CN101710384A - Improved particle filtering method based on niche genetic algorithm - Google Patents

Abstract

The invention relates to an improved particle filtering method based on the niche genetic algorithm. The method comprises the following steps of: (1) sampling based on the initial probability distribution to obtain initial particles and setting the initial weight; (2) based on the filtering estimations of M particles at (k-1)th moment, carrying out EKF or UKF on each sampled particle to obtain the mean value and the covariance matrix corresponding to the kth moment, and respectively sampling n particles from each disposal distribution by using Gaussian density as the proposal probability density and using the mean value and the covariance matrix of each particle as the mean value and the covariance matrix of the distribution to obtain a set formed by nM particles; wherein n and M are natural numbers; (3) respectively updating the weights of the Nm particles to obtain the weight of each particle; and (4) when the obtained particle set has particles are less than the effective sample capacity, resampling with the niche genetic algorithm. The invention improves the particle filtering, inhibits the degeneracy phenomena and the particle-lack problem caused by simple random resampling, and improves the diversity and the adaptability of the particles, thereby improving the performance accuracy of the particle filtering.

Description

Improvement particle filter method based on niche genetic algorithm

Technical field

The present invention relates to the nonlinear filtering algorithm field, be specifically related to a kind of improvement particle filter method of niche genetic algorithm.

Background technology

The nonlinear filtering wave technology has all obtained using widely at interior numerous areas in satellite navigation, target following, image recognition, economic analysis etc., the method of the solution nonlinear system estimation problem of Ti Chuing is EKF (EKF) the earliest, and this also is the most extensive proven technique of current application.Along with the expansion of research field and improving constantly of engineering application standard, state under the complication system environment, parameter estimation precision, performance are also had higher requirement, and because the basic thought of EKF comes approximate model by nonlinear system being carried out first-order linearization, and it is based on the hypothesis of Gaussian noise, so its precision property can not be satisfied the demand under many circumstances.In the last few years, proposed no track Kalman filtering (UKF) and continuous Monte Carlo method by continuous research improvement and innovation, and the latter promptly was usually said particle filter (PF) to existing Filtering Estimation theoretical method.

Particle filter is compared with traditional Filtering Estimation method has the advantages that to be applicable to non-linear, non-Gauss system, thereby is applied to numerous areas such as signal trace, voice audio signals enhancing, robot control, fault diagnosis and navigator fix gradually.But it also has the problem of self, and is big as calculated amount, and real-time is relatively poor, the deficient problem that has degradation phenomena in addition and cause because of resampling.Therefore relevant scholar pays close attention to the improvement of PF algorithm always, spreading kalman particle filter (EKPF), no track particle filter (UPF) have appearred, these improved PF utilize EKF or UKF to help determine the significance distribution of particle filter, realize the combination of the two, can effectively improve the degradation phenomena of filtering.In addition, development along with simulated annealing, particle swarm optimization algorithm (PSO), genetic algorithm intelligent evolution algorithm technology such as (GA), in order to solve the deficient problem of particle that resamples and cause, some scholars attempt adopting evolution algorithm that the resampling of particle filter is improved, and then improve the diversity of particle colony in the filtering, and obtain global optimum and estimate, also reached certain effect really by these improvement.

Genetic algorithm (GA) is that algorithm is carried out in a kind of application optimizing quite widely, combines application with PF, can effectively improve the deficient problem of particle.But in the method for utilizing GA to resample, because its hybridization is at random fully, though this randomized hybridization form guarantees the diversity of understanding in the primary stage of optimizing, after the number evolution in generation, a large amount of individualities concentrate on some extreme points, their offspring has caused inbreeding, finally may cause the optimizing result to converge on local optimum, introduces niche technique and then helps to address this problem.Microhabitat (Niche) technology is divided into some classes with each for individuality exactly, select the bigger individuality of some fitness in each class and form a population as an outstanding representative, produce groups of individuals of new generation through hybridization, variation in population and between the different population again, adopt preselected mechanism (Preselection) simultaneously, squeeze mechanism (Crowding), shared mechanism (Sharing) or restriction competition choice mechanism finish selection operation.Genetic algorithm (NGA) based on this niche technique can better keep the diversity of population, has very high global optimizing ability and speed of convergence simultaneously, and obtains the probability of optimum solution can improve GA and carry out Multiple hump function optimization the time.

