CN108152812B

CN108152812B - Improved AGIMM tracking method for adjusting grid spacing

Info

Publication number: CN108152812B
Application number: CN201711330498.8A
Authority: CN
Inventors: 曹运合; 潘媚媚; 吴春林; 卢毅; 吴文华; 龚作豪; 王宇
Original assignee: Xidian University; Xian Cetc Xidian University Radar Technology Collaborative Innovation Research Institute Co Ltd
Current assignee: Xidian University; Xian Cetc Xidian University Radar Technology Collaborative Innovation Research Institute Co Ltd
Priority date: 2017-12-13
Filing date: 2017-12-13
Publication date: 2021-06-29
Anticipated expiration: 2037-12-13
Also published as: CN108152812A

Abstract

The invention discloses an improved AGIMM tracking method for adjusting grid spacing, which has the following ideas: determining a radar, wherein a target exists in a radar scanning range, and the radar scans and detects the target; establishing a radar rectangular coordinate system and a radar polar coordinate system, determining the measurement value of the radar to the target under the radar rectangular coordinate system, and respectively obtaining the initial state vector of the radar to the target

And an initial error covariance matrix P of the radar to the target₀(ii) a Respectively determining an interaction model set and an initial value of the interaction model set; determining a Kalman filtering initial value of an interaction model set; calculating the state vector filtering output value of the jth maneuvering turning sub-model at the 3-sampling moment

State vector filtering output value of jth maneuvering turning sub-model to N sampling time

And 3. the error covariance filter output value P of the jth maneuvering turning submodel at the sampling moment_jError covariance filtering output value P of jth maneuvering turning submodel from (3|3) to N sampling time_j(N | N) and is reported as an improved AGIMM tracking result for adjusting the grid spacing.

Description

Improved AGIMM tracking method for adjusting grid spacing

Technical Field

The invention belongs to the field of maneuvering target tracking, and particularly relates to an improved AGIMM tracking method for adjusting grid spacing, namely an improved self-adaptive grid interaction multi-model tracking method for adjusting grid spacing, which is suitable for adaptively tracking a high-speed large maneuvering target at a high speed by adjusting grid spacing through maneuvering judgment and searching relatively optimal angular velocity so as to obtain a good tracking and filtering effect.

Background

Target tracking is always the key research field in military affairs, and is the difficulty in key research for high-speed large-mobility target tracking; the tracking filtering modeling of the high-speed large maneuvering target has great difficulty due to the characteristics of large speed, quick maneuvering and difficult prediction.

Aiming at the maneuverability of a target, scholars propose a plurality of maneuvering models, so that the tracking precision of the maneuvering target is improved to a certain extent, and the tracking of a strong maneuvering target cannot be well adapted; the problem of poor tracking adaptability of a single model is solved by the interactive multi-model, and the interactive multi-model has good robust characteristic; researchers also propose a series of improved algorithms on the basis of interactive multiple models in succession, and the main improved directions are selection of a model set, calculation of model probability, improvement of a tracking filter algorithm and the like.

At present, an adaptive grid interactive multi-model (AGIMM) algorithm combining a graph theory idea according to an interactive multi-model algorithm is one of the most effective tracking algorithms, and the AGIMM algorithm is mainly characterized in that turning models with different turning rates are used as a tracking model set, the turning rates are adaptively updated by using the idea of the graph theory so that the model set is always in dynamic updating, and the adaptive multi-model algorithm has better adaptive capacity relative to an interactive multi-model algorithm with a fixed model; similarly, the AGIMM algorithm has problems that the model mesh pitch cannot be adaptively adjusted and the angular velocity estimation speed is slow.

Disclosure of Invention

In view of the above problems in the prior art, an object of the present invention is to provide an improved AGIMM method for adjusting a grid interval, which is an improved AGIMM method for performing maneuver discrimination according to residual information and further adaptively adjusting the grid interval, and which searches for an optimal angular velocity to adapt to a real angular velocity by using a one-dimensional search concept at a non-maneuver time, so as to improve not only an adaptive capability to a maneuver time, but also a tracking accuracy and an estimation speed of the angular velocity at the non-maneuver time.

In order to achieve the technical purpose, the invention is realized by adopting the following technical scheme.

An improved AGIMM method for adjusting grid spacing, comprising the steps of:

step 1, determining a radar, wherein a target exists in a radar scanning range, and the radar scans and detects the target; establishing a radar rectangular coordinate system and a radar polar coordinate system, determining the measurement value of the radar to the target under the radar rectangular coordinate system, and respectively obtaining the initial state vector of the radar to the target

And an initial error covariance matrix P of the radar to the target₀(ii) a Respectively determining an interaction model set and initial values of the interaction model set, wherein the interaction model set comprises L motorized turning sub-models

Step 2, initialization: let k 'denote the sampling time k', k 'is 3 to N, and the initial value of k' is 3; according to initial state vector of radar to target

And an initial error covariance matrix P of the radar to the target₀Determining a Kalman filtering initial value of an interaction model set;

step 3, calculating an innovation value V of the jth maneuvering turning sub-model at the k 'sampling moment according to the Kalman filtering value of the k' -1 sampling moment interaction model set_j(k ') and k' sampling time the innovation covariance matrix S of the jth maneuver turning submodel_j(k') and the state vector filter output value of the jth maneuvering turning sub-model at the sampling time of k

And the error covariance filter output value P of the jth maneuvering turning sub-model at the k' sampling moment_j(k'|k')；j＝1,2,…,L；

Step 4, according to the innovation value V of the jth maneuvering turning submodel at the k' sampling moment_jInnovation covariance matrix S of jth maneuvering turning submodel at sampling time of (k'), k_j(k'), calculating to obtain the probability p of the interactive model set at the sampling moment of k_k'；

Step 5, according to the k' sampling time interactive model set probability p_k'Respectively calculating the turning angular velocity omega of the 1 st motor-driven turning sub-model at the k' sampling moment₁(k'), turning angular velocity ω of the 2 nd motor turning submodel at the sampling time of k₂(k ') and k' sampling time the turning angular velocity ω of the 3 rd motor turning submodel₃(k')；

Step 6, adding 1 to k', and repeatedly executing the steps 3 to 5 until the state vector filtering output value of the jth maneuvering turning sub-model at the sampling moment of 3 is obtained

Compared with the prior art, the invention has the main advantages that:

(1) the method solves the defect that the grid spacing of the model cannot be adjusted in a self-adaptive manner, can ensure that the grid spacing of the model can be adjusted on line in real time according to the maneuvering size of the target, enhances the adaptability to the maneuvering change of the target, and improves the tracking performance to maneuvering time.

(2) At the non-maneuvering moment, the idea of one-dimensional search is utilized to enable the model set to converge towards the direction of the optimal angular velocity, the estimation of the angular velocity is accelerated, and the tracking precision at the non-maneuvering moment is improved.

Drawings

The invention is described in further detail below with reference to the following description of the drawings and the detailed description.

