A depth self-adaptive correction method for a cast-off ocean current profiler

Abstract

The present application relates to the technical field of ocean exploration, and discloses a depth self-adaptive correction method for a cast-off type ocean current profiler, to solve the technical problem of vertical deviation of the ocean current profile caused by the traditional fixed resistance coefficient. The present application designs the global working condition through orthogonal test, constructs a multi-dimensional resistance small sample data set in combination with CFD simulation, establishes a resistance coefficient prediction model based on the Gaussian process regression algorithm, and completes precision verification through 5-fold cross-validation. In the actual observation process, the ocean temperature and salinity and the probe motion parameters are collected in real time, the resistance coefficient is dynamically predicted, the speed and depth are iteratively corrected in combination with the kinetic equation, and the profile depth self-adaptive calibration is realized. The present application eliminates the speed and depth calculation errors caused by the traditional fixed resistance coefficient, improves the measurement accuracy and reliability of the vertical profile data of the cast-off type ocean current profiler, and is suitable for high-precision observation operation of the ocean current profile under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of marine exploration technology and discloses a depth adaptive correction method for a drop-out current profiler. Background Technology

[0002] Droppable current profilers are a type of marine observation equipment used to rapidly acquire information about the vertical profile of ocean currents. These instruments are typically deployed into the sea by ships, aircraft, unmanned platforms, or other offshore platforms, relying on their own gravity to fall through the water and continuously collecting current-related measurement data during their descent. Because droppable instruments are not retrieved after deployment or do not rely on complex towing systems, they are characterized by rapid deployment, low platform requirements, and suitability for large-scale, rapid observation.

[0003] For drop-type current profilers, the instrument's descent is itself part of the measurement process. Current signals collected at different depths need to be accurately matched with their corresponding depths to form a valid current profile. If the depth calculation corresponding to a measurement signal at a particular moment is inaccurate, the final current profile will have a vertical offset. Especially in areas with strong vertical shear in the thermocline, halocline, densitoid, or other regions, depth errors on the order of several meters can lead to deviations in the interpretation of the current profile.

[0004] The depth of traditional drop-out oceanographic instruments is usually not obtained through direct, continuous measurement, but rather estimated by the relationship between descent time and descent rate. Common methods include assuming the instrument falls at a certain nominal speed, or using an empirical descent rate equation to convert the time after deployment into depth.

[0005] Existing methods mainly include the following two approaches: The simplest method for depth calculation using a fixed descent speed is to multiply the time since the instrument was deployed by a nominal descent speed to obtain the instrument's depth. This method is expressed as: ; In the formula, For depth, The nominal falling speed, Time of descent; The empirical equation method is used in most existing drop-rate current profilers for depth calculation. This method is expressed as: ; In the formula, For depth, For the time of descent, The nominal fall velocity coefficient is obtained from experimental calibration. This is an empirical correction factor used to compensate for changes in drag, buoyancy, and falling speed. Summary of the Invention

[0006] This invention discloses a depth adaptive correction method for a drop-type ocean current profiler to solve the technical problems of inaccurate probe descent velocity calculation and vertical offset of ocean current profile caused by traditional fixed drag coefficients.

[0007] A depth adaptive correction method for a drop-out current profiler, characterized by comprising the following steps: S1. Obtain the structural and mass parameters of the drop-off current profiler, construct a three-dimensional geometric model of the probe, and determine the basic parameters of the probe's descent dynamics. S2. Define the range of values ​​for seawater environmental parameters and probe descent velocity, and use orthogonal experimental design to simulate multiple working conditions. Solve the seawater density and dynamic viscosity corresponding to each working condition using the seawater temperature, salinity and physical property formulas. S3. Based on the three-dimensional geometric model, a fluid simulation numerical model is built. Flow field simulation is carried out on multiple sets of simulation conditions to obtain the resistance data of the probe under different conditions. The resistance coefficient of each condition is calculated based on the resistance data. A multi-dimensional small sample dataset is constructed by combining the flow characteristic parameters. S4. Based on a small sample dataset, a Gaussian process regression algorithm is used to train a resistance coefficient prediction model, and cross-validation is used to verify the model accuracy. S5. During the descent, the ocean environment and motion parameters are acquired in real time, and the prediction model is called to dynamically output the real-time drag coefficient. The real-time drag coefficient is substituted into the probe descent dynamic equation to iteratively correct the probe's real-time descent speed. The corrected descent depth is obtained by velocity integration, and the ocean current profile depth adaptive calibration is completed.