Therefore use that EKF, UKF are auxiliary to be chosen the PF significance distribution and improve the filtering degradation phenomena, utilize the NGA technology to combine simultaneously with PF, the deficient problem of the particle that improvement causes because of PF resamples, to can play better action to degradation phenomena and the deficient problem of particle that solves PF, thereby improve the overall performance of filtering algorithm.

Summary of the invention

Technology of the present invention is dealt with problems: overcome the deficiencies in the prior art, a kind of improvement particle filter method based on niche genetic algorithm is provided, to improve particle filter, the deficient problem of particle that suppresses its degradation phenomena and resample and to cause because of simple randomization, improve particle diversity and adaptivity, and then improve the precision performance of particle filter.

The objective of the invention is to be achieved through the following technical solutions: based on the improvement particle filter method of niche genetic algorithm, performing step is as follows:

(1) obtains M primary around initial probability distribution sampling, and set initial weight;

(2) estimate by k-1 M particle filter constantly, each sampling particle is carried out EKF or UKF obtain k corresponding constantly average and covariance matrix, adopt Gaussian density as the suggestion probability density, with the average of each particle, covariance matrix average, covariance matrix as this distribution, and from each suggestion distributes, adopt n particle respectively, obtain the set that nM particle formed altogether; N and M are natural number;

(3) a described nM particle is carried out weight respectively and upgrade, obtain the weight of each particle;

(4) the effective sample capacity by the weight calculation population, if the effective sample capacity that obtains is less than effective sample capacity lower limit, lower limit may be selected to be half of sample number, carries out niche genetic algorithm and resamples, and described niche genetic algorithm method for resampling performing step is:

(4.1) producing nM the individual population of forming by step (3) is initial population, and the particle weight that calculates is initial fitness;

(4.2) calculate the fitness of progeny population individuality, according to each individual fitness it is carried out descending sort, M individuality before the memory carries out hereditary selection, intersection, mutation operation to the parent population and obtains progeny population;

(4.3) adopt microhabitat restriction competition selection operation, to merge by the progeny population of genetic manipulation and the parent population of memory, in the population after the merging, the penalty function of sentencing that fitness in the adjacent nearer particle is low is accelerated it and eliminated, particle is moved to high fitness direction faster;

(4.4) by after the above operation, according to this (n+1) M individual new fitness individuality is carried out descending sort, M individuality if the population fitness reaches the setting thresholding, then exported result of calculation, otherwise forwarded step (4.2) to before the memory.

In the described step (4.2) parent population genetic selection operation is comprised the steps:

(4.2.1) to the G of parent colony _tSelect computing, at first calculate all individual fitness summations in the colony $\mathrm{\ΣF}=\underset{i=1}{\overset{\mathrm{nM}}{\mathrm{\Σ}}}F\left(x_i\right);$

(4.2.2) calculate each ideal adaptation degree size, promptly each individuality is genetic to the Probability p in the colony of future generation _Si=F (x _i)/∑ F, wherein F (x _i) be each individual fitness, ∑ F is the summation of ideal adaptation degree;

(4.2.3) simulation gambling dish operates to determine whether each individuality is selected, simulates the shared zone that structure is fitted out each individuality according to selection rate, is equivalent to the covering of the fan in the wheel disc, and each individuality comprises range of choice bound A _i[min, max] is created on [0,1] space equally distributed decimal at random γ ∈ [0,1] at random, judges which individual choice scope A γ drops in, and be then should individuality selected;

(4.2.4) use optimum conversation strategy, find out fitness is the highest in the current colony individuality and the minimum individuality of fitness; If the fitness of optimized individual is more taller than the fitness of total best individuality up to now in the current colony, then with the optimized individual in the current colony as new best individuality up to now; Usefulness best individuality up to now replaces the poorest individuality in the current colony at last, promptly obtains progeny population.

