CN103927584A - Resource scheduling optimization method based on genetic algorithm - Google Patents

Abstract

The embodiment of the invention provides a resource scheduling optimization method based on a genetic algorithm. According to the method, hierarchical region division is carried out according to individual fitness values to work out the average fitness value of hierarchical regions. In the process of selecting population individuals, firstly, roulette planning is carried out according to the average fitness value of the hierarchical regions to determine the number of candidate individuals to be selected from each hierarchical region, and then roulette individual selection is carried out in the same hierarchical region. By means of the method, the purpose of multiple-peak-value resource scheduling is achieved, a self-adaptation mechanism can be established, similar population individuals are prevented from being too concentrated, the diversity of the population individuals is improved, and therefore compared with a traditional elitist selection roulette selection algorithm, a better function solution can be obtained.

Description

A kind of scheduling of resource optimization method based on genetic algorithm

Technical field

The present invention relates to areas of information technology, be specifically related to a kind of scheduling of resource optimization method based on genetic algorithm.

Background technology

Nowadays, on the earth that the mankind depend on for existence and live, various accidents are just presenting the highly trend of unconventionalization, never occur in history or a-hundred-year accident and disaster, but occur more and more frequently now.In the face of various disasteies, emergency resources guarantee is the condition precedent of launching accident emergency management and rescue, requires certain area or trans-regional resource to be rationalized to layout and dynamically allotment, the rapid Optimum scheduling to all kinds of emergency resources of overall importance.

The top priority of solution of emergent event is that the loss that accident is caused is reduced to minimum, emergency resources delivery system towards this complexity, the emergency resources that how effectively scheduling source is various, kind is numerous and jumbled, quantity is huge, to meet disaster area rescue demand, and as much as possible by the damage control in minimum, be the current huge challenges that face of governments at all levels.

Emergency resources scheduling is a kind of of unconventional resource scheduling, and genetic algorithm, as the random search bionic method of a kind of simulating nature biological evolution or mass society behavior, is often applied in scheduling of resource.Genetic algorithm forward the power of genetic evolution from the effect of selecting operator.Roulette selection strategy is a kind of system of selection of selecting and adopting at random based on ratio, mainly according to contemporary population at individual fitness ratio weight, determines that individual inheritance, to follow-on chance, then determines individual selection according to the random pointer of roulette.

This case inventor finds in the process of the research and learning of roulette being selected to operator, although traditional roulette is selected operator simple structure, is widely used, but there are some defects: 1, roulette is selected the fitness function that requires structure adaptive value to be greater than zero, situation that cannot anticipation in objective function codomain, fitness function is chosen has considerable influence to algorithm performance; 2, because evaluation information relies on ideal adaptation degree ratio, easily there is fraud problem, the individual survival ability with higher adaptive value is very strong, and the population at individual diversity choosing is poor, causes entering interlace operation link and occurs that the individual equal probabilities of intersecting is between two higher.Therefore, prior art haves much room for improvement and improves.

Summary of the invention

The embodiment of the present invention provides a kind of scheduling of resource optimization method based on genetic algorithm, to improving the too early speed of convergence of traditional roulette algorithm, abandons too early some search subspaces, sets up adaptation mechanism, improves the diversity of population at individual.

First aspect, the embodiment of the present invention provides a kind of scheduling of resource optimization method based on genetic algorithm, and described method comprises:

Step S1, generate initial population S at random: the N in population S species resource individuality encoded, and all resources are individual forms a population;

Step S2, according to default fitness function, calculate each individual fitness value in population;

Step S3, carry out individual selection, hybridization, mutation operation: the fitness value to all individualities sorts according to size, record best fitness value and the poorest fitness value, by best fitness value and the difference of poor fitness value be divided into M hierarchical region, the individuality that fitness value is best is directly inserted to progeny population, remaining individuality is dispensed to corresponding hierarchical region according to fitness value separately;

Calculate the average fitness value A1 of each hierarchical region, described A1 be in current hierarchical region each individual fitness value sum divided by the individual amount of this grade district inclusion;

Carry out roulette for the first time and select operation, the average fitness value A1 that the selected probability B1 of each hierarchical region is current hierarchical region is divided by the average fitness value sum of all hierarchical regions;

Carry out roulette for the second time and select operation, calculate each individual selected probability B2 in each hierarchical region, described B2 is that this individual fitness value is divided by the fitness sum of all individualities of current hierarchical region;

Calculate each individual selected probability B3 in population S, B3 equals B1 and is multiplied by B2;

Step S4, judge whether current individuality meets end condition, if so, decoding obtains the resource scheduling scheme of local optimum, and if not, the progeny population generating based on individuality returns to step S2 and carries out next iteration processing.

