A method for in-situ calibration of waves based on GNSS buoys

Abstract

The application discloses a wave in-situ calibration method based on a GNSS buoy and belongs to the field of wave in-situ calibration. The method comprises the following steps: obtaining buoy deviation data based on position and attitude data of the GNSS buoy under different sea conditions; constructing a deviation compensation model and a deviation sliding model based on the buoy deviation data; obtaining a deviation compensation result based on the deviation compensation model and the deviation sliding model, wherein the deviation compensation result comprises first deviation compensation and second deviation compensation; and fusing the deviation compensation result with GNSS positioning data to obtain a wave parameter estimation result.

Description

Technical Field

[0001] This invention belongs to the field of wave in-situ calibration technology, and particularly relates to a wave in-situ calibration method based on GNSS buoys. Background Technology

[0002] In in-situ wave calibration based on GNSS buoys, the buoy's own deviation is a key technical issue. When a buoy floats at sea, it is affected by complex sea conditions such as wind, waves, and ocean currents, which can cause deviations in its position and attitude, resulting in inconsistencies between GNSS positioning data and actual wave conditions. This deviation is time-varying and nonlinear, and closely related to sea state. The buoy deviation can be decomposed into horizontal and vertical vector components, with the horizontal deviation being more significantly affected by ocean currents, while the vertical deviation is more strongly correlated with ocean waves. Furthermore, the buoy deviation may exhibit a certain slip characteristic, meaning that the deviation changes over time with a certain trend. This slip characteristic is closely related to the changing sea state process, reflecting the dynamic cumulative effect of the deviation. Therefore, accurately modeling and compensating for buoy deviation is crucial to improving the accuracy of in-situ wave calibration. Traditional deviation models often assume that the deviation is static or linear, making it difficult to fully characterize the buoy deviation characteristics under complex sea conditions, thus limiting calibration accuracy. There is an urgent need to conduct in-depth research on the spatiotemporal characteristics and nonlinear mechanisms of buoy deviation, and to develop deviation modeling and compensation methods that adapt to dynamic sea conditions, so as to provide technical support for improving the accuracy of in-situ wave calibration. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a wave in-situ calibration method based on GNSS buoys, thereby resolving the issues present in the prior art.

[0004] To achieve the above objectives, this invention provides a wave in-situ calibration method based on GNSS buoys, comprising:

[0005] Buoy deviation data were obtained based on the position and attitude data of GNSS buoys under different sea states;

[0006] Based on the buoy deviation data, a deviation compensation model and a deviation sliding model are constructed.

[0007] The deviation compensation result is obtained based on the deviation compensation model and the deviation sliding model, wherein the deviation compensation result includes a first deviation compensation and a second deviation compensation;

[0008] The deviation compensation result is fused with GNSS positioning data to obtain the wave parameter estimation result.

[0009] Optionally, a deviation compensation model is constructed based on the buoy deviation data to obtain a first deviation compensation. The process of obtaining the first deviation compensation includes:

[0010] Acquire real-time location information and historical trajectory data of GNSS buoys;

[0011] Based on the real-time location information and historical trajectory data, the horizontal and vertical displacements of the GNSS buoys are analyzed using a Kalman filter algorithm.

[0012] Calculate the horizontal and vertical components of the buoy deviation based on the horizontal and vertical displacements of the GNSS buoy;

[0013] Acquire multi-source marine environmental information, and extract features from the multi-source marine environmental information to obtain the key environmental factors affecting buoy deviation;

[0014] A horizontal deviation compensation model for the horizontal component is constructed based on the ocean current factor among the key environmental factors mentioned above.

[0015] A vertical deviation compensation model for the vertical component is constructed based on the wave factor among the key environmental factors mentioned above.

[0016] The first deviation compensation is obtained based on the horizontal deviation compensation model and the vertical deviation compensation model.

[0017] Optionally, the process of constructing the deviation sliding model includes:

[0018] Buoy deviation data within a preset time range is obtained based on a sliding window.

[0019] The feature vector of the buoy deviation sliding characteristics is obtained by analyzing the buoy deviation data based on the feature extraction algorithm;

[0020] The feature vector is input into the deviation modeling module to obtain the cumulative change law and construct the deviation sliding model;

[0021] The predicted deviation value is calculated based on the cumulative change pattern, and the compensation value for the buoy deviation is also calculated.

[0022] Based on the aforementioned deviation sliding model, the compensation value and the current buoy deviation data are weighted and fused to obtain the second deviation compensation.

[0023] Optionally, the process of inputting the feature vector into the deviation modeling module to obtain the deviation sliding model further includes model optimization, which includes:

[0024] Historical deviation data under different sea state conditions are obtained, the historical deviation data are analyzed to obtain the correlation strength between each sea state factor and the deviation characteristics, and typical sea states are obtained based on the correlation strength.

[0025] A deviation prediction model is constructed based on the aforementioned typical sea conditions;

[0026] The comprehensive deviation prediction results under typical sea conditions are calculated based on the deviation prediction model.