[0008] Furthermore, the use of orthogonal experimental design for multiple simulation conditions specifically includes: The simulation of the working conditions was carried out by a hybrid horizontal orthogonal experimental design, abandoning the full-element traversal method and selecting multiple representative working conditions with orthogonal characteristics to eliminate multivariate collinearity interference and cover the temperature and salinity environment of the entire sea area and the normal falling velocity range of the probe.

[0009] Furthermore, the formula for calculating the seawater density is as follows: ; In the formula, Density based on pure water This is the first-order coupling coefficient between temperature and salinity. These are higher-order nonlinear correction coefficients. This is a constant salinity correction factor. For seawater temperature, This refers to seawater salinity. The dynamic viscosity of seawater was calculated using a temperature-salinity coupling formula: ; In the formula, The dynamic viscosity of pure water at different temperatures; , This is the salinity coefficient.

[0010] Furthermore, in S3, the fluid simulation numerical model includes mesh generation and boundary condition setting, constructing a cylindrical fluid computational domain, and adaptively refining the probe wall mesh; the fluid domain inlet is set with a velocity inlet boundary, the outlet with a pressure outlet boundary, the probe wall with a no-slip wall boundary, and the outer boundary with a symmetrical boundary, employing... Flow field simulation is carried out using a turbulence model.

[0011] Furthermore, in step S3, calculating the resistance coefficient for each operating condition based on the resistance data specifically includes: Extract the total resistance of the probe along the axial direction The drag coefficient is calculated based on the axial drag output by Fluent: ; In the formula, For axial resistance, The density of seawater under operating conditions. The probe's descent speed, This is the reference area for the incoming flow; Obtain CFD data samples ; The construction of a multi-dimensional small sample dataset by combining flow feature parameters specifically includes: The flow Reynolds number is calculated based on operating parameters, whereby the Reynolds number is: ; In the formula, For the first Reynolds number of the group of samples For the first The density of the group samples, For the first The dynamic viscosity of the group of samples, The characteristic diameter of the probe; The input feature vector is composed of seawater density, dynamic viscosity, falling velocity, and Reynolds number: ; Use the drag coefficient as the output target: ; Forming a training sample set ; Input features for the GPR model. GPR is the target output of the model.

[0012] Furthermore, in step S4, the resistance coefficient prediction model is trained using a Gaussian process regression algorithm based on a small sample dataset, specifically including: A GPR resistance coefficient prediction model was established using the Gaussian process regression model in MATLAB based on a small sample dataset. The GPR drag coefficient prediction model is expressed as follows: ; In the formula, The drag coefficient is predicted by the GPR model. The GPR prediction model is obtained through MATLAB training. The density of seawater, For seawater dynamic viscosity, The probe's descent speed, is the flow Reynolds number.

[0013] Furthermore, in step S4, model accuracy is verified through cross-validation, specifically including: The GPR drag coefficient prediction model was validated using 5-fold cross-validation, and the model accuracy was verified using root mean square error (RMSE) and mean absolute percentage error (MAPE). The root mean square error (RMSE) is: ; In the formula, The drag coefficient is obtained from CFD calculation. The number of samples; The mean absolute percentage error (MAPE) is: ; In the formula, For the first The drag coefficients of the sample groups were predicted by the GPR model in 5-fold cross-validation.

[0014] Furthermore, in S5, during the descent process, the real-time acquisition of ocean environment and motion parameters, and the dynamic output of the real-time drag coefficient by calling the prediction model, specifically includes: Real-time collection of seawater temperature at each sampling time ,salinity and current speed estimate ; according to and Calculate the current seawater density With dynamic viscosity : ; ; Based on the current seawater density Dynamic viscosity Speed ​​estimates and characteristic diameter Calculate the real-time Reynolds number, construct the real-time feature input vector and import it into the trained prediction model, and dynamically output the real-time axial drag coefficient.

