A method and system for matching broadband modal phase velocity inversion of geoacoustic parameters based on sparse bayesian learning

Abstract

The application provides a sparse Bayesian learning-based matching wideband modal phase velocity inversion geoacoustic parameter method and system, which uses the receiving signals of a vertical array, extracts local modes by using multi-distance sparse Bayesian learning, and removes the narrowband limitation of block sparse Bayesian; the wideband modal phase velocity is used for matching inversion of geoacoustic parameters, and the decoupling of geoacoustic parameters is realized. The application has the advantages that prior knowledge of seabed parameters is not required, the depth range of mode extraction is not limited to the aperture of the VLA and can be applied to a non-uniform vertical array; when the geoacoustic parameters are inverted, the influence of the terrain on the propagation path is not considered, and the geoacoustic parameters of any terrain can be inverted; the position information of the sound source is not required, and the problem that the inversion result is inaccurate due to inaccurate sound source position is avoided; meanwhile, the sound source signals of multiple distances are used for inversion, so that the inversion of the geoacoustic parameters is more robust and accurate.

Description

Technical Field

[0001] The present application belongs to the field of marine acoustic technology, and specifically relates to a method and system for matching broadband modal phase velocity inversion of geoacoustic parameters based on sparse Bayesian learning. Background Art

[0002] The inversion of seafloor acoustic parameters is a crucial issue in ocean acoustics. Inversion is a fast, cost-effective method for obtaining shallow-water geoacoustic parameters. It can determine acoustic parameters such as seafloor velocity, density, and attenuation coefficient. These parameters have a crucial influence on sound propagation in the ocean and are crucial for the design of various intelligent oceanographic instruments. The most widely used method for seafloor sediment inversion is matched-field inversion (MFI). Commonly used matching quantities include acoustic pressure, dispersion curves, and propagation loss. Traditional MFI is affected by the propagation path, and terrain mismatch can affect the accuracy of the inversion results. Furthermore, traditional MFI requires the sound source location to be included as a priori information in the inversion, or the sound source information is used as the inverted quantity, exacerbating the multi-value problem. When inverting multiple parameters, the coupling between geoacoustic parameters cannot be ignored. Furthermore, most MFIs use isolated data from a single point at a specific distance for inversion, failing to fully exploit the potential correlations between multiple acoustic data segments to generate more reliable results.

[0003] Modal functions and horizontal wavenumbers are shape functions and propagation characteristic parameters of normal mode, which are closely related to the parameters of the acoustic field (such as the sound speed in water, the depth of seawater and the characteristics of the seabed, etc.), determine the distribution and propagation characteristics of the acoustic field, and contain important information of the acoustic field. For modal extraction, horizontal arrays can estimate horizontal wavenumbers, but the classic spatial Fourier or Hankel transform method requires a large distance aperture. For vertical arrays, among many modal extraction methods, the mode filtering method requires prior environmental information, the SVD (Singular Value Decomposition) method requires a vertical array to span the full sea depth, and the warping transform method can only be used for pulse signals. In shallow sea waveguides, the acoustic field can be regarded as a weighted superposition of multiple-order normal modes. Therefore, modal extraction can be regarded as a sparse signal recovery problem and can be solved by sparse Bayesian learning (SBL). Sparse Bayesian learning does not need to know the seabed parameters and sound source information in advance when extracting modes using a vertical array, the depth range of the extracted modes is not limited to the aperture of the vertical array, and it can also be applied to non-uniform vertical arrays, which is more flexible and efficient in practical applications. Block sparse Bayesian learning uses a block structure of multi-frequency point modal depth functions as a dictionary matrix, but due to the use of an approximate dispersion relation, the frequency band is limited to a narrow band, and only narrow-band modes can be extracted. In addition, when a hydrophone is used to construct a multi-frequency point dictionary matrix to extract modes, the distance of the sound source needs to be known. SUMMARY

[0004] The purpose of the present application is to eliminate the dependence of traditional matched field inversion on the propagation path topography and sound source position information, alleviate the problem of representative loss existing in isolated point inversion, realize the decoupling of geoacoustic parameters, and improve the robustness and accuracy of inversion.

[0005] To achieve the above purpose, the present application proposes a matched broadband modal phase velocity inversion geoacoustic parameter method based on sparse Bayesian learning, comprising:

[0006] Step 1: Fourier transform the time-domain acoustic propagation data received by the vertical array of hydrophones at multiple distances to obtain the multi-distance frequency-domain signals of each frequency point within the required bandwidth;

[0007] Step 2: Collect the horizontal wavenumbers in the set interval and use the shooting method to generate the corresponding local modal depth functions to form the modal depth function matrix A;

[0008] Step 3: Group the multi-distance time-domain signals of each frequency point into a matrix Y, and use the sparse Bayesian learning iterative method to calculate the real excited local modes and local horizontal wavenumbers by using the matrix A and Y;

[0009] Step 4: Perform multi-distance sparse Bayesian learning on each frequency point within the required bandwidth and convert the obtained local horizontal wavenumber into local phase velocity;

[0010] Step 5: Perform data cleaning and curve fitting on the broadband phase velocity points;

[0011] Step 6: Substitute the environmental parameters to be searched into the sound propagation model to obtain the phase velocity at each frequency point; by minimizing the mean square error between the frequency-phase velocity curve obtained from the sound propagation model and the curve fitted in step 5, the environmental parameters including seabed sound velocity, density, and water depth are inverted.

[0012] Step 7: Perform depth correction on the actually measured water depth by using the inverted water depth;

[0013] Step 8: Use the matched propagation loss method to invert the seabed attenuation coefficient.