FIG. 1 is a flow chart of an improved AGIMM method of adjusting grid spacing in accordance with the present invention;

FIG. 2 is a schematic diagram of an established rectangular coordinate system and a polar coordinate system of a radar;

FIG. 3 is a graph comparing distance RMSE using the AGIMM algorithm and the method of the present invention when set at a survey noise of 1 for a cornering maneuver;

FIG. 4 is a graph of the speed RMSE comparison using the AGIMM algorithm and the method of the present invention when set at a survey noise of 1 for a cornering maneuver;

FIG. 5a is a graph of angular velocity variation for a cornering maneuver with the measurement noise 1 set using the AGIMM algorithm;

FIG. 5b is a graph of the angular velocity variation corresponding to the method of the present invention when the measured noise 1 is set for a cornering maneuver;

FIG. 6 is a comparison graph of distance RMSE using the AGIMM algorithm and the method of the present invention when set at a survey noise 2 for a cornering maneuver;

FIG. 7 is a graph comparing the speed RMSE using the AGIMM algorithm and the method of the present invention when set at a survey noise 2 for a cornering maneuver;

FIG. 8 is a comparison graph of distance RMSE using the AGIMM algorithm and the method of the present invention when setting up an initial model set 1 for a near space vehicle;

FIG. 9 is a graph comparing the velocity RMSE for an initial set of models 1, targeted for near space vehicles, using the AGIMM algorithm and the method of the present invention;

FIG. 10 is a graph comparing distance RMSE using the AGIMM algorithm when setting initial model set 1 and initial model set 2 for a near space vehicle;

FIG. 11 is a comparison graph of distance RMSE using the method of the present invention when setting initial model set 1 and initial model set 2 for a near space vehicle;

FIG. 12a is a graph of angular velocity change using the AGIMM algorithm when setting up the initial set of models 2 for a near space vehicle;

fig. 12b is a graph of the change in angular velocity using the method of the invention when setting up the initial set of models 2, targeted for a near space vehicle.

Detailed Description

Referring to fig. 1, a flow chart of an improved AGIMM method for adjusting grid spacing according to the present invention; the improved AGIMM method for adjusting the grid spacing comprises the following steps:

And an initial error covariance matrix P of the radar to the target₀(ii) a And respectively determining an interaction model set and an initial value of the interaction model set.

The substep of step 1 is:

(1a) determining a radar, wherein a target exists in a radar scanning range, and the radar scans and detects the target; any point in a track where a target is located is marked as a point target P, then a radar rectangular coordinate system is established by taking the central position where the radar is located as an origin o, the direction of the east of the radar as an x axis, the direction of the north of the radar as a y axis and the z axis determined according to the right-hand rule, wherein the process of determining the z axis is as follows: the thumb and index finger of the right hand are set to point to the x-axis and y-axis, respectively, and the middle finger points to the z-axis.

And establishing a radar polar coordinate system by taking the central position of the radar as an origin o, the radial distance between the point target P and the radar as P, the azimuth angle of the point target P relative to the radar as theta and the pitch angle epsilon of the point target P relative to the radar, wherein schematic diagrams of a radar rectangular coordinate system and a radar polar coordinate system are shown in fig. 2.

(1b) The method comprises the steps that a target exists in a radar scanning range, the radar scans and detects the target, measurement information of the target under a radar polar coordinate system (rho-theta-epsilon) can be obtained through a radar sensor, rho represents the radial distance between a point target P and the radar, theta represents the azimuth angle of the point target P relative to the radar, and epsilon represents the pitch angle of the point target P relative to the radar.

And then converting the measured value under the radar polar coordinate system into a measured value under a radar rectangular coordinate system (x-y-z), wherein if the scanning period of the radar is T ' and the total detection time of the radar on the target is T, the method is equivalent to that the radar carries out N times of sampling with the period of T ' on the target track, and N is T/T '.

The method comprises the following steps that a radar obtains a measured value of a target under a radar polar coordinate system by using a sensor carried by the radar, and specifically comprises a target radial distance measured value rho (k) obtained by the radar under the radar polar coordinate system at a k sampling moment, a target azimuth angle measured value theta (k) obtained by the radar under the radar polar coordinate system at the k sampling moment and a target pitch angle measured value epsilon (k) obtained by the radar under the radar polar coordinate system at the k sampling moment; and converting the obtained measurement value of the target under the radar polar coordinate system into the measurement value of the target by the radar under the radar rectangular coordinate system at the k sampling moment:

Z_x(k)＝ρ(k)cosε(k)cosθ(k),Z_y(k)＝ρ(k)cosε(k)sinθ(k),Z_z(k)＝ρ(k)sinε(k)

wherein k is 1,2, …, N, Z_x(k) The measured value Z of the radar to the target x direction at the k sampling moment under the radar rectangular coordinate system is represented_y(k) The measured value Z of the radar to the target y direction at the k sampling moment under the rectangular coordinate system of the radar is represented_z(k) The method comprises the steps of representing a measurement value of a radar to a target in the z direction at a sampling moment of k under a radar rectangular coordinate system, representing rho (k) a measurement value of a target radial distance obtained by the radar at the sampling moment of k under a radar polar coordinate system, representing theta (k) a measurement value of a target azimuth angle obtained by the radar at the sampling moment of k under the radar polar coordinate system, and representing epsilon (k) a measurement value of a target azimuth angle obtained by the radar at the sampling moment of k under the radar polar coordinate systemAnd (3) obtaining a target pitch angle measurement value, wherein cos represents a cosine function, and sin represents a sine function.

(1c) Calculating the initial state vector of the radar to the target by using the position measurement value of the radar to the target under the radar rectangular coordinate system at the sampling time 1 and the sampling time 2

Wherein Z is_x(2) The method comprises the steps of (1) representing a measurement value of a radar to a target in the x direction at the sampling moment 2 under a radar rectangular coordinate system, namely the distance of the target in the x direction under the radar rectangular coordinate system at the sampling moment 2; z_x(1) The method comprises the steps of (1) representing a measurement value of a radar to a target x direction under a radar rectangular coordinate system at a sampling moment; z_y(2) The measurement value of the radar to the target y direction at the sampling moment 2 under the radar rectangular coordinate system is represented, namely the distance of the target y direction under the radar rectangular coordinate system at the sampling moment 2; z_y(1) The method comprises the steps of (1) representing a measurement value of a radar to a target y direction under a radar rectangular coordinate system at a sampling moment; z_z(2) The measurement value of the radar to the target in the z direction at the sampling moment 2 under the radar rectangular coordinate system is represented, namely the distance of the target in the z direction under the radar rectangular coordinate system at the sampling moment 2; z_z(1) And the measurement value of the radar to the target z direction at the sampling moment of 1 in a radar rectangular coordinate system is represented, T' represents the radar scanning period, and superscript T represents transposition.

(Z_x(2)-Z_x(1) T' corresponds to the speed of the target in the x direction under the radar rectangular coordinate system at the 2 sampling time, (Z)_y(2)-Z_y(1) T') corresponds to the speed of the target in the y direction under the radar rectangular coordinate system at the 2 sampling time,

(Z_z(2)-Z_z(1) t') corresponds to the speed of the target in the z direction under the radar rectangular coordinate system at the 2 sampling time.