[0015] Furthermore, in S5, the dynamic equation is: ; In the formula, For instrument quality, For the falling speed, It is the acceleration due to gravity. The density of seawater, For drainage volume, The drag coefficient, This is the reference area for the incoming flow; The predicted drag coefficient Substituting into the falling dynamics equation, we can solve for the real-time acceleration of the probe. The acceleration calculation formula is as follows: ; In the formula, This refers to the real-time falling acceleration of the probe. For real-time seawater density, This represents the real-time descent speed of the probe.

[0016] Furthermore, in S5, the falling velocity is calculated based on real-time updated seawater physical parameters and dynamically predicted drag coefficients to determine the real-time acceleration, and then iteratively updated using a discrete time step recursive method: ; In the formula, To adjust the falling speed for the next moment, The current falling speed, The preset sampling time interval; The continuous form of the corrected fall depth obtained through velocity integration is as follows: ; In the formula, for The probe's descent depth after adaptive correction at all times. for The probe's descent speed is dynamically corrected in real time; The discrete form is: ; In the formula, For the first The correction depth at each sampling time. For the first The correction depth at each sampling time. The falling speed after iterative correction. The preset sampling time interval; The corrected depth data is matched one-to-one with the ocean current observation data to complete the depth calibration and correction of the ocean current profile data.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention utilizes CFD software to numerically simulate the external flow field of a drop-type current profiler probe under different seawater densities, dynamic viscosities, and relative descent velocities, obtaining probe drag data. Then, the drag coefficient is calculated based on axial drag, using parameters such as seawater density, dynamic viscosity, relative descent velocity, and Reynolds number as inputs and the drag coefficient as output, to establish a continuously callable drag coefficient prediction model in MATLAB. Finally, during the actual descent, the current seawater properties and Reynolds number are calculated based on real-time or preset temperature, salinity, pressure, and velocity conditions to predict the current drag coefficient. This predicted drag coefficient is then substituted into the descent dynamics equation to recursively correct the drop velocity of the drop-type current profiler, and the corrected descent depth is obtained through velocity integration. This complete process constitutes CFD drag sample construction – drag coefficient prediction model training – actual descent velocity recursive correction – depth integration calculation, improving the vertical positioning accuracy of ocean current data from the drop-type current profiler. Attached Figure Description

[0018] Figure 1 This is a flowchart of the technology of the present invention; Figure 2 The simulation results of the axial resistance of the drop-type ocean current profiler probe under the orthogonal test conditions in Group 5; Figure 3 A graph comparing the accuracy of the GPR model's predicted drag coefficient with the CFD simulation calculation; Figure 4 The curves show the changes in the drag coefficient calculated by CFD and predicted by the GPR model under various simulation conditions. Detailed Implementation

[0019] The present invention will be further illustrated below with reference to embodiments. These embodiments are for illustrative purposes only and are not intended to limit the invention in any way. It should be understood that the described embodiments are merely some, not all, of the embodiments described in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0020] This embodiment provides a depth adaptive correction method for a drop-type ocean current profiler, such as... Figure 1 As shown, it includes the following steps: To determine the geometric and mass parameters of the disposable ocean current profiler, the basic structural and mass parameters of the probe are first obtained, including the probe's total length, maximum outer diameter, head shape, cylindrical shell dimensions, tail stabilization structure, and instrument mass. Drainage volume Reference area facing the current and characteristic diameter Among them, the probe's external dimensions, head structure, tail structure, and external protrusion structure are used to establish the three-dimensional geometric model of the discard-type residual profiler; instrument mass Drainage volume Frontal area and characteristic diameter Used for subsequent fall dynamics calculations and drag coefficient calculations. The 3D model of the drop-out current profiler provides the geometric basis for CFD external flow field simulation.

