CN103049805A - Vehicle route optimization method with time window constraint based on improved particle swarm optimization (PSO) - Google Patents

Abstract

The invention relates to a vehicle route optimization method with time window constraint based on improved particle swarm optimization (PSO). The method comprises the following steps of: step 10, initially setting parameters; step 20, constructing a particle swarm; step 30, decoding particles according to a decoding rule; step 40, calculating fitness for a distribution route obtained by decoding; step 50, searching an individual optimal state and a group optimal state to select the group optimal state Pg, which is the optimal route of the vehicle route problem under the current iteration conditions, determining the condition that the optimal position searched by the particle swarm is the optimal route in the current state, entering step 601 if a stop condition is not reached, and otherwise, entering step 603; step 601, updating the state; step 602, introducing crossover operator operation, and entering step 30 to repeat particle decoding; and step 603, stopping iteration, and outputting the optimal route result of the vehicle route problem.

Description

The vehicle route optimization method with the time window constraint based on Modified particle swarm optimization

Technical field

The invention belongs to computer science, relate to a kind of method for solving of the optimum Vehicle Routing Problems with time window constraint, also relate to simultaneously genetic algorithm and the particle swarm optimization algorithm in using artificial intelligence field.

Background technology

Vehicle Routing Problems (Vehicle Routing Problems, VRP) refer to the client of some, the goods demand of each own varying number, home-delivery center provides goods to the client, be responsible for sending goods by a fleet, organize suitable driving route, make vehicle in an orderly manner by them, satisfying certain constraint condition (such as the goods demand, traffic volume, hand over delivery availability, the vehicle capacity restriction, the distance travelled restriction, time restrictions etc.) under, the target that reaches some problems is (the shortest such as distance, expense is minimum, time is as far as possible few, use that vehicle number lacks as far as possible etc.).Under the time that the consideration demand point arrives for vehicle requires to some extent, among the vehicle routing problem, add the restriction of fashionable window, just become band time window Vehicle Routing Problems (VRP with Time Windows, VRPTW).

The accessed time window constraint that has added the client at VRP with the time window Vehicle Routing Problems.In the VRPTW problem, except running cost, cost function also will comprise owing to early to stand-by period that certain client causes and the service time of client's needs.The VRP problem has caused the subject experts and scholars' such as Geographical Information Sciences, management, operational research, applied mathematics, logistics science and computer utility great attention, and oneself has obtained larger progress, and its achievement is widely applied in emergency route planning system, Material Distributing Systems: An, transportation system and postal delivery receive-transmit system.Yet, complicacy (being proved to be the problem into NP-hard) by Vehicle Routing Problems, when node is larger, be difficult to the exact solution of the problem that obtains, especially for the related a large amount of emergency materials delivery services of emergency service, except considering cost factor, also to consider the factor of the aspects such as distribution time and environment, this is with regard to the modeling that makes problem and find the solution more complex, thereby the research of this difficult problem is also just had more learning value.

Roughly can be divided into two classes for the algorithm of finding the solution Vehicle Routing Problems at present: exact algorithm and heuritic approach.Exact algorithm comprises that mainly branch defines method, collection partitioning algorithm, dynamic programming, integer programming etc., and heuritic approach mainly comprises saves algorithm, scanning algorithm, two-phase method, tabu search algorithm, genetic algorithm, simulated annealing, ant group algorithm, optimum algorithm of multi-layer neural network, particle cluster algorithm etc.Although exact algorithm can obtain exact solution, but calculated amount is very large, and generally the increase along with problem scale is exponential growth, finds the solution overlong time, can only solve the limited simple VRP problem of nodes, and traditional heuritic approach, although shortened computing time, operand has also reduced, but often all can only obtain the approximate solution close to optimum solution, and the scope of application also can only be limited to small-scale VRP problem, and when interstitial content increased, solving precision was often very poor.The tradition heuritic approach is through being usually used in local optimum, and combines with the meta-heuristic algorithm, and local improvement is carried out in the path that oneself has.

