-t_i)\&rsqb;+{\mathrm{cq\&eta;}}_q\left(t_q\right)\end{array}$

Bring old material amount into:

$f_2\left(x^{\&prime;}\right)=f_2(x_{k+1},x_{k+2}\mathrm{...}{x}_n)=\underset{i=k+1}{\overset{m}{\&sigma;}}{\mathrm{\&alpha;}}_i{x}_i+\underset{i=m+1}{\overset{p}{\&sigma;}}{\mathrm{\&beta;}}_i{x}_i+\underset{i=p+1}{\overset{n}{\&sigma;}}{\mathrm{\&delta;}}_i{x}_i$

Wherein, η_i(t_q-t_i)=g_iln(h_it_i+ 1) the scaling loss time function for raw material j；η_q(t_q)=g_qln(h_qt_q+1) For playing function heat time heating time of melt burn out rate.α is the penalty factor that this trade mark old material brings deficiency into；β is it The penalty factor of deficiency brought into by his board for old material；δ brings the penalty factor of deficiency into for chemical waste.

In conjunction with both the above formula, the multi-goal optimizing function minimizing form can be written as:

min{f₁(x,t,t_q),-f₂(x')}

2.2) non-dominated ranking

The purpose of this method be according to solution between dominance relation come to whole population sequence.In order to which finds A little solutions are non-domination solution, and each solution will be compared with other solutions.For each solution, will calculate Two below: (a) n_x, the number of the solution of domination solution x；(b)s_x, the set of the solution that solution p can arrange.Right In a set p, quick non-dominated ranking process is as follows:

Step 1: setSymbol q represents in one of set p solution.If x is permissible Domination q, then s_x=s_x∪{q}；If q arranges x, n_x=n_x+1.If n_x=0, then x_rank=1, f₁=f₁∪ { p }, here f represent every layer of leading surface.P belongs to the leading surface of ground floor.

Step 2:i=1.

Step 3: setX ∈ f is solved for each_iIt is carried out following operation, each solves q ∈ s_x, make n_q=n_q–1.If n_p=0, then q_rank=i+1, q=q ∪ { q }.

Step 4: set i=i+1, f_i=q.Judge set q, if q is not empty set, return to step 3； Otherwise, program determination.

2.3) crowding distance calculates

Crowding distance be a point along object function direction to the average distance of its every side point, be used to estimate Calculate the density of other solutions around certain special solution in population.First, the set of non-domination solution is according to each mesh Target functional value ascending order arranges.Then calculate the crowding distance i of i-th solution_distance, it is to be solved by the i-th -1 Become cubical average side length with the i+1 system of solutions.Boundary Solutions (the minimum and maximum function of each object function The solution of value) crowding distance be appointed as infinity.Each object function corresponding, overall crowding distance is individual The summation of body crowding distance value.Before calculating crowding distance, each object function needs to be normalized place Reason.

2.4) formulate a fitness evaluation standard

In this algorithm, carried out towards the selection course in Pareto forward position according to crowded comparison operator, it Symbolization ＜_n.In population, each individual i has two attributes: non-dominant grade (i_rank) and crowded away from From (i_distance).If two solutions, in same grade, will be preferred positioned at more not crowded region. Otherwise, it is in different grades of two solutions, will be preferred in the solution of lower grade.Formulation statement As follows:

I ＜_nj iff i_rank＜ j_rankor(i_rank=j_rankand i_distance＞ j_distance)

Step 3: data updates

Each bee colony is divided into three types: employs honeybee, follows honeybee and investigation honeybee.In these people workers bee, Half is to employ honeybee, and second half is to follow honeybee and search bee.Employ honeybee and follow honeybee and execute in search space Development process, and search bee carries out the exploration in space.Around each nectar source, only one of which employs honeybee.

In this algorithm, randomly generate n people worker bee in search space first, n represents the big of each population Little.For feasible solution x_ij, i=1,2 ..., n, j are vector dimension, and its maximum is set to d.Complete initially After the change stage, following three iterative process will be executed:

3.7) employ the neighbour in nectar source corresponding to the random prospecting of honeybee, and nectar source position in its memory is repaiied Change.In artificial bee colony algorithm, this meaning is to find a new candidate solution according to an existing solution, This candidate solution is the neighbour having solved.New nectar source v_ijProduce according to formula below:

v_ij=x_ij+d_ij(x_ij-x_kj)

Wherein, k=1,2 ..., n；J=1,2 ..., d is the index randomly selecting.K is random determination, but can not Identical with i.δ_ijRandomly generate in the range of [- 1,1], this state modulator in x_ijThe product in the new nectar source of surrounding Raw, vivid illustrates the comparison to two nectar sources for the Apiss.

