An environmental perception ocean buoy GNSS measurement dynamic modeling method and system

Abstract

This invention relates to the field of marine observation and satellite navigation positioning technology, and includes the following steps: S1, multi-source data acquisition and time synchronization; S2, buoy boom effect compensation and GNSS preprocessing; S3, marine environmental feature extraction and encoding; S4, deep learning-based covariance process noise prediction; S5, Lyapunov covariance dynamics propagation and spectral constraint smoothing; S6, adaptive extended Kalman filter positioning; S7, tidal and wave parameter extraction and early warning output. This invention utilizes a diagonalized transfer matrix structure to transform recursion into a parallel convolutional form, supporting fast Fourier transform to accelerate computation and meeting the real-time inference requirements of the buoy-embedded processor. It constructs an integrated processing link from raw multi-source data acquisition to high-precision positioning, marine parameter extraction, and disaster early warning, providing reliable data support for marine environmental monitoring and emergency response.

Description

Technical Field

[0001] This invention relates to the field of marine observation and satellite navigation positioning technology, and in particular to a dynamic modeling method and system for environmentally perceptive marine buoy GNSS measurements. Background Technology

[0002] GNSS positioning technology is the core technology for ocean buoy observation. By installing a GNSS receiver on the buoy, the three-dimensional position of the buoy can be obtained in real time, and then the tidal changes and wave parameters can be extracted from the vertical displacement sequence.

[0003] Compared with traditional pressure tide gauges and acoustic wave meters, GNSS buoys have the advantages of absolute positioning, all-weather operation and simultaneous acquisition of multiple parameters, and have been widely used in marine observation networks in many nearshore waters around the world.

[0004] In GNSS positioning, the accurate estimation of the measurement noise covariance matrix directly determines the performance of Kalman filter-type state estimators. Covariance estimation bias can lead to the filter over-relying on low-quality observations or responding slowly to effective observations, resulting in significant positioning errors.

[0005] GNSS measurement noise in the marine environment is affected by a variety of factors, including sea surface multipath reflection, ionospheric delay, tropospheric changes, satellite geometry, and the buoy's own motion, and these factors change dynamically over time.

[0006] Existing methods for estimating noise covariance in GNSS measurements mainly include fixed covariance methods, heuristic methods based on the DOP index, adaptive filtering methods based on innovation sequences, and direct prediction methods based on deep learning. Among these, the direct prediction method based on deep learning is the closest to this invention. This method can learn the mapping relationship between environmental characteristics and covariance from sensor data, but it treats the covariance prediction at each moment as an independent regression problem, does not model the time evolution dynamics of covariance, and cannot capture the continuous impact of multi-scale periodic changes in the marine environment on measurement quality. Summary of the Invention

[0007] The purpose of this invention is to provide a dynamic modeling method and system for environmentally perceptive ocean buoy GNSS measurements to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A dynamic modeling method for environmentally perceptive ocean buoy GNSS measurements includes the following steps:

[0010] S1. Multi-source data acquisition and time synchronization: Deploy a multi-source sensor array on the ocean buoy to acquire raw data from GNSS receivers, inertial measurement units, and meteorological sensors. Using GNSS time as the reference time axis, perform time alignment on the inertial measurement unit data and meteorological data respectively using linear interpolation to generate a multi-source synchronized data stream with a unified timestamp.

[0011] S2. Buoy boom arm effect compensation and GNSS preprocessing: The real-time attitude of the buoy is calculated using inertial measurement unit data, the boom arm deviation of the antenna phase center relative to the buoy buoyancy center is calculated, the boom arm deviation is deducted from the GNSS position observation, and the GNSS quality index is normalized and preprocessed.

[0012] S3. Marine environmental feature extraction and coding: Estimate the intensity of multipath reflection on the sea surface, encode the tidal phase, and use a multi-scale time attention mechanism to encode features at the tidal, wave, and meteorological scales respectively, and then concatenate them to generate an environmental feature vector.

[0013] S4. Covariance process noise prediction based on deep learning: The environmental feature vector, normalized GNSS quality index and buoy motion state vector are concatenated and input into a multilayer perceptron to generate a symmetric positive definite process noise matrix.

[0014] S5. Lyapunov covariance dynamics propagation and spectral constraint smoothing: The evolution of GNSS measurement noise covariance is modeled as a Lyapunov differential equation. Spectral constraints are applied to the transition matrix, which is discretized with the GNSS sampling period, and the current GNSS measurement covariance matrix is ​​obtained through propagation.

