CN107124159A - A kind of implementation method of the particle filter based on adaptive KLD cassette lengths - Google Patents

A kind of implementation method of the particle filter based on adaptive KLD cassette lengths Download PDF

Info

Publication number
CN107124159A
CN107124159A CN201710290178.8A CN201710290178A CN107124159A CN 107124159 A CN107124159 A CN 107124159A CN 201710290178 A CN201710290178 A CN 201710290178A CN 107124159 A CN107124159 A CN 107124159A
Authority
CN
China
Prior art keywords
kld
particle
particles
cassette
max
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201710290178.8A
Other languages
Chinese (zh)
Other versions
CN107124159B (en
Inventor
殷波
祁艳杰
臧睦君
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ludong University
Original Assignee
Ludong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ludong University filed Critical Ludong University
Priority to CN201710290178.8A priority Critical patent/CN107124159B/en
Publication of CN107124159A publication Critical patent/CN107124159A/en
Application granted granted Critical
Publication of CN107124159B publication Critical patent/CN107124159B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • H—ELECTRICITY
    • H03—ELECTRONIC CIRCUITRY
    • H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
    • H03H21/00—Adaptive networks
    • H03H21/0012—Digital adaptive filters
    • H03H21/0043—Adaptive algorithms

Landscapes

  • Feedback Control In General (AREA)

Abstract

The invention discloses a kind of implementation method of the particle filter based on adaptive KLD cassette lengths, this method is realized by series of steps, the present invention proposes alternately resampling and KLD sampling algorithms, avoid the extra amount of calculation of particle filter, improve the execution efficiency of algorithm, in the case where number of particles is very big, the present invention greatly improves the performance of particle filter.The implementation method of the adaptive KLD cassette lengths of the present invention, can be distributed according to the state of each moment particle, automatically adjust the size of KLD cassette lengths.When particle distribution is wider, KLD cassette lengths can adaptively increase, and prevent that the number of particles that KLD samples is excessive, reduce computation complexity；When particle is densely distributed, KLD cassette lengths can be reduced adaptively, the number of particles of increase KLD samplings, improve estimated accuracy.