Described step (4.3) microhabitat restriction competition selection operation comprises the steps:

(4.3.1) will combine by M individuality individual and that step (4.2) is remembered, and obtain one and contain (n+1) M individual new colony through nM of obtaining after the hereditary variation operation;

(4.3.2) this (n+1) M individuality obtained per two individual x _iAnd x _jBetween distance

$\left|\right|x_i-x_j\left|\right|=\sqrt{\underset{k=1}{\overset{M}{\mathrm{\Σ}}}{(x_i-x_j)}^2}\left(\begin{array}{c}i=\mathrm{1,2},...,(n+1)M-1\\ j=i+1,...,(n+1)M\end{array}\right)$

(4.3.3) as ‖ x _i-x _jDuring ‖＜L, the fitness size of this two individuality relatively, and the individuality that fitness is little in two individualities sentenced penalty function F (min{x _i, x _j)=Penalty; Wherein L is the habitat radius, and its value is big more, then will be discrete more through the individual distribution of the population after squeezing, and Penalty is a penalty function, x _iAnd x _jBe any two individualities in the colony.

The present invention's advantage compared with prior art is: it is tactful that the present invention utilizes preferred significance distribution of suboptimum nonlinear filtering and niche genetic algorithm to resample, the degradation phenomena that can effectively suppress particle filter, promote particle diversity and adaptability, improve the deficient problem of particle, thereby the performance of particle filter is greatly improved.

Description of drawings

Fig. 1 is the implementation method schematic flow sheet according to the described improvement particle filter of embodiments of the invention;

Fig. 2 is according to the preferred significance distribution synoptic diagram of the described suboptimum nonlinear filtering of embodiments of the invention;

Fig. 3 is according to the described niche genetic algorithm method for resampling of embodiments of the invention process flow diagram.

Embodiment

Particle filter adopts sequential importance sampling algorithm (SIS), is continuous Monte Carlo method, integral operation can be converted into the summation operation of finite point.Its crucial thought is to utilize one group of random sample that has the obedience significance distribution of relevant weights, represents posterior probability density based on the estimation of these samples.When sample number is enough big, this probability estimation will be equal to posterior probability density.

Problems such as sampling can't be resolved, be difficult to but optimum importance function in common existence, therefore can utilize weights importance sampling theory, makes up importance function.Traditional particle filter is to adopt state transition probability as importance function, but because it does not utilize the up-to-date observation to system, make particle depend critically upon model, so the sample bias that produces with actual posteriority distribution is bigger, particularly observation data appears at the afterbody of transition probability or likelihood function is compared with transition probability when too concentrating, at this moment wave filter might lose efficacy, and this also is one of major reason that causes the particle filter degradation phenomena.

Mainly take the method that resamples in order to suppress degradation phenomena, method for resampling commonly used is the stochastic sampling method, and its thought is to come particle is redistributed according to weight, promptly the bigger particle of weights is duplicated, and the particle that weights are less is given up.This can suppress degradation phenomena to a certain extent, reduces the redundant computation for a large amount of little weights particles, has improved efficient.But after application stochastic sampling method resampled, particle was no longer independent, and big weights particle is by massive duplication, and little weights particle fades away.Final all particles all collapse on one or several aspect through after the several times iteration, so just make the insufficient of the sample set change of describing the posterior probability density function, i.e. the remarkable variation of the diversity of particle, and this has just brought particle scarcity problem.

Niche technique is that each is divided into some classes for individuality, select the bigger individuality of some fitness values in each class and form a population, produce groups of individuals of new generation by hybridization, variation in population and between the different population again as the outstanding representative of a class.Adopt certain mechanism that the outstanding representative in the population is remained simultaneously, the fitness that only comes from same population in filial generation is better than the individual just qualified individuality that substitutes in its parent of its parent.Like this, in evolutionary process simultaneously, owing to, make population constantly optimised constantly with excellent individual new population more.By adopting niche technique, make that the genetic algorithm after improving has kept the diversity of population, have very high global optimizing ability and speed of convergence simultaneously.And the problem of particle scarcity is mainly reflected in the multifarious deficiency of particle, has feasibility so use niche genetic algorithm technological improvement resampling.

