CN101359382A - Dynamic partner selecting method based on ant colony algorithm - Google Patents

Abstract

The invention relates to a dynamic alliance partner selection method based on the ant colony algorithm and proposes a <microhabitat ant colony optimization> (MACO). While the MACO utilizes the positive feedback, the MACO imports the time-varying parameter to utilize the experience information and the heuristic information and integrates the microhabitat information difference thoughts during the local optimization, so as to effectively prevent the <premature> problem in the genetic algorithm and the <stagnation> state occurs in the ant algorithm. The cooperative partner selection during the dynamic alliance construction process of the manufacturing enterprises is abstracted into the multi-objective optimization problem; the optimized selection target function is built. The MACO is adopted to calculate the multi-objective problem to obtain the optimum solution; thereby, the optimum combination of the dynamic alliance partner is searched efficiently and accurately.

Description

A kind of dynamic partner selecting method based on ant group algorithm

Technical field

The present invention relates to a kind of data computation method, particularly a kind of dynamic partner selecting method.

Background technology

Ant has existed and has reached 100,000,000 years, is one of the most successful population of nature.Cooperation behavior between the simple ant individuality shows as the social organization of whole colony highly structural, can finish the complex task of ant individual capability head and shoulders above under many circumstances.The cooperation set of this individual behavior from organic sphere, the surprising lifting of colony's ability of the highly structural that is embodied is just agreed without prior consultation mutually with the intension of the dynamic alliance of industry member.This also impels the present invention to adopt the reason of this algorithm just.

(Ant Colony Optimization, ACO) thought is a kind of global optimization approach from the behavior of ant searching food to ant group algorithm.The ant group algorithm optimizing process has four kinds of mechanism: 1. path selection mechanism.The path that quantity of information is big more, selecteed probability is big more.2. pheromones update mechanism.Quantity of information concentration on the path can increase with the process of ant, simultaneously also As time goes on decay gradually.3. exchange the coordination system.Be by information mutual communication usually, collaborative work between the ant individuality.4. optimal solution search mechanism.What the independent search loop of ant group obtained separates, just local optimum.Suitable local search algorithm can enlarge the search volume, helps finding globally optimal solution.Under these four kinds of mechanism, ant group algorithm organically combines information positive feedback principle and heuritic approach, constantly exchanges between the ant group individuality, transmission information, and cooperation helps finding better solutions mutually.

At occurring in nature, the ant group is at certain ecologic environment among a small circle---it is movable to carry out in the microhabitat (Microhabitat).There is various factors to influence the optimizing activity of ant in the microhabitat.The enterprise that manufacturing enterprise's dynamic alliance in fact also can be regarded as in particular range connects each other, interactional colony.Various factors in this scope affects the selection of enterprise to the ally, such as production time, cost, quality, candidate buddy credit worthiness, productive capacity, and factor such as logistics cost, logistics time.And will consider respectively according to corresponding importance of each factor and characteristics, so that candidate buddy is optimized combination.So selecting the affiliate is the problem of a multiple-objection optimization.The quick establishment of enterprise dynamic alliance quite has similarity with the optimizing activity of ant under little ecologic environment.

Yet basic ant group algorithm also has problem to be solved.1. the information positive feedback progressively is confined to the search volume may have to locally optimal solution in the very little scope when strengthening the separating of better performances.2. owing to posterior infromation trend between node is stable, cause transition probability to remain unchanged substantially, very easily be absorbed in stagnation.

Summary of the invention

The present invention be directed to existing basic ant group algorithm and exist the problem that very easily is absorbed in stagnation, proposed a kind of dynamic partner selecting method, propose " microhabitat ant group algorithm " (MACO) based on ant group algorithm.MACO is when utilizing positive feedback, introduce time-varying parameter and utilize posterior infromation and heuristic information, and when local optimal searching, combine the thought of microhabitat information gap, thereby prevent " stagnation " state of taking place in " precocity " problem that occurs in the genetic algorithm and the ant algorithm effectively.

