CN108152812A - A kind of improvement AGIMM trackings for adjusting grid spacing - Google Patents

A kind of improvement AGIMM trackings for adjusting grid spacing Download PDF

Info

Publication number
CN108152812A
CN108152812A CN201711330498.8A CN201711330498A CN108152812A CN 108152812 A CN108152812 A CN 108152812A CN 201711330498 A CN201711330498 A CN 201711330498A CN 108152812 A CN108152812 A CN 108152812A
Authority
CN
China
Prior art keywords
radar
motor
target
submodel
turning
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
CN201711330498.8A
Other languages
Chinese (zh)
Other versions
CN108152812B (en
Inventor
曹运合
潘媚媚
吴春林
卢毅
吴文华
龚作豪
王宇
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.)
Xidian University
Xian Cetc Xidian University Radar Technology Collaborative Innovation Research Institute Co Ltd
Original Assignee
Xidian University
Xian Cetc Xidian University Radar Technology Collaborative Innovation Research Institute Co Ltd
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 Xidian University, Xian Cetc Xidian University Radar Technology Collaborative Innovation Research Institute Co Ltd filed Critical Xidian University
Priority to CN201711330498.8A priority Critical patent/CN108152812B/en
Publication of CN108152812A publication Critical patent/CN108152812A/en
Application granted granted Critical
Publication of CN108152812B publication Critical patent/CN108152812B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/66—Radar-tracking systems; Analogous systems
    • G01S13/68—Radar-tracking systems; Analogous systems for angle tracking only

Landscapes

  • Engineering & Computer Science (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention discloses a kind of improvement AGIMM trackings for adjusting grid spacing, thinking is：Determine radar, there are targets, radar in the range of the radar scanning to be scanned target detection；Radar rectangular coordinate system and radar polar coordinate system are established, determines that radar to the measuring value of target, respectively obtains initial state vector of the radar to target under radar rectangular coordinate systemAnd radar is to the initial error covariance matrix P of target0；Interaction models collection and interaction models collection initial value are determined respectively；Determine the Kalman filtering initial value of interaction models collection；Calculate the state vector filtering output value of 3 sampling instant, j-th of motor-driven turning submodelTo the state vector filtering output value of j-th of motor-driven turning submodel of N sampling instantsWith the error covariance filtering output value P of 3 sampling instant, j-th of motor-driven turning submodelj(3 | 3) are to the error covariance filtering output value P of j-th of motor-driven turning submodel of N sampling instantsj(N | N), and it is denoted as a kind of improvement AGIMM tracking results for adjusting grid spacing.

