CN102323949A - Keyword optimization classification method based on fuzzy genetic algorithm - Google Patents

Abstract

The invention discloses a keyword optimization classification method based on a fuzzy genetic algorithm, comprising the steps of: (1) randomly assorting units to be optimized into each island, and initializing the N units to be different units in each island; (2) selecting the units which simultaneously participate the internal genetic operation of the island and transfer to other islands; (3) crossing the units selected by the step (2); (4) performing nonuniform mutation operation to all units in each island; (5) calculating assessed values for all units in each island; (6) searching local area for the units in each island; and (7) selecting the units in each island, and judging whether the assessed value satisfies optimization rules and then judging whether repeating the abovementioned steps. According to the method, the problem of solving large-scale engineering can be solved, the intelligence is high, the optimization algorithm is clear and simple, and the optimization accuracy is high.

Description

Keyword classification optimization method based on fuzzy genetic algorithm

Technical field

The invention belongs to the engineering information technical field, relate to a kind of classification optimization method, a kind of specifically keyword classification optimization method based on fuzzy genetic algorithm.

Background technology

Along with the needs of modern science technical development, promoted developing rapidly of optimization method, and be penetrated into every field very soon.The seventies in 20th century, optimization method begins to produce branches such as optimal design, optimum control and Optimal Management.To the eighties, in these branches, develop again and new thinner branch: the optimal design in the Optimization of Mechanical Design of field of engineering technology, building structure optimal design and chemical petroleum field etc.In all kinds of scientific researches and engineering practice; Exist the combinatorial optimization problem and complicated scheduling decision problem of a lot of difficulties; These controls all can be summed up as optimization problem in essence with decision problem; The method for solving of research engineering problem optimum solution or satisfactory solution, attractive always and challenge.

With method of steepest descent, Newton method and conjugate direction method etc. is that the traditional optimal algorithm of representative is one type of widely used optimized Algorithm.Method of steepest descent restrains with linear speed; Its method is simple, calculated amount is little, memory space is little; But it requires objective function continuously differentiable, and speed of convergence is very slow near the function minimal point, because gradient is the local property of function; See it is descend fast from the part, be not necessarily the fastest for the whole process of optimum solution of asking.Newton method requirement objective function Second Order Continuous can be little, in the process of iteration, calculate the inverse matrix of hessian matrix, and this is a step of comparison difficulty, and simultaneously, selected initial point can not be too far away from minimal point, otherwise iterative process possibly not restrain.Conjugate direction method is one type of algorithm between method of steepest descent and Newton method; It is than the former fast convergence rate; Avoided Newton method to calculate the difficulty of hessian matrix simultaneously; But still need the first order derivative information of objective function, can only n within the step effectively, the calculating of n after the step is also nonsensical.

This type optimized Algorithm universal demand objective function derivative is continuous; Have calculation of complex, serial and characteristics such as find the solution; In the face of discrete, discontinuous, no derivative, highly during the optimization problem of morbid state, they are powerless, and they are based upon on the basis of part decline; Only can make the approximate processing of simplification to problem, usually can't try to achieve globally optimal solution effectively.

Summary of the invention

Technical matters to be solved by this invention is; Overcome the shortcoming of prior art; A kind of keyword classification optimization method based on fuzzy genetic algorithm with intelligent characteristic that is suitable for extensive problem is provided; Can be applicable to the optimal design in Optimization of Mechanical Design, building structure optimal design and the chemical petroleum field of field of engineering technology; Technical matterss such as the object properties parameter in the building structure optimal design, scheduling of resource decision-making have been played positive effect, and can solve parameter weights in the Optimization of Mechanical Design preferably and select the process model multi-target Combinatorial Optimization in the optimal design in parameter optimization technical matters such as responsive, that optimum weights are difficult to seek and chemical petroleum field, complicated problems such as scheduling decision.

