CN104537425A - Optimized scheduling method for production and assembly process of automobile air conditioner air outlet - Google Patents

Abstract

The invention relates to an optimized scheduling method for a production and assembly process of an automobile air conditioner air outlet, and belongs to the technical field of intelligent optimized scheduling in production workshops. An optimization objective and a scheduling model in the production and assembly process of the automobile air conditioner air outlet are determined, and the optimized scheduling method for an adaptive hybrid distribution estimation algorithm is used for optimizing the optimization objective, wherein the scheduling model is built according to the processing time of parts of the automobile air conditioner air outlet on each machine and the final assembly completion time, and the minimum average completion time is used as the optimization objective. The optimized scheduling method has the advantages that the expression of the production and assembly process of the automobile air conditioner air outlet is clear and accurate, and high-quality regions found through global searching are subjected to intensive local searching, so that better balance is achieved between the global searching and the local searching.

Description

A kind of Optimization Scheduling of production assembling process of vehicle air conditioning outlet

Technical field

The present invention relates to a kind of Optimization Scheduling of production assembling process of vehicle air conditioning outlet, belong to workshop intelligent optimization dispatching technique field.

Background technology

Current domestic automobile interior trim part production company has entered meagre profit developing stage, can the product how producing more high-quality, more low cost be related to enterprise and have survived, in the face of so vital problem, each company is all in the various countermeasure of consideration, and one of them good method is exactly, making great efforts to produce assembly process by optimizing, shortening man-hour with reduced cost, improving the price competitiveness of product.

In Modern Manufacturing Enterprise, assembly work amount accounts for the 20%-70% of whole product work amount, and average out to 45% especially accounts for the 40%-60% of whole production time with the installation time such as automobile, locomotive.In recent years because main engine plants more and more pay close attention to integral vehicle cost, and the technology going out automotive air outlet is more and more ripe, only provide more attractive in appearance, hommization, high-quality, air outlet at a low price, just can make oneself to establish oneself in an unassailable position.And in the market of China, price is undoubtedly the deadly defect of competition among enterprises.Current China is in the stage of manufacturing industry fast development, and automaticity is quite low, and therefore, reasonable balance assembling line, raises the efficiency and can bring huge interests for enterprise.

In the production assembling process of vehicle air conditioning outlet, mainly comprise the processing of the part of vehicle air conditioning outlet, transport and 3 stages of assembling.Often kind of vehicle air conditioning outlet needs 4 parts processed respectively on first stage 4 equipment to assemble in the phase III; 4 parts corresponding to often kind of vehicle air conditioning outlet after the first stage all machines, are then collected by subordinate phase at once, to be transported to phase III etc. to be assembled; The part that various vehicle air conditioning outlet is corresponding processes through three phases successively by vehicle air conditioning outlet processing sequence; Arbitrary process equipment of first stage can only process a kind of part at synchronization, and be with sequence to be correlated with setup times between different part, setup times depends on processing sequence; The mounting equipment of phase III can only assemble same vehicle air conditioning outlet at synchronization, and between different vehicle air conditioning outlet, setup times is 0.

The production assembling process of vehicle air conditioning outlet belongs to three stage assembling line production runes of a quasi-representative, the defined this kind of assembling line of academia is three stage assembling lines (Three-Stage Assembly Flowshop, TSAF), and prove that the TSAF scheduling problem that first stage contains more than two machines belongs to np hard problem, namely there is not the algorithm in a polynomial time and soluble challenge.Obviously, the first stage contains the TSAF scheduling problem (that is: the production assembling line scheduling problem of vehicle air conditioning outlet) of 4 number of machines, also belongs to np hard problem category.This problem is reasonably dispatched, the production efficiency of the production assembling line system of vehicle air conditioning outlet can be significantly improved.

Because the production assembling line scheduling problem of vehicle air conditioning outlet is np hard problem, make traditional mathematic programming methods cannot solve this problem, therefore, the present invention designs a kind of ADAPTIVE MIXED Estimation of Distribution Algorithm (ImprovedAdaptive Estimation of Distribution Algorithm of improvement, IAEDA) Optimization Scheduling, can obtain the excellent solution of the scheduling problem of the production assembling process of vehicle air conditioning outlet within a short period of time.

Summary of the invention

Technical matters to be solved by this invention is the problem of the excellent solution of the production assembling process scheduling problem obtaining vehicle air conditioning outlet within a short period of time, provides a kind of Optimization Scheduling of production assembling process of vehicle air conditioning outlet.

Technical scheme of the present invention is: a kind of Optimization Scheduling of production assembling process of vehicle air conditioning outlet, by determining production assembling process scheduling model and the optimization aim of vehicle air conditioning outlet, and the Optimization Scheduling of ADAPTIVE MIXED Estimation of Distribution Algorithm is used to be optimized optimization aim; Wherein scheduling model is set up, with minimized average completion date for optimization aim with the final assembling deadline according to the part of vehicle air conditioning outlet the process time on each machine:

$C_T\left({\mathrm{\π}}_i^p\right)=\max \left\{\underset{k=1,...,m}{\max }\right\{{\mathrm{\Σ}}_{j=1}^i(S({\left[{\mathrm{\π}}_{j-1}^p\right]}^k,{\left[{\mathrm{\π}}_j^p\right]}^k)+p\left(\right[{{\mathrm{\π}}_j^p]}^k))\},C_T\left({\mathrm{\π}}_{i-1}^p\right)\}+p_T\left({\mathrm{\π}}_i^p\right),$

$C\left({\mathrm{\π}}_i^p\right)=\max (C_T\left({\mathrm{\π}}_i^p\right),C\left({\mathrm{\π}}_{i-1}^p\right))+p_A\left({\mathrm{\π}}_i^p\right)$

