A multi-parameter fusion positioning optimization method and system based on a genetic algorithm

Abstract

The present application relates to target positioning technical field, especially to a kind of multi-parameter fusion positioning optimization method and system based on genetic algorithm, the method, including obtaining multi-parameter observation data, based on positioning area sound velocity profile data, establish the propagation path model of constant gradient sound ray correction, construct time delay prediction model by the output of propagation path model;The output result of time delay prediction model and other parameter observation data are used to calculate multi-parameter error vector, utilize maximum likelihood estimation criterion to construct multi-parameter fusion positioning optimization model, establish genetic algorithm and quasi-newton iteration method collaborative constraint mechanism, give consideration to global search capability and local convergence precision, output the optimal estimate of target position, the present application combines the global optimization ability of genetic algorithm, effectively overcome the defect that traditional algorithm is easy to fall into local optimum, ensure that the global optimality and physical rationality of positioning result.

Description

Technical Field

[0001] This invention relates to the field of target localization technology, and in particular to a multi-parameter fusion localization optimization method and system based on genetic algorithms. Background Technology

[0002] Underwater target positioning technology is a core supporting technology in fields such as marine engineering, underwater navigation, and marine resource exploration. Its positioning accuracy directly determines the safety and effectiveness of underwater operations. Due to its advantages of compact structure, flexible deployment, and controllable cost, short baseline positioning system has become the mainstream solution for three-dimensional positioning of underwater targets.

[0003] However, existing short baseline positioning methods do not fully utilize complementary information from multiple parameters, have weak anti-interference capabilities, and the ocean channel is a random time-varying and space-varying channel. Acoustic signal propagation is susceptible to multipath effects, non-uniform sound speed, and environmental noise interference. When relying solely on the time delay parameter, the multipath effect causes the aliasing of multipath signals and direct signals, and the time delay estimation error can reach the millisecond level. Moreover, changes in sound speed with temperature, salinity, and depth can destroy the straight-line propagation assumption, causing deviations in time delay and distance conversion. When relying solely on azimuth or elevation angle parameters, the curvature of the sound speed field causes the incident angle to deviate from the true direction. The false signals generated by the multipath effect further interfere with the angle estimation. In particular, the vertical direction is prone to depth estimation deviations due to the lack of effective constraints, making it impossible to achieve accurate three-dimensional positioning. A single parameter is highly sensitive to environmental interference, and once the measurement is wrong, it will lead to positioning distortion, resulting in extremely poor model robustness.

[0004] Furthermore, in existing technologies, optimization methods based on genetic algorithms have demonstrated advantages in solving complex nonlinear problems. For example, Chinese patent CN201510677240.X discloses an indoor underwater target localization method based on genetic algorithms. This method uses the range of the water area as a constraint and optimizes the objective function using genetic algorithms to achieve underwater target localization. However, the constraint range of this method is very wide, and it only achieves two-dimensional target localization. The paper "Research on Distance Ambiguity Resolution Method for High Frame Rate Underwater Acoustic Synchronous Localization" proposes a time-delay azimuth fusion localization optimization model under the maximum likelihood criterion, but it is still limited to a two-dimensional plane. None of these methods effectively solve the multi-parameter error coupling problem in three-dimensional space. In particular, the localization accuracy in the depth direction needs to be improved. At present, a multi-parameter fusion localization optimization method and system based on genetic algorithms is needed. Summary of the Invention

[0005] To address the challenge of achieving both high accuracy and robustness in short baseline positioning within complex underwater environments, this invention provides a multi-parameter fusion positioning optimization method and system based on a genetic algorithm.

[0006] Firstly, the present invention provides a multi-parameter fusion localization optimization method based on a genetic algorithm, which adopts the following technical solution: A multi-parameter fusion localization optimization method based on genetic algorithm includes: Acquire the underwater target radiated acoustic signals received by each element of the short baseline array, and use signal processing methods to extract the time delay, azimuth angle and elevation angle between the array elements and the underwater target to obtain multi-parameter observation data; Based on the sound velocity profile data of the local area, a propagation path model with constant gradient sound ray correction is established. The corrected propagation path length is output by the propagation path model, and a time delay prediction model is constructed based on the corrected propagation path length. The error vector is obtained by calculating the error between the multi-parameter observation data and the prediction parameter data output by the time delay prediction model. A multi-parameter fusion localization optimization model is constructed using the maximum likelihood estimation criterion, including calculating the corresponding error covariance matrix based on the multi-parameter error vector, and adaptively weighting the errors of each parameter using the error covariance matrix. The maximum likelihood function of the multi-parameter fusion localization optimization model is used as the fitness function. The optimal estimated value of the target position is output by iteratively solving the problem through a genetic algorithm.

[0007] Furthermore, obtaining multi-parameter observation data includes preprocessing the acoustic signals received by each array element, removing environmental noise through an adaptive filtering algorithm, eliminating array element clock deviations using signal synchronization, calculating the cross-correlation function of the signals received by different array elements, determining the time delay by locating the peak position of the cross-correlation function, constructing a spatial geometric model using the known coordinates of each array element of the short baseline array, employing a near-field MVDR focusing beamforming algorithm based on the spatial geometric model, obtaining target coordinate estimates through spectral peak search within a three-dimensional search area, and calculating azimuth and elevation angle observations based on the geometric relationship between the target coordinate estimates and the known coordinates of each array element.

[0008] Furthermore, the step of constructing a delay prediction model based on the corrected propagation path length includes establishing a delay measurement equation for N array elements in the system: , in, This is the predicted time delay from the target to the i-th element, calculated based on the corrected propagation path length. Let be the actual bending path length from the target to the i-th element, calculated by the constant gradient acoustic ray correction model. For the equivalent underwater speed of sound, It follows a Gaussian distribution. The time delay measurement noise, The variance is the normal distribution. The time delay measurement equations for each array element are then stacked into a vector form: , in, For the TOA observation vector, For nonlinear observation functions, Let the measurement noise vector follow a zero-mean Gaussian distribution, and its covariance matrix be denoted as . .

