A method for correcting ocean salinity data based on spatiotemporal gradient adaptive filtering

Abstract

The present invention relates to the field of ocean observation data processing, and discloses a method for correcting ocean salinity data based on spatiotemporal gradient adaptive filtering, comprising the following steps: calculating motion-compensated salinity gradients and motion-compensated temperature gradients; constructing a dual-gradient driven filtering window, and introducing a gradient-sensitive weight distribution strategy to perform adaptive dynamic filtering on salinity data; generating an initial hyperparameter combination of a GBDT model; inputting the salinity data after adaptive dynamic filtering into the model, optimizing the initial hyperparameter combination of the model using a gray wolf optimization algorithm, obtaining the optimal hyperparameter combination through a dynamic search strategy, and obtaining a final salinity correction value at the same time; performing index verification on the final salinity correction value, and when it does not meet the standard, correcting the model hyperparameters or adaptive dynamic filtering parameters until it meets the standard. The method disclosed by the present invention retains the physical mutation signal of the thermocline zone to the maximum extent while suppressing noise, and significantly improves the accuracy of salinity profile correction.

Description

Technical Field

[0001] The present invention belongs to the field of ocean observation data processing, and in particular relates to an ocean salinity data correction method based on spatiotemporal gradient adaptive filtering. Background Art

[0002] Argo floats, core equipment in the global real-time ocean observation network, are equipped with CTD sensors. They rise and fall vertically in the ocean along a preset trajectory, collecting data at intervals and transmitting it to ground stations via satellite. The CTD sensors measure the conductivity, temperature, and pressure of seawater, and use the conductivity data to invert salinity values ​​through formulas.

[0003] Salinity data, a fundamental physical quantity of the marine environment, possesses irreplaceable scientific value. It is a core parameter for studying ocean dynamics, a key indicator of climate change, and an essential factor influencing ecosystems and fishery resources. However, due to factors such as sensor hardware issues and interference from the seawater environment, errors can occur between sensor-measured salinity series data and the actual seawater salinity. Therefore, salinity data correction is necessary to meet the high-precision data requirements of marine scientific research and engineering applications.

[0004] Traditionally, static filtering methods have been used to correct salinity data. This involves applying a fixed-window median filter, taking the median value of each data point within a fixed range (e.g., 5-15 points) and suppressing noise through statistical filtering. However, fixed windows cannot match variations in stratification intensity, and inappropriate filter window sizes can lead to feature loss. Furthermore, previous correction methods have overlooked the fact that when the sensor is raised or lowered, salinity data from the same water mass can be misinterpreted as spatial gradients due to temporal sampling errors. If these spurious gradients are not eliminated, salinity changes caused by motion can be misinterpreted as true ocean stratification. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides an ocean salinity data correction method based on spatiotemporal gradient adaptive filtering to achieve the purpose of high-precision dynamic correction of ocean salinity data.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] A method for correcting ocean salinity data based on spatiotemporal gradient adaptive filtering includes the following steps:

[0008] Step 1: Calculate the temporal and spatial gradients of the salinity and temperature data collected by the CTD sensor on the Argo float, and compensate for the pseudo gradient caused by the sensor motion based on the actual vertical motion speed to obtain the motion-compensated salinity gradient and motion-compensated temperature gradient.

[0009] Step 2: A dual-gradient driven filtering window is constructed using the motion-compensated salinity gradient and the motion-compensated temperature gradient, and a gradient-sensitive weight distribution strategy is introduced to perform adaptive dynamic filtering on the salinity data.

[0010] Step 3: Map the motion-compensated salinity gradient as a control parameter to the learning rate, decision tree depth, and number of decision trees of the GBDT model to generate the initial hyperparameter combination of the GBDT model;

[0011] Step 4: Input the adaptive dynamic filtered salinity data into the GBDT model and optimize the initial hyperparameter combination of the model using the Gray Wolf Optimization Algorithm. In each iteration, the predicted salinity correction value output by the model is used to calculate the physical constraint loss function. The hyperparameter performance is evaluated based on the physical constraint loss function. The optimal hyperparameter combination of the model is obtained through a dynamic search strategy, and the final salinity correction value is obtained at the same time.

[0012] Step 5: Verify the final salinity correction value. If it does not meet the standard, modify the GBDT model hyperparameters or adaptive dynamic filter parameters until the standard is met.

