A method and system for predicting the interference structure of direct sound zones taking into account wind wave conditions

Abstract

The application discloses a kind of direct sound zone interference structure prediction method and system considering wind wave condition, the method includes: based on wind wave spectrum space-time sequence with direction spectrum, complete dynamic sea surface reflection coefficient prediction, construct three-dimensional bubble particle size spectrum distribution function, calculate to obtain wideband sound propagation attenuation compensation factor;Determine parameter library index dimension, combined with the above two, complete the quick prediction and matching of interference structure characteristic parameter library, obtain corresponding characteristic parameters;Combined with the acoustic environment parameters and characteristic parameters at each forecast time, complete the advance forecast, generate time series interference structure prediction result;Finally, construct the hybrid prediction model of physical model and high-precision reference model, realize direct sound zone interference structure prediction by weighted fusion strategy.The application can realize full-scale quick prediction of direct sound zone interference structure under wind wave condition.The application is a kind of direct sound zone interference structure prediction method and system considering wind wave condition, and can be widely applied in the field of underwater acoustic engineering technology.

Description

Technical Field

[0001] This invention relates to the field of underwater acoustic engineering technology, and in particular to a method and system for predicting interference structures in the direct sound zone considering wind and wave conditions. Background Technology

[0002] Target detection and localization in deep-sea environments is a research hotspot and challenge in the field of underwater acoustics, directly related to major strategic needs such as safeguarding national maritime rights, resource exploration, and national defense security. In the deep-sea acoustic propagation environment, the direct sound zone refers to the area covered by direct sound rays that have not been ejected from the seabed. Within this zone, acoustic signal propagation loss is low and propagation characteristics are relatively stable, making it of significant application value for achieving rapid and accurate localization of medium- and long-range underwater targets. Therefore, the study of the sound field characteristics of the direct sound zone has always been a core issue of concern in the fields of underwater acoustic physics and underwater acoustic engineering.

[0003] Sound field interference structures are a significant and exploitable physical phenomenon in direct-path sound regions. When sound waves excited by a sound source reach the receiver via a direct path and a sea surface reflection path, the signals from different paths superimpose at the receiving point, interfering in the frequency domain and forming alternating bright and dark interference fringes on the frequency-distance (or frequency-pitch angle) plane. This interference structure, particularly the classic Lloyd's mirror effect, exhibits a clear physical correspondence between the fringe period, number, and distribution characteristics and parameters such as the sound source depth and distance. Utilizing this characteristic for sound source localization (i.e., interference structure matching localization) offers advantages over traditional matched-field localization methods, including lower computational complexity and greater tolerance to environmental parameter mismatches, making it a crucial development direction for passive deep-sea sound source localization technology. However, the actual marine environment is far from ideal, especially since the sea surface, as the upper boundary of the ocean waveguide, has a crucial impact on sound propagation. In practical applications, the sea surface is often affected by wind and waves, becoming a rough, randomly undulating interface, and forming a bubble-mixing layer near the sea surface caused by wave breaking. These factors together constitute a complex marine environment, posing a severe challenge to acoustic field interferometric structure prediction and matching localization methods under ideal conditions.

[0004] However, existing correction methods are mostly empirical or statistical corrections, lacking a deep understanding of the real-time coupling mechanism between the dynamic processes of wind and waves and the acoustic field interference structure, and the physical foundation of the forecast model needs to be further strengthened. Secondly, in terms of "rapid forecasting," high-precision models (such as parabolic equation models considering rough sea surfaces) have a large computational load, making it difficult to meet the requirements of real-time processing; while methods based on analytical formulas are fast, their accuracy is insufficient under complex sea conditions. Summary of the Invention

[0005] To address the aforementioned technical problems, the present invention aims to provide a method and system for predicting direct sound region interferometric structures considering wind and wave conditions, which can achieve rapid full-scale prediction of direct sound region interferometric structures under wind and wave conditions.

[0006] The first technical solution adopted in this invention is: a method for predicting the interferometric structure of a direct acoustic region considering wind and wave conditions, comprising the following steps:

[0007] The dynamic sea surface reflection coefficient is obtained by acquiring the spatiotemporal sequence of wind and wave spectra with directional spectra and predicting the dynamic sea surface reflection coefficient.

[0008] A three-dimensional bubble particle size distribution function is constructed and broadband sound propagation attenuation compensation is performed to obtain the broadband sound propagation attenuation compensation factor.

[0009] The index dimension of the parameter library is determined, and the dynamic sea surface reflection coefficient and broadband acoustic propagation attenuation compensation factor are combined to quickly predict and match the interferometric structure feature parameter library to obtain the interferometric structure feature parameters.

[0010] Acoustic environment parameters at each forecast time are obtained, and advanced forecasts are made by combining them with interferometric structure characteristic parameters to generate interferometric structure forecast results at each time.

[0011] A hybrid prediction model is constructed by combining a physical model and a high-precision reference model. Based on the prediction results of the interferometric structure at each time point, a weighted summation fusion strategy is used to predict the interferometric structure in the direct sound region.

[0012] Furthermore, the step of obtaining the dynamic sea surface reflection coefficient by acquiring the spatiotemporal sequence of the wind and wave spectrum with directional spectrum and predicting the dynamic sea surface reflection coefficient specifically includes:

[0013] Obtain sea surface wind field data;

[0014] Considering wind energy input, wave breaking dissipation, and nonlinear wave-wave interaction terms, sea surface wind field data is input into the wave action balance equation of the wave spectrum model to obtain a wind wave spectrum spatiotemporal sequence with directional spectrum.

[0015] The spatiotemporal sequence of wind and wave spectrum is divided into several frequency bands. Considering the energy weight and directional distribution of each frequency component, random initial phases are assigned to the wave number components corresponding to several frequency bands. The spectral domain is converted to the spatial domain by fast Fourier transform to obtain the geometrically reconstructed sea surface geometry.

[0016] Based on the geometrically reconstructed sea surface geometry, the curvature information and normal direction of the local sea surface at that point are extracted from the sea surface elevation field according to the location and time of the intersection of the sound ray and the sea surface.

[0017] The curvature information and normal direction of the local sea surface at this point are input into a rough surface scattering model based on the Huygens-Fresnel principle, and the dynamic sea surface reflection coefficient is obtained by combining the incident angle and frequency of the sound rays.

[0018] Furthermore, the step of constructing a three-dimensional bubble size distribution function and performing broadband sound propagation attenuation compensation to obtain a broadband sound propagation attenuation compensation factor specifically includes:

[0019] A mathematical model of bubble particle size distribution as a function of depth was established based on wind speed and wave breaking intensity parameters.

[0020] Based on the fundamental distribution characteristics of nearshore surface bubble size distribution, differentiated attenuation coefficients are matched for bubbles of different sizes. Combined with a mathematical model of bubble size distribution with depth, a three-dimensional bubble size distribution function is constructed.

[0021] Based on the three-dimensional bubble size distribution function, the bubble volume fraction of the corresponding seawater layer is obtained by integrating the bubble number concentration with the average bubble volume.

[0022] The equivalent sound velocity of the seawater depth layer is obtained by correcting the bubble volume fraction using the Wood formula.

[0023] Considering the resonant scattering loss and thermoviscous dissipation loss generated by bubbles, the acoustic attenuation contribution of bubbles of all different sizes is superimposed to obtain the broadband attenuation coefficient of this seawater depth layer.

[0024] The propagation path of the sound ray is determined based on the equivalent sound velocity. The path of the sound ray through the bubble mixing layer is discretized into several small segments. The attenuation of the sound pressure amplitude is calculated for each small segment based on the corresponding broadband attenuation coefficient.

[0025] The attenuation of the sound pressure amplitude is discretized, and the attenuation is integrated along the sound propagation path to obtain the total attenuation of the frequency component along the entire path, which is used as the broadband sound propagation attenuation compensation factor.