[0021] In this embodiment, it is assumed that the drop-out current profiler maintains an approximately vertical and stable descent under the action of the tail stabilizing structure, and the probe axis is aligned with the direction of the relative incoming current. Attitude angle changes are not considered. Let the probe's descent direction be the -x direction, and the relative incoming current direction be the +x direction.

[0022] The descent dynamics equations are established. When a drop-type ocean current profiler falls vertically in seawater, it is mainly subjected to gravity, buoyancy, and axial hydrodynamic drag. The descent dynamics equations are as follows: ; In the formula, For instrument quality, For the falling speed, It is the acceleration due to gravity. The density of seawater, For drainage volume, The drag coefficient, This is the reference area for the incoming flow; The descent dynamics equation shows that the instrument's descent velocity is related not only to its mass and displacement volume, but also to seawater density, drag coefficient, and the descent velocity itself. Changes in seawater temperature and salinity affect seawater density. Changes or different speed states If the situation changes, the actual falling speed of the instrument will also change.

[0023] Based on temperature and salinity, seawater physical parameters are calculated. In actual marine environments, seawater density and dynamic viscosity vary with temperature and salinity. Based on typical global ocean dynamics and the structural parameters of the drop-alert current profiler, to ensure that any temperature and salinity combination encountered by the drop-alert current profiler during actual operation remains strictly within the calibration space, intervals and step sizes are defined for temperature, salinity, and descent velocity: Temperature (0-30℃): Covers the vast majority of physical profile environments in the global oceans. 0℃ corresponds to the extreme low temperatures of the polar deep sea or high-latitude seas in winter; 30℃ corresponds to the extreme high temperatures of the shallow surface of the equatorial tropics. The step size is 5°C.

[0024] Salinity (20-40 PSU): Covers an extreme range from nearshore / estuarine freshwater (low salinity, approximately 20 PSU) to deep-sea / high-salinity areas (such as the Red Sea and the Mediterranean Sea, approaching 40 PSU). The step size is 5 PSU.

[0025] Current velocity (3.0-5.5 m / s): This range covers the nominal terminal fall velocity of different models of discardable ocean current profilers in seawater. The step size is 0.5 m / s.

[0026] Current variables: 3 factors (temperature, salinity, and flow rate), with 7 levels for temperature, 5 levels for salinity, and 6 levels for flow rate. This is a mixed-level orthogonal experiment. To ensure the accuracy of the prediction model while minimizing the computational burden of CFD simulation, the traditional blind traversal method based on all factors was abandoned, and a method based on... Orthogonal experimental design method using orthogonal arrays. This method selects 25 representative combinations of operating conditions with orthogonal characteristics from a multidimensional parameter space, ensuring that the probability of each level combination between any two factors is completely equal. This design effectively eliminates collinearity interference among multiple variables, ensuring that the extracted drag coefficient can maximize the coverage of the dynamic characteristic space of the XCP probe's falling flow field. In this embodiment, 25 sets of CFD operating condition samples are used to establish an initial drag model, as shown in Table 1; depending on the required correction accuracy, the number of samples can be expanded to more sets, and the model can be updated.

[0027] Table 1. Parameters of 25 Simulation Conditions for Orthogonal Experimental Design .

[0028] Changes in temperature and salinity alter seawater density and dynamic viscosity, thus affecting the probe's descent. Seawater density and salinity are calculated based on temperature and salinity, and used as input for simulation to estimate the drag of the drop-type current meter. The formula for calculating seawater density is: ; In the formula, Density based on pure water This is the first-order coupling coefficient between temperature and salinity. These are higher-order nonlinear correction coefficients. This is a constant salinity correction factor. For seawater temperature, This refers to seawater salinity. ; ; ; ; The dynamic viscosity of seawater was calculated using a temperature-salinity coupling formula: ; In the formula, The dynamic viscosity of pure water at different temperatures; , This is the salinity coefficient.

[0029] ; ; ; The seawater density and dynamic viscosity of 25 working conditions were calculated, as shown in Table 2.

[0030] Table 2. Seawater density and dynamic viscosity parameters corresponding to each orthogonal simulation condition. .

