CN102542411A - Method for carrying out dispatching control on multi-variety multi-process manufacturing enterprise workshop on basis of ACA (Automatic Circuit Analyzer) model - Google Patents

Abstract

The invention relates to a method for carrying out dispatching control on a multi-variety multi-process manufacturing enterprise workshop on the basis of an AC (Automatic Circuit Analyzer) model, which comprises the following steps of: 1, setting a target function of the multi-variety multi-process manufacturing enterprise workshop, and when task information of a task Agent is sent to a decision Agent by a coordination Agent, representing the task information with a directed graph G by the decision Agent and using a dispatching rule which is obtained by the decision Agent through a random algorithm as an initial solution of a tabu search algorithm; 2, by a rule decision element Agent, firstly, calculating a target value F, then searching other dispatching methods from the dispatching field by the tabu search algorithm, and using a solution with the minimum target value as the adopted rule; and 3, carrying out state changing on each cell (i.e. a station) in a cell space according to the dispatching rule awarded by the rule decision element. According to the invention, a complex dispatching system can be described, the practical production is well reflected, the computational efficiency of a model is also considered, and the problem of dispatching the dynamic and complex multi-variety multi-process manufacturing enterprise workshop is rapidly and accurately solved.

Description

Technology how wide in variety manufacturing enterprise workshop dispatch control method based on the ACA model

Technical field

The present invention relates to scheduling controlling field, a kind of workshop, especially a kind of technology how wide in variety manufacturing enterprise workshop dispatch control method.

Background technology

In the prior art; Application number/patent No. for example: 201010226408.2; Disclose a kind of flexible job shop dispatching method, mainly overcome the deficiency that the flexible job shop dispatching method performance based on genetic algorithm is not in full use based on coevolution on multiple populations.This invention can obtain the scheduling scheme of high-quality suitable workshop actual production, shortens the production time, can be used for the management and running and the optimization of Workshop Production process.But it is very sensitive to initial population, and speed of convergence is slow.

Solve job shop scheduling problems is one type of complicacy and has representational engineering problem; It also is typical NP-hard problem; When finding the solution solve job shop scheduling problems, often utilize mathematical programming approach, branch-and-bound method, artificial intelligence approach, neural net method, genetic algorithm, ant group algorithm and simulated annealing etc.

Summary of the invention

In order to overcome the deficiency that applicability is relatively poor, speed of convergence is slow, practicality is relatively poor of existing existing workshop dispatch control method, the technology how wide in variety manufacturing enterprise workshop dispatch control method that the present invention provides that a kind of applicability is good, speed of convergence is very fast, practical based on the ACA model.

The technical solution adopted for the present invention to solve the technical problems is:

A kind of technology how wide in variety manufacturing enterprise workshop dispatch control method based on the ACA model, in the technology said how wide in variety manufacturing enterprise workshop, the operation deadline note of workpiece j on machine i made C _Ij, workpiece j leaves the time note of whole machining process system and makes C _j, the duration of workpiece j is d _j, the delay of workpiece j is defined as:

L _j＝C _j-d _j(1)

Accomplishing when lagging behind when work just is, and when work is fulfiled ahead of schedule, is negative;

The hysteresis of workpiece j is defined as:

T _j＝max(C _j-d _j，0)＝max(L _j，0)(2)

Adopt manufacturing phase C _Max, maximum-delay L _Max, total retardation time ∑ T _jAs regulation goal, add legal being expressed as of general objective function that obtain dispatching through weight:

min(θ ₁C _max+θ ₂L _max+θ ₃∑T _j)(3)

Said control method may further comprise the steps:

1) objective function in technology said how wide in variety manufacturing enterprise workshop is:

minF＝min(θ ₁C _max+θ ₂L _max+θ ₃∑T _j)，j＝1，2，...，n (4)

S.t.y _Kj-y _Ij>=p _IjTo all (i, j) → (k, j) ∈ A

C _Max-y _Ij>=p _IjTo all (i, j) ∈ N

y _Ij-y _Il>=p _Il∨ y _Il-y _Ij>=p _IjTo all (i, l) with (i, j), i=1,2 ..., m

y _Ij>=0 couple of all (i, j) ∈ N

L _j＝C _j-d _j

T _j＝max(L _j，0)