(1d) According to a target radial distance measurement value rho (2) acquired by a radar under a radar polar coordinate system at 2 sampling moments, a target azimuth angle measurement value theta (2) acquired by the radar under the radar polar coordinate system at 2 sampling moments and a pitch angle measurement value epsilon (2) acquired by the radar under the radar polar coordinate system at 2 sampling moments, a noise covariance matrix R (2) measured by the radar to a target under the radar rectangular coordinate system at 2 sampling moments is calculated, and the expression is as follows:

wherein r is₁₁(2) Representing the 1 st row and 1 st column elements, R, in the measured noise covariance matrix R (2) of the radar to the target at the sampling time 2 under the rectangular coordinate system of the radar₁₂(2) Representing the 1 st row and 2 nd column elements, R, in the measured noise covariance matrix R (2) of the radar to the target under the rectangular coordinate system of the radar at the sampling moment 2₁₃(2) Representing the 1 st row and 3 rd column elements, R, in the measured noise covariance matrix R (2) of the radar to the target at the sampling time 2 under the rectangular coordinate system of the radar₂₂(2) Representing 2-row and 2-column elements, R, in a measured noise covariance matrix R (2) of the radar to the target at the sampling time under a radar rectangular coordinate system₂₃(2) Representing 2-row and 3-column elements, R, in a measured noise covariance matrix R (2) of the radar to the target at the sampling time under a radar rectangular coordinate system₃₃(2) Representing the 3 rd row and 3 rd column elements in a measured noise covariance matrix R (2) of the radar to the target at the sampling moment 2 under a radar rectangular coordinate system, A (2) representing a noise conversion matrix from a radar polar coordinate system to the radar rectangular coordinate system at the sampling moment 2,

a measured noise variance value representing the radial distance P between the point target P and the radar,

representing the measured noise variance value of the point target P with respect to the azimuth angle theta of the radar,

and the measured noise variance value of the pitch angle epsilon of the point target P relative to the radar is shown, and the point target P is any point in the track where the target is located in the radar scanning range.

(1e) According to the measured noise covariance matrix R (2) of the radar to the target under the radar rectangular coordinate system at the sampling moment 2, calculating the initial error covariance matrix P of the radar to the target₀The expression is as follows:

wherein

Wherein r is_ij(2) Representing the ith row and jth column elements, P, in the measured noise covariance matrix R (2) of the radar to the target at the sampling time 2 under the rectangular coordinate system of the radar_ijRepresenting the initial error covariance matrix P₀The i-th row and the j-th column have the elements, i is 1,2,3, and j is 1,2, 3.

(1f) Setting L maneuvering turning submodels with different turning angular velocities, wherein the maneuvering turning submodel is a cooperative turning model mentioned in an adaptive grid interaction multi-model (AGIMM) algorithm, the AGIMM algorithm mainly takes the cooperative turning models with different turning rates as a tracking model set, and adaptively updates the turning rate of the cooperative turning model by using the thought of graph theory, so that the tracking model set is in dynamic updating all the time; and each maneuvering turning sub-model corresponds to 1 Kalman filter respectively.

Taking L motor-driven turning submodels as an interaction model set, wherein L is 3, and determining an initial value of the interaction model set, wherein the initial value of the interaction model set is the turning angular velocity omega of the 1 st motor-driven turning submodel at the 2 sampling moment₁(2) And 2. the turning angular velocity omega of the 2 nd motor-driven turning sub-model at the sampling time₂(2) And 2 the turning angular velocity omega of the 3 rd motor turning sub-model at the sampling moment₃(2)，ω₁(2)＝-ω_max,ω₂(2)＝0,ω₃(2)＝ω_max，ω_maxRepresenting the maximum value, ω, of the jump change in the angular velocity of the preset turning angle_maxThe value range of (a) is between-180 DEG and 180 DEG, in particular omega_maxThe value of (a) can be set empirically according to the motion characteristics of the target to be tracked actually.

Step 2, determining state transition matrixes of the L motorized turning submodels according to the turning angular velocity, and setting a process noise matrix; and determining initial probability of the interaction model set and Markov state transition probability among the L maneuvering turning submodels.

The substep of step 2 is:

(2a) setting the turning angular velocity corresponding to the k sampling time interactive model set as M_k，

M_k＝{ω₁(k),ω₂(k),ω₃(k)}，ω₁(k) Represents the turning angular velocity, omega, of the 1 st motor-driven turning sub-model at the k-sampling time₂(k) Represents the turning angular velocity, omega, of the 2 nd motor-driven turning sub-model at the k-sampling time₃(k) The turning angular velocity of the 3 rd motor turning sub-model at the k sampling time is shown.

Setting the corresponding probability of the interactive model set at the sampling moment k as p_k，p_k＝{μ₁(k),μ₂(k),μ₃(k)}，μ₁(k) Represents the probability of the 1 st motor-driven turning sub-model at the k sampling moment, mu₂(k) Represents the probability of the 2 nd motor-driven turning sub-model at the k sampling moment, mu₃(k) The probability of the 3 rd motor-driven turning sub-model at the k sampling time is shown.

(2b) Determining the transition probability p of the Markov model from the ith maneuvering turning submodel to the jth maneuvering turning submodel_ijThe expression is as follows:

wherein, i is 1,2,3, j is 1,2, 3.

(2c) The interactive model set comprises 3 motorized turning sub-models, and a state transition matrix F of the jth motorized turning sub-model at the k sampling moment is calculated_j(k)。

Specifically, the state transition matrix of each maneuver turning submodel can be decomposed into two parts, namely a horizontal direction and a vertical direction in a radar rectangular coordinate system: the horizontal direction is regarded as uniform turning motion and is related to the changes of the directions of the x axis and the y axis; and if the vertical direction is regarded as uniform motion and is only related to the change of the z-axis direction, obtaining a continuous state equation of the jth maneuvering turning sub-model at the t moment under the radar rectangular coordinate system as follows:

wherein x (t) represents the distance of the target in the x-axis direction at the time t, y (t) represents the distance of the target in the y-axis direction at the time t, and z (t) represents the distance of the target in the z-axis direction at the time t;

representing the speed of the target in the x-axis direction at time t,

indicating the speed of the target in the y-axis direction at time t,

representing the speed of the target in the z-axis direction at the moment t;

represents the acceleration of the target in the x-axis direction at time t,

represents the acceleration of the target in the y-axis direction at time t,

representing the acceleration of the target in the z-axis direction at the moment t; omega_j(t) represents the turning angular velocity of the jth motor-driven turning sub-model at time t,

a state transition matrix representing the horizontal direction of the jth maneuvering turning sub-model at the time t,

and the state transition matrix represents the state transition matrix in the vertical direction of the jth maneuvering turning sub-model at the time t, and t represents a time variable.