Summary of the invention

The objective of the invention is to adopt the stronger Modified particle swarm optimization Algorithm for Solving of global search performance with the Vehicle Routing Problems of time window, realize the higher complicated optimum problem of requirement of real-time is found the solution.

For solving the problems of the technologies described above, the invention provides a kind of vehicle route optimization method with the time window constraint based on Modified particle swarm optimization, comprise the steps:

Step 10: the parameter initialization setting, set study factor c ₁And c ₂, maximum evolutionary generation n _Max, inertial factor initial value ω ₀, inertial factor final value ω _e, crossover probability p _c, total number of particles n in the population;

Step 20: the structure population, in the initial value scope, according to the coded system initialization population X=(X of particle and speed ₁, X ₂..., X _c), the position x of each particle _IdAnd speed v _Id

Step 30: the particle decoding, according to the decoding rule particle is decoded;

Step 40: fitness calculates, and decodes according to the method for " dividing into groups behind the first circuit ", and the distribution route for decoding obtains calculates desired value z, and chromosomal fitness is defined as Fitness=1/z;

Step 50: search individual optimum state and colony's optimum state, after the fitness of particle calculated and finishes, individual optimum state P was selected in oneself optimal-adaptive degree comparison of knowing of the current fitness of each particle and self _i, P _iBe the optimal path of i particle under the current iteration condition; The optimal-adaptive degree that oneself knows with the optimal particle fitness of this iteration in the population and population is relatively selected the optimum state P of colony _g, P _gBe the optimal path of Vehicle Routing Problems under the current iteration condition;

Step 60: condition judgment, termination condition are elected as and are reached maximum iteration time T _Max, perhaps the optimal location that searches of population satisfies evaluation index, and the optimal location that population searches is the optimal path under the current state, if do not reach end condition, then turns to step 601, otherwise turns to step 603;

Step 601: state upgrades, and according to current state new formula more, the state that carries out particle upgrades;

Step 602: introduce the crossover operator operation, according to crossover probability p _c, form constantly population X of t+1 ^T+1Change step 30 over to and repeat the particle decoding;

Step 603: iteration finishes, the optimal path result of output Vehicle Routing Problems.

The present invention has obtained following technique effect:

The present invention is directed to that the integer coding calculated amount is larger in the particle cluster algorithm, comparison of computational results is poor, proposed a kind of coding, coding/decoding method based on the real number ordering; By utilizing the linear decrease inertia weight function based on iterations, both can the balance overall situation and local search ability, speed of convergence and convergence precision that also can balanced algorithm make algorithm find optimum solution with minimum iterations; By using for reference the intersection principle in the genetic algorithm, the parent particle is carried out the crossover operator computing, avoid the particle cluster algorithm Premature Convergence and be absorbed in Local Extremum, increased the diversity of particle.What the present invention proposed calculates the advantages such as simple, that programming realizes easily, stronger robustness and computing velocity are fast based on having with time window Vehicle Routing Problems algorithm of Modified particle swarm optimization, is the study hotspot in the fields such as path planning, goods and materials dispensing and postal delivery transmitting-receiving.

Description of drawings

Fig. 1 is the decoding principle of particle of the present invention.

Fig. 2 is the real coding principle of particle of the present invention.

Fig. 3 is based on the process flow diagram with time window Vehicle Routing Problems algorithm of Modified particle swarm optimization.

Embodiment

Understand and enforcement the present invention for the ease of those of ordinary skills, the present invention is described in further detail below in conjunction with the drawings and the specific embodiments.

Method of the present invention adopts particle coding, the coding/decoding method based on the real number ordering, has avoided population to sink into Local Extremum; Utilization is based on the linear decrease inertia weight function of iterations, both can the balance overall situation and local search ability, and speed of convergence and convergence precision that again can balanced algorithm make algorithm find optimum solution with minimum iterations; Use for reference the intersection principle in the genetic algorithm, the parent particle is carried out the crossover operator computing, avoid the particle cluster algorithm Premature Convergence, increased the diversity of particle.Algorithm after the improvement has good global optimizing performance, has improved the computational accuracy of algorithm, can seek more excellent feasible solution.