C _max＝max(C ₁，C ₂，L，C _j)

In the formula: y _IjExpression operation (i, zero-time j), N represent that operation is all arranged (i, set j), A for all routes constraints (i, j) → (it needs work j elder generation before machine k processing on machine i, to process p for k, set j) _IjRepresent process time;

When the mission bit stream with task Agent sent to decision agent by coordination Agent, decision agent showed mission bit stream earlier with digraph G; Here the scheduling rule that obtains through any algorithm of decision agent can be as the initial solution of TABU search method.

2) after the Agent of rule decision unit receives the information of the scheduling rule that coordination Agent sends, at first calculate desired value F; In the neighborhood of this scheduling, search out other dispatching methods then;

(2.1) neighbour structure of scheduling

(i is j) with (on the machine i after exchange, (i is k) in task (i, j) execution before for task for i, the adjacency pair exchange between k) for two processing tasks on critical path; For workpiece k, (i k) belongs to this workpiece to task, and promptly workpiece k finishes the work on machine h, and (h k) operates afterwards at once; On machine h, switching task (h, k) and (h, the task of k) on machine h, carrying out before be designated as (h, l);

(2.2) searching method

When the acquisition initial schedule was separated, the taboo table was empty, and the search procedure of the Agent of rule decision unit is following: make π, V (π), N (π) and π _bRepresent processing sequence, mobile set, neighborhood and current known best separating respectively; Suppose mobile set V (π) non-NULL, these are moved be divided into two types: moving of non-taboo and moving of taboo.V (π) T in mobile the taboo, moving among V (π) the ∩ T then avoided.If V (π) is empty, then optimum solution has been found in explanation, otherwise checks moving among the V (π) successively according to corresponding working procedures ascending order on-stream time; Adopt first than former not moving of taboo of separating, and in search procedure, write down current best separating all the time; If each mobile separating of obtaining of not avoiding all is no better than former separating, then select wherein least poor moving; If there be not moving of not taboo, moving that just V (π) is all avoided, and then adopts the mobile v of taboo " at most " in the mobile set _o, and the taboo table modified: in the taboo table with v _oWith than v _oAvoid entry deletion more of a specified duration;

3) in the scheduling body layer, each cellular in the cellular space (being station) will carry out state according to the scheduling rule that rule decision unit authorizes and change.

(1) in the unimpeded not wait of workshop logistics, adopt the rule that is introduced into to go out earlier with First Come First Served, the lattice point state variation is following:

$S_{c\left(d\right)}^{\mathrm{\τ}+1}(q,t,d,e,\mathrm{wn})=\left\{\begin{array}{cc}{S}_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e=1,\mathrm{wn}+1)& \\ & S_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,Q\∈t,d,e,\mathrm{wn})\\ \mathrm{if}& \\ & \mathrm{and}{S}_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e=0,\mathrm{wn})\\ S_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e,\mathrm{wn})& \mathrm{else}\end{array}\right.---\left(5\right)$

In the formula: when particle machines at origin node c (i, j (i)), and the process equipment of destination node c (d) is in idle condition, and for the moment to destination node move down by the step for particle; Simultaneously, the state variation of origin node c (i, j (i)) is following:

$S_{c(i,j\left(i\right))}^{\mathrm{\τ}+1}(q,t,d,e,\mathrm{wn})=\left\{\begin{array}{cc}{S}_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,t=0,d=00,e=0,\mathrm{wn})& \\ & S_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,Q\∈t,d,e,\mathrm{wn})\\ \mathrm{if}& \\ & \mathrm{and}{S}_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e=0,\mathrm{wn})\\ S_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,t,d,e,\mathrm{wn})& \mathrm{else}\end{array}\right.---\left(6\right)$