Technical scheme of the present invention is: a kind of dynamic partner selecting method based on ant group algorithm comprises the steps:

1) according to the concrete condition of manufacturing enterprise's dynamic alliance, set up the corresponding collaborative chain-MCC that makes, then according to the characteristics of demand of each link, determine the index binding occurrence of assessment indicator system and each link of each link to resource;

2) determine link v according to physical condition _iThe composite behaviour matrix of candidate role unit, calculate v _iBut the 2-recombination that can produce of candidate role unit, and link v _iCandidate role unit to v _I+1Candidate role unit between the channel performance matrix, set up on this basis candidate buddy digraph CEC=(CEV, CEA), CEV=(1,2 ..., n} is candidate's node set, and CEA={ (i, j) } be the set on limit;

3) utilize microhabitat ant group algorithm (MACO) to select the combination of optimal candidate role unit.

The performing step of described microhabitat ant group algorithm (MACO):

1) initialization: set maximum cycle Cyclemax, ant is counted m.During beginning, all m ant all concentrates on starting point S.Give the quantity of information τ of equal amount on every limit ₀=C, the C constant;

2) routing: ant k is from the S point, according to selection strategy, from the set of the node that is associated with S, selects a node a, and the follow-up limit of a; Then, another node b that connects from this edge again from the set on the limit that is associated with b, selects another limit.If ant is gone to certain node, this node does not have successor node, and the record of this ant through the path deleted in then ant death, and this node of mark is no longer selected this node next time for opening circuit;

By that analogy, up to searching terminal point E.So ant k obtains separating from S to E.After ant k had searched for, remaining ant searched out other path from S to E according to the method identical with k;

3) Local Search: after m ant searched for, try to achieve m and separate, comprise repetition, exchange of information between the ant individuality " explores " the path by the microhabitat information gap again.The microhabitat information gap is by this m being separated the employing genetic manipulation, and heredity intersects, and hereditary variation realizes, utilizes genetic algorithm, carries out Local Search, obtains locally optimal solution;

4) lastest imformation element: when search cycle each time finishes, when promptly all ants arrive destinations, upgrade the pheromones in whole paths;

5) innings optimum solution of demanding perfection: till the current iteration number of times, in all locally optimal solutions of being set up, minimum value separate globally optimal solution as the current iteration number of times, the i.e. distance value of shortest path.

Described 2-can recombinate and be combined into: the situation that the link role is joined together to bear by two role units at most, when two role units unite, if both precedence does not influence the logistics index between them, and the combination of same role unit is equal to single role unit.

Beneficial effect of the present invention is: a kind of dynamic partner selecting method of the present invention based on ant group algorithm, and manufacturing enterprise's dynamic alliance built the selection of affiliate in alliance's process is abstract to be the problem of multiple-objection optimization, and set up the optimized choice objective function.Utilize MACO to resolve this multi-objective problem, can obtain optimum solution.Thereby seek the best of breed of dynamic partner efficiently and accurately.

Description of drawings

Fig. 1 the present invention is a kind of based on search three stage synoptic diagram in the dynamic partner selecting method of ant group algorithm.

Embodiment

On the basis of the research work of analyzing and summing up ant group algorithm, the present invention improves basic ant group algorithm from four kinds of optimization mechanism of ant group algorithm, and proposition microhabitat ant group algorithm (Microhabitat ACO, MACO), in order to select optimum alliance partner.

The negative effect of information positive feedback in the basic ant group algorithm when continually strengthening that better performances is more excellent and separating, should make that the search volume of algorithm is big as much as possible, to seek the interval of separating that may there be optimum solution in those; Also to make full use of effective information current in the colony, make the emphasis of algorithm search be placed in the interval at those individual places that may have higher adaptive value, thereby be retracted to globally optimal solution with " big probability ", " explorations " (exploration), " utilization " (exploitation), set up an equilibrium point between " exploration again "." utilization " is meant that selecting the highest path of quantity of information, " explorations " to be meant by quantity of information height, principle that probability is high goes discovery and select more shortest path.

The search procedure of optimum solution can be divided into initial stage, mid-term and later stage three phases, and the information feature difference that each stage had is searched for three stage synoptic diagram as shown in Figure 1.Ant group algorithm changes along with search evolution process the utilization of posterior infromation and heuristic information, and the value of each parameter is also answered adaptively modifying, to set up the equilibrium point of appropriate " exploration ", " utilization ", " exploring again ".

One, to analysis of the present invention

As social biology, not only exchange between the ant individuality by pheromones, and between any two also with the direct exchange of information of antenna.Because there are various natural causes (wind, temperature, humidity etc.) in the microhabitat of ant activity, and there are differences between the ant individuality, in exchange of information, variation has taken place in information that message that transmit leg sends and take over party receive probably.

