Method for identifying turbulence influence temperature-salinity-depth data of portable underwater glider

Abstract

The invention relates to the field of electrical digital processing, and discloses a turbulence influence temperature-salinity-depth data identification method for a portable underwater glider, which comprises the following steps of: acquiring temperature, salinity and depth data acquired by the portable underwater glider, and preprocessing the temperature, salinity and depth data; determining a speed threshold range of the portable underwater glider; calculating a strain rate tensor, a vortex tensor, a pulsation strain rate tensor and a pulsation vortex tensor for data points outside a speed threshold range, and jointly constructing a feature matrix with temperature, salinity and depth data; inputting the constructed feature matrix into an XGBoost algorithm model, and training to obtain a trained model; evaluating the performance of the model; and inputting temperature, salinity and depth data acquired by the portable underwater glider to be identified into the qualified model, and outputting data influenced by turbulence by the model. According to the method disclosed by the invention, the data points influenced by turbulence can be effectively and accurately identified, and misjudgment during passing through the thermocline and halocline can be avoided.

Description

Technical Field

[0001] The invention relates to the field of electrical digital processing, and in particular to a method for identifying temperature, salinity and depth data of turbulence effects on a portable underwater glider. Background Art

[0002] With the development of marine technology, portable underwater gliders are more widely used, mainly for measuring ocean temperature, salinity and depth data. However, underwater gliders will pass through the thermohaline layer during operation in the ocean, and will be affected by turbulence during movement, which can easily cause errors or singular values ​​in the obtained temperature, salinity and depth data.

[0003] At present, the data identification of portable underwater gliders mainly uses conventional statistical methods to determine the existence of spikes or singular values, and processes the temperature-salinity-depth data through median filtering and peace filtering. However, the temperature-salinity-depth data processed by the above method can only identify abnormal values ​​when there are large fluctuations in the thermohaline jump layer, and cannot identify data affected by turbulence. And if the above traditional method is used to identify data points affected by turbulence, not only will the correct data when passing through the thermohaline jump layer be misidentified, but it will also cause the problem of distortion of temperature-salinity-depth data due to inaccurate parameter settings. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for identifying turbulence-affected temperature, salinity and depth data of a portable underwater glider, so as to achieve the purpose of accurately identifying data affected by turbulence.

[0005] To achieve the above object, the technical solution of the present invention is as follows:

[0006] A method for identifying temperature, salinity and depth data of turbulence effects on a portable underwater glider comprises the following steps: Step 1, obtaining temperature, salinity and depth data collected by a portable underwater glider and preprocessing them; Step 2, calculating the underwater movement speed of the portable underwater glider according to the acquired depth data, and determining a speed threshold; Step 3, for data points outside the velocity threshold range, the strain rate tensor, vortex tensor, pulsating strain rate tensor and pulsating vortex tensor are calculated, and a feature matrix is ​​constructed together with the temperature, salinity and depth data; Step 4, model training: input the constructed feature matrix into the XGBoost algorithm model, the model outputs the identified data affected by turbulence, uses the real data of temperature, salinity, and depth affected by turbulence to calculate the loss, and adjusts the model parameters through the optimization algorithm to obtain a trained model; Step 5: Evaluate the performance of the model at different depths using recall, precision, and F1 score metrics; Step 6, the temperature, salinity and depth data collected by the portable underwater glider to be identified are input into the qualified model, and the model outputs the data affected by turbulence.

[0007] In the above solution, the salinity data is the collected conductivity data, and the depth data is the collected pressure data.

[0008] In the above scheme, in step 1, the preprocessing method is as follows: fill in the missing data by linear interpolation, divide the overall temperature, conductivity and pressure data into up and down segments, obtain the diving and floating data respectively, eliminate obvious abnormal data points, eliminate conductivity data exceeding 0 to 6S / m, eliminate pressure data less than or equal to 0dbar, and eliminate temperature data outside the temperature range of -2.5 to 40°C.

[0009] In the above scheme, in step 2, the calculation formula for the underwater movement speed of the portable underwater glider is as follows: ; in, represents the moving speed of the portable underwater glider, Respectively represent the depth data of the current moment and the depth data of the previous moment, Indicates the sampling frequency.

