Method and system for accurate measurement of underflow based on acoustic doppler current profiler

Abstract

This invention belongs to the field of hydrodynamics technology and relates to a method and system for accurate measurement of subsurface currents based on an acoustic Doppler current profiler. The method includes: measuring a target water area using an acoustic Doppler current profiler; extracting at least one boundary feature parameter at the measurement point; constructing and pre-training a physics-data fusion model; inputting the effective velocity data above the measurement blind zone from the original velocity profile data and the boundary feature parameter into the pre-trained physics-data fusion model; and outputting the subsurface current velocity distribution within the measurement blind zone, as well as the key statistics and confidence intervals of this velocity distribution. This invention overcomes the shortcomings of existing physical models and data-driven models, providing a precise subsurface current measurement method that combines physical interpretability and environmental adaptability. It also provides a traceable and verifiable benchmark for model training and evaluation, significantly improving the accuracy of subsurface current measurement in complex environments.

Description

Technical Field

[0001] This invention belongs to the field of hydrodynamics technology, and more specifically, relates to a method and system for accurate measurement of undercurrent based on an acoustic Doppler current profiler. Background Technology

[0002] Subsurface currents refer to water flows close to the bottom of a water body. The velocity distribution and energy transfer characteristics of subsurface currents directly determine the erosion and deposition of riverbed materials, the transport flux of suspended sediment, and the survival environment of benthic organisms. Acoustic Doppler Current Profiler (ADCP), as a highly efficient velocity measurement device, can quickly obtain vertical velocity profiles of water bodies. However, the inherent limitation of ADCP is its bottom measurement blind zone. Due to changes in the scattering characteristics of acoustic signals near the bed surface and equipment installation limitations, ADCP cannot directly measure the velocity within a certain range above the bed surface, which is precisely the core distribution area of ​​subsurface currents, resulting in the loss of crucial information in direct measurement data.

[0003] To address this issue, existing technologies are mainly divided into two categories: pure physical model extrapolation methods and pure data-driven model methods. Pure physical model extrapolation methods assume that the velocity distribution above the blind zone follows a specific theoretical formula. The formula parameters are determined by fitting measured velocity data above the blind zone, and then this formula is extrapolated to the bed surface to estimate the subsurface velocity. However, theoretical assumptions are disconnected from the actual environment. Water flow in natural water bodies is often non-uniform, and bed roughness is spatially heterogeneous due to various factors, leading to a mismatch between the model and the actual flow field. The extrapolation process further amplifies the fitting error above the blind zone, and the magnitude of the error cannot be quantified. For complex boundary conditions, the extrapolation error severely affects the reliability of the measurement results. Furthermore, the same physical model is difficult to adapt to flow field changes in different water bodies and seasons, resulting in significant fluctuations in measurement accuracy. Pure data-driven modeling utilizes machine learning algorithms to learn the mapping relationship between the overall shape of the velocity profile and boundary conditions through a large amount of field-measured data or numerical simulation data. It directly predicts the undercurrent distribution in the blind zone from the velocity data above the blind zone. However, it ignores the basic physical conservation laws of fluid motion, and the prediction results may violate common sense physics. The model performance is highly dependent on the coverage and representativeness of the training data, and the prediction results in new environments will be completely distorted. There is a lack of high-quality field data to directly verify the model, and the prediction accuracy of the model cannot be accurately evaluated. The data collection cycle is long and costly, and the data quality is uneven, which affects the model training effect.

[0004] Therefore, developing a precise measurement method for subsurface currents that combines physical interpretability and environmental adaptability with a traceable verification system has become an urgent technical problem to be solved. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method and system for accurate measurement of undercurrents based on an acoustic Doppler velocity profiler.

[0006] In a first aspect, the present invention provides a method for accurate measurement of undercurrent based on an acoustic Doppler current profiler, comprising:

[0007] The target water area was measured using an acoustic Doppler current profiler, and raw current profile data and acoustic echo intensity data were acquired simultaneously.

[0008] Based on acoustic echo intensity data, at least one boundary feature parameter is extracted at the measurement point.

