CN114966525A - Target orientation estimation method based on artificial intelligence smart city sensor array - Google Patents

Abstract

The invention belongs to the technical field of artificial intelligence positioning processing, and in particular relates to a target orientation estimation method based on an artificial intelligence smart city sensor array. The invention includes: (1) building a uniform array with urban sensors, and confirming the data array L received by the sensor array; (2) constructing the prior distribution of each variable in the data array, and performing hierarchical prior distribution on the data array signals. The invention provides a target orientation estimation method based on an artificial intelligence smart city sensor array, which fully considers the influence of various noises, uses the distribution characteristics of multiple noises controlled by Bernoulli distribution, and is closer to the actual smart city environment through the Bayesian model. , to achieve accurate azimuth estimation, with ultra-high discrimination ability and multi-target resolution ability, more suitable for wave arrival azimuth estimation in complex urban environments.

Description

Target orientation estimation method based on artificial intelligence smart city sensor array

technical field

The invention belongs to the technical field of artificial intelligence positioning processing, and in particular relates to a target orientation estimation method based on an artificial intelligence smart city sensor array.

Background technique

Due to the complex and changeable environmental impact of smart cities, the sensor operation will encounter interference such as crowds, heavy rain, strong wind, and traffic flow, and sometimes there will be impulse noise. The complex noise environment reduces the direction finding accuracy of traditional methods and cannot handle many the simultaneous influence of various noises. In urban environment detection, a variety of sensors are used to form an array to observe or receive positioning signals to complete the azimuth estimation of the target. The traditional method is not enough to meet the high-precision requirements. It already has high-resolution orientation estimation, and also requires the number of sources as a priori information.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a target orientation estimation method based on artificial intelligence smart city sensor array under the joint influence of multiple noises.

The object of the present invention is achieved in this way:

The target orientation estimation method based on artificial intelligence smart city sensor array includes the following steps:

(1) Build a uniform array with urban sensors, and confirm the data array L received by the sensor array;

(1.1) Measure the length Y of the received signal of the array, and the number of grids A traversing the azimuth space;

(1.2) Constructing the sensor array flow pattern matrix

(1.3) Collect the expected signal matrix K after sensor detection;

(1.4) Collect the pulse noise occurrence state matrix M of each channel of the sensor array;

(1.5) Collect the non-uniform noise matrix I of each element of the sensor array;

(1.6) Collect the impulse noise matrix R at each element of the sensor array;

(1.7) Confirm the data array received by the sensor array;

1 _Y×Z is the Y×Z dimension unit matrix; Y×Z is the dimension of the data array;

(2) Constructing the prior distribution of each variable in the data array, and performing hierarchical prior distribution on the data array signal;

(2.1) Construct the system truth control impulse noise U _y and the hidden variable matrix τ _y at the y-th time, and collect the z-th array element to control the impulse noise U _z,y and the hidden variable matrix τ _z,y at the y-th time;

(2.2) Detect the non-uniform noise variance vector ο of the artificial intelligence system array and the non-uniform noise variance vector ο z detected by the zth array element ο _z , shape parameter t _z , inverse scale parameter u _z , manifold matrix

(2.3) When the a-th scanning azimuth in the array is acquired, the variance matrix ξ _a of the expected signal;

(2.4) collect the data ly received by the array at the yth moment, and the data _{lz, y received} by the zth array element at the yth moment and the data ky transmitted at the _yth moment; _ly constitutes a data matrix L;

(2.5) Collect the state vector zy of the noise at the _yth moment;

(2.6) Collect the impulse noise of the zth array element at the yth moment m _{z, y} ;

(2.7) Construct the hierarchical prior distribution of the signal received by the sensor array at the yth moment:

CN stands for complex Gaussian distribution;

(2.8) Construct the hierarchical gamma distribution of the non-uniform noise variance vector ο:

G stands for gamma distribution;

(2.9) Construct a hierarchical Gamma distribution for the variance matrix ξ _a of the expected signal, the latent variable matrix τ _y at the y-th moment, and the control impulse noise U _y respectively:

Among them, na , σ _{z ,y} /3, π _z,y , p _z,y are the shape parameters of the corresponding distribution, o _a , q _z,y , δ _z,y are the inverse scale parameters of the corresponding distribution, σ ₁ is the variance vector of the element τ _y of the constrained latent variable matrix;

(2.10) Constructing the Bernoulli distribution for the noise state vector z _y at the y-th moment is:

γ _y is the probability vector of _zy occurrence;

(2.11) Construct a hierarchical Beta distribution for γ _y :

c _z,y and d _z,y are the Beta distribution parameters obeyed by the zth array element at the yth time, respectively.