For the state transition matrix of the horizontal direction

And a state transition matrix in the vertical direction

Respectively carrying out discretization processing with a sampling period of T' to respectively obtain a state transition matrix of the jth maneuvering turning sub-model at the k sampling time in the horizontal direction

And the state transition matrix of the jth maneuvering turning sub-model at the k sampling time in the vertical direction

The expression is as follows:

then calculating to obtain a state transition matrix F of the jth maneuvering turning sub-model at the k sampling moment_j(k) The expression is as follows:

wherein j is 1,2,3, 0_2×4Represents a 2 × 4 dimensional full 0 matrix, 0_4×2To represent4 x 2 dimensional full 0 matrix, omega_j(k) Represents the turning angular velocity of the jth maneuvering turning sub-model at the k sampling time, T' represents the radar scanning period,

(2d) calculating the process noise covariance matrix Q of the jth maneuvering turning sub-model at the k sampling moment_j(k) The expression is as follows:

wherein,

wherein j is 1,2,3, q_jRepresenting the process noise variance, process noise variance q, of the jth maneuver Turn submodel_jBased on the measured noise and the motor-driven turning sub-model attribute setting, generally taken at σ_ρTo

σ of_ρA measured noise standard deviation representing the radial distance P between the point target P and the radar,

the method comprises the steps of representing a measurement noise variance value of a radial distance rho between a point target P and a radar, wherein the point target P is any point in a track where a target is located in a radar scanning range; process noise variance q_jThe larger the setting is, the more the model can adapt to the target under the background of large maneuver or large noise under the condition that other parameters are unchanged; otherwise, the process noise variance q_jThe smaller the setting is, the better the tracking precision of the model on the target under the weak maneuvering or low noise background is shown under the condition that other parameters are not changed;

represents the process noise covariance matrix of the jth maneuvering turning submodel at the k sampling time in the horizontal direction,

and representing the process noise covariance matrix of the jth maneuvering turning sub-model at the k sampling time in the vertical direction.

Initialization: let k 'denote the sampling time k', k 'is 3 to N, and the initial value of k' is 3; according to initial state vector of radar to target

And an initial error covariance matrix P of the radar to the target₀And determining a Kalman filtering initial value of the interaction model set.

Step 3, carrying out weighted interactive calculation on the state vector and the error covariance matrix of the sampling time of the L motorized turning submodels k-1 by using the model set probability and the model transfer probability to serve as filtering input of the sampling time of the L motorized turning submodels k, and then respectively carrying out Kalman filtering on each motorized turning submodel by using a state transfer frame and a measurement frame of a motorized target under a rectangular coordinate system; because the initial sampling time of the kalman filter is 2 sampling times, the kalman filtering starts from 3 sampling times; the sampling time participating in the kalman filter is set to k ', k ' is 3 to N, and the initial value of k ' is 3.

The substep of step 3 is:

(3a) determining a Kalman filtering value of a k ' -1 sampling moment interactive model set, wherein the Kalman filtering value of the k ' -1 sampling moment interactive model set is a Kalman filtering initial value of the interactive model set when k ' is 3, and the Kalman filtering initial value of the interactive model set comprises a state filtering output value of a 1 st maneuvering turning sub-model at a 2 sampling moment

2 state filtering output value of 2 nd motor-driven turning submodel at sampling moment

2 state filtering output value of 3 rd maneuvering turning submodel at sampling moment

2 sampling moment 1 st motor-driven turning submodel error covariance filtering output value P₁Error covariance filter output value P of (2|2) and 2 sampling time 2 nd maneuvering turning submodel₂Error covariance filter output value P of (2|2) and 2 sampling time 3 rd maneuvering turning submodel₃(2|2), which are respectively expressed as:

P₁(2|2)＝P₀，P₂(2|2)＝P₀，P₃(2|2)＝P₀

wherein,

representing the initial state vector, P, of the radar to the target₀Representing initial error co-square of radar to targetA difference matrix.

(3b) Calculating the interactive state vector input value of the jth maneuvering turning sub-model at the k' sampling moment

The expression is as follows:

wherein, mu_i|j(k ' -1| k ' -1) represents the one-step state transition probability from the ith to the jth motorised turning submodels at the k ' -1 sampling instant,

representing the state filter output value, mu, of the ith motor-driven turning submodel at the k' -1 sampling time_i(k '-1) represents the probability of the ith maneuvering turning submodel at the sampling time of k' -1, p_ijAnd representing the transition probability of the Markov model from the ith maneuvering turning submodel to the jth maneuvering turning submodel.

(3c) Calculating the input value P of the interactive error covariance matrix of the jth maneuvering turning sub-model at the k' sampling moment_oj(k '-1| k' -1), which is expressed as:

wherein, mu_i|j(k ' -1| k ' -1) represents the one-step state transition probability from the ith to the jth maneuvering turn submodels at the sampling time of k ' -1, p_ijRepresenting the Markov model transition probability, P, from the ith maneuver turning submodel to the jth maneuver turning submodel_j(k ' -1| k ' -1) represents an error covariance filter output value of the jth maneuver turning submodel at the sampling time k ' -1,

state filtering output value of j-th maneuvering turning sub-model representing k' -1 sampling moment

Interaction state vector input value of jth maneuvering turning sub-model at k' sampling moment

The difference between the values of the two signals,

the state filtering output value of the jth maneuvering turning sub-model at the k' -1 sampling moment is represented,

and (4) representing the interactive state vector input value of the jth maneuvering turning sub-model at the k' sampling moment.

(3d) Inputting the interactive state vector of the jth maneuvering turning sub-model at the k' sampling moment

And the input value P of the interactive error covariance matrix of the jth maneuvering turning submodel at the k' sampling moment_oj(k '-1| k' -1) is used as the Kalman filter input value of the jth maneuvering turning sub-model at the k sampling moment, and then Kalman filtering is carried out.

Respectively calculating the predicted values of the state vectors of the jth maneuvering turning sub-model from the k '-1 sampling moment to the k' sampling moment

And the predicted value P of the error covariance matrix of the jth maneuvering turning sub-model from the sampling time k-1 to the sampling time k_j(k '| k' -1), which are respectively expressed as:

P_j(k'|k'-1)＝F_j(k')P_oj(k'-1|k'-1)[F_j(k')]^T+Q_j(k')

wherein Q is_j(k ') represents the process noise covariance matrix of the jth motorized turning submodel at the sampling instant k', F_j(k ') represents the state transition matrix of the jth maneuver turning sub-model at the k' sampling time.

Then respectively calculating the innovation value V of the jth maneuvering turning sub-model at the k' sampling moment_jInnovation covariance matrix S of jth maneuvering turning submodel at sampling time of (k'), k_j(K'), gain matrix K of jth maneuvering turning submodel at sampling time of K_j(k') expressed as:

S_j(k')＝H_j(k')P_j(k'|k'-1)[H_j(k')]^T+R_j(k')

wherein Z is_j(k ') represents a measurement value of the jth maneuver turning sub-model at the sampling time k',

Z_j(k')＝[Z_x(k'),Z_y(k'),Z_z(k')]^T，Z_x(k ') represents the measured value of the radar to the target x direction under the radar rectangular coordinate system at the sampling time k', Z_y(k ') represents the measured value of the radar to the target y direction under the rectangular coordinate system of the radar at the sampling time k', Z_z(k ') represents a measurement value H of the radar to the target z direction under the radar rectangular coordinate system at the sampling time k', and_j(k ') represents a measurement matrix of the jth maneuver turning sub-model at the sampling time k',