[0009] Furthermore, the step of constructing a delay prediction model based on the corrected propagation path length also includes, in three-dimensional space, decomposing the azimuth vector of the target relative to the observation node into azimuth angles on the horizontal plane. Pitch angle in the vertical plane For the i-th sensor and the target to be measured Its angle measurement equation is: , , in, and These are the measurement errors for azimuth and elevation angles, respectively, and both follow a Gaussian distribution. Let be the two-dimensional coordinates of the underwater target to be measured in the horizontal plane. Let i be the known two-dimensional coordinates of the i-th array element in the horizontal plane. This is the corrected propagation path length; The angle measurement equations of each array element are stacked into observation vector form to obtain the azimuth observation vector. and pitch angle observation vector : , , in, For azimuth angle nonlinear observation function, For azimuth measurement noise vector, For pitch angle nonlinear observation function, The noise vector is used to measure the pitch angle.

[0010] Furthermore, the step of calculating the error between the multi-parameter observation data and the prediction parameter data output by the time delay prediction model includes obtaining the azimuth and elevation prediction values, combining them with the time delay prediction values ​​output by the time delay prediction model to form prediction parameter data, calculating the difference between the observation value and the corresponding prediction value for each type of parameter based on the prediction parameter data, and arranging the three types of differences in vector form to obtain a multi-parameter error vector containing time delay error, azimuth error, and elevation error. The expression for the multi-parameter error vector is as follows: , in, Let i be the multi-parameter error vector corresponding to the i-th array element. , and These represent the time delay observation, azimuth observation, and elevation observation for the i-th array element, respectively. The three-dimensional coordinates of the underwater target , and These are the time delay prediction value, azimuth prediction value, and elevation prediction value, respectively.

[0011] Furthermore, the construction of a multi-parameter fusion localization optimization model using the maximum likelihood estimation criterion includes: Calculate the error covariance matrix based on the multi-parameter error vector, and determine the joint error vector. The joint error vector follows a matrix with zero mean and a dense covariance matrix. The multivariate Gaussian distribution, i.e. ; Based on the maximum likelihood estimation criterion, the inverse of the dense covariance matrix is ​​used as the adaptive weight of the multi-parameter model to construct a multi-parameter fusion localization optimization model. The optimal estimation of the target coordinates is achieved by minimizing the objective function, which is expressed as follows: , Maximum likelihood estimation is equivalent to minimizing the high-dimensional Mahalanobis distance of the residual vector: , in, Let Z be the likelihood function of the globally joint observation column vector given the target position p. It is a dense covariance matrix. Let N be the inverse of the dense covariance matrix, and N be the number of elements. For the global joint observation column vector of the system, This is a joint prediction vector composed of the time delay prediction vector, azimuth prediction vector, and elevation prediction vector derived from the target position p. This is the optimal estimate of the target location obtained based on the maximum likelihood estimation criterion.

[0012] Furthermore, the optimal estimate of the output target location includes taking the negative of the objective function of the maximum likelihood estimation as the fitness function of the genetic algorithm, and selecting the individual with the best fitness as the optimal estimate of the target location through multiple generations of iteration of the genetic algorithm combined with the quasi-Newton method for fine optimization. The expression of the fitness function is: , in, Here is the fitness function of the genetic algorithm. This is the objective function for the original maximum likelihood estimation.

[0013] Furthermore, the optimal estimate of the output target position also includes obtaining the true position coordinates of the underwater target, calculating the Euclidean distance between the optimal estimate and the true position, and combining this with the Euclidean distance from the true position to the origin to obtain the three-dimensional positioning relative error. The positioning accuracy is then quantitatively evaluated based on this relative error. The expression for the three-dimensional positioning relative error is as follows: , in, These are the coordinates of the optimal estimated target location obtained through a genetic algorithm. These are the actual position coordinates of the underwater target.

[0014] Secondly, a multi-parameter fusion localization optimization system based on a genetic algorithm includes: The data acquisition module is configured to: acquire the underwater target radiated acoustic signals received by each element of the short baseline array, and use signal processing methods to extract the time delay, azimuth angle and elevation angle between the array elements and the underwater target to obtain multi-parameter observation data; The preprocessing module is configured to: establish a propagation path model with constant gradient sound ray correction based on the sound velocity profile data of the positioning area, and construct a time delay prediction model by combining multi-parameter observation data and the output of the propagation path model; The conversion module is configured to: perform error calculation between the multi-parameter observation data and the prediction parameter data output by the time delay prediction model to obtain a multi-parameter error vector; The model module is configured to: construct a multi-parameter fusion localization optimization model using the maximum likelihood estimation criterion; The feature extraction module is configured to: establish a collaborative constraint mechanism between the genetic algorithm and the quasi-Newton algorithm; The transformation module is configured to use a multi-parameter fusion localization optimization model as the fitness function, perform iterative solutions through a genetic algorithm, and output the optimal estimate of the target position.

[0015] In summary, the present invention has the following beneficial technical effects: 1. This invention breaks through the limitations of existing technologies that rely on single-parameter positioning by simultaneously extracting three types of observation parameters: time delay, azimuth, and elevation. It combines preprocessing methods such as adaptive filtering and noise reduction, and signal synchronization calibration with precise extraction techniques such as generalized cross-correlation time delay estimation algorithm and MUSIC algorithm to ensure the reliability of multi-parameter observation data. Furthermore, it achieves information complementarity through multi-parameter joint modeling, which significantly reduces the impact of single-parameter errors on positioning results and enhances the model's anti-interference ability and robustness in complex marine environments.

[0016] 2. This invention establishes a propagation path model with constant gradient sound ray correction, which fully considers the variation characteristics of underwater sound speed with depth, corrects the error caused by the curved propagation trajectory of the sound signal. The time delay prediction model constructed based on this propagation path model can better fit the real acoustic propagation law, providing a high-precision theoretical basis for subsequent error calculation and effectively improving the physical consistency of the positioning model.

[0017] 3. This invention quantifies the deviation between observed and predicted values ​​by calculating multi-parameter error vectors, and achieves adaptive weighting of each parameter error by combining the error covariance matrix. Based on the maximum likelihood estimation criterion, a multi-parameter fusion positioning optimization model is constructed, which gives greater weight to parameters with high measurement accuracy and adaptively reduces the weight of parameters with low accuracy. This achieves reasonable fusion of multiple parameters and precise error control, providing a scientific objective function for positioning optimization.

[0018] 4. This invention establishes a genetic algorithm collaborative constraint mechanism, extracts the spatial coupling relationship between multiple parameters to construct geometric consistency constraint terms, performs physical constraint verification and correction on the candidate solutions generated during the iteration process of the genetic algorithm, and uses the constrained objective function as the fitness function. Combined with the global optimization capability of the genetic algorithm, it effectively overcomes the defect of traditional algorithms being prone to local optima, and ensures the global optimality and physical rationality of the positioning results.