[0027] The comprehensive deviation prediction result is introduced into the control system to optimize the second deviation compensation, resulting in an optimized deviation sliding model.

[0028] Optionally, the process of fusing the deviation compensation result with GNSS positioning data to obtain the wave parameter estimation result includes:

[0029] Acquire in-situ wave observation data and synchronized GNSS positioning data;

[0030] An in-situ wave calibration model is constructed based on the aforementioned in-situ wave observation data.

[0031] The deviation compensation result is input into the wave in-situ calibration model to obtain the deviation compensation result of the wave in-situ observation data.

[0032] The particle filter algorithm is used to fuse the deviation compensation results of the in-situ wave observation data with the synchronous GNSS positioning data to obtain the fused wave observation data;

[0033] The least squares method is used to estimate the parameters of the fused wave observation data to obtain the wave parameter estimation results.

[0034] Optionally, after parameter estimation of the fused wave observation data, the method further includes establishing a deviation compensation effect evaluation system by introducing external high-precision wave observation data;

[0035] Specifically, the wave parameter estimation results are compared with the in-situ wave observation data to calculate the improvement in wave in-situ calibration accuracy.

[0036] A preset threshold is set based on the external high-precision wave observation data;

[0037] The wave in-situ calibration accuracy is determined by comparing the improvement in the wave in-situ calibration accuracy with the preset threshold.

[0038] Optionally, when the improvement in the in-situ wave calibration accuracy reaches a preset threshold, the wave parameter estimation result is output as the final in-situ wave calibration result.

[0039] If the improvement in the accuracy of the in-situ wave calibration does not reach the preset threshold, then the in-situ wave observation data and GNSS positioning data are reacquired, and the in-situ wave calibration model is optimized until the preset threshold is met.

[0040] The present invention also provides a computer terminal device, comprising:

[0041] One or more processors;

[0042] A memory, coupled to the processor, for storing one or more programs;

[0043] When the one or more programs are executed by the one or more processors, the one or more processors implement a wave in-situ calibration method based on GNSS buoys.

[0044] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a wave in-situ calibration method based on a GNSS buoy.

[0045] Compared with the prior art, the present invention has the following advantages and technical effects:

[0046] This invention discloses a machine learning-based GNSS buoy deviation compensation method to improve the accuracy of in-situ wave calibration. The method analyzes buoy position and attitude data under different sea states and establishes a nonlinear mapping model to describe the spatiotemporal characteristics of buoy deviation. Deviation compensation models are constructed for both horizontal and vertical directions, considering the influence of ocean currents and waves. A sliding window technique and an adaptive weight adjustment strategy are employed to improve the timeliness and robustness of deviation compensation. An ensemble learning method is used to integrate prediction results from multiple algorithms, forming a more reliable deviation compensation model. Finally, the deviation compensation results are fused with GNSS positioning data to generate high-precision wave parameter estimates, and an evaluation system is established to continuously optimize the compensation effect, achieving a significant improvement in the accuracy of in-situ wave calibration. Attached Figure Description

[0047] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0048] Figure 1 This is a flowchart of the wave in-situ calibration method based on GNSS buoys according to an embodiment of the present invention;

[0049] Figure 2 This is a flowchart illustrating the prediction of overall deviation and attitude compensation in an embodiment of the present invention;

[0050] Figure 3 This is a flowchart of the wave in-situ calibration and optimization process according to an embodiment of the present invention. Detailed Implementation

[0051] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0052] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0053] Example 1

[0054] like Figures 1-3 As shown, this embodiment provides a wave in-situ calibration method based on GNSS buoys, including the following steps:

[0055] Based on the position and attitude data of GNSS buoys under different sea conditions, a nonlinear mapping model between buoy deviation and sea condition factors is established using machine learning algorithms. This results in a dynamic model describing the spatiotemporal characteristics of buoy deviation, which can characterize the nonlinear and time-varying features of buoy deviation as sea conditions change.

[0056] For the horizontal and vertical components of buoy deviation, deviation compensation models are constructed respectively. The horizontal deviation compensation model focuses on the influence of ocean currents, while the vertical deviation compensation model focuses on the influence of ocean waves. By integrating multi-source marine environmental information, refined compensation for the horizontal and vertical components of buoy deviation is achieved.

[0057] By employing the sliding window technique, buoy deviation data within a certain time range is analyzed to extract the sliding characteristics of the deviation. By establishing a deviation sliding model, the dynamic cumulative effect of buoy deviation can be characterized and compensated, thereby improving the timeliness and continuity of deviation compensation.

[0058] In the deviation modeling process, an adaptive weight adjustment strategy is introduced to dynamically adjust the weights of each sea state factor in the deviation model based on the significant differences in deviation characteristics under different sea state conditions, thereby improving the adaptability and robustness of the deviation model.

[0059] A wave in-situ calibration model based on deviation compensation is constructed, and the deviation compensation results are fused with GNSS positioning data to generate high-precision wave parameter estimation results, thereby improving the accuracy of wave in-situ calibration.