(2.12) Find the posterior probability of each variable:

(2.13) Substitute the distribution matrix constructed by the steps into the following formula to solve the posterior probability of each variable:

lnq(K)=<lnp(L|K,M,U,τ,ο)+lnp(L|ξ)> _q(ν≠L) +CONST

lnq(ξ)=<lnp(L|ξ)+lnp(ξ)> _q(ν≠L) +CONST

lnq(U)=<lnp(L|K,M,U,τ,ο)+lnp(U)> _q(ν≠L) +CONST

lnq(K)=<lnp(L|K,M,U,τ,ο)+lnp(τ|α)> _q(ν≠L) +CONST

lnq(ο)=<lnp(L|K,M,U,τ,ο)+lnp(ο)> _q(ν≠L) +CONST

lnq(α)=<lnp(τ|α)+lnp(α)> _q(ν≠L) +CONST

lnq(M)=<lnp(L|K,M,U,τ,ο)+lnp(M|γ)> _q(ν≠L) +CONST

lnq(γ)=<lnp(M|γ)+lnp(γ)> _q(ν≠L) +CONST

ν=(K,ξ,U,τ,α,ο,M,γ)

Among them, q() is the posterior probability of the variable, ln is the logarithm, <> represents the expectation, p(∣) represents the probability of the element, q(ν≠L) is the part of the set that does not contain variables. calculation, CONST is a constant term;

(3) According to the probability distribution of the variable in step (2), calculate the mean and variance of the system variable;

(3.1) Parameter initialization:

Set the initial iteration to 1, initialize the maximum number of iterations, the number of grids to traverse the azimuth space A, the shape parameter n ₀ of the expected signal distribution variance, the inverse scale parameter ο ₀ of the expected signal distribution variance, and the shape parameters p ₀ , t of the noise variance distribution ₀ , π ₀ , the inverse scale parameters q ₀ , u ₀ , δ ₀ of the noise variance distribution, control the occurrence probability c ₀ , d ₀ ;

和均值

(3.2) Update the variance of the expected signal

and mean

Q _Δ =diag(zy ·ο _y +(1 _Y×Z + _{z y} )·τ _y ·U _y )

Diag is a diagonal operation;

(3.3) Update each distribution parameter:

(3.4) Update the mean of each variable

(4) Update the number of iterations plus 1; judge whether the iteration termination condition is met, when the iteration termination condition is met, jump out of the iteration and output

Toul is the termination threshold, ξ is the expected signal variance, and te is the number of iterations; if the iteration termination conditions are not met, continue with steps (3.2)-(3.4);

(5) Carry out azimuth estimation and output the azimuth estimation result:

Among them, ‖·‖ ₁ , ‖·‖ _∞ are the infinite norm operation on the matrix.

The beneficial effects of the present invention are:

The invention provides a target orientation estimation method based on an artificial intelligence smart city sensor array, which fully considers the influence of various noises, utilizes the distribution characteristics of multiple noises controlled by Bernoulli distribution, and is closer to the actual smart city environment through the Bayesian model. , to achieve accurate azimuth estimation, with super high discrimination ability and multi-target resolution ability, more suitable for wave arrival azimuth estimation in complex urban environments.

Description of drawings

Fig. 1 is the flow chart of orientation estimation of the present invention;

Fig. 2 is a model diagram constructed by the present invention;

3 is a comparison result of root mean square error between the present invention and the traditional sparse method and the sparse method based on impulse noise;

FIG. 4 shows the detection probability results of the present invention, the traditional sparse method, and the impulsive noise-based sparse method.

Detailed ways

The present invention will be described in further detail below with reference to the accompanying drawings.