In order to solve above technical matters, the present invention provides a kind of keyword classification optimization method based on fuzzy genetic algorithm, comprises the steps:

(1) initialization colony: to each island, initialization N individuals becomes individualities different in each island, can limit the individuality that comprises specified quantity in the different island with individual Random assignment to be optimized.

(2) individual choice and migration: carry out fitness according to fitness function and detect, and select and to participate in inner genetic manipulation in island and the individuality that migrates to other island simultaneously, duplicate selected individuality and it is moved.

(3) the individual intersection: will intersect through the individuality individual and that come from other island migrations of the inner genetic manipulation in participation island that selection operation screens, be combined into the new individuality under the linguistic context; The new individuality that is produced does; Make up through between speech; The filename that comprises two or more sensitive words be mapped as two or more sensitive words represent to become one new individual, the new individuality of all generations is indirect replacements of all filename that comprises two or more sensitive words.

(4) individual variation: to all individualities in the island according to the Probability p of confirming through periodicity modulus gelatinization rule _mUnder varying environment, carry out the non-uniform mutation operation, produce the new individuality of variation.Certain sensitive word occurs in the different files name, and has produced the different contexts meaning, and perhaps two or more sensitive words occur in the different files name, and has produced the different contexts meaning.

(5) individual assessment: to each island, calculate all individual assessed values in the island based on the fitness function in the step (2), all individualities comprise the new individuality that former generation's individuality and intersection, mutation operation produce.

(6) Local Search: after calculating each individual assessed value,, carry out local area search, upgrade the assessed value of individual one to the new individuality that process is intersected in the island, mutation operation produced.

(7) optimizing criterion judges: in each island, select individuality, judge whether assessed value satisfies the principle of optimality, if do not satisfy the principle of optimality, then return step (2), the loop optimization step; If satisfy the principle of optimality, then finish optimization step.

Be that the method on the initialization island described in the said step (1) is represented the individuality in the island for the filename that all is contained certain sensitive word is mapped as this sensitive word to further qualification scheme of the present invention.

Further, the roulette back-and-forth method is adopted in the individual choice operation in the said step (2), and detailed process is:

(1) calculate each individual adaptive value, ideal adaptation degree function is:

$\mathrm{Fit}\left(f\left(x\right)\right)=\left\{\begin{array}{cc}1-0.5*{\left[\right|\frac{f\left(x\right)-b}{a}\left|\right]}^{\mathrm{\&alpha;}},& |f\left(x\right)-b|<a\\ \frac{1}{1+{\left[\right|\frac{f\left(x\right)-b}{a}\left|\right]}^{\mathrm{\&beta;}}},& |f\left(x\right)-b|\&GreaterEqual;a\end{array}\right.$

Wherein

b gets the minimum value of current i in generation; Be b=f (x) * i; α, β is constant value;

(2) calculate this adaptive value proportion in colony's adaptive value summation, represent this individuality selected probability in selection course: $p\left(x_i\right)=f\left(x_i\right)/\underset{n=1}{\overset{i}{\mathrm{\&Sigma;}}}f\left(x_n\right);$

(3) through the selection probability of each individuals, calculate its accumulated probability, the accumulated probability of i individuals is: $p_x\left(x_i\right)=\underset{n=1}{\overset{i}{\mathrm{\&Sigma;}}}p\left(x_n\right);$

(4) random number ε and the p between the generation 0 to 1 _x(x _i) compare the individuality that decides selection.If p _x(x _I-1)＜e＜p _x(x _i), then select the i individuals.

Further, the individual intersection adopted the interleaved mode based on fuzzy mechanism in the said step (3), and concrete grammar is:

$p_c=\left\{\begin{array}{cc}{p}_{c1}-\frac{(p_{c1}-0.6)(f_{\max }-f)}{f_{\max }-f},& f\&GreaterEqual;\stackrel{\&OverBar;}{f}\\ p_{c1},& f<\stackrel{\&OverBar;}{f}\end{array}\right.$

P wherein _C1=0.9, f _MaxBe the maximum adaptation degree value in the colony,

Be the average fitness value of per generation colony, f is the fitness value that gets into the individuality that intersects.