[0015] S6. Adaptive Extended Kalman Filter Positioning: The dynamic covariance matrix is ​​used as the measurement noise covariance and fed into the extended Kalman filter. It integrates GNSS observations and inertial navigation predictions to output a high-precision estimation sequence of the buoy's three-dimensional position and velocity.

[0016] S7. Tidal and wave parameter extraction and early warning output: Perform tidal harmonic analysis and wave parameter extraction on the buoy position time series, monitor the rate of change of vertical displacement and residual water level in real time, and trigger early warning of abnormal events.

[0017] As a further improvement to this technical solution: In step S1, the data collected by the GNSS receiver includes pseudorange observations, carrier phase observations, the number of satellites, and geometric accuracy dilution factors, horizontal accuracy dilution factors, and vertical accuracy dilution factors; the inertial measurement unit collects triaxial acceleration and triaxial angular velocity; the meteorological sensor collects wind speed, wind direction, air pressure, and air temperature; when the data of a certain sensor is missing for more than a preset time threshold, the time period is marked as a data missing state, and the covariance estimate of the previous valid time remains unchanged.

[0018] As a further improvement to this technical solution: In step S2, the pitch angle, roll angle and heading angle of the buoy are calculated using complementary filtering or extended Kalman filtering, the boom deviation is calculated by the rotation matrix and deducted from the GNSS position observation; the DOP value is logarithmically transformed, the number of satellites is normalized according to the maximum value of the sequence, and when the number of satellites is lower than the preset threshold, the moment is marked as a GNSS degraded state.

[0019] As a further improvement to this technical solution: In step S3, the root mean square wave height of the sea surface is estimated by wind speed, and the multipath power attenuation factor of the specular reflection of the sea surface is calculated by combining the satellite elevation angle. The signal-to-noise ratio weighted average of the multipath attenuation factors of all visible satellites is used to obtain the comprehensive multipath index. The tidal phases of the M2 and S2 tides are sine and cosine encoded. The multi-scale time attention mechanism generates attention outputs at three scales: tidal, wave, and meteorological, which are then concatenated with the multipath index and tidal phase encoding to obtain the environmental feature vector.

[0020] As a further improvement to this technical solution: In step S4, the buoy motion state vector includes the buoy pitch angle, roll angle, and heave speed; the process noise matrix is ​​generated by LDL decomposition parameterization, wherein the elements of the unit lower triangular matrix are generated by linear projection, and the elements of the diagonal matrix are guaranteed to be strictly positive by the softplus function.

[0021] As a further improvement to this technical solution: in step S5, the Lyapunov-type differential equation is:

[0022] ;

[0023] in Let be the transition matrix. The process noise matrix is ​​defined; spectral constraints are applied to the transition matrix so that the real parts of all its eigenvalues ​​are negative and not less than a preset minimum value, ensuring the exponential stability of the covariance evolution; the discretized recursion is expanded into a parallel convolution form using the diagonalized structure of the transition matrix, and the calculation is accelerated by fast Fourier transform.

[0024] As a further improvement to this technical solution: In step S6, the state vector of the extended Kalman filter includes buoy position, velocity, attitude quaternion, accelerometer bias and gyroscope bias; before the update step, the standardized information is calculated, and when it exceeds the chi-square test threshold (α=0.01), it is judged as an abnormal observation and the update is skipped.

[0025] As a further improvement to this technical solution: In step S7, least squares harmonic fitting is used to extract the harmonic constants of the main tidal constituents M2, S2, K1, and O1; after removing the tidal component from the vertical displacement sequence, the wave energy spectral density is obtained through piecewise fast Fourier transform, and the effective wave height and average wave period are calculated; when the rate of change of vertical displacement or the residual water level exceeds a preset threshold, an early warning is triggered; when the wave period exceeds 300 seconds and the water level deviation continues to exceed the threshold, a tsunami warning signal is issued.

[0026] A dynamic modeling system for environmentally aware ocean buoy GNSS measurements, used to perform the above method, includes:

[0027] The multi-source data acquisition module is used to acquire raw data from GNSS receivers, inertial measurement units, and meteorological sensors and to perform time synchronization.