[0060] By employing an ensemble learning approach, the prediction results of multiple machine learning algorithms, such as support vector machines and random forests, are combined to form a more robust bias compensation model, thereby improving the reliability and accuracy of bias compensation.

[0061] Establish a deviation compensation effect evaluation system. By introducing external high-precision wave observation data, evaluate the wave in-situ calibration accuracy before and after deviation compensation. Based on this, iteratively optimize the deviation compensation model and strategy to achieve continuous improvement in wave in-situ calibration accuracy.

[0062] As a specific implementation method of this embodiment, the following steps are included:

[0063] Buoy deviation data is obtained based on the position and attitude data of GNSS buoys under different sea states. This includes the following steps:

[0064] Based on the position and attitude data of GNSS buoys under different sea conditions, a nonlinear mapping model between buoy deviation and sea condition factors is established using machine learning algorithms. This results in a dynamic model describing the spatiotemporal characteristics of buoy deviation, which can characterize the nonlinear and time-varying features of buoy deviation as sea conditions change.

[0065] Furthermore, position and attitude data collected by GNSS buoys under different sea conditions are obtained to construct training and test datasets. The training dataset is used to train the machine learning model, and the test dataset is used to evaluate the model performance.

[0066] A nonlinear mapping model was trained using a support vector machine regression algorithm, with sea state parameters as input features and buoy deviation as the output target.

[0067] During training, the hyperparameters of the nonlinear mapping model are optimized through cross-validation to improve the generalization ability of the nonlinear mapping model.

[0068] The trained nonlinear mapping model was applied to the test dataset to evaluate the accuracy of the nonlinear mapping model in predicting buoy deviations under different sea conditions.

[0069] If the prediction accuracy of the nonlinear mapping model meets the preset threshold, then the nonlinear mapping model is determined as the optimal model.

[0070] If the prediction accuracy of the nonlinear mapping model does not meet the preset threshold, the structure or hyperparameters of the nonlinear mapping model are adjusted, and the training and testing are repeated.

[0071] By using the optimal model and combining real-time sea state parameters, the buoy deviation can be dynamically predicted to vary with time and space.

[0072] Based on the prediction results, a dynamic model describing the spatiotemporal characteristics of buoy deviation is generated;

[0073] The dynamic model was analyzed to study the nonlinear and time-varying characteristics of buoy deviation as sea state changes.

[0074] For different sea conditions, a dynamic model is used to predict the changing trend of buoy deviation;

[0075] If the prediction results show that the buoy deviation exceeds the preset threshold, an early warning mechanism will be triggered to prompt the user to take appropriate measures.

[0076] We continuously collect position and attitude data of GNSS buoys under actual sea conditions, regularly update the training dataset, and retrain the nonlinear mapping model. Through continuous iterative optimization, we improve the dynamic model's ability to characterize the spatiotemporal characteristics of buoy deviations and enhance the robustness and adaptability of the dynamic model.

[0077] Furthermore, to construct training and testing datasets, 1000 sets of position and attitude data were collected using GNSS buoys under different sea conditions, such as wave heights of 5 meters, 1 meter, and 5 meters, and wind speeds of 5 m / s, 10 m / s, and 15 m / s. 800 sets were used as the training set, and 200 sets as the testing set. When training the nonlinear mapping model, sea state parameters such as wave height and wind speed were used as input features, and buoy deviation distance was used as the output target. A support vector machine regression algorithm was employed, and 5-fold cross-validation was used to optimize the model's hyperparameters, such as the penalty coefficient C and kernel function type, to improve the model's generalization ability. The trained model was applied to the testing set to evaluate the prediction accuracy under different sea conditions. If the mean absolute error was less than 1 meter, it was determined to be the optimal model; otherwise, the model structure was adjusted, such as changing the kernel function or increasing the number of samples, and retrained for optimization. Using the optimal model, combined with real-time sea state parameters (updated every 10 minutes), the changes in buoy deviation over the next hour were predicted to generate a spatiotemporal dynamic model. Analysis revealed a non-linear relationship between buoy deviation and wave height. For example, when wave height increased from 5 meters to 5 meters, the deviation distance increased from 8 meters to 5 meters, an increase of 215%. The deviation distance also exhibited time-varying characteristics; for instance, at a wind speed of 10 m / s, the deviation distance first increased and then decreased over time, showing periodic fluctuations with a period of approximately 20 minutes. For situations where wave heights exceed 5 meters, a warning mechanism is triggered if the buoy deviation exceeds 3 meters, prompting the user to take measures such as increasing the anchor chain length or adjusting the buoy position. 100 new sets of position and attitude data are collected monthly to update the training set, retrain and optimize the model, and continuously improve the dynamic model's characterization ability and adaptability.