Therefore, the present invention is directed to degradation phenomena and the deficient problem of particle that particle filter exists, adopt the suboptimum non-linear filtering method, utilize EKF or UKF to produce the significance distribution function, and then guiding sampling particle moves to high likelihood region, propose niche genetic algorithm technological improvement resampling, and designed the improvement particle filter based on above method, Fig. 1 is the implementation method schematic flow sheet that improves particle filter.

Describe the specific embodiment of the present invention in detail below in conjunction with accompanying drawing.

Here it is as follows to set nonlinear discrete systems motion and observation model:

State model

x _k＝f(x _k-1，w _k-1)，

Observation model

z _k＝h(x _k，v _k)。

Wherein f, h are respectively the motion and the observation model equation of system, x _k, z _kRepresent k state and observation vector constantly respectively, and w _k, v _kRepresent k system and observation noise constantly respectively.Below will describe the filtering algorithm process at this model.

It is as follows that the present invention improves the particle filter algorithm process:

1, initialization: moment k=0

A) primary x ₀ ⁱ～p (x ₀), (i=1,2 ..., N _s);

X wherein ₀ ⁱI sampling of expression initial time particle, p (x ₀) expression initialization distribution of particles, N _sBe the sampling population;

B) calculate the particle weight ${\mathrm{\ω}}_0^i=p\left(z_0\right|x_0^i),$ Normalization ${\stackrel{\‾}{\mathrm{\ω}}}_0^i=\frac{{\mathrm{\ω}}_0^i}{\underset{j}{\mathrm{\Σ}}{\mathrm{\ω}}_0^j};$

ω wherein ₀ ⁱ, ω ₀ ⁱThe weight and the normalized weight of i sampling of expression initial time particle, p (z ₀| x ₀) expression primary probability distribution function;

2, extrapolation is upgraded: k＞0 constantly, and Fig. 2 has shown and has used the preferred significance distribution of suboptimum nonlinear filtering and the process that particle upgrades is finished in sampling;

A) a last moment particle is carried out EKF or UKF respectively and obtain corresponding estimation average and covariance

$[{\hat{x}}_k^i,{\hat{P}}_k^i]=\mathrm{EKF}/\mathrm{UKF}(x_{k-1}^i,P_{k-1}^i),(i=\mathrm{1,2},...,N_s)---\left(1\right)$

Wherein

Represent that respectively i particle is in k state estimation and covariance estimation constantly;

B) adopt gaussian density as the suggestion probability density, average and covariance that estimation average that obtains with corresponding particle EKF or UKF and covariance are significance distribution, therefrom sampling produces new particle $x_k^i~q\left(x_k^i\right|x_{k-1}^i,z_k)=N(x_k^i;{\hat{x}}_k^i,{\hat{P}}_k^i);$

C) calculate the weight of sampling particle newly

${\mathrm{\ω}}_k^i\∝{\mathrm{\ω}}_{k-1}^i\frac{p\left(z_k\right|x_k^i)p\left(x_k^i\right|x_{k-1}^i)}{q\left(x_k^i\right|x_{k-1}^i,z_k)}---\left(2\right)$

D) to weights ω _k ⁱBe normalized to

${\stackrel{\‾}{\mathrm{\ω}}}_k^i\frac{{\mathrm{\ω}}_k^i}{\underset{j}{\mathrm{\Σ}}{\mathrm{\ω}}_k^j}---\left(3\right)$

3, resample:

To one of degradation phenomena appropriate test is effective sample capacity N _Eff, be defined as follows

$N_{\mathrm{eff}}=\frac{N_s}{1+\mathrm{Var}\left({\mathrm{\ω}}_k^{*i}\right)}---\left(4\right)$

Wherein ${\mathrm{\ω}}_k^{*i}=p\left(x_k^i\right|z_{1:k})/q\left(x_k^i\right|x_{k-1}^i,z_k)$ It is weight.Though can not calculate N definitely _Eff, but but can draw N _EffEstimated value

${\hat{N}}_{\mathrm{eff}}=\frac{1}{\underset{i=1}{\overset{N_s}{\mathrm{\Σ}}}{\left({\stackrel{\‾}{\mathrm{\ω}}}_k^i\right)}^2}$