[0009] The physical-data fusion model is constructed and pre-trained, including: establishing a fluid dynamics governing equation solver with parameterized boundary conditions as input; designing standardized test bodies with different known roughness and shapes, measuring the complete flow field including the boundary layer and near the substrate under set environmental conditions, and simultaneously recording the corresponding boundary parameters to form a training dataset; based on the training dataset, using the boundary parameters as input, calculating the initial flow field using the fluid dynamics governing equation solver; using machine learning components to learn the residual between the initial flow field and the actual measured complete flow field, and simultaneously optimizing the parameters of the machine learning components and the key turbulence parameters in the fluid dynamics governing equation solver through backpropagation algorithm, so that the sum of the initial flow field and the residual output by the fusion model approximates the complete flow field, thus obtaining the pre-trained physical-data fusion model;

[0010] The effective velocity data located above the measurement blind zone in the original velocity profile data, along with the boundary feature parameters, are input into a pre-trained physical-data fusion model.

[0011] The physical-data fusion model is used to output the subsurface velocity distribution within the measurement blind zone, as well as the key statistics and confidence intervals of this subsurface velocity distribution.

[0012] Secondly, the present invention provides a precise measurement system for undercurrent based on an acoustic Doppler current profiler, including an acquisition unit, an extraction unit, a model building and training unit, an input unit, and an output unit;

[0013] The acquisition unit is used to measure the target water area using an acoustic Doppler current profiler, and simultaneously acquire raw current profile data and acoustic echo intensity data.

[0014] The extraction unit is used to extract at least one boundary feature parameter at the measurement point based on acoustic echo intensity data.

[0015] The model building and training unit is used to construct and pre-train a physics-data fusion model, including: establishing a fluid dynamics governing equation solver with parameterized boundary conditions as input; designing standardized test bodies with different known roughness and shapes, measuring the complete flow field including the boundary layer and near the substrate under set environmental conditions, and simultaneously recording the corresponding boundary parameters to form a training dataset; based on the training dataset, using the boundary parameters as input, calculating the initial flow field using the fluid dynamics governing equation solver; using machine learning components to learn the residual between the initial flow field and the actual measured complete flow field, and simultaneously optimizing the parameters of the machine learning components and the key turbulence parameters in the fluid dynamics governing equation solver through backpropagation algorithm, so that the sum of the initial flow field and the residual output by the fusion model approximates the complete flow field, thus obtaining the pre-trained physics-data fusion model;

[0016] The input unit is used to input the effective velocity data and boundary feature parameters located above the measurement blind zone in the original velocity profile data into the pre-trained physical-data fusion model;

[0017] The output unit is used to output the subsurface velocity distribution within the measurement blind zone, as well as the key statistics and confidence intervals of the subsurface velocity distribution, using a physical-data fusion model.

[0018] Based on the above technical solution, the present invention can be further improved as follows.

[0019] Furthermore, after acquiring the acoustic echo intensity data, the acoustic echo intensity data is denoised using wavelet transform. The denoised echo intensity profile is then smoothed using the moving average method.

[0020] Furthermore, the boundary characteristic parameters include at least one of the following: equivalent sand roughness height, bottom shape index, and riverbed dip angle, which are obtained by inversion from acoustic echo intensity data.

[0021] Furthermore, at least one boundary feature parameter is extracted at the measurement point, including:

[0022] A correlation model between acoustic echo intensity data and sand roughness height based on standardized test body calibration, or a multi-parameter inversion model established using machine learning algorithms, can be used as input for acoustic echo intensity profile, water depth and water temperature, and output for equivalent sand roughness height.

[0023] The wave height and wavelength of the base shape are calculated by the lateral variation of the acoustic echo intensity profile, and the base shape index is obtained.

[0024] By combining attitude sensor data from an acoustic Doppler current profiler with GPS positioning information, a three-dimensional terrain model is constructed, and the riverbed inclination angle is calculated.

[0025] Furthermore, based on the Reynolds-averaged Navier-Stokes equations and using a parameterized turbulent closed model, a solver for the fluid dynamics governing equations is obtained.