[0014] As an improvement to the above method, the setting interval is: [2πf / c b ,2πf / c w ]; where f represents the signal frequency, c b and c w represent the speed of sound on the seabed and in water respectively.

[0015] As an improvement to the above method, the formula of the shooting method is:

[0016] Φ(0,k rm )=0,

[0017] Φ(z1,k rm )=h,

[0018]

[0019] Where ω = 2πf is the angular frequency; z j =jh, where h and J are the width and total number of discrete grids respectively; c(z j ) is the sound velocity profile; Φ(z j ,k rm ) is the mth order normal mode depth function at the jth grid depth z j The discrete sampling value at k rm represents the m-th order horizontal wave number.

[0020] As an improvement to the above method, the local mode and horizontal wave number of the real excitation are expressed as:

[0021]

[0022] Among them, γ m To characterize the amount of local modal energy, Indicates the value of this iteration, Indicates the value of the last iteration; L indicates the number of sound sources; Φ(z,k rm ) is the discrete sampling of the m-th order normal mode depth function at depth z at the receiver; σ 2 is the variance of the noise; Σ y is the data covariance; is a matrix composed of the K excited modal functions in the matrix A, + represents pseudo-inverse calculation, and K represents the number of real propagation modes; I N represents the N-dimensional identity matrix, where N is the number of array elements; the superscript H indicates conjugate calculation; ||·||2 represents the 2-norm formula; and tr represents the trace of the matrix.

[0023] As an improvement to the above method, the obtained local horizontal wave number is converted into the local phase velocity using the formula:

[0024]

[0025] Among them, c pm represents the local phase velocity.

[0026] As an improvement to the above method, the inversion obtains environmental parameters including seabed sound velocity, density and water depth, and the formula is:

[0027]

[0028] Among them, M d is the total number of modes used; F is the total number of frequencies; represents the set of seabed environmental parameters to be inverted; is the fitted frequency-phase velocity curve; t m (f i ,u) is the copy field frequency-phase velocity curve generated by the sound propagation model when the environmental parameter set is u; f i is the frequency.

[0029] The present application also provides a system for matching broadband modal phase velocity inversion of geoacoustic parameters based on sparse Bayesian learning, which is implemented based on the above method and includes:

[0030] The module for obtaining multi-range frequency domain signals is used to perform Fourier transform on the time domain data of sound propagation at multiple distances received by the vertical array of hydrophones to obtain multi-range frequency domain signals for each frequency point within the required bandwidth;

[0031] The module for obtaining the modal depth function matrix is ​​used to collect horizontal wave numbers within a set interval and generate the corresponding local modal depth function using the shooting method to form the modal depth function matrix A;

[0032] The module for calculating local modes and local horizontal wavenumbers is used to combine the multi-distance time domain signals of each frequency point into a matrix Y. The matrices A and Y are used to calculate the local modes and local horizontal wavenumbers of the actual excitation through the iterative method of sparse Bayesian learning.

[0033] The local phase velocity acquisition module is used to perform multi-distance sparse Bayesian learning on each frequency point within the required bandwidth and convert the obtained local horizontal wavenumber into local phase velocity;

[0034] The curve fitting module is used to perform data cleaning and curve fitting on broadband phase velocity points;

[0035] The inversion environmental parameter module is used to bring the environmental parameters to be searched into the sound propagation model to obtain the phase velocity of each frequency point. By minimizing the mean square error between the frequency-phase velocity curve calculated by the sound propagation model and the fitted curve, the environmental parameters including seabed sound velocity, density and water depth are inverted.

[0036] A depth correction module is used to perform depth correction on the water depth obtained by inversion compared with the actual measured water depth;

[0037] The seabed attenuation coefficient inversion module is used to invert the seabed attenuation coefficient using the matching propagation loss method.

[0038] Compared with the prior art, the advantages of this application are:

[0039] 1. When extracting modes, the bandwidth restriction is removed, so that the modes, horizontal wave numbers, and phase velocities of broadband data can be calculated;

[0040] 2. No prior knowledge of seabed parameters is required, the depth range of the extracted modes is not limited by the aperture of the VLA and can be applied to non-uniform vertical arrays;

[0041] 3. When inverting geoacoustic parameters, the method of matching local modal phase velocity is used. The inversion is not affected by the terrain along the propagation path and the geoacoustic parameters of any terrain can be inverted.

[0042] 4. No need to know the location of the sound source, thus avoiding the problem of inaccurate inversion results caused by inaccurate sound source location;

[0043] 5. The decoupling of geoacoustic parameters is achieved, avoiding the mutual influence between parameters during inversion;

[0044] 6. Using multiple distances of sound source signals for inversion simultaneously makes the inversion of geoacoustic parameters more robust and accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Shown is a schematic diagram of the ocean environment and sound source - hydrophone deployment;

[0046] Figure 2(a) shows the multi-distance sparse Bayesian learning result curve at 250Hz and the actual horizontal wave number (red ×) in the simulation environment;

[0047] Figure 2(b) shows the modal depth function estimated by multi-distance sparse Bayesian learning (blue circle) and the actual modal depth function in the simulation environment (black line), from left to right, the 1st to 5th order respectively;

[0048] Figure 3 Shown are the scatter plots and fitted curves of broadband multi-distance sparse Bayesian learning;

[0049] Figure 4 The figure shows the parameter optimization diagram of the Bayesian optimization algorithm; the orange dots are the estimated points of each iteration, the red stars are the true values ​​set in the simulation, and the blue boxes are the final inversion results;

[0050] Figure 5(a) shows the frequency-phase velocity curve corresponding to the fitting curve (red line) and the matching phase velocity inversion result;

[0051] Figure 5(b) shows a comparison of the matching phase velocity inversion results and the actual frequency-phase velocity curve under the simulation environment;

[0052] Figure 6(a) shows the mid-range positioning result of the matching field positioning result. The red star is the true position of the sound source, and the blue circle is the MFP positioning result;

[0053] Figure 6(b) shows the depth positioning result in the matching field positioning result. The red star is the true position of the sound source, and the blue circle is the MFP positioning result.

