CN-122021255-A - Mirror image type acoustic chromatography step-by-step three-dimensional field inversion method based on empirical orthogonal function

CN122021255ACN 122021255 ACN122021255 ACN 122021255ACN-122021255-A

Abstract

The invention relates to a mirror image type acoustic chromatography step-by-step three-dimensional field inversion method based on an empirical orthogonal function, which comprises the following steps of collecting ocean environment data, carrying out modal extraction and dimension reduction characterization, establishing a field characterization system, constructing an acoustic propagation forward model, constructing a core observation quantity, carrying out joint inversion, carrying out robust solution by adopting a regularization algorithm to obtain a time coefficient of a cross section vertical structure, carrying out parameterization on a three-dimensional acoustic velocity field inversion problem, carrying out inversion of discrete cross section acoustic velocity into a globally optimal field estimation, carrying out parameterization on a three-dimensional flow velocity field inversion problem, carrying out inversion of discrete cross section flow velocity into a globally optimal field estimation, carrying out three-dimensional acoustic velocity field inversion problem solution and reconstruction, estimating global coefficients describing acoustic velocity abnormal fields, finally generating quantitative ocean acoustic velocity field products, carrying out solution and reconstruction on a three-dimensional flow velocity field inversion problem, estimating global coefficients describing east component and north component abnormal fields, and finally reconstructing a three-dimensional flow velocity field.

Inventors

ZHANG WEIMIN
Xu Weishuai
CHEN YAN
LI ZHANGLONG
YU YI
WANG XIAOHUI
JIANG YIFEI
LI BO

Assignees

中国人民解放军国防科技大学

Dates

Publication Date: 20260512
Application Date: 20251228

Claims (9)

