CN102117441A - Intelligent logistics distribution and delivery based on discrete particle swarm optimization algorithm - Google Patents

Abstract

The invention discloses intelligent logistics distribution and delivery based on a discrete particle swarm optimization algorithm and aims at scheduling the path of a transportation vehicle so as to save the transportation cost. A coding mode and an operator based on the set and the probability are introduced on the basis of a framework of a standard particle swarm optimization algorithm, the particle swarm optimization algorithm which is originally suitable for a continuous space can be introduced into a discrete combined optimized space, so that the problem that the path of the vehicle is scheduled is solved and the advantages of high operation efficiency, strong optimization capacity, strong robustness, and the like endowed by the traditional particle swarm algorithm can be maintained. In addition, by using heuristic information to construct the positions of particles and introduce a local operator search, features of the problem and information contained in the data are utilized, and thereby the solving result of the algorithm is further enhanced. By adopting normalization weighting and decision idea to deal with the target, the transportation path is strived to be shortest while the number of transportation vehicles are required to be minimum, therefore, the transportation cost of logistics distribution and delivery businessmen can be reduced to the maximum extent.

Description

Intelligent logistics distribution based on the discrete particle colony optimization algorithm

Technical field:

The present invention relates to intelligence computation and logistics distribution two big fields, mainly use a kind of discretize particle swarm optimization algorithm that the haulage vehicle in the logistics distribution is dispatched and path optimization based on set and probability.

Background technology:

Vehicle route scheduling is an important content in the logistics distribution research, and the goal in research of this problem is, to the suitable route of a series of customer demand site design, vehicle is passed through in an orderly manner, is satisfying under certain constraint condition, reaches certain optimization aim.Its constraint condition is generally: goods demand, traffic volume, friendship delivery availability, vehicle capacity restriction, distance travelled restriction, time restriction etc., optimization aim is generally: mileage is the shortest, expense is minimum, the time tries one's best less, Fleet size is as far as possible little, the vehicle utilization factor is high.The vehicle route scheduling has comprised the subproblem of the difficult combinatorial optimization problem traveling salesman problem of classical NP as it, so it also is the NP difficulty.In addition, the vehicle route scheduling of band time window is difficult to resolve owing to related to more constraint condition very much, and being proved to be even having found a feasible solution at present all is the NP difficulty.

The vehicle route scheduling of band time window because it presses close to the current demand of logistics company more, has been subjected to extensive concern in the past, the existing diverse ways of having researched and proposed.The method in past mainly can be divided into following two classes: exact algorithm and approximate data.Exact algorithm refers to obtain the algorithm of optimum solution, cuts apart and column-generation algorithm, branch and bound method, pull-type relaxation method, dynamic programming etc. as collection.They have introduced rigorous mathematical method when finding the solution, can guarantee to find the preferred plan of dispensing.But this class algorithm can't be avoided the index exploding problem, can only effectively find the solution small-scale logistics distribution.And all at a certain particular problem design, it is relatively poor to be suitable for ability, thereby its range of application is very limited in practice for these algorithms usually.Development along with the modern computing method, some approximate datas such as local search algorithm, tabu search algorithm, simulated annealing, genetic algorithm etc. all are applied to finding the solution the vehicle path planning problem, can find the solution the haulage vehicle planning of fairly large client site, but they exist shortcomings such as local optimum, algorithm parameter poor robustness.In addition, ant colony optimization algorithm naturally and understandably is applied to vehicle path planning as a kind of method that solves combinatorial optimization problem natively, has also obtained comparatively outstanding result.But ant colony optimization algorithm exists computation process complexity, speed of convergence to wait shortcoming slowly, still has certain limitation.Particle swarm optimization algorithm is as a kind of emerging algorithm in intelligence computation field, and its algorithm performance has been approved extensively that at these several years application is constantly expanded.Therefore, the vehicle route scheduling problem that recently also has the researchist to attempt finding the solution the band time window with particle swarm optimization algorithm, but these researchs only are simply the particle position of continuous space to be rounded the scheme of transporting of describing, it is relatively poor to find the solution effect.