[0010] In the above scheme, the method for determining the speed threshold range is as follows: according to the obtained motion speed , calculate the average value u and variance α of the speed, the speed threshold range S is: u-α≤S≤u+α, and the data points outside the speed threshold range are data points less than u-α and greater than u+α.

[0011] In the above scheme, the strain rate tensor is calculated as follows: ; The calculation formula of the vortex tensor is as follows: ; The calculation formula of the pulsating strain rate tensor is as follows: ; The calculation formula of the pulsation vortex tensor is as follows: ; in, Represents two different directions in the spatial coordinate system; They are The velocity component in the direction; They are The spatial component of direction; Represents the velocity component The rate of change of Represents the velocity component The rate of change of Represents spatial component The rate of change of Represents spatial component The rate of change of and The velocity components are The pulsating part, namely the velocity pulsating component; Indicates the velocity pulsation component The rate of change, Indicates the velocity pulsation component The rate of change.

[0012] In the above scheme, each time the model recognizes the input data, at each node of each decision tree, XGBoost will evaluate all available features and select a feature for classification to achieve the minimum value of the objective function; starting from an empty decision tree, all data are at the root node. For the current node, XGBoost will traverse all features and possible split points and calculate the split gain Gain of each split point: ; in, are the sum of the first-order derivatives of the left and right child nodes after splitting, are the sum of the second-order derivatives of the left and right child nodes after splitting; is a regularization parameter used to control the complexity of the model; is a regularization parameter used to control the number of leaf nodes;

[0013] If the split gain Gain is positive, splitting will occur; if it is negative or zero, no splitting will occur.

[0014] In a further technical solution, the objective function is as follows: ; Where n represents the total number of samples, i represents the i-th sample, is the recognition result of turbulence by the t-1th decision tree, represents the eigenvalue of the i-th sample, is the prediction function of the tth decision tree; represents the error function, The turbulence state actually observed for the i-th sample is is the regularization function, and the formula is as follows: ; Where T represents the number of decision trees, j represents the jth decision tree, Represents the node value of the jth decision tree;

[0015] By splitting and dividing the nodes of the decision tree, the minimum node value can be obtained, so that the objective function can achieve the optimal solution.

[0016] Through the above technical solution, the method for identifying temperature, salinity and depth data of turbulence effects on a portable underwater glider provided by the present invention has the following beneficial effects: 1. The present invention adopts the XGBoost algorithm model for turbulence data recognition. The algorithm will reconstruct the decision tree for the data points that do not perfectly identify turbulence. The features selected when each decision tree is split are different. This is to better fit the residual. Each decision tree will present the final recognition result. However, due to the existence of the learning rate, even if a decision tree is not accurately recognized, it will not dominate the entire result. The final result is jointly constructed by the recognition of all decision trees. Therefore, this method can effectively and accurately identify the data points affected by turbulence; 2. Compared with conventional data statistics methods, due to the setting of the velocity threshold, this method can only process data points outside the velocity threshold range, which can avoid wrong judgment when passing through the thermohaline layer; 3. The present invention selects 7 characteristic quantities to construct a characteristic matrix, which can provide a more comprehensive description of flow characteristics, help the model identify turbulence from multiple angles, and avoid the problem of inaccurate turbulence identification caused by selecting a single characteristic quantity. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0018] Figure 1 A schematic flow chart of a method for identifying temperature, salinity and depth data of turbulence effects on a portable underwater glider disclosed in an embodiment of the present invention; Figure 2 A flowchart of a decision tree; Figure 3 The temperature data obtained by the portable underwater glider from 0 to 100 meters is obtained by the XGBoost algorithm model, which is affected by turbulence; Figure 4 The conductivity data obtained by the portable underwater glider from 0-100 meters are used to obtain the data points affected by turbulence through the XGBoost algorithm model. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0020] The present invention provides a method for identifying temperature, salinity and depth data of turbulence effects on a portable underwater glider. Figure 1 As shown, the following steps are included: Step 1, obtaining temperature, salinity and depth data collected by a portable underwater glider and preprocessing them; The salinity data collected by the portable underwater glider is conductivity data, and the depth data is pressure data.