3.8) after finding new nectar source, x_ijNeed and v_ijIt is compared.Two nectar sources only preferably that Individual can remain.One follow honeybee select which nectar source depend on be associated with nectar source probit p_i, its calculation is as follows:

$p_i=\frac{{\mathrm{fit}}_i}{\underset{i=1}{\overset{n}{\&sigma;}}{\mathrm{fit}}_i}$

Wherein, fit_iIt is solution x_iFitness value, this fitness value is proportional to the number of Mel in the nectar source of position i Amount, its numerical expression is as follows:

${\mathrm{fit}}_i=\left\{\begin{array}{cc}{v}_i,& if\left(f\right(x_i)\&greaterequal;f(v_i\left)\right)\\ x_i,& if\left(f\right(x_i)<f(v_i\left)\right)\end{array}\right.$

3.9) if in a nectar source Mel quantity do not have in certain iterationses any improve if, this Individual nectar source will be employed honeybee to abandon by corresponding.The preset value of this iterations is artificial bee colony algorithm In an important control parameter.After nectar source is abandoned, employ honeybee will become search bee, and search bee will Find a new nectar source to replace that abandoned nectar source.The behavior of these people workers bee can be with public affairs the following Formula is describing:

$x_i^j=x_{\min }^j+\mathrm{\&sigma;}(x_{max}^j-x_{\min }^j)$

Wherein, σ is the random number in the range of [- 1,1].When search bee finds the abundanter honey of honey content Behind source, this search bee will become one and employ honeybee.

Step 4: in many Nidus Vespaes multiple target artificial bee colony is calculated, whole group is divided into several populations, often Individual population executes standard intraocular's ant colony algorithm respectively.After the optimization presetting algebraically, each population The individuality that selected section carries excellent Information is used for information exchange.Selected one list of individual composition, this Individual list will be sent in another population.Additionally, each population will prepare a replacement list, Individuality in this list will be replaced by the individuality coming from other populations.In order to preferably obtain transmission list, if Count a strategy considering the factor such as nectar source position, non-dominant level and crowding distance near Nidus Vespae.

4.5) scale of transmission list is fixed value set in advance.Each population is according to lower column major order It is ready for sending list.

P1: select individual First Principles to be non-dominant level, the individuality in low level will preferably select.

P2: if the individuality in the first level is bigger than the preset value of transmission list size, need to calculate The individual Hamming distance of each pair and crowding distance in population.Calculate and have maximum Hamming distance apart from other individualities Individuality, then select from it nearest l individual enter transmission list, finally select to have larger crowded The individuality of distance.The size of l depends on size k of transmission list and exchange factor δ (0 < δ < 1):

$l=\mathrm{\&delta;}\&times;\frac{k}{3}-1$

P3: if the individuality in the first level is less than the preset value of transmission list size, in ground floor Individuality in secondary will be first placed into transmission list, and remaining individuality will put into transmission row according to domination hierarchical sequence Table.The individuality of same level, preferably selects the individuality with larger crowding distance.This process is until sending Position is filled up in list.

4.6) each population is based on non-dominant level and crowding distance prepares to replace list.According to following preferential Order prepares to replace list.

P1: will be the first to be replaced in the individuality of last level, then replace individual in layer second from the bottom in order Body, by that analogy it is known that replace in list individual all use till.

P2: the individuality of same level, the individuality positioned at less congested area is replaced first.

Step 5: if end condition not up to set in advance, return to step 2), if reaching iteration ends Condition, then stop calculating, last output result.

The solution procedure of multiple target artificial bee colony algorithm and flow process, from initialization, assess fitness, data is more Newly, until iteration ends, as shown in Figure 2.

Claims (6)

1. a kind of blending optimization method based on new multi objective artificial bee colony algorithm is it is characterised in that comprise the following steps:

Step 1: dispensing parameter initialization；

Step 2: set up blending optimization multi-objective Model, and assess fitness value；

Step 3: data updates；

Step 4: after the optimization presetting algebraically, the individuality that each population selected section carries excellent Information is used for information exchange；Selected one list of individual composition, this list will be sent in another population；Each population will prepare a replacement list, and the individuality in this list will be replaced by the individuality coming from other populations；

Step 5: if end condition not up to set in advance, return to step 2, if reaching stopping criterion for iteration, stop calculating, last output result.

2. a kind of blending optimization method based on new multi objective artificial bee colony algorithm according to claim 1, it is characterised in that described dispensing parameter initialization, comprises the following steps:

Carry out real coding using 2 × (n+1) matrix forms, n is raw material type number；

Determine initial population number, each population scale, each body position of random initializtion, determine maximum cycle mnc；

Determine variable dimension d, i.e. the number of decision variable；

Initialization important parameter " limit ":

limit=mnc×d/2

Wherein, d is the number of decision variable, and mnc is maximum cycle.