[0054] Figure 7 Shown are the scatter plots and fitting curves of phase velocity estimation using broadband multi-distance sparse Bayesian learning;

[0055] Figure 8 Shown are the frequency-phase velocity fitting curve estimated by broadband multi-distance sparse Bayesian learning (red line) and the frequency-phase velocity curve corresponding to the broadband matching phase velocity inversion result (black line);

[0056] Figure 9(a) shows the distance positioning result of the matched phase velocity inversion; the red star is the sound source position recorded by GPS, and the blue circle is the MFP positioning result;

[0057] Figure 9(b) shows the depth positioning result of the matched phase velocity inversion; the red star is the sound source position recorded by GPS, and the blue circle is the MFP positioning result;

[0058] Figure 9(c) shows the distance positioning result of traditional matched field inversion; the red star is the sound source position recorded by GPS, and the blue circle is the MFP positioning result;

[0059] Figure 9(d) shows the depth positioning result of traditional matched field inversion; the red star is the sound source position recorded by GPS, and the blue circle is the MFP positioning result;

[0060] Figure 10(a) shows the cost function of the traditional matching field inversion for 59 points;

[0061] Figure 10(b) shows the results of the traditional matched field inversion for water depth at 59 points;

[0062] Figure 10(c) shows the results of the traditional matched field inversion for the seabed sound velocity at 59 points;

[0063] Figure 10(d) shows the results of the traditional matched field inversion for the seafloor density at 59 points;

[0064] Figure 10(e) shows the results of the traditional matched field inversion for the seafloor attenuation coefficient at 59 points;

[0065] Figure 11 Shown is the flow chart of the method for matching broadband modal phase velocity inversion for geoacoustic parameters based on sparse Bayesian learning. DETAILED DESCRIPTION

[0066] The technical solution of this application is described in detail below with reference to the accompanying drawings.

[0067] The present invention proposes a method and system for inverting geoacoustic parameters using matched broadband modal phase velocity based on sparse Bayesian learning. This method utilizes the received signals of a vertical array and uses multi-distance sparse Bayesian learning to extract local modes, removing the narrowband limitation of block sparse Bayesian learning. Broadband modal phase velocity is used for matched inversion of geoacoustic parameters, achieving decoupling of geoacoustic parameters. Furthermore, the inversion is not affected by the terrain along the propagation path and does not require known sound source location information. During the multidimensional parameter search process, a Bayesian optimization algorithm is used to improve search efficiency. Matched-field processing (MFP) positioning is performed based on the inversion results in simulation and experimental data, verifying the feasibility of this method and providing a new and feasible method for geoacoustic parameter inversion.

[0068] The principle and derivation process of the method for inverting geoacoustic parameters using matched broadband modal phase velocity based on sparse Bayesian learning are as follows:

[0069] 1. Signal Model and Multi-Distance Sparse Bayesian Learning

[0070] According to the simple normal wave theory, the point sound source at the lth sound source position Receive signal y in the frequency domain at depth z (l) (z) consists of the weighted superposition of M modes:

[0071]

[0072] in, is the sound source distance of the lth sound source position, is the sound source depth of the lth sound source position; Φ(z,k rm ) is the discrete sampling of the mth-order normal mode depth function at the receiver at depth z, often referred to as the local mode. cm =k rm +ik im is the complex eigenvalue, real part k rm represents the m-order horizontal wave number, and the imaginary part k im Represents the modal attenuation. Formula (1) is applicable in both distance-independent and distance-dependent scenarios, and essentially decomposes the sound pressure into local modes Φ(z,k rm ), the difference is that the modal amplitude in the formula There are different expressions under adiabatic and coupled conditions, that is, different decomposition coefficients. When the sound velocity profile is known, for a given horizontal wave number k rm , modal depth function Φ(z,k rm ) can be obtained by shooting method.

[0073] We use the local modal depth function at the same receiving array to express the received signals at different distances. Let the received signal of N array elements at each distance be y (l) , and consider the received signals of L sound sources at different distances to form the received signal matrix Y = [y (1) ,…,y (l) ,…,y (L) ], l=1,2,...L. Represents the amplitude of each mode at L distances, as shown below:

[0074]

[0075] At this time, the decomposition of signals at different sound source distances can use the same set of modal depth function matrices, written as matrix A, and the received signal set can be expressed as:

[0076] Y=AX+N (3)

[0077] A is composed of the M-order modal depth function:

[0078]

[0079] It follows a complex Gaussian distribution noise.

[0080] The signal mode is estimated using sparse Bayesian method. In the interval [2πf / c b ,2πf / cw ](f represents the signal frequency) a large number of acquisitions (M n Horizontal wave number k rm And use the shooting method to generate the corresponding local modal depth function, c b and c w are the speed of sound on the seabed and in water respectively. b is unknown and can be set to a larger value during execution. At this time, the number of columns of A is also expanded to M n Columns, the number of rows of X also expanded to M n OK. In A's M n Among the modes in the column, only K are real propagation modes, and the rest are virtual. n is much larger than the actual number of propagating modes K, so most of the rows in X are zero, making it K-sparse.

[0081] According to the sparse Bayesian learning theory and calculation formula, the dictionary matrix A is used to perform sparse Bayesian learning on the multi-distance receiving data set Y to estimate the modal coefficient set X. The non-zero value in X represents the energy of the mode. The appearance indicates the existence of the corresponding propagation mode. The position of the non-zero value can be used to calculate the local horizontal wave number and local modal depth function.