[0026] Furthermore, the step of determining the index dimension of the parameter library, combining the dynamic sea surface reflection coefficient and the broadband acoustic propagation attenuation compensation factor, and performing rapid prediction and matching of the interferometric structure feature parameter library to obtain the interferometric structure feature parameters specifically includes:

[0027] The index dimensions of the parameter library are determined, and key feature parameters are extracted by combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor to construct a feature parameter library.

[0028] Based on the current environmental parameters, multidimensional interpolation retrieval is performed in the feature parameter library to obtain preliminary interference structure feature parameters;

[0029] The measured interference structure characteristic parameters are obtained by measuring actual data or by determining the characteristic parameters of parameter points, and then compared with the preliminary interference structure characteristic parameters to construct correction coefficients.

[0030] The preliminary interference structure characteristic parameters are locally calibrated based on the correction coefficients to obtain the interference structure characteristic parameters.

[0031] Furthermore, the step of determining the index dimension of the parameter library, extracting key feature parameters by combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor, and constructing the feature parameter library specifically includes:

[0032] The index dimensions of the parameter library are determined, including sound source depth, horizontal distance, sound wave frequency, receiver array depth, and wind speed level.

[0033] Based on the index dimensions of the parameter library, combined with the dynamic sea surface reflection coefficient and the broadband acoustic propagation attenuation compensation factor, frequency-distance interferograms and angle-of-arrival interferograms are generated.

[0034] Key feature parameters were extracted from frequency-range interferograms and angle-of-arrival interferograms to construct a feature parameter library.

[0035] Furthermore, the step of obtaining acoustic environment parameters at each forecast time, combining them with interferometric structure characteristic parameters for advanced forecasting, and generating interferometric structure forecast results for each time specifically includes:

[0036] Establish a data interface with the meteorological and oceanographic forecasting center to obtain numerical forecast products;

[0037] Based on numerical weather prediction products, the data on the evolution of sea surface conditions in the future are determined. The acoustic environmental parameters for each forecast time are derived by combining the response time scale of bubble layer formation and dissipation.

[0038] Based on the acoustic environment parameters at each forecast time, combined with the dynamic sea surface reflection coefficient, broadband sound propagation attenuation compensation factor, and interference structure characteristic parameters, the sound field is driven by a preset accelerated calculation strategy to generate the interference structure forecast results at each time.

[0039] Furthermore, the preset accelerated computing strategy specifically includes:

[0040] The process of rapid prediction and matching of the interference structure feature parameter library is time-series reused, that is, the parameter library retrieval and matching are only performed again when the environmental parameters change beyond the preset threshold.

[0041] The physical model calculations are coarse-grained in time, and linear interpolation is performed between adjacent time points to reduce the number of actual sound field calculations.

[0042] Furthermore, the step of constructing a hybrid prediction model by combining a physical model and a high-precision reference model, and using a weighted summation fusion strategy based on the interferometric structure prediction results at each time point to achieve the prediction of the interferometric structure in the direct acoustic region, specifically includes:

[0043] By combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor, the sea surface reflection is equivalent to virtual source radiation, and coherent superposition compensation is performed to generate frequency-distance or frequency-elevation angle interferograms.

[0044] Based on the frequency-distance or frequency-elevation angle interferometry pattern, determine the physical model and high-precision reference model, and determine the forecast lead time requirement based on the interferometric structure prediction results at each time point;

[0045] Training data is generated by sampling within a typical environmental parameter space. The interference structure is calculated using both a physical model and a high-precision reference model to obtain the environmental parameter residual compensation values.

[0046] The neural network is trained based on the residual compensation values ​​of environmental parameters, and the fusion weights of the physical model output and the neural network residual output are dynamically determined according to the current environmental parameters and forecast timeliness requirements.

[0047] If the environmental parameters are within the coverage of the training data and the forecast lead time requirement is higher than the mean, the weight of the residual output of the neural network is greater than the weight of the physical model output. If the environmental parameters are in the sparse region of the training data or long-term series advance forecasting is required, the weight of the residual output of the neural network is less than the weight of the physical model output.

[0048] By employing a weighted summation fusion strategy, the superposition result of the physical model calculation and the residual compensation value is output, thereby enabling the prediction of interference structures in the direct acoustic region.

[0049] Furthermore, the step of combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor to treat sea surface reflection as equivalent to virtual source radiation, performing coherent superposition compensation, and generating frequency-distance or frequency-elevation angle interferograms specifically includes:

[0050] Given the location of the sound source, the location of the receiving array, the frequency range, and the wind speed level, the sea surface reflection is equivalent to virtual source radiation by combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor.

[0051] Based on virtual source radiation, considering the modulation of amplitude and phase by the dynamic reflection coefficient and the compensation of amplitude by path attenuation, the sound pressure of the direct wave and the reflected wave are coherently superimposed and compensated to obtain the sound pressure spectrum at the receiving point.

[0052] Based on the sound pressure spectrum at the receiving point, frequency-distance or frequency-elevation angle interference patterns are generated.

[0053] The second technical solution adopted in this invention is: a direct acoustic region interferometric structure prediction system considering wind and wave conditions, comprising:

[0054] The first module is used to obtain the spatiotemporal sequence of wind and wave spectra with directional spectra to predict the dynamic sea surface reflection coefficient, thereby obtaining the dynamic sea surface reflection coefficient.

[0055] The second module is used to construct a three-dimensional bubble particle size distribution function and perform broadband sound propagation attenuation compensation to obtain a broadband sound propagation attenuation compensation factor.

[0056] The third module is used to determine the index dimension of the parameter library, and combine the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor to quickly predict and match the interference structure feature parameter library to obtain the interference structure feature parameters.

[0057] The fourth module is used to obtain acoustic environment parameters at each forecast time, combine them with interferometric structure characteristic parameters to make advance forecasts, and generate interferometric structure forecast results at each time.

[0058] The fifth module is used to construct a hybrid prediction model by combining the physical model and the high-precision reference model. Based on the prediction results of the interferometric structure at each time, a weighted summation fusion strategy is used to predict the interferometric structure in the direct sound region.

[0059] The beneficial effects of the method and system of this invention are as follows: This invention obtains the dynamic sea surface reflection coefficient by acquiring the spatiotemporal sequence of wind and wave spectra with directional spectra and predicting the dynamic sea surface reflection coefficient. The calculation process of the dynamic reflection coefficient incorporates the differences in the contribution of different frequency wave components to sound reflection during the evolution of the wind and wave spectrum into the same physical framework, realizing the real-time dynamic characterization of sea surface reflection characteristics as they evolve with wind and waves. Furthermore, a three-dimensional bubble size distribution function is constructed and broadband sound propagation attenuation compensation is performed to obtain a broadband sound propagation attenuation compensation factor. In the ray tracing or virtual source method framework of sound field prediction, the layered attenuation coefficient is integrally compensated along the propagation path to form a frequency-dependent and path-dependent total attenuation correction, realizing refined compensation for broadband sound propagation loss under wind and wave conditions. Further, the index dimension of the parameter library is determined, and the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor are combined to perform rapid prediction and matching of the interference structure feature parameter library, obtaining the interference structure feature parameters. The search results are then evaluated using a small amount of measured sound field data or high-precision numerical model calculation results. Local corrections are made to eliminate errors caused by parameter library discretization. Then, acoustic environment parameters at each forecast time are obtained, and advanced forecasts are performed using interferometric structure characteristic parameters to generate interferometric structure forecast results for each time period. These results can guide the adjustment of underwater platform detection and positioning strategies in future periods. Finally, a hybrid forecast model is constructed by combining a physical model and a high-precision reference model. Based on the interferometric structure forecast results at each time period, a weighted summation fusion strategy is used to predict the interferometric structure in the direct sound zone. Through adaptive fusion, the optimal balance between accuracy and speed can be achieved in different application scenarios, while ensuring the physical rationality of the forecast results. This invention comprehensively considers the coupling effects of the air-sea interface, water acoustics, and stochastic statistical multiphysics fields. It considers both the influence of wind and waves on sea surface reflection and bubble layers, and also characterizes the changes in the sound velocity profile of shallow water caused by wind and waves. Simultaneously, the interferometric structure is treated as a stochastic field, and through the fusion of the physical model and neural network, the statistical characteristics of the interferometric structure are predicted, meeting the practical needs of engineering for forecast reliability and positioning algorithm parameter selection. Attached Figure Description

[0060] Figure 1 This is a flowchart of the steps of the method for predicting the interference structure in the direct sound region considering wind and wave conditions according to the present invention;

[0061] Figure 2 This is a structural block diagram of a direct acoustic region interference structure prediction system that takes into account wind and wave conditions, according to the present invention. Detailed Implementation

[0062] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.