Band time window Vehicle Routing Problems method based on Modified particle swarm optimization provided by the invention may further comprise the steps as shown in Figure 3.

Step 10: method begins, and initiation parameter is set.Set study factor c ₁And c ₂, maximum evolutionary generation n _Max, inertial factor initial value ω ₀, inertial factor final value ω _e, crossover probability p _c, total number of particles n in the population.

Step 20: structure population.In the initial value scope, according to the coded system initialization population X=(X of particle and speed ₁, X ₁..., X _n), the position x of each particle _IdAnd speed v _Id

Each particle represents a solution of D dimension space, and establishing population is I, and t represents the time of particle, then the state representation X of i particle _i=(x _I1, x _I2..., x _ID), the velocity vector of i particle is expressed as V _i=(v _I1, v _I2..., v _ID), the optimum state that i particle lives through is designated as P _i=(p _I1, p _I2..., p _ID), the optimum state P that colony lives through _gExpression at t+1 moment state renewal equation is:

$\left\{\begin{array}{c}{v^{t+1}}_{\mathrm{id}}={v^t}_{\mathrm{id}}+{\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="HSA00000843375800011.png,HSA00000843375800012.png"\; state="\{\{state\}\}">c</figure-callout>}}_1\&times;\mathrm{rand}\left(\right)\&times;({p^t}_{\mathrm{id}}-{x^t}_{\mathrm{id}})+{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HSA00000843375800011.png,HSA00000843375800012.png"\; state="\{\{state\}\}">c</figure-callout>}}_2\&times;\mathrm{rand}\left(\right)\&times;({p^t}_{\mathrm{gd}}-{x^t}_{\mathrm{id}})\\ {x^{t+1}}_{\mathrm{id}}={x^t}_{\mathrm{id}}+{v^{t+1}}_{\mathrm{id}}\end{array}\right.$

Wherein, rand () is evenly distributed on [0,1] interval random number, c ₁, c ₂The study factor, c ₁The expression particle is to the dependence situation of self experience, c ₂The expression particle is to the dependence situation of community information.To speed V _t, maximal rate V is arranged _MaxAs restriction.

In order better to control search and the development ability of the method, improve the convergence of basic particle group algorithm, in the speed evolution equation, introduce inertia weight ω, so rate equation becomes:

v ⁱ⁺¹ _id＝ωv ^t _id+c ₁×rand()×(p ^t _id-x ^t _id)+c ₂×rand()×(p ^t _gd-x ^t _id)

The definition inertia weight is the linear decrease function of iterations:

$\mathrm{\&omega;}={\mathrm{\&omega;}}_0-\frac{{\mathrm{\&omega;}}_0-{\mathrm{\&omega;}}_e}{n_{\max }}n$

In the formula: ω ₀Be the initial value of inertial factor, ω _eBe the end value of inertial factor, n _MaxBe maximum iteration time, n is the current iteration number of times.

Step 30: particle decoding.According to the decoding rule particle is decoded.

Particle decoding rule supposes that the client task number is 8 as shown in Figure 1, and vehicle number is 3, if the position vector X of certain particle is:

Client task number: 12345678

X _v：1 2 2 2 2 2 3 3

X _r：1 5 3 2 4 1 1 2

X wherein _vRepresent the vehicle that each task is corresponding, X _rRepresent the execution order of each task in the vehicle route of correspondence.Position vector X represents that vehicle 1 finishes 1, and vehicle 2 finishes the work 2,3,4,5,6, and execution order is 6,4,3,5,2, and vehicle 3 finishes the work 7,8, and execution order is 7,8.

Then the vehicle route of this particle homographic solution is:

Car 1:0 → 1 → 0

Car 2:0 → 6 → 4 → 3 → 5 → 2 → 0

Car 3:0 → 7 → 8 → 0

Fig. 2 is the real coding principle, and the result is as follows after certain particle iteration:

Client task number: 12345678

X _v：1 2 2 2 2 2 3 3

X _r：1 5.3 3.2 3.3 3.1 0.5 2.1 1.2

Vehicle 2 executes the task 2,3,4,5,6, task 6 corresponding X _rMinimum, so vehicle executes the task 6, its corresponding X after the integer order specification task 6 at first _rBecome 1, task 5 corresponding X _rInferior little, therefore second is performed, its corresponding X after the integer order specification task 5 _rBecome 2, the rest may be inferred for other task.X _rAgain become after the standard:

X _r：1 5 3 4 2 1 2 1

This particle expression is simple, is convenient to program and realizes.