[0028] The attitude compensation and preprocessing module is used to calculate the real-time attitude of the buoy, compensate for the arm deviation of the antenna phase center, and perform normalization preprocessing of GNSS quality indicators.

[0029] The marine environment feature encoding module is used to estimate the intensity of multipath reflection from the sea surface, encode the tidal phase, and generate environmental feature vectors through a multi-scale time attention mechanism.

[0030] The covariance dynamics learning module includes a multilayer perceptron inference unit and a Lyapunov propagation unit. The multilayer perceptron inference unit is used to predict the process noise matrix, and the Lyapunov propagation unit is used to propagate the dynamic GNSS measurement noise covariance matrix.

[0031] An adaptive Kalman filter positioning module is used to fuse GNSS observations and inertial navigation predictions to output a high-precision estimation sequence of the buoy's three-dimensional position and velocity;

[0032] The ocean parameter extraction and early warning module is used to perform tidal harmonic analysis, wave spectrum analysis, and abnormal event detection.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] 1. This invention combines Lyapunov dynamic system modeling with deep learning-driven covariance prediction, reducing the negative log-likelihood of covariance estimation by 55.0% compared to the fixed covariance method and by 28.3% compared to the MLP direct prediction method; the root mean square error of vertical positioning under severe sea conditions is reduced to 5.3 cm, a reduction of 73.2% compared to the fixed covariance method; the M2 tidal amplitude extraction error is only 0.8 cm, and the relative error of wave effective wave height is controlled within 6.3%, meeting the centimeter-level accuracy requirements for marine observation;

[0035] 2. This invention applies spectral constraints to the transition matrix, fundamentally eliminating the problem of covariance jumps in independent predictions at each time step; the covariance dynamic system has incremental exponential stability, and it converges quickly to a unique steady-state trajectory starting from any positive definite initial covariance, eliminating the sensitivity to initial parameters, greatly reducing the difficulty of on-site deployment and debugging, and the eigenvalues ​​can be manually adjusted to balance sensitivity and smoothness.

[0036] 3. This invention achieves mast antenna arm effect compensation through IMU real-time attitude calculation, eliminating displacement amplification distortion caused by buoy tilt; based on sea surface roughness and satellite elevation angle, parameterized multipath modeling improves the uncertainty estimation quality under low elevation angle signal degradation conditions; multi-scale time attention mechanism captures the periodic effects of tides, waves, and weather respectively, enabling the model to have a forward-looking perception capability of environmental changes.

[0037] 4. This invention utilizes the diagonalization structure of the transfer matrix to transform the recursion into a parallel convolution form, supporting fast Fourier transform to accelerate calculation and meet the real-time inference requirements of the embedded processor at the buoy end; it constructs an integrated processing link from raw multi-source data acquisition to high-precision positioning, marine parameter extraction and disaster early warning, providing reliable data support for marine environmental monitoring and emergency response.

[0038] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0039] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0040] Figure 1 This is a flowchart illustrating the technical route of the method of the present invention;

[0041] Figure 2 A time series comparison of the covariance matrix traces of the various methods of this invention (top) and sea state changes (bottom);

[0042] Figure 3 A comparison of vertical positioning errors of various methods in this invention;

[0043] Figure 4 To verify the exponential convergence characteristics of the method of the present invention under different initial covariances R(0);

[0044] Figure 5The image shows a comparison of tidal extraction (top), effective wave height estimation (bottom left), and M2 amplitude error (bottom right) in this invention. Detailed Implementation

[0045] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically in the following paragraphs by way of example with reference to the accompanying drawings. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0046] The model training method of this invention:

[0047] Define GNSS measurement residuals as ,in For the true location, the training loss function is the sum of the negative log-likelihood and the smoothing penalty term:

[0048]

[0049] in, To smooth out the penalty weights, the negative log-likelihood term is:

[0050]

[0051] The smoothing penalty term is the squared hinge loss:

[0052]

[0053] Within the complete Lyapunov dynamics framework, the spectral constraints are designed to ensure smooth conditions, and the penalty term is zero. The training data comes from GNSS observations during buoy field deployment, with post-processing results of differential GNSS or precise point positioning used as reference positions.