In microhabitat, the influence owing to extraneous factor makes the change that information takes place between information source and information receiving terminal be called microhabitat information gap (Information Differece) with this in the present invention.The microhabitat information gap has increased the diversity of information, can think once to enlarge the chance of search volume.Use for reference genetic algorithm genetic manipulation (intersecting and variation), handle the information that is exchanged between the ant individuality, produce information gap, can enlarge the search volume of separating, and select optimum solution wherein.

1, the realization of microhabitat ant group algorithm

For convenience of description, be without loss of generality, increase by two skies (empty) node in ant is looked for food the path digraph, i.e. starting point S and end point E are in order to the beginning and the end of sign node search.It is as follows that MACO finds the solution the main performing step of shortest route problem of digraph:

(1) initialization: set maximum cycle Cyclemax, ant is counted m.During beginning, all m ant all concentrates on starting point S.Give the quantity of information τ of equal amount on every limit ₀=C, the C constant.

(2) routing: ant k is from the S point, according to selection strategy, from the set of the node that is associated with S, selects a node a, and the follow-up limit of a; Then, another node b that connects from this edge again from the set on the limit that is associated with b, selects another limit.If ant is gone to certain node, this node does not have successor node, and the record of this ant through the path deleted in then ant death, and this node of mark is no longer selected this node next time for opening circuit.

By that analogy, up to searching terminal point E.So ant k obtains separating from S to E.After ant k had searched for, remaining ant searched out other path from S to E according to the method identical with k.

When Local Search, ant had both been considered the length of stretch down, also will consider the distribution intensity of its pheromones, selected next to arrive node with certain probability in neighborhood of nodes.Be positioned at the ant k of node i, select next node j by following selection strategy:

$p_{\mathrm{ij}}=\left\{\begin{array}{cc}\frac{{\mathrm{\&tau;}}_{\mathrm{ij}}^{\mathrm{\&alpha;}\left(t\right)}\&CenterDot;{\mathrm{\&eta;}}_{\mathrm{ij}}^{\mathrm{\&beta;}\left(t\right)}}{\underset{s\&Element;{\mathrm{allowed}}_k}{\mathrm{\&Sigma;}}{\mathrm{\&tau;}}_{\mathrm{ij}}^{\mathrm{\&alpha;}\left(t\right)}{\mathrm{\&eta;}}_{\mathrm{ij}}^{\mathrm{\&beta;}\left(t\right)}}& j\&Element;{\mathrm{allowed}}_k\\ 0& \mathrm{otherwise}\end{array}\right.---\left(1\right)$

Wherein, allowedk (k=1,2 ..., m) the current node that allows ant k to pass through of expression, τ _IjExpression is spread out over highway section between node i and the node j, and (a kind of trend information of representing solution space to search for is the power that ant is moved to node j from node i for i, the j) pheromones on; η _IjBe that (i, j) the search cost on is the heuristic function that moves to node j from node i, η in the highway section of evaluation ant individuality between node i node j _Ij=1/d _Ijα (t) is used to describe the importance of posterior infromation, and β (t) is used to describe the importance of heuristic function, is arithmetic number.

Basic ant group algorithm α and β in the whole process of search all remain unchanged.Its consequence is the local feature that algorithm more and more depends on problem, and algorithm later stage appearing suddenly of separating of new more outstanding is more and more difficult.Local message is constantly accumulated, and finally makes algorithm convergence separate with the problem local feature is closely-related in one.And this demand to posterior infromation and heuristic information in evolutionary process of algorithm changes, and the algorithm of invariant parameter is difficult to satisfy these needs.