When particle was in transportation node, operation processing preparation station node, operation processing station node, under the unobstructed situation of logistics, particle got into next adjacent node by flow direction, so have:

d＝(i+1，j(i+1))

(2) SPT and LPL rule:

The shortest or the longest process time, priority rule was given m platform machine with m work allocation short or the longest constantly in τ=0; After this, arbitrary machine free time, what process time was short or the longest in the remaining work will distribute to this idle machine;

P _S＝P _A，P _A＝{k|t _k＝min(max)T}(T∈t)

$S_{c\left(d_x\right)}^{\mathrm{\τ}+1}(q,t,d,e,\mathrm{wn})=\left\{\begin{array}{cc}{S}_{c\left(d_x\right)}^{\mathrm{\τ}}(q,t,d,e=1,\mathrm{wn}+1)& \\ \mathrm{if}& S_{c\left(d_x\right)}^{\mathrm{\τ}}(q,t,d,e=0,\mathrm{wn})\\ S_{c\left(d_x\right)}^{\mathrm{\τ}}(q,t,d,e,\mathrm{wn})& \mathrm{else}\end{array}\right.---\left(7\right)$

In the formula: P _SSelect workpiece P through rule _A, supposing that here the time of each workpiece processing is inequality, T is that workpiece is at c (d _x) last set of processing the required time, c (d _x) be the machine of idle arbitrary same alike result, k for selected at c (d _x) the last particle of processing.

Technical conceive of the present invention is: cellular automaton: cellular automaton (Cellular Automata is called for short CA) has powerful spatial operation ability, is usually used in the research of self-organizing system evolution process.

Cellular automation method is the present unique a kind of booleanization discrete dynamical systems method that can simulate on computers that designs according to the complication system characteristics, and the sign that the modeling method of multiagent system has solved the individual intelligent behavior in the dispatching system of workshop preferably and emerged in large numbers behavior from small-scale character to large scale system.

For this reason, consider the multiagent system method that combination is simulated the workshop dispatching system with cellular automaton.Workshop condition layer, some Agent and immovable cellular automaton layers that independent behaviour is arranged have been comprised in the model.Whole workshop dispatching system person mainly is divided into two on scheduling body layer (cellular automaton) and scheduling controlling layer (multiagent system); Part element in the dispatching system of workshop is expanded; Make them have the characteristic of intelligent body; Can calculate the running status of oneself through oneself decision-making and control, and not adopt the general state transition function of cellular automaton to calculate, the state of station is the expression of more newly arriving through cellular state still.And still adopt the synchronous Calculation Method of cellular automaton for total system, and promptly upgrade the state of all stations in the next cycle, can reach the simulation requirements of the speed of taking into account and precision like this.

(1) scheduling controlling layer

Merging is called task Agent with search Agent will to plan Agent; Main being responsible for worked out MRP according to order, demand forecast, productive capacity and the existing inventories that search; Be transformed into processing tasks working out good plan; And mission bit stream registration is stored in the mission bit stream storehouse, then relevant information is stored in the Relational database, and information sent to coordinate Agent and decision agent.

(2) scheduling body layer

The cellular Automation Model expression formula of setting up after expanding is following:

A′ _S＝{L ₂，S，N，R，F _S，P _S}

In the formula: A ' _SThe Agent-cellular Automation Model of-technology how wide in variety manufacturing enterprise workshop dispatching system;

L ₂-workshop dispatching system is the network of a two dimension;

The S-lattice point is the state set of station, comprises type, state and the process time etc. of equipment;

Station in the N-station neighborhood zone;

The constraint condition that the R-scheduling process need satisfy;

F _S-intend the scheduling rule of adopting to gather;

P _S-rule decision unit.

For the ease of observing; Whole Agent-cellular Automation Model is represented to become the two-layer physical model on the Virtual Space at present; The upper strata is made up of a plurality of main bodys and other non intelligent elements, and lower floor is that make a strategic decision first Agent of a cellular automaton and more rules forms, and is as shown in Figure 1.