Description

A kind of improvement AGIMM trackings for adjusting grid spacing
Technical field
The invention belongs to maneuvering target tracking field, more particularly to a kind of improvement AGIMM track sides for adjusting grid spacing Method, i.e., a kind of improvement adaptive mesh Interactive Multiple-Model tracking for adjusting grid spacing, suitable for being adjusted by motor-driven differentiation Whole grid spacing and the relatively best angular speed adaptive tracing high speed high maneuvering targets of search are filtered with obtaining tracking well Wave effect.
Background technology
All the time, target following is all militarily priority research areas, and is then to the tracking of high speed high maneuvering targets Difficult point in primary study；High speed high maneuvering targets because its speed it is big, it is motor-driven rapid and it is difficult to predict the characteristics of tracking is filtered Wave modeling has very big difficulty.
For the mobility of target, scholars propose some maneuver modelings, improve maneuvering target to a certain extent Tracking accuracy, but still the tracking of strong maneuvering target can not be well adapted to；And the proposition of Interactive Multiple-Model then solves list The problem of model following bad adaptability, has good robust property；Researcher is also in succession on the basis of Interactive Multiple-Model A series of innovatory algorithm is proposed, main improvement direction is the selection of Models Sets, the calculating of model probability and tracking filter Improvement of algorithm etc..
It is according to adaptive mesh Interactive Multiple-Model (AGIMM) algorithm of interacting multiple algorithm combination graph theory thought at present The characteristics of one of most effective track algorithm, AGIMM algorithms are main is the Turn Models that will have different rates of turn as tracking Models Sets, use for reference the thought adaptive updates rate of turn of graph theory make Models Sets be constantly in dynamic update in, relative to mould The fixed interacting multiple algorithm of type has better adaptive ability；But similarly, there is model nets for AGIMM algorithms Compartment is away from cannot adaptively adjust and the slow-footed problem of Attitude rate estimator.
Invention content
In view of the above-mentioned problems of the prior art, it is an object of the invention to propose a kind of improvement for adjusting grid spacing AGIMM methods, the improvement AGIMM methods of this kind adjustment grid spacing be it is a kind of according to residual information carry out motor-driven differentiation so that from The improvement AGIMM methods of adjustment grid spacing are adapted to, and optimum angle is searched for using the thought of linear search at the non-maneuver moment Speed can not only improve the adaptive ability to the motor-driven moment, and can improve to non-machine to adapt to true angular velocity The tracking accuracy at dynamic moment and the estimating speed of angular velocity.
To reach above-mentioned technical purpose, the present invention is realised by adopting the following technical scheme.
A kind of improvement AGIMM methods for adjusting grid spacing, include the following steps：
Step 1, radar is determined, there are targets, radar in the range of the radar scanning to be scanned target detection；It establishes Radar rectangular coordinate system and radar polar coordinate system determine that radar to the measuring value of target, obtains respectively under radar rectangular coordinate system To radar to the initial state vector of targetAnd radar is to the initial error covariance matrix P of target0；Interaction is determined respectively Models Sets and interaction models collection initial value, interaction models collection include L motor-driven turning submodels
Step 2, it initializes：K ' expression k' sampling instants are enabled, k '=3~N, the initial value of k' is 3；According to radar to target Initial state vectorAnd radar is to the initial error covariance matrix P of target0Determine Kalman's filter of interaction models collection Wave initial value；
Step 3, according to the Kalman filtering value of k'-1 sampling instant interaction models collection, k' j-th of machine of sampling instant is calculated The new breath value V of turn bend modelj(k') and the new of j-th of motor-driven turning submodel of k' sampling instants ceases covariance matrix Sj (k') and the state vector filtering output value of j-th of motor-driven turning submodel of k' sampling instantsWhen being sampled with k' Carve the error covariance filtering output value P of j-th of motor-driven turning submodelj(k'|k')；J=1,2 ..., L；
Step 4, according to the new breath value V of j-th of motor-driven turning submodel of k' sampling instantsj(k'), j-th of k' sampling instants The new breath covariance matrix S of motor-driven turning submodelj(k'), k' sampling instant interaction models collection Probability ps are calculatedk'；
Step 5, according to k' sampling instant interaction models collection Probability psk', it is motor-driven that k' sampling instants the 1st are calculated respectively The turning rate ω of turning submodel1(k'), the turning rate ω of the 2nd motor-driven turning submodel of k' sampling instants2(k') With the turning rate ω of the 3rd motor-driven turning submodel of k' sampling instants3(k')；
Step 6, k' is enabled to add 1, step 3 is repeated to step 5, until obtaining 3 sampling instant, j-th of motor-driven turning submodule The state vector filtering output value of typeState vector to j-th of motor-driven turning submodel of N sampling instants filters output ValueWith the error covariance filtering output value P of 3 sampling instant, j-th of motor-driven turning submodelj(3 | 3) are sampled to N The error covariance filtering output value P of j-th of motor-driven turning submodel of momentj(N | N), and it is denoted as a kind of adjustment grid spacing Improve AGIMM tracking results.
Major advantage embodies both ways the present invention compared with prior art：
(1) it solves the disadvantage that model meshes spacing cannot be adjusted adaptively, model can be allowed according to target maneuver size reality When on-line tuning grid spacing, enhance the adaptability changed to target maneuver, improve the tracking performance to the motor-driven moment.
(2) at the non-maneuver moment, Models Sets is made to be restrained to the direction of best angular speed using the thought of linear search, accelerated The estimation of angular velocity improves the tracking accuracy at non-maneuver moment.
Description of the drawings
Explanation and specific embodiment are described in further detail the present invention below in conjunction with the accompanying drawings.
Fig. 1 is a kind of improvement AGIMM method flow diagrams of adjustment grid spacing of the present invention；
Fig. 2 is the radar rectangular coordinate system and radar polar coordinate system schematic diagram established；
Fig. 3 is that target is motor-driven to turn, and when setting measures noise 1, uses AGIMM algorithms and the distance of the method for the present invention RMSE comparison diagrams；
Fig. 4 is that target is motor-driven to turn, and when setting measures noise 1, uses AGIMM algorithms and the speed of the method for the present invention RMSE comparison diagrams；
Fig. 5 a are that target is motor-driven to turn, and when setting measures noise 1, use the corresponding angular speed variation diagram of AGIMM algorithms；
Fig. 5 b are that target is motor-driven to turn, and when setting measures noise 1, are changed using the corresponding angular speed of the method for the present invention Figure；
Fig. 6 is that target is motor-driven to turn, and when setting measures noise 2, uses AGIMM algorithms and the distance of the method for the present invention RMSE comparison diagrams；
Fig. 7 is that target is motor-driven to turn, and when setting measures noise 2, uses AGIMM algorithms and the speed of the method for the present invention RMSE comparison diagrams；
Fig. 8 is that target is near space vehicle, when setting initial model collection 1, uses AGIMM algorithms and the method for the present invention Distance RMSE comparison diagrams；
Fig. 9 is that target is near space vehicle, when setting initial model collection 1, uses AGIMM algorithms and the method for the present invention Speed RMSE comparison diagrams；
Figure 10 is that target is near space vehicle, when setting initial model collection 1 and initial model collection 2, is calculated using AGIMM The distance RMSE comparison diagrams of method；
Figure 11 is that target is near space vehicle, when setting initial model collection 1 and initial model collection 2, uses the present invention The distance RMSE comparison diagrams of method；
Figure 12 a are that target is near space vehicle, set and are become during initial model collection 2 using the angular speed of AGIMM algorithms Change figure；
Figure 12 b are that target is near space vehicle, set and are become during initial model collection 2 using the angular speed of the method for the present invention Change figure.
Specific embodiment
With reference to Fig. 1, a kind of improvement AGIMM method flow diagrams of adjustment grid spacing of the invention；Wherein described adjustment net Compartment away from improvement AGIMM methods, include the following steps：
Step 1, radar is determined, there are targets, radar in the range of the radar scanning to be scanned target detection；It establishes Radar rectangular coordinate system and radar polar coordinate system determine that radar to the measuring value of target, obtains respectively under radar rectangular coordinate system To radar to the initial state vector of targetAnd radar is to the initial error covariance matrix P of target0；Interaction is determined respectively Models Sets and interaction models collection initial value.
The sub-step of step 1 is：
(1a) determines radar, and there are targets, radar in the range of the radar scanning to be scanned target detection；By target Any point in the track of place is denoted as point target P, then criticizes east as origin o, with radar in center using radar Direction is x-axis, refers to direct north using radar establishes radar rectangular coordinate system as y-axis, according to the z-axis that the right-hand rule determines, In determine z-axis process be：The thumb and forefinger for setting the right hand are respectively directed to x-axis and y-axis, then middle finger is oriented to z-axis.
In center be origin o using radar, the radial distance between point target P and radar as ρ, with point target P Radar polar coordinate system is established relative to the pitch angle ε of radar for θ, with point target P relative to the azimuth of radar, wherein radar is straight Angular coordinate system and radar polar coordinate system schematic diagram are as shown in Figure 2.
There are targets, radar in the range of (1b) radar scanning to be scanned target detection, can be with by radar sensor Measurement information of the target under radar polar coordinate system (ρ-θ-ε) is obtained, ρ represents the radial distance between point target P and radar, θ Represent azimuths of the point target P relative to radar, ε represents pitch angles of the point target P relative to radar.