[0063] First, it should be noted that the existing technical solutions have the following shortcomings, for example:

[0064] 1) Existing technologies simplify and fragment the description of the impact mechanism of wind and waves, lacking a dynamically coupled physical model:

[0065] Existing direct-sound interferometric structure prediction techniques generally suffer from shortcomings in handling sea surface boundary conditions, including idealized assumptions and insufficient mechanistic characterization. On one hand, most existing methods still treat the sea surface as an ideal, smooth horizontal interface, considering only the coherent superposition of the direct wave and the first reflected wave, neglecting the random undulations of the real sea surface under the influence of wind and waves, and the near-surface bubble layer generated by wave breaking. While this idealized boundary assumption simplifies the model and improves computational efficiency, it directly leads to significant deviations in the prediction of reflected wave amplitude and phase under actual sea conditions. When the sea surface undulations are comparable to the sound wave wavelength, the reflected wave energy is scattered, reducing the contrast of the interference fringes. Furthermore, the presence of the bubble layer not only alters the near-surface sound velocity profile but also causes additional attenuation of high-frequency sound waves, distorting the amplitude and frequency characteristics of the interferometric structure. On the other hand, the few studies that have attempted to consider the influence of sea surface wind and waves mostly employ a step-by-step decoupled approach: first, a randomly undulating sea surface is generated using the wind and wave spectrum; then, an empirical bubble layer attenuation model is introduced; and finally, statistical averaging or empirical correction is performed in the sound field calculation. While this approach reflects the macroscopic effects of wind and waves to some extent, it essentially treats sea surface undulations, bubble layers, and sound field propagation as relatively independent components in a series calculation. It fails to reveal the dynamic coupling and nonlinear interaction mechanisms among these three elements, particularly lacking a physical description of the real-time correspondence between dynamic wind and wave processes (such as the evolution of sea surface undulations over time and the formation and dissipation of bubble layers) and the sound field interference structure. This simplification and fragmentation at the mechanistic level limits the physical consistency and generalization ability of the forecast model under different sea states, making it difficult to fundamentally guarantee the reliability of the forecast results in real, complex marine environments.

[0066] 2) Existing forecasting methods struggle to achieve an effective balance between accuracy and speed, resulting in insufficient rapid forecasting capabilities:

[0067] In the engineering forecasting of interferometric structures in direct sound regions, the contradiction between computational accuracy and computational efficiency is the core bottleneck currently facing the technology. On the one hand, high-precision numerical models of sound fields based on parabolic equations, finite element methods, or normal modes can theoretically handle undulating sea surfaces and random media problems with relatively high accuracy. However, their computational load increases dramatically with increasing frequency, expanding sea area, and increasing number of sea surface realizations, making it difficult to meet the needs of real-time or near-real-time rapid forecasting. Especially in application scenarios that require parameter scanning or probability statistics for different wind speeds and sea states, the computational overhead of such methods makes them almost impractical for engineering. On the other hand, existing rapid forecasting methods mostly follow the framework of traditional ray theory and virtual source methods, using analytical formulas to calculate the coherent superposition of direct waves and sea surface reflected waves. These methods have high computational efficiency and clear physical meaning of parameters, but their ability to handle sea surface boundaries is severely limited. When wind and wave factors are introduced, complex effects such as the reflection coefficient caused by rough sea surfaces changing with the grazing angle, phase dispersion caused by random undulations, and frequency-selective attenuation of bubble layers are difficult to be uniformly represented by simple analytical expressions. To compensate for this deficiency, existing technologies often approximate the impact of wind and waves by introducing empirical reflection loss correction terms or statistical averaging factors. However, this a posteriori correction method is essentially a "patchwork" strategy, lacking both physical prior predictive capabilities and often sacrificing forecast accuracy for computational speed improvements. Therefore, in the current technological system, high-precision models have an excessive computational burden and are difficult to apply in real time, while fast models are severely inadequate in representing the impact of wind and waves. There is a lack of a middle ground that balances physical realism and computational efficiency. This imbalance between accuracy and speed often puts the prediction of direct acoustic interferometric structures under wind and wave conditions in a dilemma in practical engineering, making it difficult to meet the dual requirements of real-time performance and accuracy for applications such as rapid underwater target detection and location.

[0068] 3) Existing technologies have weak closed-loop capabilities for interferometric structure prediction and sound source localization applications:

[0069] One of the ultimate goals of direct-arrival interferometric structure prediction is to provide reliable characteristic data for sound source localization. However, existing technologies exhibit a significant disconnect between prediction and localization. First, most existing prediction methods only output macroscopic features of the sound field interferometric structure (such as the spacing and contrast of interference fringes), failing to fully consider the inherent matching relationship between these features and the receiver array configuration and signal processing methods in localization applications. For example, in the angle-of-arrival domain interferometric structure localization method based on vertical linear arrays, the array manifold, beamwidth, and sidelobe characteristics of the beamformer significantly affect the actual observed interferometric pattern. Existing prediction models often only provide the sound pressure interferometric structure at the ideal receiver point, without incorporating the array processing stage into the unified prediction framework, leading to a systematic deviation between the predicted features and the actually processed features. Second, existing localization methods generally exhibit poor robustness to environmental mismatch when facing wind and waves. Traditional interferometric structure matching localization methods are typically based on an interferometric structure template library under ideal sea surface conditions. When wind and waves are present in actual sea conditions, distortion and mismatch occur between the measured interferometric pattern and the ideal template, resulting in a sharp decrease in localization accuracy. While some studies have proposed statistically tolerant matching methods, these are essentially passive tolerances of template mismatch, rather than actively incorporating wind and wave influences into the forecast model to generate more environmentally adaptive positioning templates. In other words, existing technologies lack an integrated "forecast-positioning" design approach, failing to construct a complete technical chain from environmental parameters (wind speed, wave height) to interferometric structure forecasting, and then to sound source positioning feature extraction and matching. This fragmented technical architecture makes it difficult for forecast results to directly serve subsequent positioning calculations, and the positioning system cannot adaptively adjust the forecast model and matching strategy according to dynamic changes in wind and wave conditions, thus limiting its robustness in real marine environments.

[0070] 4) Existing technologies do not adequately consider the comprehensive effects of multiphysics coupling:

[0071] The direct-reaching sound zone interference structure under wind and wave conditions is actually a product of the intersection of multiple disciplines such as underwater acoustics, oceanography, and signal processing. However, current technologies still fall short in the comprehensive processing of multi-physics coupling effects. On the one hand, wind and waves not only change the reflection characteristics of the sea surface but also affect the sound velocity profile and sound propagation conditions of the entire shallow water body through near-surface water mixing and enhanced turbulence. Existing research often focuses on the correction of the sea surface boundary itself, ignoring the dynamic changes in the acoustic characteristics inside the water body during wind and wave processes. This boundary-dominated simplified model is approximately valid at low wind speeds, but when the wind speed increases to the point of causing significant water mixing, the change in the sound velocity profile will directly affect the angle of arrival and time delay difference of the direct-reaching sound zone, thereby changing the overall morphology of the interference structure. On the other hand, the randomness of wind and waves determines that the interference structure is essentially a random field, rather than a deterministic fringe pattern. Current forecasting technologies for wind and wave conditions primarily rely on deterministic models. Even when statistical averaging is introduced, it is mostly a post-processing step, lacking the ability to systematically predict the statistical characteristics of interferometric structures (such as the probability distribution of fringe contrast and the variance of the interferometric period). However, in practical engineering applications, these statistical characteristics are crucial for assessing forecast reliability and guiding the selection of parameters for subsequent positioning algorithms. Therefore, current technologies have significant shortcomings in integrating the coupling effects of multiple physics fields, including air-sea interface processes, water acoustic properties, and stochastic statistical characteristics of the sound field, making it difficult to achieve rapid, full-element, full-scale forecasting of interferometric structures directly reaching the sound zone under wind and wave conditions.