[0019] 5. This invention uses a genetic algorithm combined with the quasi-Newton method to iteratively solve a multi-parameter fusion positioning optimization model, and introduces a three-dimensional positioning relative error quantification evaluation mechanism, so that the relative error of underwater target three-dimensional positioning is less than one-thousandth of the slant range, which significantly improves the positioning accuracy. In particular, in the vertical depth direction, the problem of inaccurate underwater target positioning is solved by pitch angle constraints and acoustic correction. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall process of a multi-parameter fusion localization optimization method based on genetic algorithm according to an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the genetic algorithm processing in an embodiment of the present invention.

[0022] Figure 3 This is a flowchart of the global coarse estimation stage of the genetic algorithm in an embodiment of the present invention.

[0023] Figure 4 This is a distribution map of the 3D spatial positioning results of an embodiment of the present invention.

[0024] Figure 5 These are box plots of the algorithm errors in various embodiments of the present invention.

[0025] Figure 6This is a comparison chart of the implementation effects of various algorithms in this invention embodiment; wherein, Figure 6 (a) is a graph of the cumulative error distribution function. Figure 6 (b) is the XY plane positioning offset projection diagram.

[0026] Figure 7 This is a comparison chart of the average absolute errors of various algorithms in the embodiments of the present invention.

[0027] Figure 8 This is a comparison chart of the average relative positioning accuracy of various algorithms in the embodiments of the present invention. Detailed Implementation

[0028] The present invention will be further described in detail below with reference to the accompanying drawings.

[0029] Example 1 Reference Figure 1 This embodiment of a multi-parameter fusion localization optimization method based on a genetic algorithm includes: S1. Acquire multi-parameter observation data, including extracting the time delay, azimuth angle and elevation angle between the array elements and the underwater target based on the underwater target radiated acoustic signals received by each element of the short baseline array using signal processing techniques; S2. Based on the sound velocity profile data of the positioning area, establish a propagation path model with constant gradient sound ray correction, and combine it with multi-parameter observation data to construct a time delay prediction model through the output of the propagation path model. S3. Calculate the multi-parameter error vector using the output of the time delay prediction model, including calculating the deviation between the multi-parameter observation data and the prediction parameter data; S4. Construct a multi-parameter fusion positioning optimization model using the maximum likelihood estimation criterion, including calculating the corresponding error covariance matrix based on the multi-parameter error direction, and using the error covariance matrix to adaptively weight the errors of each parameter. S5. Using a multi-parameter fusion positioning optimization model as the fitness function, the optimal estimated value of the target position is output through iterative solution using a genetic algorithm and a quasi-Newton method.

[0030] Specifically, a multi-parameter fusion localization optimization method based on joint optimization of genetic algorithm and quasi-Newton iterative method includes the following steps: S1. Acquire multi-parameter observation data, including extracting the time delay, azimuth angle and elevation angle between the array elements and the underwater target based on the echo acoustic signals received by each element of the short baseline array from the underwater target using signal processing technology. like Figure 1 , Figure 2 As shown, this step is for a short baseline positioning system (consisting of at least 3 array elements, with the array element coordinates pre-calibrated and stored). ,in, Let be the coordinate of the i-th element along the x-axis. Let be the coordinate of the i-th element along the y-axis. Let be the coordinate of the i-th element in the z-axis direction. Let be the three-dimensional spatial coordinate vector of the i-th array element. Through signal preprocessing, parameter extraction algorithms, and geometric transformation, the observation data of underwater target time delay, azimuth angle, and elevation angle are obtained. The specific implementation process is as follows: First, the raw acoustic signals received by each array element are preprocessed. In a short-baseline array, each element synchronously receives acoustic signals radiated by underwater targets, such as actively radiated pulse acoustic signals or passively radiated continuous acoustic signals. The raw acoustic signals received by each element are preprocessed as follows: an adaptive filtering algorithm is used to filter the signals. The filter dynamically suppresses marine environmental noise, such as ocean current noise, biological noise, and reflection interference signals caused by multipath effects, by adjusting the weights in real time, while retaining the direct acoustic signal components. Simultaneously, clock synchronization technology is employed. In this embodiment, a GPS-based time synchronization module is used to align the local clock of each array element with the system reference clock, eliminating signal time difference errors caused by clock deviations between array elements and ensuring the consistency of the time reference for signals received by different array elements. The preprocessed acoustic signal is denoted as... Where t is the time variable and i is the array element number. N represents the total number of array elements.

[0031] Then, based on the preprocessed acoustic signal The time delay between each array element and the underwater target is extracted using a generalized cross-correlation time delay estimation algorithm. Using the target sound source's transmitted signal s(t) as the reference signal, calculate the difference between the reference signal and the received signal of the i-th array element. Cross-correlation function: , in, The time difference to be estimated As a time variable, the cross-correlation function is located using a peak detection algorithm. The time difference corresponding to the maximum value This time difference is the time delay observation between the i-th array element and the reference array element. Based on the known coordinates of each element of the short baseline array Given the initial underwater sound speed c, the theoretical time delay range is estimated based on the geometric distance relationship between the target and the i-th array element. The extracted time delay observations are then analyzed. Perform validity checks, remove outliers that exceed the theoretical delay range, and retain valid delay observations.

[0032] Next, azimuth angle extraction is performed based on the preprocessed acoustic signal. and the known coordinates of each element of the short baseline array A spatial geometric model is constructed (with the horizontal plane of the positioning area as the xy plane and the vertical direction as the z-axis), and the target azimuth angle is determined using near-field MVDR focusing beamforming technology. and target pitch angle Specifically, this includes: constructing the covariance matrix of the array's received signals: , in, Let be the received signal vector of n array elements at the k-th sampling time. for The conjugate transpose of . Let H be the number of sampling points, and H denote the conjugate transpose. To improve the robustness of matrix inversion with a finite number of samples, a diagonal loading process is introduced to perform diagonal loading on the covariance matrix, resulting in the corrected covariance matrix: , in, This is the corrected covariance matrix after diagonal loading. This is an estimate of the covariance matrix of the array received signal. This is the diagonal loading amount, and its value is set based on the noise power level. It is an n×n dimensional identity matrix, with diagonal elements being 1 and off-diagonal elements being 0.