[0031] An external flow field model of the drop-off current profiler was established and CFD conditions were set. A 1:1 3D model of the drop-off current profiler was constructed in SolidWorks 3D modeling software. The constructed 3D geometric model was imported into CFD ICEM mesh generation software. A cylindrical fluid region was established with the drop-off current profiler as the center, and each component was named. Tetrahedral meshes were generated, and the boundaries of the drop-off current profiler were adaptively refined. The mesh quality was checked, and the mesh file was obtained. Import the mesh file into Fluent, fix the probe in the computational domain in Fluent, and make the water flow through the probe along the probe axis. The inlet velocity is equivalent to the falling velocity of the drop current profiler relative to the seawater. In the fluid simulation computational domain, the inlet is set as a velocity outlet, the outlet as a pressure outlet, the probe surface as a no-slip wall, and the outer boundary can be set as a symmetry. The k-wSST model can be used for the turbulence model.

[0032] In this embodiment, 25 sets of CFD operating condition samples were constructed; each set of operating conditions included seawater density. Dynamic viscosity and relative falling speed Three types of input parameters. By changing these parameters, the axial drag response of the probe under different seawater properties and descent velocities can be obtained. Figure 2 The simulation results of the axial drag of the drop-type ocean current profiler probe under the orthogonal experimental conditions of Group 5 are provided by... Figure 2 The simulation results show the characteristics of surface resistance distribution, flow field pressure, and viscous resistance distribution under this typical working condition, clearly reflecting the interaction between seawater flow and the probe structure. This provides an intuitive simulation basis for accurately extracting axial resistance data and calculating the resistance coefficient. The simulation results are shown in Table 3.

[0033] Table 3. Simulation results of axial resistance of probes under various orthogonal simulation conditions. .

[0034] Extracting axial resistance and resistance coefficient: Fluent outputs axial resistance and calculates resistance coefficient. After calculation for each Fluent operating condition, the total resistance of the probe along the axial direction is extracted. The drag coefficient is calculated based on the axial drag output by Fluent: ; In the formula, For axial resistance, The density of seawater under operating conditions. The probe's descent speed, This is the reference area for the incoming flow; This yields a CFD data sample: ; The sample drag coefficient data are shown in Table 4.

[0035] Table 4. Calculation results of drag coefficients for each orthogonal simulation condition. .

[0036] Table 4 shows the variation of axial drag under different seawater properties and different falling speeds for specific probe shapes, providing training data for subsequent MATLAB drag coefficient prediction models.

[0037] The discrete CFD samples obtained from Fluent are transformed into a continuously callable drag coefficient prediction model, enabling the prediction of the current drag coefficient based on the current seawater properties and velocity state during actual descent, rather than relying solely on fixed empirical coefficients.

[0038] A GPR drag coefficient prediction model was established based on MATLAB. MATLAB training samples were constructed, and 25 sets of CFD samples were imported into MATLAB. For each set of samples, its Reynolds number was calculated. ; In the formula, For the first Reynolds number of the group of samples For the first The density of the group samples, For the first The dynamic viscosity of the group of samples, The characteristic diameter of the probe; The input feature vector is composed of seawater density, dynamic viscosity, falling velocity, and Reynolds number: ; Use the drag coefficient as the output target: ; This forms the training sample set: ; In the formula, Input features for the GPR model. Output the target for the model.

[0039] The training sample set is shown in Table 5.

[0040] Table 5. Statistical Table of Parameters for GPR Model Training Sample Set under Various Simulation Conditions .

[0041] A Gaussian process regression model is used to predict the drag coefficient. Since the Fluent sample size in this implementation is 25 groups, it falls under the category of small-sample modeling. Compared to neural networks, the Gaussian process regression model is more suitable for small-sample data, can establish nonlinear mapping relationships under limited sample conditions, and has better predictive stability. Therefore, this implementation preferably uses the Gaussian process regression model in MATLAB to establish the GPR drag coefficient prediction model.