[0082] 2. Matching phase velocity inversion

[0083] The horizontal wave number k of the mth mode rm and phase velocity c pm There are the following relationships:

[0084]

[0085] Therefore, the local horizontal wave number estimated by multi-distance sparse Bayesian learning can be used to calculate the local phase velocity. r (or c p ) is affected by factors such as geoacoustic parameters, water depth and sound velocity profile. These environmental parameters can be inverted by matching the phase velocity of the sound propagation model. Because multi-distance sparse Bayesian learning involves local modes and local phase velocities, the inversion result is the parameters at the receiving point. Since the geoacoustic parameters are approximately unchanged within a certain distance range, the inverted geoacoustic parameters can be regarded as the values ​​within the sea area. The inverted sea depth H at the receiving array can be used to correct the instrument's depth measurement results. If the depth measurement is unknown, H is the water depth when the sea depth level remains unchanged; if the depth measurement is known, the inverted H can be used to correct the depth measurement when the sea depth level remains unchanged or changes. Theoretically, the matching phase velocity inversion method is not affected by the terrain and can be used to invert the geoacoustic parameters of any terrain. The inversion accuracy is higher than the traditional matching sound pressure inversion method that is affected by terrain changes along the propagation path.

[0086] In the existing methods, because the block sparse Bayesian learning method uses multi-frequency signals to estimate the horizontal wavenumber, an approximate dispersion relationship is required, so the frequency band needs to be a narrow band, and only the horizontal wavenumber and phase velocity of the narrow band can be estimated. However, the multi-distance sparse Bayesian method of the present application removes the narrowband limitation because each frequency point uses multi-distance receiving data to independently estimate the horizontal wavenumber, so the phase velocity of a wider frequency band can be estimated. We use broadband data to reduce the error influence of multi-distance sparse Bayesian learning in estimating the horizontal wavenumber of a single frequency point. Select the receiving data within a large bandwidth, and use multi-distance sparse Bayesian learning to estimate the horizontal wavenumber and phase velocity for each frequency point. Since the phase velocity has a trend of decreasing with increasing frequency, we perform curve fitting on the estimated multi-frequency phase velocity. Then, a series of environmental parameters to be searched are brought into the sound propagation model (KRAKEN) to obtain the phase velocity of each frequency point. By minimizing the frequency-phase velocity (fc p ) curve with the above fitting curve, and invert the ocean environmental parameters. We call this method matched-phase-velocity inversion (MPVI), and the formula is as follows:

[0087]

[0088] Among them, M d is the total number of modes used; F is the total number of frequencies; Represents the set of seabed environmental parameters to be inverted. is the fitted fc p Curve, t m (f i ,u) is the copy field fc generated by KRAKEN when the environmental parameter set is u p Curve, f i is the frequency. Since c p When , only the complex eigenvalue k is used c The real part k r , without the imaginary part k representing the modal attenuation i , so the calculation process does not involve attenuation. And because the horizontal wave number k is estimated using multi-distance sparse Bayesian learning r When k contains the attenuation information i It is implicit in the modal coefficient X, and we only use the position of the non-zero value of X, not the value of the non-zero value itself, so k is not involved. i , so the final estimated horizontal wave number k rThe sea bottom does not contain information of attenuation. For the above two reasons, the matched phase velocity inversion method is not sensitive to the sea bottom attenuation coefficient, i.e. the attenuation coefficient is decoupled from other environmental parameters such as the sea bottom sound velocity, density and water depth, etc.

[0089] Embodiment 1

[0090] Based on the above principle, as shown in Figure 11 , the matched broadband modal phase velocity inversion method for geophysical parameters based on sparse Bayesian learning includes:

[0091] Step 1: Fourier transform the sound propagation time-domain data received by the vertical array of hydrophones at multiple distances to obtain the multi-distance frequency-domain signals at each frequency point within the required bandwidth.

[0092] Step 2: Collect a large number of horizontal wave numbers in the interval [2πf / c b ,2πf / c w ] and use the shooting method to generate the corresponding local modal depth function to form a matrix A, c b and c w are the sound velocities in the sea bottom and water respectively. c b is unknown and can be set to a large value during execution. The shooting method formula is as follows:

[0093]

[0094] where ω = 2πf is the angular frequency and f is the frequency; z j = jh, h and J are the width and total number of discrete grids respectively, and c(z) is the sound velocity profile.

[0095] Step 3: Group the multi-distance time-domain signals at each frequency point into a matrix Y, and use the iteration formula of sparse Bayesian learning to calculate the real excited local mode and local horizontal wave number by using the matrix A and Y.

[0096]

[0097]

[0098] where γ m is a quantity representing the energy of the mode, represents the value of this iteration, represents the value of the last iteration; Φ m is a short form of Φ(z, k rm ); σ 2 is the variance of noise, Σ y is the data covariance, is a matrix composed of K excited mode functions in A, + represents pseudo-inverse calculation; I Nrepresents the N-dimensional identity matrix; ||·||2 represents the 2-norm formula; tr represents the trace of the matrix.

[0099] Step 4: After performing multi-distance sparse Bayesian learning on each frequency point within the required bandwidth (the frequency range in which the local horizontal wavenumber is required), the obtained local horizontal wavenumber is converted into the local phase velocity.

[0100]

[0101] Step 5: Perform data cleaning and curve fitting on the broadband phase velocity points.