Then measuring value under radar polar coordinate system is converted to the measuring value under radar rectangular coordinate system (x-y-z), it is false If the radar scanning period is T', a length of t during the radar detection total to target is then equivalent to radar and has carried out the period to target trajectory N times for T' sample, N=t/T'.
Radar obtains measuring value of the target under radar polar coordinate system using self-contained sensor, specifically includes k and adopts Target radial distance measurements measured value ρ (k) that the sample moment, radar obtained under radar polar coordinate system, the k sample moment is in radar polar coordinates Azimuth of target measuring value θ (k) and k sample the moment target of radar acquisition under radar polar coordinate system that the lower radar of system obtains are bowed Elevation angle measuring value ε (k)；And it is straight in radar that measuring value of the target of acquisition under radar polar coordinate system is converted into the k sample moment Radar is to the measuring value of target under angular coordinate system：
Zx(k)=ρ (k) cos ε (k) cos θ (k), Zy(k)=ρ (k) cos ε (k) sin θs (k), Zz(k)=ρ (k) sin ε (k)
Wherein, k=1,2 ..., N, Zx(k) represent the k sample moment under radar rectangular coordinate system radar to target x directions Measuring value, Zy(k) represent the k sample moment under radar rectangular coordinate system radar to the measuring value in target y directions, Zz(k) it represents The k sample moment, radar was to the measuring value in target z directions under radar rectangular coordinate system, and ρ (k) the expression k sample moment is in radar pole The target radial distance measurements measured value that radar obtains under coordinate system, θ (k) represent that k sample moment radar under radar polar coordinate system obtains The azimuth of target measuring value taken, ε (k) represent the target pitch angular amount of k sample moment radar acquisition under radar polar coordinate system Measured value, cos represent cosine function, and sin represents SIN function.
(1c) using 1 sampling instant and 2 sampling instants under radar rectangular coordinate system radar to the position measuring value of target, Calculate initial state vector of the radar to target
Wherein, Zx(2) representing 2 sampling instants, radar is to the measuring value in target x directions under radar rectangular coordinate system, i.e., and 2 The distance in sampling instant target x directions under radar rectangular coordinate system；Zx(1) represent 1 sampling instant in radar rectangular coordinate system Lower radar is to the measuring value in target x directions；Zy(2) represent 2 sampling instants under radar rectangular coordinate system radar to target y directions Measuring value, i.e. 2 sampling instant target y directions under radar rectangular coordinate system distance；Zy(1) represent 1 sampling instant in thunder Radar is to the measuring value in target y directions under up to rectangular coordinate system；Zz(2) 2 sampling instants thunder under radar rectangular coordinate system is represented Up to the distance of the measuring value to target z directions, i.e. 2 sampling instant target z directions under radar rectangular coordinate system；Zz(1) 1 is represented Sampling instant radar under radar rectangular coordinate system represents radar scanning period, subscript T tables to the measuring value in target z directions, T' Show transposition.
(Zx(2)-Zx(1))/T' corresponds to the speed in 2 sampling instant target x directions under radar rectangular coordinate system, (Zy(2)- Zy(1))/T' corresponds to the speed in 2 sampling instant target y directions under radar rectangular coordinate system,
(Zz(2)-Zz(1))/T' corresponds to the speed in 2 sampling instant target z directions under radar rectangular coordinate system.
(1d) is adopted according to the 2 sampling instants target radial distance measurements measured value ρ (2) that radar obtains under radar polar coordinate system, 2 The sample moment under radar polar coordinate system radar obtain azimuth of target measuring value θ (2) and 2 sampling instants in radar polar coordinate system The pitch angle measuring value ε (2) that lower radar obtains calculates the 2 sampling instants measurement of radar to target under radar rectangular coordinate system Noise covariance matrix R (2), expression formula are：
Wherein, r11(2) represent 2 sampling instants under radar rectangular coordinate system radar to the measurement noise covariance square of target 1st row, the 1st column element in battle array R (2), r12(2) the 2 sampling instants measurement of radar to target under radar rectangular coordinate system is represented 1st row, the 2nd column element in noise covariance matrix R (2), r13(2) 2 sampling instants radar under radar rectangular coordinate system is represented To the 1st row, the 3rd column element in the measurement noise covariance matrix R (2) of target, r22(2) represent 2 sampling instants at radar right angle Radar is to the 2nd row, the 2nd column element in the measurement noise covariance matrix R (2) of target, r under coordinate system23(2) when representing 2 sampling Radar is engraved under radar rectangular coordinate system to the 2nd row, the 3rd column element in the measurement noise covariance matrix R (2) of target, r33(2) Represent 2 sampling instants under radar rectangular coordinate system radar to the 3rd row, the 3rd in the measurement noise covariance matrix R (2) of target Column element, the noise transition matrix of A (2) 2 sampling instant radar polar coordinate systems of expression to radar rectangular coordinate system,Represent point mesh The measuring noise square difference value of the radial distance ρ between P and radar is marked,Represent amounts of the point target P relative to the azimuth angle theta of radar Survey noise variance value,Represent measuring noise square difference values of the point target P relative to the pitch angle ε of radar, point target P is swept for radar Any point in the range of retouching in track where target.
(1e) according to 2 sampling instants under radar rectangular coordinate system radar to the measurement noise covariance matrix R of target (2), initial error covariance matrix P of the radar to target is calculated0, expression formula is：
Wherein
Wherein, rij(2) represent 2 sampling instants under radar rectangular coordinate system radar to the measurement noise covariance square of target I-th row, jth column element in battle array R (2), PijRepresent initial error covariance matrix P0In the i-th row, jth column element, i=1,2,3, J=1,2,3.
(1f) sets the different motor-driven turning submodel of L turning rate, wherein motor-driven turning submodel is adaptive The turn model mentioned in grid Interactive Multiple-Model (AGIMM) algorithm, AGIMM algorithms will mainly have different turning speed The turn model of rate uses for reference the turning speed of the thought adaptive updates turn model of graph theory as trace model collection Rate, so that trace model collection is constantly in dynamic update；Each motor-driven turning submodel corresponds to 1 Kalman respectively Kalman filter.
Using L motor-driven turning submodels as interaction models collection, L=3, and determine interaction models collection initial value, the friendship Mutual Models Sets initial value is the turning rate ω of the 1st motor-driven turning submodel of 2 sampling instants1(2), 2 sampling instants the 2nd The turning rate ω of motor-driven turning submodel2(2) and the turning rate ω of the 3rd motor-driven turning submodel of 2 sampling instants3 (2), ω1(2)=- ωmax,ω2(2)=0, ω3(2)=ωmax, ωmaxRepresent preset turning rate jump variation most Big value, ωmaxValue range between -180 ° to 180 °, specific ωmaxValue can be transported according to the target of actual tracking Dynamic characteristic does empirical setting.
Step 2, the state-transition matrix of L motor-driven turning submodels, setting up procedure noise are determined according to turning rate Matrix；Determine the Markov state transition probability between the probability of interaction models collection and a motor-driven turning submodels of L.
The sub-step of step 2 is：
(2a) sets k sample moment interaction models and integrates corresponding turning rate as Mk,
Mk={ ω1(k),ω2(k),ω3(k) }, ω1(k) turning of the 1st motor-driven turning submodel of k sample moment is represented Angular speed, ω2(k) turning rate of the 2nd motor-driven turning submodel of k sample moment, ω are represented3(k) the k sample moment is represented The turning rate of 3rd motor-driven turning submodel.
It sets k sample moment interaction models and integrates and correspond to probability as pk, pk={ μ1(k),μ2(k),μ3(k) }, μ1(k) k is represented The 1st motor-driven model probability of making a turn of sampling instant, μ2(k) the 2nd motor-driven model probability of making a turn of k sample moment, μ are represented3 (k) the 3rd motor-driven model probability of making a turn of k sample moment is represented.
(2b), the Markov model transfer for determining i-th of motor-driven turning submodel to j-th of motor-driven turning submodel are general Rate pij, expression formula is：
Wherein, i=1,2,3, j=1,2,3.
(2c), interaction models collection include 3 motor-driven turning submodels, calculate j-th of motor-driven turning submodel of k sample moment State-transition matrix Fj(k)。
Specifically, the state-transition matrix of each motor-driven turning submodel can be decomposed into water in radar rectangular coordinate system Square to vertical direction two parts：Regard at the uniform velocity turning motion in horizontal direction as, it is all related with the variation of x-axis, y-axis direction； Vertical direction regards uniform motion as, only related with z-axis direction change, then obtains j-th of motor-driven turning submodel of t moment in radar Continuous state equation under rectangular coordinate system is as follows：
Wherein, x (t) represent t moment target x-axis direction distance, y (t) represent t moment target y-axis direction away from From z (t) represents distance of the t moment target in z-axis direction；Represent speed of the t moment target in x-axis direction,Represent t Moment target y-axis direction speed,Represent speed of the t moment target in z-axis direction；Represent t moment target in x The acceleration of axis direction,Represent acceleration of the t moment target in y-axis direction,Represent t moment target in z-axis direction Acceleration；ωj(t) turning rate of j-th of motor-driven turning submodel of t moment is represented,Represent that j-th of t moment is motor-driven It makes a turn the state-transition matrix in model level direction,Represent j-th of motor-driven turning submodel vertical direction of t moment State-transition matrix, t represent time variable.
To the state-transition matrix of the horizontal directionWith the state-transition matrix of the vertical directionPoint It Jin Hang not be the sliding-model control of T ' in the sampling period, respectively obtain j-th of motor-driven turning submodel of k sample moment in the horizontal direction On state-transition matrixSquare is shifted with the state of j-th of motor-driven turning submodel of k sample moment in vertical direction Battle arrayIts expression formula is：
Then the state-transition matrix F of j-th of motor-driven turning submodel of k sample moment is calculatedj(k), expression formula For：
Wherein, j=1,2,3,02×42 × 4 dimension full 0 matrixes of expression, 04×2Represent 4 × 2 dimension full 0 matrixes, ωj(k) represent that k is adopted The turning rate of j-th of motor-driven turning submodel of sample moment, T' represent the radar scanning period,