Step 40: fitness calculates.At first decode according to the method for " dividing into groups behind the first circuit ".Distribution route for decoding obtains calculates desired value z.Chromosomal fitness is defined as Fitness=1/z.

Desired value is defined as follows:

$\mathrm{Min\; z}=\underset{i=0}{\overset{n}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{n}{\mathrm{\&Sigma;}}}\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}{c}_{\mathrm{ij}}{x}_{\mathrm{ijk}}$

Wherein:

$\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ijk}}\&le;K,(i=0)$

$\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ijk}}=1,(i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;n;i\&NotEqual;j)$

$\underset{i=0}{\overset{n}{\mathrm{\&Sigma;}}}{d}_i\underset{j=0}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ijk}}\&le;Q,(i\&NotEqual;j,\&ForAll;k\&Element;[0,K-1\left]\right)$

t ₀＝0

t _i+t _ij+s _i-M(1-x _ijk)≤t _j (i，j∈[1，n]；i≠j；k∈[0，K-1])

b _i≤t _i≤e _i

$\underset{i=0}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{jhk}}-\underset{j=0}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{hjk}}=0,(j\&Element;V,k\&Element;V_e)$

Variable and parameter symbol definition are as follows:

N: the total quantity of customer's location;

I: the customer's location sequence number of setting out, i=1,2 ..., n;

J: purpose customer's location sequence number, j=1,2 ..., n;

K: vehicle fleet;

K: the vehicle sequence number, i=1,2 ..., K;

C _u: the distribution cost from i to j;

x _Ijk: represent that k car is from from i to j;

d _i: the demand of customer's location i;

Q: the delivered payload capability of car;

t _Ij: vehicle is from customer's location i to the j running time;

t _i: from customer's location i constantly;

t _j: the moment that arrives customer's location j;

M: normal number;

s _i: customer's location i service time;

b _i, e _i: [b _i, e _i] be the time window of client i, wherein, b _iThe initial point of customer requirement arrival time section, e _iIt is the terminal point of customer requirement arrival time section;

Step 50: search individual optimum state and colony's optimum state.After the fitness of particle calculated and finishes, individual optimum state P was selected in oneself optimal-adaptive degree comparison of knowing of the current fitness of each particle and self _i, P _iBe the optimal path of i particle under the current iteration condition; The optimal-adaptive degree that oneself knows with the optimal particle fitness of this iteration in the population and population is relatively selected the optimum state P of colony _g, P _gBe the optimal path of Vehicle Routing Problems under the current iteration condition.

Step 60: condition judgment.If do not reach end condition, turn to step 30, otherwise stop, iteration finishes.Termination condition is generally elected as and is reached maximum iteration time T _Max, perhaps the optimal location that searches of population satisfies evaluation index, and the optimal location that population searches is the optimal path under the current state.

Step 601: state upgrades.According to state renewal equation formula, the state that carries out particle upgrades.

Step 602: crossover operator.According to crossover probability p _c, form constantly population X of t+1 ^T+1

In order to enlarge the search volume, find more excellent solution, the while is for fear of Premature Convergence and be absorbed in Local Extremum, increase the diversity of particle, the thought of genetic algorithm is dissolved in the particle cluster algorithm, introduce the crossover operator operation, every one dimension crossover operator of particle state and speed is as follows.P in the formula _ChildExpression filial generation particle, p _ParntExpression parent particle, p _cThe expression crossover probability.