[0033] Considering that underwater targets with short baseline arrays are often located in the Fresnel zone, and wavefront curvature cannot be ignored, a precise spherical wavefront model is used to construct the steering vector: for spatial candidate points Define its distance to the i-th element. and reference array element distance Then the guiding vector The i-th element is , λ is the wavelength of the acoustic signal, and j is the imaginary unit. The MVDR spatial spectrum function is derived from... Given, among which, As the guide vector, in the three-dimensional search area determined by prior information The target coordinates are estimated by searching the spectral peaks. , in, The target's three-dimensional coordinates are estimated through spectral peak search. `arg max` is the operator that maximizes the objective function. This process does not require a pre-defined number of sound sources, and under single-source conditions, the estimate approximates the maximum likelihood solution. The target's horizontal coordinates are... The horizontal coordinates of the i-th element Calculate the azimuth angle, satisfying the formula In the formula Let be the target azimuth angle at the i-th array element, that is, the angle between the line connecting the target and the array element on the horizontal plane and the positive x-axis, ranging from . , It is the arctangent function in the fourth quadrant. The azimuth measurement error follows a zero-mean Gaussian distribution, and its variance is determined by the peak curvature of the MVDR spatial spectrum, the signal-to-noise ratio, and the geometric factors of the array elements. Consistency checks are performed on the azimuth observations of multiple array elements. The mean and standard deviation of each azimuth are calculated, outliers exceeding ±3 times the standard deviation are removed, and valid azimuth observations are retained. .

[0034] Pitch angle extraction: Also based on near-field MVDR focused beamforming technology and spatial geometry model, it directly uses the obtained target three-dimensional coordinates for estimation. Extract the target's elevation angle relative to each array element. The vertical height difference between the target and the i-th element is... The horizontal Euclidean distance is The elevation angle corresponding to each array element is obtained through geometric conversion, and its calculation satisfies the formula. In the formula Let be the elevation angle of the target at the i-th element, that is, the angle between the line connecting the target and the element on the vertical plane and the horizontal plane, ranging from . , The pitch angle measurement error follows a zero-mean Gaussian distribution, with variance determined by the vertical array aperture and the sharpness of the MVDR spectral peaks. The extracted pitch angle... A validity test was performed, outliers exceeding the mean ± 3 standard deviations were removed, and valid pitch angle observations were retained. .

[0035] S2. Based on the sound velocity profile data of the positioning area, establish a propagation path model with constant gradient sound ray correction, and combine it with multi-parameter observation data to construct a time delay prediction model through the output of the propagation path model. Before deploying the short baseline positioning system, underwater sound velocity sensors are used to detect sound velocity in the positioning area, collecting sound velocity data at different depths. The depth sampling interval is set to 0.5m~2m, adjusted according to the water depth of the positioning area: 0.5m interval for water depth ≤100m, and 2m interval for water depth >100m. The sampling depth range covers the depth of the short baseline array elements (usually 5m~20m underwater) to the maximum depth at which the underwater target may appear, acquiring at least 50 sets of depth-sound velocity data pairs. , Let the depth be the k-th sampling point. To determine the sound velocity at corresponding depths, a linear fitting algorithm was used to fit the collected depth-sound velocity data pairs. Based on the assumption of constant gradient sound ray, i.e., sound velocity varies linearly with depth, a sound velocity-depth distribution function was constructed. ,in, Let be the underwater sound speed at depth z. For the underwater sound speed at the reference depth, The sound velocity gradient is represented by this distribution function, which can reflect the changes in sound velocity at different depths within the positioning area in real time.

[0036] Then, based on the sound speed-depth distribution function Using multi-parameter observation data and combining the principles of ray acoustics, the bending path equation for sound signal propagation is derived, and the corrected actual propagation path length is calculated. A local coordinate system is established with any array element as the origin, where... The axis points horizontally towards the target azimuth. The direction, z' axis is along the vertical direction (consistent with the z-axis of the global coordinate system), and the initial coordinates of the target in the local coordinate system are determined by time-delayed observations. and speed of sound Based on the constant gradient acoustic ray theory, the curved path of the acoustic signal propagating from the target to the i-th array element satisfies a differential equation. The parametric equation of the acoustic ray trajectory is obtained by solving the method of separation of variables. Then, the corrected propagation path length of the acoustic signal from the target to the i-th array element is calculated by arc length integration. ,satisfy:: , in, For the sound speed gradient, Let z represent the square of the underwater sound speed at depth z where the underwater target is located. To represent the underwater sound speed at a reference depth, The depth of the i-th element underwater sound speed at that location The initial incident angle of the sound signal at the target is given. This path length takes into account the curvature of the sound ray caused by the change in sound speed with depth. Compared with the traditional straight-line propagation assumption, it is closer to the real propagation law. Thus, the propagation path model with constant gradient sound ray correction is completed.

[0037] Next, based on the corrected propagation path length Construct a delay prediction model; for N array elements in the system, establish a delay measurement equation: , in, Based on the corrected propagation path length The calculated target to the number i The prediction delay of the array element For the equivalent underwater speed of sound, The time delay measurement noise follows a Gaussian distribution. The time delay measurement noise, The variance is given by a normal distribution. The time delay measurement equations for each array element are then stacked into a vector form: , in, For the TOA observation vector, To the target location Let be the nonlinear observation function of the independent variable. Let the measurement noise vector follow a zero-mean Gaussian distribution, and its covariance matrix be denoted as . .

[0038] Simultaneously, in three-dimensional space, the azimuth vector of the target relative to the observation node is decomposed into the azimuth angle in the horizontal plane and the pitch angle in the vertical plane. For the i-th sensor and the target to be measured... Its angle measurement equation is: , , in, and These are the measurement errors for azimuth and elevation angles, respectively, and both follow a Gaussian distribution. Let be the two-dimensional coordinates of the underwater target to be measured in the horizontal plane. Let i be the known two-dimensional coordinates of the i-th array element in the horizontal plane. This is the corrected propagation path length; The angle measurement equations of each array element are stacked into observation vector form to obtain the azimuth observation vector. and pitch angle observation vector : , , in, For azimuth angle nonlinear observation function, For azimuth measurement noise vector, For pitch angle nonlinear observation function, The pitch angle is measured as a noise vector. At this point, the propagation path model and time delay prediction model based on constant gradient sound ray correction are completed. The time delay prediction model can output the corresponding time delay prediction value, azimuth angle prediction value and pitch angle prediction value according to the target coordinates, which can be used for subsequent calculation of multi-parameter error vector.