Summary of the invention:

In order to overcome the problem that existing account form is not high enough at computation rate, the scheduling quality is not good, be not suitable for aspects such as extensive logistics distribution, the present invention proposes a kind ofly can be efficiently haulage vehicle to be dispatched discrete particle colony optimization algorithm with path planning, use intelligentized computing method, it is the shortest also to make every effort to the path of transporting when minimizing required haulage vehicle number, thereby reduces logistics distribution merchant's transportation cost substantially.

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

(1) adopts a kind of particle coded system, particle swarm optimization algorithm is applicable to solves discrete combinatorial optimization problem (the vehicle route scheduling belongs to a kind of of combinatorial optimization problem) based on set and probability.Combinatorial optimization problem can be defined as that (Ω), wherein S represents the set of all feasible solutions for S, f, and f is an objective function, and Ω is a constraint condition.The target of problem is exactly under the condition that satisfies constraint Ω, finds one group of feasible solution X ^*∈ S makes the f optimization.In discrete particle cluster algorithm of the present invention, combinatorial optimization problem (S, f Ω) are associated with following feature:

● a universal set E, E can be divided into n dimension, i.e. E=E ¹∪ E ²∪ ... ∪ E ⁿ

● a candidate solution set X ∈ S is associated with general collection E.X ∈ E and X ¹∈ E ¹, X ²∈ E ²..., X ⁿ∈ E ⁿ

● when X satisfied constraint condition Ω, X was a feasible solution.

● the target of algorithm finds one to make the optimized feasible solution X of f exactly ^*

According to as above definition, the process of finding the solution a combinatorial optimization problem with particle swarm optimization algorithm can be considered to select some elements to constitute the process of the subclass of general collection E with the optimization aim function.

(2) adopt particle swarm optimization algorithm with integrated learning strategy, particle by in the search volume alternately " study " and " flight " constantly approach the optimum solution position, promptly seek out the scheduling arrangement scheme of optimum.The basic operation of particle swarm optimization algorithm mainly comprises two steps: the first, and Velocity Updating---each particle is by self search experience or the search experience of the colony size and Orientation (i.e. " study ") of adjusting own flying speed; The second, position renewal---each particle is according to self current location and the present speed reposition (i.e. " flight ") that calculates oneself.In the particle swarm optimization algorithm that the present invention proposes, particle has adopted a kind of integrated learning strategy: the different dimensional of same particle is to learn to different models, Mo Fan selection has covered whole particle colony in addition, rather than the particle of simple particle self and current optimum.A benefit of doing like this is the diversity that has strengthened colony's search.Because the search volume of vehicle dispatching problem is often very complicated and changeable, strengthens the population diversity and help to prevent that algorithm is absorbed in local optimum, thereby improve the quality of separating.

(3) decision thought of a kind of normalization weighted sum of employing is considered to minimize vehicle number and minimize two targets of path distance simultaneously.The adaptive value of each particle is its represented weighted sum of separating associated vehicle number and transportation range.Wherein, transportation range has been carried out normalized, made that minimize vehicle number has precedence over and minimize transportation range.This be because, in actual applications, the fees of maintenance of fleet's automobile and driver and express delivery personnel's salary is a very important part in the logistics company expenditure often, therefore logistics company is all paid close attention to most the minimizing that transport point needs vehicle number mostly, when need in the different schemes that obtains be the vehicle of similar number the time, other target such as total distance of transporting will be considered.

The invention has the beneficial effects as follows: adopt the coding strategy based on set and probability, the combinatorial optimization problem space in the vehicle route scheduling can very snugly be characterized; Adopt the particle swarm optimization algorithm of being with the integrated learning strategy to carry out optimizing and find the solution, have the advantage of finding the solution quality height, strong robustness; Adopt normalization weighted sum decision thought guiding searching process, help to minimize the transportation expense simultaneously from vehicle number and two different angles of transportation range.