(2d) calculates the process noise covariance matrix Q of j-th of motor-driven turning submodel of k sample momentj(k), it expresses Formula is：
Wherein,
Wherein, j=1,2,3, qjRepresent the process-noise variance of j-th of motor-driven turning submodel, process-noise variance qjRoot According to noise and motor-driven model attributes setting of making a turn is measured, generally take in σρIt arrivesBetween, σρRepresent point target P and radar it Between radial distance ρ measurement noise criteria it is poor,Represent the measurement noise side of the radial distance ρ between point target P and radar Difference, point target P are any point in track where target in the range of radar scanning；Process-noise variance qjThe bigger table of setting Representation model more adapts to the target under high maneuver or big noise background in the case where other parameter is constant；Conversely, process noise Variance qjIt sets smaller, then it represents that model has the target under weak motor-driven or low noise background in the case where other parameter is constant Better tracking accuracy；Represent the process noise association of j-th of motor-driven turning submodel of k sample moment in the horizontal direction Variance matrix,Represent the process noise covariance square of j-th of motor-driven turning submodel of k sample moment in vertical direction Battle array.
Initialization：K ' expression k' sampling instants are enabled, k '=3~N, the initial value of k' is 3；According to radar to the initial of target State vectorAnd radar is to the initial error covariance matrix P of target0Determine that the Kalman filtering of interaction models collection is initial Value.
Step 3, using Models Sets probability and Model transfer probability to the shape of L motor-driven turning submodel k-1 sampling instants State vector and error co-variance matrix are weighted filtering of the interactive computing as L motor-driven turning submodel k sample moment Then input shifts frame and metrology frame to each motor-driven turning submodule using state of the maneuvering target under rectangular coordinate system Type carries out Kalman's Kalman filter respectively；Because the initial samples moment of Kalman filter is 2 sampling instants, karr Graceful filtering is since 3 sampling instants；The sampling instant for participating in Kalman filter is set as k ', k '=3~N, the initial value of k' It is 3.
The sub-step of step 3 is：
(3a) determines the Kalman filtering value of k'-1 sampling instant interaction models collection, when k' is 3 during k'-1 sampling The Kalman filtering value that integrates of interaction models is carved as the Kalman filtering initial value of interaction models collection, the karr of the interaction models collection Graceful filtering initial value includes the state filtering output valve of the 1st motor-driven turning submodel of 2 sampling instants2 sampling instants The state filtering output valve of 2nd motor-driven turning submodelThe shape of the 3rd motor-driven turning submodel of 2 sampling instants State filtering output valueThe error covariance filtering output value P of the 1st motor-driven turning submodel of 2 sampling instants1(2| 2), the error covariance filtering output value P of the 2nd motor-driven turning submodel of 2 sampling instants2(2 | 2), the 3rd machine of 2 sampling instants The error covariance filtering output value P of turn bend model3(2 | 2), expression formula is respectively：
P1(2 | 2)=P0, P2(2 | 2)=P0, P3(2 | 2)=P0
Wherein,Represent radar to the initial state vector of target, P0Represent initial error covariance of the radar to target Matrix.
(3b) calculates the interaction mode vector input value of k' j-th of motor-driven turning submodel of sampling instantIts expression formula is：
Wherein, μi|j(k'-1 | k'-1) represent i-th of motor-driven turning submodel of k'-1 sampling instants to j-th of motor-driven turning One step state transition probability of submodel,Represent that k'-1 is adopted The state filtering output valve of i-th of motor-driven turning submodel of sample moment, μi(k'-1) motor-driven turn of i-th of k'-1 sampling instants are represented Bend model probability, pijRepresent Markov model transfer of i-th of motor-driven turning submodel to j-th of motor-driven turning submodel Probability.
(3c) calculates the interaction error co-variance matrix input value P of k' j-th of motor-driven turning submodel of sampling instantoj(k'- 1 | k'-1), expression formula is：
Wherein, μi|j(k'-1 | k'-1) represent i-th of motor-driven turning submodel of k'-1 sampling instants to j-th of motor-driven turning One step state transition probability of submodel, pijRepresent i-th of motor-driven turning submodel to the Ma Er of j-th of motor-driven turning submodel Section husband Model transfer probability, Pj(k'-1 | k'-1) represents the error covariance of k'-1 j-th of motor-driven turning submodel of sampling instant Filtering output value,Represent the state filtering output valve of k'-1 j-th of motor-driven turning submodel of sampling instantWith the interaction mode vector input value of j-th of motor-driven turning submodel of k' sampling instants Between difference,Represent the state filtering output valve of k'-1 j-th of motor-driven turning submodel of sampling instant,Represent the interaction mode vector input value of k' j-th of motor-driven turning submodel of sampling instant.
(3d) is by the interaction mode vector input value of j-th of motor-driven turning submodel of k' sampling instants With the interaction error co-variance matrix input value P of j-th of motor-driven turning submodel of k' sampling instantsoj(k'-1 | k'-1) it is adopted as k The Kalman filter input value of j-th of motor-driven turning submodel of sample moment, then carries out Kalman filtering.
K'-1 sampling instants are calculated respectively to the predicted value of the state vector of j-th of motor-driven turning submodel of k' sampling instantsWith k-1 sampling instants to the error co-variance matrix predicted value of j-th of motor-driven turning submodel of k' sampling instants Pj(k'| k'-1), expression formula is respectively：
Pj(k'| k'-1)=Fj(k')Poj(k'-1|k'-1)[Fj(k')]T+Qj(k')
Wherein, Qj(k') the process noise covariance matrix of k' j-th of motor-driven turning submodel of sampling instant, F are representedj (k') state-transition matrix of k' j-th of motor-driven turning submodel of sampling instant is represented.
Then the new breath value V of k' j-th of motor-driven turning submodel of sampling instant is calculated respectivelyj(k'), k' sampling instants jth The new breath covariance matrix S of a motor-driven turning submodelj(k'), the gain square of j-th of motor-driven turning submodel of k' sampling instants Battle array Kj(k'), expression formula is respectively：
Sj(k')=Hj(k')Pj(k'|k'-1)[Hj(k')]T+Rj(k')
Wherein, Zj(k') measuring value of k' j-th of motor-driven turning submodel of sampling instant is represented,
Zj(k')=[Zx(k'),Zy(k'),Zz(k')]T, Zx(k') represent k' sampling instants under radar rectangular coordinate system Radar is to the measuring value in target x directions, Zy(k') represent k' sampling instants under radar rectangular coordinate system radar to target y directions Measuring value, Zz(k') represent k' sampling instants under radar rectangular coordinate system radar to the measuring value in target z directions, Hj(k') Represent the measurement matrix of k' j-th of motor-driven turning submodel of sampling instant,Rj(k') k' is represented J-th of motor-driven turning submodel of sampling instant under radar rectangular coordinate system radar to the measurement noise covariance matrix of target, It can represent as follows：
Wherein, r11(k ') represents k' j-th of motor-driven turning submodel radar pair under radar rectangular coordinate system of sampling instant The measurement noise covariance matrix R of targetj(k') the 1st row, the 1st column element, r in12(k ') represents that j-th of k' sampling instants are motor-driven Turn submodel under radar rectangular coordinate system radar to the measurement noise covariance matrix R of targetj(k') the 1st row, the 2nd row in Element, r13(k ') represent k' j-th of motor-driven turning submodel of sampling instant under radar rectangular coordinate system radar to the amount of target Survey noise covariance matrix Rj(k') the 1st row, the 3rd column element, r in22(k ') represents k' j-th of motor-driven turning submodule of sampling instant Type under radar rectangular coordinate system radar to the measurement noise covariance matrix R of targetj(k') the 2nd row, the 2nd column element, r in23 (k ') represents that k' j-th of motor-driven turning submodel of sampling instant radar under radar rectangular coordinate system assists the measurement noise of target Variance matrix Rj(k') the 2nd row, the 3rd column element, r in33(k ') represents k' j-th of motor-driven turning submodel of sampling instant in radar Radar is to the measurement noise covariance matrix R of target under rectangular coordinate systemj(k') the 3rd row, the 3rd column element in, ρ (k') represent k' The sampling instant target radial distance measurements measured value that radar obtains under radar polar coordinate system, θ (k') represent k' sampling instants in thunder The azimuth of target measuring value that radar obtains under up to polar coordinate system, ε (k') represent k' sampling instants thunder under radar polar coordinate system Up to the target pitch angular amount measured value of acquisition, A (k') represents k' sampling instant radar polar coordinate systems making an uproar to radar rectangular coordinate system Sound transition matrix,Represent the measuring noise square difference value of the radial distance ρ between point target P and radar,Represent point target P Relative to the measuring noise square difference value of the azimuth angle theta of radar,Represent that point target P makes an uproar relative to the measurement of the pitch angle ε of radar Sound variance yields, point target P are any point in track where target in the range of radar scanning.
The state vector filtering output value of j-th of motor-driven turning submodel of k' sampling instants is finally calculatedWith the error covariance filtering output value P of j-th of motor-driven turning submodel of k' sampling instantsj(k'| k'), expression Formula is is respectively：
Pj(k'| k')=Pj(k'|k'-1)-Kj(k')Sj(k')[Kj(k')]T
Step 4, the new breath of three model filterings output and the spy of corresponding new breath covariance Gaussian distributed are utilized Property construction likelihood function；Then utilize the likelihood function of each model k' sampling instants, the model probabilities of k'-1 sampling instants with And the model probability of Model transfer probability updating k' sampling instants；Maximum probability, utilizes model in last preference pattern probability The distance function of the model of maximum probability carries out motor-driven differentiation to target.
The sub-step of step 4 is：
(4a) is according to the new breath value V of j-th of motor-driven turning submodel of k' sampling instantsj(k'), j-th of machine of k' sampling instants The new breath covariance matrix S of turn bend modelj(k'), calculate k' j-th of motor-driven turning submodel of sampling instant apart from letter Number Dj(k'),Then j-th of motor-driven turning submodel of k' sampling instants is calculated seemingly Right function Λj(k'), expression formula is：