[0013] In the above scheme, in step 1, the temporal gradient and spatial gradient of salinity data and temperature data include salinity temporal gradient, temperature temporal gradient, salinity spatial gradient, and temperature spatial gradient, which are specifically calculated as follows:

[0014] Salinity time gradient calculation formula:

[0015] ;

[0016] in, Indicates the sampling points, and Respectively represent Hedi The time of the sampling point, and Respectively indicate time points and Salinity measurements at

[0017] Temperature time gradient calculation formula:

[0018] ;

[0019] in, and Respectively indicate time points and The temperature measurement value at

[0020] The calculation formula of salinity spatial gradient is:

[0021] ;

[0022] in, and Respectively represent Hedi The pressure measurement value of each sampling point, and Represents pressure points and The salinity measurement at Indicates the density of the current layer of seawater, represents the gravitational acceleration of seawater;

[0023] The calculation formula of temperature spatial gradient is:

[0024] ;

[0025] in, and Represents pressure points and The temperature measurement value at .

[0026] In the above scheme, in step 1, the motion-compensated salinity gradient is calculated as follows:

[0027] ;

[0028] in, represents the salinity time gradient, represents the spatial gradient of salinity, Indicates the vertical movement speed of the sensor in seawater:

[0029] ;

[0030] in, and Represent the pressure measurement values ​​of the i+1th and i-1th sampling points respectively, and Represent the time of the i+1th and i-1th sampling points respectively, Indicates the density of the current layer of seawater, represents the gravitational acceleration of seawater;

[0031] The motion-compensated temperature gradient is calculated as follows:

[0032] ;

[0033] in, represents the temperature-time gradient, Represents the spatial temperature gradient.

[0034] In the above scheme, in step 2, the dual gradient driven filter window is constructed as follows:

[0035] ;

[0036] in, Indicates the filter window size, Indicates the minimum value of the window, Indicates the maximum value of the window, represents the salinity gradient weight, represents the temperature gradient weight, represents the normalized value of salinity gradient, represents the normalized value of temperature gradient;

[0037] ;

[0038] ;

[0039] in, represents the motion-compensated salinity gradient, represents the motion-compensated temperature gradient, 、 They represent the maximum and minimum absolute values ​​of the motion-compensated salinity gradient on the seawater profile, 、 They represent the maximum and minimum absolute values ​​of the motion-compensated temperature gradient on the seawater profile, respectively.

[0040] In the above scheme, in step 2, the calculation formula for adaptive dynamic filtering of salinity data is as follows:

[0041] ;

[0042] in, represents the salinity data after filtering, represents the raw salinity data, Display window Salinity data within represents the salinity gradient sensitivity weight;

[0043] ;

[0044] in, Indicates the rate at which the control weight changes with the gradient, represents the motion-compensated salinity gradient.

[0045] In the above solution, step 3 is as follows:

[0046] (1) The learning rate mapping formula is as follows:

[0047] ;

[0048] in, represents the learning rate of the model, represents the maximum learning rate, represents the attenuation coefficient, represents the motion-compensated salinity gradient, It represents the maximum absolute value of the motion-compensated salinity gradient on the seawater profile;

[0049] (2) The decision tree depth mapping formula is as follows:

[0050] ;

[0051] in, Indicates the depth of the decision tree;

[0052] (3) The decision tree quantity mapping formula is as follows:

[0053] ;

[0054] in, represents the number of decision trees, represents the minimum number of decision trees, represents the complexity scaling factor, Indicates the number of current section layered interfaces.

[0055] In the above solution, step 4 is as follows:

[0056] (1) Wolf pack initialization:

[0057] The initial hyperparameter combination generated in step 3 is used as the initial solution of the alpha wolf, and the remaining individuals are distributed within the neighborhood of the alpha wolf solution. The formula is as follows:

[0058] ;

[0059] in, represents the candidate hyperparameter vector, represents the initial solution vector of the wolf leader, Represents the basic neighborhood range vector, corresponding to the perturbation range of learning rate, decision tree depth, and number of decision trees, represents the hyperbolic tangent function, which is used to map the temperature gradient to a neighborhood scaling factor, represents the motion-compensated temperature gradient, It represents the maximum absolute value of the motion-compensated temperature gradient on the seawater profile;

[0060] (2) Iterative optimization:

[0061] For each candidate hyperparameter vector in each generation of the wolf pack , do the following:

[0062] use Train a temporary GBDT model with filtered salinity data as input , the output is the temporary predicted salinity correction value , calculate the physical constraint loss function; use the physical constraint loss function to evaluate the hyperparameter performance, so that each iteration finds the parameter combination with the minimum salinity error and the gradient and density that best conform to the physical laws: ;

[0063] (3) Dynamic search:

[0064] If the current layer motion compensates the salinity gradient , The standard deviation of the motion-compensated salinity gradient is determined to be a strong gradient area. At this time, the dynamic search strategy is triggered and the convergence factor is updated:

[0065] ;