[0026] Furthermore, standardized test bodies with different known roughnesses and shapes were designed. Under set environmental conditions, the complete flow field, including the boundary layer and the vicinity of the bed, was measured, and the corresponding boundary parameters were recorded simultaneously to form a training dataset. This included: installing standardized test bodies in a circulating water tank, and for each standardized test body, measuring the two-dimensional instantaneous velocity field in the central region of the cross-section using a high frame rate PIV system under several different water flow velocities. The time-averaged flow field was obtained after time averaging. The time-averaged flow field contains complete flow field information from the water surface to the bed surface, and the corresponding boundary parameters and water flow conditions were recorded.

[0027] Furthermore, the machine learning components are fully connected neural networks, convolutional neural networks, or attention mechanism networks.

[0028] Furthermore, before the effective flow velocity data is input into the pre-trained physical-data fusion model, the effective flow velocity data is normalized. The normalization parameters are the vertical average flow velocity and water depth measured using an acoustic Doppler current profiler.

[0029] Furthermore, the acoustic Doppler current profiler is mounted on a mobile measurement platform; the mobile measurement platform is an unmanned vessel or an underwater autonomous vehicle.

[0030] The beneficial effects of this invention are as follows: This invention overcomes the shortcomings of existing physical models and data-driven models, providing a precise method for measuring subsurface currents that combines physical interpretability and environmental adaptability. It also provides a traceable and verifiable benchmark for model training and evaluation, fundamentally improving the reliability of subsurface current measurement results. This invention employs a physical-data fusion model for precise subsurface current measurement, significantly improving measurement accuracy in complex environments compared to a single model. The physical-data fusion model output includes confidence intervals based on training residual statistics, quantifying the uncertainty of predictions. The standardized test case design greatly reduces the difficulty of obtaining high-quality training data, making this solution feasible for engineering application. Attached Figure Description

[0031] Figure 1 A schematic diagram of the principle of the precise measurement method of undercurrent based on an acoustic Doppler current profiler provided in Embodiment 1 of the present invention;

[0032] Figure 2 This is a schematic diagram of the undercurrent precision measurement system based on an acoustic Doppler velocity profiler provided in Embodiment 1 of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0034] Example 1

[0035] As an example, to solve the above-mentioned technical problems, as shown in the appendix Figure 1 As shown, this embodiment provides a method for accurate measurement of undercurrent based on an acoustic Doppler current profiler, including the following steps:

[0036] Step 110: Measure the target water area using an acoustic Doppler current profiler, and simultaneously acquire raw current profile data and acoustic echo intensity data;

[0037] In some optional embodiments, after acquiring the acoustic echo intensity data, the data is denoised using wavelet transform. The denoised echo intensity profile is then smoothed using a moving average method. An unmanned surface vessel (USV) equipped with a 1200kHz acoustic Doppler current profiler is used for onboard measurements across the target river cross-section. The vertical cell size of the acoustic Doppler current profiler is set to 0.1m, with a blank area of ​​0.2m, simultaneously recording flow velocity, echo intensity, and bottom tracking data.

[0038] Step 120: Based on the acoustic echo intensity data, extract at least one boundary feature parameter at the measurement point.

[0039] In practical applications, the boundary characteristic parameters include at least one of the following: equivalent sand roughness height obtained by inversion from acoustic echo intensity data, bottom shape index, and riverbed dip angle.

[0040] In some alternative embodiments, a correlation model between acoustic echo intensity data calibrated based on a standardized test body and sand roughness height is used, or a multi-parameter inversion model is established using machine learning algorithms, with input acoustic echo intensity profile, water depth and water temperature, and output equivalent sand roughness height.

[0041] The wave height and wavelength of the base shape are calculated by the lateral variation of the acoustic echo intensity profile, and the base shape index is obtained.

[0042] By combining attitude sensor data from an acoustic Doppler current profiler with GPS positioning information, a three-dimensional terrain model is constructed, and the riverbed inclination angle is calculated.