Wherein, described N in population S species resource individuality encoded and comprised: adopt matrix to encode to the N in population S species resource individuality, each individual j kind resource quantity that represents to be dispatched to from i resource provision point destination, wherein, j ∈ N.

Further, described individuality hybridized and comprised:

A point of crossing j of random selection, the individual Parent1 of the parent of matrix coder and Parent2 are divided into respectively to two parts, Parent1 is cut apart to the part that obtains and Parent2 and cut apart the corresponding part obtaining and carry out coordinated transposition, thereby obtain two new offspring individuals.

Preferably, described M equals 3;

Wherein, each self-contained individual amount of a described M hierarchical region is identical or not identical.

Described N in population S species resource individuality encoded and comprised and take scale-of-two or real number, or tree structure is encoded to resource individuality.

The scheduling of resource optimization method based on genetic algorithm that the embodiment of the present invention provides,, take to carry out hierarchical region division according to ideal adaptation degree value, set up the average fitness value of hierarchical region.When selected population is individual, first according to the average fitness value of hierarchical region, carry out roulette planning, the quantity of candidate's individuality is chosen in decision from each hierarchical region, then in same hierarchical region, carrying out roulette selects individual again, by the method, on the resource scheduling that solves multi-peak, can set up adaptation mechanism, automatically suppress population similar individuals too concentrated, improve population at individual diversity, thereby obtain, than traditional elite, select the better Function Solution of roulette selection algorithm.

Accompanying drawing explanation

Fig. 1 is a kind of scheduling of resource optimization method schematic flow sheet based on genetic algorithm that the embodiment of the present invention provides;

Fig. 2 is the individual choice method flow schematic diagram based on genetic algorithm that the embodiment of the present invention provides.

Embodiment

The embodiment of the present invention provides a kind of scheduling of resource optimization method based on genetic algorithm, to improving the too early speed of convergence of traditional roulette algorithm, abandons too early some search subspaces, sets up adaptation mechanism, improves the diversity of population at individual.

In order to make those skilled in the art person understand better the present invention program, below in conjunction with the accompanying drawing in the embodiment of the present invention, technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is only the embodiment of a part of the present invention, rather than whole embodiment.Embodiment based in the present invention, those of ordinary skills, not making the every other embodiment obtaining under creative work prerequisite, should belong to the scope of protection of the invention.

The embodiment of the present invention provides a kind of scheduling of resource optimization method based on genetic algorithm, shown in Figure 1, comprising:

Step S1, generate initial population S at random: the N in population S species resource individuality encoded, and all resources are individual forms a population;

Step S2, according to default fitness function, calculate each individual fitness value in population;

Step S3, carry out individual selection, hybridization, mutation operation;

Step S4, judge whether current individuality meets end condition, if so, decoding obtains the resource scheduling scheme of local optimum, and if not, the progeny population generating based on individuality returns to step S2 and carries out next iteration processing.

The selection step of wherein said step S3 is shown in Figure 2, comprising:

Step D1, the fitness value of all individualities is sorted according to size, record best fitness value and the poorest fitness value, by best fitness value and the difference of poor fitness value be divided into M hierarchical region, the individuality that fitness value is best is directly inserted to progeny population, remaining individuality is dispensed to corresponding hierarchical region according to fitness value separately;

Step D2, calculate the average fitness value A1 of each hierarchical region, described A1 is the individual amount that in current hierarchical region, each individual fitness value sum comprises divided by this stratum;

Step D3, carry out for the first time roulette and select operation, the average fitness value A1 that the selected probability B1 of each hierarchical region is current hierarchical region is divided by the average fitness value sum of all hierarchical regions;

Step D4, carry out for the second time roulette and select operation, calculate each individual selected probability B2 in each hierarchical region, described B2 is that this individual fitness value is divided by the fitness sum of all individualities of current hierarchical region;

Each individual selected next step hereditary probability B3 that enters in step D5, calculating population S, B3 equals B1 and is multiplied by B2.