[0066] in, represents the updated convergence factor, represents the initial convergence factor, Indicates the current iteration number, Indicates the maximum number of iterations;

[0067] Position update coefficient calculation:

[0068] ;

[0069] in, , express The random disturbance coefficient, represents the convergence factor; and Represents the position update coefficient, which is used to control the search range and adjust the influence of the optimal solution on the current solution;

[0070] The formula for updating the position after scaling is:

[0071] ;

[0072] in, Indicates the Iteration candidate parameter vector, represents the scaling factor, Indicates the The alpha wolf solution of the iteration, that is, the current optimal parameter vector;

[0073] (4) Termination and output:

[0074] When the maximum number of iterations is reached or the loss value of the leader changes for 5 consecutive times When , stop the iteration and output the optimal hyperparameters , use this hyperparameter to train the final GBDT model and output the final salinity correction value:

[0075] ;

[0076] in, represents the final salinity correction value, Indicates the The predicted correction value of the decision tree, represents the learning rate, represents the number of decision trees, Indicates the salinity value after filtering.

[0077] In the above scheme, in step 4, the physical constraint loss function formula is as follows:

[0078] ;

[0079] in, represents the absolute error of salinity, represents the density penalty term, represents the gradient similarity, and represents the weight coefficient;

[0080] Absolute error of salinity:

[0081] ;

[0082] Where N represents the total amount of data. Indicates the reference salinity value of the data, represents the predicted salinity correction value;

[0083] Gradient Similarity:

[0084] ;

[0085] in, Indicates the Data points predicted corrected data motion compensated salinity gradient, Indicates the Data points are referenced to the motion-compensated salinity gradient;

[0086] Density penalty:

[0087] ;

[0088] in, Indicates the Temporary predicted salinity correction value for data points, Indicates the The reference salinity value for each data point is Seawater density calculated for the UNESCO seawater equation of state.

[0089] In the above scheme, in step 5, the final salinity correction value is verified by indicators including statistical accuracy verification, as follows:

[0090] When the mean absolute error of salinity exceeds the threshold, the hyperparameters of the GBDT model are adjusted as follows:

[0091] Dynamic adjustment of model complexity:

[0092] ;

[0093] ;

[0094] in, is the corrected tree depth, is the corrected number of trees, Indicates the depth search range of the original decision tree, Indicates the search range of the original decision tree number, Represents the model complexity correction coefficient;

[0095] ;

[0096] in, Indicates the current mean absolute error of salinity, represents the statistical accuracy threshold, represents the absolute value of the motion-compensated salinity gradient, It represents the maximum absolute value of the motion-compensated salinity gradient on the seawater profile.

[0097] In the above scheme, in step 5, the final salinity correction value is verified by the index including the characteristic retention verification, as follows:

[0098] The gradient retention rate is used to measure the degree of retention of salinity gradient characteristics during the calibration process. The gradient retention rate calculation formula is as follows:

[0099] ;

[0100] Where GPR is the gradient retention rate, represents the motion-compensated salinity gradient of the i-th data point after correction, represents the salinity gradient of the motion-compensated reference data at the i-th data point, and N represents the total number of data points in the salinity profile;

[0101] When the gradient retention rate is less than the threshold When , the adaptive dynamic filter parameters are modified:

[0102] Maximum filter window adjustment:

[0103] ;

[0104] Salinity gradient sensitivity weight adjustment:

[0105] ;

[0106] in, represents the maximum filter window after correction, represents the maximum filter window, represents the salinity gradient sensitivity weight, represents the modified salinity gradient sensitivity weight, represents the adjustment factor;

[0107] ;

[0108] in, Represents the gradient retention rate threshold.

[0109] Through the above technical solution, the ocean salinity data correction method based on spatiotemporal gradient adaptive filtering provided by the present invention has the following beneficial effects:

[0110] 1. This method distinguishes sensor motion artifacts from true ocean stratification through the motion-compensated gradients of step 1 and the gradient similarity constraints of step 4, addressing the problem of water mass temporal deviations caused by Argo float sampling being misinterpreted as spatial gradients. This method effectively removes motion-induced spurious gradient components, making the spatial distribution of ocean stratification characteristics more realistic, and significantly improving the ability of salinity profile data to depict ocean dynamic processes.

[0111] 2. Through the dual-gradient driven filter window in step 2 and the closed-loop feature retention correction in step 5, this method preserves the true stratification characteristics in the thermohaline region, solving the problems of oversmoothing and feature loss caused by traditional fixed-window filtering methods. This method suppresses noise while maximally preserving the physical mutation signals in the thermohaline region, improving the accuracy of ocean stratification structure analysis based on salinity data.