[0043] Step 130: Construct and pre-train a physics-data fusion model, including: establishing a fluid dynamics governing equation solver with parameterized boundary conditions as input; designing standardized test bodies with different known roughness and shapes, measuring the complete flow field including the boundary layer and near the substrate under set environmental conditions, and simultaneously recording the corresponding boundary parameters to form a training dataset; based on the training dataset, using the boundary parameters as input, calculating the initial flow field using the fluid dynamics governing equation solver; using machine learning components to learn the residual between the initial flow field and the actual measured complete flow field, and simultaneously optimizing the parameters of the machine learning components and the key turbulence parameters in the fluid dynamics governing equation solver through backpropagation algorithm, so that the sum of the initial flow field and the residual output by the fusion model approximates the complete flow field, thus obtaining the pre-trained physics-data fusion model.

[0044] When designing standardized test specimens, for example, three sets of test specimens are prepared, including: a smooth plexiglass plate, closely packed sandpaper with an average particle size of 5 mm, and a regularly arranged array of cylinders with a height of 10 mm. The hydraulic roughness height of each set of test specimens is measured.

[0045] In the circulating water tank, standardized test bodies are installed. For each standardized test body, the two-dimensional instantaneous velocity field of the central region of the cross section is measured using a high frame rate PIV system under several different water flow velocities. The time-averaged flow field is obtained after time averaging. The time-averaged flow field contains complete flow field information from the water surface to the bed surface. The corresponding boundary parameters and water flow conditions are recorded.

[0046] In some optional embodiments, for example, each test body is installed in a circulating water tank with a length of 30m, a width of 0.5m, and a height of 0.8m. For each test body, a two-dimensional instantaneous velocity field in the central region of the cross-section is measured using a high frame rate particle image velocimetry system at five different water flow velocities. After time averaging, a high-resolution time-averaged flow field is obtained. The time-averaged flow field contains complete information from the water surface to the bed surface, especially the accurate flow velocity near the bottom layer. The corresponding boundary parameters (mainly based on the equivalent sand roughness height) and water flow conditions (average flow velocity and water depth) are recorded, and a total of 15 sets of standard data pairs are obtained.

[0047] Based on the Reynolds-averaged Navier-Stokes equations and using a parameterized turbulent closed model, a solver for the fluid dynamics governing equations is obtained.

[0048] Combine the mixing length turbulence model and the mixing length parameters therein Let it be a learnable parameter, denoted as , This represents the set of physical parameters to be optimized. Learnable parameters. The initial values ​​are taken from the classic empirical values ​​of the turbulent closed model, and the parameters of the machine learning component are... Random initialization (following a normal distribution) is used.

[0049] The high-resolution time-averaged flow field obtained after time averaging includes longitudinal, lateral, and vertical velocities at different vertical heights, achieving blind-spot-free coverage from the water surface to the bed surface.

[0050] The corresponding boundary parameters are recorded synchronously, including equivalent sand roughness height, bottom shape index (the ratio of wave height to wavelength), riverbed dip angle, average flow velocity, and water depth.

[0051] Let the time-averaged flow field be , The height is [height], and the longitudinal flow velocity is [velocity]. The lateral flow velocity is The vertical flow velocity is Boundary parameters are The average flow velocity of the water is The water depth is The equivalent sand roughness height is ,high The flow velocity correction at that point is .

[0052] In some alternative embodiments, the machine learning component is a fully connected neural network, a convolutional neural network, or an attention mechanism network.

[0053] Suppose the machine learning component uses a 3-layer fully connected neural network, with 64, 32, and 16 neurons in each layer, and ReLU activation function to avoid the vanishing gradient problem. The input is normalized […]. , , The output is the corresponding height. Flow rate correction at the location .

[0054] Step 140: Input the effective velocity data and boundary feature parameters located above the measurement blind zone in the original velocity profile data into the pre-trained physical-data fusion model.

[0055] In some alternative embodiments, the effective velocity data is normalized before being input into the pre-trained physical-data fusion model, with the normalization parameters being the vertical average velocity and water depth measured using an acoustic Doppler current profiler.

[0056] For each measuring point, the equivalent sand roughness height is estimated from the echo intensity profile using an inversion algorithm based on signal attenuation and scattering; the effective acoustic Doppler velocity profiler velocity data above the blind zone (e.g., above 0.3m from the bed surface) is extracted, and the average flow velocity along that vertical line is calculated. and water depth ;Will The normalized flow velocity data is then input into the pre-trained physical-data fusion model.