[0039] S3. Calculate the multi-parameter error vector using the output of the time delay prediction model, including calculating the deviation between the multi-parameter observation data and the prediction parameter data; Based on the constant gradient acoustic ray correction propagation path model constructed in the second step, and combined with the preliminary estimate of the three-dimensional coordinates of the underwater target, denoted as... Deriving the predicted azimuth angle value Pitch angle prediction value And extract the latency prediction value output by the latency prediction model. The data constitutes the prediction parameters.

[0040] Using the known coordinates of the i-th element in the second-step propagation path model Based on the initial coordinates of the target The horizontal spatial geometric relationship between the target and the i-th array element is constructed. The differences in the horizontal coordinates and vertical coordinates between the target and the i-th array element in the horizontal plane are calculated. Based on trigonometric function relationships, the predicted azimuth angle is derived to ensure the consistency of the calculation logic between the observed and predicted values. At the same time, the horizontal distance corrected by the propagation path model in the second step is introduced to correct the differences in the horizontal and vertical coordinates. The difference eliminates the influence of sound ray curvature on the estimation of the horizontal position, and finally the corrected predicted azimuth angle is obtained. ,in, To make a preliminary estimate of the horizontal coordinate components along the x-axis of the underwater target, To make a preliminary estimate of the horizontal coordinate components along the y-axis for underwater targets, To make a preliminary estimate of the horizontal coordinate components in the z-axis direction for underwater targets.

[0041] Based on the target and the corrected propagation path length of the i-th element output by the second-step propagation path model, and combined with the vertical height difference between the target and the i-th element, a vertical spatial geometric relationship is constructed, the horizontal Euclidean distance is calculated, and the predicted pitch angle is derived based on trigonometric function relationships. The calculation formula is as follows: , in, The three-dimensional coordinates of the underwater target to be optimized. To correct the propagation path length, we obtain the corrected pitch angle prediction value. , The components of the horizontal coordinate in the z-axis direction are known for the i-th array element.

[0042] Directly call the time delay prediction model built in the second step, and input the initial coordinates of the target. Given the coordinates of the i-th element, obtain the predicted delay between the i-th element output by the model and the target. To ensure that the predicted time delay perfectly matches the path length corrected by the propagation path model, based on the multi-parameter observation data of the i-th element obtained in the first step, and the predicted parameter data derived above. The deviations are calculated one by one to obtain the error values ​​of three types of parameters, including time delay error, azimuth error, and elevation error. These errors are arranged in a fixed order of time delay-azimuth-elevation angle in column vector form to construct the multi-parameter error vector corresponding to the i-th element. Its expression is as follows: , in, Let i be the multi-parameter error vector corresponding to the i-th array element. , and These represent the time delay observation, azimuth observation, and elevation observation between the i-th array element and the target, respectively. The three-dimensional coordinates of the underwater target , and These are the predicted time delay, azimuth, and elevation angles, respectively. Each element corresponds to the error value of a type of parameter, resulting in a multi-parameter error vector for all array elements. This forms a set of error vectors; simultaneously, outliers in the error vectors are removed, and the mean value of the error for each type of parameter is calculated. with standard deviation Remove excess The error vectors corresponding to the error values ​​within the range ensure the effectiveness and reliability of the error vector set, ultimately yielding multi-parameter error vector data for subsequent optimization model construction. This data fully quantifies the deviation between observed values ​​and theoretical predictions.

[0043] S4. Construct a multi-parameter fusion positioning optimization model using the maximum likelihood estimation criterion, including calculating the corresponding error covariance matrix based on the multi-parameter error direction, and using the error covariance matrix to adaptively weight the errors of each parameter. The set of multi-parameter error vectors of all effective array elements obtained in the third step Statistical analysis was performed to determine the error distribution type and calculate the total covariance matrix. A typical time period during the positioning process, where the target and the receiving array remained relatively stationary, was selected. Operations S1 to S3 were repeated to form a stationary observation window of length N. Multiple parameter error data were acquired by repeating operations S1 to S3, with each data set including the time delay error, azimuth error, and elevation error of each array element. The three types of errors from all array elements during a single positioning process were integrated into a total error vector. For an array containing n effective array elements, a complete observation vector was defined. Its dimensions are Based on the observation sample set of this error vector, the KS test was used to verify the zero-mean Gaussian fit of the marginal distributions of the three types of scalar observation data: time delay error, azimuth error, and pitch error. The results showed that none of the three were significantly different from the zero-mean Gaussian distribution, thus confirming the total error vector. It follows a multidimensional zero-mean Gaussian distribution, i.e. ,in The fully coupled covariance matrix has dimensions of . Calculate the covariance matrix. First, the sample mean of each component is estimated based on the target's error vector samples. Due to the zero-mean property, its expected value approaches zero. Then, the matrix elements are calculated according to the definition of covariance. , , in, Let be the centering bias vector of the k-th sample group. For the multi-parameter error vector observations of the k-th sample, Given a multi-parameter error vector, the sample mean vector, due to the zero-mean property, has an expected value that approaches zero. Let be the time delay error component of the k-th sample group. This represents the sample mean of the time delay error component. Let be the azimuth error component of the k-th sample group. This represents the sample mean of the azimuth error components. Let k be the pitch angle error component of the k-th sample group. The sample mean and covariance matrix of the pitch angle error components. The dimension is 3N×3N, with diagonal elements representing the variance of each error component, and off-diagonal elements characterizing the correlation structure between different array elements and between different types of errors such as time delay, azimuth, and elevation. The estimated fully coupled covariance matrix... A positive definiteness check is performed to ensure that it satisfies positive definiteness and invertibility, providing an effective and numerically stable second-order moment constraint for subsequent fusion positioning based on error statistics.

[0044] Then, determine the joint error vector. ,in, For the global joint observation column vector of the system, The joint prediction vector is composed of the time delay prediction vector, azimuth prediction vector, and elevation prediction vector derived from the target position. The joint error vector E follows a multivariate Gaussian distribution with zero mean and a dense covariance matrix Q. .

[0045] Based on the maximum likelihood estimation criterion, the likelihood function for multi-parameter fusion positioning is derived. The physical meaning of the likelihood function is the probability of observing the current global joint observation column vector given the coordinates of the target to be estimated. Since the observation processes of the three parameters—time delay, azimuth, and elevation—are independent, the joint likelihood function is expressed as: , in, Let Z be the likelihood function of the global joint observation column vector observed given the target location p. Let Q be the determinant of the covariance matrix. This represents the total dimension of the joint observation vector.