At the initial stage of MACO algorithm operation, for guarantee to separate relatively than dominance, need more utilize heuristic information, this is the needs that dwindle the search volume; And,, need more " utilization " posterior infromation in order to guarantee more excellent appearing suddenly of separating in the algorithm later stage, and the influence of desalination local message.Therefore, be provided with,

α(t)＝α ₀+k _α×t (2)

β(t)＝β ₀-k _β×t (3)

α wherein ₀And β ₀Be the parameter values of initial setting up, span is same as basic ant group algorithm, 0≤α ₀≤ 5,1≤β ₀≤ 5.k _α, k _βBe two positive constant (0＜k _α, k _β＜1), for dispersed problem, t equals the cyclic algebra Cycle in the algorithm, and requires k _β＜β ₀/ Cyclemax, Cyclemax are maximum cycle.k _αGenerally can get k _α=k _β

(3) Local Search.After m ant searched for, try to achieve m and separate (comprising repetition).Exchange of information between the ant individuality " is explored " the path by the microhabitat information gap again.The microhabitat information gap is to adopt genetic manipulation (heredity intersects, hereditary variation) to realize by this m is separated.Utilize genetic algorithm, carry out Local Search, obtain locally optimal solution.

(4) lastest imformation element.The mode of ant lastest imformation element has two kinds: the local updating and the overall situation are upgraded.The local updating rule is that ant is whenever shifted to next node, will stay pheromones on this path; It then is when the search cycle finishes each time (when all ants arrive destinations) that the overall situation is upgraded, and upgrades the pheromones in whole paths.The overall situation is upgraded the result considered every each pathfinding of ant, has implied information feedback, makes that algorithm is easier to become excellent.The present invention adopts the overall situation to upgrade.The local optimum result that previous step is obtained reacts on problem space, promptly carries out track concentration and upgrade on digraph.

If the pheromones volatility is (1-ρ), highway section (i, j) the pheromones intensity τ on during each loop ends _Ij(t+n) be:

τ _ij(t+n)＝ρ·τ _ij(t)+Δτ _ij (4)

${\mathrm{\&Delta;\&tau;}}_{\mathrm{ij}}=\underset{k\&Element;{\mathrm{Visited}}_{\mathrm{ij}}}{\mathrm{\&Sigma;}}{\mathrm{\&Delta;\&tau;}}_{\mathrm{ij}}^k---\left(5\right)$

Visited _IjFor in this time circulation through the highway section (i, ant set j),

${\mathrm{\&tau;}}_{\mathrm{ij}}^k=Q\left(t\right)/L_k---\left(6\right)$

Wherein, L _kBe the local optimum target function value of k ant in this circulation.

Because the utilization of posterior infromation and heuristic information changes along with search evolution process, so adopt time-varying function Q (t) to replace basic ant group algorithm adjustment information element ${\mathrm{\&Delta;\&tau;}}_{\mathrm{ij}}^k=Q/L_k$ In be the pheromones intensity Q of constant term.Q (t) is along with search becomes big gradually, and it is big that renewal amount also becomes gradually.

Q(t)＝Q ₀+k _Q×t (7)

Wherein, k _QValue according to experience.Generally be taken as and k _α, k _βThe same order of magnitude makes that the variation of Q (t) is not too big.

Calculate the global information increment according to locally optimal solution.After the overall situation is upgraded, continue iteration up to satisfying stop condition.Stop condition is that maximum iteration time or separate stays cool.

(5) innings optimum solution of demanding perfection.Till the current iteration number of times, in all locally optimal solutions of being set up, minimum value separate globally optimal solution as the current iteration number of times, the i.e. distance value of shortest path.

The MACO algorithm is at the search initial stage, to utilize heuristic information; Along with the evolution of search, posterior infromation increases, and then to utilize posterior infromation, quickens the convergence of separating; And arrived the later stage of search, and avoid the phenomenon of algorithm precocity, stagnation, adopt genetic algorithm to realize the microhabitat information gap, enlarge the search volume of understanding, guarantee more excellent appearing suddenly of separating.On the optimizing performance in algorithm later stage, be better than single genetic algorithm or ant group algorithm, can effectively prevent " stagnation " state that takes place in " precocity " problem that occurs in the genetic algorithm and the ant group algorithm.

2, the ant group system model of partner's optimized choice is set up

Manufacturing enterprise's dynamic partner selects optimization problem can be described as: from the set CE of candidate enterprise, be every link role v _iSelect suitable affiliate, and consider logistics time and cost, (Manufacturing Cooperative Chain MCC) finishes with optimum performance (time, cost, quality etc.) to make whole manufacturing work in coordination with chain.This is a multi-objective optimization question.