Further, the concrete grammar of periodicity modulus gelatinization rule is in the said step (4):

$p_m\left(t\right)=\frac{\mathrm{\&alpha;}[t-(k+0.5)T_c{]}^2}{{T_c}^2}$

Wherein t is an evolutionary generation, and k is a periodicity, T _cBe period of change, α is a probability adjustment coefficient.

Further; In the said step (6) during Local Search; Appear at the individuality that shines upon in the filename under a plurality of different context meanings for the individual combined crosswise of two or more former generation and carried out assessment; If, last or the assessed value of lower unit than the good situation of the assessed value of artificial selected unit under, keep the better evaluate value; If the value of artificially selected unit keeps the assessed value of artificially selecting the unit than last or under all good situation of the assessed value of lower unit.

Further, judge that the method whether assessed value satisfies the principle of optimality is in the said step (7):

Assessed value error e and assessed value error change Δ e are divided into five fuzzy subsets; NB (big negative); NS (little negative); Z (zero), PS (just little), PB (greatly just); Assessed value error e and assessed value error change Δ e standard are interval [1; 1], then five fuzzy subsets are defined as any fuzzy subset on the interval [1,1]; Rule is constructed as follows, and wherein i rule is: if

and

so

E wherein _iBe the subordinate function of e, Δ E _iBe the subordinate function of Δ e, n is the number of fuzzy rule, u ^*Be total output, c ₃(i) for the output membership function value be 1 center.

The invention has the beneficial effects as follows: the present invention can solve large-scale engineering and find the solution problem, and the scope of application is wider, and intelligent degree is high, and optimized Algorithm is clear simple, and it is high to optimize accurate rate.The present invention is used for genetic algorithm to fuzzy logic theory, and simulation biological evolution process and mechanism come keyword is optimized classification, can obtain fast keyword classification optimization efficiency, robust and optimize performance reliably.The fuzzy mechanism of fuzzy genetic algorithm utilization is applied to individual selection, intersection, variation; The genetic manipulation factor is constantly adjusted in variation according to environment, conforms dynamically, and swarm optimization avoids the part more excellent; Filter out stronger population of adaptability or individuality, and then improve on the whole and optimize accurate rate.The present invention can be applied to the optimal design in Optimization of Mechanical Design, building structure optimal design and the chemical petroleum field of field of engineering technology; Technical matterss such as the object properties parameter in the building structure optimal design, scheduling of resource decision-making have been played positive effect, and can solve parameter weights in the Optimization of Mechanical Design preferably and select the process model multi-target Combinatorial Optimization in the optimal design in parameter optimization technical matters such as responsive, that optimum weights are difficult to seek and chemical petroleum field, complicated problems such as scheduling decision.

Description of drawings

Fig. 1 is a fuzzy genetic algorithm rendering of the present invention;

Fig. 2 is an operational flow diagram of the present invention;

Fig. 3 is a flogic system schematic diagram of the present invention.

Embodiment

A kind of distributed point to point network active probe method that present embodiment provides based on reciprocal feedback structure, the design of this method is as shown in Figure 1: the individuality of existence is: the individuality 101,102,103 etc. in the island 1; Individuality 201,202,203 etc. in the island 2; And the individuality in the island 3 301,302,303 etc.Select individuality 101, the individuality 203 in the island 2 and the individuality 302 in the island 3 in the island 1 respectively, and individuality 101 is migrated to island 2, individuality 203 is migrated to island 3, individuality 302 is migrated to island 1.After this, in each island, intersect, make a variation, assess and judge all individualities, and repeat this flow process, up to obtaining optimal value.Detailed process is shown in following algorithm:

P wherein _cFor intersecting the probability that takes place, p _mBe the probability that variation takes place, N is a population scale, the algebraically of G for stopping evolving, and Tf any one individual fitness function for evolving and producing surpasses Tf, or reproductive order of generation surpasses G, or satisfies optimal conditions, then can stop evolutionary process.