[0078] A deviation compensation model and a deviation sliding model are constructed based on buoy deviation data. This includes the following steps:

[0079] Furthermore, a deviation compensation model is constructed based on the buoy deviation data to obtain the first deviation compensation. The process of obtaining the first deviation compensation includes: acquiring the real-time position information and historical trajectory data of the GNSS buoy; analyzing the horizontal and vertical displacements of the GNSS buoy using the Kalman filter algorithm based on the real-time position information and historical trajectory data; calculating the horizontal and vertical components of the buoy deviation based on the horizontal and vertical displacements of the GNSS buoy; acquiring multi-source marine environmental information and extracting features from the multi-source marine environmental information to obtain the key environmental factors affecting the buoy deviation; constructing a horizontal deviation compensation model for the horizontal component based on the ocean current factor among the key environmental factors; constructing a vertical deviation compensation model for the vertical component based on the wave factor among the key environmental factors; and obtaining the first deviation compensation based on the horizontal and vertical deviation compensation models.

[0080] For the horizontal and vertical components of buoy deviation, deviation compensation models are constructed respectively. The horizontal deviation compensation model focuses on the influence of ocean currents, while the vertical deviation compensation model focuses on the influence of ocean waves. By integrating multi-source marine environmental information, refined compensation for the horizontal and vertical components of buoy deviation is achieved, namely, the first deviation compensation.

[0081] Furthermore, real-time position information and historical trajectory data of the buoy are acquired. By analyzing the horizontal and vertical displacements of the buoy, the horizontal and vertical components of the buoy deviation are determined. Multi-source marine environmental information, including ocean current data, wave data, wind speed data, and wind direction data, is acquired. Through data fusion and feature extraction of this multi-source marine environmental information, the key environmental factors affecting the buoy deviation are identified. For the horizontal component of the buoy deviation, a horizontal deviation compensation model incorporating the influence of ocean currents is constructed, and a machine learning algorithm is used to establish a nonlinear mapping relationship between the ocean current factor and the horizontal deviation. For the vertical component of the buoy deviation, a vertical deviation compensation model incorporating the influence of waves is constructed, and time series analysis is used to establish this model. The method establishes a dynamic relationship between wave factors and vertical deviation, which are key environmental factors. An adaptive adjustment mechanism is introduced into the horizontal and vertical deviation compensation models to adjust the model parameters in real time based on changes in buoy position and the dynamic changes in key environmental factors, resulting in adjusted horizontal and vertical deviation compensation amounts. These adjusted amounts are then fused to obtain a comprehensive compensation vector for buoy deviation. This comprehensive compensation vector is applied to the buoy's position information to obtain compensated buoy position information. The compensated buoy position information is transmitted to the monitoring center for refined tracking and management of the buoy.

[0082] Furthermore, real-time buoy location information is acquired via a GNSS positioning module, and historical trajectory data is stored in a cloud database. The horizontal and vertical displacements of the buoy are estimated using a Kalman filter algorithm, yielding the horizontal and vertical components of the buoy deviation. The root mean square error (RMSE) of the horizontal displacement estimation is 5 meters, and the RMSE of the vertical displacement estimation is 8 meters. Multi-source marine environmental information, including ocean currents, waves, wind speed, and wind direction, is obtained from marine environmental monitoring stations. Data fusion technology is used to normalize the heterogeneous data, extracting key environmental factors affecting buoy deviation. Principal component analysis reveals that ocean current velocity and wave height are the main factors influencing both horizontal and vertical buoy deviations, with a cumulative contribution rate exceeding 85%. For buoy horizontal deviation, a compensation model incorporating ocean current influences is constructed. A support vector machine regression algorithm is used to establish a nonlinear mapping relationship between ocean current velocity and horizontal deviation, with a mean absolute error of 2 meters after 5-fold cross-validation. To address buoy vertical deviation, a compensation model incorporating wave influence was constructed. An ARIMA time series model was used to establish the dynamic relationship between wave height and vertical deviation, with a root mean square error (RMSE) of 6 meters for the fitting residuals. An adaptive adjustment mechanism was introduced into the compensation model, updating model parameters every 10 minutes based on changes in buoy position and environmental factors, adjusting the horizontal and vertical deviation compensation amounts in real time. After 100 iterations, the absolute error of the horizontal deviation compensation model decreased by 3 meters, and the RMS error of the vertical deviation compensation model decreased by 2 meters. Finally, the adjusted horizontal and vertical deviation compensation amounts were vector-synthesized to obtain a comprehensive compensation vector for buoy deviation, which was applied to correct the buoy position information. The corrected buoy position information was transmitted in real time to the monitoring center via a BeiDou communication module. The monitoring center then performed refined tracking and management of the buoy based on its precise position. If the buoy position deviated from the preset route by more than 50 meters, an alarm mechanism was activated, notifying staff to inspect and maintain the buoy.

[0083] By employing the sliding window technique, buoy deviation data within a certain time range is analyzed to extract the sliding characteristics of the deviation. By establishing a deviation sliding model, the dynamic cumulative effect of buoy deviation can be characterized and compensated, thereby improving the timeliness and continuity of deviation compensation.