Description

A kind of implementation method of the particle filter based on adaptive KLD cassette lengths
Technical field
The present invention relates to particle filter field, more particularly to a kind of particle based on adaptive KLD cassette lengths The implementation method of wave filter.
Background technology
The estimated accuracy of particle filter is improved with the increase of number of particles, therefore in order to improve estimated accuracy, grain The substantial amounts of particle of subfilter, and in practice, the high dimension of system mode and complicated posterior density function are with greater need for a large amount of Particle to carry out, state description and probability distribution are approximate, therefore the calculating and renewal of particle and its weights in a large number make The complexity for obtaining particle filter is very high.Too high complexity limits the further development and application of particle filter, especially Be under the higher application environment of some real-times, therefore how to reduce the complexity of particle filter be badly in need of solve pass Key problem.
In order to reduce the complexity of particle filter, current modified hydrothermal process includes edge wave filter strategy and adaptive Particle filter is answered, wherein：Rao-Blackwellization technologies are a kind of edge technologies, and it has reduction system side The characteristic of difference, can improve the performance of particle filter.The general principle of Rao-Blackwellization technologies is：According to being Unite the particular module of state-space model, integrality spatial decomposition is respectively processed into multiple spaces, for example, by bar The state space of the linear Gauss model of part is divided into two parts：A part is linear Gauss's, it is possible to use Kalman filtering To be handled；Another part is non-linear Gauss, it is possible to use particle filter is handled.Kalman filtering processing is linear During Gauss problem, estimated accuracy is high and speed is fast, while the dimension reduction of the non-linear Gaussian portion after decomposing so that grain The computation complexity reduction of son filtering, therefore Rao-Blackwellization technologies can reduce the complexity of particle filter.
Adaptive particle filter algorithm can automatically select number of particles, reduce computation complexity, and its guiding theory is to work as When probability density concentrates on the small range of state space (uncertainty of state distribution is smaller), using a small amount of number of particles, It is on the contrary then using more number of particles.The key technology of adaptive particle filter is to employ Kullback-Leibler Distance (KLD) adaptively sampled method.
In existing particle filter algorithm, what the setting of number of particles was usually fixed, in order to ensure estimation essence Degree, often sets a larger number of particles, however some noise circumstances are preferable or true Posterior probability distribution simultaneously In the case of not sufficiently complex, the estimated accuracy that substantial amounts of particle is improved is limited, and also creates very high calculating Complexity.
In existing KLD particle filters device algorithm, particle resampling is first carried out, KLD samplings are then carried out again.By Can give up some particles in KLD post-samplings, the re-sampling operations for being rejected particle be it is unnecessary, therefore existing KLD particles filter Ripple device device algorithm contains some redundant operations, adds the computation complexity of algorithm.
In existing KLD particle filters device algorithm, the parameter value of KLD cassette lengths is changeless, is fixed Parameter value setting can cause the appearance of some two kinds of situations：When particle is densely distributed, the box number that KLD samplings are obtained is inadequate； Particle is distributed distributing, and the box number that KLD samplings are obtained is excessive.Both of these case can cause respectively estimated accuracy decline and The increased adverse consequences of computation complexity.
The content of the invention
In order to reduce the complexity of particle filter, the present invention proposes a kind of based on the low of adaptive KLD cassette lengths The implementation method of complexity KLD particle filters, number of particles is automatically selected according to particle distribution, when probability density collection In in the small range of state space, using a small amount of number of particles, it is on the contrary then using more number of particles.
To achieve the above object, the present invention is realized according to following technical scheme：
A kind of implementation method of the particle filter based on adaptive KLD cassette lengths, it is characterised in that including as follows Step：
1) input：Based on KLD criterions, from the judgement formula of the adaptively sampled cut-offs of KLD：
Wherein, nxIt is the adaptively sampled number of particles of KLD, z1-δThe expression upper bound can for 1- δ standardized normal distribution Reliability, k is KLD box numbers, and ε is less arithmetic number；KLD box number k and ε, z1-δAnd nxValue it is relevant, KLD boxes Number k and nxε value is approximately directly proportional, wherein, z1-δRepresent confidence level of the upper bound for 1- δ standardized normal distribution；
Input the t-1 moment particle collection beε and δ value is set, number of particles needed for setting is most Small value is nmin, the particle number at t-1 moment is nt-1；
2) initialize：Before KLD samplings are carried out, particle assembly, sampling number of particles, KLD boxes number and power are defined It is worth accumulation amount, makes the prediction particle collection of tnt=0, KLD box number k=0, weights accumulation amount α=0, box Length is b；
3) resampling：To particle collectionResampling is carried out, i-th of particle is produced
4) time updates：According to the probability density function p (x of particlet|xt-1), by the particle at t-1 momentCalculating obtains t The particle at moment
5) measure and update：By the particle of tBring function intoObtaining particle weights is
6) obtained particle is put into prediction particle to concentrate,
7) weights add upIt is continuously increased a value；
8) box number is calculated：Judged using following if functions, work as particleIt is distributed in a new box In the range of, box number increase by 1：
If(Fall into one section of empty cassette length b) then k=k+1
End If；
9) number of particles is updated：nt=nt+ 1, one cycle is often done, increases a number of particles；