[0042] The GPR drag coefficient prediction model is expressed as follows: ; In the formula, The GPR prediction model is obtained through MATLAB training. The axial drag coefficient is obtained from the model prediction. The density of seawater, For seawater dynamic viscosity, The probe's descent speed, is the flow Reynolds number.

[0043] With axial drag coefficient As output, a continuous mapping relationship is established between seawater properties, flow state, and probe resistance characteristics.

[0044] To verify the accuracy of the GPR model, this embodiment employs a 5-fold cross-validation method to validate the GPR drag coefficient prediction model. Specifically, the 25 samples are divided into 5 subsets. Each time, 4 subsets are selected for model training, and the remaining subset is used for model validation. This process is repeated 5 times, ensuring that each sample serves as a validation sample in one prediction. The cross-validation predicted drag coefficient for each sample is thus obtained.

[0045] To evaluate the prediction performance, the root mean square error (RMSE) and mean absolute percentage error (MAPE) were used as evaluation metrics. RMSE characterizes the average deviation between the drag coefficient predicted by the GPR model and the drag coefficient calculated by the CFD; a smaller RMSE indicates a higher prediction error. In this embodiment, the model's preset accuracy requirements are set as follows: root mean square error of cross-validation RMSE ≤ 0.02, and mean absolute percentage error MAPE < 3%. Figure 3 This is a graph comparing the accuracy of the drag coefficient predictions from the GPR model with those calculated by CFD simulation. Figure 3 It can intuitively present the degree of fit between the model prediction results and the CFD simulation true values, and intuitively verify that the GPR model constructed by this invention has excellent single-point prediction accuracy and meets the accuracy requirements for engineering use.

[0046] When the model satisfies both of the above conditions, it is considered that the established GPR drag coefficient prediction model can meet the drag coefficient prediction accuracy requirements for subsequent recursive correction of the drop velocity of the drop-out current profiler.

[0047] The root mean square error (RMSE) is: ; In the formula, The drag coefficient is obtained from CFD calculation. The number of samples; The mean absolute percentage error (MAPE) is: ; In the formula, For the first The drag coefficients of the sample groups were predicted by the GPR model in 5-fold cross-validation. Figure 4 The curves showing the changes in drag coefficient CFD calculated values ​​and GPR model predicted values ​​under various simulation conditions are provided. Figure 4It can be seen that the fitting effect of the drag coefficient with the changes in seawater properties and flow state across the entire operating range fully verifies the generalization fitting ability of the GPR model for complex nonlinear operating conditions. The calculated RMSE in this embodiment is 0.00245, and the MAPE is 0.2032%, meeting the accuracy requirements.

[0048] The velocity recursive correction and depth integral calculation during the actual descent process are applied to the actual descent data processing of the drop-out current profiler. The descent velocity is dynamically corrected based on real-time or preset marine environmental parameters, and the corrected depth is further calculated.

[0049] During the actual descent of the drop-off current profiler, environmental parameters and initial velocity are acquired. The data processing module also acquires the temperature at each sampling moment during the actual descent of the drop-off current profiler. salinity and current speed estimate ; If the instrument itself has the capability to measure temperature, salinity, and pressure, then real-time acquired data can be used directly; if the instrument is not equipped with a complete temperature, salinity, and pressure sensor, then preset sea area CTD profiles, historical oceanographic data, mother ship synchronous observation data, or ocean reanalysis data can be used as input.

[0050] according to , Calculate the current seawater density and dynamic viscosity: ; ; Predict the drag coefficient at the current moment based on the current seawater density. Dynamic viscosity Speed ​​estimates and characteristic diameter Calculate the current Reynolds number: ; Get the current input: ; Input the established GPR drag coefficient prediction model to obtain the predicted drag coefficient at the current time: ; The drag coefficient at that moment It is dynamically predicted based on the current seawater properties and velocity state, and can reflect the impact of changes in working conditions on the probe's axial resistance during the actual descent.