[0102] Step 6: Substitute a series of environmental parameters to be searched into the sound propagation model (KRAKEN) to obtain the phase velocity of each frequency point. By minimizing the frequency-phase velocity (fc p ) curve and the fitting curve in step 5, and invert the environmental parameters including seabed sound velocity and density, and water depth.

[0103]

[0104] Step 7: Perform depth correction on the actually measured water depth using the inverted water depth to reduce the error caused by instrument measurement.

[0105] Step 8: Use the matching propagation loss method to invert the seafloor attenuation coefficient. This inversion result, along with the environmental parameters obtained in step 6, can be used in matching field structures and subsequent positioning processing. The matching propagation loss method can adopt existing mature methods.

[0106] Example 2

[0107] The present application also provides a system for matching broadband modal phase velocity inversion of geoacoustic parameters based on sparse Bayesian learning, which is implemented based on the above method and includes:

[0108] The module for obtaining multi-range frequency domain signals is used to perform Fourier transform on the time domain data of sound propagation at multiple distances received by the vertical array of hydrophones to obtain multi-range frequency domain signals for each frequency point within the required bandwidth;

[0109] The module for obtaining the modal depth function matrix is ​​used to collect horizontal wave numbers within a set interval and generate the corresponding local modal depth function using the shooting method to form the modal depth function matrix A;

[0110] The module for calculating local modes and local horizontal wavenumbers is used to combine the multi-distance time domain signals of each frequency point into a matrix Y. The matrices A and Y are used to calculate the local modes and local horizontal wavenumbers of the actual excitation through the iterative method of sparse Bayesian learning.

[0111] The local phase velocity acquisition module is used to perform multi-distance sparse Bayesian learning on each frequency point within the required bandwidth and convert the obtained local horizontal wavenumber into local phase velocity;

[0112] The curve fitting module is used to perform data cleaning and curve fitting on broadband phase velocity points;

[0113] The inversion environmental parameter module is used to bring the environmental parameters to be searched into the sound propagation model to obtain the phase velocity of each frequency point. By minimizing the mean square error between the frequency-phase velocity curve calculated by the sound propagation model and the fitted curve, the environmental parameters including seabed sound velocity, density and water depth are inverted.

[0114] A depth correction module is used to perform depth correction on the water depth obtained by inversion compared with the actual measured water depth;

[0115] The seabed attenuation coefficient inversion module is used to invert the seabed attenuation coefficient using the matching propagation loss method.

[0116] This invention addresses the problem of inverting seafloor acoustic parameters and ocean depth using a vertical array in shallow water environments. Using simulation and experimental data, the feasibility and accuracy of multi-distance sparse Bayesian learning for extracting local modes and horizontal wavenumbers are verified. A process for matching phase velocity inversion environmental parameters (geoacoustic parameters and ocean depth) is presented. The inverted parameters are then incorporated into a matching field positioning model for positioning, verifying the reliability of the inverted parameters. Experimental data also compares the method with traditional matching field inversion, demonstrating the improved accuracy of the proposed method through positioning results.

[0117] Example 1: Simulation Environment

[0118] In order to ensure the consistency between the simulation and the sea experiment, we refer to the actual situation of the sea experiment in the simulation (such as Figure 1 The launching ship conducted a distance experiment from the receiving point to the east, launching a total of 59 flares with a nominal depth of 7m within 60km ( Figure 1 The experimental sea depth is about 40m, and the speed of sound in seawater is about 1507m / s. The receiving point is a 16-element vertical array distributed in most of the water body (4.44~34.65m). In the simulation, we use Figure 1 The left part shows a uniform seabed model. Below the water layer with a depth of H meters is a semi-infinite space with a seabed sound velocity c. b , seafloor density ρ b and the seabed sound attenuation coefficient α b Three seabed acoustic parameters. Among them, c b Set to 1680m / s, ρ b 1.9g / cm 3 , α b According to the empirical formula αb =0.29f 1.91 dB / m setting, H is 40m. SNR = 27dB.

[0119] A single-frequency signal with a frequency of 250 Hz is selected, and the γ curve representing the modal energy is calculated using multi-distance sparse Bayesian learning. As shown in Figure 2(a), the peak of the γ curve is very sharp and has a high resolution. The curve has five obvious peaks, corresponding to the five horizontal wave numbers k. rm , indicating that five modes were retrieved. The red “×” in the figure represents the actual horizontal wave number of the sound field generated by KRAKEN in this simulation environment. It can be seen that the horizontal wave value estimated by multi-distance sparse Bayesian learning is consistent with the actual value. The k values ​​of each order at this time are calculated. rm The error rates are 0.0372%, 0.0643%, 0.1291%, 0.0187%, and 0.0537%, respectively, with an average error rate of 0.0606%. The corresponding five modal depth functions are shown in Figure 2(b). It can be seen that the modal functions estimated by multi-distance sparse Bayesian learning agree well with the modal functions of the KRAKEN simulation field. The simulation shows that the multi-distance sparse Bayesian learning method can effectively estimate the horizontal wavenumber and modal function with high accuracy.

[0120] The 50-500 Hz broadband signal is used to perform multi-distance sparse Bayesian learning to estimate the horizontal wave number and calculate the estimated broadband horizontal wave number k rm and the true horizontal wave number error, and obtain the fifth-order k rm The average frequency error rates are 0.1624%, 0.3279%, 0.5929%, 0.9958% and 2.7046% respectively, and the overall average error rate is 0.9567%. It can be seen that the multi-distance sparse Bayesian learning method for estimating horizontal wavenumbers has a low overall error rate. The phase velocities at multiple frequency points are finally obtained as follows Figure 3 The black scattered points are shown in the figure, and the red line is the fitting curve. It can be seen that the final fitting result is consistent with the trend of the original data.