Description of drawings:

The synoptic diagram of Fig. 1 vehicle path planning problem

The overall flow figure of discrete particle colony optimization algorithm among Fig. 2 the present invention

Fig. 3 is based on the synoptic diagram of the demoder of vehicle-mounted and time window constraint

Embodiment:

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

The synoptic diagram of vehicle path planning problem is as shown in Figure 1: to the client of a series of appointments, determine the vehicle delivery travel route, make vehicle from freight house, pass through a series of customer's locations in an orderly manner, and return freight house.Certain constraint condition (vehicle load, customer demand, time window etc.) is satisfied in requirement, total transportation cost minimum.Vehicle path planning can be expressed on mathematics like this: (C L) is a complete graph, wherein C=(c to G= ₀, c ₁..., c _n) be set of node, L=＜c _i, c _jBe the fillet collection, c _i, c _j∈ C, i ≠ j.In G, c ₀The expression parking lot, remaining node is represented the client.Each node all with a goods demand q _i(the goods demand q in parking lot is associated ₀=0).Each bar limit＜c _i, c _jAll with a t _IjRelevant, t _IjExpression client c _iAnd c _jBetween running time.The vehicle that the parking lot has some to convey goods, the load of each vehicle is constant to be Q.In the vehicle path planning problem of band time window, to parking lot and each client c _i(i=0 ..., n), all a time window [e has been gone up in association _i, l _i], must in this time window, begin to carry out (e to their service ₀Be set to zero-time 0, l0 is all vehicles times of return at the latest).

In particle swarm optimization algorithm, each particle i keeps a velocity vector V _i, a current location vector X _iWith a historical optimal location vector pBest _iAs its name suggests, velocity vector V _iThe speed and the direction of particle i current flight have been determined; Position vector X _iExpression particle i current in the search volume residing position, be the basis of assessing its represented quality of separating; Historical optimal location vector pBest _iWrite down the optimal location (corresponding to the best position of separating assessed value) that particle i finds in search procedure, be used to preserve the search posterior infromation of particle i.Fig. 2 has provided the overall flow figure of particle swarm optimization algorithm.The embodiment of whole algorithm is described below:

1, individual coding

1.1 particle position

Particle position is represented as:

$X_i=[{X_i}^0,{X_i}^1,...,{X_i}^n]---\left(1\right)$

${X_i}^d=[<{\mathrm{<figure-callout\; id="1"\; label="nb"\; filenames="HSA00000367220300012.png"\; state="\{\{state\}\}">nb</figure-callout>}}_1,d>,<d,{\mathrm{nb}}_2>],$ nb ₁，nb ₂∈{0，1，...d-1，d+1，n}，nb ₁≠nb ₂ (2)

Nb wherein ₁, nb ₂=(1 ... j-1, j+1, n) expression is nb with adjacent two nodes of node d ₁And nb ₂, promptly vehicle can be according to＜0 ..., nb ₁, d, nb ₂..., 0〉route carry out goods delivery.This moment, each particle position constituted an oriented Hamilton loop.

By introducing one based on demoder vehicle-mounted and the time window constraint, we can be divided into a series of haulage track with each Hamilton loop, thereby obtain a feasible solution of problem.The principle of work of this demoder is very simple, as shown in Figure 3: from the limit that goes out in parking lot, to each the bar limit in the Hamilton loop, if this limit is satisfied vehicle-mounted and the time window constraint, then keep; Otherwise between two summits of this frontier juncture connection, insert the parking lot, and replace former limit with two new limits that connect the parking lot.This process is equal to along Hamilton loop shipping goods, and when vehicle can't be for next customer service, vehicle went back to the field, and from the parking lot send another car for after customer service, promptly created a new haulage track.

By this particle coded system with based on the demoder of constraint condition, each particle is all characterizing a feasible solution in the problem space.

1.2 particle's velocity

Particle's velocity is defined as:

$V_i=[{V_i}^0,{V_i}^1,...,{V_i}^n]---\left(3\right)$

${V_i}^d=\{<u,v>/p(u,v)|<u,v>\&Element;A^d\}---\left(4\right)$

Wherein p (u, v) ∈ [0,1] is every limit＜u, v〉dependent probability, represent that this limit is in structure selecteed possibility during particle position.In case p (u, v)=0, then with limit＜u, v〉concentrate deletion from speed.

