Temperature time alignment and error compensation method for a ctd

Abstract

The application discloses a temperature time domain alignment and error compensation method of a temperature-salinity-depth instrument, and relates to the technical field of temperature-salinity-depth instrument measurement, and the specific steps are as follows: firstly, the original temperature sequence of the temperature-salinity-depth instrument and the temperature true value sequence of a high-precision reference instrument are synchronously collected, time domain alignment of the original temperature sequence is completed through root mean square error minimization search, based on the aligned temperature sequence, the first-order temperature change rate and the second-order temperature change rate dynamic characteristics are calculated, and a multi-dimensional feature vector is constructed; the actual residual error of the aligned temperature sequence and the temperature true value sequence is taken as a learning target, a random forest nonlinear regression model is trained, a residual error compensation amount is fitted, and a final corrected temperature sequence is output. Compared with the traditional linear model method, the root mean square error of temperature measurement of the model method is significantly reduced, the nonlinear dynamic error under the strong dynamic variable temperature working condition is effectively inhibited, the measurement precision and the robustness of the temperature-salinity-depth instrument are improved, and the application requirements of ocean observation are met.

Description

Technical Field

[0001] This invention relates to the field of temperature, salinity, and depth (TDM) measurement technology, and specifically to a method for temperature time-domain alignment and error compensation in TDM. Background Technology

[0002] CTD (Conductivity, Temperature, Depth) meters are fundamental equipment for oceanographic observation. Low-cost CTDs are widely used in ocean profiling and long-term observation due to their low cost and flexible deployment. However, in actual profiling, the temperature field may change rapidly, causing insufficient measurement response speed and resulting in dynamic errors.

[0003] Existing technologies typically employ time alignment correction combined with physical model or data fitting compensation: First, a low-cost temperature sequence and a high-precision reference sequence are obtained. Similarity analyses, such as cross-correlation, are used to estimate the systematic time delay between the two. The low-cost temperature sequence is then shifted as a whole according to this time delay to achieve initial waveform alignment. Subsequently, to further reduce errors, a fixed-parameter thermal hysteresis correction model is used to compensate for the aligned sequence, or a regression model is directly established to fit the error and output the compensation amount to form a corrected temperature sequence. While this approach is easy to implement in engineering, under conditions of rapid temperature changes, nonlinear dynamic residuals related to the rate of change may still exist after alignment, and fixed-parameter models struggle to cover complex temperature variations. Summary of the Invention

[0004] This invention provides a method for temperature time-domain alignment and error compensation of a temperature, salinity, and depth instrument, in order to solve the problem in the prior art that fixed parameter models are difficult to cover complex temperature variations under rapidly changing temperature conditions.

[0005] A method for temperature time-domain alignment and error compensation of a temperature-salinity-depth (TDI) instrument includes the following steps:

[0006] S1. Collect the original temperature sequence of the temperature, salinity and depth instrument and the true temperature sequence of the high-precision reference instrument. Perform timestamp unification, resampling and noise reduction and smoothing on the original temperature sequence and the true temperature sequence to ensure that the time base of the original temperature sequence and the true temperature sequence are consistent.

[0007] S2. By calculating the root mean square error between the original temperature sequence and the true temperature sequence after translation under different time shifts, the optimal systematic time delay of the temperature, salinity and depth instrument is determined by minimizing the root mean square error. The original temperature sequence is then time-shifted according to the time delay to obtain the time-domain aligned temperature sequence.

[0008] S3. Based on the time-domain aligned temperature sequence, calculate the first-order and second-order temperature change rates that characterize the dynamic temperature change, and combine them with the aligned temperature sequence to construct a multi-dimensional feature vector.

[0009] S4. Using the residual between the time-domain aligned temperature sequence and the true sequence as the learning objective, a nonlinear regression model is trained based on multidimensional feature vectors. The residual compensation amount is obtained by fitting the nonlinear regression model. The time-domain aligned temperature sequence is superimposed with the compensation amount to obtain the final corrected temperature sequence.