[0112] 3. The present invention maps the motion compensation gradient to the GBDT model hyperparameters in step 3 and uses the Gray Wolf Optimization algorithm in step 4 to perform hyperparameter tuning, thus solving the problem of over-reliance on experience in parameter tuning in traditional machine learning methods. This method automatically associates physical gradient characteristics with algorithm parameters, achieving adaptive optimization of model hyperparameters. This significantly enhances the robustness of the salinity correction algorithm in different marine environments and significantly reduces the cost of manual tuning in engineering applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0113] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for describing the embodiments or the prior art.

[0114] Figure 1 This is a flow chart of a method for correcting ocean salinity data based on spatiotemporal gradient adaptive filtering disclosed in an embodiment of the present invention;

[0115] Figure 2 This is a diagram showing the effect of the salinity data profile correction of the present invention. DETAILED DESCRIPTION

[0116] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0117] The present invention provides a method for correcting ocean salinity data based on spatiotemporal gradient adaptive filtering, such as Figure 1 As shown, the following steps are included:

[0118] Step 1: First, the Argo float was equipped with a temperature, salinity, and depth sensor (CTD), named SHY and Sea-Bird, which meet international measurement standards. The sensors sampled the seawater at 19.5°N, 116.0°E vertically at regular intervals, acquiring data on conductivity, temperature, and pressure. The sensors moved at a constant vertical speed of 0.1 m / s, collecting data every meter at a depth interval of 0.1 m / s.

[0119] The salinity data obtained by the SHY sensor was used as the data to be corrected, and the salinity data obtained by the Sea-Bird sensor was used as the reference data.

[0120] The temporal and spatial gradients of the salinity and temperature data collected by the CTD sensor on the Argo float are calculated, and the pseudo gradient caused by the sensor motion is compensated in combination with the actual vertical motion velocity to obtain the motion-compensated salinity gradient and motion-compensated temperature gradient.

[0121] The temporal and spatial gradients of salinity and temperature data include salinity temporal gradient, temperature temporal gradient, salinity spatial gradient, and temperature spatial gradient. The specific calculations are as follows:

[0122] Salinity time gradient calculation formula:

[0123] ;

[0124] in, Indicates the sampling points, and Respectively represent Hedi The time of the sampling point, and Respectively indicate time points and Salinity measurements at

[0125] Temperature time gradient calculation formula:

[0126] ;

[0127] in, and Respectively indicate time points and The temperature measurement value at

[0128] The calculation formula of salinity spatial gradient is:

[0129] ;

[0130] in, and Respectively represent Hedi The pressure measurement value of each sampling point, and Represents pressure points and The salinity measurement at Indicates the density of the current layer of seawater, represents the gravitational acceleration of seawater;

[0131] The calculation formula of temperature spatial gradient is:

[0132] ;

[0133] in, and Represents pressure points and The temperature measurement value at .

[0134] The motion-compensated salinity gradient is calculated as follows:

[0135] ;

[0136] in, represents the salinity time gradient, represents the spatial gradient of salinity, Indicates the vertical movement speed of the sensor in seawater:

[0137] ;

[0138] in, and Represent the pressure measurement values ​​of the i+1th and i-1th sampling points respectively, and Represent the time of the i+1th and i-1th sampling points respectively, Indicates the density of the current layer of seawater, represents the gravitational acceleration of seawater;

[0139] Based on the UNESCO seawater state equation, the density of seawater at any pressure is calculated by coupling the standard atmospheric pressure density with the pressure compression effect:

[0140] ;

[0141] in, represents the density of seawater at standard atmospheric pressure, Indicates seawater pressure, It represents the bulk elastic modulus, which describes the ability of seawater to resist compression. It is calculated using the UNESCO standard formula and is applicable to the global ocean pressure range:

[0142] ;

[0143] in, is the bulk elastic modulus under standard atmospheric pressure, P is the seawater pressure, and S is the salinity. represents the bulk elastic modulus, and A and B represent the pressure correction coefficients.

[0144] The motion-compensated temperature gradient is calculated as follows:

[0145] ;

[0146] in, represents the temperature-time gradient, Represents the spatial temperature gradient.

[0147] Step 2: A dual-gradient driven filtering window is constructed using the motion-compensated salinity gradient and the motion-compensated temperature gradient, and a gradient-sensitive weight distribution strategy is introduced to perform adaptive dynamic filtering on the salinity data.