Each Agent in the scheduling controlling layer conveys to the initial result of decision in proper order according to the information flow flow direction and coordinates Agent; And search for new scheduling rule again according to desired value through rule decision unit; Here the Agent of rule decision unit will adopt the method for TABU search to obtain last comparatively ideal separating; Change the corresponding lattice point state of scheduling body layer then; Thereby reach the effect of particle movement on the indirect control body layer, and the Agent of rule decision unit to give coordination Agent with information feedback at last.Evolutionary model is as shown in Figure 2.

Therefore whole workshop scheduling model is obtained developing by the final scheduling rule decision-making and the station state evolution of the initial rules decision process drive scheduling body layer (cellular automaton) of scheduling controlling layer (multiagent system).

Beneficial effect of the present invention mainly shows: in solve job shop scheduling problems; The decision maker wants comprehensive task information, inventory information, machinery and equipment situation to wait to select to adopt which kind of scheduling rule; The selection of scheduling rule is dynamic a, process initiatively; Each element in the selection course all has the work of oneself to do, and will work in coordination, coordinate, and accomplishes scheduler task jointly.And multiagent system just ability fine the modeling method that interprets these phenomena; But it is in service in system; Communication usually need take a large amount of system times and resource; And adopting cellular automaton to carry out emulation, computation schema and synchronous Calculation Method through the space-time discretize can improve simulation speed greatly.Fast, solved dynamic, complicated technology how wide in variety manufacturing enterprise solve job shop scheduling problems more accurately.Evaded other and solved dispatching methods in extensive, complex schedule problem, had that computing velocity is slow, structural parameters are difficult to confirm, speed of convergence slow, Search Results can't guarantee shortcomings such as global optimum.

Description of drawings

Fig. 1 is the synoptic diagram of technology how wide in variety manufacturing enterprise workshop dispatching system ACA physical model.

Fig. 2 is the synoptic diagram of technology how wide in variety manufacturing enterprise workshop dispatching system ACA evolutionary model.

Fig. 3 is the synoptic diagram of digraph G.

Fig. 4 is to Jm||C _MaxOne step of problem is recalled the synoptic diagram of adjacent exchange.

Embodiment

Below in conjunction with accompanying drawing the present invention is further described.

With reference to Fig. 1～Fig. 4, a kind of technology how wide in variety manufacturing enterprise workshop dispatch control method based on the ACA model, said dispatch control method need be considered multiple performance index simultaneously, is the multiple goal scheduling problem.The operation deadline note of workpiece j on machine i made C _Ij, workpiece j leaves the time note of whole machining process system and makes C _j, the duration of workpiece j is d _j, the delay of workpiece j is defined as

L _j＝C _j-d _j(1)

Accomplishing when lagging behind when work just is, and when work is fulfiled ahead of schedule, is negative.The hysteresis of workpiece j is defined as

T _j＝max(C _j-d _j，0)＝max(L _j，0)(2)

Lag behind and in fact bear never afterwards with different being that postpones.According to the complicacy and the multiple goal property characteristics of explained hereafter how wide in variety system, adopt manufacturing phase C _Max, maximum-delay L _Max, total retardation time ∑ T _jAs regulation goal, add legal being expressed as of general objective function that obtain dispatching through weight:

min(θ ₁C _max+θ ₂L _max+θ ₃∑T _j)(3)

The present invention adopts the foundation of TABU search method as Agent selection scheduling rule.The TABU search method is to move for jumping out a kind of deterministic algorithm that local optimum designs, when being absorbed in local optimum, do to rise, and it can find the optimum solution of some instances in the scheduling problem.

At first by decision agent according to mission bit stream and inventory information; Select the initial solution of a scheduling rule as scheduling process; Coordinated Agent sends to the Agent of rule decision unit; Carry out the judgement of scheduling rule by the Agent of rule decision unit then and reselect, adopt minimum the separating of desired value at last as the rule of adopting.