As can be seen ${\hat{N}}_{\mathrm{eff}}\&le;N_s,$ Work as N _EffHour mean that serious degradation phenomena is arranged.If N _ThBe effective sample capacity lower limit, if ${\hat{N}}_{\mathrm{eff}}<N_{\mathrm{th}}$ Then resampling more, new particle is $\{x_k^i,{\stackrel{\&OverBar;}{\mathrm{\&omega;}}}_k^i\}\&RightArrow;{\{x_k^i,{\stackrel{\&OverBar;}{\mathrm{\&omega;}}}_k^i\}}^{\&prime;},$ { x wherein _k ⁱ, ω _k ⁱI particle of } ' be is in k particle state and normalized weight through resampling and obtaining constantly;

4, obtain the result at last

${\hat{x}}_k=\underset{i=1}{\overset{N_s}{\mathrm{\&Sigma;}}}{\stackrel{\&OverBar;}{\mathrm{\&omega;}}}_k^i{x}_k^i---\left(6\right)$

${\hat{P}}_k=\underset{i=1}{\overset{N_s}{\mathrm{\&Sigma;}}}{\stackrel{\&OverBar;}{\mathrm{\&omega;}}}_k^i({\hat{x}}_k-x_k^i){({\hat{x}}_k-x_k^i)}^T---\left(7\right)$

Wherein

Being illustrated respectively in k filter state estimation and covariance constantly estimates;

Resampling certain applications niche genetic algorithm technology at wave filter.The present invention adopts the niche technique based on restriction competition selection strategy, basic idea is by comparing the distance between each individuality in the colony, if in certain distance, then the lower individuality of fitness is applied stronger penalty function, greatly reduce its fitness, the probability that this individuality is eliminated in the evolutionary process of back just greatly like this.By this mechanism, in certain distance, will only there be a good individuality in the colony, thereby have both safeguarded the diversity of colony, make again to maintain a certain distance between each individuality, and feasible individuality can spread out in whole individual space.The algorithm flow that is the niche genetic algorithm resampling shown in Figure 3.

The process that niche genetic algorithm resamples is as follows:

1, initialization: evolutionary generation t=0;

A) by prior imformation, set the probability distribution of initial population, sample according to prior probability distribution, thereby produce by nM individual x _i(i=1,2 ..., nM) the initial population G of Zu Chenging ₀

B), set corresponding fitness function F (x), and calculate each individual fitness F (x in the population according to different problems _i) (i=1,2 ..., nM);

2, according to each individual fitness it is carried out descending sort, M individuality before the memory;

3, select operator, to the G of colony _tSelect computing (referring to A);

4, crossover operator is done crossing operation (referring to B) to the individual collections of selecting;

5, mutation operator is done variation computing (referring to C) to the individual collections after intersecting;

6, microhabitat restriction competition selection operation, nM M the individuality individual and that the 2nd step was remembered that variation is obtained combines, and obtains one and contains (n+1) M individual new colony; This (n+1) M individuality obtained per two individual x _iAnd x _jBetween distance, as ‖ x _i-x _jDuring ‖＜L, the fitness size of this two individuality relatively, and the individuality that wherein fitness is lower sentenced penalty function F (min{x _i, x _j)=Penalty.Wherein the L value is big more, then will be discrete more through the individual distribution of the population after squeezing, and if will supplant incompatible individually as early as possible, penalty function Penalty then should the little value of the amount of exhausting;

7, according to this (n+1) M individual new fitness individuality is carried out descending sort, M individuality before the memory;

8, end condition judgment.Can set the termination condition and arrive certain algebraically for the population fitness reaches to set thresholding or evolve, if do not satisfy end condition, evolutionary generation t=t+1 then, and with preceding M individuality in the previous step ordering as the new G of colony of future generation _t, forwarded for the 2nd step then to; If satisfy the termination condition, then export result of calculation, algorithm finishes.

A, selection operator

Genetic algorithm use to select operator to come the operation of selecting the superior and eliminating the inferior of the individuality in the colony.Selection operation is based upon on the basis that the ideal adaptation degree is estimated.Selection course adopts classical roulette method (ratio selection) and optimum conversation strategy.