[0057] Step 150: Use the physical-data fusion model to output the subsurface velocity distribution within the measurement blind zone, as well as the key statistics and confidence intervals of the subsurface velocity distribution.

[0058] The physics-data fusion model outputs a complete velocity profile including blind zones, and simultaneously outputs a 95% confidence interval (e.g., 0.12 ± 0.03 m / s) for the subsurface velocity at a height of 0.5 m above the bed surface, calculated based on statistical error propagation results validated on standardized test bodies during the training phase. By integrating data from all measurement points, a spatial distribution map of the subsurface velocity across the river cross-section is generated, and uncertainties in key areas are marked. Key statistics are statistical parameters that quantify the core characteristics of the subsurface velocity distribution, such as: average velocity, representing the average velocity magnitude of the subsurface velocity within the measurement blind zone, reflecting the overall flow intensity of the subsurface velocity in the region; median velocity, representing the median value of the velocity distribution, reflecting the level of the subsurface velocity; maximum and minimum velocities; and kurtosis, representing the morphological characteristics of the velocity distribution.

[0059] In some alternative embodiments, the acoustic Doppler current profiler is mounted on a mobile measurement platform; the mobile measurement platform is an unmanned surface vessel or an autonomous underwater vehicle.

[0060] This invention overcomes the shortcomings of existing physical models and data-driven models, providing a precise method for measuring subsurface current that combines physical interpretability and environmental adaptability. It also provides a traceable and verifiable benchmark for model training and evaluation, fundamentally improving the reliability of subsurface current measurement results.

[0061] The physics-data fusion model inherits the universality of physical laws, while correcting local deviations under specific boundaries through data learning. Compared with a single model, it significantly improves the accuracy of undercurrent measurement in complex environments.

[0062] The physical-data fusion model output includes confidence intervals based on training residual statistics, which quantifies the uncertainty of prediction. The design of standardized test cases greatly reduces the difficulty of obtaining high-quality training data, making this solution feasible for engineering promotion.

[0063] Example 2

[0064] Based on the same principle as the method shown in Embodiment 1 of the present invention, as illustrated in the appendix. Figure 2 As shown, an embodiment of the present invention also provides a precise measurement system for undercurrent based on an acoustic Doppler current profiler, including an acquisition unit, an extraction unit, a model building and training unit, an input unit, and an output unit;

[0065] The acquisition unit is used to measure the target water area using an acoustic Doppler current profiler, and simultaneously acquire raw current profile data and acoustic echo intensity data.

[0066] The extraction unit is used to extract at least one boundary feature parameter at the measurement point based on acoustic echo intensity data.

[0067] The model building and training unit is used to construct and pre-train a physics-data fusion model, including: establishing a fluid dynamics governing equation solver with parameterized boundary conditions as input; designing standardized test bodies with different known roughness and shapes, measuring the complete flow field including the boundary layer and near the substrate under set environmental conditions, and simultaneously recording the corresponding boundary parameters to form a training dataset; based on the training dataset, using the boundary parameters as input, calculating the initial flow field using the fluid dynamics governing equation solver; using machine learning components to learn the residual between the initial flow field and the actual measured complete flow field, and simultaneously optimizing the parameters of the machine learning components and the key turbulence parameters in the fluid dynamics governing equation solver through backpropagation algorithm, so that the sum of the initial flow field and the residual output by the fusion model approximates the complete flow field, thus obtaining the pre-trained physics-data fusion model;

[0068] The input unit is used to input the effective velocity data and boundary feature parameters located above the measurement blind zone in the original velocity profile data into the pre-trained physical-data fusion model;

[0069] The output unit is used to output the subsurface velocity distribution within the measurement blind zone, as well as the key statistics and confidence intervals of the subsurface velocity distribution, using a physical-data fusion model.

[0070] Optionally, after acquiring the acoustic echo intensity data, the acoustic echo intensity data is denoised using wavelet transform, and the denoised echo intensity profile is smoothed using the moving average method.