[0084] Furthermore, buoy deviation data within a preset time range is obtained, and the preset time range is determined using sliding window technology;

[0085] Feature extraction algorithms are used to extract features from buoy deviation data to obtain feature vectors that reflect the buoy deviation sliding characteristics;

[0086] The feature vector is input into the deviation modeling module, which is based on machine learning algorithm and uses the feature vector as training sample to obtain the buoy deviation sliding model through training. The buoy deviation sliding model is used to characterize the dynamic cumulative change of buoy deviation over time.

[0087] When new buoy deviation data is obtained, the new buoy deviation data is input into the buoy deviation sliding model, and the dynamic cumulative trend of buoy deviation over a future period of time is predicted by the buoy deviation sliding model.

[0088] Determine the compensation value for buoy deviation based on the future trend of buoy deviation;

[0089] The compensation value is then fused with the current buoy deviation data to obtain the compensated buoy deviation data.

[0090] The system continuously tracks changes in buoy deviation data. When the distribution characteristics of the buoy deviation data change significantly, it triggers the retraining of the buoy deviation sliding model and updates the parameters of the buoy deviation sliding model using the latest buoy deviation data.

[0091] Furthermore, a sliding window technique is used to determine the buoy deviation data for the past 30 days within a preset time range. Wavelet transform is employed to extract features from the buoy deviation data, yielding a 10-dimensional feature vector reflecting the buoy deviation sliding characteristics, including statistical features such as mean, variance, skewness, and kurtosis. The extracted feature vector is input into a deviation modeling module based on a Long Short-Term Memory (LSTM) neural network. Using the feature vectors from the past 30 days as training samples, the buoy deviation sliding model is trained through 100 iterations. This model is then used to predict the buoy deviation trend for the next 7 days, with a predicted mean absolute percentage error (MAPE) of 8%. Based on the predicted future trend, exponential smoothing is used to determine the buoy deviation compensation value, i.e., the second deviation compensation. The compensation value is calculated as: Compensation value = Current deviation value × (1 + Predicted rate of change). The calculated compensation value is then weighted and fused with the current buoy deviation data at a weight ratio of 6:4 to obtain the compensated buoy deviation data. We continuously tracked the changes in buoy deviation data and performed a Kolmogorov-Smirnov test on the data distribution every 7 days. When the p-value of the test result was less than 0.5, we triggered the retraining of the buoy deviation sliding model and updated the model parameters using the buoy deviation data of the most recent 30 days. The updated model reduced the MAPE to 6% on the test set.

[0092] The deviation compensation result is obtained based on the deviation compensation model and the deviation sliding model. The deviation compensation result includes first deviation compensation and second deviation compensation. The process includes the following steps:

[0093] In the deviation modeling process, an adaptive weight adjustment strategy is introduced to dynamically adjust the weights of each sea state factor in the deviation model based on the significant differences in deviation characteristics under different sea state conditions, thereby improving the adaptability and robustness of the deviation model.

[0094] Furthermore, the process of inputting the feature vector into the deviation modeling module to obtain the deviation sliding model also includes model optimization. The model optimization process includes: acquiring historical deviation data under different sea state conditions, analyzing the historical deviation data to obtain the correlation strength between each sea state factor and the deviation characteristics, and obtaining typical sea states based on the correlation strength; constructing a deviation prediction model based on typical sea states; calculating the comprehensive deviation prediction result under typical sea states based on the deviation prediction model; and introducing the comprehensive deviation prediction result into the control system to optimize the second deviation compensation to obtain the optimized deviation sliding model.

[0095] Furthermore, historical deviation data under different sea state conditions is acquired and analyzed to determine the correlation strength between various sea state factors and deviation characteristics. For each sea state condition, a corresponding deviation prediction model is constructed based on the correlation strength, and the initial weights of each sea state factor are determined. Current sea state data is acquired to determine which typical sea state type it belongs to. Based on the typical sea state type, the corresponding deviation prediction model is invoked, and the weights of each sea state factor in the deviation prediction model are dynamically adjusted according to the current sea state data. Specifically, the weights of sea state factors with significant deviation characteristics are increased, while the weights of sea state factors with insignificant deviation characteristics are decreased. By weighted combination of the deviation prediction values ​​of each sea state factor, a comprehensive deviation prediction result is obtained. The comprehensive deviation prediction result is introduced into the control system to adjust and compensate for the navigation attitude in real time. Deviation feedback data is continuously collected, and the deviation prediction model and weight values ​​are periodically verified and updated.