2, Velocity Updating

Particle carries out Velocity Updating according to following formula:

$v_i^d=\mathrm{\&omega;}\&times;v_i^d+c\&times;{\mathrm{rand}}^d\&times;({\mathrm{pBest}}_{\mathrm{fi}\left(d\right)}^d-x_i^d)---\left(5\right)$

ω and c are respectively inertia weight and speedup factor parameter, f _i(d) ∈ 1,2 ..., M} (M is a population size) is referred to as model, the d dimension that has defined particle i will be learnt by the pBest of which particle in population.f _i(d) it is determined by a study probability P c: when Velocity Updating, each dimension of each particle in the population all has the probability of Pc to learn to the pBest of oneself, and the probability of (1-Pc) utilizes this one dimension of the historical optimal location of certain companion's particle of algorithm of tournament selection policy selection to learn in addition.

Operational symbol in the formula (5) is to be based upon on the set and the basis of probability, and the definition of " constant * speed " operational symbol, " speed+speed " operational symbol, " position-position " operational symbol, " constant * position " operational symbol is respectively shown in formula (6)-(9)

$c\&times;{v_i}^d=\{<u,v>/p^{\&prime;}(u,v)|<u,v>\&Element;A^d\},$

$p^{\&prime;}(u,v)=\left\{\begin{array}{cc}1,& \mathrm{ifc}\&times;p(u,v)>1\\ c\&times;p(u,v),& \mathrm{otherwise}\end{array}\right.$

${V_i}^d+{V_j}^d=\{<u,v>/\max (p_i(u,v),p_j(u,v))|<u,v>\&Element;A^d---\left(7\right)$

$X_i^d-X_j^d=U^d=\{<u,v>|<u,v>\&Element;X_i^d\mathrm{and}<u,v>\&NotElement;X_j^d\}---\left(8\right)$

c×U ^d＝{<u，v>/p′(u，v)|<u，v>∈A ^d}，

$p^{\&prime;}(u,v)=\left\{\begin{array}{cc}1,& \mathrm{if}<u,v>\&Element;U^d\mathrm{andc}>1\\ c,& \mathrm{if}<u,v>\&Element;U^d\mathrm{and}0\&le;c\&le;1\\ 0,& \mathrm{if}<u,v>\&NotElement;U^d\end{array}\right.---\left(9\right)$

3, position renewal

At first, particle's velocity set (the limit collection of band probability) is converted into the limit collection:

$\mathrm{Cut}\left({v_i}^d\right)=\{<u,v>|<u,v>/p(u,v)\&Element;{v_i}^d\mathrm{andp}(u,v)\&GreaterEqual;\mathrm{rand}\}---\left(10\right)$

The renewal of particle position is a structure property.We are by utilizing the current location X of particle _i(t) and speed V _iMake up the reposition X of particle _i(t+1).At first, from parking lot p ₀=0 beginning, iteration is determined the client of next visit.If current automobile position is k, its next access location k ' can derive from 3 set: speed V _iIn expression with k the client's that the limit links to each other S set is arranged _VParticle current location X _iThe S set that the client that the limit links to each other is arranged with k in the represented limit _X, obviously | S _X|=2; And in the problem space complete graph with the client set S of k adjacency _AIn addition, require k ' not in search taboo table (client who had visited).The privilege of access order of three set reduces successively, promptly at first seeks S _VIn whether the k ' that satisfies condition is arranged, then choose and carry out next step operation if having, otherwise seek S _XIn whether the k ' that satisfies condition is arranged, work as S _VAnd S _XIn all do not satisfy condition k ' time, can be at S _AIn choose k '.If S _V, S _XAnd S _AIn the k ' that all do not satisfy condition, suppose that the parking lot has been inserted into after the k, and reselect k ' (promptly having rebulid a transportation route).