10) judge：By the number of particles n at current timetWith the smallest particles number n of settingminIt is compared, works as satisfaction During condition, the number of particles n that KLD samplings need is calculatedx, then judged, current population reaches KLD sampling needs Number of particles when, jump out circulation, terminate KLD samplings：
If(nt＞ nmin)then
End If；
If(nt≥nx)then
Jump out circulation
End If；
11) circulate：Circulation performs step 3- steps 10, and the decision condition until meeting step 10 jumps out circulation；
12) weights are normalized：
In above-mentioned technical proposal, the control of the adaptive KLD cassette lengths is realized according to following steps：
1) set the particle collection of initial time asNmaxFor the maximum population of subsequent sampling；
2) initial KLD cassette lengths b is set0, by the b of initial time0Value is arranged to smaller value so that t=1 moment adopts Like-particles number is N, to obtain large range of particle distributed area, setup parameter ε and δ value；
3) according to the particle collection of tSelect maximum particle xmax(t) with minimum particle xmin (t)；
4) adaptive KLD cassette lengths：
According to formulaObtain the KLD cassette lengths b at t+1 momentadptive(t+1)；Its In, xmax(t) the maximum particle of t, x are representedmin(t) the minimum particle of t is represented, | | xmax(t)-xmin(t) | | represent xmaxAnd x (t)minThe distance between (t)；
5) according to KLD cassette lengths badptive(t+1) KLD samplings are carried out to the particle at t+1 moment.
In above-mentioned technical proposal, the number of particles maximum during particle filter is Nmax, approximately obtain in population In the case that mesh is most, required maximum box number KmaxFor：kmax≈Nmaxε。
The present invention compared with prior art, has the advantages that：
The present invention proposes alternately resampling and KLD sampling algorithms, and resampling process often produces a particle and just carried out KLD samples, and then carries out KLD criterion judgements, if being unsatisfactory for stop condition, carries out resampling and the KLD of next particle Sampling process, if meeting the stop condition of KLD criterions, follow-up particle will be rejected, and will not carry out resampling and KLD Sampling process, it is to avoid particle filter extra amount of calculation, improves the execution efficiency of algorithm, in number of particles very big feelings Under condition, the present invention greatly improves the performance of particle filter.
In addition, the implementation method of the adaptive KLD cassette lengths of the present invention, can be according to the state of each moment particle Distribution, automatically adjusts the size of cassette length.When particle distribution is wider, adaptive KLD cassette lengths increase automatically Greatly；When particle is densely distributed, it can adaptively reduce cassette length, increase sampling number of particles improves estimated accuracy.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will to embodiment or The accompanying drawing used required in description of the prior art is briefly described, it should be apparent that, drawings in the following description are only Some embodiments of the present invention, for those of ordinary skill in the art, on the premise of not paying creative work, also Other accompanying drawings can be obtained according to these accompanying drawings.
Fig. 1 is the flow chart of the implementation method of the particle filter of the present invention；
Fig. 2 is the number of particles contrast schematic diagram of experimental verification of the present invention.
Embodiment
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is A part of embodiment of the present invention, rather than whole embodiments.
The criterion of the adaptively sampled methods of KLD is described as：
Assuming that n particle is to sample to obtain from k the effectively discrete distribution of subspace, vector x={ x is made1, x2..., xkRepresent to sample from every sub-spaces obtained number of particles, vectorial P={ p1, p2..., pkRepresent each effective The true probability of subspace, then be by the maximal possibility estimation on P that n particle is obtained Further obtain P likelihood probability statistical value λnFor：
According to above formula, defining K-L distances is：
When number of particles n tends to be infinite, P likelihood probability statistical value λnCentered difference distribution is converged on, i.e.,：
What K-L distances were represented is the difference between real Posterior distrbutionp and maximal possibility estimation.DefinitionRepresent that K-L distances are less than or equal to ε probability (ε is the arithmetic number of a very little), then have：
The confidence level of centered difference distributionIt can be expressed as：
Select suitable n so thatSo as to obtain：
Due to needing to calculate centered difference distribution function, directly calculating n value is cumbersome, is that this can be by Wilson-Hilferty conversion obtains accurate approximation
Wherein, z1-δRepresent confidence level of the upper bound for 1- δ standardized normal distribution.
The present invention proposes alternately resampling and KLD sampling algorithms, and resampling process often produces a particle and just carried out KLD samples, and then carries out KLD criterion judgements, if being unsatisfactory for stop condition, carries out resampling and the KLD of next particle Sampling process, if meeting the stop condition of KLD criterions, follow-up particle will be rejected, and will not carry out resampling and KLD Sampling process.Compared with existing algorithm, the present invention carries algorithm and avoids the extra amount of calculation of particle filter, improves algorithm Execution efficiency, in the case where number of particles is very big, such improvement has greatly improved to systematic function.
A kind of implementation method of particle filter based on adaptive KLD cassette lengths of the present invention, including following step Suddenly：
1) input：Based on KLD criterions, from the judgement formula of the adaptively sampled cut-offs of KLD：
Wherein, nxIt is the adaptively sampled number of particles of KLD, z1-δThe expression upper bound can for 1- δ standardized normal distribution Reliability, k is KLD box numbers, and ε is less arithmetic number；KLD box number k and ε, z1-δAnd nxValue it is relevant, KLD boxes Number k and nxε value is approximately directly proportional, wherein, z1-δRepresent confidence level of the upper bound for 1- δ standardized normal distribution；