[0010] Further, S1 specifically includes: setting the temperature-salinity-depth meter and the high-precision reference instrument to use the same sampling rate, and acquiring the original temperature sequence output by the temperature-salinity-depth meter under the condition of large gradient dynamic temperature variation. Temperature true value sequence output by high-precision reference instrument For the original temperature sequence With temperature true value sequence Perform timestamp unification, noise reduction, smoothing, and outlier removal.

[0011] Furthermore, S2 specifically includes: setting a time shift amount. The search interval is used to calculate the original temperature sequence after translation. With temperature true value sequence The root mean square error between them is selected by the time shift that makes the root mean square error reach its global minimum. As the optimal systematic time delay, the original temperature sequence Time shifting is performed according to the optimal systematic delay to obtain the time-domain aligned temperature sequence.

[0012] Further, S3 specifically involves: first performing smoothing filtering on the time-domain aligned temperature sequence, then using the central difference method to calculate the first-order temperature change rate and the second-order temperature change rate, and combining the time-domain aligned temperature value, the first-order temperature change rate, and the second-order temperature change rate at each moment to form a multi-dimensional feature vector.

[0013] Furthermore, the formula for calculating the root mean square error is as follows:

[0014] ;

[0015] In the formula, The root mean square error, Original temperature sequence Translation The corresponding temperature value at that time. For the true temperature sequence exist Temperature value at any given time.

[0016] Furthermore, the formula for calculating the first-order rate of temperature change is:

[0017] ;

[0018] In the formula, For the first Each sampling time The corresponding first-order rate of temperature change, For the first Each sampling time Temperature value after time-domain alignment For the first Each sampling time Temperature value after time-domain alignment The sampling time interval;

[0019] The formula for calculating the second-order rate of temperature change is:

[0020] ;

[0021] In the formula, For the first Each sampling time The corresponding second-order rate of temperature change, For the first Each sampling time The corresponding first-order rate of temperature change, For the first Each sampling time The corresponding first-order rate of temperature change.

[0022] Furthermore, the multidimensional feature vector is:

[0023] ;

[0024] In the formula, For the first Each sampling time The corresponding multidimensional feature vector, For the first Each sampling time Temperature values ​​after time-domain alignment For the first Each sampling time The corresponding first-order rate of temperature change, For the first Each sampling time The corresponding second-order rate of temperature change.

[0025] Furthermore, the residual, residual compensation amount, and final correction temperature sequence in S4 satisfy:

[0026] ;

[0027] In the formula, Sampling time Temperature residual, Sampling time The corresponding true temperature value, Sampling time Temperature value after time-domain alignment;

[0028] ;

[0029] In the formula, Sampling time The residual compensation amount, It is a nonlinear regression model. Sampling time The corresponding multidimensional feature vector;

[0030] ;

[0031] In the formula, Sampling time The corresponding final temperature correction sequence, Sampling time Temperature values ​​after time-domain alignment.

[0032] Furthermore, the nonlinear regression model is a random forest regression model.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows: the temperature dynamic error of the temperature-salinity-depth instrument is decomposed into two parts: phase misalignment caused by systematic time delay and nonlinear dynamic residual after time domain alignment, and processed separately. Global time domain registration is achieved through automatically determined time delay parameters, and the residual error after alignment is compensated on this basis. At the same time, a residual compensation model is established with physical consistency dynamic characteristics such as temperature change rate as the core driver. Finally, the corrected temperature sequence is output in the form of aligned temperature + predicted residual, which significantly improves the temperature measurement accuracy and stability under strong gradient dynamic profile conditions. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the sensitivity curve of the error index for the thermal hysteresis delay parameter.

[0035] Figure 2 This is a schematic diagram comparing the temperature sequences before time-domain alignment.

[0036] Figure 3 This is a schematic diagram showing the comparison of temperature sequences after time-domain alignment.

[0037] Figure 4 A comparison chart of time series forecasts;

[0038] Figure 5 A comparison chart showing the effects of dynamic response error correction;

[0039] Figure 6 A comparison chart of the predicted residual distributions from the model;

[0040] Figure 7 This is a graph showing the rate of change of prediction error with temperature.