[0054] Experimental data: To verify the effectiveness of the method of the present invention, GNSS buoys were deployed in the nearshore waters of Beibu Gulf in Guangxi for experimental verification. The experiment lasted for 72 hours, covering multiple complete tidal cycles. During the experiment, the sea state level changed from level 2 to level 5. The buoys were equipped with a dual-frequency GNSS receiver (sampling rate 10Hz), MEMS-IMU (sampling rate 100Hz), and meteorological sensors. The post-hoc precise single-point positioning results of the shore-based differential GNSS reference station were used as the reference true value.

[0055] The method of this invention is compared with the following baseline methods: fixed covariance method, DOP-based heuristic method, innovation sequence-based adaptive method, and MLP-based direct prediction method. Specific Implementation

[0056] Please see Figures 1 to 5In the embodiments of the present invention:

[0057] Step S1: Multi-source data acquisition and time synchronization:

[0058] A multi-source sensor array is deployed on an ocean buoy to collect the following data: pseudorange observations from the GNSS receiver, carrier phase observations, and the number of satellites. And various DOP metrics, including the geometric precision dilution factor. Horizontal precision dilution factor and vertical precision dilution factor GNSS sampling rate The value range is 1. 20Hz; triaxial acceleration and triaxial angular velocity output by the Inertial Measurement Unit (IMU), with a sampling rate of 20Hz. The value range is 100. 200Hz; wind speed output by the weather sensor ,wind direction air pressure and temperature .

[0059] Using GNSS time as the reference time axis, linear interpolation is applied to IMU data and meteorological data for time alignment, generating a multi-source synchronized data stream with a unified timestamp. Specifically, for GNSS time... Find two adjacent sampling times in the IMU data sequence and ,satisfy The interpolation result of the IMU data is:

[0060]

[0061] in, Indicates that IMU is in The six-axis output vector at any given time is interpolated using the same method for meteorological data. When data from a particular sensor is missing for more than a preset time threshold... When a data gap occurs, the time period is marked as missing. The covariance learning module maintains the covariance estimate from the previous valid time point for that time period.

[0062] Input: Raw data streams from each sensor; Output: Time-synchronized multi-source observation data. ,in Represents a vector of meteorological parameters. This represents the total number of sampling points.

[0063] Step S2: Buoy boom arm effect compensation and GNSS preprocessing:

[0064] The GNSS antenna of the buoy is installed at the top of the mast. There is a fixed arm vector between the phase center of the antenna and the center of the buoy. When the buoy pitches and rolls with the waves, the arc drawn by the antenna at the top of the mast is greater than the actual water surface movement at the center of the buoy. That is, the arm effect amplifies and distorts the displacement measured by the antenna relative to the real water surface displacement. The purpose of this step is to eliminate the measurement deviation introduced by the arm effect and restore the real water surface displacement at the center of the buoy.

[0065] The pitch angle is obtained by calculating the buoy's real-time attitude using IMU data through complementary filtering or extended Kalman filtering. yaw angle and heading angle Buoy attitude is achieved using a rotation matrix. This represents the transformation from the buoy volume coordinate system to the navigation coordinate system:

[0066]

[0067] in, , , Respectively, the coordinate systems around the body , , The basic rotation matrix of the axis.

[0068] The arm vector of the antenna phase center relative to the buoy's buoyancy center is expressed in volume coordinates as follows: When the buoy changes attitude, the lever effect causes the antenna phase center to deviate from its static position directly above the buoyancy center. The deviation is:

[0069]

[0070] in, To reconstruct the water surface displacement at the center of buoyancy by subtracting the arm deviation from the GNSS position observations when the buoy is in still water and projecting it onto the navigation coordinate system:

[0071]

[0072] Simultaneously, the GNSS quality indicators are preprocessed using normalization: the DOP value undergoes a logarithmic transformation.

[0073]

[0074] Eliminate magnitude differences; normalize the number of satellites to the maximum value in the sequence:

[0075]

[0076] When the number of satellites is below a preset threshold When this time is marked as a GNSS degradation state, the corresponding covariance will be driven by a larger process noise.

[0077] Input: IMU data and GNSS observations from the synchronization data stream; Output: Attitude-compensated GNSS observations. And normalized quality index vector:

[0078] .

[0079] Step S3: Marine environmental feature extraction and coding:

[0080] This step extracts marine environmental features from meteorological parameters, sea state observations, and time information, and generates environmental feature vectors through a multi-scale time attention mechanism.