Wherein, j=1,2,3；Then the likelihood function Λ of j-th of motor-driven turning submodel of k' sampling instants is utilizedj(k') With i-th of motor-driven model probability μ that makes a turn of k'-1 sampling instantsi(k'-1), k' j-th of motor-driven turning submodule of sampling instant is calculated Type probability μj(k'), expression formula is：
And then obtain k' sampling instant interaction models collection Probability psk', pk'={ μ1(k'),μ2(k'),μ3(k') }, μ1(k') table Show the 1st motor-driven model probability of making a turn of k' sampling instants, μ2(k') represent that k' the 2nd motor-driven turning submodel of sampling instant is general Rate, μ3(k') k' the 3rd motor-driven model probability of making a turn of sampling instant is represented.
(4b) finds out k' sampling instant interaction models collection Probability psk'In maximum probability max (pk'),
max(pk')=max { μ1(k'),μ2(k'),μ3(k')}。
The corresponding motor-driven turning submodel of (4c) selection maximum probability is denoted as k' sampling instants jth ' a motor-driven turning submodule Type utilizes the distance function D of k' sampling instants jth ' a motor-driven turning submodelj'(k') motor-driven differentiation is carried out to target；In view of Dj'(k') χ for measuring that dimension is m is obeyed2Distribution, it is assumed that it is 0.01 that strong motor-driven probability, which occurs, for target, then inquires χ2Distribution table can It is 6.637 to obtain thresholding, and it is M=7 to set motor-driven judgement threshold.
Work as Dj'(k')>During M, judgement target is motor-driven in k' sampling instants；Otherwise, it is determined that target is in the non-machine of k' sampling instants It is dynamic.
Step 5, when judgement target is when k' sampling instants are motor-driven, the square root of the difference of distance function and motor-driven thresholding is calculated Adjustment multiple as minimum grid；When judging target when k' sampling instants are non-maneuver, using model probability search non- The angular speed of realistic model is best suited in Models Sets in the motor-driven period.
The sub-step of step 5 is：
(5a) sets G as grid spacing, G0For the minimum grid spacing of setting, between generally taking 0.1 °~1 °, Ke Yigen According to the mobility size value for the target to be tracked, mobility is bigger to be taken greatly, otherwise is taken smaller；Grid spacing G can basis Target does the adjustment of adaptive scaling in the mobility of k' sampling instants, target k' sampling instants it is motor-driven big when, grid spacing G It can become larger；Otherwise become smaller.
(5b) calculates k' sampling instant interaction models collection turning rates Mk', Mk'={ ω1(k'),ω2(k'),ω3 (k') }, sub-step is：
(5b.1) is if setting max (pk')=μ2(k'), μ2(k') represent that k' the 2nd motor-driven turning submodel of sampling instant is general Rate；If D2(k')>M, then
ω2(k')=ω2(k'-1),ω1(k')=ω2(k')-G,ω3(k')=ω2(k')+G；Conversely, k' is sampled Moment interaction models collection carries out optimizing probabilistic approximation, obtains：
Wherein, D2(k') distance function of k' the 2nd motor-driven turning submodel of sampling instant is represented, G represents grid spacing, ω2(k') turning rate of k' the 2nd motor-driven turning submodel of sampling instant, ω are represented2(k'-1) k'-1 sampling instants are represented The turning rate of 2nd motor-driven turning submodel, ω1(k') turning for k' the 1st motor-driven turning submodel of sampling instant is represented Bent angle speed, ω1(k'-1) turning rate of k'-1 the 1st motor-driven turning submodel of sampling instant, ω are represented3(k') it represents The turning rate of the 3rd motor-driven turning submodel of k' sampling instants, ω3(k'-1) represent that k'-1 sampling instants the 3rd are motor-driven The turning rate of turning submodel.
(5b.2) setting max (pk')=μ1(k'), μ1(k') represent that k' the 1st motor-driven turning submodel of sampling instant is general Rate；If D1(k')>M, then
ω2(k')=ω1(k'-1),ω1(k')=ω2(k')-G,ω3(k')=ω2(k')+G；Conversely, k' is sampled Moment interaction models collection carries out optimizing probabilistic approximation, obtains：
Wherein, D1(k') distance function of k' the 1st motor-driven turning submodel of sampling instant is represented, G represents grid spacing, ω1(k') turning rate of k' the 1st motor-driven turning submodel of sampling instant, ω are represented1(k'-1) k'-1 sampling instants are represented The turning rate of 1st motor-driven turning submodel, ω2(k') turning for k' the 2nd motor-driven turning submodel of sampling instant is represented Bent angle speed, ω2(k'-1) turning rate of k'-1 the 2nd motor-driven turning submodel of sampling instant, ω are represented3(k') it represents The turning rate of the 3rd motor-driven turning submodel of k' sampling instants, ω3(k'-1) represent that k'-1 sampling instants the 3rd are motor-driven The turning rate of turning submodel, μ2(k') k' the 2nd motor-driven model probability of making a turn of sampling instant, μ are represented3(k') it represents The 3rd motor-driven model probability of making a turn of k' sampling instants.
(5b.3) setting max (pk')=μ3(k '), μ3(k ') represents that k ' the 3rd motor-driven turning submodel of sampling instant is general Rate；If D3(k ') ＞ M, then
ω2(k ')=ω3(k ' -1), ω1(k ')=ω2(k ')-G, ω3(k ')=ω2(k′)+G；Conversely, to k ' samplings Moment interaction models collection carries out optimizing probabilistic approximation, obtains：
Wherein, D3(k ') represents the distance function of k ' the 3rd motor-driven turning submodel of sampling instant, and G represents grid spacing, ω2(k ') represents the turning rate of k ' the 2nd motor-driven turning submodel of sampling instant, ω3(k ' -1) represents -1 sampling instants of k ' The turning rate of 3rd motor-driven turning submodel, ω3(k ') represents turning for k ' the 3rd motor-driven turning submodel of sampling instant Bent angle speed, ω1(k ') represents the turning rate of k ' the 1st motor-driven turning submodel of sampling instant, ω1(k ' -1) expression k ' - The turning rate of the 1st motor-driven turning submodel of 1 sampling instant, ω2(k ' -1) represents the 2nd motor-driven turn of -1 sampling instants of k ' The turning rate of bend model, μ2(k ') represents k ' the 2nd motor-driven model probability of making a turn of sampling instant, μ2(k ') represents k ' The 2nd motor-driven model probability of making a turn of sampling instant.
Step 6, k ' plus 1 is enabled, and by k ' sampling instant interaction models collection Probability psk′Turn with k ' sampling instant interaction models collection Bent angle speed Mk′As new interaction models collection probability set and interaction models collection turning rate, step 3 is repeated to step 5, until obtaining the state vector filtering output value of 3 sampling instant, j-th of motor-driven turning submodelTo N sampling instants The state vector filtering output value of j-th of motor-driven turning submodelWith 3 sampling instant, j-th of motor-driven turning submodule The error covariance filtering output value P of typejThe error covariance of (3 | 3) to j-th of motor-driven turning submodel of N sampling instants filters Output valve Pj(N | N), and it is denoted as a kind of improvement AGIMM tracking results for adjusting grid spacing.
Further verification explanation makees effect of the present invention by following emulation.
(1) simulated environment 1：Target is set as turning maneuver modeling, and the original state of target isA length of 350s during emulation, every 70s, target changes an angle Speed, angular speed, which specifically changes, is set as w=[0.5 °, -10 °, 0.7 °, 6 °, -0.7 °], sampling time interval T=1s.Setting Measure noise 1：Criterion distance difference is 50m, and azimuth and pitching mean angular deviation are 0.1 °.
Initial model collection is set as W1={ -10 °, 0.1 °, 10 ° }, i.e. ωmax=10 °, minimum grid spacing is set as G0= 0.5°。
(2) simulated environment 2：Change on the basis of simulated environment 1 and measure noise, setting measures noise 2：Criterion distance Difference is 100m, and azimuth and pitching mean angular deviation are 0.2 °.
(3) simulated environment 3：Target is disposed proximate to space hypersonic vehicle, and target original state isTarget does uniformly accelerated motion always in z-axis.
T=0~20s：Target does linear uniform motion in x-axis direction, and y directional accelerations are 1g；T=21~380s：Mesh Mark does at the uniform velocity turning motion along x-axis, and the angular speed at the uniform velocity turned is 0.098rad/s.T=81~400s：Target does even accelerate directly Line moves, and y directional accelerations are -1g；It measures noise and is set as noise 1, initial model collection is set as W1, minimum grid spacing sets It is set to G0=0.5 °.
(4) simulated environment 4：Change the setting of initial model collection on the basis of simulated environment 3, initial model collection is set For W2={ -20 °, 0.1 °, 20 ° }, i.e. ωmax=20 °.
(5) analysis of simulation result：
5.1,1 interpretation of result of simulated environment:Can be seen that in the case of noise 1 with reference to Fig. 3 and Fig. 4, i.e., noise compared with In the case of small, although improving, AGIMM algorithms are more much better than AGIMM algorithm, and advantage is not apparent.But it compares The angle of turn of Fig. 5 a and Fig. 5 b combinations simulated environment setting, it can be seen that two kinds of algorithms are in target maneuver moment 71s and 211s When, variation is also ensued in the estimation of angular velocity, but improves AGIMM algorithms relative to AGIMM algorithms to target angular velocity Estimation it is relatively more accurate.
5.2,2 interpretation of result of simulated environment：Observation Fig. 6 and Fig. 7 can be seen that measures noise in increase, i.e., noise is set When being set to noise 2, AGIMM algorithms substantially reduce the tracking performance of target, and improved AGIMM algorithms still can maintain performance Stabilization, illustrate that the interference free performance of innovatory algorithm is better.,
5.3,3 interpretation of result of simulated environment：Observation Fig. 8 and Fig. 9 can be seen that either distance RMSE comparison diagrams, still Speed RMSE comparison diagrams, innovatory algorithm will be more preferable to the tracking effect of near space hypersonic aircraft.
5.4,4 interpretation of result of simulated environment：Setting initial model collection W respectively1And W2, Figure 10 and Figure 11 is observed, can be seen Go out, improve AGIMM algorithms is influenced very little by initial model collection, and AGIMM algorithms are to initial model collection setting requirements higher, when The turning rate that initial model collection and the target movement of setting may occur, which deviates the too big tracking filter that can then seriously affect, imitates Fruit；Figure 12 a and Figure 12 b are observed, even if it can also be seen that improving AGIMM algorithms in the very big feelings of initial model collection setting range Still the angular speed found close to target actual motion can be restrained under condition rapidly.
In conclusion emulation experiment demonstrates the correctness of the present invention, validity and reliability.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art God and range；In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies Within, then the present invention is also intended to include these modifications and variations.