The selection initial solution of decision agent: multiple goal job shop scheduling problem can be described as:

minF＝min(θ ₁C _max+θ ₂L _max+θ ₃∑T _j)，j＝1，2，...，n

(4)

S.t. y _Kj-y _Ij>=p _IjTo all (i, j) → (k, j) ∈ A

C _Max-y _Ij>=p _IjTo all (i, j) ∈ N

y _Ij-y _Il>=p _Il∨ y _Il-y _Ij>=p _IjTo all (i, l) with (i, j), i=1,2 ..., m

y _Ij>=0 couple of all (i, j) ∈ N

L _j＝C _j-d _j

T _j＝max(L _j，0)

C _max＝max(C ₁，C ₂，L，C _j)

In the formula: y _IjExpression operation (i, zero-time j), N represent that operation is all arranged (i, set j), A for all routes constraints (i, j) → (it needs work j elder generation before machine k processing on machine i, to process p for k, set j) _IjRepresent process time.

Any to satisfy separating of formula can be a scheduling, and the scheduling rule that obtains through any algorithm of decision agent can be as the initial solution of TABU search method here, and the final goal of scheduling is to locate an as far as possible little F.

When with the mission bit stream of task Agent when coordinating Agent and send to decision agent, decision agent shows mission bit stream earlier with digraph G, the example of 3 workpiece of one 4 machines for example, the digraph G that obtains is as shown in Figure 3:

The TABU search process of the Agent of rule decision unit: after the Agent of rule decision unit receives the information of the scheduling rule that coordination Agent sends, at first calculate desired value F; In the neighborhood of this scheduling, search out other dispatching method then.

(2.1) neighbour structure of scheduling

(i is j) with (i, the adjacency pair exchange between k) obtains the neighbours of existing scheduling with its two processing tasks on critical path.On the machine i after exchange, (i, k) (i j) carries out task before in task.For workpiece k, (i k) belongs to this workpiece to task, and promptly (h k) (carries out) afterwards operating on machine h workpiece k finishing the work at once.On machine h, (h is k) with in that (h, the task of k) on machine h, carrying out before (is designated as (h, l)) to switching task.Fig. 4 can clearly find out, even exchange for the first time is not (in that (i j) draws improvement with (i, k) between), for the second time in that (h is k) with (h, exchange l) also can obtain an overall improvement.

(2.2) searching method

When the acquisition initial schedule was separated, the taboo table was empty, and the search procedure of the Agent of rule decision unit is following: make π, V (π), N (π) and π _bRepresent processing sequence, mobile set, neighborhood and current known best separating respectively.Suppose mobile set V (π) non-NULL, these are moved be divided into two types: moving of non-(not) taboo and moving of taboo.V (π) T in mobile the taboo, moving among V (π) the ∩ T then avoided.If V (π) is empty, then optimum solution has been found in explanation, otherwise checks moving among the V (π) successively according to corresponding working procedures ascending order on-stream time.Adopt first than former not moving of taboo of separating, and in search procedure, write down current best separating all the time.If each mobile separating of obtaining of not avoiding all is no better than former separating, then select wherein least poor moving.If there be not moving of not taboo, moving that just V (π) is all avoided, and then adopts the mobile v of taboo " at most " in the mobile set _o, and the taboo table modified: in the taboo table with v _oWith than v _oAvoid entry deletion more of a specified duration.

ACA model station state variation: in the scheduling body layer, each cellular in the cellular space (being station) will have more scheduling rule that rule decision unit authorizes to carry out state and changes.

FIFO and FCFS rule: logistics is unimpeded in the workshop, does not have the time of wait, the general rule that is introduced into to go out earlier with First Come First Served that adopts.The lattice point state variation is following:

$S_{c\left(d\right)}^{\mathrm{\τ}+1}(q,t,d,e,\mathrm{wn})=\left\{\begin{array}{cc}{S}_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e=1,\mathrm{wn}+1)& \\ & S_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,Q\∈t,d,e,\mathrm{wn})\\ \mathrm{if}& \\ & \mathrm{and}{S}_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e=0,\mathrm{wn})\\ S_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e,\mathrm{wn})& \mathrm{else}\end{array}\right.---\left(5\right)$