1. The mirror image type acoustic chromatography step-by-step three-dimensional field inversion method based on the empirical orthogonal function is characterized by comprising the following steps of: Step S1, acquiring and preprocessing multi-mode marine environment data related to a target sea area from various sources to acquire space-time matched acoustic and marine observation data; s2, extracting and reducing dimension representation by using an empirical orthogonal function mode, extracting and researching a spatial structure mode which is dominant in a sea area sound velocity field and a flow velocity field by using historical observation or numerical mode data, and establishing a field representation system on the basis; S3, forward modeling and sensitive kernel calculation, namely constructing an acoustic propagation forward model for simulating the propagation behavior of sound waves in a complex ocean medium, and quantifying the sensitivity of acoustic observation propagation time difference to target flow velocity and sound velocity based on the propagation behavior; s4, constructing a core observed quantity, performing joint inversion, and performing robust solution by adopting a regularization algorithm to obtain a time coefficient of the vertical structure of the section; S5, parameterizing a three-dimensional sound velocity field inversion problem, namely constructing a basis function representing a horizontal sound velocity space structure by utilizing historical data, and assimilating a discrete section sound velocity inversion result into a globally optimal field estimation; s6, parameterizing a three-dimensional flow velocity field inversion problem, namely constructing a basis function representing a horizontal flow velocity space structure by utilizing historical data, and assimilating a discrete section flow velocity inversion result into a globally optimal field estimation; S7, solving and reconstructing a three-dimensional sound velocity field inversion problem, namely estimating a global three-dimensional sound velocity field from limited noisy section observation data by a regularization technology from a section sound velocity inversion result; And step S8, solving and reconstructing a three-dimensional flow velocity field inversion problem, namely estimating global coefficients for describing the abnormal fields of the east component and the north component from a section flow velocity inversion result through a regularization technology, and finally reconstructing the three-dimensional flow velocity field.
2. The empirical orthogonal function based mirrored acoustic tomography step three dimensional field inversion method of claim 1 wherein step S1 comprises: Integrating the public database, the historical voyage observation and CROCO numerical mode output, and collecting temperature, salt, deep profile and flow rate data of a target sea area to construct a training data set, wherein the quality control of the training data comprises the steps of removing abnormal values, unifying data formats, and performing space-time matching and gridding interpolation; The system comprises an on-board mobile platform, an A platform, an M platform, a high-precision sound source and receiver, a GPS positioning system, a motion sensor and a surface layer thermal salt meter, wherein the A platform is used for carrying out sound signal transmitting and receiving tasks under a preset track; In the data acquisition process, receiving MOAT original acoustic signals observed by a system, processing the original acoustic signals to obtain bidirectional propagation time of each path, preprocessing the acoustic signals through band-pass filtering, coherent detection and pulse compression to improve the signal to noise ratio and accurately identify the signal arrival time on each path, further, obtaining bidirectional propagation time observation values by calculating the time difference of forward and reverse propagation signals on the same path, wherein the time difference comprises flow velocity and sound velocity change information obtained by integration along the acoustic path, the physical basis is a ray travel time chromatographic model, and the acoustic propagation time Along the acoustic propagation path Is shown as follows: Wherein, the The arc length of the sound ray is indicated, And Representing the coordinates of the horizontal plane, The depth is indicated as such, Represents the average sound velocity of the sound, Representing the position A deviation of the sound velocity from the average sound velocity, Representing position along sound ray A flow velocity component tangential to the point sound ray path; the propagation time disturbance with respect to the average state is given by Mean flow rate: furthermore, in the target sea area, the kinetic energy is concentrated on the mesoscale fluctuations, and therefore the average flow velocity is negligible, i.e The propagation time disturbance is reduced to: Adopting a double-station reciprocal observation configuration, namely, the A platform and the M platform are provided with transmitting and receiving functions, and combining symmetrical paths constructed by the seabed mirror image transponder to inhibit the influence of vertical flow velocity on sound propagation time; For MOAT systems, a simplified two-way propagation time model is built, with propagation time disturbance differences expressed as: is a travel time disturbance along the positive direction of the section, Is the disturbance of travel time along the reverse direction of the section, and establishes the time difference And horizontal flow along the acoustic line path Calculating an average flow velocity along the section using the propagation time difference data of the marine acoustic tomography observations; constructing the sum of the sound velocity deviation and the propagation time disturbance by considering only the influence of the sound velocity field change on the propagation time The sum of the propagation disturbance times is expressed as the following relation model: In addition to acoustic observations, data from other on-site observation platforms is received, including at least one of temperature, salt, and flow profiles provided by submerged buoy, argo buoy, on-board temperature and salt depth CTD, for verification of inversion results and supplemental constraints.
3. The empirical orthogonal function based mirrored acoustic tomography step three dimensional field inversion method of claim 2 wherein step S2 includes: Using the collected long-term historical temperature, salt and deep profile data of the target sea area to calculate and obtain a historical sound velocity field based on a Chen-Millero empirical formula; On the basis of obtaining historical sound velocity field and corresponding flow velocity field data, extracting horizontal average flow velocity profile and sound velocity profile of each section direction, respectively carrying out EOF decomposition on the horizontal average flow velocity profile and the sound velocity profile and extracting vertical spatial modes of the horizontal average flow velocity profile and the sound velocity profile, specifically, decomposing the deviation of the vertical flow velocity profile and the sound velocity profile from the average profile into linear superposition of the EOF modes for each section: Where N is the total number of modes, J is the number of successive layers, Is the first The flow rate of the individual successive layers, Is the first The sound velocity of the successive layers, Is the first The depth of the successive layers is such that, Is the flow rate of The number of EOF modes is one, Is the sound velocity of The number of EOF modes is one, Is the flow rate of The time coefficients of the individual EOF modes, Is the sound velocity of Time coefficients of the EOF modes.