[0148] According to historical data testing, a window width less than 5 results in residual noise, and a window width greater than 15 results in a smooth transition between layers. Therefore, the initial window minimum value is =5, maximum value The window size is 15, and the temperature gradient is introduced to compensate for the thermal expansion effect while the salinity gradient dominates the window size. The temperature gradient and salinity gradient dual gradient driven filter window is constructed:

[0149] ;

[0150] in, Indicates the filter window size, Indicates the minimum value of the window, Indicates the maximum value of the window, represents the salinity gradient weight, represents the temperature gradient weight, represents the normalized value of salinity gradient, Represents the normalized value of the temperature gradient; normalizing the gradient can avoid the influence of different physical dimensions on the window calculation;

[0151] ;

[0152] ;

[0153] in, represents the motion-compensated salinity gradient, represents the motion-compensated temperature gradient, 、 They represent the maximum and minimum absolute values ​​of the motion-compensated salinity gradient on the seawater profile, 、 They represent the maximum and minimum absolute values ​​of the motion-compensated temperature gradient on the seawater profile, respectively.

[0154] The calculation formula for adaptive dynamic filtering of salinity data is as follows:

[0155] ;

[0156] in, represents the salinity data after filtering, represents the raw salinity data, Display window Salinity data within Represents the salinity gradient sensitivity weight; in order to prevent the traditional median filter from smoothing the real steep change characteristics of the jump layer, The balance between retaining the original data in the thermocline region and suppressing noise in the uniform region is achieved by adaptively weighting the salinity gradient.

[0157] ;

[0158] in, Indicates the rate at which the control weight changes with the gradient, represents the motion-compensated salinity gradient.

[0159] In step 3, the motion-compensated salinity gradient is mapped as a control parameter to the learning rate, decision tree depth, and number of decision trees of the GBDT (gradient boosted decision tree) model to generate the initial hyperparameter combination of the GBDT model.

[0160] The details are as follows:

[0161] (1) The learning rate is adaptively adjusted according to the salinity gradient. A smaller learning rate is used in the jump layer area with a larger gradient, and a larger learning rate is used in the uniform area. The learning rate mapping formula is as follows:

[0162] ;

[0163] in, represents the learning rate of the model, represents the maximum learning rate, represents the attenuation coefficient, represents the motion-compensated salinity gradient, It represents the maximum absolute value of the motion-compensated salinity gradient on the seawater profile;

[0164] (2) The depth of the decision tree is set hierarchically according to the size of the salinity gradient. The larger the gradient, the greater the tree depth, in order to capture complex nonlinear relationships. The decision tree depth mapping formula is as follows:

[0165] ;

[0166] in, Indicates the depth of the decision tree;

[0167] (3) The number of decision trees increases with the increase of stratification intensity and the number of stratification interfaces. The mapping formula for the number of decision trees is as follows:

[0168] ;

[0169] in, represents the number of decision trees, represents the minimum number of decision trees, represents the complexity scaling factor, Indicates the number of current section layered interfaces.

[0170] In step 4, the salinity data after adaptive dynamic filtering is input into the GBDT model, and the initial hyperparameter combination of the model is optimized using the gray wolf optimization algorithm. In each iteration, the predicted salinity correction value output by the model is used to calculate the physical constraint loss function, and the hyperparameter performance is evaluated based on the physical constraint loss function. The optimal hyperparameter combination of the model is obtained through a dynamic search strategy, and the final salinity correction value is obtained at the same time.

[0171] The details are as follows:

[0172] (1) Wolf pack initialization:

[0173] The initial hyperparameter combination generated in step 3 is used as the alpha wolf's initial solution. The remaining individuals are distributed within the neighborhood of the alpha wolf's solution. The greater the temperature gradient, the more concentrated the initial solution distribution. The formula is as follows:

[0174] ;

[0175] in, represents the candidate hyperparameter vector, represents the initial solution vector of the wolf leader, Represents the basic neighborhood range vector, corresponding to the perturbation range of learning rate, decision tree depth, and number of decision trees, represents the hyperbolic tangent function, which is used to map the temperature gradient to a neighborhood scaling factor, represents the motion-compensated temperature gradient, Indicates the maximum absolute value of the motion-compensated temperature gradient on the seawater profile

[0176] (2) Iterative optimization:

[0177] For each candidate hyperparameter vector in each generation of the wolf pack , do the following:

[0178] use Train a temporary GBDT model with filtered salinity data as input , the output is the temporary predicted salinity correction value , calculate the physical constraint loss function;

[0179] The formula of the physical constraint loss function is as follows:

[0180] ;

[0181] in, represents the absolute error of salinity, represents the density penalty term, represents the gradient similarity, and represents the weight coefficient;

[0182] Specifically, the absolute error of salinity:

[0183] ;

[0184] Where N represents the total amount of data. Indicates the reference salinity value of the data, represents the predicted salinity correction value;

[0185] Gradient Similarity:

[0186] ;

[0187] in, Indicates the Data points predicted corrected data motion compensated salinity gradient, Indicates the Data points are referenced to the motion-compensated salinity gradient;

[0188] Density penalty:

[0189] ;

[0190] in, Indicates the Temporary predicted salinity correction value for data points, Indicates the The reference salinity value for each data point is Seawater density calculated for the UNESCO seawater equation of state.