Combined with Figure 1 and Figure 2, the target orientation estimation method based on artificial intelligence smart city sensor array includes the following steps:

(1) Build a uniform array with urban sensors, and confirm the data array L received by the sensor array;

(1.1) Measure the length Y of the received signal of the array, and the number of grids A traversing the azimuth space;

(1.2) Constructing the sensor array flow pattern matrix

(1.3) Collect the expected signal matrix K after sensor detection;

(1.4) Collect the pulse noise occurrence state matrix M of each channel of the sensor array;

(1.5) Collect the non-uniform noise matrix I of each element of the sensor array;

(1.6) Collect the impulse noise matrix R at each element of the sensor array;

(1.7) Confirm the data array received by the sensor array;

1 _Y×Z is the Y×Z dimension unit matrix; Y×Z is the dimension of the data array;

(2) Constructing the prior distribution of each variable in the data array, and performing hierarchical prior distribution on the data array signal;

(2.1) Construct the system truth control impulse noise U _y and the hidden variable matrix τ _y at the y-th time, and collect the z-th array element to control the impulse noise U _z,y and the hidden variable matrix τ _z,y at the y-th time;

(2.2) Detect the non-uniform noise variance vector ο of the artificial intelligence system array and the non-uniform noise variance vector ο z detected by the zth array element ο _z , shape parameter t _z , inverse scale parameter u _z , manifold matrix

(2.3) When the a-th scanning azimuth in the array is acquired, the variance matrix ξ _a of the expected signal;

(2.4) collect the data ly received by the array at the yth moment, and the data _{lz, y received} by the zth array element at the yth moment and the data ky transmitted at the _yth moment; _ly constitutes a data matrix L;

(2.5) Collect the state vector zy of the noise at the _yth moment;

(2.6) Collect the impulse noise of the zth array element at the yth moment m _{z, y} ;

(2.7) Construct the hierarchical prior distribution of the signal received by the sensor array at the yth moment:

CN stands for complex Gaussian distribution;

(2.8) Construct the hierarchical gamma distribution of the non-uniform noise variance vector ο:

G stands for gamma distribution;

(2.9) Construct a hierarchical Gamma distribution for the variance matrix ξ _a of the expected signal, the latent variable matrix τ _y at the y-th moment, and the control impulse noise U _y respectively:

Among them, na , σ _{z ,y} /3, π _z,y , p _z,y are the shape parameters of the corresponding distribution, o _a , q _z,y , δ _z,y are the inverse scale parameters of the corresponding distribution, σ ₁ is the variance vector of the element τ _y of the constrained latent variable matrix;

(2.10) Constructing the Bernoulli distribution for the noise state vector z _y at the y-th moment is:

γ _y is the probability vector of _zy occurrence;

(2.11) Construct a hierarchical Beta distribution for γ _y :

c _z,y and d _z,y are the Beta distribution parameters obeyed by the zth array element at the yth time, respectively.

(2.12) Find the posterior probability of each variable:

(2.13) Substitute the distribution matrix constructed by the steps into the following formula to solve the posterior probability of each variable:

lnq(K)=<lnp(L|K,M,U,τ,ο)+lnp(L|ξ)> _q(ν≠L) +CONST

lnq(ξ)=<lnp(L|ξ)+lnp(ξ)> _q(ν≠L) +CONST

lnq(U)=<lnp(L|K,M,U,τ,ο)+lnp(U)> _q(ν≠L) +CONST

lnq(K)=<lnp(L|K,M,U,τ,ο)+lnp(τ|α)> _q(ν≠L) +CONST

lnq(ο)=<lnp(L|K,M,U,τ,ο)+lnp(ο)> _q(ν≠L) +CONST

lnq(α)=<lnp(τ|α)+lnp(α)> _q(ν≠L) +CONST

lnq(M)=<lnp(L|K,M,U,τ,ο)+lnp(M|γ)> _q(ν≠L) +CONST

lnq(γ)=<lnp(M|γ)+lnp(γ)> _q(ν≠L) +CONST

ν=(K,ξ,U,τ,α,ο,M,γ)

Among them, q() is the posterior probability of the variable, ln is the logarithm, <> represents the expectation, p(∣) represents the probability of the element, q(ν≠L) is the part of the set that does not contain variables. calculation, CONST is a constant term;

(3) According to the probability distribution of the variable in step (2), calculate the mean and variance of the system variable;

(3.1) Parameter initialization:

Set the initial iteration to 1, initialize the maximum number of iterations, the number of grids to traverse the azimuth space A, the shape parameter n ₀ of the expected signal distribution variance, the inverse scale parameter ο ₀ of the expected signal distribution variance, and the shape parameters p ₀ , t of the noise variance distribution ₀ , π ₀ , the inverse scale parameters q ₀ , u ₀ , δ ₀ of the noise variance distribution, control the occurrence probability c ₀ , d ₀ ;

和均值

(3.2) Update the variance of the expected signal

and mean

Q _Δ =diag(zy ·ο _y +(1 _Y×Z + _{z y} )·τ _y ·U _y )