Input the t-1 moment particle collection beε and δ value is set, number of particles needed for setting is most Small value is nmin, the particle number at t-1 moment is nt-1；
2) initialize：Before KLD samplings are carried out, particle assembly, sampling number of particles, KLD boxes number and power are defined It is worth accumulation amount, makes the prediction particle collection of tnt=0, KLD box number k=0, weights accumulation amount α=0, box Length is b；
3) resampling：To particle collectionResampling is carried out, i-th of particle is produced
4) time updates：According to the probability density function p (x of particlet|xt-1), by the particle at t-1 momentCalculating obtains t The particle at moment
5) measure and update：By the particle of tBring function intoObtaining particle weights is
6) obtained particle is put into prediction particle to concentrate,
7) weights add upIt is continuously increased a value；
8) box number is calculated：Judged using following if functions, work as particleIt is distributed in a new box In the range of, box number increase by 1：
Work f (Fall into one section of empty cassette length b) then k=k+1
End If；
9) number of particles is updated：nt=nt+ 1, one cycle is often done, increases a number of particles；
10) judge：By the number of particles n at current timetWith the smallest particles number n of settingminIt is compared, works as satisfaction During condition, the number of particles n that KLD samplings need is calculatedx, then judged, current population reaches KLD sampling needs Number of particles when, jump out circulation, terminate KLD samplings：
If(nt＞ nmin)then
End If；
If(nt≥nx)then
Jump out circulation
End If；
11) circulate：Circulation performs step 3- steps 10, and the decision condition until meeting step 10 jumps out circulation；
12) weights are normalized：
In KLD adaptive particle filter algorithms, the sampling number of particles at each moment is different, in particle point In the case that cloth is more concentrated, the sampling particle needed for KLD is less, in the case where particle distribution is relatively broad, needed for KLD Sampling particle it is more.During actual particle filter, the situation of change of state parameter and the distribution situation ratio of particle It is more complicated.
In the present invention, KLD particle filters realize adaptive KLD boxes using the control method of adaptive KLD cassette lengths Sub- length, is realized according to following steps：
1) set the particle collection of initial time asNmaxFor the maximum population of subsequent sampling；
2) initial KLD cassette lengths b is set0, by the b of initial time0Value is arranged to smaller value so that t=1 moment adopts Like-particles number is N, to obtain large range of particle distributed area, setup parameter ε and δ value；
3) according to the particle collection of tSelect maximum particle xmax(t) with minimum particle xmin (t)；
4) adaptive KLD cassette lengths：
According to formulaObtain the KLD cassette lengths b at t+1 momentadptive(t+1)； Wherein, xmax(t) the maximum particle of t, x are representedmin(t) the minimum particle of t is represented, | | xmax(t)-xmin(t) | | table Show xmaxAnd x (t)minThe distance between (t)；
Adaptive KLD cassette lengths badptiveBasic thought be distribution situation according to current time particle, come adaptive Answer ground that the KLD cassette lengths of subsequent time are set.When | | xmax(t)-xmin(t) | | value it is larger when, illustrate particle now point Cloth scope is relatively broad, and now KLD cassette lengths are adaptively reduced, and sampling number of particles increases；When | | xmax(t)-xmin (t) | | value it is smaller when, illustrate that particle distribution now is more concentrated, now KLD cassette lengths adaptively increase, Number of particles of sampling is reduced.In addition, the adaptive KLD cassette lengths institute maximum box number of the getable limit is kmax, correspondence Maximum sampling population be Nmax, therefore using adaptive KLD cassette lengths can effectively prevent in existing algorithm due to Sampling number of particles continues saturation and makes the situation of KLD sampling failures.
5) according to KLD cassette lengths badptive(t+1) KLD samplings are carried out to the particle at t+1 moment.
Number of particles maximum during particle filter is Nmax, approximately obtain in the case where number of particles is most, Required maximum box number KmaxFor：
kmax≈Nmaxε
The present invention, can be with adjust automatically cassette length in the case where that can not obtain state variable and process noise information Size, can reduce sampling particle number, will not cause number of particles too small again and influence the estimated accuracy of algorithm.
In order to verify the performance of the present invention, of the invention and existing KLD particle filter algorithms are entered using MATLAB softwares Row contrast simulation, simulation result is as shown in Figure 2.From fig. 2 it can be seen that in existing fixed cassette length KLD particle filters In algorithm, as cassette length b=0.5, because cassette length is excessive so that the particle of actual samples only has 100--200；When During cassette length b=0.1, because cassette length is too small, the population at each moment can reach the upper limit 500 so that KLD is certainly Adapt to sampling failure.Adaptive cassette length KLD particle filter algorithms proposed by the present invention, can efficiently solve both Problem：When the setting of initial cassette length is excessive, adaptive cassette length can automatically reduce, to increase sampling population； When the setting of initial cassette length is too small, adaptive cassette length can automatically increase, to reduce sampling population.
In addition, from fig. 2 it can be seen that due to the change of each moment state variable, when sampling KLD is adaptively sampled, The population that each moment needs is also change, and is had many times, and the sampling number of particles of subsequent time is less than previous Moment, therefore alternately resampling and KLD samplings can be effectively prevented from the resampling process of the unwanted particle of subsequent time, The amount of calculation of algorithm is reduced, complexity is reduced.
Although above with general explanation and specific embodiment, the present invention is described in detail, at this On the basis of invention, it can be made some modifications or improvements, this will be apparent to those skilled in the art.Therefore, These modifications or improvements, belong to the scope of protection of present invention without departing from theon the basis of the spirit of the present invention.