To a MCC of certain product, and C=(V, A), cardV=n wherein.The set CE of candidate role unit is arranged, cardCE=m.Each link role v _i(there is m in the candidate role unit of 0≤i≤n) _i(m _i≤ m) individual.CE _i={ ce _Ij/ i=1,2 ..., n, j=1,2 ..., m _iBe to be competent at link role v _iCandidate role's unit set, ${\mathrm{CE}}_i\&SubsetEqual;\mathrm{CE}.$

Optimal combined situation is v _iBear by an independent role unit, and v _iArbitrary candidate role unit can select v _I+1Arbitrary candidate role unit unite.At this moment, as long as, just can obtain optimum alliance partner's combination in each node selection link evaluation index of MCC and the role unit of logistics index best performance.

But in actual conditions, because cost, delivery date etc. factor, a role unit makes when can not finish the link task separately, perhaps a plurality of role unit unites and bears when bearing certain link and can obtain better benefit than an independent role unit, the situation that a plurality of role unit bears a link just may occur.Simultaneously, owing to reasons such as economy, law, technology, bear link v _iA certain candidate role unit often can not choose at random successor node v _I+1Candidate role unit unite.And be based on some rule or some condition, selectively and v _I+1Some candidate role units unite.

If link v _i, v _I+1M is arranged respectively _i, m _I+1Individual candidate role unit.For simplified model, the present invention investigates the situation that a link role is joined together to bear by two role units at most.When the associating of two roles unit, the precedence of establishing both does not influence the logistics index between them, and the combination of same role unit is equal to single role unit, but this situation belongs to the 2-recombination.Be located under the particular constraints v _iCandidate role unit can produce cm _iBut (0≤i≤n) kind 2-recombination $(m_i\&le;{\mathrm{cm}}_i\&le;F_2^{m_i}).$

For not having when constraint, m _iBut the individual candidate role 2-of unit recombination by combinatorial theory knowledge, has,

$F_r^{m_i}=\frac{(m_i+r-1)r}{r!},$

$\frac{{\left(m_i\right)}_r}{r!}=\frac{m_i(m_i-1)\&CenterDot;\&CenterDot;\&CenterDot;(m_i-r+1)}{r!}---\left(8\right)$

Link v _iThe property capable of being combined of candidate role unit by the composite behaviour matrix

(1≤i≤n) expression has ${\mathrm{cb}}_{\mathrm{kl}}\&Element;{\mathrm{CB}}_{m_i\&times;m_i}^i,$ 1≤k，l≤m _i，

${\mathrm{cb}}_{\mathrm{kl}}=\left\{\begin{array}{cc}1& \mathrm{k\; joint\; l}\\ 0& \mathrm{otherwise}\end{array}\right.---\left(9\right)$

According to the composite behaviour matrix

Can obtain v _iThe combination of candidate role unit.If can not have the constraint combination between any two candidate role units of same link, then ${\mathrm{cm}}_i=F_2^{m_i}.$ The present invention investigates situation that can not have the constraint combination between any two candidate role units of same link.

If under the particular constraints, v _i, v _I+1Candidate role unit can produce cm respectively _i, cm _I+1But kind of 2-recombination, v _iCandidate role unit to v _I+1Candidate role unit between the passage situation by the channel performance matrix (1≤i≤n) expression has ${\mathrm{pa}}_{\mathrm{rs}}\&Element;{\mathrm{PA}}_{{\mathrm{cm}}_i\&times;{\mathrm{cm}}_{i+1}}^i,$ 1≤r≤cm _i，1≤s≤cm _i+1。

${\mathrm{pa}}_{\mathrm{rs}}=\left\{\begin{array}{cc}1& \mathrm{r\; can\; joint\; s}\\ 0& \mathrm{otherwise}\end{array}\right.---\left(10\right)$

If

Middle nonzero element number is nzp, from v _iTo v _I+1Passage the nzp bar is just arranged.This possible number of combinations of MCC candidate role unit is: The maximum number of combinations of candidate role unit is ${\mathrm{nzp}}_{\max }=F_2^{m_i}\&times;F_2^{m_{i+1}}.$

Respectively the set of candidate role unit is mapped as node set CEV, each node weights that the functional value of the link evaluation index objective function of candidate role unit is mapped as, the functional value of logistics target goals function is mapped as the weights on limit.According to v _iOrder, set up candidate buddy digraph CEC=(CEV, CEA).CEV={1,2 ..., n} is candidate's node set, and CEA={ (i, j) } be the set on limit.D={d _iBe the node weight matrix, DL={dl _IjIt is the limit weight matrix.