[0071] Optionally, the boundary characteristic parameters include at least one of the following: equivalent sand roughness height, bottom shape index, and riverbed dip angle, which are obtained by inversion from acoustic echo intensity data.

[0072] Optionally, at least one boundary feature parameter is extracted at the measurement point, including:

[0073] A correlation model between acoustic echo intensity data and sand roughness height based on standardized test body calibration, or a multi-parameter inversion model established using machine learning algorithms, can be used as input for acoustic echo intensity profile, water depth and water temperature, and output for equivalent sand roughness height.

[0074] The wave height and wavelength of the base shape are calculated by the lateral variation of the acoustic echo intensity profile, and the base shape index is obtained.

[0075] By combining attitude sensor data from an acoustic Doppler current profiler with GPS positioning information, a three-dimensional terrain model is constructed, and the riverbed inclination angle is calculated.

[0076] Optionally, a solver for the fluid dynamics governing equations can be obtained based on the Reynolds-averaged Navier-Stokes equations and using a parameterized turbulent closed model.

[0077] Optionally, standardized test bodies with different known roughness and shape are designed. Under set environmental conditions, the complete flow field including the boundary layer and the vicinity of the bed is measured, and the corresponding boundary parameters are recorded simultaneously to form a training dataset. This includes: installing standardized test bodies in a circulating water tank, and for each standardized test body, measuring the two-dimensional instantaneous flow velocity field in the central region of the cross-section using a high frame rate PIV system under several different water flow velocities. The time-averaged flow field is obtained after time averaging. The time-averaged flow field contains complete flow field information from the water surface to the bed surface, and the corresponding boundary parameters and water flow conditions are recorded.

[0078] Optionally, the machine learning component can be a fully connected neural network, a convolutional neural network, or an attention mechanism network.

[0079] Optionally, before the effective velocity data is input into the pre-trained physical-data fusion model, the effective velocity data is normalized by using the vertical average velocity and water depth measured by an acoustic Doppler current profiler.

[0080] Optionally, the acoustic Doppler current profiler is mounted on a mobile measurement platform; the mobile measurement platform is an unmanned surface vessel or an autonomous underwater vehicle.

[0081] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for accurate measurement of a submarine flow based on an acoustic Doppler current profiler, characterized in that, include: The target water area was measured using an acoustic Doppler current profiler, and raw current profile data and acoustic echo intensity data were acquired simultaneously. Based on acoustic echo intensity data, at least one boundary feature parameter is extracted at the measurement point. The physical-data fusion model is constructed and pre-trained, including: establishing a fluid dynamics governing equation solver with parameterized boundary conditions as input; designing standardized test bodies with different known roughness and shapes, measuring the complete flow field including the boundary layer and near the substrate under set environmental conditions, and simultaneously recording the corresponding boundary parameters to form a training dataset; based on the training dataset, using the boundary parameters as input, calculating the initial flow field using the fluid dynamics governing equation solver; using machine learning components to learn the residual between the initial flow field and the actual measured complete flow field, and simultaneously optimizing the parameters of the machine learning components and the key turbulence parameters in the fluid dynamics governing equation solver through backpropagation algorithm, so that the sum of the initial flow field and the residual output by the fusion model approximates the complete flow field, thus obtaining the pre-trained physical-data fusion model; The effective velocity data located above the measurement blind zone in the original velocity profile data, along with the boundary feature parameters, are input into a pre-trained physical-data fusion model. The physical-data fusion model is used to output the subsurface velocity distribution within the measurement blind zone, as well as the key statistics and confidence intervals of this subsurface velocity distribution.

2. The method for accurately measuring the underflow according to the acoustic Doppler current profiler of claim 1, wherein, After acquiring the acoustic echo intensity data, the acoustic echo intensity data is denoised using wavelet transform. The denoised echo intensity profile is then smoothed using the moving average method.

3. The method for accurately measuring the underflow according to the acoustic Doppler current profiler of claim 1, wherein, The boundary characteristic parameters include at least one of the following: equivalent sand roughness height, bottom shape index, and riverbed dip angle, which are obtained by inversion from acoustic echo intensity data.