[0121] After obtaining the fitting curve, we can then invert the environmental parameters matching the phase velocity. We invert the seabed sound velocity c b , density ρ b and sea depth H, while the attenuation is considered to be known. In the process of searching for 3D environmental parameters, in order to reduce the amount and improve the inversion efficiency, we use the Bayesian optimization algorithm. Figure 4 Graphs of the Bayesian optimization process and results. Figure 4In the figure, the orange dots represent the evaluation points selected in each of the 1000 iterations. In that iteration, the expected return corresponding to that set of environmental parameters was the highest. The red stars represent the true values ​​of the environmental parameters, and the blue squares represent the final parameter inversion results from Bayesian optimization, corresponding to the minimum cost function. During the optimization process, the inversion results of the environmental parameters converged towards the true values.

[0122] Finally, the inversion result obtained by Bayesian optimization algorithm for matching phase velocity inversion is H = 39.86m, c b =1685.0m / s,ρ b =1.92g / cm 3 , very close to the real H=40m,c b =1680m / s,ρ b =1.9g / cm 3 Multi-distance sparse Bayesian learning fitting curve and matching inversion results corresponding to fc p The curves are shown in Figure 5(a) below. The two are very close, indicating that minimizing MSE can find the curve closest to the fitting curve in the matching field, and then find the closest environmental parameters. Compare the fc corresponding to the environmental parameters set in the simulation with the fc corresponding to the matching inversion. p The curves are shown in Figure 5(b) below. The two are very close and almost overlap. This shows that matching the broadband phase velocity inversion environmental parameters has a very good effect and can obtain relatively accurate environmental parameters.

[0123] The inverted water depth, seabed sound velocity, density, and known attenuation coefficient were then entered into Kraken to calculate the copy field for MFP positioning in the 100-300 Hz frequency band. The sound source localization results are shown in Figures 6(a) and 6(b).

[0124] As can be seen, both the MFP positioning distance (Figure 6(a)) and depth (Figure 6(b)) agree well with the true values. The average error in distance positioning is 0.23 km, with an average error rate of 0.78%. The average error in depth positioning is 0.55 m, accounting for 1.38% of the full ocean depth of 40 m. This indicates that the inverted seafloor parameters are valid, demonstrating the feasibility of the matched broadband phase velocity inversion method based on multi-range sparse Bayesian learning.

[0125] Example 2: Experimental Environment

[0126] The experimental environment has been described in Example 1. Because the seafloor attenuation coefficient is insensitive to the proposed matched phase velocity method and is decoupled from other parameters, we first use the matched phase velocity to invert the seafloor sound velocity and density without involving the seafloor attenuation coefficient. We then use the matched propagation loss to invert the attenuation coefficient, allowing subsequent matching field positioning to verify the inversion results.

[0127] Using the received signals of 59 distances to perform multi-distance sparse Bayesian learning in the range of 50-500 Hz, the first five horizontal wave numbers are selected to obtain the corresponding phase velocity as follows Figure 7 As shown, it can be seen Figure 7 There is an obvious fifth-order curve in the fitting curve. Figure 7 The red line shows the original data, and the black scattered points are the original data.

[0128] Then, the 100-450 Hz signal was selected to match the broadband phase velocity inversion seabed parameters. The Bayesian optimization algorithm was used in the parameter tour process. The search results showed that H was 38.96 m and c b is 1608.97 m / s, ρ b is 1.97g / cm3. Now draw the fc of broadband multi-distance sparse Bayesian learning p The fitting curve (red line) and the inverted seafloor parameter results are introduced into KRAKEN to obtain fc p The curve (black line) is as follows Figure 8 As shown, it can be seen that the two are highly similar.

[0129] The attenuation coefficient was inverted using the matching propagation loss method. 59 distance points on the survey line were selected to calculate the propagation loss. Eight frequency points within 100-350 Hz were selected as the center frequency of the propagation loss. The attenuation coefficient was searched by matching the experimental propagation loss with the propagation loss calculated by KRAKEN (the water depth, seabed sound velocity, and density obtained by the above inversion were introduced into KRAKEN at this time). The analytical expression of the nonlinear fitting curve of the attenuation coefficient changing with frequency is α b =0.8390f 1.94 , where the unit of f is kHz.

[0130] The inverted local water depth is 38.96 m, while the experimentally measured water depth at the VLA location is 38.5 m, resulting in a measurement error of 0.46 m. ​​The measured water depth along the survey line is then added to the total depth by 0.46 m to obtain the new depth. The inverted seafloor acoustic parameters (seafloor velocity, density, and attenuation coefficient) are then combined and fed into KRAKEN to generate a copy field. Broadband matched field positioning (range and depth) is performed on 59 sound source points using a frequency of 100 to 300 Hz. The results are shown in Figures 9(a) and 9(b), where the red stars represent the true locations of the sound sources and the blue circles represent the MFP positioning results. As can be seen, the distance positioning results are generally consistent with the GPS records, with an average error of 0.81 km and an average error rate of 2.94%. The depth positioning is also close to the true depth of 7 m, with an average error of 1.46 m, representing 3.64% of the full sea depth of 40 m. The above results show that the attenuation coefficient obtained by matching the seabed sound velocity, density and water depth obtained by matching the phase velocity inversion and matching the propagation loss is relatively reliable. To a certain extent, it illustrates the feasibility and effectiveness of the method of multi-distance sparse Bayesian learning to match broadband phase velocity inversion environmental parameters.