$\stackrel{\‾}{C}\left({\mathrm{\π}}^P\right)={\mathrm{\Σ}}_{i=1}^{n}C\left({\mathrm{\π}}_i^P\right)/n$

${\mathrm{\π}}^{P*}=\arg \left\{\stackrel{\‾}{C}\left({\mathrm{\π}}^P\right)\right\}\→\min ,\∀{\mathrm{\π}}^P\∈\∏$

Wherein, ${\mathrm{\π}}^P=[{\mathrm{\π}}_1^p,{\mathrm{\π}}_2^p,...,{\mathrm{\π}}_n^p]({\mathrm{\π}}_i^p\∈\{1,...,n\},i=1,...,n)$ For a n to be processed vehicle air conditioning outlet is based on the arrangement of processing sequence, ${\left[{\mathrm{\π}}_i^p\right]}^k({\mathrm{\π}}_i^p\⊃\{{\left[{\mathrm{\π}}_i^p\right]}^k|k=\mathrm{1,2},...,m\},i=1,...,n)$ For assembling motor vehicle air-conditioner air outlet need the vehicle air conditioning outlet part processed on first stage kth platform equipment, for part and part between setup times and $S\left(\right[{{\mathrm{\π}}_0^p]}^k,{\left[{\mathrm{\π}}_1^p\right]}^k)>0,p\left({\left[{\mathrm{\π}}_i^p\right]}^k\right)$ For part process time, for subordinate phase collection, transport belong to vehicle air conditioning outlet all parts to time of phase III mounting equipment, for belonging to vehicle air conditioning outlet all parts through the first stage processing and subordinate phase collect, transport point need the maximum processing time and for vehicle air conditioning outlet at the built-up time of phase III, for vehicle air conditioning outlet deadline and for the average completion time of all vehicle air conditioning outlets; Optimization aim is in the set Π of all vehicle air conditioning outlet processing sequences, find a π ^p*, make objective function minimum.

The concrete steps of the Optimization Scheduling of described ADAPTIVE MIXED Estimation of Distribution Algorithm are as follows:

Step1, coded system: encode with the process and assemble of often kind of vehicle air conditioning outlet order wherein n is the number of vehicle air conditioning outlet to be processed, for vehicle air conditioning outlet to be assembled need the part processed on first stage kth platform equipment;

Step2, population and probability model initialization: population scale is M, adopt random device to produce initialization population, until the quantity of initial solution reaches the requirement of population scale; The matrix P (gen) of n × n dimension is adopted to represent the probability Distribution Model in algorithm gen generation;

Wherein, P _i(gen)=[P _i1(gen), P _i2(gen) ..., P _in(gen)] be the i-th every trade vector in P (gen), Pi _j(gen) be P (gen) the i-th row jth column element and represent gen for time vehicle air conditioning outlet j individual or separate i-th on the probability that occurs p (gen) from the processing precedence relationship numerically reflecting different vehicle air conditioning outlet, P _ij(gen) larger, represent gen for time the vehicle air conditioning outlet j probability that occurs on i-th of individuality larger;

Step3, sampling produce new population: adopt the mode of roulette to sample to probability model;

Step4, jump out " Insert " mutation operation of principle based on improving first: use and improve " Insert " mutation operation of jumping out principle first to the optimum individual in population or separate and perform Local Search;

Step5, update probabilistic model: first judge that " optimum individual " whether continuous ten generations obtained do not upgrade, and if so, then carry out from new initialization to probability matrix; Otherwise adopt the learning rate based on information entropy and aberration rate self-adaptative adjustment mechanism, " optimum individual " that use algorithm to find in search procedure upgrades probability matrix;

Step6, end condition: the maximum iteration time of setting end condition is 200, if met, then export " optimum individual "; Otherwise go to step Step3, iterate, until meet end condition.

Principle of work of the present invention is:

Step 1: production assembling process scheduling model and the optimization aim of setting up vehicle air conditioning outlet.

Scheduling model is set up, with minimized average completion date for optimization aim with the final assembling deadline according to the part of vehicle air conditioning outlet the process time on each machine

$C_T\left({\mathrm{\π}}_i^p\right)=\max \left\{\underset{k=1,...,m}{\max }\right\{{\mathrm{\Σ}}_{j=1}^i(S({\left[{\mathrm{\π}}_{j-1}^p\right]}^k,{\left[{\mathrm{\π}}_j^p\right]}^k)+p\left(\right[{{\mathrm{\π}}_j^p]}^k))\},C_T\left({\mathrm{\π}}_{i-1}^p\right)\}+p_T\left({\mathrm{\π}}_i^p\right),$

$C\left({\mathrm{\π}}_i^p\right)=\max (C_T\left({\mathrm{\π}}_i^p\right),C\left({\mathrm{\π}}_{i-1}^p\right))+p_A\left({\mathrm{\π}}_i^p\right)$

$\stackrel{\‾}{C}\left({\mathrm{\π}}^P\right)={\mathrm{\Σ}}_{i=1}^{n}C\left({\mathrm{\π}}_i^P\right)/n$

${\mathrm{\π}}^{P*}=\arg \left\{\stackrel{\‾}{C}\left({\mathrm{\π}}^P\right)\right\}\→\min ,\∀{\mathrm{\π}}^P\∈\∏$