Claims (3)

1. a kind of implementation method of the particle filter based on adaptive KLD cassette lengths, it is characterised in that including following step Suddenly：
1) input：Based on KLD criterions, from the judgement formula of the adaptively sampled cut-offs of KLD：
Wherein, nxIt is the adaptively sampled number of particles of KLD, z1-δRepresent confidence level of the upper bound for 1- δ standardized normal distribution, k For KLD box numbers, ε is less arithmetic number；KLD box number k and ε, z1-δAnd nxValue it is relevant, KLD box numbers k and nx ε value is approximately directly proportional, wherein, z1-δRepresent confidence level of the upper bound for 1- δ standardized normal distribution；
Input the t-1 moment particle collection beSet ε and δ value, set required number of particles minimum value as nmin, the particle number at t-1 moment is nt-1；
2) initialize：Before KLD samplings are carried out, particle assembly is defined, sampling number of particles, KLD boxes number and weights are tired Dosage, makes the prediction particle collection of tnt=0, KLD box number k=0, weights accumulation amount α=0, cassette length is b；
3) resampling：To particle collectionResampling is carried out, i-th of particle is produced
4) time updates：According to the probability density function p (x of particlet|xt-1), by the particle at t-1 momentCalculating obtains t Particle
5) measure and update：By the particle of tBring function intoObtaining particle weights is
6) obtained particle is put into prediction particle to concentrate,
7) weights add upIt is continuously increased a value；
8) box number is calculated：Judged using following if functions, work as particleIt is distributed in a new box scope It is interior, box number increase by 1:
If(Fall into one section of empty cassette length b) then k=k+1
End If；
9) number of particles is updated：nt=nt+ 1, one cycle is often done, increases a number of particles；
10) judge：By the number of particles n at current timetWith the smallest particles number n of settingminIt is compared, when meeting condition When, calculate the number of particles n that KLD samplings needx, then judged, current population reaches the particle that KLD samplings need During number, circulation is jumped out, terminates KLD samplings：
If(nt＞ nmin)then
End If；
If(nt≥nx)then
Jump out circulation
End If；
11) circulate：Circulation performs step 3- steps 10, and the decision condition until meeting step 10 jumps out circulation；
12) weights are normalized：
2. a kind of implementation method of particle filter based on adaptive KLD cassette lengths according to claim 1, it is special Levy and be：The control of the adaptive KLD cassette lengths is realized according to following steps：
1) set the particle collection of initial time asNmaxFor the maximum population of subsequent sampling；
2) initial KLD cassette lengths b is set0, by the b of initial time0Value is arranged to smaller value so that the sampling grain at t=1 moment Subnumber is N, to obtain large range of particle distributed area, setup parameter ε and δ value；
3) according to the particle collection of tSelect maximum particle xmax(t) with minimum particle xmin(t)；
4) adaptive KLD cassette lengths：
According to formulaObtain the KLD cassette lengths b at t+1 momentadptive(t+1)；Wherein, xmax(t) the maximum particle of t, x are representedmin(t) the minimum particle of t is represented, | | xmax(t)-xmin(t) | | represent xmax And x (t)minThe distance between (t)；
5) according to KLD cassette lengths badptive(t+1) KLD samplings are carried out to the particle at t+1 moment.
3. a kind of implementation method of adaptive KLD cassette lengths according to claim 1, it is characterised in that：Particle filter During number of particles maximum be Nmax, approximately obtain in the case where number of particles is most, required maximum box number Mesh KmaxFor：kmax≈Nmaxε。
CN201710290178.8A 2017-04-27 2017-04-27 Implementation method of particle filter based on self-adaptive KLD box length Expired - Fee Related CN107124159B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710290178.8A CN107124159B (en) 2017-04-27 2017-04-27 Implementation method of particle filter based on self-adaptive KLD box length

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710290178.8A CN107124159B (en) 2017-04-27 2017-04-27 Implementation method of particle filter based on self-adaptive KLD box length

Publications (2)

Publication Number Publication Date
CN107124159A true CN107124159A (en) 2017-09-01
CN107124159B CN107124159B (en) 2020-06-05