[0096] Furthermore, buoy deviation data under different sea states were obtained from historical databases, including sea state factors such as wind speed, wind direction, ocean current speed, ocean current direction, and wave height. The Pearson correlation coefficient method was used to analyze the correlation strength between each sea state factor and the deviation characteristics. The results showed that the correlation coefficient between wind speed and deviation amplitude was 8, the correlation coefficient between ocean current speed and deviation direction was 7, and the correlation coefficient between wave height and deviation frequency was 6. For three typical sea states—strong winds, strong ocean currents, and large waves—deviation prediction models based on Support Vector Machines (SVM) were constructed, and the initial parameters of the models were determined through cross-validation. The current sea state data was clustered and identified as a strong ocean current state. The SVM deviation prediction model under strong ocean current conditions was called, and the weight values ​​of the two features—ocean current speed and direction—were dynamically adjusted using the sigmoid function based on real-time ocean current speed and direction data, with a weight range of 6–0. The predicted deviation values ​​of the three features—wind speed, ocean current speed, and wave height—were weighted and combined with a weight ratio of 3:5:2 to obtain the comprehensive deviation prediction result. The prediction result is fed into the autopilot control system, where a PID control algorithm is used to adjust the heading and speed in real time to compensate for navigation deviations. After each voyage, GNSS positioning data and inertial navigation data are collected, and the root mean square error (RMSE) between the actual deviation value and the predicted deviation value is calculated. When the RMSE exceeds 2 nautical miles for three consecutive voyages, the deviation prediction model is retrained. The latest 100 historical deviation data samples are used to optimize the parameters and update the weights of the SVM model, continuously improving the accuracy and adaptability of deviation prediction.

[0097] The wave parameter estimation results are obtained by fusing the deviation compensation results with GNSS positioning data. This includes the following steps:

[0098] A wave in-situ calibration model based on deviation compensation is constructed, and the deviation compensation results are fused with GNSS positioning data to generate high-precision wave parameter estimation results, thereby improving the accuracy of wave in-situ calibration.

[0099] Furthermore, in-situ wave observation data and synchronized GNSS positioning data were acquired; both the in-situ wave observation data and the GNSS positioning data were time series data.

[0100] For in-situ wave observation data, a wave in-situ calibration model based on machine learning is constructed. The machine learning model selects at least one algorithm from support vector machine, neural network and random forest.

[0101] The wave in-situ calibration model was applied to the wave in-situ observation data to obtain the deviation compensation results of the wave in-situ observation data.

[0102] Kalman filtering or particle filtering algorithms are used to fuse the deviation compensation results with GNSS positioning data to obtain fused wave observation data;

[0103] Based on the fused wave observation data, the wave parameters are estimated using the least squares method or Bayesian estimation method. The wave parameters include at least one of wave height, wave period and wave direction, and the wave parameter estimation results are obtained.

[0104] The wave parameter estimation results are compared with the in-situ wave observation data to calculate the improvement in the accuracy of the in-situ wave calibration.

[0105] If the improvement in wave in-situ calibration accuracy reaches a preset threshold, the wave parameter estimation result will be output as the final wave in-situ calibration result.

[0106] If the improvement in wave in-situ calibration accuracy does not reach the preset threshold, return to the data acquisition step, reacquire wave in-situ observation data and GNSS positioning data, and optimize the wave in-situ calibration model until the accuracy requirements are met.

[0107] Furthermore, one month of in-situ wave observation data and synchronous GNSS positioning data were obtained from wave buoys, with a data sampling frequency of 1 Hz. For the in-situ wave observation data, a support vector regression (SVR) algorithm was used to construct an in-situ wave calibration model. The input features included six dimensions such as acceleration and angular velocity, and the output was the wave height deviation compensation value. The kernel function type and penalty coefficient of the SVR model were optimized using a grid search method, and the optimal model was determined through 5-fold cross-validation. The trained SVR model was applied to the in-situ wave observation data to obtain the wave height deviation compensation result. Using a Kalman filter algorithm, with GNSS positioning data as the observation and the deviation compensation result as the control variable, a state-space model was established. The optimal wave height value was recursively estimated through two steps: prediction and update, achieving the fusion of the in-situ wave observation data and the GNSS positioning data. Based on the fused wave observation data, the least squares method was used to estimate wave parameters, including meaningful wave height, mean wave period, and mean wave direction. The estimated wave parameters were compared with in-situ observations to calculate the improvement in wave height calibration accuracy. Results showed a 3m reduction in wave height error after calibration, meeting the preset threshold of 2m. The estimated wave parameters were then used as the final in-situ wave calibration result. To further improve calibration accuracy, new observation data was continuously acquired, and the SVR model was dynamically updated using online learning. This continuous optimization of in-situ wave calibration performance, after three months of iterative optimization, resulted in wave height calibration accuracy stabilizing within 1m, meeting engineering application requirements.

[0108] By employing an ensemble learning approach, the prediction results of multiple machine learning algorithms, such as support vector machines and random forests, are combined to form a more robust bias compensation model, thereby improving the reliability and accuracy of bias compensation.

[0109] Furthermore, historical datasets are acquired and preprocessed to obtain training and testing datasets. For the training dataset, at least two machine learning algorithms are used to train at least two prediction models. The prediction results of the at least two prediction models are weighted and averaged to obtain the prediction result of the ensemble learning model. Based on the prediction result of the ensemble learning model, a bias compensation model is constructed. This bias compensation model is used to analyze the prediction error distribution characteristics of the ensemble learning model and correct its prediction results. An adaptive weight adjustment strategy is adopted to dynamically adjust the weights of the at least two prediction models in the ensemble learning model based on their prediction performance. New data is acquired, and an online learning mechanism is used to incrementally update the ensemble learning model based on the new data. Finally, the prediction results of the ensemble learning model and the output of the bias compensation model are combined to obtain the final prediction result.