Wherein, described the fitness value of all individualities is comprised according to order from small to large and being arranged and order is from big to small arranged according to size sequence.

Wherein, described N in population S species resource individuality encoded and comprised: adopt matrix to encode to the N in population S species resource individuality, each individual j kind resource quantity that represents to be dispatched to from i resource provision point destination, wherein, j ∈ N.

Further, described individuality hybridized and comprised:

A point of crossing j of random selection, the individual Parent1 of the parent of matrix coder and Parent2 are divided into respectively to two parts, Parent1 is cut apart to the part that obtains and Parent2 and cut apart the corresponding part obtaining and carry out coordinated transposition, thereby obtain two new offspring individuals.

Preferably, described M equals 3;

Wherein, each self-contained individual amount of a described M hierarchical region is identical or not identical.

Described N in population S species resource individuality encoded and comprised and take scale-of-two or real number, or tree structure is encoded to resource individuality.

It should be noted that, the method that the embodiment of the present invention provides is all suitable for for solving Richest problem, and Richest problem mainly comprises:

1, the whole search volume of Richest problem is the whole world 6,000,000,000 people, and search volume may be partitioned into again the constantly subspace of segmentation such as continent, country, area;

2,100 people of initial random selection, wherein have Dane, American from developed country, have Russian, Mexican and Chinese from developing country, also have some people from less developed country;

3, the case study of traditional roulette selection strategy.

Due to country variant, different regions in global range, even people's gap between the rich and the poor of areal is huge, and obviously Richest problem model is a multi-peak problem very intuitively.

For example, Denmark is country the richest in western developed country, wealth is very high per capita, compare with Denmark, wealth is lower per capita in the countries such as developing Mexico, Russia, China, if according to traditional roulette selection strategy, the individuality that belongs to sub-search volume Denmark just has compared with more options probability.Obviously, Denmark is the richest in the whole world per capita, but does not have so-called top rich man; On the contrary, Russia, Mexico, China is the very high place of Gini coefficient like this, and gap between the rich and the poor very large (be Richest problem function precipitous and concuss), but there will be the richest somebody in the world, as Mexico telecommunication magnate Carlos James Slim He Lu etc.

Therefore, with traditional roulette selection strategy, solve similar Richest problem model, be easier to cause being absorbed in locally optimal solution.

In real world, according to economic development level, the index such as rich level divides global various countries for a plurality of grades such as " developed country ", " developing country ", " less developed countries " per capita, according to individual wealth index, be divided into rich stratum, middle class and the poor stratum again, its middle-class is divided according to being relevant average fitness.

Using such method, the present invention improves traditional roulette selection strategy, according to ideal adaptation degree value information, divides individual affiliated stratum, further obtains the average fitness of corresponding stratum.When solving Richest problem, first according to stratum's average fitness, carry out roulette selection and from each stratum, choose the quantity of candidate's individuality with planning, then in same stratum, again carry out roulette and select individual.For avoiding random crossover and mutation to cause each to take turns, select individual destruction, adopt elite to select genetic algorithm, to substitute the poorest individuality after crossover and mutation operation.

Thisly according to ideal adaptation degree, contemporary population at individual is carried out to stratum's division, set up stratum's average fitness, first according to stratum's average fitness, carry out roulette planning, determine the quantity of each stratum's individuality; Then the individuality in same stratum is carried out the selection strategy of roulette according to ideal adaptation degree ratio, the present invention is referred to as sublevel secondary roulette selection strategy.Sublevel secondary roulette selection strategy, has certain flexibility, can effectively avoid because ideal adaptation degree difference is excessive, causes and selects diversity decline and poor " potentiality " individuality of fitness to be eliminated.

Therefore the scheduling of resource optimization method based on genetic algorithm that the embodiment of the present invention provides, takes to carry out hierarchical region division according to ideal adaptation degree value, sets up the average fitness value of hierarchical region.When selected population is individual, first according to the average fitness value of hierarchical region, carry out roulette planning, the quantity of candidate's individuality is chosen in decision from each hierarchical region, then in same hierarchical region, carrying out roulette selects individual again, by the method, on the resource scheduling that solves multi-peak, can set up adaptation mechanism, automatically suppress population similar individuals too concentrated, improve population at individual diversity, thereby obtain, than traditional elite, select the better Function Solution of roulette selection algorithm.