Family

ID=59725472

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710290178.8A Expired - Fee Related CN107124159B (en) 2017-04-27 2017-04-27 Implementation method of particle filter based on self-adaptive KLD box length

Country Status (1)

Country Link
CN (1) CN107124159B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108562287A (en) * 2018-01-08 2018-09-21 哈尔滨工程大学 A kind of Terrain-aided Underwater Navigation based on adaptively sampled particle filter
CN110334322A (en) * 2019-06-26 2019-10-15 电子科技大学 A kind of population adaptive approach of particle filter

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090238406A1 (en) * 2006-09-29 2009-09-24 Thomson Licensing Dynamic state estimation
CN103684350A (en) * 2013-12-04 2014-03-26 北京理工大学 Particle filter method
CN104616318A (en) * 2015-01-22 2015-05-13 重庆邮电大学 Moving object tracking method in video sequence image
CN106447692A (en) * 2016-09-30 2017-02-22 中国科学院光电技术研究所 Sampling self-adaptive immune genetic particle filter weak target tracking method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090238406A1 (en) * 2006-09-29 2009-09-24 Thomson Licensing Dynamic state estimation
CN103684350A (en) * 2013-12-04 2014-03-26 北京理工大学 Particle filter method
CN104616318A (en) * 2015-01-22 2015-05-13 重庆邮电大学 Moving object tracking method in video sequence image
CN106447692A (en) * 2016-09-30 2017-02-22 中国科学院光电技术研究所 Sampling self-adaptive immune genetic particle filter weak target tracking method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
罗元等: "基于自适应多提议分布粒子滤波的蒙特卡洛定位算法", 《计算机应用》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108562287A (en) * 2018-01-08 2018-09-21 哈尔滨工程大学 A kind of Terrain-aided Underwater Navigation based on adaptively sampled particle filter
CN110334322A (en) * 2019-06-26 2019-10-15 电子科技大学 A kind of population adaptive approach of particle filter

Also Published As

Publication number Publication date
CN107124159B (en) 2020-06-05

Similar Documents

Publication Publication Date Title
Yang et al. Netadapt: Platform-aware neural network adaptation for mobile applications
CN106803071A (en) Object detecting method and device in a kind of image
KR102413028B1 (en) Method and device for pruning convolutional neural network
Elsayed et al. Differential evolution with automatic parameter configuration for solving the CEC2013 competition on real-parameter optimization
CN111226234B (en) Method, apparatus and computer program for creating deep neural networks
CN110942248B (en) Training method and device for transaction wind control network and transaction risk detection method
CN110942483B (en) Function rapid convergence model construction method, device and terminal
Nafar et al. Using modified fuzzy particle swarm optimization algorithm for parameter estimation of surge arresters models
CN112348155A (en) Optimization method and system of fuzzy neural network model
CN111539458A (en) Feature map processing method and device, electronic equipment and storage medium
CN107124159A (en) A kind of implementation method of the particle filter based on adaptive KLD cassette lengths
Gadsden et al. A fuzzy-smooth variable structure filtering strategy: For state and parameter estimation
Pavlenko et al. Criterion of cyber-physical systems sustainability
CN110996365B (en) Heterogeneous network vertical switching algorithm and system based on multi-objective optimization model
Kaur et al. Design of Low Pass FIR Filter Using Artificial NeuralNetwork
CN112884142B (en) Neural network training method, target detection method, device, equipment and storage medium
CN112988275A (en) Task perception-based mobile edge computing multi-user computing unloading method
CN117170980A (en) Early warning method, device, equipment and storage medium for server hardware abnormality
Abdulkadir et al. Analysis of recurrent neural networks for henon simulated time-series forecasting
CN108921207B (en) Hyper-parameter determination method, device and equipment
CN113807541B (en) Fairness repair method, system, equipment and storage medium for decision system
CN109697511B (en) Data reasoning method and device and computer equipment
Yaghini et al. HIOPGA: a new hybrid metaheuristic algorithm to train feedforward neural networks for prediction
US11645519B2 (en) Filtering data in orthogonal directions through a convolutional neural network
Liu et al. S-Cyc: A Learning Rate Schedule for Iterative Pruning of ReLU-based Networks

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
CB03 Change of inventor or designer information
CB03 Change of inventor or designer information

Inventor after: Zang Mujun

Inventor after: Yin Bo

Inventor after: Qi Yanjie

Inventor before: Yin Bo

Inventor before: Qi Yanjie

Inventor before: Zang Mujun

GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20200605