[0131] For comparison, the received data was subjected to traditional matching field inversion. Traditional matching field inversion can simultaneously invert multiple parameters such as water depth, seabed sound speed, density, attenuation, etc., but it is more complicated to apply in scenarios with horizontal changes. Here, it is generally assumed that the water depth is horizontally constant, and the water depth obtained by inversion is the average depth of the entire area. Broadband matching field inversion was performed on 59 distance points respectively. Unlike the method proposed in the present invention, traditional matching field inversion requires accurate sound source location information, so we brought the 59 sound source distances recorded by GPS and the nominal depth of 7m into the inversion. The cost function of the 59 points is shown in Figure 10 (a). It can be seen that the inverted cost function is the smallest when the third distance point is taken, reaching 0.2143. At the same time, the inversion results of water depth (Figure 10(b)), seabed sound velocity (Figure 10(c)), density (Figure 10(d)), and attenuation coefficient (Figure 10(e)) at different distances are plotted. It can be seen that the four parameters are very scattered within the search range. This is because the solution of the matched field inversion has a multi-valued characteristic (that is, the copy field calculated using different combinations of multi-dimensional parameters will have similar sound field characteristics to the measured sound field), and the matched field inversion is a separate inversion for each distance. Therefore, even points with close distances may have very different inversion results. However, the results of the received signal inversion using different distances in the same sea area should not vary greatly. Therefore, it is necessary to consider which distance points to select as the representative final results. This is also a problem usually faced by using matched field inversion. Here, since the inversion results of the parameters at each distance point are relatively scattered, the seabed parameter corresponding to the minimum cost function is taken as the inversion result. The corresponding seabed parameter is H = 39.90m, c b=1601.60m / s,ρ b =2.05g / cm3,α b = 0.13dB / λ. The same KRAKEN model is substituted to generate a copy field, yielding the positioning results shown in Figures 9(c) and 9(d). It can be seen that range positioning performance deteriorates at long distances. The average error for range positioning is 2.28 km, with an average error rate of 6.35%. Depth positioning, while somewhat better at close range, is significantly poor at long distances, with an average error of 13.54 m, representing 33.86% of the total 40 m water depth. Traditional matched field inversion is generally used in environments with minimal horizontal variation. When inverted using a horizontally invariant model, the inversion result approximates an average. Therefore, using this model in situations where horizontal variation is significant can result in errors. When used for matched field positioning, the results of broadband phase velocity inversion using multi-range sparse Bayesian learning matching achieve superior depth positioning results compared to matched field inversion, resolving the difficulty in accurately determining depth often encountered in MFP. Furthermore, the distance results are superior to those of traditional methods. We also need to note that the traditional matching field inversion of seabed parameters uses relatively accurate sound source information, while the matching phase velocity inversion method does not use sound source information and has achieved good results in subsequent positioning.

[0132] In summary, in the present invention, multi-distance sparse Bayesian learning does not require geoacoustic parameters or sound source information when extracting normal modes, making it simpler and more efficient. Using information from multiple distance points at each frequency point to estimate the horizontal wavenumber removes the bandwidth limitation and allows the use of broadband signals to improve accuracy. During inversion, the present invention decouples the seabed attenuation coefficient so that it can be inverted independently, reducing the interaction between parameters. Matching phase velocity inversion does not require sound source location information, and the inversion accuracy does not have to be affected by the accuracy of the sound source location like traditional matching field inversion. The proposed method uses the received data of multiple distance points at the same time, which not only reduces the large amount of calculation required for matching field inversion to perform separate inversions of different distance signals one by one, but also overcomes the problem of lack of representativeness of isolated point inversion results. Matching phase velocity inversion matches the local phase velocity and is not affected by terrain changes on the propagation path, so that matching phase velocity inversion can perform geoacoustic inversion and depth correction regardless of whether the terrain level remains unchanged or changes. In contrast, matched-field inversion matches the acoustic pressure, which is affected by the propagation path and is sensitive to topographic mismatch, especially when a horizontally invariant model is applied to a horizontally varying seafloor environment. These advantages make matched phase velocity inversion more robust and accurate than traditional matched-field inversion.

[0133] The present application may also provide a computer device comprising: at least one processor, memory, at least one network interface, and a user interface. The various components in the device are coupled together via a bus system. It will be understood that the bus system is used to enable connectivity and communication between these components. In addition to a data bus, the bus system also includes a power bus, a control bus, and a status signal bus.

[0134] The user interface may include a display, a keyboard, or a pointing device, such as a mouse, a trackball, a touchpad, or a touch screen.

[0135] It is understood that the memory in the embodiments disclosed in the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDRSDRAM), enhanced synchronous DRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DRRAM). The memories described herein are intended to include, but are not limited to, these and any other suitable types of memory.

[0136] In some embodiments, the memory stores the following elements, executable modules or data structures, or a subset or an extension thereof: an operating system and applications.

[0137] The operating system includes various system programs, such as the framework layer, core library layer, and driver layer, which are used to implement various basic services and handle hardware-based tasks. Application programs include various application programs, such as media players and browsers, which are used to implement various application services. The program that implements the method of the embodiment of the present disclosure can be included in the application program.

[0138] In the above embodiment, the processor may also call a program or instruction stored in the memory, specifically, a program or instruction stored in the application program, to:

[0139] Perform the steps of the above method.

[0140] The above method can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in the processor or by software instructions. The above processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The above-disclosed methods, steps, and logic block diagrams can be implemented or executed. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the above-disclosed method can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method.

[0141] It is understood that the embodiments described herein may be implemented using hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit may be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, or other electronic units or combinations thereof for performing the functions described herein.

[0142] For software implementation, the technology of the present application can be implemented by executing the functional modules (e.g., procedures, functions, etc.) of the present application. The software code can be stored in a memory and executed by a processor. The memory can be implemented in the processor or external to the processor.