4. The empirical orthogonal function-based mirrored acoustic tomography step-wise three-dimensional field inversion method of claim 3 wherein step S3 includes: Based on historical background field data, performing high-precision sound ray tracking simulation by adopting Bellhop ray acoustic models to identify and count sound ray paths with stability among mirror image chromatography pairs under different typical ocean environment conditions; inversion of the horizontal flow velocity of the section is carried out, and a sensitivity kernel function of section observation is calculated based on the acoustic ray path, wherein for each acoustic section, layered discretization is firstly carried out in the vertical direction, and the water depth dimension is equally divided into Successive layers, the first Flow rate in layer And sound velocity For uniform distribution, the sound ray path is then discretized into intra-layer line segments, the The sound line is at the first The length of the layer is The discretized propagation delay difference is approximately characterized by a path integral: I is the total number of sound rays, Is the first Propagation delay of the strip sound line is poor; substituting the intra-layer flow rate in the above formula with the EOF expansion formula, and combining discretization to obtain an acoustic-flow projection equation as follows: Wherein the method comprises the steps of Is a nuclear matrix constructed from acoustic line geometry and EOF modalities; And establishing a matrix equation by combining observation equations of all sound rays, wherein the matrix equation is shown as follows: inversion of section sound velocity is carried out, sound velocity deviation fields are expressed as linear combinations of a plurality of modes based on an EOF (object oriented function), and section inversion problems are parameterized based on the EOF: Wherein G IN is the nuclear matrix of the Nth mode of the I sound ray, Is the sum of the propagation disturbance times of the first I sound ray, Is the flow rate of The time coefficients of the individual EOF modes, Is the sound velocity of Time coefficients of the EOF modes.
5. The empirical orthogonal function based mirrored acoustic tomography step three dimensional field inversion method of claim 4 wherein step S4 includes: obtaining MOAT core observables of all the preprocessed effective paths, including the preprocessed acoustic signal bidirectional propagation time difference and the preliminary estimation of the flow velocity along the section area deduced from the physical relationship, and performing quality control and error estimation based on 3 sigma principle to construct an observation vector ; Based on the linear relation established by the EOF dimension reduction representation and the forward modeling, constructing a joint inversion equation: Wherein, the For a set vector of MOAT observed profile travel time differences or profile flow rates, Is a sparse solution vector of EOF coefficients, In order to project the matrix of the light, Characterizing an observation error and an error in a model solving process; Solving inversion problems by using Tikhonov regularized least square method, wherein the method overcomes discomfort of the problems by introducing additional constraint conditions and is characterized in that an objective function formed by data mismatch terms and solution constraint terms is minimized The following formula is shown: Wherein the method comprises the steps of Is Lagrangian multiplier, determined by L-curve method, Is the weighted average coefficient of the solution in the vertical or horizontal plane when When the minimum value is taken, the best solution of the equation set is as follows: Finally, based on the EOF time coefficient obtained by solving And Reconstructing a horizontal flow velocity vertical section in the section direction between each mirror image acoustic chromatographic node pair through linear superposition operation by combining pre-extracted EOF spatial modes And sound velocity vertical profile : Wherein, the Is the horizontal flow rate of the j-th layer, Is the average horizontal flow rate of the j-th layer, Is the speed of sound of the j-th layer, Is the average sound velocity of the j-th layer.
6. The empirical orthogonal function based mirrored acoustic tomography step three dimensional field inversion method of claim 5 wherein step S5 includes: Preprocessing and decomposing the collected historical three-dimensional flow velocity and sonic field data for each depth Extracting horizontal fields of all time steps on the depth level in the historical data, and independently carrying out EOF decomposition in the horizontal direction; calculating the sound velocity C (x, y, z, t) at each spatial point on the depth level Obtaining a climatic state background field of the depth level for the reconstruction of the sound velocity field The following formula is shown: Wherein the method comprises the steps of For the total length of the time series, the climatic state field represents the equilibrium or average state of the depth-level sound velocity; Subtracting the climatic state field from the original instantaneous field to obtain an abnormal field of sound velocity The following formula is shown: each time step Flattening of horizontal anomaly field of (2) into column vector Splicing column vectors of all time steps to construct an abnormal data matrix of the depth layer By solving the spatial covariance matrix corresponding to the data matrix To obtain a horizontal EOF modality, as shown in the following formula: Wherein, the Is the first Characteristic values of the individual horizontal EOF modes, Is the first Individual EOF spatial modalities or feature vectors; At any time Is a complete horizontal sound velocity abnormal field of (c), from the front Modality with highest variance contribution rate and time coefficient thereof Linear reconstruction is performed as shown in the following formula: By time coefficient Replaced by coefficients to be inverted which vary with depth but remain unchanged in the horizontal direction Thereby obtaining a parameterized expression as follows: under this framework, the inversion problem of three-dimensional field reconstruction, from estimating values at innumerable grid points, is transformed to solve a set of depth-dependent modal coefficients , Is the first A modal coefficient; Finally, by minimizing the following cost function To determine : Wherein the method comprises the steps of Is a linear observer, from a complete three-dimensional field Extracting data of the position of the r-th section; i.e. the sound velocity value on the r-th section predicted by the current model, Is at the first On each section The actual observed value of the position, s, is the position of the position point in the horizontal direction of the section, Is a constraint weight that is set to be equal to the constraint weight, Is an L2 norm regularization term, avoiding A severe oscillation occurs that is not physically present, Is the first And modal coefficients.