4. The method for accurate measurement of undercurrent based on an acoustic Doppler velocity profiler according to claim 3, characterized in that, Extract at least one boundary feature parameter at the measurement point, including: A correlation model between acoustic echo intensity data and sand roughness height based on standardized test body calibration, or a multi-parameter inversion model established using machine learning algorithms, can be used as input for acoustic echo intensity profile, water depth and water temperature, and output for equivalent sand roughness height. The wave height and wavelength of the base shape are calculated by the lateral variation of the acoustic echo intensity profile, and the base shape index is obtained. By combining attitude sensor data from an acoustic Doppler current profiler with GPS positioning information, a three-dimensional terrain model is constructed, and the riverbed inclination angle is calculated.

5. The method for accurate measurement of undercurrent based on an acoustic Doppler velocity profiler according to claim 1, characterized in that, Based on the Reynolds-averaged Navier-Stokes equations and using a parameterized turbulent closed model, a solver for the fluid dynamics governing equations is obtained.

6. The method for accurate measurement of undercurrent based on an acoustic Doppler velocity profiler according to claim 1, characterized in that, Design standardized test bodies with different known roughnesses and shapes, measure the complete flow field including the boundary layer and the vicinity of the bed under set environmental conditions, and simultaneously record the corresponding boundary parameters to form a training dataset. This includes: installing standardized test bodies in a circulating water tank, and for each standardized test body, measuring the two-dimensional instantaneous flow velocity field in the central region of the cross-section using a high frame rate PIV system under several different water flow velocities, obtaining the time-averaged flow field after time averaging. The time-averaged flow field contains complete flow field information from the water surface to the bed surface, and recording the corresponding boundary parameters and water flow conditions.

7. The method for accurate measurement of undercurrent based on an acoustic Doppler velocity profiler according to claim 1, characterized in that, The machine learning components are fully connected neural networks, convolutional neural networks, or attention mechanism networks.

8. The method for accurate measurement of undercurrent based on an acoustic Doppler velocity profiler according to claim 1, characterized in that, Before the effective flow velocity data is input into the pre-trained physical-data fusion model, the effective flow velocity data is normalized. The normalization parameters are the vertical average flow velocity and water depth measured using an acoustic Doppler current profiler.

9. The method for accurate measurement of undercurrent based on an acoustic Doppler velocity profiler according to claim 1, characterized in that, The acoustic Doppler current profiler is mounted on a mobile measurement platform; the mobile measurement platform is an unmanned surface vessel or an autonomous underwater vehicle.

10. A precise measurement system for undercurrent based on an acoustic Doppler velocity profiler, characterized in that, It includes acquisition units, extraction units, model building and training units, input units, and output units: The acquisition unit is used to measure the target water area using an acoustic Doppler current profiler, and simultaneously acquire raw current profile data and acoustic echo intensity data. The extraction unit is used to extract at least one boundary feature parameter at the measurement point based on acoustic echo intensity data. The model building and training unit is used to construct and pre-train a physics-data fusion model, including: establishing a fluid dynamics governing equation solver with parameterized boundary conditions as input; designing standardized test bodies with different known roughness and shapes, measuring the complete flow field including the boundary layer and near the substrate under set environmental conditions, and simultaneously recording the corresponding boundary parameters to form a training dataset; based on the training dataset, using the boundary parameters as input, calculating the initial flow field using the fluid dynamics governing equation solver; using machine learning components to learn the residual between the initial flow field and the actual measured complete flow field, and simultaneously optimizing the parameters of the machine learning components and the key turbulence parameters in the fluid dynamics governing equation solver through backpropagation algorithm, so that the sum of the initial flow field and the residual output by the fusion model approximates the complete flow field, thus obtaining the pre-trained physics-data fusion model; The input unit is used to input the effective velocity data and boundary feature parameters located above the measurement blind zone in the original velocity profile data into the pre-trained physical-data fusion model; The output unit is used to output the subsurface velocity distribution within the measurement blind zone, as well as the key statistics and confidence intervals of the subsurface velocity distribution, using a physical-data fusion model.

Images