[0041] Figure 8 A robustness comparison chart across consecutive data blocks;

[0042] Figure 9 A comparison chart of absolute errors under different rates of change;

[0043] Figure 10 A scatter plot showing the physical consistency between the compensation amount and the rate of temperature change;

[0044] Figure 11 A comparison chart of root mean square errors for different feature combinations and model structures;

[0045] Figure 12 This is a flowchart of the technology of the present invention. Detailed Implementation

[0046] 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.

[0047] like Figure 12 As shown, a method for temperature time-domain alignment and error compensation of a temperature-salinity-depth (TDI) instrument includes:

[0048] S1. A self-developed low-cost temperature, salinity, and depth instrument was selected as the observation device to be evaluated, and a high-precision reference instrument was simultaneously installed to obtain true temperature data; the sampling rate of both the low-cost self-developed temperature, salinity, and depth instrument and the high-precision reference instrument was set to 1Hz.

[0049] This embodiment conducted a water tank experiment simulating dynamic temperature variation measurement. Standard seawater was injected into the tank, and a temperature-controlled reflux system was used to induce controllable, continuous, and rapid temperature changes within the tank, simulating the large-gradient dynamic temperature variation conditions experienced by rapidly traversing the thermocline in a real ocean. During the dynamic temperature variation process, observational data acquired by a high-precision reference instrument was used as the true temperature sequence. The original temperature series observed by the low-cost, self-developed temperature, salinity, and depth instrument is... The collected raw temperature sequence With temperature true value sequence Timestamp unification and resampling are performed to bring them to a unified time base, while noise reduction and smoothing are also completed to lay the foundation for subsequent steps.

[0050] S2. Based on S1 and The optimal systematic time delay of the temperature, salinity, and depth instrument was determined by minimizing the root mean square error (RMSE). The specific process is as follows: Set the time shift amount. The search interval is used to calculate the result of translation with respect to different translation amounts. RMSE between; such as Figure 1 As shown, RMSE in translation amount The global minimum is reached at 79s. Figure 1 The curves clearly reflect the RMSE variation trends corresponding to different delay parameters. The error is minimized at 79s, indicating that this low-cost temperature, salinity, and depth instrument has a stable systematic physical time delay of approximately 79s. Shifting the original temperature sequence by 79s yields a time-domain aligned temperature sequence, denoted as... The time-domain alignment operation is completed; the deviation between the original temperature sequence and the true temperature sequence is as follows: Figure 2 As shown, the consistency between the true temperature sequence and the time-domain aligned temperature sequence is as follows: Figure 3 As shown.

[0051] S3. Time-domain aligned temperature sequence obtained from S2 Based on this, we calculate the features representing dynamic temperature changes and concatenate them to construct a multi-dimensional feature vector. The specific steps are as follows:

[0052] Dynamic feature calculation: Obtaining the sampling time interval of the temperature, salinity, and depth instrument. To suppress the amplification effect of high-frequency observation noise during the differentiation process, the time-domain aligned temperature sequence is first... Smoothing filtering is performed; then the central difference method is used to calculate the first-order temperature change rate (temperature gradient) and the second-order temperature change rate (temperature acceleration) of each sampling point.

[0053] First-order rate of temperature change:

[0054] ;

[0055] In the formula, For the first Each sampling time The corresponding first-order rate of temperature change, For the first Each sampling time Temperature value after time-domain alignment For the first Each sampling time Temperature value after time-domain alignment The sampling time interval;

[0056] Second-order temperature change rate:

[0057] ;

[0058] In the formula, For the first Each sampling time The corresponding second-order rate of temperature change, For the first Each sampling time The corresponding first-order rate of temperature change, For the first Each sampling time The corresponding first-order rate of temperature change;

[0059] For each sampling time Extract the time-domain aligned temperature value at the corresponding time point. First-order temperature change rate and second-order temperature change rate Concatenate to construct a multidimensional feature vector The specific expression is:

[0060] ;

[0061] In the formula, For the first Each sampling time The corresponding multidimensional feature vector, For the first Each sampling time Temperature values ​​after time-domain alignment For the first Each sampling time The corresponding first-order rate of temperature change, For the first Each sampling time The corresponding second-order rate of temperature change;

[0062] Multidimensional feature vectors are used to drive the training of subsequent nonlinear regression models.