[0021] The preprocessing method is as follows: fill in the missing data through linear interpolation, divide the overall temperature, conductivity and pressure data into up and down segments, obtain the diving and floating data respectively, and eliminate obvious abnormal data points: eliminate conductivity data exceeding 0-6S / m, eliminate pressure data less than or equal to 0dbar, and eliminate temperature data outside of -2.5-40℃, to ensure that the data for analysis have no obvious abnormal values.

[0022] Step 2, calculating the underwater movement speed of the portable underwater glider according to the acquired depth data, and determining a speed threshold; The calculation formula for the underwater speed of a portable underwater glider is as follows: ; in, represents the moving speed of the portable underwater glider, Respectively represent the depth data of the current moment and the depth data of the previous moment, Indicates the sampling frequency.

[0023] Since the speed of the portable underwater glider does not change significantly without the influence of turbulence, the obtained movement speed , calculate the average value u and variance α of the speed, and the speed threshold range S is: u-α≤S≤u+α. Data points outside the speed threshold range refer to data points that are less than u-α and greater than u+α.

[0024] The speed threshold setting is not fixed. The threshold can be changed by judging the performance of the algorithm through the recall rate, precision rate and F1 score. By introducing the speed threshold setting, it can be ensured that the mutation point when passing through the thermohaline layer will not affect the judgment of whether it is turbulence or not, which can effectively ensure the authenticity of the data.

[0025] Step 3: For data points outside the velocity threshold range, the strain rate tensor, vortex tensor, pulsating strain rate tensor and pulsating vortex tensor are calculated, and a feature matrix is ​​constructed together with the temperature, salinity and depth data.

[0026] The selection of characteristic quantities in the XGBoot algorithm uses a combination of original data and flow invariants. The original data are temperature, pressure, and conductivity. The invariants are strain rate tensor and vortex tensor, pulsating strain rate tensor, and pulsating vortex tensor.

[0027] The strain rate tensor contains the tensile and shear deformation information of the fluid. The calculation formula of the strain rate tensor is as follows: ; The vortex tensor describes the rotational motion of the fluid element. The calculation formula of the vortex tensor is as follows: ; The pulsating strain rate tensor is used to describe the local deformation rate in turbulence. The calculation formula of the pulsating strain rate tensor is as follows: ; The pulsating vortex tensor is used to describe the local rotation rate in turbulence. The calculation formula of the pulsating vortex tensor is as follows: ; in, Represents two different directions in the spatial coordinate system; They are The velocity component in the direction; They are The spatial component of direction; Represents the velocity component The rate of change of Represents the velocity component The rate of change of Represents spatial component The rate of change of Represents spatial component The rate of change of and The velocity components are The pulsating part, namely the velocity pulsating component; Indicates the velocity pulsation component The rate of change, Indicates the velocity pulsation component The rate of change.

[0028] Step 4, model training: Input the constructed feature matrix into the XGBoost algorithm model, the model outputs the identified data affected by turbulence, uses the real data of temperature, salinity, and depth affected by turbulence to calculate the loss, and adjusts the model parameters through the optimization algorithm to obtain a trained model.

[0029] Each time the model recognizes the input data, at each node of each decision tree, XGBoost will evaluate all available features and select a feature for classification to achieve the minimum value of the objective function; starting from an empty decision tree, all data are at the root node. For the current node, XGBoost will traverse all features and possible split points and calculate the split gain Gain of each split point: ; in, are the sum of the first-order derivatives of the left and right child nodes after splitting, are the sum of the second-order derivatives of the left and right child nodes after splitting; is a regularization parameter used to control the complexity of the model; is a regularization parameter used to control the number of leaf nodes;

[0030] The split gain is an indicator to measure the split effect. It measures the reduction in loss before and after a node splits. It is calculated based on the reduction in the objective function. If the split gain is positive, the node can split. If it is negative or zero, the node will not split.