Wherein, ${\mathrm{\π}}^P=[{\mathrm{\π}}_1^p,{\mathrm{\π}}_2^p,...,{\mathrm{\π}}_n^p]({\mathrm{\π}}_i^p\∈\{1,...,n\},i=1,...,n)$ For a n to be processed vehicle air conditioning outlet is based on the arrangement of processing sequence, ${\left[{\mathrm{\π}}_i^p\right]}^k({\mathrm{\π}}_i^p\⊃\{{\left[{\mathrm{\π}}_i^p\right]}^k|k=\mathrm{1,2},...,m\},i=1,...,n)$ For assembling motor vehicle air-conditioner air outlet need the vehicle air conditioning outlet part processed on first stage kth platform equipment, $(S\left(\right[{{\mathrm{\π}}_0^p]}^k,{\left[{\mathrm{\π}}_1^p\right]}^k)>0)$ For part and part between setup times, for part process time, for subordinate phase collection, transport belong to vehicle air conditioning outlet all parts to time of phase III mounting equipment, for belonging to vehicle air conditioning outlet all parts through the first stage processing and subordinate phase collect, transport point need the maximum processing time, for vehicle air conditioning outlet at the built-up time of phase III, for vehicle air conditioning outlet deadline, for the average completion time of all vehicle air conditioning outlets; Optimization aim is in the set Π of all vehicle air conditioning outlet processing sequences, find a π ^p*, make objective function minimum.

Step 2: the expression of solution.

The present invention proposes the coded system based on vehicle air conditioning outlet processing sequence.For example, for the production assembling process scheduling problem of the vehicle air conditioning outlet of n=6, m=3, just separate or arrangement for one of this problem.This arrangement represents that the vehicle air conditioning outlet being numbered 6 is processed at first, is namely numbered the part [6] corresponding to vehicle air conditioning outlet of 6 ¹, [6] ², [6] ³process at first, then at first through subordinate phase and the process of phase III on first stage the 1st, 2,3 equipment respectively; Next is that the vehicle air conditioning outlet the 2nd being numbered 3 is processed, and is namely numbered the part [3] corresponding to vehicle air conditioning outlet of 3 ¹, [3] ², [3] ³and then part [6] on first stage the 1st, 2,3 equipment respectively ¹, [6] ², [6] ³process, then the 2nd through subordinate phase and the process of phase III; The ensuing vehicle air conditioning outlet being numbered 5,1,4,2 is processed successively.For the problem of other different scales, just the value of n with m may be different, expression way with go up above together.

Step 3: initialization of population strategy.

When initialization of population, a part is individual or separate the SPT generate rule adopting expansion, and it is individual that this can guarantee that initial population comprises certain high-quality, can accumulate high-quality solution information when making probability distribution matrix utilize the sub-population of the advantage in population to upgrade for the first time; Remainder individuality adopts random fashion to generate, and this is conducive to the diversity and the dispersiveness that keep population.Wherein, the SPT generate rule mode of expansion is: 1) carry out ascending sort by process time respectively to the part that first stage every platform equipment is processed, so just there is m sequence, then the vehicle air conditioning outlet of the part in each sequence belonging to it is replaced, m individuality can be obtained; 2) 1) in " carrying out ascending sort by process time " replace with " carrying out ascending sort by process time and setup times sum ", then perform and 1) identical operation, other m individuality can be obtained; 3) by built-up time, ascending order arrangement is carried out to the vehicle air conditioning outlet that phase III equipment is assembled, 1 individuality can be obtained; 4) average processing time of all parts of each vehicle air conditioning outlet on first stage m platform machine is obtained, ask each vehicle air conditioning outlet first stage average processing time again and it is in rear two stage processing time sum, then vehicle air conditioning outlet is sorted by this and value, 1 individuality can be obtained.2m+2 individuality can be produced like this.

Step 4: probability Distribution Model initialization strategy.

IAEDA adopts the matrix P (gen) of n × n dimension to represent the probability Distribution Model in algorithm gen generation, that is:

Wherein, P _i(gen)=[P _i1(gen), P _i2(gen) ..., P _in(gen)] be the i-th every trade vector in P (gen), Pi _j(gen) be P (gen) the i-th row jth column element and represent gen for time vehicle air conditioning outlet j individual or separate i-th on the probability that occurs p (gen) from the processing precedence relationship numerically reflecting different vehicle air conditioning outlet, Pi _j(gen) larger, represent gen for time the vehicle air conditioning outlet j probability that occurs on i-th of individuality larger;

When algorithm is initial (gen=0), setting Pi _j(0)=1/ (n × n), i, j=1 ..., n.Relative to usual manner setting Pi _j(0)=1/n, setting P _ij(0)=1/ (n × n) can make P (gen) upgrade (being updated to P (1) by P (0)) first and accumulate high-quality individual information in initial population more after carrying out " OK " normalization, i-th (i=1 in high-quality individuality can be increased, n) corresponding product bj (bj ∈ { 1,, n}) and at P _i(gen) in, the numerical value of bj row, makes vehicle air conditioning outlet bj to P _i(gen) when roulette sampling generates new individual i-th, selected probability increases.This is conducive to bootstrap algorithm and searches near high-quality individuality, suitably can improve the search capability at algorithm initial stage.

Step 5: probability Distribution Model adaptive updates mechanism.