[0081] First, the multipath reflection intensity of the sea surface is estimated. The multipath reflection intensity of the sea surface on GNSS signals is closely related to the sea surface roughness and the satellite elevation angle. The root mean square wave height of the sea surface is also considered. By wind speed Estimated using empirical relationships:

[0082]

[0083] For the elevation angle is For satellites, the multipath power attenuation factor for specular reflection from the sea surface is:

[0084]

[0085] in, For the GNSS signal carrier wavelength, a weighted average of the multipath attenuation factors of all visible satellites is taken, with the weights being the normalized values ​​of the signal-to-noise ratio of each satellite, to obtain the comprehensive multipath index. .

[0086] Then, tidal phase encoding is performed. Ocean tides are dominated by the semi-diurnal equinox M2, with a period of [missing information]. Hours, the current time The tidal phase is encoded as follows:

[0087]

[0088] in, The hour corresponds to the period of the solar semi-diurnal tide S2. This encoding enables the network to sense the current tidal phase, thereby predicting the impact of tidal-related water level changes on the quality of GNSS observations.

[0089] Finally, multi-scale temporal attention embedding is performed, and a multi-scale temporal attention mechanism is designed to encode features for tidal, wave, and meteorological scales respectively. For each time scale... Define query vector The key vector is the sinusoidal encoding of the current time at this scale. Sum value vector The attention output is calculated as follows, since the parameters are learnable:

[0090]

[0091] in, To output the projection matrix, For temperature parameters, The number of keys. For the dimension of the key, For the dimension of the value, As an output dimension, the temperature parameter controls the degree of concentration of attention weights: lower temperatures allow attention to be focused on the bonds most relevant to the current moment, which is suitable for capturing localized environmental features.

[0092] The attention outputs at the three scales are concatenated with multipath metrics and tidal phase encodings to form an environmental feature vector:

[0093]

[0094] Input: meteorological parameters, sea state observations, current time; Output: environmental feature vector ,

[0095] Step S4: Deep learning-based prediction of noise in the covariance process:

[0096] Environmental feature vectors Normalized GNSS quality indicators and buoy motion state vector After being stitched together, the data is input into a multi-layer perceptron (MLP), where... Representing the buoy's heave speed, the MLP maps the input to a high-dimensional embedding space:

[0097]

[0098] in, It is a high-dimensional embedding vector. For the embedding dimension, MLP consists of multiple fully connected layers and ReLU activation functions.

[0099] Using embedding vectors Generate a symmetric positive definite process noise matrix LDL decomposition parameterization is used to ensure positive definiteness:

[0100]

[0101] in, It is a unit lower triangular matrix. It is a diagonal matrix, and the diagonal elements are guaranteed to be strictly positive using the softplus function:

[0102]

[0103] in, For learnable parameter vectors, The lower triangular elements are generated by linear projection:

[0104] ( )

[0105] in A learnable parameter vector; diagonal elements Upper triangular element ( This parameterization method guarantees For any All are positive definite matrices.

[0106] Inputs: Environmental feature vector, GNSS quality index, buoy motion state; Output: Process noise matrix .

[0107] Step S5: Lyapunov covariance dynamics propagation and spectral constraint smoothing

[0108] GNSS measurement noise covariance The evolution model is a Lyapunov-type differential equation:

[0109]

[0110] in, The transition matrix controls the dissipation dynamics of the covariance. The process noise matrix output in step S4 is used as the driving term.

[0111] right Apply spectral constraints to ensure the exponential stability and temporal smoothness of covariance evolution. Parameterized by eigenvalue decomposition

[0112]

[0113] in It is a diagonal eigenvalue matrix. It is an invertible matrix, and its eigenvalues ​​satisfy the following constraints:

[0114]

[0115] in, , The maximum rate of contraction of the covariance ellipsoid volume. Given the dimension of the covariance matrix, this constraint guarantees that the time derivative of the logarithmic determinant of the covariance satisfies the smoothness condition:

[0116]

[0117] With GNSS sampling period Discretization yields the discrete-time Lyapunov recurrence relation:

[0118]

[0119] in, and For the discretized transition matrix and process noise matrix, using The diagonalized structure is unfolded recursively into a parallel convolutional form:

[0120]

[0121] in, , This convolutional structure eliminates the sequential dependency of recursion and can be used for parallel computation using the Fast Fourier Transform.