$p_{\mathrm{child}}^1\left(x_i\right)=p_c\&times;p_{\mathrm{parent}}^1\left(x_i\right)+(1-p_c)\&times;p_{\mathrm{parent}}^2\left(x_i\right)$

$p_{\mathrm{child}}^2\left(x_i\right)=p_c\&times;p_{\mathrm{parent}}^2\left(x_i\right)+(1-p_c)\&times;p_{\mathrm{parent}}^1\left(x_i\right)$

$p_{\mathrm{child}}^1\left(v_i\right)=\frac{p_{\mathrm{parent}}^1\left(v_i\right)+p_{\mathrm{parent}}^2\left(v_i\right)}{|p_{\mathrm{parent}}^1\left(v_i\right)+p_{\mathrm{parent}}^2\left(v_i\right)|}\left|p_{\mathrm{parent}}^1\left(v_i\right)\right|$

$p_{\mathrm{child}}^2\left(v_i\right)=\frac{p_{\mathrm{parent}}^1\left(v_i\right)+p_{\mathrm{parent}}^2\left(v_i\right)}{|p_{\mathrm{parent}}^1\left(v_i\right)+p_{\mathrm{parent}}^2\left(v_i\right)|}\left|p_{\mathrm{parent}}^2\left(v_i\right)\right|$

p _cBe the random number (empirical value is about 0.2) between [0,1], two parents intersect the filial generation particles that obtain, and calculate its fitness, if its fitness greater than the particle of parent, then the parent particle is replaced, if less than, then give up.

Step 603: finish Output rusults.

Table 1 is to utilize Solomon Benchmark data set to the vehicle route result of algorithms of different test.Solomon Benchmark data set is at present with the most frequently used standard testing collection of time window Vehicle Routing Problems, is that Solomon is at the VPRTW standard testing exam pool of nineteen eighty-three design.Solomon Benchmark test set has 56 groups of test datas, position relationship according to node can be divided into test data 3 large classes: R class, C class and RC class, wherein node is stochastic distribution in the R class data, concern without obvious bunch of collection between node location, node is collection bunch formula distribution in the C class data, and node is distributed near several centers, and RC class data fall between, part of nodes is stochastic distribution, and part of nodes is collection bunch formula and distributes.Different according to test data time scheduling level, test data further can be subdivided into 6 groups again: R class data can be divided into R1 class, R2 class, C class data can be divided into C1 class, C2 class, RC class data can be divided into RC1 class, RC2 class, wherein R1 class data are the short term scheduling data, R2 class data are long-term data dispatching, and C1, C2, RC1, RC2 criteria for classification are similar with it.Solomon Benchmark test set has covered the various aspects with the time window Vehicle Routing Problems substantially.Overstriking is present best arithmetic result in the table, and overstriking and italic are that the present invention can not get present best result.As can be seen from the table: algorithm of the present invention has 3 to be better than at present best result, and 49 can get best up till now result, and 4 can not get at present best result.

Table 2 is average vehicle route results of Solomon Benchmark data set algorithms of different.The result adds up according to C1, C2, R1, R2, these 6 types of RC1, RC2.As can be seen from the table: the average result of algorithm of the present invention is better than genetic algorithm, classical particle group's algorithm and classical ant colony optimization algorithm, and the most approaching at present best result.

Table 3 is the algorithm average operating times according to C1, C2, R1, R2, these 6 type statistics of RC1, RC2.As can be seen from the table: classical ant colony optimization algorithm is consuming time the longest, genetic algorithm secondly, classical particle group's algorithm takes second place, algorithm of the present invention is consuming time the shortest.

In sum, algorithm difference with the prior art of the present invention is integrated use based on the coding of real number ordering, coding/decoding method, linear decrease inertia weight function based on iterations, intersection principle in the genetic algorithm, so that obtaining 3 for Solomon Benchmark test set, algorithm of the present invention is better than at present best result, and with classical ant group algorithm, classical particle group's algorithm is compared with genetic algorithm, speed of convergence obviously improves, iterations significantly reduces, ability of searching optimum also increases to some extent, and performance obviously is better than classical ant group algorithm, classical particle group's algorithm and genetic algorithm.