[0191] The physical constraint loss function is used to evaluate the hyperparameter performance, so that each iteration finds the parameter combination with the minimum salinity error and the gradient and density that best conform to the physical laws: ;

[0192] (3) Dynamic search:

[0193] If the current layer motion compensates the salinity gradient , The standard deviation of the motion-compensated salinity gradient is determined to be a strong gradient area. At this time, the traditional GWO is prone to fall into the local optimum, triggering the dynamic search strategy and updating the convergence factor:

[0194] ;

[0195] in, represents the updated convergence factor, represents the initial convergence factor, Indicates the current iteration number, Indicates the maximum number of iterations;

[0196] Position update coefficient calculation:

[0197] ;

[0198] in, , express The random disturbance coefficient, represents the convergence factor; and Represents the position update coefficient, which is used to control the search range and adjust the influence of the optimal solution on the current solution;

[0199] The formula for updating the position after scaling is:

[0200] ;

[0201] in, Indicates the Iteration candidate parameter vector, represents the scaling factor, Indicates the The alpha wolf solution of the iteration, that is, the current optimal parameter vector;

[0202] (4) Termination and output:

[0203] When the maximum number of iterations is reached or the loss value of the leader changes for 5 consecutive times When , stop the iteration and output the optimal hyperparameters , use this hyperparameter to train the final GBDT model and output the final salinity correction value:

[0204] ;

[0205] in, represents the final salinity correction value, Indicates the The predicted correction value of the decision tree, represents the learning rate, represents the number of decision trees, Indicates the salinity value after filtering.

[0206] Step 5: Verify the final salinity correction value. If it does not meet the standard, modify the GBDT model hyperparameters or adaptive dynamic filter parameters until the standard is met.

[0207] The final salinity correction value is verified by statistical accuracy verification, as follows:

[0208] When the mean absolute error of salinity exceeds the threshold, the hyperparameters of the GBDT model are adjusted as follows:

[0209] Dynamic adjustment of model complexity:

[0210] ;

[0211] ;

[0212] in, is the corrected tree depth, is the corrected number of trees, Indicates the depth search range of the original decision tree, Indicates the search range of the original decision tree number, Represents the model complexity correction coefficient;

[0213] ;

[0214] in, Indicates the current mean absolute error of salinity, Indicates the statistical accuracy threshold, such as 0.02PSU; represents the absolute value of the motion-compensated salinity gradient, It represents the maximum absolute value of the motion-compensated salinity gradient on the seawater profile.

[0215] The final salinity correction value is verified by the following indicators, including characteristic retention verification:

[0216] The gradient retention rate is used to measure the degree of retention of salinity gradient characteristics during the calibration process. The gradient retention rate calculation formula is as follows:

[0217] ;

[0218] Where GPR is the gradient retention rate, represents the motion-compensated salinity gradient of the i-th data point after correction, represents the salinity gradient of the motion-compensated reference data at the i-th data point, and N represents the total number of data points in the salinity profile;

[0219] When the gradient retention rate is less than the threshold ( %), the adaptive dynamic filter parameters are modified:

[0220] Maximum filter window adjustment:

[0221] ;

[0222] Salinity gradient sensitivity weight adjustment:

[0223] ;

[0224] in, represents the maximum filter window after correction, represents the maximum filter window, represents the salinity gradient sensitivity weight, represents the modified salinity gradient sensitivity weight, represents the adjustment factor;

[0225] ;

[0226] in, Represents the gradient retention rate threshold.

[0227] After correction, the salinity correction effect of the embodiment of the present invention is shown in FIG. Figure 2It can be seen that the method of the present invention makes the corrected salinity curve closely fit the reference data curve, the steep gradient of the thermocline zone is retained, and the noise in the uniform zone is effectively suppressed, which fully demonstrates the high-precision correction capability of the present invention in complex marine environments.