The separating of " MCC partner selection " multi-objective optimization question is one and meets v _iThe node sequence of the weights minimum of order.Separating under the multiple goal meaning is a kind of " compromise solution ", " noninferior solution ".Therefore, " MCC partner selection " multi-objective optimization question mathematical problem can be described as:

To certain bar MCC, and C=(V, A), cardV=n wherein.The set CE of candidate role unit is arranged, cardCE=m.Each link role v _i(there is m in the candidate role unit of 0≤i≤n) _i(m _i≤ m) individual, establish link v _i, v _I+1M is arranged respectively _i, m _I+1Individual candidate role unit, v _i, v _I+1Candidate role unit can produce cm _iBut kind of 2-recombination,

Be respectively node cm _i, cm _I+1And limit cm _i-cm _I+1Weights.From v _iTo v _I+1Passage channel performance matrix be PA _{I, j+1}Supposing has an orderly node to separate H, if do not exist any other orderly node to separate Q, makes Z _r(Q)≤Z _r(H), r=1,2 ..., n wherein has the strict (Z of establishment of an inequality at least _rBe corresponding target function value), then H is that a noninferior solution or Pareto separate.

" MCC partner selection " multi-objective optimization question can be write as following form:

$\min Z=\underset{i=1}{\overset{n-1}{\mathrm{\&Sigma;}}}{\mathrm{PA}}_{i,i+1}(d_{{\mathrm{cm}}_i}+d_{{\mathrm{cm}}_{i+1}}+{\mathrm{dl}}_{{\mathrm{cm}}_i,{\mathrm{cm}}_{i+1}})---\left(11\right)$

Wherein

Be respectively each node weights, the weights on each limit.Like this, manufacturing enterprise's dynamic partner selects optimization problem promptly to be converted into: find the solution (CEV, CEA) problem of a shortest path of last search at weighted and directed diagraph CEC=.

As can be seen, manufacturing enterprise's dynamic partner selects optimization problem and TSP problem that similar part is arranged.Yet it has the binding feature that some TSP problems are not had:

(1) search of every ant the condition of ending no longer is all nodes of traversal, but arrives end node.

(2) accurate aeoplotropism.But the recombination arbitrarily in twos of the candidate role of same link unit node does not have succession.Simultaneously, arranging of node of candidate role unit is oriented between the different links.Be v _iBut any two candidate's nodes between can produce recombination.v _iCandidate's node lead to v _I+1Candidate's node, otherwise not all right.

(3) all there is weight on node and limit.Weighing computation method difference when candidate role unit node makes up between the candidate role unit node combination of same link, the different link.

Because it is the MCC company choice problem has these characteristics, just no longer suitable at the ant group algorithm of TSP.

The present invention adopts microhabitat ant group algorithm (MACO) to select the affiliate of MCC.

3, partner selection step

From the position of leader role unit, the step of partner selection is as follows:

In the 1st step,, set up corresponding M CC according to the concrete condition of manufacturing enterprise's dynamic alliance.According to the characteristics of demand of each link, determine the index binding occurrence of assessment indicator system and each link of each link then to resource.

In the 2nd step, determine link v according to physical condition _iThe composite behaviour matrix of candidate role unit, calculate v _iBut the 2-recombination that can produce of candidate role unit, and link v _iCandidate role unit to v _I+1Candidate role unit between the channel performance matrix, set up on this basis candidate buddy digraph CEC=(CEV, CEA).

In the 3rd step, utilize microhabitat ant group algorithm (MACO) to select the combination of optimal candidate role unit.

4, application examples

Certain enterprise obtains the market opportunity, produces a kind of injection mold.The link of the MCC of this injection mold has, raw material supply, standard component supply, injection mould overall design, die cavity processing, mold integral assembling etc.If the leader is with regard to this market opportunity bid.There are 46 candidate role units to participate in submitting a tender.According to actual conditions and above analysis, can calculate candidate buddy may combined number be about in theory:

$\underset{i=1}{\overset{9}{\mathrm{\&Pi;}}}{\mathrm{nzp}}_i\&ap;8\&times;{10}^{11}$

Here obviously occurred so-called " shot array ".