[0051] The falling velocity is recursively corrected based on the falling dynamics equation, and the predicted drag coefficient is then used. Substituting into the equations of motion for falling, calculate the acceleration at the current moment: ; Based on time step Recursively calculate the corrected falling velocity for the next time step: ; In the formula, To adjust the falling speed for the next moment, The current falling speed, The preset sampling time interval; The recursive process allows the descent speed to be dynamically updated as the seawater density, dynamic viscosity, and drag coefficient change.

[0052] The corrected depth is obtained by integrating the velocity. After obtaining the corrected fall velocity sequence, the velocity is integrated over time to obtain the corrected fall depth. The continuous form is: ; In the formula, for The probe's descent depth after adaptive correction at all times. for The probe's descent speed is dynamically corrected in real time; The discrete form is: ; In the formula, For the first The correction depth at each sampling time. For the first The correction depth at each sampling time. The falling speed after iterative correction. The preset sampling time interval; Finally, the ocean current data collected by the drop-out current profiler and the corrected depth were compared. A one-to-one correspondence was established to complete the depth calibration and correction of the ocean current profile data.

[0053] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A depth adaptive correction method for a drop-type ocean current profiler, characterized in that, Includes the following steps: S1. Obtain the structural and mass parameters of the drop-off current profiler, construct a three-dimensional geometric model of the probe, and determine the basic parameters of the probe's descent dynamics. S2. Define the range of values ​​for seawater environmental parameters and probe descent velocity, and use orthogonal experimental design to simulate multiple working conditions. Solve the seawater density and dynamic viscosity corresponding to each working condition using the seawater temperature, salinity and physical property formulas. S3. Based on the three-dimensional geometric model, a fluid simulation numerical model is built. Flow field simulation is carried out on multiple sets of simulation conditions to obtain the axial resistance data of the probe under different conditions. The resistance coefficient of each condition is calculated based on the axial resistance data. A multi-dimensional small sample dataset is constructed by combining the flow characteristic parameters. S4. Based on a small sample dataset, a Gaussian process regression algorithm is used to train a resistance coefficient prediction model, and cross-validation is used to verify the model accuracy. S5. During the descent, the ocean environment and motion parameters are acquired in real time, and the prediction model is called to dynamically output the real-time drag coefficient. The real-time drag coefficient is substituted into the probe descent dynamic equation to iteratively correct the probe's real-time descent speed. The corrected descent depth is obtained by velocity integration, and the ocean current profile depth adaptive calibration is completed.

2. The depth adaptive correction method for a drop-type ocean current profiler according to claim 1, characterized in that, The specific implementation of multiple sets of simulation conditions using orthogonal experimental design includes: The simulation of the working conditions was carried out by a hybrid horizontal orthogonal experimental design, abandoning the full-element traversal method and selecting multiple representative working conditions with orthogonal characteristics to eliminate multivariate collinearity interference and cover the temperature and salinity environment of the entire sea area and the normal falling velocity range of the probe.

3. The depth adaptive correction method for a drop-type ocean current profiler according to claim 1, characterized in that, The formula for calculating seawater density is: ； In the formula, Density based on pure water This is the first-order coupling coefficient between temperature and salinity. These are higher-order nonlinear correction coefficients. This is a constant salinity correction factor. For seawater temperature, Seawater salinity; The dynamic viscosity of seawater was calculated using a temperature-salinity coupling formula: ； In the formula, The dynamic viscosity of pure water at different temperatures; , This is the salinity coefficient.

4. The depth adaptive correction method for a drop-type ocean current profiler according to claim 1, characterized in that, In S3, the fluid simulation numerical model includes mesh generation and boundary condition setting, constructing a cylindrical fluid computational domain, and adaptively refining the probe wall mesh; a velocity inlet boundary is set at the fluid domain inlet, a pressure outlet boundary is set at the outlet, a no-slip wall boundary is set at the probe wall, and a symmetrical boundary is set at the outer boundary. Flow field simulation is carried out using a turbulence model.