R_j(k ') represents the measured noise of the jth maneuvering turning sub-model at the k' sampling moment to the target by the radar under the rectangular coordinate system of the radarCovariance matrix, which can be expressed as follows:

wherein r is₁₁(k ') represents the measured noise covariance matrix R of the j th maneuvering turning submodel at the k' sampling moment to the target by the radar under the rectangular coordinate system of the radar_j(k') line 1, column 1 element, r₁₂(k ') represents the measured noise covariance matrix R of the j th maneuvering turning submodel at the k' sampling moment to the target by the radar under the rectangular coordinate system of the radar_j(k') line 1 and column 2 elements, r₁₃(k ') represents the measured noise covariance matrix R of the j th maneuvering turning submodel at the k' sampling moment to the target by the radar under the rectangular coordinate system of the radar_j(k') line 1 and column 3 elements, r₂₂(k ') represents the measured noise covariance matrix R of the j th maneuvering turning submodel at the k' sampling moment to the target by the radar under the rectangular coordinate system of the radar_j(k') row 2, column 2 elements, r₂₃(k ') represents the measured noise covariance matrix R of the j th maneuvering turning submodel at the k' sampling moment to the target by the radar under the rectangular coordinate system of the radar_j(k') row 2, column 3 elements, r₃₃(k ') represents the measured noise covariance matrix R of the j th maneuvering turning submodel at the k' sampling moment to the target by the radar under the rectangular coordinate system of the radar_j(k ') 3 rd row and 3 rd column elements, ρ (k ') represents a measured value of a radial distance of a target acquired by the radar in a radar polar coordinate system at a sampling moment k ', θ (k ') represents a measured value of an azimuth angle of the target acquired by the radar in the radar polar coordinate system at the sampling moment k ', ε (k ') represents a measured value of an elevation angle of the target acquired by the radar in the radar polar coordinate system at the sampling moment k ', A (k ') represents a noise conversion matrix from the radar polar coordinate system to the radar rectangular coordinate system at the sampling moment k ',

Finally, the state vector filtering output value of the jth maneuvering turning sub-model at the k' sampling moment is obtained through calculation

And the error covariance filter output value P of the jth maneuvering turning sub-model at the k' sampling moment_j(k '| k') expressed as:

P_j(k'|k')＝P_j(k'|k'-1)-K_j(k')S_j(k')[K_j(k')]^T

step 4, constructing a likelihood function by utilizing the information output by the three model filtering and the corresponding information covariance and the characteristic of Gaussian distribution; then, updating the model probability at the sampling moment of k ' by using the likelihood function at the sampling moment of each model k ', the model probability at the sampling moment of k ' -1 and the model transition probability; and finally, selecting the maximum probability in the model probabilities, and performing maneuvering judgment on the target by using the distance function of the model with the maximum model probability.

The substep of step 4 is:

(4a) according to the innovation value V of the jth maneuvering turning sub-model at the k' sampling moment_jInnovation covariance matrix S of jth maneuvering turning submodel at sampling time of (k'), k_j(k '), calculating the distance of the jth maneuvering turning sub-model at the k' sampling momentFunction D_j(k')，

Then calculating the likelihood function Lambda of the jth maneuvering turning sub-model at the k' sampling moment_j(k') expressed as:

wherein j is 1,2, 3; then, the likelihood function lambda of the jth maneuvering turning sub-model at the k' sampling moment is utilized_j(k ') and k' -1 sampling time ith probability mu of motor-driven turning submodel_i(k '-1), calculating the probability mu of the jth maneuvering turning sub-model at the k' sampling moment_j(k') expressed as:

further obtaining the probability p of the interactive model set at the k' sampling moment_k'，p_k'＝{μ₁(k'),μ₂(k'),μ₃(k')}，μ₁(k') represents the probability, μ, of the 1 st motor-driven turning sub-model at the sampling time k₂(k') represents the probability, μ, of the 2 nd motor-driven turning sub-model at the sampling time k₃(k ') represents the probability of the 3 rd motor-driven turning sub-model at the sampling time k'.

(4b) Finding out k' sampling moment interactive model set probability p_k'Maximum probability max (p) of_k')，

max(p_k')＝max{μ₁(k'),μ₂(k'),μ₃(k')}。

(4c) Selecting the maneuvering turning submodel corresponding to the maximum probability, recording as the jth maneuvering turning submodel at the k 'sampling time, and utilizing the distance function D of the jth maneuvering turning submodel at the k' sampling time_j'(k') performing maneuver discrimination on the target; in view of D_j'(k') compliance with measuring χ of dimension m²Distribution, assuming that the probability of strong maneuvering of the target is 0.01, query χ²Setting the available threshold of the distribution table to 6.637 and setting maneuver judgmentThe difference threshold is M-7.

When D is present_j'(k')>When M, judging that the target is mobile at the sampling moment of k'; otherwise, the target is judged to be non-maneuvering at the sampling moment k'.

Step 5, when the target is judged to be maneuvering at the sampling moment of k', calculating the square root of the difference between the distance function and the maneuvering threshold as the adjustment multiple of the minimum grid; and when the target is judged to be non-maneuvering at the sampling moment k', searching the angular speed which is most consistent with the actual model in the model set in the non-maneuvering period by using the model probability.

The substep of step 5 is:

(5a) setting G as the grid spacing, G₀The set minimum grid interval is generally between 0.1 and 1 degrees, and can be selected according to the mobility of the target to be tracked, wherein the mobility is larger than that, and is smaller on the contrary; the grid distance G can be adaptively scaled and adjusted according to the maneuverability of the target at the sampling time k ', and when the maneuverability of the target at the sampling time k' is large, the grid distance G is enlarged; and vice versa, becomes smaller.

(5b) Calculating the turning angular velocity M of the k' sampling moment interactive model set_k'，M_k'＝{ω₁(k'),ω₂(k'),ω₃(k') }, which substeps are:

(5b.1) setting max (p)_k')＝μ₂(k')，μ₂(k ') represents the probability of the 2 nd motor-driven turning sub-model at the sampling moment of k'; if D is₂(k')>M, then

ω₂(k')＝ω₂(k'-1),ω₁(k')＝ω₂(k')-G,ω₃(k')＝ω₂(k') + G; otherwise, performing optimization probability approximation on the k' sampling moment interaction model set to obtain:

wherein D is₂(k ') represents the distance of the 2 nd motor-driven turning sub-model at the k' sampling timeDistance function, G denotes the grid spacing, ω₂(k') represents the turning angular velocity, ω, of the 2 nd motor-turning submodel at the sampling time k₂(k '-1) represents the turning angular velocity, ω, of the 2 nd motor turning submodel at the sampling time of k' -1₁(k') represents the turning angular velocity, ω, of the 1 st motor-driven turning sub-model at the sampling time k₁(k '-1) represents the turning angular velocity, ω, of the 1 st maneuver turning submodel at the sampling time of k' -1₃(k') represents the turning angular velocity, ω, of the 3 rd motor turning sub-model at the sampling time k₃(k '-1) represents the turning angular velocity of the 3 rd motor turning sub-model at the k' -1 sampling time.

(5b.2) setting max (p)_k')＝μ₁(k')，μ₁(k ') represents the probability of the 1 st motor-driven turning sub-model at the sampling time k'; if D is₁(k')>M, then

ω₂(k')＝ω₁(k'-1),ω₁(k')＝ω₂(k')-G,ω₃(k')＝ω₂(k') + G; otherwise, performing optimization probability approximation on the k' sampling moment interaction model set to obtain:

wherein D is₁(k ') denotes the distance function of the 1 st motor-driven turning sub-model at the sampling instant k', G denotes the grid spacing, ω₁(k') represents the turning angular velocity, ω, of the 1 st motor-driven turning sub-model at the sampling time k₁(k '-1) represents the turning angular velocity, ω, of the 1 st maneuver turning submodel at the sampling time of k' -1₂(k') represents the turning angular velocity, ω, of the 2 nd motor-turning submodel at the sampling time k₂(k '-1) represents the turning angular velocity, ω, of the 2 nd motor turning submodel at the sampling time of k' -1₃(k') represents the turning angular velocity, ω, of the 3 rd motor turning sub-model at the sampling time k₃(k '-1) represents the turning angular velocity, μ, of the 3 rd motor turning sub-model at the sampling time of k' -1₂(k ') denotes the k' sampling instantProbability of 2 motorized turning submodels, mu₃(k ') represents the probability of the 3 rd motor-driven turning sub-model at the sampling time k'.