[0110] Furthermore, data from the past year, totaling 10,000 records, was retrieved from a historical database. 80% was randomly selected as the training dataset, and the remaining 20% ​​as the test dataset. Data preprocessing included missing value imputation, outlier handling, and data standardization. The KNN algorithm was used to impute missing values, and outliers exceeding three standard deviations were replaced with the mean. The Z-score method was used for data standardization. For the training dataset, Support Vector Machine (SVM) and Random Forest algorithms were used for training. The hyperparameters of the models were optimized using a grid search method. The SVM model used the RBF kernel function with a penalty coefficient C = 10, and the Random Forest model had 100 decision trees with a maximum tree depth of 5. The predictions from the two models were weighted and averaged in a 2:1 ratio to obtain the prediction results of the ensemble learning model. To address the prediction results of the ensemble learning model, a bias compensation model is constructed using the Locally Weighted Linear Regression (LWLR) algorithm. Based on the distribution characteristics of the prediction error, a compensation function is fitted within a local region to correct the prediction results of the ensemble learning model. The compensation coefficient is adaptively adjusted according to the magnitude of the prediction error, with a maximum compensation margin not exceeding 20%. Based on the prediction performance of the SVM and Random Forest models on the test dataset, an adaptive weight adjustment strategy is adopted: when the prediction error of the SVM model is less than that of the Random Forest model, the weight of the SVM model is increased; otherwise, the weight of the Random Forest model is increased, with a weight adjustment margin of 5%. Whenever 100 new data points are acquired, an online learning mechanism is used to add the new data to the training dataset, while simultaneously incrementally updating the SVM and Random Forest models. The updated models continue to be used for ensemble learning. The prediction results of the ensemble learning model and the output of the bias compensation model are combined to obtain the final prediction result. Evaluation shows that the prediction accuracy reaches over 95%, meeting business requirements.

[0111] Establish a deviation compensation effect evaluation system. By introducing external high-precision wave observation data, evaluate the wave in-situ calibration accuracy before and after deviation compensation. Based on this, iteratively optimize the deviation compensation model and strategy to achieve continuous improvement in wave in-situ calibration accuracy.

[0112] Furthermore, high-precision external wave observation data is acquired as a reference standard for evaluating the deviation compensation effect. Based on the high-precision external wave observation data, the in-situ calibration accuracy of the current deviation compensation model is evaluated. If the in-situ calibration accuracy of the current deviation compensation model does not reach a preset threshold, machine learning algorithms are used to optimize the parameters of the deviation compensation model based on the evaluation results. Based on the optimized deviation compensation model, the wave observation data is re-processed for deviation compensation to obtain compensated wave observation data. The compensated wave observation data is input into the in-situ calibration model to obtain new calibration results. The new calibration results are compared with the high-precision external wave observation data to evaluate the improvement effect of the optimized deviation compensation model on the in-situ calibration accuracy. If the improvement effect reaches a preset threshold, the optimized deviation compensation model is applied to actual operations. Otherwise, the deviation compensation model is iteratively optimized until the calibration accuracy meets the preset requirements.

[0113] Furthermore, 10,000 high-precision wave observation records from the past year were obtained from the National Marine Environmental Monitoring Center and used as a reference standard for evaluating the bias compensation effect. The root mean square error (RMSE) was used to assess the in-situ calibration accuracy of the current bias compensation model. If the RMSE was greater than 2m, the calibration accuracy was considered to have failed to meet the preset threshold. A genetic algorithm was used to optimize the parameters of the bias compensation model, with a population size of 50, a crossover probability of 8, a mutation probability of 1, 100 iterations, and the objective function being to minimize the RMSE. The optimized bias compensation model was applied to the original wave observation data to perform bias compensation processing, resulting in compensated wave observation data. The compensated data was then input into an in-situ calibration model based on support vector regression (SVR), using a radial basis function kernel, a penalty coefficient C = 10, 8000 training samples, and 2000 test samples to obtain new calibration results. The new calibration results were compared with the high-precision wave observation data, and the RMSE was calculated to evaluate the improvement effect of the optimized bias compensation model on the in-situ calibration accuracy. If the RMSE decreases by more than 30%, the improvement is considered to have reached the preset threshold, and the optimized deviation compensation model is applied to actual operations. Otherwise, the deviation compensation model is iteratively optimized, and the parameters of the genetic algorithm are adjusted, such as increasing the population size and improving the crossover probability, until the calibration accuracy meets the requirement of RMSE less than 1m. Through the optimization of the deviation compensation model and the retraining of the in-situ calibration model, high-precision calibration of wave observation data is finally achieved. The consistency between the calibrated wave parameters and the high-precision observation data reaches more than 95%, which can meet the needs of various marine engineering and scientific research applications.