In addition, concentrate in next accessing points of same priority, may have a plurality of k ' that satisfy condition, this moment, we introduced elapsed time amount timespan (k, k ') as heuristic information (heuristic), chose the k ' that makes timespan (k, k ') minimum.The computing method of timespan are as follows:

timespan(i，j)＝max{currtime+t _ij，e _j}-currtime (11)

What its was represented is to beginning to serve the needed time for next client j from present node i.Wherein currtime represents the current time in system, t _IjBe that vehicle travels time of required cost e between i, j node _jBeginning window service time of expression client j.

4, initialization

At the initial phase of algorithm, the particle's velocity quilt is initialize at random; Particle position is with probability

Use greedy algorithm (heuristic information is with reference to (11)) initialize, with probability

Initialize at random; The historical optimal value of particle is made as the current location of particle.

5, objective function

The present invention is when carrying out the scheduling planning in vehicle transport path, and considered two targets simultaneously: primary goal is to minimize vehicle number; When required vehicle number is identical, minimize total transportation route length.This also is the primary demand of the industry of logistics distribution in the past.In algorithm of the present invention, the adaptive value of particle is by the vehicle number of separating (NV) of its representative and travel round total distance (TD) decision.Wherein, vehicle number is as primary optimization aim, and it is a positive integer, and this NV value of representing different feasible solutions differs 1 at least, promptly | and NV (X _i)-NV (X _j) | 〉=1.So, as long as will travel round time normalization in (0,1) interval, again with itself and NV addition as objective function, it must not can have influence on the optimization of NV.Realized that promptly vehicle number is minimised as first optimization aim, and under the identical situation of vehicle number, separating that the time of traveling round is short more will be regarded as more excellent separating, and satisfies the demand of logistics distribution.Used the arc cotangent normalized function among the present invention:

normalize(x)＝arctan(x)/(π/2) (12)

Objective function is defined as:

fitness(X _i)＝NV(X _i)+normalize(TD(X _i))(13)

6, local optimum strategy

In addition, in every the wheel, each particle position is introduced a local optimum strategy after upgrading.Mode is as follows: 1) select the travel route through the minimum automobile of client's number, to attempt inserting traveling round in the route of all the other automobiles by all clients that it is responsible for, inserting prerequisite is the script service time that does not influence all the other clients, and time window and the vehicle-mounted constraint satisfied.2) if all clients of certain automobile process all can be inserted into traveling round in the route of all the other automobiles to be served, then cancel this and travel round automobile.3) give the particle assignment according to the new scheme of always traveling round.

Based on the discrete particle colony optimization algorithm of set and probability,, more be adapted to characterize the search volume of vehicle route problem among the present invention compared with traditional particle swarm optimization algorithm.And, owing to adopted a kind of Velocity Updating mode of integrated learning, can effectively avoid being absorbed in local optimum, the global optimizing ability of algorithm is stronger, and the quality of the distribution project of generation is higher.In addition, the present invention has considered the optimization of fleet vehicle number and total transportation range simultaneously, has greatly saved transportation cost, is of value to the economic benefit that improves logistics company.

Claims (4)

1. at the vehicle path planning problem of being with time window in the logistics distribution industry, a kind of intelligentized scheduling scheme based on the discrete particle colony optimization algorithm has been proposed, it is characterized in that: the main frame of using particle swarm optimization algorithm, and based on the set and the coded system and the operational symbol of probability, the vehicle route problem is found the solution, and the algorithm that the present invention proposes may further comprise the steps and operates:

(1) based on the set and the coded system of probability: the search volume of particle colony is the limit collection of the complete graph that defines of parking lot and client node; Particle position is a subclass of the limit collection of complete graph, limit in this subclass joins end to end and constitutes an oriented Hamilton loop, this Hamilton loop can obtain one group based on demoder vehicle-mounted and the time window constraint by one and send route with charge free, i.e. problem feasible solution; Particle's velocity is the limit collection of band probability, the reposition of the selected structure particle of limit possibility in the sets of speeds, and every associated probability in limit is then represented the possibility of this limit selected structure particle reposition when position renewal;