[0228] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for correcting ocean salinity data based on spatiotemporal gradient adaptive filtering, characterized in that: The steps include: Step 1: Calculate the temporal and spatial gradients of the salinity and temperature data collected by the CTD sensor on the Argo float, and compensate for the pseudo gradient caused by the sensor motion based on the actual vertical motion speed to obtain the motion-compensated salinity gradient and motion-compensated temperature gradient. Step 2: A dual-gradient driven filtering window is constructed using the motion-compensated salinity gradient and the motion-compensated temperature gradient, and a gradient-sensitive weight distribution strategy is introduced to perform adaptive dynamic filtering on the salinity data. Step 3: Map the motion-compensated salinity gradient as a control parameter to the learning rate, decision tree depth, and number of decision trees of the GBDT model to generate the initial hyperparameter combination of the GBDT model; Step 4: Input the adaptive dynamic filtered salinity data into the GBDT model and optimize the initial hyperparameter combination of the model using the Gray Wolf Optimization Algorithm. In each iteration, the predicted salinity correction value output by the model is used to calculate the physical constraint loss function. The hyperparameter performance is evaluated based on the physical constraint loss function. The optimal hyperparameter combination of the model is obtained through a dynamic search strategy, and the final salinity correction value is obtained at the same time. Step 5: Verify the final salinity correction value. If it does not meet the standard, modify the GBDT model hyperparameters or adaptive dynamic filter parameters until the standard is met.

2. The ocean salinity data correction method based on spatiotemporal gradient adaptive filtering according to claim 1, characterized in that: In step 1, the temporal gradient and spatial gradient of salinity data and temperature data include salinity temporal gradient, temperature temporal gradient, salinity spatial gradient, and temperature spatial gradient, which are specifically calculated as follows: Salinity time gradient calculation formula: ； in, Indicates the sampling points, and Respectively represent Hedi The time of the sampling point, and Respectively indicate time points and Salinity measurements at Temperature time gradient calculation formula: ； in, and Respectively indicate time points and The temperature measurement value at The calculation formula of salinity spatial gradient is: ； in, and Respectively represent Hedi The pressure measurement value of each sampling point, and Represents pressure points and The salinity measurement at Indicates the density of the current layer of seawater, represents the gravitational acceleration of seawater; The calculation formula of temperature spatial gradient is: ； in, and Represents pressure points and The temperature measurement value at .

3. The ocean salinity data correction method based on spatiotemporal gradient adaptive filtering according to claim 1 is characterized in that: In step 1, the motion-compensated salinity gradient is calculated as follows: ； in, represents the salinity time gradient, represents the spatial gradient of salinity, Indicates the vertical movement speed of the sensor in seawater: ； in, and Represent the pressure measurement values ​​of the i+1th and i-1th sampling points respectively, and Represent the time of the i+1th and i-1th sampling points respectively, Indicates the density of the current layer of seawater, represents the gravitational acceleration of seawater; The motion-compensated temperature gradient is calculated as follows: ； in, represents the temperature-time gradient, Represents the spatial temperature gradient.

4. The ocean salinity data correction method based on spatiotemporal gradient adaptive filtering according to claim 1, characterized in that: In step 2, the dual gradient driven filter window is constructed as follows: ； in, Indicates the filter window size, Indicates the minimum value of the window, Indicates the maximum value of the window, represents the salinity gradient weight, represents the temperature gradient weight, represents the normalized value of salinity gradient, represents the normalized value of temperature gradient; ； ； in, represents the motion-compensated salinity gradient, represents the motion-compensated temperature gradient, 、 They represent the maximum and minimum absolute values ​​of the motion-compensated salinity gradient on the seawater profile, 、 They represent the maximum and minimum absolute values ​​of the motion-compensated temperature gradient on the seawater profile, respectively.

5. The ocean salinity data correction method based on spatiotemporal gradient adaptive filtering according to claim 1, characterized in that: In step 2, the calculation formula for adaptive dynamic filtering of salinity data is as follows: ； in, represents the salinity data after filtering, represents the raw salinity data, Display window Salinity data within represents the salinity gradient sensitivity weight; ； in, Indicates the rate at which the control weight changes with the gradient, represents the motion-compensated salinity gradient.

6. The ocean salinity data correction method based on spatiotemporal gradient adaptive filtering according to claim 1, characterized in that: Step 3 is as follows: (1) The learning rate mapping formula is as follows: ； in, represents the learning rate of the model, represents the maximum learning rate, represents the attenuation coefficient, represents the motion-compensated salinity gradient, It represents the maximum absolute value of the motion-compensated salinity gradient on the seawater profile; (2) The decision tree depth mapping formula is as follows: ； in, Indicates the depth of the decision tree; (3) The decision tree quantity mapping formula is as follows: ； in, represents the number of decision trees, represents the minimum number of decision trees, represents the complexity scaling factor, Indicates the number of current section layered interfaces.