Claims (8)

  1. A kind of 1. improvement AGIMM methods for adjusting grid spacing, which is characterized in that include the following steps：
    Step 1, radar is determined, there are targets, radar in the range of the radar scanning to be scanned target detection；Establish radar Rectangular coordinate system and radar polar coordinate system determine that radar to the measuring value of target, respectively obtains thunder under radar rectangular coordinate system Up to the initial state vector to targetAnd radar is to the initial error covariance matrix P of target0；Interaction models are determined respectively Collection and interaction models collection initial value, interaction models collection include L motor-driven turning submodels；
    Step 2, it initializes：K ' expression k' sampling instants are enabled, k '=3~N, the initial value of k' is 3；According to radar to the first of target Beginning state vectorAnd radar is to the initial error covariance matrix P of target0At the beginning of the Kalman filtering for determining interaction models collection Initial value；
    Step 3, according to the Kalman filtering value of k'-1 sampling instant interaction models collection, motor-driven turn of j-th of k' sampling instants are calculated The new breath value V of bend modelj(k') and the new of j-th of motor-driven turning submodel of k' sampling instants ceases covariance matrix Sj(k'), with And the state vector filtering output value of j-th of motor-driven turning submodel of k' sampling instantsWith j-th of k' sampling instants The error covariance filtering output value P of motor-driven turning submodelj(k'|k')；J=1,2 ..., L；
    Step 4, according to the new breath value V of j-th of motor-driven turning submodel of k' sampling instantsj(k'), j-th of k' sampling instants are motor-driven The new of submodel of turning ceases covariance matrix Sj(k'), k' sampling instant interaction models collection Probability ps are calculatedk'；
    Step 5, according to k' sampling instant interaction models collection Probability psk', the 1st motor-driven turning of k' sampling instants is calculated respectively The turning rate ω of submodel1(k'), the turning rate ω of the 2nd motor-driven turning submodel of k' sampling instants2(k') and k' The turning rate ω of the 3rd motor-driven turning submodel of sampling instant3(k')；
    Step 6, k' is enabled to add 1, step 3 is repeated to step 5, until obtaining 3 sampling instant, j-th of motor-driven turning submodel State vector filtering output valueTo the state vector filtering output value of j-th of motor-driven turning submodel of N sampling instantsWith the error covariance filtering output value P of 3 sampling instant, j-th of motor-driven turning submodelj(3 | 3) to N sample when Carve the error covariance filtering output value P of j-th of motor-driven turning submodelj(N | N), and it is denoted as a kind of changing for adjustment grid spacing Into AGIMM tracking results.
  2. 2. a kind of improvement AGIMM methods for adjusting grid spacing as described in claim 1, which is characterized in that in step 1, Described to establish radar rectangular coordinate system and radar polar coordinate system, process is respectively：
    Any point in track where target is denoted as point target P, then using radar center as origin o, with thunder It is x-axis, refers to direct north using radar and establish radar right angle as y-axis, according to the z-axis that the right-hand rule determines up to signified due east direction Coordinate system, wherein determining that z-axis process is：The thumb and forefinger for setting the right hand are respectively directed to x-axis and y-axis, then middle finger is oriented to z Axis；
    In center be origin o using radar, the radial distance between point target P and radar as ρ, it is opposite with point target P In the azimuth of radar radar polar coordinate system is established relative to the pitch angle ε of radar for θ, with point target P；
    The radar under radar rectangular coordinate system is to the measuring value of target, and the specially k sample moment is in radar rectangular coordinate system Lower radar is to the measuring value of target：
    Zx(k)=ρ (k) cos ε (k) cos θ (k), Zy(k)=ρ (k) cos ε (k) sin θs (k), Zz(k)=ρ (k) sin ε (k)
    Wherein, k=1,2 ..., N, Zx(k) measurement of radar to target x directions under radar rectangular coordinate system of k sample moment is represented Value, Zy(k) represent the k sample moment under radar rectangular coordinate system radar to the measuring value in target y directions, Zz(k) k sample is represented Moment, radar was to the measuring value in target z directions under radar rectangular coordinate system, and ρ (k) the expression k sample moment is in radar polar coordinate system The target radial distance measurements measured value that lower radar obtains, θ (k) represent the mesh of k sample moment radar acquisition under radar polar coordinate system Azimuth measuring value is marked, ε (k) represents the target pitch angular amount measured value of k sample moment radar acquisition under radar polar coordinate system, Cos represents cosine function, and sin represents SIN function.
  3. 3. a kind of improvement AGIMM methods for adjusting grid spacing as claimed in claim 2, which is characterized in that in step 1, The radar is to the initial state vector of targetAnd radar is to the initial error covariance matrix P of target0, obtain process For：
    (1a) acquisition k sample moment is under radar polar coordinate system when the target radial distance measurements measured value ρ (k) of radar acquisition, k sample It is engraved in the azimuth of target measuring value θ (k) and k sample moment thunder under radar polar coordinate system that radar obtains under radar polar coordinate system Up to the target pitch angular amount measured value ε (k) of acquisition, and obtain the k sample moment under radar rectangular coordinate system radar to the amount of target Measured value：
    Zx(k)=ρ (k) cos ε (k) cos θ (k), Zy(k)=ρ (k) cos ε (k) sin θs (k), Zz(k)=ρ (k) sin ε (k)
    Wherein, k=1,2 ..., N, Zx(k) measurement of radar to target x directions under radar rectangular coordinate system of k sample moment is represented Value, Zy(k) represent the k sample moment under radar rectangular coordinate system radar to the measuring value in target y directions, Zz(k) k sample is represented Moment, radar was to the measuring value in target z directions under radar rectangular coordinate system, and ρ (k) the expression k sample moment is in radar polar coordinate system The target radial distance measurements measured value that lower radar obtains, θ (k) represent the mesh of k sample moment radar acquisition under radar polar coordinate system Azimuth measuring value is marked, ε (k) represents the target pitch angular amount measured value of k sample moment radar acquisition under radar polar coordinate system, Cos represents cosine function, and sin represents SIN function；
    (1b) calculates initial state vector of the radar to target
    Wherein, Zx(2) representing 2 sampling instants, radar is to the measuring value in target x directions under radar rectangular coordinate system, i.e., during 2 sampling Carve the distance in target x directions under radar rectangular coordinate system；Zx(1) 1 sampling instant radar under radar rectangular coordinate system is represented To the measuring value in target x directions；Zy(2) the 2 sampling instants measurement of radar to target y directions under radar rectangular coordinate system is represented Value, the i.e. distance in 2 sampling instant target y directions under radar rectangular coordinate system；Zy(1) represent 1 sampling instant at radar right angle Radar is to the measuring value in target y directions under coordinate system；Zz(2) represent 2 sampling instants under radar rectangular coordinate system radar to mesh Mark the distance of the measuring value, i.e. 2 sampling instant target z directions under radar rectangular coordinate system in z directions；Zz(1) when representing 1 sampling The measuring value of radar under radar rectangular coordinate system to target z directions is engraved in, T' represents the radar scanning period, and subscript T represents transposition；
    (1c) calculate 2 sampling instants under radar rectangular coordinate system radar to the measurement noise covariance matrix R (2) of target, Expression formula is：
    Wherein, r11(2) represent 2 sampling instants under radar rectangular coordinate system radar to the measurement noise covariance matrix R of target (2) the 1st row, the 1st column element, r in12(2) represent 2 sampling instants under radar rectangular coordinate system radar to the measurement noise of target 1st row, the 2nd column element in covariance matrix R (2), r13(2) represent 2 sampling instants under radar rectangular coordinate system radar to mesh Target measures the 1st row, the 3rd column element in noise covariance matrix R (2), r22(2) represent 2 sampling instants in radar rectangular co-ordinate The lower radar of system is to the 2nd row, the 2nd column element in the measurement noise covariance matrix R (2) of target, r23(2) represent that 2 sampling instants exist Radar is to the 2nd row, the 3rd column element in the measurement noise covariance matrix R (2) of target, r under radar rectangular coordinate system33(2) it represents 2 sampling instants radar under radar rectangular coordinate system is first to the 3rd row, the 3rd row in the measurement noise covariance matrix R (2) of target Element, the noise transition matrix of A (2) 2 sampling instant radar polar coordinate systems of expression to radar rectangular coordinate system,Represent point target P The measuring noise square difference value of radial distance ρ between radar,Represent measurements of the point target P relative to the azimuth angle theta of radar Noise variance value,Represent measuring noise square difference values of the point target P relative to the pitch angle ε of radar, point target P is radar scanning In the range of any point in track where target；
    (1d) calculates initial error covariance matrix P of the radar to target0, expression formula is：