5. The depth adaptive correction method for a drop-type ocean current profiler according to claim 1, characterized in that, In step S3, the calculation of the resistance coefficient for each working condition based on the axial resistance data specifically includes: Extract the total resistance of the probe along the axial direction The drag coefficient is calculated based on the axial drag output by Fluent: ； In the formula, For axial resistance, The density of seawater under operating conditions. The probe's descent speed, This is the reference area for the incoming flow; Obtain CFD data samples ; The construction of a multi-dimensional small sample dataset by combining flow feature parameters specifically includes: The flow Reynolds number is calculated based on operating parameters, whereby the Reynolds number is: ； In the formula, For the first Reynolds number of the group of samples For the first The density of the group samples, For the first The dynamic viscosity of the group of samples, The characteristic diameter of the probe; The input feature vector is composed of seawater density, dynamic viscosity, falling velocity, and Reynolds number: ； Use the drag coefficient as the output target: ； Forming a training sample set ; Input features to the GPR model. GPR is the target output of the model.

6. The depth adaptive correction method for a drop-type ocean current profiler according to claim 1, characterized in that, In step S4, the drag coefficient prediction model is trained using a Gaussian process regression algorithm based on a small sample dataset, specifically including: A GPR resistance coefficient prediction model was established using the Gaussian process regression model in MATLAB based on a small sample dataset. The GPR drag coefficient prediction model is expressed as follows: ； In the formula, The drag coefficient is predicted by the GPR model. The GPR prediction model is obtained through MATLAB training. The density of seawater, For seawater dynamic viscosity, The probe's descent speed, Here is the flow Reynolds number.

7. The depth adaptive correction method for a drop-type ocean current profiler according to claim 1, characterized in that, In step S4, model accuracy is verified through cross-validation, specifically including: The GPR drag coefficient prediction model was validated using 5-fold cross-validation, and the model accuracy was verified using root mean square error (RMSE) and mean absolute percentage error (MAPE). The root mean square error (RMSE) is: ； In the formula, The drag coefficient is obtained from CFD calculation. The number of samples; The mean absolute percentage error (MAPE) is: ； In the formula, For the first The drag coefficients of the sample groups were predicted by the GPR model in 5-fold cross-validation.

8. The depth adaptive correction method for a drop-type ocean current profiler according to claim 1, characterized in that, In step S5, during the descent, real-time acquisition of ocean environment and motion parameters, and dynamic output of the real-time drag coefficient using a prediction model, specifically includes: Real-time collection of seawater temperature at each sampling time ,salinity and current speed estimate ; according to and Calculate the current seawater density With dynamic viscosity : ； ； Based on the current seawater density Dynamic viscosity Speed ​​estimates and characteristic diameter Calculate the real-time Reynolds number, construct the real-time feature input vector and import it into the trained prediction model, and dynamically output the real-time axial drag coefficient.

9. The depth adaptive correction method for a drop-type ocean current profiler according to claim 1, characterized in that, In S5, the dynamic equation is: ； In the formula, For instrument quality, For the falling speed, It is the acceleration due to gravity. The density of seawater, For drainage volume, The drag coefficient, This is the reference area for the incoming flow; The predicted drag coefficient Substituting into the falling dynamics equation, we can solve for the real-time acceleration of the probe. The acceleration calculation formula is as follows: ； In the formula, This refers to the real-time falling acceleration of the probe. For real-time seawater density, This represents the real-time descent speed of the probe.

10. The depth adaptive correction method for a drop-type ocean current profiler according to claim 1, characterized in that, In S5, the falling speed is: The real-time acceleration is calculated based on real-time updated seawater physical parameters and dynamically predicted drag coefficients, and the real-time descent velocity of the probe is iteratively updated using a discrete time step recursive method. ； In the formula, To adjust the falling speed for the next moment, The current falling speed, The preset sampling time interval; The continuous form of the corrected fall depth obtained through velocity integration is as follows: ； In the formula, for The probe's descent depth after adaptive correction at all times. for The probe's descent speed is dynamically corrected in real time; The discrete form is: ； In the formula, For the first The correction depth at each sampling time. For the first The correction depth at each sampling time. The falling speed after iterative correction. The preset sampling time interval; The corrected depth data is matched one-to-one with the ocean current observation data to complete the depth calibration and correction of the ocean current profile data.

Images