[0046] The core idea of ​​maximum likelihood estimation is to find a set of target coordinates. This maximizes the likelihood function. Taking the natural logarithm of the likelihood function and discarding the constant term independent of p, the maximum likelihood estimation is equivalent to minimizing the high-dimensional Mahalanobis distance of the residual vector. A multi-parameter fusion positioning optimization model is constructed, and the optimal estimation of the target coordinates is achieved by minimizing the objective function, which is: , , in, The optimal estimate of the target position is obtained based on the maximum likelihood estimation criterion, under the ideal assumption of independent noise. Given a dense covariance matrix, the cost function is... The joint likelihood function can be simply degenerated into a scalar superposition of the squared residuals of the four sensors. However, in a real SBL engineering environment, due to the strong correlation between ranging and angle measurement, and between array elements, the joint likelihood function cannot be decomposed into the independent summation of each sensor. Cost function Must be written as residual vector Weight precision matrix and Quadratic matrix multiplication of residual vectors is obtained by using the empirical covariance matrix. .

[0047] According to the maximum likelihood estimation criterion, the goal of localization optimization is to find a function that minimizes the cost function. Maximum target coordinates .in, The coordinate constraint range of the location area is determined by the detection radius of the short baseline system. Since the exponential function is a monotonically increasing function, The maximum value corresponds to its minimum value. Thus, based on the likelihood function probability, the objective function is optimized. The multi-parameter fusion positioning optimization model has been constructed. This model not only reflects the probabilistic characteristics of multi-parameter observations, but also realizes adaptive weighting of errors through the inverse of the covariance matrix.

[0048] S6. Using a multi-parameter fusion positioning optimization model as the fitness function, the optimal estimated value of the target position is output through iterative solution by combining genetic algorithm and quasi-Newton iteration method.

[0049] First, the original maximum likelihood estimation cost function obtained in step four is... As the fitness function of genetic algorithms Because genetic algorithms employ a fitness-maximizing optimization strategy, and The smaller the value, the smaller the Mahalanobis distance of the observed residuals and the higher the estimation accuracy.

[0050] Based on the positioning accuracy requirements and computational resource constraints of the short baseline positioning system, the key parameters of the genetic algorithm are configured as follows: the population size is set to 150 individuals, and each individual is encoded as the three-dimensional coordinate vector of the underwater target to be optimized. The crossover probability is set to 0.7 to 0.9, and an arithmetic crossover operator is used to generate offspring by linearly combining parent individuals based on random weight coefficients; the mutation probability is set to 0.2, and an adaptive Gaussian mutation operator is used to apply Gaussian random perturbations to random positions on the individual gene chain, with the mutation length decreasing exponentially with the number of iterations; the maximum number of iterations is set to 150 generations.

[0051] During the population initialization phase, a uniform random generation strategy is used to construct the initial population. A set number of three-dimensional coordinates are randomly generated as initial individuals.

[0052] The multi-generation iterative process of the genetic algorithm is executed according to the following sub-steps.

[0053] (1) Parent selection: Using the tournament selection operator, 3 to 5 individuals are randomly selected from the current population to form a competition group. The individual with the highest fitness value in the group is selected as the parent individual. The above operation is repeated until the number of parent individuals selected is consistent with the population size, so as to ensure that the superior genes are stably inherited. (2) Crossover and mutation: For the selected parent individuals, perform arithmetic crossover operation according to the preset crossover probability to generate offspring individuals, and then perform adaptive Gaussian mutation operation on all offspring individuals according to the preset mutation probability to expand the search range of the solution space and maintain population diversity; among which, the Gaussian mutation is asynchronous. The initial step size, The attenuation coefficient is... For the current algebra, The maximum number of iterations; (3) Elite retention strategy: retain the top 20% of elite individuals in the current population in terms of fitness value and directly copy and pass them to the next generation population to ensure the monotonicity of algorithm convergence and that the optimal solution is not lost; (4) Population renewal: Elite individuals and the remaining offspring individuals selected in descending order of fitness value are combined to form a new generation of population; (5) Genetic Algorithm Termination Judgment: After each generation of iteration, the optimal fitness value of the current population is recorded. When the change in the optimal fitness value is less than a preset threshold for 10 consecutive generations, the genetic algorithm terminates. The iteration process of the genetic algorithm phase is terminated when the current iteration number reaches the preset maximum iteration number.

[0054] After the genetic algorithm phase terminates, the individual with the highest fitness value in the current population is selected. As the initial value for the local refinement stage of the L-BFGS-B quasi-Newton method.

[0055] L-BFGS-B local refinement stage, with As initial values, with cost function The algorithm minimizes the objective function; in each iteration, it calculates the gradient vector at the current point. ,in The Jacobian matrix of the observation model is given; the L-BFGS-B algorithm's double-loop recursive mechanism implicitly approximates the inverse of the Hessian matrix using the historical coordinate differences and gradient differences of the most recent steps, thus constructing the second-order descent direction. ,in To approximate the Hessian matrix using the L-BFGS method; after each iteration update, a preset physical space boundary is used. Projection constraints are applied to the iteration points to ensure that the optimization process always takes place within the effective feasible region.

[0056] The L-BFGS-B algorithm iteration terminates when any of the following convergence conditions are met: gradient infinity norm. , or the change in coordinate position between two adjacent iterations The value is set to 0.01 meters to meet the centimeter-level positioning accuracy requirement; after the algorithm terminates, the current iteration point is output as the final positioning solution. .

[0057] By employing a hybrid hierarchical optimization strategy that combines the aforementioned genetic algorithm with the L-BFGS-B quasi-Newton method, the genetic algorithm first performs a parallel breadth-first search in the global solution space, effectively avoiding multiple local minima on the non-convex cost surface and stably falling into the global optimal attraction domain. Then, the L-BFGS-B algorithm utilizes second-order curvature approximation information to achieve rapid quadratic convergence in the neighborhood of the optimal solution, ultimately outputting a target position estimate that combines global optimality and high accuracy.