    Wherein
    Wherein, rij(2) represent 2 sampling instants under radar rectangular coordinate system radar to the measurement noise covariance matrix R of target (2) the i-th row, jth column element, P inijRepresent initial error covariance matrix P0In the i-th row, jth column element, i=1,2,3, j= 1,2,3；
    Wherein, Zx(2) represent 2 sampling instants under radar rectangular coordinate system radar to the measuring value in target x directions, Zx(1) it represents 1 sampling instant under radar rectangular coordinate system radar to the measuring value in target x directions；Zy(2) represent that 2 sampling instants are straight in radar Radar is to the measuring value in target y directions, Z under angular coordinate systemy(1) 1 sampling instant radar pair under radar rectangular coordinate system is represented The measuring value in target y directions；Zz(2) the 2 sampling instants measurement of radar to target z directions under radar rectangular coordinate system is represented Value, Zz(1) represent that 1 sampling instant radar under radar rectangular coordinate system represents radar scanning to the measuring value in target z directions, T' Period, subscript T represent transposition.
  4. 4. a kind of improvement AGIMM methods for adjusting grid spacing as claimed in claim 3, which is characterized in that in step 1, The interaction models collection and interaction models collection initial value, determination process are：
    The motor-driven turning submodel for setting L turning rate different, wherein motor-driven turning submodel is interacted for adaptive mesh The turn model mentioned in Multiple Models Algorithm, using L motor-driven turning submodels as interaction models collection, L=3, and determine Interaction models collection initial value, turning angle speed of the interaction models collection initial value for the 1st motor-driven turning submodel of 2 sampling instants Spend ω1(2), the turning rate ω of the 2nd motor-driven turning submodel of 2 sampling instants2(2) and the 3rd motor-driven turn of 2 sampling instants The turning rate ω of bend model3(2), ω1(2)=- ωmax,ω2(2)=0, ω3(2)=ωmax, ωmaxRepresent preset The maximum value of turning rate jump variation.
  5. 5. a kind of improvement AGIMM methods for adjusting grid spacing as claimed in claim 4, which is characterized in that in step 2, The Kalman filtering initial value of the interaction models collection, specifically includes：The state of the 1st motor-driven turning submodel of 2 sampling instants Filtering output valueThe state filtering output valve of the 2nd motor-driven turning submodel of 2 sampling instants2 samplings The state filtering output valve of the 3rd motor-driven turning submodel of momentThe 1st motor-driven turning submodel of 2 sampling instants Error covariance filtering output value P1The error covariance filtering output of the 2nd (2 | 2), 2 sampling instants motor-driven turning submodel Value P2The error covariance filtering output value P of the 3rd (2 | 2), 2 sampling instants motor-driven turning submodel3(2 | 2), expression formula point It is not：
    P1(2 | 2)=P0, P2(2 | 2)=P0, P3(2 | 2)=P0
    Wherein,Represent radar to the initial state vector of target, P0Represent initial error covariance matrix of the radar to target.
  6. 6. a kind of improvement AGIMM methods for adjusting grid spacing as claimed in claim 5, which is characterized in that in step 3, The new breath value V of described j-th of motor-driven turning submodel of k' sampling instantsj(k') and j-th of motor-driven turning submodule of k' sampling instants The new breath covariance matrix S of typej(k') and the filtering of the state vector of j-th of motor-driven turning submodel of k' sampling instants exports ValueWith the error covariance filtering output value P of j-th of motor-driven turning submodel of k' sampling instantsj(k'| k'), table It is respectively up to formula：
    Sj(k')=Hj(k')Pj(k'|k'-1)[Hj(k')]T+Rj(k')
    Pj(k'| k')=Pj(k'|k'-1)-Kj(k')Sj(k')[Kj(k')]T
    Wherein, Zj(k') measuring value of k' j-th of motor-driven turning submodel of sampling instant, H are representedj(k') k' sampling instants are represented The measurement matrix of j-th of motor-driven turning submodel,Represent k'-1 sampling instants to j-th of machine of k' sampling instants The predicted value of the state vector of turn bend model, Pj(k'| k'-1) represents k-1 sampling instants to j-th of machine of k' sampling instants The error co-variance matrix predicted value of turn bend model, Rj(k') represent that k' j-th of motor-driven turning submodel of sampling instant exists Radar is to the measurement noise covariance matrix of target, K under radar rectangular coordinate systemj(k') represent that j-th of k' sampling instants are motor-driven The gain matrix of turning submodel.
  7. 7. a kind of improvement AGIMM methods for adjusting grid spacing as claimed in claim 6, which is characterized in that in step 4, The k' sampling instants interaction models collection Probability pk', the process of obtaining is：
    (4a) is according to the new breath value V of j-th of motor-driven turning submodel of k' sampling instantsj(k'), motor-driven turn of j-th of k' sampling instants The new breath covariance matrix S of bend modelj(k'), the distance function D of k' j-th of motor-driven turning submodel of sampling instant is calculatedj (k'),Then the likelihood letter of k' j-th of motor-driven turning submodel of sampling instant is calculated Number Λj(k'), expression formula is：
    Wherein, j=1,2,3；Then the likelihood function Λ of j-th of motor-driven turning submodel of k' sampling instants is utilizedj(k') and k'-1 I-th of motor-driven model probability μ that makes a turn of sampling instanti(k'-1), k' j-th of motor-driven model probability of making a turn of sampling instant is calculated μj(k'), expression formula is：
    And then obtain k' sampling instant interaction models collection Probability psk', pk'={ μ1(k'),μ2(k'),μ3(k') }, μ1(k') k' is represented The 1st motor-driven model probability of making a turn of sampling instant, μ2(k') k' the 2nd motor-driven model probability of making a turn of sampling instant, μ are represented3 (k') k' the 3rd motor-driven model probability of making a turn of sampling instant is represented.
  8. A kind of 8. improvement AGIMM methods for adjusting grid spacing as claimed in claim 7, which is characterized in that the sub-step of step 5 Suddenly it is：
    (5a) sets G as grid spacing, G0Minimum grid spacing for setting；
    (5b) calculates k' sampling instant interaction models collection turning rates Mk',
    Mk'={ ω1(k'),ω2(k'),ω3(k') }, sub-step is：
    (5b.1) is if setting max (pk')=μ2(k'), μ2(k') k' the 2nd motor-driven model probability of making a turn of sampling instant is represented；Such as Fruit D2(k')>M, then
    ω2(k')=ω2(k'-1),ω1(k')=ω2(k')-G,ω3(k')=ω2(k')+G；Conversely, to k' sampling instants Interaction models collection carries out optimizing probabilistic approximation, obtains：
    Wherein, D2(k') distance function of k' the 2nd motor-driven turning submodel of sampling instant is represented, G represents grid spacing, ω2 (k') turning rate of k' the 2nd motor-driven turning submodel of sampling instant, ω are represented2(k'-1) k'-1 sampling instants the are represented The turning rate of 2 motor-driven turning submodels, ω1(k') turning of k' the 1st motor-driven turning submodel of sampling instant is represented Angular speed, ω1(k'-1) turning rate of k'-1 the 1st motor-driven turning submodel of sampling instant, ω are represented3(k') k' is represented The turning rate of the 3rd motor-driven turning submodel of sampling instant, ω3(k'-1) the 3rd motor-driven turn of k'-1 sampling instants are represented The turning rate of bend model；
    (5b.2) setting max (pk')=μ1(k'), μ1(k') k' the 1st motor-driven model probability of making a turn of sampling instant is represented；If D1(k')>M, then
    ω2(k')=ω1(k'-1),ω1(k')=ω2(k')-G,ω3(k')=ω2(k')+G；Conversely, to k' sampling instants Interaction models collection carries out optimizing probabilistic approximation, obtains：
    Wherein, D1(k') distance function of k' the 1st motor-driven turning submodel of sampling instant is represented, G represents grid spacing, ω1 (k'-1) turning rate of k'-1 the 1st motor-driven turning submodel of sampling instant is represented,
    ω2(k'-1) turning rate of k'-1 the 2nd motor-driven turning submodel of sampling instant, ω are represented3(k'-1) k'-1 is represented The turning rate of the 3rd motor-driven turning submodel of sampling instant, μ2(k') k' the 2nd motor-driven turning submodule of sampling instant is represented Type probability, μ3(k') k' the 3rd motor-driven model probability of making a turn of sampling instant is represented；
    (5b.3) setting max (pk')=μ3(k'), μ3(k') k' the 3rd motor-driven model probability of making a turn of sampling instant is represented；If D3(k')>M, then
    ω2(k')=ω3(k'-1),ω1(k')=ω2(k')-G,ω3(k')=ω2(k')+G；Conversely, to k' sampling instants Interaction models collection carries out optimizing probabilistic approximation, obtains：
    Wherein, D3(k') distance function of k' the 3rd motor-driven turning submodel of sampling instant, ω are represented3(k'-1) represent that k'-1 is adopted The turning rate of the 3rd motor-driven turning submodel of sample moment, ω1(k'-1) k'-1 the 1st motor-driven turning of sampling instant is represented The turning rate of submodel, ω2(k'-1) turning rate of k'-1 the 2nd motor-driven turning submodel of sampling instant, μ are represented2 (k') k' the 2nd motor-driven model probability of making a turn of sampling instant, μ are represented2(k') it represents that k' sampling instants the 2nd are motor-driven to make a turn Model probability.
CN201711330498.8A 2017-12-13 2017-12-13 Improved AGIMM tracking method for adjusting grid spacing Active CN108152812B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711330498.8A CN108152812B (en) 2017-12-13 2017-12-13 Improved AGIMM tracking method for adjusting grid spacing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711330498.8A CN108152812B (en) 2017-12-13 2017-12-13 Improved AGIMM tracking method for adjusting grid spacing