The process flow diagram of method of the present invention is as shown in Figure 2, and the flogic system schematic diagram is as shown in Figure 3, and concrete steps are following:

The first step, initialization colony: according to the part of speech characteristic of individuality, initialization N individuals is as island; Individual Random assignment to be optimized to each island, is become different individualities, can limit the individuality that comprises specified quantity in the different island.Here, all filenames of certain sensitive word that contain separately are mapped as this sensitive word and represent to become population, are appreciated that to all sensitive words it is the indirect replacement of all filename that contains certain sensitive word separately.

Second step, individual choice and migration: carry out fitness based on fitness function and detect, and select and will participate in inner genetic manipulation in island and the individuality that migrates to other island simultaneously, duplicate selected individuality and it is moved.

In island 1, produce the individuality of predetermined number, carry out fitness according to fitness function and detect, and select and to participate in inner genetic manipulation in island and the individuality that migrates to other island (in this case, being island 2) simultaneously.

The method of selection operation adopts the roulette back-and-forth method, and detailed process is shown in following algorithm:

1) calculate each individual adaptive value, ideal adaptation degree function is:

$\mathrm{Fit}\left(f\left(x\right)\right)=\left\{\begin{array}{cc}1-0.5*{\left[\right|\frac{f\left(x\right)-b}{a}\left|\right]}^{\mathrm{\&alpha;}},& |f\left(x\right)-b|<a\\ \frac{1}{1+{\left[\right|\frac{f\left(x\right)-b}{a}\left|\right]}^{\mathrm{\&beta;}}},& |f\left(x\right)-b|\&GreaterEqual;a\end{array}\right.$

Wherein the value of

b is got the minimum value of current i in generation; Be b=f (x) * i

α, β is constant value, and get α=0.5 here, β=2;

2) calculate this adaptive value proportion in colony's adaptive value summation, represent this individuality selected probability in selection course: $p\left(x_i\right)=f\left(x_i\right)/\underset{n=1}{\overset{i}{\mathrm{\&Sigma;}}}f\left(x_n\right);$

3) through the selection probability of each individuals, calculate its accumulated probability, the accumulated probability of i individuals is: $p_x\left(x_i\right)=\underset{n=1}{\overset{i}{\mathrm{\&Sigma;}}}p\left(x_n\right);$

4) random number ε and the p between the generation 0 to 1 _x(x _i) compare the individuality that decides selection.If p _x(x _I-1)＜e＜p _x(x _i), then select the i individuals.

The 3rd step, the individual intersection: will intersect through the individuality individual and that come from other island migrations of the inner genetic manipulation in participation island that selection operation screens, be combined into the new individuality under the linguistic context.

Duplicate selected individuality, participate in the genetic manipulation of 1 inside, island, simultaneously it is migrated to island 2, participate in the genetic manipulation of 2 inside, island.

In island 2, the individuality that screens through selection operation the individuality that comes from island 1 migration and the island 2 intersects, and is combined into the new individuality under the linguistic context.Here the new individuality that is produced does; Make up through between speech; The filename that comprises two or more sensitive words be mapped as two or more sensitive words represent to become one new individual, be appreciated that new individuality for all generations is the indirect replacement of all filename that comprises two or more sensitive words.Employing is based on the interleaved mode of fuzzy mechanism, not only consider when the former generation population just when situation (lateral comparison), and will with reference in the some generations process just when situation of change (vertical analysis), be easy to judge present locally optimal solution and globally optimal solution.Different individualities is adopted different crossover probabilities; For the individuality of fitness value far above the average fitness value of colony; Corresponding to lower crossover probability, make it be able to protect entering to satisfy the ranks of the principle of optimality, and the individuality of relatively low fitness; Give higher crossover probability, increase the probability that makes up between its participation speech.