[0058] After the iteration terminates, the individual with the highest fitness value is selected from the final population. The three-dimensional coordinates corresponding to this individual are the optimal estimate of the underwater target's location. Subsequently, the accuracy of this optimal estimate is evaluated: the true location coordinates of the underwater target are obtained using an underwater high-precision positioning device. The three-dimensional Euclidean distance between the optimal estimate and the true location is calculated using the following formula: , Next, calculate the three-dimensional Euclidean distance from the actual position to the origin, using the formula: The ratio of the two is taken as the relative error of 3D positioning, and the expression is: , in, These are the coordinates of the optimal estimated target location obtained through a genetic algorithm. The system takes the actual position coordinates of the underwater target as an example, evaluates the positioning accuracy based on the relative error, and continues until the positioning accuracy meets the requirements of practical applications. Finally, it outputs the optimal estimated value of the target position and the corresponding accuracy evaluation result, thus completing the entire high-precision three-dimensional positioning process for underwater targets.

[0059] Example 2 This embodiment provides a simulation experiment of a multi-parameter fusion localization optimization method based on a genetic algorithm; To verify the effectiveness and superiority of the proposed multi-parameter joint GA-BPGS algorithm, this embodiment selects several existing mainstream localization algorithms for comparative experiments, including: linear least squares method, nonlinear least squares method, conventional BP neural network, SSA algorithm, VO2 algorithm, and SOR algorithm. Based on measured data, a comprehensive evaluation is conducted through multiple dimensions such as absolute error, relative error, cumulative error distribution, and spatial visualization. The specific comparison results are as follows: (1) Comparison of Mean Absolute Error (MAE) and Relative Positioning Error like Figure 4 As shown, the GA-BPGS algorithm proposed in this embodiment exhibits extremely high positioning accuracy. In terms of MAE (mean absolute error), the mean absolute error of this embodiment is only 0.18m, while the conventional BP neural network has an error as high as 14.53m, the SSA algorithm 7.55m, and the nonlinear least squares method 2.42m. Regarding the mean relative positioning error, the relative error of the algorithm of this invention is as low as 0.24%, significantly better than the 6.14% of the conventional BP neural network, the 7.34% of the SSA algorithm, and the 1.53% of the linear least squares method. The data shows that the MAE of this invention is reduced by approximately 98.7% compared to the conventional BP neural network, demonstrating an overwhelming accuracy advantage.

[0060] (2) Comparison of Cumulative Error Distribution Function (CDF) and Stability like Figure 5 As shown, the GA-BPGS algorithm in this embodiment not only has a low mean error but also exhibits optimal stability. Figure 1 In the CDF curves, the solid green line in this embodiment converges rapidly to 1.0 at a very small error, proving that the vast majority of positioning results can be controlled within a very small error range; while the curves of the other comparative algorithms are flatter and converge slowly. Figure 5 From the error box plot distribution, the algorithm in this embodiment has extremely high error concentration, very narrow boxes, and no obvious outliers; for example Figure 7 , Figure 8As shown, in comparison, although conventional BP neural networks and SSA algorithms have some correct estimates, their error fluctuations are extremely large, and there are a large number of outliers with abnormally deviated positions, making it difficult to guarantee the stability of positioning.

[0061] (3) Three-dimensional spatial positioning and XY plane projection analysis like Figure 6 As shown, in the positioning projection in three-dimensional space and the XY two-dimensional plane, the red asterisk represents the true target position, the black equilateral triangle represents the array position, and the green dot represents the target position calculated by the algorithm of this invention. It can be clearly seen from the figure that most of the green positioning points are closely and evenly clustered around the red true value points, with minimal overall offset and no obvious divergence. This fully demonstrates that the algorithm of this invention can effectively achieve accurate convergence of positioning points in complex three-dimensional spatial environments, not only meeting high precision requirements but also verifying the high reliability and feasibility of the method in practical engineering applications.

[0062] Example 3 The difference between this embodiment and Embodiment 1 is that this embodiment provides a multi-parameter fusion localization optimization system based on a genetic algorithm, including: The data acquisition module is configured to acquire multi-parameter observation data, including the underwater target radiated acoustic signal received by each element of the short baseline array, and to extract the time delay, azimuth angle and elevation angle between the array elements and the underwater target using signal processing technology. The preprocessing module is configured to: establish a propagation path model with constant gradient sound ray correction based on the sound velocity profile data of the positioning area, and construct a time delay prediction model by combining multi-parameter observation data and the output of the propagation path model; The conversion module is configured to: calculate a multi-parameter error vector using the output of the time delay prediction model, including calculating the deviation between the multi-parameter observation data and the prediction parameter data; The model module is configured to: construct a multi-parameter fusion localization optimization model using the maximum likelihood estimation criterion, including calculating the corresponding error covariance matrix based on the multi-parameter error direction, and adaptively weighting the error of each parameter using the error covariance matrix; The feature extraction module is configured to: establish a genetic algorithm collaborative constraint mechanism, design collaborative constraint operators for the multi-parameter fusion localization optimization model during the genetic algorithm iteration process, and perform constraint verification on the candidate solutions generated by the genetic operation; The transformation module is configured to use a multi-parameter fusion localization optimization model as the fitness function, perform iterative solutions through a genetic algorithm, and output the optimal estimate of the target position.

[0063] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A multi-parameter fusion localization optimization method based on genetic algorithm, characterized in that, include: Acquire the underwater target radiated acoustic signals received by each element of the short baseline array, and use signal processing methods to extract the time delay, azimuth angle and elevation angle between the array elements and the underwater target to obtain multi-parameter observation data; Based on the sound velocity profile data of the local area, a propagation path model with constant gradient sound ray correction is established. The corrected propagation path length is output by the propagation path model, and a time delay prediction model is constructed based on the corrected propagation path length. The error vector is obtained by calculating the error between the multi-parameter observation data and the prediction parameter data output by the time delay prediction model. A multi-parameter fusion localization optimization model is constructed using the maximum likelihood estimation criterion, including calculating the corresponding error covariance matrix based on the multi-parameter error vector, and adaptively weighting the errors of each parameter using the error covariance matrix. The maximum likelihood function of the multi-parameter fusion localization optimization model is used as the fitness function. The optimal estimated value of the target position is output by iteratively solving the problem through a genetic algorithm.

2. The multi-parameter fusion localization optimization method based on genetic algorithm according to claim 1, characterized in that, The process of obtaining multi-parameter observation data includes preprocessing the acoustic signals received by each array element, removing environmental noise through an adaptive filtering algorithm, eliminating array element clock deviations through signal synchronization, calculating the cross-correlation function of the signals received by different array elements, determining the time delay by locating the peak position of the cross-correlation function, constructing a spatial geometric model using the known coordinates of each array element of the short baseline array, employing a near-field MVDR focusing beamforming algorithm based on the spatial geometric model, obtaining the target coordinate estimate through spectral peak search within the three-dimensional search area, and calculating the azimuth and elevation angle observations based on the geometric relationship between the target coordinate estimate and the known coordinates of each array element.