(5b.3) setting max (p)_k′)＝μ₃(k′)，μ₃(k ') represents the probability of the 3 rd motor-driven turning sub-model at the sampling moment of k'; if D is₃(k') is > M, then

ω₂(k′)＝ω₃(k′-1)，ω₁(k′)＝ω₂(k′)-G，ω₃(k′)＝ω₂(k') + G; otherwise, performing optimization probability approximation on the k' sampling moment interaction model set to obtain:

wherein D is₃(k ') denotes the distance function of the 3 rd motor-driven turning sub-model at the sampling time k', G denotes the grid spacing, ω₂(k') represents the turning angular velocity, ω, of the 2 nd motor-turning submodel at the sampling time k₃(k '-1) represents the turning angular velocity, ω, of the 3 rd motor turning submodel at the sampling time of k' -1₃(k') represents the turning angular velocity, ω, of the 3 rd motor turning sub-model at the sampling time k₁(k') represents the turning angular velocity, ω, of the 1 st motor-driven turning sub-model at the sampling time k₁(k '-1) represents the turning angular velocity, ω, of the 1 st maneuver turning submodel at the sampling time of k' -1₂(k '-1) represents the turning angular velocity, μ, of the 2 nd motor turning submodel at the sampling time of k' -1₂(k') represents the probability, μ, of the 2 nd motor-driven turning sub-model at the sampling time k₂(k ') represents the probability of the 2 nd motor-driven turning sub-model at the sampling time k'.

Step 6, adding 1 to k ', and sampling the k' to obtain the probability p of the interactive model set at the moment_k′And k' sampling time interactive model set turning angular velocity M_k′Repeatedly executing the steps 3 to 3 as a new interactive model set probability set and an interactive model set turning angular velocity5, until obtaining the state vector filtering output value of the jth maneuvering turning sub-model at the 3 sampling time

The effects of the present invention are further demonstrated by the following simulations.

Simulation environment 1: the target is set as a turning maneuver model, and the initial state of the target is

The simulation time period is 350s, the target is changed in angular velocity every 70s, and the angular velocity is specifically changed by setting w to [0.5 °, -10 °, 0.7 °, 6 °, -0.7 ° ]]The sampling time interval T is 1 s. Setting measurement noise 1: the standard deviation of the distance is 50m, and the standard deviation of the azimuth angle and the pitch angle is 0.1 degrees.

Initial model set to W₁-10 °,0.1 °, 10 °, i.e. ω_max10 °, minimum grid spacing is set to G₀＝0.5°。

(II) simulation environment 2: changing the measurement noise on the basis of the simulation environment 1, and setting the measurement noise 2: the standard deviation of the distance is 100m, and the standard deviation of the azimuth angle and the pitch angle is 0.2 degrees.

(III) simulation environment 3: the target is set as a hypersonic aircraft model in the near space, and the initial state of the target is

The target always makes uniform acceleration motion in the z-axis.

t is 0-20 s: the target is straightened at a constant speed in the direction of the x axisLinear motion, acceleration in the y direction is 1 g; t is 21-380 s: the target makes uniform turning motion along the x axis, and the angular speed of the uniform turning is 0.098 rad/s. t is 81-400 s: the target makes uniform acceleration linear motion, and the acceleration in the y direction is-1 g; the metrology noise is set to noise 1 and the initial model set is set to W₁With minimum grid spacing set to G₀＝0.5°。

(IV) simulation Environment 4: changing the setting of the initial model set on the basis of the simulation environment 3, setting the initial model set as W₂-20 °,0.1 °,20 °, i.e. ω_max＝20°。

(V) simulation result analysis:

5.1 analysis of simulation environment 1 results, it can be seen in conjunction with fig. 3 and 4 that in case of noise 1, i.e. in case of less noise, the improved AGIMM algorithm is better than the AGIMM algorithm, but the advantages are not very obvious. However, comparing the turning angles set in conjunction with the simulation environment in fig. 5a and 5b, it can be seen that the estimation of the angular velocity at the target maneuver times 71s and 211s by the two algorithms changes, but the improved AGIMM algorithm is relatively more accurate than the estimation of the target angular velocity by the AGIMM algorithm.

5.2, simulation environment 2 result analysis: as can be seen from observing fig. 6 and fig. 7, when the measurement noise is increased, i.e., the noise is set as noise 2, the tracking performance of the AGIMM algorithm on the target is greatly reduced, and the improved AGIMM algorithm can still maintain the performance stability, which indicates that the anti-interference performance of the improved algorithm is better.

5.3, simulation environment 3 result analysis: as can be seen from the observation of FIG. 8 and FIG. 9, the improved algorithm has better tracking effect on the hypersonic flight vehicle in the near space, no matter the distance RMSE comparison graph or the speed RMSE comparison graph.

5.4, simulation environment 4 result analysis: respectively setting an initial model set W₁And W₂Observing fig. 10 and 11, it can be seen that the improved AGIMM algorithm is less affected by the initial model set, while the AGIMM algorithm has higher requirements for the initial model set, and when the set initial model set and the turning angular velocity that may occur due to the target motion deviate too much, the tracking filter is seriously affectedWave effect; looking at fig. 12a and 12b, it can also be seen that the improved AGIMM algorithm can quickly converge to find the angular velocity approaching the actual motion of the target even if the initial model set is set to a large range.

In conclusion, the simulation experiment verifies the correctness, the effectiveness and the reliability of the method.

It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention; thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.

Claims

1. An improved AGIMM method for adjusting grid spacing, comprising the steps of:

And an initial error covariance matrix P of the radar to the target₀(ii) a Respectively determining an interaction model set and an initial value of the interaction model set, wherein the interaction model set comprises L motorized turning sub models;

step 3, calculating an innovation value V of the jth maneuvering turning sub-model at the k 'sampling moment according to the Kalman filtering value of the k' -1 sampling moment interaction model set_j(k′) And the innovation covariance matrix S of the jth maneuvering turning submodel at the k' sampling moment_j(k') and the state vector filter output value of the jth maneuvering turning sub-model at the sampling time of k

And the error covariance filter output value P of the jth maneuvering turning sub-model at the k' sampling moment_j(k′|k′)；j＝1，2，…，L；

Step 4, according to the innovation value V of the jth maneuvering turning submodel at the k' sampling moment_jInnovation covariance matrix S of jth maneuvering turning submodel at sampling time of (k'), k_j(k'), calculating to obtain the probability p of the interactive model set at the sampling moment of k_k′；