[0031] The objective function is as follows: ; Where n represents the total number of samples, i represents the i-th sample, is the recognition result of turbulence by the t-1th decision tree, that is, The sum of the prediction results of the decision trees; represents the eigenvalue of the i-th sample, is the prediction function of the tth decision tree; represents the error function, The turbulence state actually observed for the i-th sample is is the regularization function, and the formula is as follows: ; Where T represents the number of decision trees, j represents the jth decision tree, Represents the node value of the jth decision tree;

[0032] The regularization function can limit the maximum depth of the decision tree to ensure the complexity of the algorithm, ensure that the results can be settled at the minimum complexity, and prevent overfitting. After simplification, the objective function can be regarded as The quadratic function of the decision tree can obtain the minimum node value by splitting and dividing the nodes, so that the objective function can achieve the optimal solution.

[0033] The final output of each decision tree of the algorithm is to output a probability value between 0 and 1 for each data point according to the two states of turbulence and non-turbulence, indicating the possibility that the point is affected by turbulence. This probability value is obtained by summing up the prediction results of all decision trees. Finally, in order to reasonably record the data points affected by turbulence, the data points affected by turbulence are finally determined by setting the threshold. Finally, according to the setting, label 1 is affected by turbulence, and label 0 is not affected by turbulence. Stored in a table containing temperature, salinity and depth data.

[0034] like Figure 2 As shown in the figure, for example, at the root node, the strain rate tensor can provide the maximum splitting gain, so this feature is used to split the node into subnode 1 and subnode 2. The data points in subnode 1 are turbulent, and the data in subnode 2 are non-turbulent. Regardless of whether the data points in node 1 or node 2 are split according to the feature that can provide the maximum splitting gain. Figure 2 As shown, during the splitting process, non-turbulent points may contain turbulence, and turbulent points may contain non-turbulence. This requires continuous splitting and refinement through different features among the 7 features (such as feature 2 or feature 3) until the data point ends or the maximum depth of the decision tree is reached (the maximum depth and complexity of the decision tree are controlled by the regularization function.) Each decision tree is constructed for the data points that the previous decision tree failed to perfectly identify, that is, to fit the residuals and continuously optimize the previous decision tree to achieve the best recognition.

[0035] Step 5: Evaluate the performance of the model at different depths using the recall, precision, and F1 score indicators.

[0036] Since temperature, salinity and depth data are depth-dependent, the performance of the model at different depths needs to be considered when evaluating the model. For example, the recall rate, precision rate and F1 score of the model at different depths can be analyzed to see whether the model can accurately capture the turbulence effects at each depth level. When evaluating, it is necessary to divide the data into different depth intervals and calculate the recall rate, precision rate and F1 score in each interval respectively. Then, if the overall evaluation is qualified, the evaluation of each depth layer is also guaranteed to be qualified.

[0037] ; R is the recall rate, TP is the number of data points correctly predicted by the model as turbulent, and FN is the number of turbulent data points incorrectly predicted by the model as non-turbulent.

[0038] ; P is the precision, and FP is the number of non-turbulent data points that the model incorrectly predicts as turbulent.

[0039] ; The F1 score is the harmonic mean of precision and recall, which comprehensively considers the balance between precision and recall.

[0040] By combining the recall rate, precision rate and F1 score obtained from the temperature, salinity and depth data at different depth levels with the overall recall rate, precision rate and F1 score for comprehensive evaluation, we can effectively avoid the misjudgment of turbulence data points caused by different temperature, salinity and depth characteristics at different depths.

[0041] Step 6: Input the temperature, salinity and depth data collected by the portable underwater glider to be identified into the qualified model, and the model outputs the data affected by turbulence. Figure 3 and Figure 4 The temperature and conductivity data obtained by the portable underwater glider at 0-100 meters are shown in the XGBoost algorithm model to obtain data points affected by turbulence; the circles represent data points affected by turbulence. The data affected by turbulence are processed using a three-level low-pass filter to reduce the impact of turbulence on temperature, salinity and depth data.