Table 3

Problem	Genetic algorithm	The classical particle colony optimization algorithm	Classical ant colony optimization algorithm	The inventive method
					The C1 class	98	85	111	62
The C2 class	312	231	338	135
					The R1 class	112	81	124	60
The R2 class	309	273	553	182
					The RC1 class	79	80	82	58
The RC2 class	297	257	333	149

Claims (5)

1. the vehicle route optimization method with the time window constraint based on Modified particle swarm optimization is characterized in that comprising the steps:

Step 10: the parameter initialization setting, set study factor c ₁And c ₂, maximum evolutionary generation n _Max, inertial factor initial value ω ₀, inertial factor final value ω _e, crossover probability p _c, total number of particles n in the population;

Step 20: the structure population, in the initial value scope, according to the coded system initialization population X=(X of particle and speed ₁, X ₂..., X _n), the position x of each particle _IdAnd speed v _Id

Step 30: the particle decoding, according to the decoding rule particle is decoded;

Step 40: fitness calculates, and decodes according to the method for " dividing into groups behind the first circuit ", and the distribution route for decoding obtains calculates desired value z, and chromosomal fitness is defined as Fitness=1/z;

Step 50: search individual optimum state and colony's optimum state, after the fitness of particle calculated and finishes, individual optimum state P was selected in oneself optimal-adaptive degree comparison of knowing of the current fitness of each particle and self _i, P _iBe the optimal path of i particle under the current iteration condition; The optimal-adaptive degree that oneself knows with the optimal particle fitness of this iteration in the population and population is relatively selected the optimum state P of colony _g, P _gBe the optimal path of Vehicle Routing Problems under the current iteration condition;

Step 60: condition judgment, termination condition are elected as and are reached maximum iteration time T _Max, perhaps the optimal location that searches of population satisfies evaluation index, and the optimal location that population searches is the optimal path under the current state, if do not reach end condition, then turns to step 601, otherwise turns to step 603;

Step 601: state upgrades, and according to current state new formula more, the state that carries out particle upgrades;

Step 602: introduce the crossover operator operation, according to crossover probability p _c, form constantly population X of t+1 ^T+1Change step 30 over to and repeat the particle decoding;

Step 603: iteration finishes, the optimal path result of output Vehicle Routing Problems.

2. the vehicle route optimization method with time window constraint based on Modified particle swarm optimization according to claim 1, it is characterized in that: in the described step 20, the method for structure population is:

Each particle represents a solution of D dimension space, and establishing population is I, and t represents the time of particle, then the state representation X of i particle _i=(x _I1, x _I2..., x _Id), the velocity vector of i particle is expressed as V _i=(v _I1, v _I2..., v _ID), the optimum state that i particle lives through is designated as P _i=(p _I1, p _I2..., p _ID), the optimum state P that colony lives through _gExpression at t+1 moment state renewal equation is:

$\left\{\begin{array}{c}{v^{t+1}}_{\mathrm{id}}={v^t}_{\mathrm{id}}+c_1\&times;\mathrm{rand}\left(\right)\&times;({p^t}_{\mathrm{id}}-{x^t}_{\mathrm{id}})+c_2\&times;\mathrm{rand}\left(\right)\&times;({p^t}_{\mathrm{gd}}-{x^t}_{\mathrm{id}})\\ {x^{t+1}}_{\mathrm{id}}={x^t}_{\mathrm{id}}+{v^{t+1}}_{\mathrm{id}}\end{array}\right.$

Wherein, rand () is evenly distributed on [0,1] interval random number, c ₁, c ₂The study factor, c ₁The expression particle is to the dependence situation of self experience, c ₂The expression particle is to the dependence situation of community information; To speed V _i, maximal rate V is arranged _MaxAs restriction.

3. each described vehicle route optimization method with the time window constraint based on Modified particle swarm optimization according to claim 1-2, it is characterized in that: in the described step 20, in the method for structure population, at constantly state renewal equation introducing of t+1 inertia weight ω, its rate equation is set to:

v ^t+1 _id＝ωv ^t _id+c ₁×rand()×(p ^t _id-x ^t _id)+c ₂×rand()×(p ^t _gd-x ^t _id)

Wherein $\mathrm{\&omega;}={\mathrm{\&omega;}}_0-\frac{{\mathrm{\&omega;}}_0-{\mathrm{\&omega;}}_e}{n_{\max }}n$

In the formula: ω ₀Be the initial value of inertial factor, ω _eBe the end value of inertial factor, n _MaxBe maximum iteration time, n is the current iteration number of times.