7. The ocean salinity data correction method based on spatiotemporal gradient adaptive filtering according to claim 1, characterized in that: Step 4 is as follows: (1) Wolf pack initialization: The initial hyperparameter combination generated in step 3 is used as the initial solution of the alpha wolf, and the remaining individuals are distributed within the neighborhood of the alpha wolf solution. The formula is as follows: ； in, represents the candidate hyperparameter vector, represents the initial solution vector of the wolf leader, Represents the basic neighborhood range vector, corresponding to the perturbation range of learning rate, decision tree depth, and number of decision trees, represents the hyperbolic tangent function, which is used to map the temperature gradient to a neighborhood scaling factor, represents the motion-compensated temperature gradient, Indicates the maximum absolute value of the motion-compensated temperature gradient on the seawater profile (2) Iterative optimization: For each candidate hyperparameter vector in each generation of the wolf pack , do the following: use Train a temporary GBDT model with filtered salinity data as input , the output is the temporary predicted salinity correction value , calculate the physical constraint loss function; use the physical constraint loss function to evaluate the hyperparameter performance, so that each iteration finds the parameter combination with the minimum salinity error and the gradient and density that best conform to the physical laws: ; (3) Dynamic search: If the current layer motion compensates the salinity gradient , The standard deviation of the motion-compensated salinity gradient is determined to be a strong gradient area. At this time, the dynamic search strategy is triggered and the convergence factor is updated: ； in, represents the updated convergence factor, represents the initial convergence factor, Indicates the current iteration number, Indicates the maximum number of iterations; Position update coefficient calculation: ； in, , express The random disturbance coefficient, represents the convergence factor; and Represents the position update coefficient, which is used to control the search range and adjust the influence of the optimal solution on the current solution; The formula for updating the position after scaling is: ； in, Indicates the Iteration candidate parameter vector, represents the scaling factor, Indicates the The alpha wolf solution of the iteration, that is, the current optimal parameter vector; (4) Termination and output: When the maximum number of iterations is reached or the loss value of the leader changes for 5 consecutive times When , stop the iteration and output the optimal hyperparameters , use this hyperparameter to train the final GBDT model and output the final salinity correction value: ； in, represents the final salinity correction value, Indicates the The predicted correction value of the decision tree, represents the learning rate, represents the number of decision trees, Indicates the salinity value after filtering.

8. The ocean salinity data correction method based on spatiotemporal gradient adaptive filtering according to claim 1, characterized in that: In step 4, the physical constraint loss function formula is as follows: ； in, represents the absolute error of salinity, represents the density penalty term, represents the gradient similarity, and represents the weight coefficient; Absolute error of salinity: ； Where N represents the total amount of data. Indicates the reference salinity value of the data, represents the predicted salinity correction value; Gradient Similarity: ； in, Indicates the Data points predicted corrected data motion compensated salinity gradient, Indicates the Data points are referenced to the motion-compensated salinity gradient; Density penalty: ； in, Indicates the Temporary predicted salinity correction value for data points, Indicates the The reference salinity value for each data point is Seawater density calculated for the UNESCO seawater equation of state.

9. The ocean salinity data correction method based on spatiotemporal gradient adaptive filtering according to claim 1, characterized in that: In step 5, the final salinity correction value is verified by indicators including statistical accuracy verification, as follows: When the mean absolute error of salinity exceeds the threshold, the hyperparameters of the GBDT model are adjusted as follows: Dynamic adjustment of model complexity: ； ； in, is the corrected tree depth, is the corrected number of trees, Indicates the depth search range of the original decision tree, Indicates the search range of the original decision tree number, Represents the model complexity correction coefficient; ； in, Indicates the current mean absolute error of salinity, represents the statistical accuracy threshold, represents the absolute value of the motion-compensated salinity gradient, It represents the maximum absolute value of the motion-compensated salinity gradient on the seawater profile.

10. The ocean salinity data correction method based on spatiotemporal gradient adaptive filtering according to claim 1, characterized in that: In step 5, the final salinity correction value is verified by the following indicators, including characteristic retention verification: The gradient retention rate is used to measure the degree of retention of salinity gradient characteristics during the calibration process. The gradient retention rate calculation formula is as follows: ； Where GPR is the gradient retention rate, represents the motion-compensated salinity gradient of the i-th data point after correction, represents the salinity gradient of the motion-compensated reference data at the i-th data point, and N represents the total number of data points in the salinity profile; When the gradient retention rate is less than the threshold When , the adaptive dynamic filter parameters are modified: Maximum filter window adjustment: ； Salinity gradient sensitivity weight adjustment: ； in, represents the maximum filter window after correction, represents the maximum filter window, represents the salinity gradient sensitivity weight, represents the modified salinity gradient sensitivity weight, represents the adjustment factor; ； in, Represents the gradient retention rate threshold.

Images