For the algorithm based on EDA, probability Distribution Model determines the direction of search of algorithm, and its update mechanism has considerable influence to algorithm performance.For IAEDA, the renewal of probability Distribution Model or matrix P (gen) depends primarily on learning rate and aberration rate.First, less learning rate or larger aberration rate are conducive to the maintenance of population diversity level, and then algorithm can obtain good search width, but speed of convergence can be slack-off, even do not restrain, and cause the search depth of algorithm to be difficult to be guaranteed; Secondly, larger learning rate or less aberration rate can accelerate convergence of algorithm speed, and algorithm will have good search depth, but algorithm is easy to be absorbed in local optimum, and then cause Premature Convergence, and the search width of algorithm cannot be kept.How reasonable set learning rate and aberration rate in algorithm evolution process, thus guarantee that algorithm finds reasonable balance between search width and the degree of depth, be the key effectively improving algorithm performance.Therefore, the learning rate based on information entropy and aberration rate self-adaptative adjustment mechanism is proposed, for upgrading P (gen).Make E (gen) be the information entropy of gen for P (gen), namely have:

$E\left(\mathrm{gen}\right)=-{\mathrm{\Σ}}_{i=1}^n{\mathrm{\Σ}}_{j=1}^n\left\{P_{\mathrm{ij}}\left(\mathrm{gen}\right)\ln P_{\mathrm{ij}}\left(\mathrm{gen}\right)\right\}$

From algorithm 1st generation, along with running or the increase of evolutionary generation gen, Pi _j(gen) will gradually to 0 or 1 near (optimum solution or element corresponding to optimum individual close to 1, other elements are close to 0), make E (gen) constantly reduce and finally trend towards 0.As can be seen here, the evolutionary process of IAEDA is also the process that E (gen) declines gradually, and E (gen) can reflect the evolution degree of algorithm to a certain extent.Therefore, the learning rate r (gen) in algorithm gen generation adopts following formula to calculate:

$r\left(\mathrm{gen}\right)=\left\{\begin{array}{cc}{r}_f,& 0<E\left(\mathrm{gen}\right)\&le;E_0\\ \exp [a\&times;(E\left(\mathrm{gen}\right)/E_0-1)](r_f-r_{\min })+r_{\min },& E_0<E\left(\mathrm{gen}\right)\end{array}\right.$

Wherein, E ₀for information entropy threshold value, be set as E ₀=0.6E _m(E _m=nlnn is the upper limit of E (gen)), r _ffor final learning rate (r _f=0.08), r _minfor minimum learning rate (r _min=0.02 and r _min< r _f< 1), α is controling parameters (value is generally between 2 to 6).At the algorithm evolution initial stage, r (gen) value is relatively large, and along with the increase of gen, E (gen) moves closer to E ₀, and r (gen) reduces gradually and is tending towards r _f.

In order to guarantee the directivity of algorithm search preferably, IAEDA adopts algorithm to gen on behalf of stopping found optimum individual ${\mathrm{\&pi;}}^{\mathrm{Pbest}}\left(\mathrm{gen}\right)=[{\mathrm{\&pi;}}_1^{\mathrm{pbest}}\left(\mathrm{gen}\right),{\mathrm{\&pi;}}_2^{\mathrm{pbest}}\left(\mathrm{gen}\right),...,{\mathrm{\&pi;}}_n^{\mathrm{pbest}}\left(\mathrm{gen}\right)]$ As the sub-population of advantage, P (gen) is upgraded.B (gen) is made to be the renewal matrix that gen ties up for n × n.Element in B (gen) is except when its subscript (i, j) belongs to time value be outside 1, all the other values are 0.P (gen) adopts following formula to upgrade:

P(gen+1)＝(1-r(gen))×P(gen)+(r(gen)/n)×B(gen)

At the algorithm evolution initial stage, r (gen) value is larger, like this P (gen) is updated to P (gen+1) and after further " OK " normalization being done to P (gen+1), more obviously can increases in Pi (gen+1) row numerical value (i=1 ..., n), make product to P _i(gen+1) when roulette sampling generates new individual i-th, selected probability increases.This contributes to population at π ^pbest(gen) near, search, can increase dynamics and the degree of depth of search.Along with the increase of gen, r (gen) value reduces gradually, at this moment not only vehicle air conditioning outlet when generating i-th of new individuality, selected probability diminishes relatively, and the speed of convergence (element namely in P (gen+1) is tending towards the speed of 0 or 1 gradually) of P (gen+1) is suitably slowed down simultaneously.This is conducive to, in the diversity of algorithm later stage maintenance population, can improving width and the precision of search.

Production assembling process scheduling problem due to vehicle air conditioning outlet is strong NP-hard problem, solution space is very complicated, for avoiding algorithm to be absorbed in local optimum too early, from gen=1, after P (gen) is updated to P (gen+1), according to adapt variance P _m(gen) certain disturbance or variation are carried out to P (gen+1), thus increase the diversity of population further.P _m(gen) following formula is adopted to calculate:

P _M(gen)＝P _min×exp[lnβ×((E _M-E(gen))/E _M)]

Wherein, P _minfor aberration rate lower limit (P _min=0.4), P _m(gen) be the aberration rate in gen generation, β is amplification controling parameters (β=2).Along with the increase of gen, E (gen) is tending towards 0 gradually, and P _m(gen) β P will be increased to gradually _min.

Make random (0,1) for producing equally distributed random number between [0,1], random (0 or 1) is random generation 0 or 1, C (gen) is the Variation Matrix that gen ties up for n × n.Each element in C (gen) in every generation respectively by random (0 or 1) assignment.After execution r (gen) is updated to P (gen+1) to P (gen), if random (0,1) < P _m(gen) set up and then adopt following formula to make a variation to P (gen+1):

P(gen+1)＝(1-r _M(gen))P(gen+1)+r _M(gen)×C(gen)

Wherein, r _m(gen) be variation rate.In order to suitably strengthen the effect of variation, therefore by r _m(gen) value is r (gen)/2.Along with the increase of gen, P _m(gen) become large gradually, then the probability performed P (gen+1) makes a variation increases, and this effectively can slow down the speed of convergence of P (gen+1), thus prevents Premature Convergence to a certain extent.