[0063] S4. Using the residual between the time-domain aligned temperature sequence and the true temperature sequence as the learning objective, a nonlinear regression model is trained based on the multidimensional feature vector constructed in S3, the residual compensation amount is fitted, and the final corrected temperature sequence is obtained.

[0064] Calculate the time-domain aligned temperature series With temperature true value sequence residual :

[0065] ;

[0066] In the formula, Sampling time Temperature residual, Sampling time The corresponding true temperature value, Sampling time Temperature value after time-domain alignment;

[0067] The residual is used to characterize the dynamic error mechanism caused by heat conduction, and the rate of temperature change, as the core feature, has a consistent physical correlation with the magnitude of the error.

[0068] Using the multidimensional feature vector constructed by S3 as input, and the residual To achieve the learning objective, train a nonlinear regression model. In this embodiment, a random forest regression model is selected, and the residual compensation amount is obtained by fitting the random forest regression model. :

[0069] ;

[0070] In the formula, Sampling time Residual compensation amount;

[0071] Time-domain aligned temperature sequence The residual compensation amount obtained by fitting the model Superimpose the results to obtain the final corrected temperature sequence. :

[0072] );

[0073] In the formula, Sampling time The corresponding final temperature correction sequence, Sampling time Temperature values ​​after time-domain alignment.

[0074] Time series prediction comparison chart as follows Figure 4 As shown, the model method of this invention has a high degree of consistency with the true temperature sequence. The correction effect of dynamic response error can be achieved through... Figure 5 Intuitive observation clearly demonstrates that the error correction method employed in this invention using a random forest regression model more closely approximates the true temperature sequence; the distribution characteristics and concentration of the model's predicted residuals are as follows: Figure 6 As shown, by Figure 6 It can be seen that the residuals of the random forest regression model of this invention are smaller, verifying the rationality and reliability of the random forest regression model of this invention. Figure 7 As shown, under rapid temperature change conditions, the maximum error of the traditional linear model can reach about 0.3℃, while the method of the present invention can control the maximum error to about 0.15℃, clearly demonstrating the error difference between the two methods under different temperature change rates. Figure 8The error performance of the traditional linear model and the model method of the present invention is shown in different continuous data blocks, and the model method of the present invention demonstrates its good robustness. Figure 9 The error levels of the model method of this invention and the traditional linear model method were quantitatively compared under various temperature variations. The temperature variations transitioned from stable to moderate to severe, corresponding to a gradual increase in the temperature gradient (rate of temperature change). Figure 9 It can be seen that the absolute error of the model method of the present invention is smaller under different temperature gradients, further highlighting the advantages of the model method of the present invention. Figure 10 The scatter plot of the physical consistency between the temperature change rate and the compensation amount verifies the physical correlation between the model compensation amount and the temperature change rate. Figure 11 By comparing the RMSE performance of different feature combinations and model structures, the superiority of the multidimensional feature + random forest model framework of this invention is verified. The experimental results fully demonstrate that the method of this invention has stronger adaptability and robustness under strong dynamic temperature conditions, can effectively correct the temperature dynamic error of the temperature, salinity and depth instrument, significantly improve the measurement accuracy, and meet the needs of practical applications.

[0075] 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 method for temperature time-domain alignment and error compensation in a temperature, salinity, and depth instrument, characterized in that, Includes the following steps: S1. Collect the original temperature sequence of the temperature, salinity and depth instrument and the true temperature sequence of the high-precision reference instrument. Perform timestamp unification, resampling and noise reduction and smoothing on the original temperature sequence and the true temperature sequence to ensure that the time base of the original temperature sequence and the true temperature sequence are consistent. S2. By calculating the root mean square error between the original temperature sequence and the true temperature sequence after translation under different time shifts, the optimal systematic time delay of the temperature, salinity and depth instrument is determined by minimizing the root mean square error. The original temperature sequence is then time-shifted according to the time delay to obtain the time-domain aligned temperature sequence. S3. Based on the time-domain aligned temperature sequence, calculate the first-order and second-order temperature change rates that characterize the dynamic temperature change, and combine them with the aligned temperature sequence to construct a multi-dimensional feature vector. S4. Using the residual between the time-domain aligned temperature sequence and the true sequence as the learning objective, a nonlinear regression model is trained based on multidimensional feature vectors. The residual compensation amount is obtained by fitting the nonlinear regression model. The time-domain aligned temperature sequence is superimposed with the compensation amount to obtain the final corrected temperature sequence.