Step 5, according to the k' sampling time interactive model set probability p_k′Respectively calculating the turning angular velocity omega of the 1 st motor-driven turning sub-model at the k' sampling moment₁(k'), turning angular velocity ω of the 2 nd motor turning submodel at the sampling time of k₂(k ') and k' sampling time the turning angular velocity ω of the 3 rd motor turning submodel₃(k′)；

2. The method of claim 1, wherein in step 1, the establishing of the radar rectangular coordinate system and the radar polar coordinate system comprises the following steps:

any point in a track where a target is located is marked as a point target P, then a radar rectangular coordinate system is established by taking the central position where the radar is located as an origin o, the direction of the east of the radar as an x axis, the direction of the north of the radar as a y axis and the z axis determined according to the right-hand rule, wherein the process of determining the z axis is as follows: setting the thumb and the index finger of the right hand to point to the x axis and the y axis respectively, and setting the middle finger point to be the z axis;

establishing a radar polar coordinate system by taking the central position of the radar as an origin o, the radial distance between a point target P and the radar as rho, the azimuth angle of the point target P relative to the radar as theta and the pitch angle epsilon of the point target P relative to the radar;

the measurement value of the radar to the target under the radar rectangular coordinate system is specifically the measurement value of the radar to the target under the radar rectangular coordinate system at the k sampling time:

Z_x(k)＝ρ(k)cosε(k)cosθ(k)，Z_y(k)＝ρ(k)cosε(k)sinθ(k)，Z_z(k) ρ (k) sin ∈ (k) where k ═ l, 2, …, N, Z_x(k) The measured value Z of the radar to the target x direction at the k sampling moment under the radar rectangular coordinate system is represented_y(k) The measured value Z of the radar to the target y direction at the k sampling moment under the rectangular coordinate system of the radar is represented_z(k) The method comprises the steps of representing a measured value of a radar to a target in the z direction at a k sampling moment under a radar rectangular coordinate system, representing a measured value of a target radial distance obtained by the radar under a radar polar coordinate system at the k sampling moment by rho (k), representing a measured value of a target azimuth angle obtained by the radar under the radar polar coordinate system at the k sampling moment by theta (k), representing a measured value of a target pitch angle obtained by the radar under the radar polar coordinate system at the k sampling moment by epsilon (k), representing a cosine function by cos, and representing a sine function by sin.

3. The method of claim 2, wherein in step 1, the radar is used to estimate the initial state vector of the target

And an initial error covariance matrix P of the radar to the target₀The obtaining process is as follows:

(1a) obtaining a target radial distance measurement value rho (k) obtained by a radar under a radar polar coordinate system at k sampling time, a target azimuth angle measurement value theta (k) obtained by the radar under the radar polar coordinate system at k sampling time, and a target pitch angle measurement value epsilon (k) obtained by the radar under the radar polar coordinate system at k sampling time, and obtaining a measurement value of the radar to a target under the radar rectangular coordinate system at k sampling time:

Z_x(k)＝ρ(k)cosε(k)cosθ(k)，Z_y(k)＝ρ(k)cosε(k)sinθ(k)，Z_z(k)＝ρ(k)sinε(k)

wherein k is l, 2, …, N, Z_x(k) The measured value Z of the radar to the target x direction at the k sampling moment under the radar rectangular coordinate system is represented_y(k) The measured value Z of the radar to the target y direction at the k sampling moment under the rectangular coordinate system of the radar is represented_z(k) The method comprises the steps of representing a measured value of a radar to a target z direction at a k sampling moment under a radar rectangular coordinate system, representing a measured value of a target radial distance obtained by the radar under a radar polar coordinate system at the k sampling moment by rho (k), representing a measured value of a target azimuth angle obtained by the radar under the radar polar coordinate system at the k sampling moment by theta (k), representing a measured value of a target pitch angle obtained by the radar under the radar polar coordinate system at the k sampling moment by epsilon (k), representing a cosine function by cos, and representing a sine function by sin;

(1b) calculating initial state vector of radar to target

Wherein Z is_x(2) The method comprises the steps of (1) representing a measurement value of a radar to a target in the x direction at the sampling moment 2 under a radar rectangular coordinate system, namely the distance of the target in the x direction under the radar rectangular coordinate system at the sampling moment 2; z_x(1) Indicating 1 sampling instant in the radarMeasuring values of the radar to the target x direction under the rectangular coordinate system; z_y(2) The measurement value of the radar to the target y direction at the sampling moment 2 under the radar rectangular coordinate system is represented, namely the distance of the target y direction under the radar rectangular coordinate system at the sampling moment 2; z_y(1) The method comprises the steps of (1) representing a measurement value of a radar to a target y direction under a radar rectangular coordinate system at a sampling moment; z_z(2) The measurement value of the radar to the target in the z direction at the sampling moment 2 under the radar rectangular coordinate system is represented, namely the distance of the target in the z direction under the radar rectangular coordinate system at the sampling moment 2; z_z(1) The method comprises the steps that 1, a measurement value of a radar to a target z direction at the sampling moment under a radar rectangular coordinate system is represented, T' represents a radar scanning period, and superscript T represents transposition;

(1c) calculating 2a measured noise covariance matrix R (2) of the radar to the target at the sampling moment under a radar rectangular coordinate system, wherein the expression is as follows:

wherein r is₁₁(2) Representing the 1 st row and 1 st column elements, R, in the measured noise covariance matrix R (2) of the radar to the target at the sampling time 2 under the rectangular coordinate system of the radar₁₂(2) Representing the 1 st row and 2 nd column elements, R, in the measured noise covariance matrix R (2) of the radar to the target under the rectangular coordinate system of the radar at the sampling moment 2₁₃(2) Representing the 1 st row and 3 rd column elements, R, in the measured noise covariance matrix R (2) of the radar to the target at the sampling time 2 under the rectangular coordinate system of the radar₂₂(2) Representing 2-row and 2-column elements, R, in a measured noise covariance matrix R (2) of the radar to the target at the sampling time under a radar rectangular coordinate system₂₃(2) Representing 2-row and 3-column elements, R, in a measured noise covariance matrix R (2) of the radar to the target at the sampling time under a radar rectangular coordinate system₃₃(2) Representing the measurement noise of the radar to the target under the radar rectangular coordinate system at the sampling moment 2The 3 rd row and 3 rd column elements in the acoustic covariance matrix R (2), A (2) represents the noise conversion matrix from the radar polar coordinate system to the radar rectangular coordinate system at the 2 sampling moments,

representing a measurement noise variance value of a point target P relative to a pitch angle epsilon of the radar, wherein the point target P is any point in a track where a target is located in a radar scanning range; rho (2) represents a target radial distance measurement value obtained by the radar under a radar polar coordinate system at 2 sampling moments, theta (2) represents a target azimuth angle measurement value obtained by the radar under the radar polar coordinate system at 2 sampling moments, and epsilon (2) represents a target pitch angle measurement value obtained by the radar under the radar polar coordinate system at 2 sampling moments;

(1d) calculating an initial error covariance matrix P of the radar to the target₀The expression is as follows:

wherein

Wherein r is_ij(2) Representing the ith row and jth column elements, P, in the measured noise covariance matrix R (2) of the radar to the target at the sampling time 2 under the rectangular coordinate system of the radar_ijRepresenting the initial error covariance matrix P₀The i-th row and the j-th column of the element have i-l, 2,3 and j-1, 2 and 3.