[0122] The construction of this dynamical system proves that it possesses incremental exponential stability: for any two sets of different initial covariances and The difference satisfies:

[0123]

[0124] in, Shrinkage rate This represents the Frobenius norm, a property that indicates the covariance evolution is insensitive to initial conditions and converges exponentially to a form starting from any positive definite initial value. The only steady-state trajectory driven by [the engine].

[0125] Eigenvalues During the training phase, the optimal value is learned through gradient descent. During the deployment phase, it can be manually adjusted to balance sensitivity and smoothness: reducing the absolute negative value of the feature makes the covariance more sensitive to process noise, while increasing it makes the evolution smoother.

[0126] Input: Process noise matrix Covariance at the previous time step Output: GNSS measurement covariance matrix at the current time. .

[0127] Step S6: Adaptive Extended Kalman Filter Localization:

[0128] The dynamic covariance matrix output in step S5 As a measurement noise covariance input to EKF, it enables GNSS / IMU tight combination positioning.

[0129] EKF state vector Including buoy position ,speed Posture Quaternions Accelerometer deviation and gyroscope deviation .

[0130] Prediction steps: State propagation using IMU data:

[0131]

[0132] in, For the IMU kinematic equations, Let Jacobian be the state transition matrix. The noise mapping matrix, For IMU process noise covariance, Let be the state estimation error covariance.

[0133] Update steps: When GNSS observations arrive, calculate the innovation and Kalman gain:

[0134]

[0135] in, For the innovation vector, For the observation matrix, For the new information covariance, For Kalman gain, This is the dynamic covariance matrix output in step S5.

[0136] Anomaly removal before update: Calculate standardized innovation When this value exceeds the chi-square test threshold If the current GNSS observation is determined to be an outlier, the update is skipped. (Parameters) The significance level is set at 0.01.

[0137] Status Update:

[0138]

[0139] in, It is an identity matrix.

[0140] Input: Attitude-compensated GNSS observations, dynamic covariance matrix IMU data, output: buoy's three-dimensional position and speed High-precision estimated sequence.

[0141] Step S7: Extraction of tidal and wave parameters and output of early warning:

[0142] Ocean parameters are extracted from the buoy position time series output in step S6.

[0143] Tidal harmonic analysis: taking the vertical displacement sequence The harmonic constants of the main tidal constituents are extracted using least squares harmonic fitting, and the tidal level model is expressed as:

[0144]

[0145] in, Mean sea level To fit the number of tidal constituents, and The first The amplitude and lag angle of each tidal component. For the first The angular frequencies of each tidal constituent, the main tidal constituents include: the angular frequency of the semidiurnal tidal constituent M2. angular frequency of S2 And the cycle of the diurnal tidal K1. Hourly and O1 cycles The amplitudes of each tidal constituent were obtained by solving the overdetermined equations using the least squares method. Hechijiao .

[0146] Wave parameter extraction: Wave displacement is obtained by removing the tidal component from the vertical displacement sequence. ,in The tidal level values ​​obtained by harmonic fitting are... Perform piecewise fast Fourier transform to obtain wave energy spectral density Then calculate the effective wave height. and mean wave period :

[0147]

[0148] in, for Spectral moments, and These are the lower and upper limits of the wave frequency band.

[0149] Anomaly detection and early warning: Real-time monitoring of the rate of change of vertical displacement and residual water level after filtering out tides ,when Exceeding the preset rate threshold or Exceeding the preset water level deviation threshold When an abnormal event warning is triggered, a long-term water level anomaly criterion is set for tsunami events: when the wave period exceeds 300 seconds and the water level deviation continues to exceed the threshold, a tsunami warning signal is issued.

[0150] Input: High-precision buoy position time series; Output: Tidal harmonic constant. Real-time wave parameters Abnormal warning signals.

[0151] Experiment 1: Comparison of Covariance Estimation Accuracy

[0152] The negative log-likelihood was used as the evaluation index, and the mean loss and standard deviation of each method were calculated on 72-hour data; the results are shown in Table 1.

[0153] Experiment 2: Comparison of Positioning Accuracy

[0154] The covariances output by each method were incorporated into the same EKF framework to evaluate the root mean square error of 3D positioning and the root mean square error of vertical positioning, and statistical analysis was performed under calm and severe sea conditions; the results are shown in Table 2.