Method is following:

$p_c=\left\{\begin{array}{cc}{p}_{c1}-\frac{(p_{c1}-0.6)(f_{\max }-f)}{f_{\max }-f},& f\&GreaterEqual;\stackrel{\&OverBar;}{f}\\ p_{c1},& f<\stackrel{\&OverBar;}{f}\end{array}\right.$

P wherein _C1=0.9, f _MaxBe the maximum adaptation degree value in the colony, Be the average fitness value of per generation colony, f is the fitness value that gets into the individuality that intersects.

The 4th step, individual variation: to all individualities in the island according to the Probability p of confirming through periodicity modulus gelatinization rule _mUnder varying environment, carry out the non-uniform mutation operation.

In island 2, the mode of all individualities in the island by periodically fuzzy variation made a variation, produce the new individuality of variation.The new individuality here be appreciated that into; Certain sensitive word occurs in different files name (being different from as the mapped file name of planting group time) lining; And produced the different contexts meaning; Perhaps two or more sensitive words occur in different files name (filename that is shone upon after being different from interlace operation) lining, and have produced the different contexts meaning.In long relatively period, keep lower variation probability, the spatial domain of more fully searching for colony and being covered, and the situation of high variation probability appears in certain period of time, make algorithm can jump out Local Search.Formula is following:

$p_m\left(t\right)=\frac{\mathrm{\&alpha;}[t-(k+0.5)T_c{]}^2}{{T_c}^2}$

Wherein t is an evolutionary generation, and k is a periodicity, T _cBe period of change, generally elect the minimum evolutionary generation of expectation as, α is a probability adjustment coefficient, is the variation Probability p _mExtreme cases.

In island 2, also produce the initial individuality of predetermined number, select to migrate to the individuality on island 3 then through selection operation.Then, duplicate individuality, and it is migrated to island 3.In island 3, the individuality of existence intersects or makes a variation the individuality that will come from island 2 migration and the island 3.In addition, also in island 3, further carry out similar operations, by that analogy until island n.

The 5th step, individual assessment:, calculate all individual assessed values in the island based on the fitness function in the step (2) to each island.

To each island, calculate all individual assessed values in the island, all individualities comprise the new individuality that former generation's individuality and intersection, mutation operation produce.Here new individuality refers to combination between speech and morphs and the new individuality that becomes.The assessed value computing method are continued to use the method in the step 2.

The 6th step, Local Search: after calculating each individual assessed value,, carry out local area search, upgrade the assessed value of individual one to individual in the island.

Local area search representes through the slight modification parameter carrying out search through using the result that fuzzy genetic algorithm obtains, and whether this result exists better result relevant with special parameter for particular individual.Here the parameter of indication is the probability that makes up between individual participation speech, and the probability of participating in making up between speech is directly proportional with the height of ideal adaptation degree.

Know from experience for new that obtains through using fuzzy genetic algorithm and duplicate phenomenon; Individual combined crosswise has appeared in the filename under a plurality of different context meanings because the new individuality that produces possibly be two or more former generation, and the new individuality that perhaps produces possibly be that former generation's individuality has appeared in the filename under a plurality of different context meanings after variation.Carry out Local Search to the individuality of selecting, whether expression search here has any better assessed value.Appear at the individuality that shines upon in the filename under a plurality of different context meanings for the individual combined crosswise of two or more former generation and carried out assessment; The assessed value of last or lower unit than the good situation of the assessed value of artificial selected unit under; Keep the better evaluate value, in addition, artificially the value of selected unit is than under all good situation of the assessed value of last or lower unit; Keep the assessed value of artificial selected unit, and accomplish the local area search in this generation.

In the 7th step, optimize criterion and judge: in each island, select individuality, judge whether assessed value satisfies the principle of optimality,, then return step (2), the loop optimization step if do not satisfy the principle of optimality; If satisfy the principle of optimality, then finish optimization step.

When accomplishing local area search, in each island, select individual and judge one by one, to accomplish generation fuzzy genetic algorithm.

The optimal conditions of present embodiment is made a strategic decision by the fuzzy rule that the filename set that sensitive word shone upon produces.