In the formula: when particle machines at origin node c (i, j (i)), and the process equipment of destination node c (d) is in idle condition, and for the moment to destination node move down by the step for particle.Simultaneously, the state variation of origin node c (i, j (i)) is following:

$S_{c(i,j\left(i\right))}^{\mathrm{\τ}+1}(q,t,d,e,\mathrm{wn})=\left\{\begin{array}{cc}{S}_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,t=0,d=00,e=0,\mathrm{wn})& \\ & S_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,Q\∈t,d,e,\mathrm{wn})\\ \mathrm{if}& \\ & \mathrm{and}{S}_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e=0,\mathrm{wn})\\ S_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,t,d,e,\mathrm{wn})& \mathrm{else}\end{array}\right.---\left(6\right)$

When particle was in transportation node, operation processing preparation station node, operation processing station node, from the technology angle, under the unobstructed situation of logistics, particle should get into next adjacent node by flow direction, so have:

d＝(i+1，j(i+1))

SPT and LPL rule: lack (length) priority rule process time most and give m platform machine with m work allocation of lacking (length) most constantly in τ=0.After this, arbitrary machine free time, what lacked (length) process time most in the remaining work will distribute to this idle machine.

P _S＝P _A，P _A＝{k|t _k＝min(max)T}(T∈t)

$S_{c\left(d_x\right)}^{\mathrm{\τ}+1}(q,t,d,e,\mathrm{wn})=\left\{\begin{array}{cc}{S}_{c\left(d_x\right)}^{\mathrm{\τ}}(q,t,d,e=1,\mathrm{wn}+1)& \\ \mathrm{if}& S_{c\left(d_x\right)}^{\mathrm{\τ}}(q,t,d,e=0,\mathrm{wn})\\ S_{c\left(d_x\right)}^{\mathrm{\τ}}(q,t,d,e,\mathrm{wn})& \mathrm{else}\end{array}\right.---\left(7\right)$

In the formula: P _SSelect workpiece P through rule _A, supposing that here the time of each workpiece processing is inequality, T is that workpiece is at c (d _x) last set of processing the required time, c (d _x) be the machine of idle arbitrary same alike result, k for selected at c (d _x) the last particle of processing.

Instance: certain company is the enterprise that all kinds of tubular electrothermal elements, high-power and tens big serial 400 multiple products such as high-performance combined type heater element, metallic matrix P.e.c. heater element are produced in family specialty research and development, is widely used in all kinds of household electrical appliance.A wherein existing product that monomer just can process, the product that also has the processing of four, five unit constructions could accomplish, employing be typical explained hereafter pattern how wide in variety.Along with the development of the said firm and the variation of customer requirement; Research for its solve job shop scheduling problems becomes more and more important and necessary; But because company size is medium; Therefore not to pay attention to very much to the scheduling link, and lack a regular dispatching system and mechanism therefore the Agent-cellular Automation Model of the technology how wide in variety manufacturing enterprise workshop dispatching system that proposes is had the excellent research meaning to the problem of this enterprise.

Adopt manufacturing phase C _Max, maximum-delay L _Max, total retardation time ∑ T _jThe weighted sum of three desired values is as the catalogue scale value, and as the foundation of model testing, and unit is all with minute representing.The result of emulation and genetic algorithm are calculated the result that obtains for 80 times respectively relatively to this example, genetic algorithms use be the optimization method of iterative computation, will adopt the emulation of three single goals and a general objective in the Swarm emulation respectively.θ _iRepresent each objective weight, confirm the relative importance grade between the target earlier,, adopt the random number method of substitution and the contrast method of average to obtain at last: θ according to the target comparator matrix that obtains ₁=0.7, θ ₂=0.2, θ ₃=0.1.Therefore the expression formula that obtains general objective does

minF＝min(0.7C _max+0.2L _max+0.1∑T _j)

Study to the workshop in the factory, this workshop produces ten kinds of structure Different products altogether at present, and their model is respectively 3015,3020 ..., 3024, their shared ten different equipment are designated as M respectively ₀, M ₁..., M ₉, belong to typical technology how wide in variety workshop.Each equipment and operation title are as shown in table 1, and each workpiece model and process route thereof are as shown in table 2, and the processing required time and the duration of each procedure of each workpiece are as shown in table 3.