Adopt basic ant group algorithm respectively, Max-Min Ant System (MMAS), three kinds of algorithms of MACO resolve this problem, compare.The computer environment that uses is pentiumII, 64MRAM, and Win2k uses the MATLAB Programming with Pascal Language.Parameter is: ant is counted antm=25, maximum iteration time Cyclemax=300, track conservation rate ρ=0.8, α=1, β=3, Q=10.Every kind of algorithm moves the data such as the table 1 that obtain 10 times.

As can be seen from Table 1, basic ant group algorithm just is absorbed in local optimum behind the certain number of times of iteration, be difficult to obtain globally optimal solution.MMAS and MACO then are easier to obtain globally optimal solution, and the two performance is suitable.On convergence, the latter slightly is better than the former.Parameter value is carried out conversion, compare, similar result is also arranged.

Compare with basic ant group algorithm, the parameter that MMAS and MACO need be provided with is more.For MMAS, τ _Max, τ _MinBe provided with the influence of convergence effect and speed of convergence very greatly, to different particular problems, in calculating, draw experience, just the reasonable value of setting can be arranged.For MACO, parameter balanced use of mainly following posterior infromation and heuristic information and the principle of suitably expanding solution space be set.

From last example as can be seen, quantity is 8 * 10 theoretically ¹¹Candidate buddy may make up, utilization MACO can search optimum combination rapidly, this has shown the actual effect of MACO.According to the practice effect, this algorithm is all obtained result preferably to the partner selection of big figure more (the candidate role unit number of each link is in 100) on efficient and effect.

Three kinds of ant group algorithms of table 1 relatively

Claims (3)

1, a kind of dynamic partner selecting method based on ant group algorithm is characterized in that, comprises the steps:

1) according to the concrete condition of manufacturing enterprise's dynamic alliance, set up the corresponding collaborative chain-MCC that makes, then according to the characteristics of demand of each link, determine the index binding occurrence of assessment indicator system and each link of each link to resource;

2) determine link v according to physical condition _iThe composite behaviour matrix of candidate role unit, calculate v _iBut the 2-recombination that can produce of candidate role unit, and link v _iCandidate role unit to v _I+1Candidate role unit between the channel performance matrix, set up on this basis candidate buddy digraph CEC=(CEV, CEA), CEV={1,2 ..., n} is candidate's node set, and CEA={ (i, j) } be the set on limit;

3) utilize microhabitat ant group algorithm (MACO) to select the combination of optimal candidate role unit.

2, according to the described dynamic partner selecting method of claim 1, it is characterized in that the performing step of described microhabitat ant group algorithm (MACO) based on ant group algorithm:

1) initialization: set maximum cycle Cyclemax, ant is counted m.During beginning, all m ant all concentrates on starting point S.Give the quantity of information τ of equal amount on every limit ₀=C, the C constant;

2) routing: ant k is from the S point, according to selection strategy, from the set of the node that is associated with S, selects a node a, and the follow-up limit of a; Then, another node b that connects from this edge again from the set on the limit that is associated with b, selects another limit.If ant is gone to certain node, this node does not have successor node, and the record of this ant through the path deleted in then ant death, and this node of mark is no longer selected this node next time for opening circuit;

By that analogy, up to searching terminal point E.So ant k obtains separating from S to E.After ant k had searched for, remaining ant searched out other path from S to E according to the method identical with k;

3) Local Search: after m ant searched for, try to achieve m and separate, comprise repetition, exchange of information between the ant individuality " explores " the path by the microhabitat information gap again.The microhabitat information gap is by this m being separated the employing genetic manipulation, and heredity intersects, and hereditary variation realizes, utilizes genetic algorithm, carries out Local Search, obtains locally optimal solution;

4) lastest imformation element: when search cycle each time finishes, when promptly all ants arrive destinations, upgrade the pheromones in whole paths;

5) innings optimum solution of demanding perfection: till the current iteration number of times, in all locally optimal solutions of being set up, minimum value separate globally optimal solution as the current iteration number of times, the i.e. distance value of shortest path.

3, according to the described dynamic partner selecting method of claim 1 based on ant group algorithm, it is characterized in that, described 2-can recombinate and be combined into: the situation that the link role is joined together to bear by two role units at most, when two role units unite, if both precedence does not influence the logistics index between them, and the combination of same role unit is equal to single role unit.

Images