Publications (2)

Publication Number Publication Date
CN108152812A true CN108152812A (en) 2018-06-12
CN108152812B CN108152812B (en) 2021-06-29

Family

ID=62467188

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711330498.8A Active CN108152812B (en) 2017-12-13 2017-12-13 Improved AGIMM tracking method for adjusting grid spacing

Country Status (1)

Country Link
CN (1) CN108152812B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110716196A (en) * 2019-11-04 2020-01-21 广东中科四创科技有限公司 Multi-point low-slow small aerial target tracking sighting system
CN111562737A (en) * 2019-12-30 2020-08-21 中国人民解放军国防科技大学 Improved AGIMM tracking method controlled by fuzzy logic
CN111650577A (en) * 2020-06-12 2020-09-11 电子科技大学 Maneuvering target tracking method containing Doppler measurement under polar coordinate system
CN111736144A (en) * 2020-07-06 2020-10-02 哈尔滨工业大学 Maneuvering turning target state estimation method only using distance observation

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103853908A (en) * 2012-12-04 2014-06-11 中国科学院沈阳自动化研究所 Self-adapting interactive multiple model mobile target tracking method

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103853908A (en) * 2012-12-04 2014-06-11 中国科学院沈阳自动化研究所 Self-adapting interactive multiple model mobile target tracking method

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
刘德虎 等: "一种修正的自适应网格交互多模型算法", 《火力与指挥控制》 *
王硕 等: "基于机动转弯目标的自适应交互式多模型算法", 《计算机仿真》 *
秦雷 等: "临近空间目标非弹道式机动模式跟踪滤波技术", 《***仿真学报》 *
秦雷 等: "基于AGIMM的临近空间机动目标跟踪滤波算法", 《***工程与电子技术》 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110716196A (en) * 2019-11-04 2020-01-21 广东中科四创科技有限公司 Multi-point low-slow small aerial target tracking sighting system
CN110716196B (en) * 2019-11-04 2023-04-25 广东中科四创科技有限公司 Multi-point low-slowness small aerial target tracking and looking method
CN111562737A (en) * 2019-12-30 2020-08-21 中国人民解放军国防科技大学 Improved AGIMM tracking method controlled by fuzzy logic
CN111562737B (en) * 2019-12-30 2022-03-11 中国人民解放军国防科技大学 Improved AGIMM tracking method controlled by fuzzy logic
CN111650577A (en) * 2020-06-12 2020-09-11 电子科技大学 Maneuvering target tracking method containing Doppler measurement under polar coordinate system
CN111650577B (en) * 2020-06-12 2022-05-24 电子科技大学 Maneuvering target tracking method containing Doppler measurement under polar coordinate system
CN111736144A (en) * 2020-07-06 2020-10-02 哈尔滨工业大学 Maneuvering turning target state estimation method only using distance observation
CN111736144B (en) * 2020-07-06 2023-09-26 哈尔滨工业大学 Maneuvering turning target state estimation method only by distance observation

Also Published As

Publication number Publication date
CN108152812B (en) 2021-06-29

Similar Documents

Publication Publication Date Title
Gao et al. Maximum likelihood principle and moving horizon estimation based adaptive unscented Kalman filter
CN108152812A (en) A kind of improvement AGIMM trackings for adjusting grid spacing
CN111797478B (en) Strong maneuvering target tracking method based on variable structure multi-model
CN109633590B (en) Extended target tracking method based on GP-VSMM-JPDA
CN108536171A (en) The paths planning method of multiple no-manned plane collaboration tracking under a kind of multiple constraint
CN104199022B (en) Target modal estimation based near-space hypersonic velocity target tracking method
CN106199580B (en) A kind of Singer model refinement algorithms based on fuzzy inference system
CN106646450A (en) Radar track anti-error correlation method based on distance stepped clustering
CN104182609B (en) The three-dimensional target tracking method that unbiased transformation based on decorrelation is measured
CN109212519B (en) Narrow-band radar target tracking method based on BF-DLSTM
CN108303095B (en) Robust volume target cooperative localization method suitable for non-Gaussian filtering
JP2023517147A (en) Systems and methods for measuring wind flow turbulence with LiDAR in complex terrain
CN116642482A (en) Positioning method, equipment and medium based on solid-state laser radar and inertial navigation
CN106546976A (en) It is a kind of to be based on long period nonuniform sampling target following processing method and processing device
CN110111356B (en) Method for calculating rotating shaft direction and rotating angular velocity of moving rotating object
CN108226887A (en) A kind of waterborne target rescue method for estimating state in the case of observed quantity transient loss
CN114608585A (en) Method and device for synchronous positioning and mapping of mobile robot
Lin et al. A convolutional neural network particle filter for UUV target state estimation
CN105701292B (en) A kind of parsing discrimination method of maneuvering target turning rate
CN114462293A (en) Method for predicting medium and long-term trajectory of hypersonic target
CN108320302A (en) CBMeMBer multi-object tracking methods based on random hypersurface
CN110333513B (en) Particle filter SLAM method fusing least square method
CN109188420B (en) Narrow-band radar target tracking method based on deep long-short term memory network
CN116907503A (en) Remote sensing formation satellite positioning method and system based on robust positioning algorithm of outlier
CN114565020B (en) Aircraft sensor signal fusion method based on deep belief network and extended Kalman filtering

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
GR01 Patent grant
GR01 Patent grant