[0072] Based on this, the embodiments of the present invention address the problems in the prior art, such as prediction bias caused by the ideal boundary assumption, difficulty in balancing accuracy and speed, disconnect between prediction and positioning applications, and insufficient multi-physics coupling. Therefore, an architecture combining hierarchical decoupling and collaborative fusion is adopted, constructed from the bottom up into five core modules: a dynamic sea surface reflection coefficient prediction module, a broadband sound propagation attenuation compensation module, a rapid prediction and matching module for the interferometric structure feature parameter library, a forward prediction module, and a hybrid prediction module. The modules form a closed loop through data flow and control flow: First, the dynamic sea surface reflection coefficient prediction module and the broadband sound propagation attenuation compensation module extract the time-varying influence factors of wind and waves on sea surface reflection and water sound propagation, respectively, as the basic input for sound field forecasting; second, the interferometric structure feature parameter library rapid forecasting and matching module generates a feature parameter library based on a preset environmental parameter space to achieve initial rapid forecasting; third, the advanced forecasting module integrates sea surface state numerical forecast products to generate interferometric structure forecasting results for future periods, providing predictive capabilities for underwater platforms; finally, the hybrid forecasting module uses the calculation results of the physical model as the backbone and utilizes deep neural networks to perform residual learning and compensation for complex wind and wave effects that the physical model cannot fully characterize, achieving high-precision and high-timeliness final forecast output.

[0073] Reference Figure 1This invention provides a method for predicting interferometric structures in direct acoustic regions considering wind and wave conditions. The method includes the following steps:

[0074] S100. Obtain the spatiotemporal sequence of wind and wave spectra with directional spectra to predict the dynamic sea surface reflection coefficient, and obtain the dynamic sea surface reflection coefficient.

[0075] S110, Obtain sea surface wind field data;

[0076] S120. Considering wind energy input, wave breaking dissipation, and nonlinear wave-wave interaction terms, the sea surface wind field data is input into the wave spectrum model to solve the wave action balance equation, and a wind wave spectrum spatiotemporal sequence with directional spectrum is obtained.

[0077] In this embodiment, the first step is to construct the wind and wave spectrum spatiotemporal evolution submodule. Wind field data, including wind speed, direction, and spatiotemporal distribution at a height of 10 meters above the sea surface, is obtained from the meteorological and oceanographic forecasting center. A third-generation wave spectrum model is used as the basic framework, with the wind field as the driving force. The spatiotemporal evolution of the wind and wave spectrum is achieved by solving the wave action balance equation. The balance equation considers wind energy input, wave breaking dissipation, and nonlinear wave-wave interaction terms to reflect the energy transfer between wave components of different frequencies. Spatial discretization uses the finite volume method, and time progression uses an explicit format with adaptive step size to ensure that the spatiotemporal sequence of the wind and wave spectrum is output with a minute-level time resolution within a typical sea area (10 km × 10 km range). The wind and wave spectrum output by this submodule is expressed in the form of a directional spectrum, with each grid point containing a frequency spectrum and a directional distribution function.

[0078] S130. Divide the wind and wave spectrum spatiotemporal sequence into several frequency bands, consider the energy weight and directional distribution of each frequency component, assign random initial phases to the wave number components corresponding to several frequency bands, and convert the spectral domain to the spatial domain through fast Fourier transform to obtain the geometrically reconstructed sea surface geometry.

[0079] In this embodiment, the next step is the implementation of the sea surface microscale geometry reconstruction submodule. Based on the directional spectrum output by the wind and wave spectrum spatiotemporal evolution submodule, a two-dimensional sea surface elevation field is generated using a linear superposition method. Specifically, the frequency range is divided into several frequency bands, and the wave number component corresponding to each frequency band is assigned a random initial phase. During superposition, the energy weight and directional distribution of each frequency component are considered. To improve computational efficiency, a fast Fourier transform is used to convert the spectral domain to the spatial domain, and the size of the generated spatial grid matches the size of the sea surface region. Since the wind and wave spectrum evolves over time, the elevation field is also updated over time, thus achieving dynamic reconstruction of the sea surface geometry. To balance computational accuracy and efficiency, the time update step can be set to 1 / 10 of the wave spectrum evolution timescale, and the spatial grid resolution can be set to 1 / 5 of the highest wave number component.

[0080] S140. Based on the geometrically reconstructed sea surface geometry, extract the curvature information and normal direction of the local sea surface at the point of intersection of the sound ray and the sea surface from the sea surface elevation field according to the location and time of the intersection.

[0081] S150. Input the curvature information and normal direction of the local sea surface at this point into the rough surface scattering model based on the Huygens-Fresnel principle, and calculate the dynamic sea surface reflection coefficient by combining the incident angle and frequency of the sound rays.

[0082] In this embodiment, the final step is the implementation of the dynamic reflection coefficient calculation submodule. Along each sound ray path, based on the position and time of the intersection point between the sound ray and the sea surface, the curvature information and normal direction of the local sea surface near that point are extracted from the sea surface elevation field. Using the sound wave incident angle, frequency, and local sea surface curvature as input, a rough surface scattering model based on the Huygens-Fresnel principle is employed to calculate the reflection coefficient. Specifically, the local sea surface is approximated as a surface with a specific radius of curvature, and the amplitude and phase changes of the reflection coefficient are obtained using the Kirchhoff approximation. For different frequency components, their reflection coefficients are calculated separately, forming a frequency-dependent reflection coefficient vector. This submodule outputs a dynamic reflection coefficient that varies with time and space, its physical significance being that it can realistically reflect the comprehensive modulation effect of waves at different scales on sound reflection during the evolution of the wind and wave spectrum.

[0083] S200. Construct a three-dimensional bubble particle size distribution function and perform broadband sound propagation attenuation compensation to obtain the broadband sound propagation attenuation compensation factor.

[0084] S210. Based on wind speed and wave breaking intensity parameters, a mathematical model of bubble particle size distribution with depth is established.

[0085] S220. Based on the basic distribution characteristics of the bubble size spectrum in the nearshore surface layer, differentiated attenuation coefficients are matched for bubbles of different sizes. Combined with a mathematical model of the bubble size spectrum changing with depth, a three-dimensional bubble size spectrum distribution function is constructed.

[0086] In this embodiment, the first step is to implement the bubble particle size distribution vertical profile modeling submodule. Based on wind speed and wave breaking intensity parameters, a mathematical model of bubble particle size distribution with depth is established. In the near-surface layer (e.g., 0-10 meters deep), bubbles are mainly entrained by wave breaking, and the particle size distribution exhibits a bimodal distribution, with small bubbles (radius < 50 micrometers) accounting for a higher proportion, while large bubbles (radius > 100 micrometers) increase significantly under strong wind conditions. As depth increases, large bubbles rapidly rise or dissolve under buoyancy, and the particle size distribution shifts towards smaller particle sizes. This module adopts a model where the bubble number density decays exponentially with depth, and bubbles of different sizes have different decay coefficients, thereby constructing a three-dimensional bubble particle size distribution function. The input parameters are wind speed and wave breaking frequency, and the output is the distribution function of bubble number concentration with particle size at each depth layer.

[0087] S230. Based on the three-dimensional bubble size distribution function, the bubble volume fraction of the corresponding seawater layer is obtained by integrating the bubble number concentration with the average bubble volume.

[0088] S240. The volume fraction of the bubbles is corrected for sound velocity using the Wood formula to obtain the equivalent sound velocity of the seawater depth layer.