2. The method for temperature time-domain alignment and error compensation of a temperature, salinity, and depth instrument according to claim 1, characterized in that, S1 specifically includes: setting the temperature, salinity, and depth instrument to use the same sampling rate as the high-precision reference instrument, and acquiring the original temperature sequence output by the temperature, salinity, and depth instrument under large gradient dynamic temperature variation conditions. Temperature true value sequence output by high-precision reference instrument For the original temperature sequence With temperature true value sequence Perform timestamp unification, noise reduction, smoothing, and outlier removal.

3. The method for temperature time-domain alignment and error compensation of a temperature, salinity, and depth instrument according to claim 1, characterized in that, S2 specifically includes: setting a time shift amount. The search interval is used to calculate the original temperature sequence after translation. With temperature true value sequence The root mean square error between them is selected by the time shift that makes the root mean square error reach its global minimum. As the optimal systematic time delay, the original temperature sequence Time shifting is performed according to the optimal systematic delay to obtain the time-domain aligned temperature sequence.

4. The method for temperature time-domain alignment and error compensation of a temperature, salinity, and depth instrument according to claim 1, characterized in that, Specifically, S3 involves: first performing smoothing filtering on the time-domain aligned temperature sequence, then using the central difference method to calculate the first-order and second-order temperature change rates, and combining the time-domain aligned temperature value, the first-order temperature change rate, and the second-order temperature change rate at each moment to form a multi-dimensional feature vector.

5. The method for temperature time-domain alignment and error compensation of a temperature, salinity, and depth instrument according to claim 3, characterized in that, The formula for calculating the root mean square error is: ； In the formula, The root mean square error, Original temperature sequence Translation Later corresponding time Temperature value, For the true temperature sequence exist Temperature value at any given time.

6. The method for temperature time-domain alignment and error compensation of a temperature, salinity, and depth instrument according to claim 4, characterized in that, The formula for calculating the first-order rate of temperature change is: ； In the formula, For the first Each sampling time The corresponding first-order rate of temperature change, For the first Each sampling time Temperature value after time-domain alignment For the first Each sampling time Temperature value after time-domain alignment The sampling time interval; The formula for calculating the second-order rate of temperature change is: ； In the formula, For the first Each sampling time The corresponding second-order rate of temperature change, For the first Each sampling time The corresponding first-order rate of temperature change, For the first Each sampling time The corresponding first-order rate of temperature change.

7. The method for temperature time-domain alignment and error compensation of a temperature, salinity, and depth instrument according to claim 4, characterized in that, The multidimensional feature vector is: ； In the formula, For the first Each sampling time The corresponding multidimensional feature vector, For the first Each sampling time Temperature values ​​after time-domain alignment For the first Each sampling time The corresponding first-order rate of temperature change, For the first Each sampling time The corresponding second-order rate of temperature change.

8. The method for temperature time-domain alignment and error compensation of a temperature, salinity, and depth instrument according to claim 1, characterized in that, The residual, residual compensation amount, and final correction temperature sequence in S4 satisfy the following: ； In the formula, Sampling time Temperature residual, Sampling time The corresponding true temperature value, Sampling time Temperature value after time-domain alignment; ； In the formula, Sampling time The residual compensation amount, It is a nonlinear regression model. Sampling time The corresponding multidimensional feature vector; ； In the formula, Sampling time The corresponding final temperature correction sequence, Sampling time Temperature values ​​after time-domain alignment.

9. The method for temperature time-domain alignment and error compensation of a temperature, salinity, and depth instrument according to claim 1, characterized in that, The nonlinear regression model is a random forest regression model.

Images