The embodiment of the present invention also provides a kind of scheduling of resource optimization method based on genetic algorithm, comprising:

(1), initialization population: $S\left(t\right)=\{x_1^t,x_2^t,\mathrm{\Λ},x_N^t\},x_i^t\∈P\left(t\right),t=0,i\∈[1,N];$

(2), according to fitness function f(x), calculate each individuality in S (t) fitness value

(3), the fitness value for all individualities of population to t according to sorting from high to low, record best fitness value f _bestthe poorest fitness value f _worst, according to elite's retention strategy, retain fitness value preferably individual, a remaining N-1 individuality is divided into M hierarchical region Z, Diff _tt=f _best-f _worstwhat be best fitness value with the poorest fitness value is poor, Diff _t〃=(f _best-f _worst)/M is divided into M decile stratum by best fitness value and the difference of poor fitness value,

$Z_j^t\left(t\right)=\{x_{1+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t,x_{2+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t,...,x_{{\mathrm{\Σ}}_{k=1}^j{n}_k}^t\},$ And $j\≤M,{\mathrm{\Σ}}_{j=1}^M{n}_j=N-1$

Therefore, N-1 current population at individual is dispensed to each hierarchical region according to ideal adaptation degree value,

$\∀x\∈Z_j^t,f\left(x\right)\∈[f_{\mathrm{worst},}{f}_{\mathrm{worst}}+{\mathrm{Diff}}_t^{\′\′}\×j]=f_{\mathrm{worst},}{f}_{\mathrm{worst}}+j\×\frac{(f_{\mathrm{best}}-f_{\mathrm{worst}})}{M},$ And meet j=(1,2 ..., M).

(4), calculate each hierarchical region interior individual amount nj and average fitness value thereof

${\stackrel{\‾}{F}}_{z_j^t}=\frac{{\mathrm{\Σ}}_{i=1}^{n_1}f\left(x_{i+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t\right)}{n_j}---(2-1)$

(5), roulette is selected operation for the first time, to the corresponding average fitness value of each hierarchical region of M carry out roulette selection, to determine to choose the quantity planning of candidate's individuality, j the probability that hierarchical region is selected from different estate

$P_j=\frac{{\stackrel{\‾}{F}}_{z_j^t}}{{\mathrm{\Σ}}_{k=1}^M{\stackrel{\‾}{F}}_{z_k^t}}---(2-2)$

(6), roulette is selected operation for the second time, because in hierarchical region be not the optimum individual on full population truly, does not therefore adopt elite's retention strategy, hierarchical region

$Z_j^t\left(t\right)=\{x_{1+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t,x_{2+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t,...,x_{{\mathrm{\Σ}}_{k=1}^j{n}_k}^t\}$

In individual selected probability

$P_{j,k}^{\′\′}=\frac{f\left(x_{i+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t\right)}{{\mathrm{\Σ}}_{i=1}^{n_j}f\left(x_{i+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t\right)}---(2-3)$

According to formula (2-1), formula (2-2), formula (2-3), can calculate population middle individuality the selected probability that enters next step genetic manipulation

$\begin{array}{c}{P}_i^t=P_j\×P_{j,k}^{\′\′}=\frac{{\stackrel{\‾}{F}}_{z_j^t}}{{\mathrm{\Σ}}_{k=1}^M{\stackrel{\‾}{F}}_{z_k^t}}\×\frac{f\left(x_{i+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t\right)}{{\mathrm{\Σ}}_{i=1}^{n_j}f\left(x_{i+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t\right)}\\ =\frac{\left({\mathrm{\Σ}}_{i=1}^{n_j}f\left(x_{i+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t\right)\right)}{{\mathrm{\Σ}}_{n_h=1}^M\left(\frac{{\mathrm{\Σ}}_{i=1}^{n_j}f\left(x_{i+{\mathrm{\Σ}}_{k=1n_k}^{j-1}}^t\right)}{n_j}\right)}\×\frac{f\left(x_{i+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t\right)}{{\mathrm{\Σ}}_{i=1}^{n_j}f\left(x_{i+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t\right)}\\ =\frac{f\left(x_{i+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t\right)}{n_j{\mathrm{\Σ}}_{n_h=1}^m\left({\mathrm{\Σ}}_{i=1}^{n_h}f{\left(x_{i+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t\right)}_{/\mathrm{nh}}{}_{}\right)}\end{array}---(2-4)$