3. The multi-parameter fusion localization optimization method based on genetic algorithm according to claim 1, characterized in that, The step of constructing a delay prediction model based on the corrected propagation path length includes establishing a delay measurement equation for N array elements in the system: ， in, This is the predicted time delay from the target to the i-th element, calculated based on the corrected propagation path length. Let be the actual bending path length from the target to the i-th element, calculated by the constant gradient acoustic ray correction model. For the equivalent underwater speed of sound, It follows a Gaussian distribution. The time delay measurement noise, The variance is the normal distribution. The time delay measurement equations for each array element are then stacked into a vector form: ， in, For TOA observation vector, For nonlinear observation functions, For a measurement noise vector that follows a zero-mean Gaussian distribution, its covariance matrix is ​​denoted as . .

4. The multi-parameter fusion localization optimization method based on genetic algorithm according to claim 3, characterized in that, The step of constructing a time delay prediction model based on the corrected propagation path length also includes, in three-dimensional space, decomposing the target's azimuth vector relative to the observation node into azimuth angles on the horizontal plane. Pitch angle in the vertical plane For the i-th sensor and the target to be measured Its angle measurement equation is: ， ， in, and These are the measurement errors for azimuth and elevation angles, respectively, and both follow a Gaussian distribution. Let be the two-dimensional coordinates of the underwater target to be measured in the horizontal plane. Let i be the known two-dimensional coordinates of the i-th array element in the horizontal plane. This is the corrected propagation path length; The angle measurement equations of each array element are stacked into observation vector form to obtain the azimuth observation vector. and pitch angle observation vector : ， ， in, For azimuth angle nonlinear observation function, For azimuth measurement noise vector, For pitch angle nonlinear observation function, The noise vector is used to measure the pitch angle.

5. The multi-parameter fusion localization optimization method based on genetic algorithm according to claim 1, characterized in that, The step of calculating the error between the multi-parameter observation data and the prediction parameter data output by the time delay prediction model includes obtaining the predicted azimuth and elevation angle values, and combining them with the time delay prediction values ​​output by the time delay prediction model to form prediction parameter data. Based on the prediction parameter data, the difference between the observed value and the corresponding predicted value for each type of parameter is calculated. The three types of differences are arranged in vector form to obtain a multi-parameter error vector containing time delay error, azimuth error, and elevation error. The expression for the multi-parameter error vector is as follows: ， in, Let i be the multi-parameter error vector corresponding to the i-th array element. , and These represent the time delay observation, azimuth observation, and elevation observation for the i-th array element, respectively. The three-dimensional coordinates of the underwater target , and These are the time delay prediction value, azimuth prediction value, and elevation prediction value, respectively.

6. The multi-parameter fusion localization optimization method based on genetic algorithm according to claim 1, characterized in that, The construction of a multi-parameter fusion localization optimization model using the maximum likelihood estimation criterion includes: Calculate the error covariance matrix based on the multi-parameter error vector, and determine the joint error vector. The joint error vector follows a matrix with zero mean and a dense covariance matrix. The multivariate Gaussian distribution, i.e. ; Based on the maximum likelihood estimation criterion, the inverse of the dense covariance matrix is ​​used as the adaptive weight of the multi-parameter model to construct a multi-parameter fusion localization optimization model. The optimal estimation of the target coordinates is achieved by minimizing the objective function, which is expressed as follows: ， Maximum likelihood estimation is equivalent to minimizing the high-dimensional Mahalanobis distance of the residual vector: ， in, Let Z be the likelihood function of the globally joint observation column vector given the target position p. It is a dense covariance matrix. Let N be the inverse of the dense covariance matrix, and N be the number of elements. For the global joint observation column vector of the system, This is a joint prediction vector composed of the time delay prediction vector, azimuth prediction vector, and elevation prediction vector derived from the target position p. This is the optimal estimate of the target location obtained based on the maximum likelihood estimation criterion.

7. The multi-parameter fusion localization optimization method based on a combination of genetic algorithm and quasi-Newton method according to claim 1, characterized in that, The optimal estimate of the target location is obtained by taking the negative of the objective function of the maximum likelihood estimation as the fitness function of the genetic algorithm, and then selecting the individual with the best fitness as the optimal estimate of the target location through multiple generations of iteration of the genetic algorithm combined with the quasi-Newton method for fine optimization. The expression of the fitness function is as follows: ， in, Here is the fitness function of the genetic algorithm. This is the objective function for the original maximum likelihood estimation.

8. The multi-parameter fusion localization optimization method based on genetic algorithm and according to claim 7, characterized in that, The optimal estimate of the target position also includes obtaining the true position coordinates of the underwater target, calculating the Euclidean distance between the optimal estimate and the true position, and combining this with the Euclidean distance from the true position to the origin to obtain the three-dimensional positioning relative error. The positioning accuracy is then quantitatively evaluated based on this relative error. The expression for the three-dimensional positioning relative error is as follows: ， in, These are the coordinates of the optimal estimated target location obtained through a genetic algorithm. These are the actual position coordinates of the underwater target.

9. A multi-parameter fusion localization optimization system based on a genetic algorithm, executed according to the method of claim 1, characterized in that, include: The data acquisition module is configured to: acquire the underwater target radiated acoustic signals received by each element of the short baseline array, and use signal processing methods to extract the time delay, azimuth angle and elevation angle between the array elements and the underwater target to obtain multi-parameter observation data; The preprocessing module is configured to: establish a propagation path model with constant gradient sound ray correction based on the sound velocity profile data of the positioning area, and construct a time delay prediction model by combining multi-parameter observation data and the output of the propagation path model; The conversion module is configured to: perform error calculation between the multi-parameter observation data and the prediction parameter data output by the time delay prediction model to obtain a multi-parameter error vector; The model module is configured to: construct a multi-parameter fusion localization optimization model using the maximum likelihood estimation criterion; The feature extraction module is configured to: establish a collaborative constraint mechanism between the genetic algorithm and the quasi-Newton algorithm; The transformation module is configured to use a multi-parameter fusion localization optimization model as the fitness function, perform iterative solutions through a genetic algorithm, and output the optimal estimate of the target position.

Images