[0089] S250. Considering the resonance scattering loss and thermal viscosity dissipation loss generated by bubbles, the acoustic attenuation contribution of bubbles of all different sizes is superimposed to obtain the broadband attenuation coefficient of the seawater depth layer.

[0090] In this embodiment, the next step is the implementation of the submodule for calculating the layered sound velocity and attenuation coefficient. Based on the bubble particle size distribution function, the influence of the bubble mixing layer on sound propagation is calculated. Sound velocity correction uses the Wood formula, calculating the equivalent sound velocity based on the bubble volume fraction, which is obtained from the bubble number concentration and the average volume integral. Sound attenuation calculation employs a sound wave and bubble resonance absorption model, considering the resonance scattering and thermoviscous dissipation of bubbles at different frequencies. The contributions of all bubble sizes at each depth layer are superimposed to obtain the broadband attenuation coefficient (i.e., the attenuation value corresponding to different frequencies) at that depth layer. Since the bubble particle size spectrum varies with depth, the attenuation coefficient also exhibits vertical layering characteristics, and the attenuation degree of each frequency component varies with depth.

[0091] S260. Determine the propagation path of the sound ray based on the equivalent sound velocity, and discretize the path of the sound ray through the bubble mixing layer into several small segments. Calculate the attenuation of the sound pressure amplitude for each small segment based on the corresponding broadband attenuation coefficient.

[0092] S270. Discretize the attenuation of the sound pressure amplitude and calculate the integral of the attenuation along the sound propagation path to obtain the total attenuation of the frequency component along the entire path, which is used as the broadband sound propagation attenuation compensation factor.

[0093] In this embodiment, the broadband attenuation compensation submodule is implemented. Within the ray tracing or virtual source method framework for sound field prediction, the layered attenuation coefficients are integrated along the propagation path. For each sound ray, its path through the bubble mixing layer is discretized into several small segments. The attenuation of the sound pressure amplitude for each segment is calculated based on the frequency-dependent attenuation coefficient corresponding to its depth layer. Since the attenuation coefficients differ for different frequency components, the integral compensation requires discretization of the frequency dimension, i.e., calculating the total path attenuation for each frequency component separately. The final output compensation value is a frequency-dependent amplitude attenuation factor, used to correct the amplitude of the direct wave and the sea surface reflected wave, ensuring that the prediction results accurately reflect the broadband sound propagation loss caused by wind and waves.

[0094] S300. Determine the index dimension of the parameter library, combine the dynamic sea surface reflection coefficient and the broadband acoustic propagation attenuation compensation factor, and perform rapid prediction and matching of the interference structure feature parameter library to obtain the interference structure feature parameters.

[0095] S310. Determine the index dimension of the parameter library, extract key feature parameters by combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor, and construct the feature parameter library.

[0096] Specifically, the index dimensions of the parameter library are determined, including sound source depth, horizontal distance, sound wave frequency, receiver array depth, and wind speed level. Based on the index dimensions of the parameter library, combined with the dynamic sea surface reflection coefficient and broadband sound propagation attenuation compensation factor, frequency-distance interferograms and angle-of-arrival interferograms are generated. Key feature parameters are extracted from the frequency-distance interferograms and angle-of-arrival interferograms to construct a feature parameter library.

[0097] In this embodiment, the first step is to implement the feature parameter library construction submodule. The index dimensions of the parameter library are determined, including five core dimensions: sound source depth, horizontal distance, sound wave frequency, receiver array depth, and wind speed level. The value range of each dimension covers typical application scenarios. For each parameter combination, the aforementioned dynamic sea surface reflection coefficient prediction module and broadband sound propagation attenuation compensation module are invoked. The complex amplitude of the sound pressure at the receiving point is calculated using a ray model or virtual source method, thereby generating frequency-distance interferograms and angle-of-arrival interferograms. Key feature parameters are extracted from the patterns, including: the period of the interference fringes on the frequency axis, the slope of the change on the distance axis, the frequency position of the interference nulls, and the spacing and width of the interference fringes in the angle-of-arrival domain. The feature parameters corresponding to all parameter combinations are stored in a structured database, and a multi-dimensional index is established to support rapid retrieval.

[0098] S320. Based on the current environmental parameters, perform multidimensional interpolation retrieval in the feature parameter library to obtain preliminary interference structure feature parameters;

[0099] In this embodiment, the next step is the implementation of the fast retrieval submodule. During real-time forecasting, current environmental parameters (sound source depth, distance range, receiver array depth, wind speed, etc.) are input, and a multi-dimensional interpolation retrieval is performed in the feature parameter library. Specifically, the parameter grid points surrounding the current input point in the parameter library are first located, and then linear interpolation or multi-dimensional interpolation algorithms are used to calculate the feature parameter values ​​corresponding to the current parameter points. The time complexity of the retrieval process is related to the parameter dimension, but with a preset grid and index structure, it can be completed in milliseconds. The retrieval results serve as the interferometric structure feature parameters for the initial forecast.

[0100] S330. Obtain the measured interference structure characteristic parameters through measured data or determine the characteristic parameters of the parameter points, compare them with the preliminary interference structure characteristic parameters, and construct correction coefficients.

[0101] S340. Based on the correction coefficient, perform local calibration on the preliminary interference structure characteristic parameters to obtain the interference structure characteristic parameters.

[0102] In this embodiment, the final step is the implementation of the matching correction submodule. Considering the potential errors introduced by the parameter library discretization, a small amount of measured sound field data is introduced, or a high-precision parabolic equation model is called for local calibration. If measured data exists, interference structure feature parameters are extracted from the measured data, compared with the search results in the library, correction coefficients are calculated, and these correction coefficients are applied to the search results to obtain the corrected feature parameters. If no measured data exists, a small number of representative parameter points (such as two wind speed levels near the current wind speed) can be selected, and a high-precision model is called to calculate accurate feature parameters. These parameters are then compared with the search results for the corresponding points in the library to construct a local correction function. This correction process trades low computational overhead for improved accuracy, avoiding the high computational burden of using a high-precision model across the entire parameter space.

[0103] S400: Obtain acoustic environment parameters at each forecast time, combine them with interferometric structure characteristic parameters to perform advanced forecasting, and generate interferometric structure forecasting results at each time.

[0104] S410. Establish a data interface with the meteorological and oceanographic forecasting center to obtain numerical forecast products;

[0105] In this embodiment, the first step is to implement the sea surface state numerical forecast product access and analysis submodule. A data interface is established with the meteorological and oceanographic forecasting center to periodically acquire numerical forecast products. The data format adopts the standard NetCDF or GRIB format, including elements such as wind field (10-meter wind speed u and v components), significant wave height, mean wave period, and sea surface pressure. The spatial resolution is no less than 10 kilometers, the temporal resolution is no less than 3 hours, and the forecast lead time covers the next 24 to 72 hours. The data analysis module converts the raw data into the local coordinate system and performs spatiotemporal interpolation to match it with the spatial grid and time step of the acoustic field forecast.

[0106] S420. Based on numerical weather prediction products, determine the sea surface state evolution data for future periods, and derive the acoustic environment parameters for each forecast time by combining the response time scale of bubble layer formation and dissipation.

[0107] In this embodiment, the next step is the implementation of the environmental parameter time-series extrapolation submodule. Based on the sea surface state evolution data for future periods provided by numerical weather prediction products, and combined with the response time scale of bubble layer formation and dissipation, the acoustic environmental parameters for each forecast time are extrapolated. Specifically, wind speed and significant wave height are directly taken from the corresponding elements of the numerical weather prediction products. The prediction of bubble layer thickness and bubble volume fraction requires the establishment of a response model: the formation of the bubble layer is mainly affected by wave breaking intensity, which is related to significant wave height and wave steepness, with a response time scale of approximately 15-30 minutes; the dissipation of the bubble layer is mainly controlled by bubble buoyancy and dissolution, with a dissipation time scale of approximately several hours to tens of hours. This module adopts a linear response model, using the current bubble layer state as the initial condition and the numerical weather prediction wind and wave sequence as the driving force, to extrapolate the bubble layer parameters for each future time.