Diag is a diagonal operation;

(3.3) Update each distribution parameter:

(3.4) Update the mean of each variable

(4) Update the number of iterations plus 1; judge whether the iteration termination condition is met, when the iteration termination condition is met, jump out of the iteration and output

Toul is the termination threshold, ξ is the expected signal variance, and te is the number of iterations; if the iteration termination conditions are not met, continue with steps (3.2)-(3.4);

(5) Carry out azimuth estimation and output the azimuth estimation result:

Among them, ‖·‖ ₁ , ‖·‖ _∞ are the infinite norm operation on the matrix.

The distinguishing feature of the present invention is that the impulse noise and its influence are fully considered, the constructed positioning model is more in line with the actual noisy environment of the smart city, and better estimation results can be obtained.

Example 1, the following data simulation is performed, the single-frequency pulse signal matrix is used as the incident signal, the incident azimuth is -45° and -30°, and the noise variance of each array element varies randomly between [0.2, 7], let The generalized signal-to-noise ratio varies in the interval [-15, 30], and the traditional sparse method, the impulsive noise-based sparse method and the method proposed by the present invention are compared and analyzed.

Figure 3 shows the variation curve of the root mean square error of each method when the GSNR changes in the impulse noise environment. Through comparison, it can be found that the traditional sparse method fails more seriously; although the sparse method based on impulse noise also has a downward trend, when a variety of noises appear alternately, there is a jump trend, which will lead to unstable estimation results; the method of the present invention RMSE The lowest and the most stable, the best estimation result among the three methods is obtained, and the deviation is the smallest.

Figure 4 shows the detection success probability curves of the three methods when the GSNR changes in the impulse noise environment, and the detection success is defined as the target deviation within 0.5°. By comparison, it can be seen that the traditional sparse method cannot accurately find the target orientation in time under the background of impulse noise and non-uniform noise; the probability of successful detection of the sparse method based on impulse noise increases, but the estimation result is unstable; the method of the present invention successfully estimates the target the highest level. Therefore, the simulation experiment also fully verifies the effectiveness and feasibility of the present invention.

In summary, the present invention provides a target orientation estimation method based on an artificial intelligence smart city sensor array under the joint influence of multiple noises, which realizes high-precision DOA estimation in the case of a mixture of non-uniform noise and impulse noise, and is more in line with practical application scenarios. . Compared with the existing similar methods, the high-precision DOA estimation method under the mixed condition of impulse noise has higher accuracy and stronger adaptability.

Claims (6)

1. The target orientation estimation method based on artificial intelligence smart city sensor array, is characterized in that, comprises the steps: (1) Build a uniform array with urban sensors, and confirm the data array L received by the sensor array; (2) Constructing the prior distribution of each variable in the data array, and performing hierarchical prior distribution on the data array signal; (3) According to the probability distribution of the variable in step (2), calculate the mean and variance of the system variable; (4) Update the number of iterations plus 1; judge whether the iteration termination condition is met, when the iteration termination condition is met, jump out of the iteration and output

If the iteration termination condition is not met, go to step (3); (5) Carry out azimuth estimation, and output the azimuth estimation result. 2. The target orientation estimation method based on artificial intelligence smart city sensor array according to claim 1, is characterized in that, described step (1) comprises: (1.1) Measure the length Y of the received signal of the array, and the number of grids A traversing the azimuth space; (1.2) Constructing the sensor array flow pattern matrix

(1.3) Collect the expected signal matrix K after sensor detection; (1.4) Collect the pulse noise occurrence state matrix M of each channel of the sensor array; (1.5) Collect the non-uniform noise matrix I of each element of the sensor array; (1.6) Collect the impulse noise matrix R at each element of the sensor array; (1.7) Confirm the data array received by the sensor array;

1 _Y×Z is the unit matrix of Y×Z dimension; Y×Z is the dimension of the data array. 3. The target orientation estimation method based on artificial intelligence smart city sensor array according to claim 1, is characterized in that, described step (2) comprises: (2.1) Construct the system truth control impulse noise U _y and the hidden variable matrix τ _y at the y-th time, and collect the z-th array element to control the impulse noise U _z,y and the hidden variable matrix τ _z,y at the y-th time; (2.2) Detect the non-uniform noise variance vector ο of the artificial intelligence system array and the non-uniform noise variance vector ο z detected by the zth array element ο _z , shape parameter t _z , inverse scale parameter u _z , manifold matrix