4. The method of claim 3, wherein in step 1, the interactive model set and the initial values of the interactive model set are determined by:

setting L motor-driven turning submodels with different turning angular speeds, wherein the motor-driven turning submodel is a cooperative turning model mentioned in the self-adaptive grid interaction multi-model algorithm, taking the L motor-driven turning submodels as an interaction model set, and determining an initial value of the interaction model set, wherein the initial value of the interaction model set is the turning angular speed omega of the 1 st motor-driven turning submodel at the 2 sampling moment₁(2) And 2. the turning angular velocity omega of the 2 nd motor-driven turning sub-model at the sampling time₂(2) And 2 the turning angular velocity omega of the 3 rd motor turning sub-model at the sampling moment₃(2)，ω₁(2)＝-ω_max，ω₂(2)＝0，ω₃(2)＝ω_max，ω_maxRepresenting the maximum value of the jump change of the preset turning angular velocity.

5. The improved AGIMM method for adjusting grid spacing according to claim 4, wherein in step 2, the initial value of kalman filtering of the interaction model set specifically includes: 2 state filtering output value of 1 st motor-driven turning sub-model at sampling moment

2 sampling moment 1 st motor-driven turning submodel error covariance filtering output value P₁Error covariance filter output value P of (2|2) and 2 sampling time 2 nd maneuvering turning submodel₂Error covariance filter output value P of (2|2) and 2 sampling time 3 rd maneuvering turning submodel₃(2|2) expressions thereof are respectivelyComprises the following steps:

P₁(2|2)＝P₀，P₂(2|2)＝P₀，P₃(2|2)＝P₀

wherein,

representing the initial state vector, P, of the radar to the target₀An initial error covariance matrix of the radar to the target is represented.

6. The method of claim 5, wherein in step 3, the k' sampling time is the innovation value V of the jth maneuver turning submodel_j(k ') and k' sampling time the innovation covariance matrix S of the jth maneuver turning submodel_j(k') and the state vector filter output value of the jth maneuvering turning sub-model at the sampling time of k

S_j(k′)＝H_j(k′)P_j(k′|k′-1)[H_j(k′)]^T+R_j(k′)

P_j(k′|k′)＝P_j(k′|k′-1)-K_j(k′)S_j(k′)[K_j(k′)]^T

wherein Z is_j(k ') represents a measurement value of the jth maneuver turning sub-model at the k' sampling time, H_j(k ') represents a measurement matrix of the jth maneuver turning sub-model at the sampling time k',

representing the predicted value, P, of the state vector of the jth motorised turning sub-model from the k '-1 sampling instant to the k' sampling instant_j(k ' | k ' -1) represents the error covariance matrix prediction value from the sampling moment of k-1 to the sampling moment of k ' for the jth maneuvering turning sub-model, R_j(K ') represents the measured noise covariance matrix of the j th maneuvering turning submodel at the K' sampling moment to the target by the radar under the rectangular coordinate system of the radar, K_j(k ') represents the gain matrix of the jth maneuver turning sub-model at the k' sampling instant.

7. The improved AGIMM method of claim 6 wherein in step 4, the k' sampling time interaction model set probability p_k′The obtaining process is as follows:

(4a) according to the innovation value V of the jth maneuvering turning sub-model at the k' sampling moment_jInnovation covariance matrix S of jth maneuvering turning submodel at sampling time of (k'), k_j(k '), calculating the distance function D of the jth maneuvering turning sub-model at the k' sampling moment_j(k′)，

further obtaining the probability p of the interactive model set at the k' sampling moment_k′，p_k′＝{μ₁(k′)，μ₂(k′)，μ₃(k′)}，μ₁(k') represents the probability, μ, of the 1 st motor-driven turning sub-model at the sampling time k₂(k') represents the probability, μ, of the 2 nd motor-driven turning sub-model at the sampling time k₃(k ') represents the probability of the 3 rd motor-driven turning sub-model at the sampling time k'.

8. The improved AGIMM method of adjusting grid spacing of claim 7, wherein the substep of step 5 is:

(5a) setting G as the grid spacing, G₀Is the set minimum grid spacing;

(5b) calculating the turning angular velocity M of the k' sampling moment interactive model set_k′，

M_k′＝{ω₁(k′)，ω₂(k′)，ω₂(k') }, which substeps are:

(5b.1) setting max (p)_k′)＝μ₂(k′)，μ₂(k ') represents the probability of the 2 nd motor-driven turning sub-model at the sampling moment of k'; if D is₂(k') is > M, then

ω₂(k′)＝ω₂(k′-1)，ω₁(k′)＝ω₂(k′)-G，ω₃(k′)＝ω₂(k') + G; otherwise, performing optimization probability approximation on the k' sampling moment interaction model set to obtain:

wherein D is₂(k ') denotes the distance function of the 2 nd motor-driven turning sub-model at the sampling time k', G denotes the grid spacing, ω₂(k') represents the turning angular velocity, ω, of the 2 nd motor-turning submodel at the sampling time k₂(k '-1) represents the turning angular velocity, ω, of the 2 nd motor turning submodel at the sampling time of k' -1₁(k') represents the turning angular velocity, ω, of the 1 st motor-driven turning sub-model at the sampling time k₁(k '-1) represents the turning angular velocity, ω, of the 1 st maneuver turning submodel at the sampling time of k' -1₃(k') represents the turning angular velocity, ω, of the 3 rd motor turning sub-model at the sampling time k₃(k '-1) represents the turning angular velocity of the 3 rd motor turning sub-model at the k' -1 sampling time;

(5b.2) setting max (p)_k′)＝μ₁(k′)，μ₁(k ') represents the probability of the 1 st motor-driven turning sub-model at the sampling time k'; if D is₁(k') is > M, then

ω₂(k′)＝ω₁(k′-1)，ω₁(k′)＝ω₂(k′)-G，ω₃(k′)＝ω₂(k') + G; otherwise, performing optimization probability approximation on the k' sampling moment interaction model set to obtain:

wherein D is₁(k ') denotes the distance function of the 1 st motor-driven turning sub-model at the sampling instant k', G denotes the grid spacing, ω₁(k '-1) represents the turning angular velocity of the 1 st maneuvering turning sub-model at the k' -1 sampling moment, and M is a maneuvering judgment threshold;

ω₂(k '-1) represents the turning angle of the 2 nd motor turning sub-model at the k' -1 sampling timeSpeed, ω₃(k '-1) represents the turning angular velocity, μ, of the 3 rd motor turning sub-model at the sampling time of k' -1₂(k') represents the probability, μ, of the 2 nd motor-driven turning sub-model at the sampling time k₃(k ') represents the probability of the 3 rd motor-driven turning sub-model at the sampling moment of k';

wherein D is₃(k') represents the distance function, ω, of the 3 rd motorised turning sub-model at the sampling instant k₃(k '-1) represents the turning angular velocity, ω, of the 3 rd motor turning submodel at the sampling time of k' -1₁(k '-1) represents the turning angular velocity, ω, of the 1 st maneuver turning submodel at the sampling time of k' -1₂(k '-1) represents the turning angular velocity, μ, of the 2 nd motor turning submodel at the sampling time of k' -1₂(k ') represents the probability of the 2 nd motor-driven turning sub-model at the sampling time k'.