N wherein _jfor place hierarchical region $Z_j^t\left(t\right)=\{x_{1+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t,x_{2+{\mathrm{\Σ}}_{k=1}^{j-1}{n}_k}^t,...,x_{{\mathrm{\Σ}}_{k=1}^j{n}_k}^t\}$ Individual amount.This shows, the individual amount of place hierarchical region is larger, its selecteed probability will decrease, on the contrary, individual amount is less, its selecteed probability will strengthen to some extent, reduce to a certain extent fitness similar individuals and entered next step genetic manipulation by mistake more options, can avoid to a certain extent too early being abandoned of the poor individuality of fitness.This genetic method will be set up adaptation mechanism, automatically suppress the too concentrated of population similarity individuality, improve the diversity of population at individual, for solving the optimization of similar Richest problem and multi-peak function, provide a new method.

One of ordinary skill in the art will appreciate that all or part of step in the various flow processs of above-described embodiment is to come the hardware that instruction is relevant to complete by program, this program can be stored in a computer-readable recording medium, storage medium can comprise: ROM (read-only memory) (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), disk or CD etc.

, in the above-described embodiments, the description of each embodiment is all emphasized particularly on different fields meanwhile, in certain embodiment, there is no the part of detailed description, can be referring to the associated description of other embodiment.

A kind of scheduling of resource optimization method based on the genetic algorithm above embodiment of the present invention being provided is described in detail, applied specific case herein mutual principle of the present invention and embodiment are set forth, the explanation of above embodiment is just for helping to understand method of the present invention and core concept thereof; , for one of ordinary skill in the art, according to thought of the present invention, all will change in specific embodiments and applications, in sum, this description should not be construed as limitation of the present invention meanwhile.

Claims (6)

1. the scheduling of resource optimization method based on genetic algorithm, is characterized in that, described method comprises:

Step S1, generate initial population S at random: the N in population S species resource individuality encoded, and all resources are individual forms a population;

Step S2, according to default fitness function, calculate each individual fitness value in population;

Step S3, carry out individual selection, hybridization, mutation operation: the fitness value to all individualities sorts according to size, record best fitness value and the poorest fitness value, by best fitness value and the difference of poor fitness value be divided into M hierarchical region, the individuality that fitness value is best is directly inserted to progeny population, remaining individuality is dispensed to corresponding hierarchical region according to fitness value separately;

Calculate the average fitness value A1 of each hierarchical region, described A1 be in current hierarchical region each individual fitness value sum divided by the individual amount of this grade district inclusion;

Carry out roulette for the first time and select operation, the average fitness value A1 that the selected probability B1 of each hierarchical region is current hierarchical region is divided by the average fitness value sum of all hierarchical regions;

Carry out roulette for the second time and select operation, calculate each individual selected probability B2 in each hierarchical region, described B2 is that this individual fitness value is divided by the fitness sum of all individualities of current hierarchical region;

Calculate each individual selected probability B3 in population S, B3 equals B1 and is multiplied by B2;

Step S4, judge whether current individuality meets end condition, if so, decoding obtains the resource scheduling scheme of local optimum, and if not, the progeny population generating based on individuality returns to step S2 and carries out next iteration processing.

2. method according to claim 1, it is characterized in that, described N in population S species resource individuality encoded and comprised: adopt matrix to encode to the N in population S species resource individuality, each individual j kind resource quantity that represents to be dispatched to from i resource provision point destination, wherein, j ∈ N.

3. method according to claim 2, is characterized in that, described individuality is hybridized and comprised:

A point of crossing j of random selection, the individual Parent1 of the parent of matrix coder and Parent2 are divided into respectively to two parts, Parent1 is cut apart to the part that obtains and Parent2 and cut apart the corresponding part obtaining and carry out coordinated transposition, thereby obtain two new offspring individuals.

4. method according to claim 1, is characterized in that, described M equals 3.

5. method according to claim 1, is characterized in that, each self-contained individual amount of a described M hierarchical region is identical or not identical.

6. method according to claim 1, is characterized in that, described N in population S species resource individuality encoded and comprised and take scale-of-two or real number, or tree structure is encoded to resource individuality.