Table 1

Table 2

Table 3

At first, combine inventories to formulate processing tasks then, the work that just task Agent will do in technology how wide in variety manufacturing enterprise workshop dispatching system ACA model according to the quantitative requirement of order numbers to various product.The order quantity required and the stockpile number of each workpiece are as shown in table 4.

Table 4

Setting up on the ACA model based, carry out the Swarm emulation experiment.Through the cooperating with each other and communicating by letter of each Agent in the dispatching system of workshop, obtain separating by the Agent of rule decision unit at last near the final scheduling of target.The maximum that obtains is at last made phase C _Max=104 minutes, maximum-delay L _Max=28 minutes, total retardation time ∑ T _j=206 minutes.Adopt the weighting multiple goal with, be about to C _Max, L _MaxWith ∑ T _jThree targets are 99 minutes through the weighting desired value that formula gets to the end, are superior to the monocular scale value 101 minutes of genetic algorithm.Therefore the Agent-cellular Automation Model of setting up technology how wide in variety manufacturing enterprise workshop dispatching system is come that multi-objective problem is dispatched in this workshop and is optimized and finds the solution separate (minimum value) that can more be approached target; Explanation is with respect to genetic algorithm, and this model has certain feasibility.

Claims (1)

1. technology how wide in variety manufacturing enterprise workshop dispatch control method based on the ACA model is characterized in that: in the technology said how wide in variety manufacturing enterprise workshop, the operation deadline note of workpiece j on machine i made C _Ij, workpiece j leaves the time note of whole machining process system and makes C _j, the duration of workpiece j is d _j, the delay of workpiece j is defined as:

L _j＝C _j-d _j(1)

Accomplishing when lagging behind when work just is, and when work is fulfiled ahead of schedule, is negative;

The hysteresis of workpiece j is defined as:

T _j＝max(C _j-d _j，0)＝max(L _j，0)(2)

Adopt manufacturing phase C _Max, maximum-delay L _Max, total retardation time ∑ T _jAs regulation goal, add legal being expressed as of general objective function that obtain dispatching through weight:

min(θ ₁C _max+θ ₂L _max+θ ₃∑T _j)(3)

Said control method may further comprise the steps:

1) objective function in technology said how wide in variety manufacturing enterprise workshop is:

minF＝min(θ ₁C _max+θ ₂L _max+θ ₃∑T _j)，j＝1，2，...，n (4)

S.t. y _Kj-y _Ij>=p _IjTo all (i, j) → (k, j) ∈ A

C _Max-y _Ij>=p _IjTo all (i, j) ∈ N

y _Ij-y _Il>=p _Il∨ y _Il-y _Ij>=p _IjTo all (i, l) with (i, j), i=1,2 ..., m

y _Ij>=0 couple of all (i, j) ∈ N

L _j＝C _j-d _j

T _j＝max(L _j，0)

C _max＝max(C ₁，C ₂，L，C _j)

In the formula: y _IjExpression operation (i, zero-time j), N represent that operation is all arranged (i, set j), A for all routes constraints (i, j) → (it needs work j elder generation before machine k processing on machine i, to process p for k, set j) _IjRepresent process time;

When the mission bit stream with task Agent sent to decision agent by coordination Agent, decision agent showed mission bit stream earlier with digraph G;

2) after the Agent of rule decision unit receives the information of the scheduling rule that coordination Agent sends, at first calculate desired value F; In the neighborhood of this scheduling, search out other dispatching methods then;

(2.1) neighbour structure of scheduling

(i is j) with (on the machine i after exchange, (i is k) in task (i, j) execution before for task for i, the adjacency pair exchange between k) for two processing tasks on critical path; For workpiece k, (i k) belongs to this workpiece to task, and promptly workpiece k finishes the work on machine h, and (h k) operates afterwards at once; On machine h, switching task (h, k) and (h, the task of k) on machine h, carrying out before be designated as (h, l);