[0155] Experiment 3: Accuracy of Tidal and Wave Parameter Extraction:

[0156] Tidal harmonic analysis and wave parameter extraction were performed using the positioning outputs of various methods. Based on tide gauge observation data and wave buoy measured data, the amplitude error and relative error of the M2 tidal constituent were evaluated. The results are shown in Table 3. Figure 5 .

[0157] Experiment 4: Covariance Smoothing and Initial Value Convergence:

[0158] The method of this invention was run from different initial covariances to verify the exponential convergence characteristics, and the smoothness of the covariance time series of each method was compared.

[0159] See results Figure 3 and Figure 4 .

[0160] Table 1: Comparison of covariance estimation accuracy of various methods (72-hour average)

[0161]

[0162] Table 2: Comparison of Root Mean Square Errors for Localization by Different Methods (Unit: cm)

[0163]

[0164] Table 3: Comparison of Tide and Wave Parameter Extraction Accuracy

[0165]

[0166] Experimental conclusion:

[0167] Under the learning eigenvalue setting, the method of this invention achieves a mean negative log-likelihood of 3.12 for covariance estimation, which is 55.0% lower than the fixed covariance method and 28.3% lower than the MLP direct prediction method. Under severe sea conditions, the root mean square error of vertical positioning using the method of this invention is 5.3 cm, which is 73.2% lower than the 19.8 cm of the fixed covariance method and 42.4% lower than the 9.2 cm of the MLP direct prediction method. The M2 tidal amplitude extraction error is 0.8 cm, and the relative error of significant wave height is 6.3%.

[0168] Adjust the eigenvalues ​​to Afterwards, the average loss rose to 4.68, but the standard deviation dropped to 0.72, indicating that the covariance estimation is more conservative and stable, and is suitable for application scenarios with higher requirements for smoothness.

[0169] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Those skilled in the art can readily implement the present invention based on the description and drawings above. However, any modifications, alterations, and variations made by those skilled in the art without departing from the scope of the present invention using the disclosed technical content are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, and variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A dynamic modeling method for environmentally perceptive ocean buoy GNSS measurements, characterized in that, Includes the following steps: S1. Multi-source data acquisition and time synchronization: Deploy a multi-source sensor array on the ocean buoy to acquire raw data from GNSS receivers, inertial measurement units, and meteorological sensors. Using GNSS time as the reference time axis, perform time alignment on the inertial measurement unit data and meteorological data respectively using linear interpolation to generate a multi-source synchronized data stream with a unified timestamp. S2. Buoy boom arm effect compensation and GNSS preprocessing: The real-time attitude of the buoy is calculated using inertial measurement unit data, the boom arm deviation of the antenna phase center relative to the buoy buoyancy center is calculated, the boom arm deviation is deducted from the GNSS position observation, and the GNSS quality index is normalized and preprocessed. S3. Marine environmental feature extraction and coding: Estimate the intensity of multipath reflection on the sea surface, encode the tidal phase, and use a multi-scale time attention mechanism to encode features at the tidal, wave, and meteorological scales respectively, and then concatenate them to generate an environmental feature vector. S4. Covariance process noise prediction based on deep learning: The environmental feature vector, normalized GNSS quality index and buoy motion state vector are concatenated and input into a multilayer perceptron to generate a symmetric positive definite process noise matrix. S5. Lyapunov covariance dynamics propagation and spectral constraint smoothing: The evolution of GNSS measurement noise covariance is modeled as a Lyapunov differential equation. Spectral constraints are applied to the transition matrix, which is discretized with the GNSS sampling period, and the current GNSS measurement covariance matrix is ​​obtained through propagation. S6. Adaptive Extended Kalman Filter Positioning: The dynamic covariance matrix is ​​used as the measurement noise covariance and fed into the extended Kalman filter. It integrates GNSS observations and inertial navigation predictions to output a high-precision estimation sequence of the buoy's three-dimensional position and velocity. S7. Tidal and wave parameter extraction and early warning output: Perform tidal harmonic analysis and wave parameter extraction on the buoy position time series, monitor the rate of change of vertical displacement and residual water level in real time, and trigger early warning of abnormal events.

2. The method for dynamic modeling of environmentally perceptive ocean buoy GNSS measurements according to claim 1, characterized in that, In step S1, the data collected by the GNSS receiver includes pseudorange observations, carrier phase observations, number of satellites, and geometric accuracy dilution factor, horizontal accuracy dilution factor, and vertical accuracy dilution factor; the inertial measurement unit collects triaxial acceleration and triaxial angular velocity; the meteorological sensor collects wind speed, wind direction, air pressure, and air temperature; when the data of a certain sensor is missing for more than a preset time threshold, the time period is marked as a data missing state, and the covariance estimate of the previous valid time remains unchanged.