7. The empirical orthogonal function based mirrored acoustic tomography step three dimensional field inversion method of claim 6 wherein step S6 includes: The climatic state average flow field of the historical three-dimensional flow velocity field data is calculated for the east component U (x, y, z, t) and the north component V (x, y, z, t) of the horizontal flow velocity And The following formula is shown: At each depth On the east component anomaly field independently And north component anomaly field And (3) carrying out horizontal EOF decomposition, and respectively extracting dominant modes representing the horizontal spatial variation characteristics of the east component and the north component, wherein the dominant modes are shown in the following formula: here, the And The m-th spatial modality of the U-component and V-component respectively, And The U component and the V component respectively The number of modal coefficients is chosen such that, And Covariance matrices of the U component and V component anomaly fields respectively; introducing EOF coefficients of U and V components to be inverted as a function of depth And Reconstruction of the transient field is expressed as: for a point on the section r Its geographic coordinates are The model flow velocity vector of the point is The flow velocity component along the cross section direction predicted by the point model Calculated from the dot product, the following formula is shown: Wherein, the Is the included angle between the r-th section and the east direction; substituting the parameterized model and taking into consideration that the climatic state average flow has normal projection, predicting the flow velocity of the model in the section direction Expressed by the following formula: Based on the model, the model is solved by adopting a Tikhonov regularized least square method, and a stable solution is obtained by minimizing the following objective function J, wherein the following objective function J is shown in the following formula: wherein R is the total number of sections, Is the average horizontal flow velocity in the cross section direction obtained through vertical inversion, And The regularization parameters corresponding to the U and V components, respectively.
8. The empirical orthogonal function based mirrored acoustic tomography step three dimensional field inversion method of claim 7 wherein step S7 includes: for the inversion of the acoustic velocity field, the parameterized model is to be obtained Substituting into a cost function to solve: discretizing the successive integration and summation operations involved in the cost function, for The whole water body is discretized into the vertical direction according to the observation data Depth layers, before use From this, discrete observation data vectors are constructed Is that Is combined into Extracting the values of the background field at all the observation points to form a background field observation vector It is Is combined into Column vector of (2), parameter vector As an unknown number Arranged in a certain order into Is used for the column vectors of (a), n=m×k; design matrix For the first A predicted abnormal sound velocity value at each observation point Is a linear combination Wherein the coefficients The value rule of (2) is as follows if the first The unknowns are And observe the point Is also the depth of Then Otherwise, 0; Thus, the discrete form of the entire observation is: Is the average sonic field, Is the background field A contributed observation vector portion; constructing a discrete form of regularization term is: is the coefficient vector to be solved The first of (3) An element; the two items are integrated, and the cost function is completely discretized The following formula is shown: determining optimal regularization parameters using L-curves Values by plotting different Under-value solution norm And residual norms The curve presents an L-shaped corner, the corner point corresponds to The value is the optimal value by selecting the optimal value Solving the values to obtain a final stable parameter vector estimation final solution The optimal coefficient obtained by solving Substituting the three-dimensional acoustic velocity field into a parameterized model to obtain a reconstructed complete three-dimensional acoustic velocity field with the following formula: 。
9. the empirical orthogonal function based mirrored acoustic tomography step three dimensional field inversion method of claim 8 wherein step S8 includes: To achieve numerical solution, the objective function of the flow-velocity level inversion is continuously integrated and summed and discretized by The effective section flow velocity observation data is dispersed into the vertical direction Depth layers and choose the front The level representation is carried out on the order EOF mode, and discrete observation data vectors are constructed according to the level representation It is Is combined into Column vector of (2), background field vector Is that Is combined into Is used for the column vectors of (a), Is the first Section where each observation point is located Included angle with east direction, parameter vector The unknowns are arranged in sequence in the following format, and the total dimension is : Is the first Coefficients of the eastern flow velocity mode of the order, Is the first The depth of the individual depth layers; construction of linear forward operators, i.e. sensitive kernel matrices The matrix establishes a relationship between the parameter vector and the predicted observed data vector for the first A plurality of observation points for predicting abnormal flow velocity values Is a linear combination Wherein, the first The unknowns are And observe the point Depth of (2) Then sensitive nuclear matrix Elements of (2) First, the The unknowns are And observe the point Depth of (2) Then Otherwise, the device can be used to determine whether the current, ; The discrete form of the entire observation is: The regularization term is constructed by considering the physical characteristics of the east and north components of the flow velocity field, and the discrete form of the regularization term is shown in the following formula: Wherein the method comprises the steps of Is one If the diagonal weight matrix of (a) Is that The element of the j th row and the j th column If (3) Is that Coefficient of (1) ; The two items are integrated, and the cost function is completely discretized The following formula is shown: determining regularization parameters using L-curve method By calculating and plotting the optimum values of (a) different values A double logarithmic curve of the solution norm and the residual norm under the value is selected, and the corner point of the double logarithmic curve is selected The value is finally solved to obtain a final solution The optimal coefficient obtained by solving And Substituting the three-dimensional flow velocity field into a parameterized model, and reconstructing to obtain a complete three-dimensional flow velocity field with the following formula: 。