[0143] The present application may also provide a non-volatile storage medium for storing a computer program. When the computer program is executed by a processor, each step in the above method embodiment can be implemented.

[0144] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions of this application and are not intended to limit the scope of the present invention. Although this application has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions to the technical solutions of this application do not depart from the spirit and scope of the technical solutions of this application and should be encompassed by the claims of this application.

Claims

1. A method for inverting geoacoustic parameters using matched broadband modal phase velocity based on sparse Bayesian learning, comprising: Step 1: Perform Fourier transform on the time domain data of sound propagation at multiple distances received by the vertical array of hydrophones to obtain the multi-range frequency domain signals for each frequency point within the required bandwidth; Step 2: Collect horizontal wave numbers within the set interval and use the shooting method to generate the corresponding local modal depth function to form the modal depth function matrix A; Step 3: The multi-distance time domain signals of each frequency point are combined into a matrix Y. The matrices A and Y are used to calculate the local modes and local horizontal wavenumbers of the actual excitation through the iterative method of sparse Bayesian learning; Step 4: Perform multi-distance sparse Bayesian learning on each frequency point within the required bandwidth and convert the obtained local horizontal wavenumber into local phase velocity; Step 5: Perform data cleaning and curve fitting on the broadband phase velocity points; Step 6: Substitute the environmental parameters to be searched into the sound propagation model to obtain the phase velocity at each frequency point; by minimizing the mean square error between the frequency-phase velocity curve obtained from the sound propagation model and the curve fitted in step 5, the environmental parameters including seabed sound velocity, density, and water depth are inverted. Step 7: Perform depth correction on the actually measured water depth by using the inverted water depth; Step 8: Invert the seabed attenuation coefficient using the matching propagation loss method; The formula of the target shooting method is: Φ(0,k rm )=0, Φ(z1,k rm )=h, j=1,2,…J Where ω = 2πf is the angular frequency, f represents the signal frequency; z j =jh, where h and J are the width and total number of discrete grids respectively; c(z j ) is the sound velocity profile; Φ(z j ,k rm ) is the mth order normal mode depth function at the jth grid depth z j The discrete sampling value at k rm represents the m-th order horizontal wave number.

2. The method for inverting geoacoustic parameters by matching broadband modal phase velocity based on sparse Bayesian learning according to claim 1 is characterized in that: The setting interval is: [2πf / c b ,2πf / c w ]; where f represents the signal frequency, c b and c w represent the speed of sound on the seabed and in water respectively.

3. The method for inverting geoacoustic parameters by matching broadband modal phase velocity based on sparse Bayesian learning according to claim 1 is characterized in that: The local mode and horizontal wave number of the real excitation are expressed as: Among them, γ m To characterize the amount of local modal energy, Indicates the value of this iteration, Indicates the value of the last iteration; L indicates the number of sound sources; Φ(z,k rm ) is the discrete sampling of the m-th order normal mode depth function at depth z at the receiver; σ 2 is the variance of the noise; Σ y is the data covariance; is a matrix composed of the K excited modal functions in the matrix A, + represents pseudo-inverse calculation, and K represents the number of real propagation modes; I N represents the N-dimensional identity matrix, where N is the number of array elements; the superscript H indicates conjugate calculation; ||·||2 represents the 2-norm formula; and tr represents the trace of the matrix.

4. The method for inverting geoacoustic parameters by matching broadband modal phase velocity based on sparse Bayesian learning according to claim 3 is characterized in that: The obtained local horizontal wave number is converted into the local phase velocity using the formula: Among them, c pm represents the local phase velocity.

5. The method for inverting geoacoustic parameters by matching broadband modal phase velocity based on sparse Bayesian learning according to claim 4 is characterized in that: The inversion obtains environmental parameters including seabed sound velocity, density and water depth, and the formula is: Among them, M d is the total number of modes used; F is the total number of frequencies; represents the set of seabed environmental parameters to be inverted; is the fitted frequency-phase velocity curve; t m (f i ,u) is the copy field frequency-phase velocity curve generated by the sound propagation model when the environmental parameter set is u; f i is the frequency.

6. A system for matching broadband modal phase velocity inversion for geoacoustic parameters based on sparse Bayesian learning, implemented based on the method of any one of claims 1 to 5, characterized in that: The system comprises: The module for obtaining multi-range frequency domain signals is used to perform Fourier transform on the time domain data of sound propagation at multiple distances received by the vertical array of hydrophones to obtain multi-range frequency domain signals for each frequency point within the required bandwidth; The module for obtaining the modal depth function matrix is ​​used to collect horizontal wave numbers within a set interval and generate the corresponding local modal depth function using the shooting method to form the modal depth function matrix A; The module for calculating local modes and local horizontal wavenumbers is used to combine the multi-distance time domain signals of each frequency point into a matrix Y. The matrices A and Y are used to calculate the local modes and local horizontal wavenumbers of the actual excitation through the iterative method of sparse Bayesian learning. The local phase velocity acquisition module is used to perform multi-distance sparse Bayesian learning on each frequency point within the required bandwidth and convert the obtained local horizontal wavenumber into local phase velocity; The curve fitting module is used to perform data cleaning and curve fitting on broadband phase velocity points; The inversion environmental parameter module is used to bring the environmental parameters to be searched into the sound propagation model to obtain the phase velocity of each frequency point. By minimizing the mean square error between the frequency-phase velocity curve calculated by the sound propagation model and the fitted curve, the environmental parameters including seabed sound velocity, density and water depth are inverted. A depth correction module, configured to perform depth correction on the water depth obtained by inversion compared with the actual measured water depth; and The seabed attenuation coefficient inversion module is used to invert the seabed attenuation coefficient using the matching propagation loss method.

Images