(2) fitness value of particle adopts and calculates as minor function

fitness(X _i)＝NV(X _i)+normalize(TD(X _i))

Wherein NV represents to transport needed vehicle number, and TD represents total transportation range of all routes, and normalize (x)=arctan (x)/(pi/2) is the arc cotangent normalized function; Particle colony is first target to minimize vehicle number in optimizing process, is second target to minimize transportation range;

(3) employed heuristic information is defined as follows in algorithm initialization stage and particle position renewal process:

timespan(i，j)＝max{currtime+t _ij，e _j}-currtime

What its was represented is to beginning to serve the needed time for next client j from present node i; Wherein currtime represents the current time in system, t _IjBe that vehicle travels time of required cost e between i, j node _jBeginning window service time of expression client j;

(4) initialization: at the initial phase of algorithm, the particle's velocity quilt is initialize at random; Particle position is with probability

Use the greedy algorithm initialize, with probability

Initialize at random; The historical optimal value of particle is made as the current location of particle;

(5) Velocity Updating: particle carries out Velocity Updating according to following formula

ω and c are respectively inertia weight and speedup factor parameter, f _i(d) ∈ 1,2 ..., M} (M is a population size) is referred to as model, the d dimension that has defined particle i will the historical optimal value of which particle be learnt in population; f _i(d) by study probability P c decision: when Velocity Updating, each dimension of each particle in the population all has the probability of Pc to learn to the historical optimal location of oneself, and the probability of (1-Pc) utilizes this one dimension of the historical optimal location of certain companion's particle of algorithm of tournament selection policy selection to learn in addition;

Operational symbol in the Velocity Updating formula is to be based upon on the set and the basis of probability, and " constant * speed " operational symbol and " speed+speed " operational symbol are defined as the change of the probability on limit in the sets of speeds; " position-position " operational symbol is defined as the reducing of limit collection; " constant * position " operational symbol is defined as the limit collection that Jiang Bianji is converted into the band probability;

(6) position renewal: particle position upgrade to be structure property, the selection that makes up the limit of particle position comes from three set: the present speed collection of particle, the current location collection of particle, complete graph limit collection, priority reduce successively; All factors being equal, preference will be give to the level set in, then rely on heuristic information avidly to select the limit of elapsed time minimum;

(7) Local Search: after each particle position upgrades, introduce a Local Search strategy; Select the travel route through the minimum automobile of client's number, will be attempted inserting traveling round in the route of all the other automobiles by all clients that it is responsible for, inserting prerequisite is the script service time that does not influence all the other clients, and time window and the vehicle-mounted constraint satisfied; If all clients of certain automobile process all can be inserted into traveling round in the route of all the other automobiles and serve, then cancel this and travel round automobile, and give the particle assignment according to the new scheme of always traveling round;

(8) the assessment population reaches stop condition if optimize, and then stops whole algorithm and obtains optimum solution; Otherwise, returned for (4) step and continue to optimize population.

2. the discrete particle colony optimization algorithm that is used to find the solution the vehicle route problem according to claim 1, it is characterized in that: adopt a kind of particle coded system based on set and probability, the process of finding the solution a combinatorial optimization problem can be considered to select some elements to constitute the process of a subclass of general collection with the optimization aim function.

3. the discrete particle colony optimization algorithm that is used to find the solution the vehicle route problem according to claim 1, it is characterized in that: adopted a kind of integrated learning strategy, when Velocity Updating, the different dimensional of same particle is to learn to different models, Mo Fan selection has covered whole particle colony in addition, rather than the particle of simple particle self and current optimum.

4. the discrete particle colony optimization algorithm that is used to find the solution the vehicle route problem according to claim 1 is characterized in that: adopt a kind of decision thought of normalization weighted sum, consider simultaneously to minimize vehicle number and minimize two targets of path distance; The adaptive value of each particle is its represented weighted sum of separating associated vehicle number and transportation range; Wherein, transportation range has been carried out normalized, made that minimize vehicle number has precedence over and minimize transportation range.

Images