[0108] S430: Based on the acoustic environment parameters at each forecast time, combined with the dynamic sea surface reflection coefficient, broadband sound propagation attenuation compensation factor, and interference structure characteristic parameters, the sound field is driven by a preset accelerated calculation strategy to generate the interference structure forecast results at each time.

[0109] In this embodiment, the final step is the implementation of the interferometric structure time-series prediction submodule. The extrapolated environmental parameters (wind speed, significant wave height, bubble layer parameters, etc.) for each future time period are sequentially input into the dynamic sea surface reflection coefficient prediction module and the broadband sound propagation attenuation compensation module to drive sound field calculations and generate interferometric structure prediction results for each time period. Since the time series of the advance prediction may be long (e.g., 24 hours), and the sound field prediction for each time period requires certain computational resources, this module adopts a two-step acceleration strategy: first, the interferometric structure feature parameter library rapid prediction and matching module is time-series reused, i.e., parameter library retrieval and matching are only performed again when environmental parameter changes exceed a threshold; second, the physical model calculations are time-coarsened, performing linear interpolation between adjacent time periods to reduce the number of actual sound field calculations. The final output is the interferometric structure prediction results in time-series form, which can be used to guide the adjustment of underwater platform detection and positioning strategies in future periods.

[0110] The S500 system combines a physical model and a high-precision reference model to construct a hybrid prediction model. Based on the prediction results of the interferometric structure at each time point, a weighted summation fusion strategy is used to predict the interferometric structure in the direct sound region.

[0111] S510. Combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor, the sea surface reflection is equivalent to virtual source radiation, and coherent superposition compensation is performed to generate frequency-distance or frequency-elevation angle interference patterns.

[0112] Specifically, given the sound source location, receiver array location, frequency range, and wind speed level, and combining the dynamic sea surface reflection coefficient and broadband sound propagation attenuation compensation factor, sea surface reflection is equivalent to virtual source radiation. Based on virtual source radiation, considering the modulation of amplitude and phase by the dynamic reflection coefficient and the compensation of amplitude by path attenuation, the sound pressure of the direct wave and the reflected wave are coherently superimposed and compensated to obtain the sound pressure spectrum at the receiving point. Based on the sound pressure spectrum at the receiving point, frequency-distance or frequency-elevation angle interferograms are generated.

[0113] In this embodiment, the implementation of the physical model calculation submodule is the first step. This submodule uses the outputs of the aforementioned dynamic sea surface reflection coefficient prediction module and broadband sound propagation attenuation compensation module as a basis, employing a ray model or virtual source method for sound field calculation. The specific calculation process is as follows: given input parameters such as sound source location, receiver array location, frequency range, and wind speed level, the dynamic sea surface reflection coefficient prediction module is first invoked to obtain the dynamic reflection coefficient under the current wind and wave conditions, and the broadband sound propagation attenuation compensation module is invoked to obtain the frequency-dependent path attenuation compensation value. Then, based on the principle of the virtual source method, sea surface reflection is equivalent to virtual source radiation, and the sound pressure of the direct wave and the reflected wave are coherently superimposed, considering the modulation of amplitude and phase by the dynamic reflection coefficient and the compensation of amplitude by path attenuation during superposition. Finally, the sound pressure spectrum at the receiving point is output, and frequency-distance or frequency-elevation angle interferometry patterns are generated accordingly. The physical model of this submodule has clear physical meaning and high computational efficiency, forming the main framework of the hybrid forecast.

[0114] S520. Based on the frequency-distance or frequency-elevation angle interferometry pattern, determine the physical model and high-precision reference model, and determine the forecast lead time requirement based on the interferometric structure prediction results at each time.

[0115] S530. Sample and generate training data in a typical environmental parameter space, and calculate the interference structure through the physical model and the high-precision reference model respectively to obtain the environmental parameter residual compensation value.

[0116] In this embodiment, the second step is the implementation of the deep neural network residual learning submodule. The goal of this submodule is to learn the residual relationship between the physical model's prediction results and the calculation results of a high-precision reference model (such as a parabolic equation model considering a rough sea surface), thereby compensating for complex wind and wave effects that the physical model fails to fully characterize. The neural network employs a lightweight fully connected or convolutional structure. Input features include wind speed, significant wave height, bubble layer thickness, sound wave frequency, sound source depth, and distance; the output is residual terms, including sound pressure amplitude residuals and phase residuals. Training data is generated by sampling within a typical environmental parameter space, calculating the interference structure for each sample using both the physical model (fast) and the high-precision reference model (slow), and calculating the difference between the two as the residual label. After training, the network can output residual compensation values ​​based on the input environmental parameters in a very short time, and its nonlinear fitting capability can capture subtle physical processes neglected in the physical model.

[0117] S540. Train the neural network based on the residual compensation values ​​of environmental parameters, and dynamically determine the fusion weights of the physical model output and the neural network residual output according to the current environmental parameters and forecast timeliness requirements.

[0118] S550. If the environmental parameters are within the coverage of the training data and the forecast timeliness requirement is higher than the mean, the weight of the residual output of the neural network is greater than the weight of the physical model output. If the environmental parameters are in the sparse region of the training data or long-term series advance forecasting is required, the weight of the residual output of the neural network is less than the weight of the physical model output.

[0119] S560: Through a weighted summation fusion strategy, the superposition result of the physical model calculation results and the residual compensation value is output to achieve the prediction of the interference structure in the direct acoustic zone.

[0120] In this embodiment, the final step is the implementation of the adaptive fusion output submodule. During actual forecasting, the fusion weights of the physical model output and the neural network residual output are dynamically determined based on current environmental parameters and forecast lead time requirements. When environmental parameters are within the training data coverage area and forecast lead time requirements are high, the neural network residual can be assigned a higher weight to fully utilize data-driven compensation capabilities. When environmental parameters are in sparse regions of the training data or require long-term series advance forecasting, the physical model is assigned a higher weight to ensure the physical consistency and stability of the forecast results. The fusion strategy uses a weighted summation method, and the final output is the superposition of the physical model calculation results and the residual compensation. Through adaptive fusion, this module can achieve the best balance between accuracy and speed in different application scenarios, while ensuring the physical rationality of the forecast results.

[0121] In summary, this embodiment of the invention first loads a pre-built interferometric structure feature parameter library and a trained deep neural network residual learning model upon system startup. Secondly, it accesses real-time sea surface state observation data and numerical forecast products via a data interface. Then, based on the user-input forecast task (including forecast lead time, target area, frequency range, etc.), it schedules various modules to work collaboratively: for current-time forecasts, the dynamic sea surface reflection coefficient prediction module and the broadband sound propagation attenuation compensation module generate environmental correction factors, which are then combined with the interferometric structure feature parameter library rapid forecasting and matching module to obtain the initial forecast results. Finally, the physical model calculation and neural network residual compensation in the hybrid forecasting module yield the final forecast output. For future-time advance forecasts, the advance forecasting module is called to obtain future environmental parameter sequences, driving other modules to generate time-series forecast results. Finally, the forecast results are output to the user terminal in the form of interferometric patterns and feature parameter sequences. This entire process achieves high-precision, high-time-response forecasting of interferometric structures directly reaching the acoustic zone under wind and wave conditions, providing reliable environmental support for underwater target detection and positioning.

[0122] Reference Figure 2 A direct-access acoustic region interferometric structure prediction system considering wind and wave conditions, comprising:

[0123] The first module 201 is used to obtain the spatiotemporal sequence of wind and wave spectrum with directional spectrum to predict the dynamic sea surface reflection coefficient and obtain the dynamic sea surface reflection coefficient.