For avoiding algorithm to be absorbed in local optimum further, from gen=1, after calculating the information entropy E of gen for P (gen) (gen), if optimum individual π in population ^pbest(gen) in continuous 10 generations, do not change or E (gen) is less than 1, then force to perform initialization operation to P (gen) to reset, guarantee that it can search for more zones of different by bootstrap algorithm, and then utilize learning rate and aberration rate to upgrade it to the P (gen) after resetting.In addition, after execution learning rate and aberration rate have upgraded it, " OK " normalized need have been carried out to P (gen+1), namely ensure P _i(gen+1) in, element sum is 1, so that next step adopts roulette to generate new population.

Step 6: new population sampling generation method.

Sampling generates new population, is exactly for the P after i-th renewal of individuality each in population to gen+1 _i(gen+1) carry out roulette sampling generate (i=1 ..., n).For guaranteeing that high-quality is individual further ${\mathrm{\&pi;}}^{\mathrm{Pbest}}\left(\mathrm{gen}\right)=[{\mathrm{\&pi;}}_1^{\mathrm{pbest}}\left(\mathrm{gen}\right),{\mathrm{\&pi;}}_2^{\mathrm{pbest}}\left(\mathrm{gen}\right),...,{\mathrm{\&pi;}}_n^{\mathrm{pbest}}\left(\mathrm{gen}\right)]$ Information retained, generate each new individual time, first from π ^pbest(gen) Stochastic choice L (L=0.2n) individual position in, remains into the element of its correspondence or product in new individual same position, and the element then remaining n-L position is determined by roulette sampling again.In roulette sampling process, if the element that a certain position is chosen is with fixed arbitrary element is identical before, then re-start sampling, until choose different element.

Step 7: based on FindFirstSkipN _insertthe Local Search of neighborhood.

For strengthening the local search ability of IAEDA, to the best individuality of history in generated new population or the search of execution based on " Insert " field can be separated.Make N _insert(π, u, v) is for being inserted into the element on u position in arrangement π or product on v position.π is based on N in arrangement _insertthe neighborhood of (π, u, v) can be expressed as:

N _insert(π)＝{π ^temp＝Insert(π,u,v)|v≠u,u-1；u,v＝1,2,…,n}

Improve the Insert field jumping out principle first to search for as searching N _insertthe more excellent neighborhood solution of (π, u, v) first then jumps out previous cycle and using this neighborhood solution as current optimum solution.Based on above-mentioned definition, FindFirstSkipN _insertthe step of (π, u, v) is as follows:

Step 7.1: make u=1, v=2,

Step 7.2: ${\mathrm{\&pi;}}^{\mathrm{temp}}=\mathrm{FindFirstSkip}{N}_{\mathrm{Insert}}({\mathrm{\&pi;}}_{\mathrm{gbest}\_1}^{\mathrm{gen}},u,v);$

Step 7.3: if $f\left({\mathrm{\&pi;}}^{\mathrm{temp}}\right)<f\left({\mathrm{\&pi;}}_{\mathrm{gbest}\_1}^{\mathrm{gen}}\right)$ Then ${\mathrm{\&pi;}}_{\mathrm{gbest}\_1}^{\mathrm{gen}}={\mathrm{\&pi;}}^{\mathrm{temp}},f\left({\mathrm{\&pi;}}_{\mathrm{gbest}\_1}^{\mathrm{gen}}\right)=f\left({\mathrm{\&pi;}}^{\mathrm{temp}}\right);$

Step 7.3.1:u=u+1;

Step 7.3.2: if u≤n, forward step 7.2 to, otherwise, forward step 7.5 to;

Step 7.4: if $f\left({\mathrm{\&pi;}}^{\mathrm{temp}}\right)<f\left({\mathrm{\&pi;}}_{\mathrm{gbest}\_1}^{\mathrm{gen}}\right)$ Then v=v+1;

Step 7.4.1: if v≤n and v ≠ u, forward step 7.2 to, otherwise forward step 7.3.1 to;

Step 7.5: export

Step 8: judge whether to export optimum results.

As reached the maximum iteration time 200 of setting, then export " optimum individual "; Otherwise, make gen=gen+1, return step 5.

The invention has the beneficial effects as follows:

1, propose scheduling model and the optimization aim of the production assembling process of vehicle air conditioning outlet, make the expression of the production assembling process of vehicle air conditioning outlet accurately clear;

2, propose probability Distribution Model initialization strategy, make P (0) to accumulate more effective information, for algorithm subsequent searches improves good initial search area;

3, propose a kind of distributed model adaptive updates mechanism, utilize information entropy to carry out the evolution degree of metric algorithm and real-time update P (gen+1) according to this, and generate according to P (gen+1) new population that retains defect mode and realize global search;

4, comparatively careful Local Search is carried out to the quality area that global search is found, thus between the overall situation and Local Search, arrive balance preferably; IAEDA is made to be expected to become the efficient algorithm of the scheduling model of the production assembling process solving vehicle air conditioning outlet.

Accompanying drawing explanation

Fig. 1 is the production assembling process schematic diagram of vehicle air conditioning outlet in the present invention;

Fig. 2 is the production assembling process Gantt chart of two kinds of vehicle air conditioning outlets;

Fig. 3 is algorithm flow chart of the present invention;

Fig. 4 is that in the present invention, problem scale is the expression schematic diagram that n=6 separates;

Fig. 5 is the variation schematic diagram based on " Insert " of the present invention.