4. each described vehicle route optimization method with the time window constraint based on Modified particle swarm optimization according to claim 1-3, it is characterized in that: in the described step 40, the concrete grammar that fitness calculates is that desired value is defined as:

$\mathrm{Min\; z}=\underset{i=0}{\overset{n}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{n}{\mathrm{\&Sigma;}}}\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}{c}_{\mathrm{ij}}{x}_{\mathrm{ijk}}$

Wherein:

$\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ijk}}\&le;K,(i=0)$

$\underset{k=0}{\overset{K-1}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ijk}}=1,(i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;n;i\&NotEqual;j)$

$\underset{i=0}{\overset{n}{\mathrm{\&Sigma;}}}{d}_i\underset{j=0}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ijk}}\&le;Q,(i\&NotEqual;j,\&ForAll;k\&Element;[0,K-1\left]\right)$

t ₀＝0

t _i+t _ij+s _i-M(1-x _ijk)≤t _j (i，j∈[1，n]；i≠j；k∈[0，K-1])

b _i≤t _i≤e _i

$\underset{i=0}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ihk}}-\underset{j=0}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{hjk}}=0,(j\&Element;V,k\&Element;V_e)$

Variable and parameter symbol definition are as follows in the formula:

N: the total quantity of customer's location;

I: the customer's location sequence number of setting out, i=1,2 ..., n;

J: purpose customer's location sequence number, j=1,2 ..., n;

K: vehicle fleet;

K: the vehicle sequence number, i=1,2 ..., K;

C _u: the distribution cost from i to j;

x _Ijk: represent that k car is from from i to j;

d _i: the demand of customer's location i;

Q: the delivered payload capability of car;

t _Ij: vehicle is from customer's location i to the j running time;

t _i: from customer's location i constantly;

t _j: the moment that arrives customer's location j;

M: normal number;

s _i: customer's location i service time;

b _i, e _i: [b _i, e _i] be the time window of client i, wherein, b _iThe initial point of customer requirement arrival time section, e _iIt is the terminal point of customer requirement arrival time section.

5. each described vehicle route optimization method with the time window constraint based on Modified particle swarm optimization according to claim 1-3, it is characterized in that: in the described step 602, the concrete grammar of introducing the crossover operator operation is:

Every one dimension crossover operator of particle state and speed is as follows:

$p_{\mathrm{child}}^1\left(x_i\right)=p_c\&times;p_{\mathrm{parent}}^1\left(x_i\right)+(1-p_c)\&times;p_{\mathrm{parent}}^2\left(x_i\right)$

$p_{\mathrm{child}}^2\left(x_i\right)=p_c\&times;p_{\mathrm{parent}}^2\left(x_i\right)+(1-p_c)\&times;p_{\mathrm{parent}}^1\left(x_i\right)$

$p_{\mathrm{child}}^1\left(v_i\right)=\frac{p_{\mathrm{parent}}^1\left(v_i\right)+p_{\mathrm{parent}}^2\left(v_i\right)}{|p_{\mathrm{parent}}^1\left(v_i\right)+p_{\mathrm{parent}}^2\left(v_i\right)|}\left|p_{\mathrm{parent}}^1\left(v_i\right)\right|$

$p_{\mathrm{child}}^2\left(v_i\right)=\frac{p_{\mathrm{parent}}^1\left(v_i\right)+p_{\mathrm{parent}}^2\left(v_i\right)}{|p_{\mathrm{parent}}^1\left(v_i\right)+p_{\mathrm{parent}}^2\left(v_i\right)|}\left|p_{\mathrm{parent}}^2\left(v_i\right)\right|$

P in the formula _ChildExpression filial generation particle, p _ArentExpression parent particle, p _cThe expression crossover probability;

p _cBe the random number between [0,1], two parents intersect the filial generation particles that obtain, and calculate its fitness, if its fitness greater than the particle of parent, then the parent particle is replaced, if less than, then give up.

Images