(2.2) searching method

When the acquisition initial schedule was separated, the taboo table was empty, and the search procedure of the Agent of rule decision unit is following: make π, V (π), N (π) and π _bRepresent processing sequence, mobile set, neighborhood and current known best separating respectively; Suppose mobile set V (π) non-NULL, these are moved be divided into two types: moving of non-taboo and moving of taboo.V (π) T in mobile the taboo, moving among V (π) the ∩ T then avoided.If V (π) is empty, then optimum solution has been found in explanation, otherwise checks moving among the V (π) successively according to corresponding working procedures ascending order on-stream time; Adopt first than former not moving of taboo of separating, and in search procedure, write down current best separating all the time; If each mobile separating of obtaining of not avoiding all is no better than former separating, then select wherein least poor moving; If there be not moving of not taboo, moving that just V (π) is all avoided, and then adopts the mobile v of taboo " at most " in the mobile set _o, and the taboo table modified: in the taboo table with v _oWith than v _oAvoid entry deletion more of a specified duration;

3) in the unimpeded not wait of workshop logistics, adopt the rule that is introduced into to go out earlier with First Come First Served, the lattice point state variation is following:

$S_{c\left(d\right)}^{\mathrm{\τ}+1}(q,t,d,e,\mathrm{wn})=\left\{\begin{array}{cc}{S}_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e=1,\mathrm{wn}+1)& \\ & S_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,Q\∈t,d,e,\mathrm{wn})\\ \mathrm{if}& \\ & \mathrm{and}{S}_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e=0,\mathrm{wn})\\ S_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e,\mathrm{wn})& \mathrm{else}\end{array}\right.---\left(5\right)$

In the formula: when particle machines at origin node c (i, j (i)), and the process equipment of destination node c (d) is in idle condition, and for the moment to destination node move down by the step for particle; Simultaneously, the state variation of origin node c (i, j (i)) is following:

$S_{c(i,j\left(i\right))}(q,t,d,e,\mathrm{wn})=\left\{\begin{array}{cc}{S}_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,t=0,d=00,e=0,\mathrm{wn})& \\ & S_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,Q\∈t,d,e,\mathrm{wn})\\ \mathrm{if}& \\ & \mathrm{and}{S}_{c\left(d\right)}^{\mathrm{\τ}}(q,t,d,e=0,\mathrm{wn})\\ S_{c(i,j\left(i\right))}^{\mathrm{\τ}}(q,t,d,e,\mathrm{wn})& \mathrm{else}\end{array}\right.---\left(6\right)$

When particle was in transportation node, operation processing preparation station node, operation processing station node, under the unobstructed situation of logistics, particle got into next adjacent node by flow direction, so have:

d＝(i+1，j(i+1))

(2) SPT and LPL rule:

The shortest or the longest process time, priority rule was given m platform machine with m work allocation short or the longest constantly in τ=0; After this, arbitrary machine free time, what process time was short or the longest in the remaining work will distribute to this idle machine;

P _S＝P _A，P _A＝{k|t _k＝min(max)T}(T∈t)

$S_{c\left(d_x\right)}^{\mathrm{\τ}+1}(q,t,d,e,\mathrm{wn})=\left\{\begin{array}{cc}{S}_{c\left(d_x\right)}^{\mathrm{\τ}}(q,t,d,e=1,\mathrm{wn}+1)& \\ \mathrm{if}& S_{c\left(d_x\right)}^{\mathrm{\τ}}(q,t,d,e=0,\mathrm{wn})\\ S_{c\left(d_x\right)}^{\mathrm{\τ}}(q,t,d,e,\mathrm{wn})& \mathrm{else}\end{array}\right.---\left(7\right)$

In the formula: P _SSelect workpiece P through rule _A, supposing that here the time of each workpiece processing is inequality, T is that workpiece is at c (d _x) last set of processing the required time, c (d _x) be the machine of idle arbitrary same alike result, k for selected at c (d _x) the last particle of processing.

Images