Description

Mirror image type acoustic chromatography step-by-step three-dimensional field inversion method based on empirical orthogonal function Technical Field The invention belongs to the technical field of mirror image acoustic tomography and the technical field of ocean parameter inversion, and particularly relates to a mirror image acoustic tomography step-by-step three-dimensional field inversion method based on an empirical orthogonal function. Background The three-dimensional fine structure of the sound velocity field and the flow velocity field in the ocean has important significance for navigation of underwater vehicles, detection of underwater targets, research on ocean power processes and climate prediction. The sound velocity field is used as a main carrier for sound wave propagation in a marine medium, and the spatial distribution and the change of the sound velocity field directly influence the propagation path, the energy attenuation and the information integrity of sound signals, so that the performance of a hydroacoustic positioning, communication and detection system is determined. The flow velocity field reflects the process of transporting and distributing the energy in the ocean, and the multi-scale structure of the flow velocity field plays a key role in understanding ocean circulation, energy cascade and material exchange. Therefore, the realization of high-resolution, synchronous and quantitative reconstruction of the two field structures has become one of the core problems in marine environment perception and scientific research. Traditional contact measurement means, such as CTD profilers and acoustic Doppler flow profilers (Acoustic Doppler Current Profiler, ADCP), although capable of providing high-precision thermal salt flow data on single points or vertical profiles, have limited spatial coverage and low temporal resolution, and are difficult to capture the rapid changes of marine environments and the detailed features of three-dimensional structures. Meanwhile, the method generally needs to rely on a survey ship for operation, is time-consuming and labor-consuming, has high layout and maintenance cost, and is difficult to implement in severe sea conditions or deep sea areas. Although the range and flexibility of data acquisition are expanded to a certain extent by the mobile observation platforms such as the main underwater vehicle and the glider in recent years, the observation is still mainly carried out by discrete points or sections, and the real-time three-dimensional monitoring in a large range is difficult to realize. Therefore, developing a technique that overcomes these limitations and enables efficient, economical, non-contact inversion of ocean parameter fields has become an urgent need in the field of ocean observations. Marine acoustic tomography (Ocean Acoustic Tomography, OAT) has proven to be a powerful tool for reconstructing large-scale average sonic and flow fields in the ocean since the 70 s of the 20 th century as a remote sensing inversion technique. The basic principle is to invert the physical properties of the medium using the time (or phase) information of the propagation of the sound wave in the ocean medium. Specifically, the propagation time of an acoustic signal is affected by integrating the speed of sound along the acoustic path along with the flow velocity, with the forward propagation time being shorter than the reverse propagation time, while the change in speed of sound affects the bi-directional propagation time. By arranging a plurality of sound sources and receivers to form an acoustic path network, acquiring propagation time differences or differential time data in different directions, establishing an integral equation about sound velocity fields and flow velocity fields, and combining a proper mathematical inversion algorithm, the spatial distribution structure of the ocean internal parameter fields can be deduced from limited acoustic observation data. However, the conventional MOAT inversion technology still has a series of key limitations, which severely restrict the application of the inversion technology from the theoretical concept to business (1) in the inversion strategy, the main stream method still severely depends on directly carrying out high-dimensional parameter estimation on an original or sparse grid, so that the size of the parameters to be solved is huge, and the stability and reliability of the solution are limited. (2) The existing method is often developed by adopting a general basis function (such as Fourier basis or wavelet basis) in mathematical sense in horizontal inversion, and lacks explicit description of spatial structural features of an actual marine power environment, so that the inversion result is difficult to maintain physical reasonable spatial continuity, and structural distortion or false fluctuation is easy to occur. (3) The current MOAT inversion flow is usually designed into one-time global solution or si