(1) roulette method

A) calculate all individual fitness summations in the colony $\mathrm{\&Sigma;F}=\underset{i=1}{\overset{\mathrm{nM}}{\mathrm{\&Sigma;}}}F\left(x_i\right);$

B) calculate each ideal adaptation degree size, promptly each individuality is genetic to the Probability p in the colony of future generation _Si=F (x _i)/∑ F, wherein F (x _i) be each individual fitness, ∑ F is the summation of ideal adaptation degree;

C) simulation gambling dish operates to determine whether each individuality is selected.Simulate the shared zone that constructs each individuality according to selection rate, be equivalent to the covering of the fan in the wheel disc.Each individuality comprises range of choice bound A _i[min, max] is created on [0,1] space equally distributed decimal at random γ ∈ [0,1] at random.Judge which individual choice scope A γ drops in, then should individuality selected.

(2) optimum conversation strategy

A) find out individuality and the minimum individuality of fitness that fitness is the highest in the current colony.

B) if the fitness of optimized individual is more taller than the fitness of total best individuality up to now in the current colony, then with the optimized individual in the current colony as new best individuality up to now.

C) usefulness best individuality up to now replaces the poorest individuality in the current colony.

B, crossover operator

Individuality is obtained per two individual x _iAnd x _jBetween distance:

$\left|\right|x_i-x_j\left|\right|=\sqrt{\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{(x_i-x_j)}^2}\left(\begin{array}{c}i=\mathrm{1,2},...,\mathrm{nM}-1\\ j=i+1,...,\mathrm{nM}\end{array}\right)---\left(8\right)$

Select the pairing of approaching individuality, if individual distance is too big, then restriction intersects, and such way helps the microhabitat environment.Two individual x of pairing _iAnd x _jBetween carry out arithmetic and intersect:

$\left\{\begin{array}{c}{x}_i^{\&prime;}=\mathrm{\&alpha;}{x}_j+(1-\mathrm{\&alpha;})x_i\\ x_j^{\&prime;}=\mathrm{\&alpha;}{x}_i+(1-\mathrm{\&alpha;})x_j\end{array}\right.----\left(9\right)$

In the formula: the coefficient when a is a linear combination is defined as [0,1] interval interior equally distributed decimal at random here.Intersection all generates a coefficient at random each time, and progeny population can have good discreteness like this.Having very little probability in the process of intersecting occurs crossing the border, it is the bound that the progeny population gene may exceed variable itself, for this situation, give up this bad subbase because of, subbase because of the bound scope in produce a stochastic variable replace this bad subbase because of; x _i, x ' _iI individual state before and after expression intersects respectively;

C, mutation operator

At first set the variation Probability p _m, produce [0,1] interval random number δ, judge as if δ less than p _m, then morph.At this moment at [x _Min, x _Max] in produce a random particles and replace original particle.That is:

x″ _i＝x _min+β(x _max-x _min) (10)

In the formula: β is [0,1] interval interior equally distributed decimal at random.The variation Probability p _mSetting will the population of filial generation be exerted an influence, variation probability big more then filial generation is big more with respect to the random variation of parent population particle; x _Min, x _MaxRepresent that respectively the state that exists in the population is minimum individual and maximum individual, x " _iI individual state after expression makes a variation respectively.

In sum, the present invention comes preferred particle filtering significance distribution function to resist the degradation phenomena of particle filter by utilizing the suboptimum nonlinear filtering algorithm, the niche genetic algorithm of integrated application simultaneously resamples and improves the deficient problem of particle, thereby further promoted the filter state estimated accuracy and, be useful improvement particle filter for the adaptive ability of system.

Below only be concrete exemplary applications of the present invention, protection scope of the present invention is not constituted any limitation.But its expanded application is in the application of all particle filter algorithms, and all employing equivalents or equivalence are replaced and the technical scheme of formation, all drop within the rights protection scope of the present invention.