(2.3) When the a-th scanning azimuth in the array is acquired, the variance matrix ξ _a of the expected signal; (2.4) collect the data ly received by the array at the yth moment, and the data _{lz, y received} by the zth array element at the yth moment and the data ky transmitted at the _yth moment; _ly constitutes a data matrix L; (2.5) Collect the state vector zy of the noise at the _yth moment; (2.6) Collect the impulse noise of the zth array element at the yth moment m _{z, y} ; (2.7) Construct the hierarchical prior distribution of the signal received by the sensor array at the yth moment:

CN stands for complex Gaussian distribution; (2.8) Construct the hierarchical gamma distribution of the non-uniform noise variance vector ο:

G stands for gamma distribution; (2.9) Construct a hierarchical Gamma distribution for the variance matrix ξ _a of the expected signal, the latent variable matrix τ _y at the y-th moment, and the control impulse noise U _y respectively:

Among them, na , σ _{z ,y} /3, π _z,y , p _z,y are the shape parameters of the corresponding distribution, o _a , q _z,y , δ _z,y are the inverse scale parameters of the corresponding distribution, σ ₁ is the variance vector of the element τ _y of the constrained latent variable matrix; (2.10) Constructing the Bernoulli distribution for the noise state vector z _y at the y-th moment is:

γ _y is the probability vector of _zy occurrence; (2.11) Construct a hierarchical Beta distribution for γ _y :

c _z,y and d _z,y are the Beta distribution parameters obeyed by the zth array element at the yth time, respectively. (2.12) Find the posterior probability of each variable:

(2.13) Substitute the distribution matrix constructed by the steps into the following formula to solve the posterior probability of each variable: Inq(K)=<lnp(L|K, M, U, τ, o)+lnp(L|ξ)> _q(v≠L) +CONST lnq(ξ)=<lnp(L|ξ)+lnp(ξ)> _q(v≠L) +CONST lnq(U)=<lnp(L|K, M, U, τ, o)+lnp(U)> _q(v≠L) +CONST lnq(K)=<lnp(L|K, M, U, τ, o)+lnp(τ|α)> _q(v≠L) +CONST lnq(o)=<lnp(L|K, M, U, τ, o)+lnp(o)> _q(v≠L) +CONST lnq(α)=<lnp(τ|α)+lnp(α)> _q(v≠L) +CONST lnq(M)=<lnp(L|K, M, U, τ, o)+lnp(M|γ)> _q(v≠L) +CONST lnq(γ)=<lnp(M|γ)+lnp(γ)> _q(v≠L) +CONST v=(K,ξ,U,τ,α,o,M,γ) Among them, q() is the posterior probability of the variable, ln is the logarithm, <> represents the expectation, p(|) represents the probability of the element, and q(v≠L) is the part of the set that does not contain variables. calculation, CONST is a constant term; 4. The target orientation estimation method based on artificial intelligence smart city sensor array according to claim 1, is characterized in that, described step (3) comprises: (3.1) Parameter initialization: Set the initial iteration to 1, initialize the maximum number of iterations, the number of grids to traverse the azimuth space A, the shape parameter n ₀ of the expected signal distribution variance, the inverse scale parameter o ₀ of the expected signal distribution variance, and the shape parameters p ₀ , t of the noise variance distribution ₀ , π ₀ , the inverse scale parameters q ₀ , u ₀ , δ ₀ of the noise variance distribution, control the occurrence probability c ₀ , d ₀ ; (3.2) Update the variance of the expected signal

and mean

Q _Δ =diag( _{zy·o y +} (1 _Y×Z + _zy )·τ _y ·U _y ) Diag is a diagonal operation; (3.3) Update each distribution parameter:

(3.4) Update the mean of each variable

5. The target orientation estimation method based on artificial intelligence smart city sensor array according to claim 1, is characterized in that, described step (4) comprises: Determine whether the iteration termination condition is met, and when the iteration termination condition is met, jump out of the iteration and output

Toul is the termination threshold, ξ is the expected signal variance, and te is the number of iterations; if the iteration termination conditions are not met, continue with steps (3.2)-(3.4). 6. The target orientation estimation method based on artificial intelligence smart city sensor array according to claim 1, is characterized in that, described step (5) orientation estimation result is:

Among them, ||·|| ₁ , ||·|| _∞ are the infinite norm operation on the matrix.

Images