Individual assessed value in each island after the selection completion local area search is as the sample data of input; A given criterion evaluation value; Two input data: e and Δ e are wherein arranged, and e is the assessed value error, and Δ e is the assessed value error change; Be output as u, u is the individual affiliated fuzzy subset in back that optimizes;

For e and Δ e it is divided into five fuzzy subsets, NB (big negative), NS (little negative), Z (zero), PS (just little), PB (greatly just).E and Δ e standard are in interval [1,1], and then five fuzzy subsets are defined as any fuzzy subset on the interval [1,1].

The fuzzy rule that the filename set that sensitive word shone upon produces through artificial audit heap file name, is confirmed its attribute and carries out the fuzzy membership area dividing that rule is constructed as follows, and wherein i rule is:

If

and

then

E wherein _iBe the subordinate function of e, Δ E _iBe the subordinate function of Δ e, n is the number of fuzzy rule, u ^*Be total output, c ₃(i) for the output membership function value be 1 center.

In this case, judge whether assessed value satisfies the principle of optimality, do not accomplish if be judged as that then flow process is returned last selection operation step, the complex phase of laying equal stress on is with handling.Then, when judging that individual assessed value has satisfied the principle of optimality, think that then this assessed value optimizes, and the end part of speech optimal control follow-up to sensitive word.

This a series of circulation is carried out on each island, and on each island, carried out individual migration.

In the optimization processing method of this embodiment, carry out following optimization process: individual number in the island is 10 island of 100 and assesses, promptly for a generation, individual number is 1000.

As stated, according to this optimization processing method, can more easily obtain the parts of speech classification optimization to sensitive word, this value has the characteristic more approaching with target property.

To those skilled in the art; Can also be according to the core concept design different and of the present invention of actual engineering problem and the intelligent optimization control algolithm of structure oneself; In actual engineering, reach best effect, thereby better solve the optimization problem of actual engineering.What need to specify is, this paper is to be optimized for example with the sensitive word qualitative classification to explanation of the present invention, but after the present invention carried out suitable adjustment, it was equally applicable to the optimizing of other similar engineering problems and finds the solution.

Except that the foregoing description, the present invention can also have other embodiments.All employings are equal to the technical scheme of replacement or equivalent transformation formation, all drop on the protection domain of requirement of the present invention.

Claims (7)

1. the keyword classification optimization method based on fuzzy genetic algorithm is characterized in that, carries out as follows:

(1) initialization colony: to each island, initialization N individuals becomes individualities different in each island with individual Random assignment to be optimized;

(2) individual choice and migration: carry out fitness according to fitness function and detect, and select and to participate in inner genetic manipulation in island and the individuality that migrates to other island simultaneously, duplicate selected individuality and it is moved;

(3) the individual intersection: will intersect through the individuality individual and that come from other island migrations of the inner genetic manipulation in participation island that selection operation screens, be combined into the new individuality under the linguistic context;

(4) individual variation: to all individualities in the island according to the Probability p of confirming through periodicity modulus gelatinization rule _mUnder varying environment, carry out the non-uniform mutation operation;

(5) individual assessment:, calculate all individual assessed values in the island according to the fitness function in the step (2) to each island;

(6) Local Search: after calculating each individual assessed value,, carry out local area search, upgrade the assessed value of individual one to the new individuality that process is intersected in the island, mutation operation produced;

(7) optimizing criterion judges: in each island, select individuality, judge whether assessed value satisfies the principle of optimality, if do not satisfy the principle of optimality, then return step (2), the loop optimization step; If satisfy the principle of optimality, then finish optimization step.

2. the keyword classification optimization method based on fuzzy genetic algorithm according to claim 1; It is characterized in that the method on the initialization island described in the said step (1) is that the filename that will contain certain sensitive word respectively is mapped as this sensitive word and representes the individuality in the island.