Embodiment

Embodiment 1: as Figure 1-5, a kind of Optimization Scheduling of production assembling process of vehicle air conditioning outlet, by determining production assembling process scheduling model and the optimization aim of vehicle air conditioning outlet, and the Optimization Scheduling of ADAPTIVE MIXED Estimation of Distribution Algorithm is used to be optimized optimization aim; Wherein scheduling model is set up, with minimized average completion date for optimization aim with the final assembling deadline according to the part of vehicle air conditioning outlet the process time on each machine:

$C_T\left({\mathrm{\&pi;}}_i^p\right)=\max \left\{\underset{k=1,...,m}{\max }\right\{{\mathrm{\&Sigma;}}_{j=1}^i(S({\left[{\mathrm{\&pi;}}_{j-1}^p\right]}^k,{\left[{\mathrm{\&pi;}}_j^p\right]}^k)+p\left(\right[{{\mathrm{\&pi;}}_j^p]}^k))\},C_T\left({\mathrm{\&pi;}}_{i-1}^p\right)\}+p_T\left({\mathrm{\&pi;}}_i^p\right),$

$C\left({\mathrm{\&pi;}}_i^p\right)=\max (C_T\left({\mathrm{\&pi;}}_i^p\right),C\left({\mathrm{\&pi;}}_{i-1}^p\right))+p_A\left({\mathrm{\&pi;}}_i^p\right)$

$\stackrel{\&OverBar;}{C}\left({\mathrm{\&pi;}}^P\right)={\mathrm{\&Sigma;}}_{i=1}^{n}C\left({\mathrm{\&pi;}}_i^P\right)/n$

${\mathrm{\&pi;}}^{P*}=\arg \left\{\stackrel{\&OverBar;}{C}\left({\mathrm{\&pi;}}^P\right)\right\}\&RightArrow;\min ,\&ForAll;{\mathrm{\&pi;}}^P\&Element;\&prod;$

Wherein, ${\mathrm{\&pi;}}^P=[{\mathrm{\&pi;}}_1^p,{\mathrm{\&pi;}}_2^p,...,{\mathrm{\&pi;}}_n^p]({\mathrm{\&pi;}}_i^p\&Element;\{1,...,n\},i=1,...,n)$ For a n to be processed vehicle air conditioning outlet is based on the arrangement of processing sequence, ${\left[{\mathrm{\&pi;}}_i^p\right]}^k({\mathrm{\&pi;}}_i^p\&Superset;\{{\left[{\mathrm{\&pi;}}_i^p\right]}^k|k=\mathrm{1,2},...,m\},i=1,...,n)$ For assembling motor vehicle air-conditioner air outlet need the vehicle air conditioning outlet part processed on first stage kth platform equipment, for part and part between setup times and $S\left(\right[{{\mathrm{\&pi;}}_0^p]}^k,{\left[{\mathrm{\&pi;}}_1^p\right]}^k)>0,p\left({\left[{\mathrm{\&pi;}}_i^p\right]}^k\right)$ For part process time, for subordinate phase collection, transport belong to vehicle air conditioning outlet all parts to time of phase III mounting equipment, for belonging to vehicle air conditioning outlet all parts through the first stage processing and subordinate phase collect, transport point need the maximum processing time and for vehicle air conditioning outlet at the built-up time of phase III, for vehicle air conditioning outlet deadline and for the average completion time of all vehicle air conditioning outlets; Optimization aim is in the set Π of all vehicle air conditioning outlet processing sequences, find a π ^p*, make objective function minimum.

The concrete steps of the Optimization Scheduling of described ADAPTIVE MIXED Estimation of Distribution Algorithm are as follows:

Step1, coded system: encode with the process and assemble of often kind of vehicle air conditioning outlet order wherein n is the number of vehicle air conditioning outlet to be processed, for vehicle air conditioning outlet to be assembled need the part processed on first stage kth platform equipment;

Step2, population and probability model initialization: population scale is M, adopt random device to produce initialization population, until the quantity of initial solution reaches the requirement of population scale; The matrix P (gen) of n × n dimension is adopted to represent the probability Distribution Model in algorithm gen generation;

Wherein, P _i(gen)=[P _i1(gen), P _i2(gen) ..., P _in(gen)] be the i-th every trade vector in P (gen), Pi _j(gen) be P (gen) the i-th row jth column element and represent gen for time vehicle air conditioning outlet j individual or separate i-th on the probability that occurs gen>=1, P (gen) from the processing precedence relationship numerically reflecting different vehicle air conditioning outlet, Pi _j(gen) larger, represent gen for time the vehicle air conditioning outlet j probability that occurs on i-th of individuality larger;

Step3, sampling produce new population: adopt the mode of roulette to sample to probability model;

Step4, jump out " Insert " mutation operation of principle based on improving first: use and improve " Insert " mutation operation of jumping out principle first to the optimum individual in population or separate and perform Local Search;

Step5, update probabilistic model: first judge that " optimum individual " whether continuous ten generations obtained do not upgrade, and if so, then carry out from new initialization to probability matrix; Otherwise adopt the learning rate based on information entropy and aberration rate self-adaptative adjustment mechanism, " optimum individual " that use algorithm to find in search procedure upgrades probability matrix;

Step6, end condition: the maximum iteration time of setting end condition is 200, if met, then export " optimum individual "; Otherwise go to step Step3, iterate, until meet end condition.