[0124] In this embodiment, the core function of the dynamic sea surface reflection coefficient prediction module is to establish a dynamic sea surface reflection coefficient prediction method coupled with the spatiotemporal evolution of the wind and wave spectrum. It comprises a wind and wave spectrum spatiotemporal evolution submodule, a sea surface microscale geometric reconstruction submodule, and a reflection coefficient dynamic calculation submodule. The wind and wave spectrum spatiotemporal evolution submodule is driven by wind field data from the sea surface state numerical forecast product and uses nonlinear wave-wave interaction theory to extrapolate the spatiotemporal evolution of the wind and wave spectrum in real time. The sea surface microscale geometric reconstruction submodule generates a two-dimensional elevation field reflecting the instantaneous morphology of the sea surface based on the evolved wind and wave spectrum. The reflection coefficient dynamic calculation submodule calculates the dynamic reflection coefficient corresponding to each sound ray path by combining the sound wave incident angle, frequency, and instantaneous sea surface curvature distribution. The dynamic reflection coefficient output by this module not only varies with time and space but also reflects the differences in the contribution of different frequency components to the reflection coefficient during the wind and wave spectrum evolution process.

[0125] The second module 202 is used to construct a three-dimensional bubble particle size distribution function and perform broadband sound propagation attenuation compensation to obtain a broadband sound propagation attenuation compensation factor.

[0126] In this embodiment, the core function of the broadband sound propagation attenuation compensation module is to establish a broadband sound propagation attenuation compensation method based on the layered evolution of bubble particle size spectrum. It comprises a bubble particle size spectrum vertical profile modeling submodule, a layered sound velocity and attenuation coefficient calculation submodule, and a broadband attenuation compensation submodule. The bubble particle size spectrum vertical profile modeling submodule establishes an evolution model of bubble particle size with depth and time based on wind speed, wave breaking intensity, and bubble buoyancy dynamics. The layered sound velocity and attenuation coefficient calculation submodule, based on the bubble particle size spectrum distribution, uses a physical model of sound wave-bubble interaction to calculate the equivalent sound velocity correction and broadband attenuation coefficient at each depth layer. The broadband attenuation compensation submodule, within the framework of ray tracing or virtual source method, integrates the layered attenuation coefficient along the propagation path to form a frequency-dependent and path-dependent total attenuation correction. The attenuation compensation value output by this module can precisely characterize the differentiated attenuation features of different frequency components under wind and wave conditions.

[0127] The third module 203 is used to determine the index dimension of the parameter library, and combine the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor to quickly predict and match the interference structure feature parameter library to obtain the interference structure feature parameters.

[0128] In this embodiment, the core function of the fast prediction and matching module based on the interferometric structure feature parameter library is to establish a fast prediction and matching method based on the interferometric structure feature parameter library. It consists of a feature parameter library construction submodule, a fast retrieval submodule, and a matching correction submodule. The feature parameter library construction submodule uses sound source depth, distance, frequency, receiver array depth, and wind speed level as index dimensions to pre-calculate and store interferometric structure feature parameters under various typical environments, such as interferometric fringe period, fringe slope, null frequency, and fringe spacing in the angle of arrival domain. The fast retrieval submodule performs rapid retrieval and interpolation in the library based on the current environmental parameters during real-time prediction to obtain the initial interferometric structure prediction result. The matching correction submodule uses a small amount of measured sound field data or the calculation results of a high-precision numerical model to locally correct the feature parameters retrieved from the library, thereby eliminating errors caused by the discretization of the parameter library.

[0129] The fourth module 204 is used to obtain the acoustic environment parameters at each forecast time, combine them with the interferometric structure characteristic parameters to make advance forecasts, and generate the interferometric structure forecast results at each time.

[0130] In this embodiment, the core function of the advanced forecasting module is to establish an advanced direct-access interferometric structure forecasting method that integrates sea surface state numerical forecasting products. It comprises a sea surface state numerical forecasting product access and analysis submodule, an environmental parameter time-series extrapolation submodule, and an interferometric structure time-series forecasting submodule. The sea surface state numerical forecasting product access and analysis submodule acquires high spatiotemporal resolution numerical forecasting products such as wind field, wave field, and pressure field from the meteorological and oceanographic forecasting center, and performs data analysis and resampling. The environmental parameter time-series extrapolation submodule extrapolates the acoustic environmental parameters for future times based on the future sea surface state evolution trend provided by the numerical forecasting products, combined with the response time scale of bubble layer formation and dissipation. The interferometric structure time-series forecasting submodule inputs the extrapolated environmental parameters into the aforementioned dynamic sea surface reflection coefficient prediction module and broadband sound propagation attenuation compensation module to generate an interferometric structure forecasting sequence for future time periods (e.g., 6 to 24 hours), thus achieving advanced forecasting.

[0131] The fifth module 205 is used to construct a hybrid prediction model by combining the physical model and the high-precision reference model. Based on the prediction results of the interferometric structure at each time, a weighted summation fusion strategy is used to predict the interferometric structure in the direct sound region.

[0132] In this embodiment, the core function of the hybrid forecasting module is to establish a hybrid forecasting method for direct-access acoustic interferometric structures by fusing a physical model with a deep neural network. It comprises a physical model calculation submodule, a deep neural network residual learning submodule, and an adaptive fusion output submodule. The physical model calculation submodule uses the aforementioned dynamic reflection coefficient and attenuation-compensated ray model or virtual source method to quickly generate baseline forecast results for the interferometric structure. The deep neural network residual learning submodule uses wind speed, wave height, bubble layer thickness, and sound wave frequency as input features, and the residual between the physical model forecast results and the calculation results of a high-precision reference model (such as a parabolic equation model) as the learning objective to train a lightweight residual compensation network. During actual forecasting, the adaptive fusion output submodule dynamically adjusts the fusion weights of the physical model and neural network outputs based on current environmental parameters and forecast lead times, outputting the final interferometric structure forecast result. This module fully considers both the prior knowledge of the physical model and the data-driven nonlinear fitting capability.

[0133] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0134] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for predicting interferometric structures in the direct acoustic region considering wind and wave conditions, characterized in that, Includes the following steps: The dynamic sea surface reflection coefficient is obtained by acquiring the spatiotemporal sequence of wind and wave spectra with directional spectra and predicting the dynamic sea surface reflection coefficient. A three-dimensional bubble particle size distribution function is constructed and broadband sound propagation attenuation compensation is performed to obtain the broadband sound propagation attenuation compensation factor. The index dimension of the parameter library is determined, and the dynamic sea surface reflection coefficient and broadband acoustic propagation attenuation compensation factor are combined to quickly predict and match the interferometric structure feature parameter library to obtain the interferometric structure feature parameters. Acoustic environment parameters at each forecast time are obtained, and advanced forecasts are made by combining them with interferometric structure characteristic parameters to generate interferometric structure forecast results at each time. A hybrid prediction model is constructed by combining a physical model and a high-precision reference model. Based on the prediction results of the interferometric structure at each time point, a weighted summation fusion strategy is used to predict the interferometric structure in the direct sound region.

2. The method for predicting the interferometric structure of a direct acoustic region considering wind and wave conditions according to claim 1, characterized in that, The step of obtaining the dynamic sea surface reflection coefficient by acquiring the spatiotemporal sequence of wind and wave spectra with directional spectra and predicting the dynamic sea surface reflection coefficient specifically includes: Obtain sea surface wind field data; Considering wind energy input, wave breaking dissipation, and nonlinear wave-wave interaction terms, sea surface wind field data is input into the wave action balance equation of the wave spectrum model to obtain a wind wave spectrum spatiotemporal sequence with directional spectrum. The spatiotemporal sequence of wind and wave spectrum is divided into several frequency bands. Considering the energy weight and directional distribution of each frequency component, random initial phases are assigned to the wave number components corresponding to several frequency bands. The spectral domain is converted to the spatial domain by fast Fourier transform to obtain the geometrically reconstructed sea surface geometry. Based on the geometrically reconstructed sea surface geometry, the curvature information and normal direction of the local sea surface at that point are extracted from the sea surface elevation field according to the location and time of the intersection of the sound ray and the sea surface. The curvature information and normal direction of the local sea surface at this point are input into a rough surface scattering model based on the Huygens-Fresnel principle, and the dynamic sea surface reflection coefficient is obtained by combining the incident angle and frequency of the sound rays.