3. the keyword classification optimization method based on fuzzy genetic algorithm according to claim 1 is characterized in that, the roulette back-and-forth method is adopted in the individual choice operation in the said step (2), and detailed process is:

(1) calculate each individual adaptive value, ideal adaptation degree function is:

$\mathrm{Fit}\left(f\left(x\right)\right)=\left\{\begin{array}{cc}1-0.5*{\left[\right|\frac{f\left(x\right)-b}{a}\left|\right]}^{\mathrm{\&alpha;}},& |f\left(x\right)-b|<a\\ \frac{1}{1+{\left[\right|\frac{f\left(x\right)-b}{a}\left|\right]}^{\mathrm{\&beta;}}},& |f\left(x\right)-b|\&GreaterEqual;a\end{array}\right.$

Wherein b gets the minimum value of current i in generation; Be b=f (x) * i; α, β is constant value;

(2) calculate this adaptive value proportion in colony's adaptive value summation, represent this individuality selected probability in selection course: $p\left(x_i\right)=f\left(x_i\right)/\underset{n=1}{\overset{i}{\mathrm{\&Sigma;}}}f\left(x_n\right);$

(3) through the selection probability of each individuals, calculate its accumulated probability, the accumulated probability of i individuals is: $p_x\left(x_i\right)=\underset{n=1}{\overset{i}{\mathrm{\&Sigma;}}}p\left(x_n\right);$

(4) random number ε and the p between the generation 0 to 1 _x(x _i) compare the individuality that decides selection, if p _x(x _I-1)＜e＜p _x(x _i), then select the i individuals.

4. the keyword classification optimization method based on fuzzy genetic algorithm according to claim 1 is characterized in that, the individual intersection adopted the interleaved mode based on fuzzy mechanism in the said step (3), and concrete grammar is:

$p_c=\left\{\begin{array}{cc}{p}_{c1}-\frac{(p_{c1}-0.6)(f_{\max }-f)}{f_{\max }-f},& f\&GreaterEqual;\stackrel{\&OverBar;}{f}\\ p_{c1},& f<\stackrel{\&OverBar;}{f}\end{array}\right.$

P wherein _C1=0.9, f _MaxBe the maximum adaptation degree value in the colony, Be the average fitness value of per generation colony, f is the fitness value that gets into the individuality that intersects.

5. the keyword classification optimization method based on fuzzy genetic algorithm according to claim 1 is characterized in that, the concrete grammar of periodicity modulus gelatinization rule is in the said step (4):

$p_m\left(t\right)=\frac{\mathrm{\&alpha;}[t-(k+0.5)T_c{]}^2}{{T_c}^2}$

Wherein t is an evolutionary generation, and k is a periodicity, T _cBe period of change, α is a probability adjustment coefficient.

6. the keyword classification optimization method based on fuzzy genetic algorithm according to claim 1; It is characterized in that; In the said step (6) during Local Search, appeared at the individuality that shines upon in the filename under a plurality of different context meanings for the individual combined crosswise of two or more former generation and carried out and assess, if; Last or the assessed value of lower unit than the good situation of the assessed value of artificial selected unit under, keep the better evaluate value; If the value of artificially selected unit keeps the assessed value of artificially selecting the unit than last or under all good situation of the assessed value of lower unit.

7. the keyword classification optimization method based on fuzzy genetic algorithm according to claim 1 is characterized in that, judges that the method whether assessed value satisfies the principle of optimality is in the said step (7):

Assessed value error e and assessed value error change Δ e are divided into five fuzzy subsets; NB (big negative); NS (little negative); Z (zero), PS (just little), PB (greatly just); Assessed value error e and assessed value error change Δ e standard are interval [1; 1], then five fuzzy subsets are defined as any fuzzy subset on the interval [1,1]; Rule is constructed as follows, and wherein i rule is: if

and

so

E wherein _iBe the subordinate function of e, Δ E _iBe the subordinate function of Δ e, n is the number of fuzzy rule, u ^*Be total output, c ₃(i) for the output membership function value be 1 center.

Images