Embodiment 2: as Figure 1-5, a kind of Optimization Scheduling of production assembling process of vehicle air conditioning outlet, by determining production assembling process scheduling model and the optimization aim of vehicle air conditioning outlet, and the Optimization Scheduling of ADAPTIVE MIXED Estimation of Distribution Algorithm is used to be optimized optimization aim; Wherein scheduling model is set up, with minimized average completion date for optimization aim with the final assembling deadline according to the part of vehicle air conditioning outlet the process time on each machine:

$C_T\left({\mathrm{\&pi;}}_i^p\right)=\max \left\{\underset{k=1,...,m}{\max }\right\{{\mathrm{\&Sigma;}}_{j=1}^i(S({\left[{\mathrm{\&pi;}}_{j-1}^p\right]}^k,{\left[{\mathrm{\&pi;}}_j^p\right]}^k)+p\left(\right[{{\mathrm{\&pi;}}_j^p]}^k))\},C_T\left({\mathrm{\&pi;}}_{i-1}^p\right)\}+p_T\left({\mathrm{\&pi;}}_i^p\right),$

$C\left({\mathrm{\&pi;}}_i^p\right)=\max (C_T\left({\mathrm{\&pi;}}_i^p\right),C\left({\mathrm{\&pi;}}_{i-1}^p\right))+p_A\left({\mathrm{\&pi;}}_i^p\right)$

$\stackrel{\&OverBar;}{C}\left({\mathrm{\&pi;}}^P\right)={\mathrm{\&Sigma;}}_{i=1}^{n}C\left({\mathrm{\&pi;}}_i^P\right)/n$

${\mathrm{\&pi;}}^{P*}=\arg \left\{\stackrel{\&OverBar;}{C}\left({\mathrm{\&pi;}}^P\right)\right\}\&RightArrow;\min ,\&ForAll;{\mathrm{\&pi;}}^P\&Element;\&prod;$

Wherein, ${\mathrm{\&pi;}}^P=[{\mathrm{\&pi;}}_1^p,{\mathrm{\&pi;}}_2^p,...,{\mathrm{\&pi;}}_n^p]({\mathrm{\&pi;}}_i^p\&Element;\{1,...,n\},i=1,...,n)$ For a n to be processed vehicle air conditioning outlet is based on the arrangement of processing sequence, ${\left[{\mathrm{\&pi;}}_i^p\right]}^k({\mathrm{\&pi;}}_i^p\&Superset;\{{\left[{\mathrm{\&pi;}}_i^p\right]}^k|k=\mathrm{1,2},...,m\},i=1,...,n)$ For assembling motor vehicle air-conditioner air outlet need the vehicle air conditioning outlet part processed on first stage kth platform equipment, for part and part between setup times and $S\left(\right[{{\mathrm{\&pi;}}_0^p]}^k,{\left[{\mathrm{\&pi;}}_1^p\right]}^k)>0,p\left({\left[{\mathrm{\&pi;}}_i^p\right]}^k\right)$ For part process time, for subordinate phase collection, transport belong to vehicle air conditioning outlet all parts to time of phase III mounting equipment, for belonging to vehicle air conditioning outlet all parts through the first stage processing and subordinate phase collect, transport point need the maximum processing time and for vehicle air conditioning outlet at the built-up time of phase III, for vehicle air conditioning outlet deadline and for the average completion time of all vehicle air conditioning outlets; Optimization aim is in the set Π of all vehicle air conditioning outlet processing sequences, find a π ^p*, make objective function minimum.

The concrete steps of the Optimization Scheduling of described ADAPTIVE MIXED Estimation of Distribution Algorithm are as follows:

Step1, coded system: encode with the process and assemble of often kind of vehicle air conditioning outlet order wherein n is the number of vehicle air conditioning outlet to be processed, for vehicle air conditioning outlet to be assembled need the part processed on first stage kth platform equipment;

Step2, population and probability model initialization: population scale is M, adopt random device to produce initialization population, until the quantity of initial solution reaches the requirement of population scale; The matrix P (gen) of n × n dimension is adopted to represent the probability Distribution Model in algorithm gen generation;

Wherein, P _i(gen)=[P _i1(gen), P _i2(gen) ..., P _in(gen)] be the i-th every trade vector in P (gen), Pi _j(gen) be P (gen) the i-th row jth column element and represent gen for time vehicle air conditioning outlet j individual or separate i-th on the probability that occurs gen>=1, P (gen) from the processing precedence relationship numerically reflecting different vehicle air conditioning outlet, Pi _j(gen) larger, represent gen for time the vehicle air conditioning outlet j probability that occurs on i-th of individuality larger;

Step3, sampling produce new population: adopt the mode of roulette to sample to probability model;

Step4, jump out " Insert " mutation operation of principle based on improving first: use and improve " Insert " mutation operation of jumping out principle first to the optimum individual in population or separate and perform Local Search;

Step5, update probabilistic model: first judge that " optimum individual " whether continuous ten generations obtained do not upgrade, and if so, then carry out from new initialization to probability matrix; Otherwise adopt the learning rate based on information entropy and aberration rate self-adaptative adjustment mechanism, " optimum individual " that use algorithm to find in search procedure upgrades probability matrix;

Step6, end condition: the maximum iteration time of setting end condition is 200, if met, then export " optimum individual "; Otherwise go to step Step3, iterate, until meet end condition.

Population scale is set to 70.

Concrete contrast experiment is as follows:

By the IAEDA designed by the present invention and current existing main flow algorithm---DPSO is (see document Tian Y, Liu D Y, Yuan D H, Wang K H.A discrete PSO for two-stage assembly scheduling problem.InternationalJournal of Advanced Manufacturing Technology, 2013,66:481-499.) contrast, the validity of checking IAEDA.Wherein, IAEDA ran for 200 generations, and DPSO Riming time of algorithm is identical with IAEDA, and test result is as shown in table 1.The target function value that table 1 is tried to achieve under giving different problem scale situation:

The target function value of trying to achieve under the different problem scale of table 1

n×m	30×2	40×2	50×3	60×3
					IAEDA	890.56	1068.31	1380.49	1679.68
DPSO	896.58	1080.58	1389.07	1690.05

From table 1, the present invention is considered index, is all obviously better than DPSO algorithm for considered problem, and this shows that IAEDA is a kind of efficient algorithm of the production Assembly process optimization solving vehicle air conditioning outlet.