3. a kind of blending optimization method based on new multi objective artificial bee colony algorithm according to claim 2 is it is characterised in that described 2 × (n+1) matrix forms are [(x, q), (t₀,t_q)]^t, matrix column number is the length of chromosome, and the first row chromosome represents inventory x of raw material, and the second row chromosome represents raw material smelting time t₀With the smelting time t playing melt_q, (n+1)th integer of inventory row chromosome is corresponding to have used melt constant q.

4. a kind of blending optimization method based on new multi objective artificial bee colony algorithm according to claim 1 it is characterised in that described blending optimization multi-objective Model particularly as follows:

Set up following two object functions, if c_iFor the price of raw material in i-th, x_iFor the input amount of i-th kind of raw material, c has been melt price, then cost of material:

Compensate and expect cost:

Wherein, t_qFor playing the smelting time of melt, t_iMaking time for i-th kind of raw material；

Feed intake totle drilling cost:

Bring old material amount into:

Wherein, η_i(t_q-t_i)=g_iln(h_it_i+ 1) the scaling loss time function for raw material j；η_q(t_q)=g_qln(h_qt_q+ 1) be melt burn out rate function heat time heating time；α is the penalty factor that this trade mark old material brings deficiency into；β is that other boards bring the penalty factor of deficiency into for old material；δ brings the penalty factor of deficiency into for chemical waste；

In conjunction with totle drilling cost f that feeds intake₁With bring old material amount f into₂, for represent blending optimization multi-objective Model minimize form multi-goal optimizing function be written as:

min{f₁(x,t,t_q),-f₂(x')} .

5. a kind of blending optimization method based on new multi objective artificial bee colony algorithm according to claim 1 is it is characterised in that the renewal of described data comprises the following steps:

Randomly generate n people worker bee in search space first, n represents the size of each population, for feasible solution x_ij, i=1,2 ..., n, j are vector dimension, and its maximum is set to d；After completing initial phase, following three iterative process will be executed:

3.1) employ the neighbour in nectar source corresponding to the random prospecting of honeybee, and modify in nectar source position during it is remembered, and finds a new candidate solution according to an existing solution, this candidate solution is the neighbour having solved, new nectar source v_ijProduce according to formula below:

v_ij=x_ij+δ_ij(x_ij-x_kj)

Wherein, k=1,2 ..., n；J=1,2 ..., d are the indexes randomly selecting, and k is random determination, but can not be identical with i；δ_ijRandomly generate in the range of [- 1,1], this state modulator in x_ijThe generation in the new nectar source of surrounding.

3.2) after finding new nectar source, x_ijNeed and v_ijIt is compared, only preferably that can remain for two nectar sources, follow honeybee for one and select what which nectar source depended on being associated with nectar source probit p_i, its calculation is as follows:

Wherein, fit_iIt is solution x_iFitness value, this fitness value is proportional to the quantity of Mel in the nectar source of position i, and its numerical expression is as follows:

3.3) if in a nectar source Mel quantity do not have in certain iterationses any improve if, this nectar source will be employed honeybee to abandon by corresponding, after nectar source is abandoned, employ honeybee will become search bee, and search bee to replace that abandoned nectar source by finding a new nectar source, the behavior of these people workers bee is described with formula below:

Wherein, σ is the random number in the range of [- 1,1], and after search bee finds honey content abundanter nectar source, this search bee will become one and employ honeybee.

6. a kind of blending optimization method based on new multi objective artificial bee colony algorithm according to claim 1, it is characterised in that described list will be sent in another population, comprises the following steps:

4.1) scale of transmission list is fixed value set in advance, and each population is ready for sending list according to lower column major order:

P1: select individual First Principles to be non-dominant level, the individuality in low level will preferably select；

P2: if the individuality in the first level is bigger than the preset value of transmission list size, need to calculate the individual Hamming distance of each pair and crowding distance in population, calculate the individuality having maximum Hamming distance apart from other individualities, then select, from its l nearest individual entrance transmission list, finally to select the individuality with larger crowding distance；The size of l depends on size k of transmission list and exchange factor δ (0 < δ < 1):

P3: if the individuality in the first level is less than the preset value of transmission list size, the individuality in the first level will be first placed into transmission list, remaining individuality will put into transmission list according to domination hierarchical sequence；The individuality of same level, preferably selects the individuality with larger crowding distance；This process fills up position until transmission list；

4.2) each population is based on non-dominant level and crowding distance prepares to replace list, according to the preparation replacement list of following priority:

P1: will be the first to be replaced in the individuality of last level, then replace the individuality in layer second from the bottom in order, by that analogy it is known that replace in list individual all use till；

P2: the individuality of same level, the individuality positioned at less congested area is replaced first.