3. The method for predicting the direct acoustic region interference structure considering wind and wave conditions according to claim 1, characterized in that, The step of constructing a three-dimensional bubble particle size distribution function and performing broadband sound propagation attenuation compensation to obtain a broadband sound propagation attenuation compensation factor specifically includes: A mathematical model of bubble particle size distribution as a function of depth was established based on wind speed and wave breaking intensity parameters. Based on the fundamental distribution characteristics of nearshore surface bubble size distribution, differentiated attenuation coefficients are matched for bubbles of different sizes. Combined with a mathematical model of bubble size distribution with depth, a three-dimensional bubble size distribution function is constructed. Based on the three-dimensional bubble size distribution function, the bubble volume fraction of the corresponding seawater layer is obtained by integrating the bubble number concentration with the average bubble volume. The equivalent sound velocity of the seawater depth layer is obtained by correcting the bubble volume fraction using the Wood formula. Considering the resonant scattering loss and thermoviscous dissipation loss generated by bubbles, the acoustic attenuation contribution of bubbles of all different sizes is superimposed to obtain the broadband attenuation coefficient of this seawater depth layer. The propagation path of the sound ray is determined based on the equivalent sound velocity. The path of the sound ray through the bubble mixing layer is discretized into several small segments. The attenuation of the sound pressure amplitude is calculated for each small segment based on the corresponding broadband attenuation coefficient. The attenuation of the sound pressure amplitude is discretized, and the attenuation is integrated along the sound propagation path to obtain the total attenuation of the frequency component along the entire path, which is used as the broadband sound propagation attenuation compensation factor.

4. The method for predicting the direct acoustic region interference structure considering wind and wave conditions according to claim 1, characterized in that, The step of determining the index dimension of the parameter library, combining the dynamic sea surface reflection coefficient and the broadband acoustic propagation attenuation compensation factor, and performing rapid prediction and matching of the interferometric structure feature parameter library to obtain the interferometric structure feature parameters specifically includes: The index dimensions of the parameter library are determined, and key feature parameters are extracted by combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor to construct a feature parameter library. Based on the current environmental parameters, multidimensional interpolation retrieval is performed in the feature parameter library to obtain preliminary interference structure feature parameters; The measured interference structure characteristic parameters are obtained by measuring actual data or by determining the characteristic parameters of parameter points, and then compared with the preliminary interference structure characteristic parameters to construct correction coefficients. The preliminary interference structure characteristic parameters are locally calibrated based on the correction coefficients to obtain the interference structure characteristic parameters.

5. The method for predicting the direct acoustic region interference structure considering wind and wave conditions according to claim 4, characterized in that, The step of determining the index dimension of the parameter library, extracting key feature parameters by combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor, and constructing the feature parameter library specifically includes: The index dimensions of the parameter library are determined, including sound source depth, horizontal distance, sound wave frequency, receiver array depth, and wind speed level. Based on the index dimensions of the parameter library, combined with the dynamic sea surface reflection coefficient and the broadband acoustic propagation attenuation compensation factor, frequency-distance interferograms and angle-of-arrival interferograms are generated. Key feature parameters were extracted from frequency-range interferograms and angle-of-arrival interferograms to construct a feature parameter library.

6. The method for predicting the interferometric structure of a direct acoustic region considering wind and wave conditions according to claim 1, characterized in that, The step of obtaining acoustic environment parameters at each forecast time, combining them with interferometric structure characteristic parameters for advance forecasting, and generating interferometric structure forecast results at each time specifically includes: Establish a data interface with the meteorological and oceanographic forecasting center to obtain numerical forecast products; Based on numerical weather prediction products, the data on the evolution of sea surface conditions in the future are determined. The acoustic environmental parameters for each forecast time are derived by combining the response time scale of bubble layer formation and dissipation. Based on the acoustic environment parameters at each forecast time, combined with the dynamic sea surface reflection coefficient, broadband sound propagation attenuation compensation factor, and interference structure characteristic parameters, the sound field is driven by a preset accelerated calculation strategy to generate the interference structure forecast results at each time.

7. The method for predicting the direct acoustic region interference structure considering wind and wave conditions according to claim 6, characterized in that, The preset accelerated computing strategy specifically includes: The process of rapid prediction and matching of the interference structure feature parameter library is time-series reused, that is, the parameter library retrieval and matching are only performed again when the environmental parameters change beyond the preset threshold. The physical model calculations are coarse-grained in time, and linear interpolation is performed between adjacent time points to reduce the number of actual sound field calculations.

8. The method for predicting the direct acoustic region interference structure considering wind and wave conditions according to claim 1, characterized in that, The step of constructing a hybrid prediction model by combining a physical model and a high-precision reference model, and using a weighted summation fusion strategy based on the interferometric structure prediction results at each time point, to predict the interferometric structure in the direct acoustic region, specifically includes: By combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor, the sea surface reflection is equivalent to virtual source radiation, and coherent superposition compensation is performed to generate frequency-distance or frequency-elevation angle interferograms. Based on the frequency-distance or frequency-elevation angle interferometry pattern, determine the physical model and high-precision reference model, and determine the forecast lead time requirement based on the interferometric structure prediction results at each time point; Training data is generated by sampling within a typical environmental parameter space. The interference structure is calculated using both a physical model and a high-precision reference model to obtain the environmental parameter residual compensation values. The neural network is trained based on the residual compensation values ​​of environmental parameters, and the fusion weights of the physical model output and the neural network residual output are dynamically determined according to the current environmental parameters and forecast timeliness requirements. If the environmental parameters are within the coverage of the training data and the forecast lead time requirement is higher than the mean, the weight of the residual output of the neural network is greater than the weight of the physical model output. If the environmental parameters are in the sparse region of the training data or long-term series advance forecasting is required, the weight of the residual output of the neural network is less than the weight of the physical model output. By employing a weighted summation fusion strategy, the superposition result of the physical model calculation and the residual compensation value is output, thereby enabling the prediction of interference structures in the direct acoustic region.

9. The method for predicting the direct acoustic region interference structure considering wind and wave conditions according to claim 8, characterized in that, The step of combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor to treat sea surface reflection as equivalent to virtual source radiation, performing coherent superposition compensation, and generating frequency-distance or frequency-elevation angle interferograms specifically includes: Given the location of the sound source, the location of the receiving array, the frequency range, and the wind speed level, the sea surface reflection is equivalent to virtual source radiation by combining the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor. Based on virtual source radiation, considering the modulation of amplitude and phase by the dynamic reflection coefficient and the compensation of amplitude by path attenuation, the sound pressure of the direct wave and the reflected wave are coherently superimposed and compensated to obtain the sound pressure spectrum at the receiving point. Based on the sound pressure spectrum at the receiving point, frequency-distance or frequency-elevation angle interference patterns are generated.

10. A direct-access acoustic region interferometric structure prediction system considering wind and wave conditions, characterized in that, Includes the following modules: The first module is used to obtain the spatiotemporal sequence of wind and wave spectra with directional spectra to predict the dynamic sea surface reflection coefficient, thereby obtaining the dynamic sea surface reflection coefficient. The second module is used to construct a three-dimensional bubble particle size distribution function and perform broadband sound propagation attenuation compensation to obtain a broadband sound propagation attenuation compensation factor. The third module is used to determine the index dimension of the parameter library, and combine the dynamic sea surface reflection coefficient and the broadband sound propagation attenuation compensation factor to quickly predict and match the interference structure feature parameter library to obtain the interference structure feature parameters. The fourth module is used to obtain acoustic environment parameters at each forecast time, combine them with interferometric structure characteristic parameters to make advance forecasts, and generate interferometric structure forecast results at each time. The fifth module is used to construct a hybrid prediction model by combining the physical model and the high-precision reference model. Based on the prediction results of the interferometric structure at each time, a weighted summation fusion strategy is used to predict the interferometric structure in the direct sound region.

Images