By reference to the accompanying drawings the specific embodiment of the present invention is explained in detail above, but the present invention is not limited to above-mentioned embodiment, in the ken that those of ordinary skill in the art possess, various change can also be made under the prerequisite not departing from present inventive concept.

Claims (2)

1. the Optimization Scheduling of the production assembling process of a vehicle air conditioning outlet, it is characterized in that: by determining production assembling process scheduling model and the optimization aim of vehicle air conditioning outlet, and use the Optimization Scheduling of ADAPTIVE MIXED Estimation of Distribution Algorithm to be optimized optimization aim; Wherein scheduling model is set up, with minimized average completion date for optimization aim with the final assembling deadline according to the part of vehicle air conditioning outlet the process time on each machine:

$C_T\left({\mathrm{\&pi;}}_i^p\right)=\max \left\{\underset{k=1,...,m}{\max }\right\{{\mathrm{\&Sigma;}}_{j=1}^i(S({\left[{\mathrm{\&pi;}}_{j-1}^p\right]}^k,{\left[{\mathrm{\&pi;}}_j^p\right]}^k)+p\left({\left[{\mathrm{\&pi;}}_j^p\right]}^k\right))\},C_T\left({\mathrm{\&pi;}}_{i-1}^p\right)\}+p_T\left({\mathrm{\&pi;}}_i^p\right),$

$C\left({\mathrm{\&pi;}}_i^p\right)=\max (C_T\left({\mathrm{\&pi;}}_i^p\right),C\left({\mathrm{\&pi;}}_{i-1}^p\right))+p_A\left({\mathrm{\&pi;}}_i^p\right)$

$\stackrel{\&OverBar;}{C}\left({\mathrm{\&pi;}}^P\right)={\mathrm{\&Sigma;}}_{i=1}^{n}C\left({\mathrm{\&pi;}}_i^P\right)/n$

${\mathrm{\&pi;}}^{P*}=\arg \left\{\stackrel{\&OverBar;}{C}\left({\mathrm{\&pi;}}^P\right)\right\}\&RightArrow;\min ,\&ForAll;{\mathrm{\&pi;}}^P\&Element;\mathrm{\&Pi;}$

Wherein, ${\mathrm{\&pi;}}^P=[{\mathrm{\&pi;}}_1^p,{\mathrm{\&pi;}}_2^p,...,{\mathrm{\&pi;}}_n^p]({\mathrm{\&pi;}}_i^p\&Element;\{1,...,n\},i=1,...,n)$ For a n to be processed vehicle air conditioning outlet is based on the arrangement of processing sequence, for assembling motor vehicle air-conditioner air outlet need the vehicle air conditioning outlet part processed on first stage kth platform equipment, for part and part between setup times and $S({\left[{\mathrm{\&pi;}}_0^p\right]}^k,{\left[{\mathrm{\&pi;}}_1^p\right]}^k)>0,$ for part process time, for subordinate phase collection, transport belong to vehicle air conditioning outlet all parts to time of phase III mounting equipment, for belonging to vehicle air conditioning outlet all parts through the first stage processing and subordinate phase collect, transport point need the maximum processing time and for vehicle air conditioning outlet at the built-up time of phase III, for vehicle air conditioning outlet deadline and for the average completion time of all vehicle air conditioning outlets; Optimization aim is in the set Π of all vehicle air conditioning outlet processing sequences, find a π ^p*, make objective function minimum.

2. the Optimization Scheduling of the production assembling process of vehicle air conditioning outlet according to claim 1, is characterized in that: the concrete steps of the Optimization Scheduling of described ADAPTIVE MIXED Estimation of Distribution Algorithm are as follows:

Step1, coded system: encode with the process and assemble of often kind of vehicle air conditioning outlet order wherein n is the number of vehicle air conditioning outlet to be processed, for vehicle air conditioning outlet to be assembled need the part processed on first stage kth platform equipment;

Step2, population and probability model initialization: population scale is M, adopt random device to produce initialization population, until the quantity of initial solution reaches the requirement of population scale; The matrix P (gen) of n × n dimension is adopted to represent the probability Distribution Model in algorithm gen generation;

Wherein, P _i(gen)=[P _i1(gen), P _i2(gen) ..., P _in(gen)] be the i-th every trade vector in P (gen), P _ij(gen) be P (gen) the i-th row jth column element and represent gen for time vehicle air conditioning outlet j individual or separate i-th on the probability that occurs p (gen) from the processing precedence relationship numerically reflecting different vehicle air conditioning outlet, P _ij(gen) larger, represent gen for time the vehicle air conditioning outlet j probability that occurs on i-th of individuality larger;

Step3, sampling produce new population: adopt the mode of roulette to sample to probability model;

Step4, jump out " Insert " mutation operation of principle based on improving first: use and improve " Insert " mutation operation of jumping out principle first to the optimum individual in population or separate and perform Local Search;

Step5, update probabilistic model: first judge that " optimum individual " whether continuous ten generations obtained do not upgrade, and if so, then carry out from new initialization to probability matrix; Otherwise adopt the learning rate based on information entropy and aberration rate self-adaptative adjustment mechanism, " optimum individual " that use algorithm to find in search procedure upgrades probability matrix;

Step6, end condition: the maximum iteration time of setting end condition is 200, if met, then export " optimum individual "; Otherwise go to step Step3, iterate, until meet end condition.