Claims (3)

1. based on the improvement particle filter method of niche genetic algorithm, it is characterized in that performing step is as follows:

(1) obtains M primary around initial probability distribution sampling, and set initial weight;

(2) estimate by k-1 M particle filter constantly, each sampling particle is carried out EKF or UKF obtain k corresponding constantly average and covariance matrix, adopt Gaussian density as the suggestion probability density, with the average of each particle, covariance matrix average, covariance matrix as this distribution, and from each suggestion distributes, adopt n particle respectively, obtain the set that nM particle formed altogether; N and M are natural number;

(3) a described nM particle is carried out weight respectively and upgrade, obtain the weight of each particle;

(4) the effective sample capacity by the weight calculation population, if the effective sample capacity that obtains is less than effective sample capacity lower limit, lower limit may be selected to be half of sample number, carries out niche genetic algorithm and resamples, and described niche genetic algorithm method for resampling performing step is:

(4.1) producing nM the individual population of forming by step (3) is initial population, and the particle weight that calculates is initial fitness;

(4.2) calculate the fitness of progeny population individuality, according to each individual fitness it is carried out descending sort, M individuality before the memory carries out hereditary selection, intersection, mutation operation to the parent population and obtains progeny population;

(4.3) adopt microhabitat restriction competition selection operation, to merge by the progeny population of genetic manipulation and the parent population of memory, in the population after the merging, the penalty function of sentencing that fitness in the adjacent nearer particle is low is accelerated it and eliminated, particle is moved to high fitness direction faster;

(4.4) by after the above operation, according to this (n+1) M individual new fitness individuality is carried out descending sort, M individuality if the population fitness reaches the setting thresholding, then exported result of calculation, otherwise forwarded step (4.2) to before the memory.

2. the improvement particle filter method based on niche genetic algorithm according to claim 1 is characterized in that: in the described step (4.2) parent population genetic selection operation is comprised the steps:

(4.2.1) to the G of parent colony _tSelect computing, at first calculate all individual fitness summations in the colony $\mathrm{\&Sigma;F}=\underset{i=1}{\overset{\mathrm{nM}}{\mathrm{\&Sigma;}}}F\left(x_i\right);$

(4.2.2) calculate each ideal adaptation degree size, promptly each individuality is genetic to the Probability p in the colony of future generation _Si=F (x _i)/∑ F, wherein F (x _i) be each individual fitness, ∑ F is the summation of ideal adaptation degree;

(4.2.3) simulation gambling dish operates to determine whether each individuality is selected, simulates the shared zone that constructs each individuality according to selection rate, is equivalent to the covering of the fan in the wheel disc, and each individuality comprises range of choice bound A _i[min, max] is created on [0,1] space equally distributed decimal at random γ ∈ [0,1] at random, judges which individual choice scope A γ drops in, and be then should individuality selected;

(4.2.4) use optimum conversation strategy, find out fitness is the highest in the current colony individuality and the minimum individuality of fitness; If the fitness of optimized individual is more taller than the fitness of total best individuality up to now in the current colony, then with the optimized individual in the current colony as new best individuality up to now; Usefulness best individuality up to now replaces the poorest individuality in the current colony at last, promptly obtains progeny population.

3. the improvement particle filter method based on niche genetic algorithm according to claim 1 is characterized in that: described step (4.3) microhabitat restriction competition selection operation comprises the steps:

(4.3.1) will combine by M individuality individual and that step (4.2) is remembered, and obtain one and contain (n+1) M individual new colony through nM of obtaining after the hereditary variation operation;

(4.3.2) this (n+1) M individuality obtained per two individual x _iAnd x _jBetween distance

$\left|\right|x_i-x_j\left|\right|=\sqrt{\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{(x_i-x_j)}^2}\left(\begin{array}{c}i=\mathrm{1,2},...,(n+1)M-1\\ j=i+1,...,(n+1)M\end{array}\right)$

(4.3.3) as ‖ x _i-x _jDuring ‖＜L, the fitness size of this two individuality relatively, and the individuality that fitness is little in two individualities sentenced penalty function F (min{x _i, x _j)=Penalty; Wherein L is the habitat radius, and its value is big more, then will be discrete more through the individual distribution of the population after squeezing, and Penalty is a penalty function, x _iAnd x _jBe any two individualities in the colony.

Images