[0042] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be 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 will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying temperature, salinity and depth data of turbulence effects on a portable underwater glider, characterized in that: The steps include: Step 1, obtaining temperature, salinity and depth data collected by a portable underwater glider and preprocessing them; Step 2, calculating the underwater movement speed of the portable underwater glider according to the acquired depth data, and determining a speed threshold range; Step 3, for data points outside the velocity threshold range, the strain rate tensor, vortex tensor, pulsating strain rate tensor and pulsating vortex tensor are calculated, and a feature matrix is ​​constructed together with the temperature, salinity and depth data; Step 4, model training: input the constructed feature matrix into the XGBoost algorithm model, the model outputs the identified data affected by turbulence, uses the real data of temperature, salinity, and depth affected by turbulence to calculate the loss, and adjusts the model parameters through the optimization algorithm to obtain a trained model; Step 5: Evaluate the recognition performance of the model at different depths through recall, precision, and F1 score indicators; Step 6, input the temperature, salinity and depth data collected by the portable underwater glider to be identified into the qualified model, and the model outputs the data affected by turbulence.

2. The method for identifying temperature, salinity and depth data of turbulence effects of a portable underwater glider according to claim 1, characterized in that: The salinity data is the collected conductivity data, and the depth data is the collected pressure data.

3. The method for identifying turbulence-affected temperature, salinity and depth data of a portable underwater glider according to claim 2, characterized in that: In step 1, the preprocessing method is as follows: fill in the missing data by linear interpolation, divide the overall temperature, conductivity and pressure data into up and down segments, obtain the diving and floating data respectively, eliminate obvious abnormal data points, eliminate the conductivity data exceeding 0 to 6 S / m, eliminate the pressure data less than or equal to 0 dbar, and eliminate the temperature data outside the temperature range of -2.5 to 40 °C.

4. The method for identifying turbulence-affected temperature, salinity and depth data of a portable underwater glider according to claim 2, characterized in that: In step 2, the calculation formula for the underwater movement speed of the portable underwater glider is as follows: ； in, represents the moving speed of the portable underwater glider, Respectively represent the depth data of the current moment and the depth data of the previous moment, Indicates the sampling frequency.

5. The method for identifying temperature, salinity and depth data of turbulence effects of a portable underwater glider according to claim 4, characterized in that: The method for determining the speed threshold range is as follows: Based on the obtained motion speed , calculate the average value u and variance α of the speed, the speed threshold range S is: u-α≤S≤u+α, and the data points outside the speed threshold range are data points less than u-α and greater than u+α.

6. The method for identifying temperature, salinity and depth data of turbulence effects of a portable underwater glider according to claim 1, characterized in that: The strain rate tensor is calculated as follows: ； The calculation formula of the vortex tensor is as follows: ； The calculation formula of the pulsating strain rate tensor is as follows: ； The calculation formula of the pulsation vortex tensor is as follows: ； in, Represents two different directions in the spatial coordinate system; They are The velocity component in the direction; They are The spatial component of direction; Represents the velocity component The rate of change of Represents the velocity component The rate of change of Represents spatial component The rate of change of Represents spatial component The rate of change of and The velocity components are The pulsating part, namely the velocity pulsating component; Indicates the velocity pulsation component The rate of change, Indicates the velocity pulsation component The rate of change.

7. The method for identifying temperature, salinity and depth data of turbulence effects of a portable underwater glider according to claim 1, characterized in that: Each time the model recognizes the input data, at each node of each decision tree, XGBoost will evaluate all available features and select a feature for classification to achieve the minimum value of the objective function; starting from an empty decision tree, all data are at the root node. For the current node, XGBoost will traverse all features and possible split points and calculate the split gain Gain of each split point: ； in, are the sum of the first-order derivatives of the left and right child nodes after splitting, are the sum of the second-order derivatives of the left and right child nodes after splitting; is a regularization parameter used to control the complexity of the model; is a regularization parameter used to control the number of leaf nodes; If the split gain Gain is positive, splitting will occur; if it is negative or zero, no splitting will occur.

8. The method for identifying temperature, salinity and depth data of turbulence effects of a portable underwater glider according to claim 7, characterized in that: The objective function is as follows: ； Where n represents the total number of samples, i represents the i-th sample, is the recognition result of turbulence by the t-1th decision tree; represents the eigenvalue of the i-th sample, is the prediction function of the tth decision tree; represents the error function, The turbulence state actually observed for the i-th sample is is the regularization function, and the formula is as follows: ； Where T represents the number of decision trees, j represents the jth decision tree, Represents the node value of the jth decision tree; By splitting and dividing the nodes of the decision tree, the minimum node value can be obtained, so that the objective function can achieve the optimal solution.

Images