[0114] Example 2

[0115] This embodiment also provides a computer terminal device, including:

[0116] One or more processors;

[0117] A memory, coupled to the processor, for storing one or more programs;

[0118] When the one or more programs are executed by the one or more processors, the one or more processors implement a wave in-situ calibration method based on GNSS buoys.

[0119] Example 3

[0120] This embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a wave in-situ calibration method based on a GNSS buoy.

[0121] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A wave in-situ calibration method based on GNSS buoys, characterized in that, Includes the following steps: Buoy deviation data were obtained based on the position and attitude data of GNSS buoys under different sea states; Based on the buoy deviation data, a deviation compensation model and a deviation sliding model are constructed. The process of constructing the deviation sliding model includes: acquiring buoy deviation data within a preset time range based on a sliding window; analyzing the buoy deviation data using a feature extraction algorithm to obtain a feature vector of the buoy deviation sliding characteristics; inputting the feature vector into the deviation modeling module for sample training to construct the deviation sliding model; predicting the current buoy deviation data based on the deviation sliding model to obtain a deviation prediction value; calculating the buoy deviation compensation value based on the deviation prediction value; and weightedly fusing the compensation value and the current buoy deviation data to obtain a second deviation compensation. The deviation compensation result is obtained based on the deviation compensation model and the deviation sliding model, wherein the deviation compensation result includes a first deviation compensation and a second deviation compensation; the first deviation compensation is obtained by constructing a deviation compensation model based on the buoy deviation data; The process of fusing the deviation compensation result with GNSS positioning data to obtain wave parameter estimation results includes: acquiring in-situ wave observation data and synchronous GNSS positioning data; constructing an in-situ wave calibration model based on the in-situ wave observation data; inputting the deviation compensation result into the in-situ wave calibration model to obtain the deviation compensation result of the in-situ wave observation data; fusing the deviation compensation result of the in-situ wave observation data and the synchronous GNSS positioning data using a particle filter algorithm to obtain fused wave observation data; and using the least squares method to estimate the parameters of the fused wave observation data to obtain wave parameter estimation results.

2. The wave in-situ calibration method based on GNSS buoys according to claim 1, characterized in that, The process of obtaining the first deviation compensation includes: Acquire real-time location information and historical trajectory data of GNSS buoys; Based on the real-time location information and historical trajectory data, the horizontal and vertical displacements of the GNSS buoys are analyzed using a Kalman filter algorithm. Calculate the horizontal and vertical components of the buoy deviation based on the horizontal and vertical displacements of the GNSS buoy; Acquire multi-source marine environmental information, and extract features from the multi-source marine environmental information to obtain the key environmental factors affecting buoy deviation; A horizontal deviation compensation model for the horizontal component is constructed based on the ocean current factor among the key environmental factors mentioned above. A vertical deviation compensation model for the vertical component is constructed based on the wave factor among the key environmental factors mentioned above. The first deviation compensation is obtained based on the horizontal deviation compensation model and the vertical deviation compensation model.

3. The wave in-situ calibration method based on GNSS buoys according to claim 1, characterized in that, The process of obtaining the deviation sliding model by inputting the feature vector into the deviation modeling module also includes model optimization, which includes: Historical deviation data under different sea state conditions are obtained, the historical deviation data are analyzed to obtain the correlation strength between each sea state factor and the deviation characteristics, and typical sea states are obtained based on the correlation strength. A deviation prediction model is constructed based on the aforementioned typical sea conditions; The comprehensive deviation prediction results under typical sea conditions are calculated based on the deviation prediction model. The comprehensive deviation prediction result is introduced into the control system to optimize the second deviation compensation, resulting in an optimized deviation sliding model.

4. The wave in-situ calibration method based on GNSS buoys according to claim 1, characterized in that, After parameter estimation of the fused wave observation data, the method also includes establishing a deviation compensation effect evaluation system by introducing external high-precision wave observation data. Specifically, the wave parameter estimation results are compared with the in-situ wave observation data to calculate the improvement in wave in-situ calibration accuracy. A preset threshold is set based on the external high-precision wave observation data; The wave in-situ calibration accuracy is determined by comparing the improvement in the wave in-situ calibration accuracy with the preset threshold.

5. The wave in-situ calibration method based on GNSS buoys according to claim 4, characterized in that, When the improvement in the in-situ wave calibration accuracy reaches a preset threshold, the wave parameter estimation result is output as the final in-situ wave calibration result. If the improvement in the accuracy of the in-situ wave calibration does not reach the preset threshold, then the in-situ wave observation data and GNSS positioning data are reacquired, and the in-situ wave calibration model is optimized until the preset threshold is met.

6. A computer terminal device, characterized in that, include: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the wave in-situ calibration method based on GNSS buoys as described in any one of claims 1-5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the wave in-situ calibration method based on GNSS buoys as described in any one of claims 1-5.

Images