3. The method for dynamic modeling of environmentally perceptive ocean buoy GNSS measurements according to claim 1, characterized in that, In step S2, the pitch, roll, and heading angles of the buoy are calculated using complementary filtering or extended Kalman filtering. The boom deviation is calculated using a rotation matrix and deducted from the GNSS position observations. The DOP value is logarithmically transformed, and the number of satellites is normalized to the maximum value in the sequence. When the number of satellites is lower than a preset threshold, the moment is marked as a GNSS degraded state.

4. The method for dynamic modeling of environmentally perceptive marine buoy GNSS measurements according to claim 1, characterized in that, In step S3, the root mean square wave height of the sea surface is estimated by wind speed, and the multipath power attenuation factor of the specular reflection of the sea surface is calculated by combining the satellite elevation angle. The signal-to-noise ratio weighted average of the multipath attenuation factors of all visible satellites is used to obtain the comprehensive multipath index. The tidal phases of the M2 and S2 tides are sine and cosine encoded. The multi-scale time attention mechanism generates attention outputs at three scales: tidal, wave, and meteorological, which are then concatenated with the multipath index and tidal phase encoding to obtain the environmental feature vector.

5. The method for dynamic modeling of environmentally perceptive marine buoy GNSS measurements according to claim 1, characterized in that, In step S4, the buoy motion state vector includes the buoy pitch angle, roll angle, and heave speed; the process noise matrix is ​​generated by LDL decomposition parameterization, wherein the elements of the unit lower triangular matrix are generated by linear projection, and the elements of the diagonal matrix are guaranteed to be strictly positive by the softplus function.

6. The method for dynamic modeling of environmentally perceptive ocean buoy GNSS measurements according to claim 1, characterized in that, In step S5, the Lyapunov-type differential equation is: ； in Let be the transition matrix. The process noise matrix is ​​defined; spectral constraints are applied to the transition matrix such that the real parts of all its eigenvalues ​​are negative and not less than a preset minimum value, thus ensuring the exponential stability of the covariance evolution. The discretized recursion is expanded into a parallel convolution form by utilizing the diagonalized structure of the transition matrix, and the computation is accelerated by the fast Fourier transform.

7. The method for dynamic modeling of environmentally perceptive ocean buoy GNSS measurements according to claim 1, characterized in that, In step S6, the state vector of the extended Kalman filter includes buoy position, velocity, attitude quaternion, accelerometer bias, and gyroscope bias. Before the update step, the standardized information is calculated. If it exceeds the chi-square test threshold (α=0.01), it is judged as an abnormal observation and the update is skipped.

8. The method for dynamic modeling of environmentally perceptive marine buoy GNSS measurements according to claim 1, characterized in that, In step S7, the harmonic constants of the main tidal constituents M2, S2, K1, and O1 are extracted using least squares harmonic fitting; after removing the tidal component from the vertical displacement sequence, the wave energy spectral density is obtained through piecewise fast Fourier transform, and the effective wave height and average wave period are calculated; when the rate of change of vertical displacement or the residual water level exceeds a preset threshold, an early warning is triggered; when the wave period exceeds 300 seconds and the water level deviation continues to exceed the threshold, a tsunami warning signal is issued.

9. A dynamic modeling system for environmentally perceptive ocean buoy GNSS measurement according to claim 1, characterized in that, For performing the method according to any one of claims 1-8, comprising: The multi-source data acquisition module is used to acquire raw data from GNSS receivers, inertial measurement units, and meteorological sensors and to perform time synchronization. The attitude compensation and preprocessing module is used to calculate the real-time attitude of the buoy, compensate for the arm deviation of the antenna phase center, and perform normalization preprocessing of GNSS quality indicators. The marine environment feature encoding module is used to estimate the intensity of multipath reflection from the sea surface, encode the tidal phase, and generate environmental feature vectors through a multi-scale time attention mechanism. The covariance dynamics learning module includes a multilayer perceptron inference unit and a Lyapunov propagation unit. The multilayer perceptron inference unit is used to predict the process noise matrix, and the Lyapunov propagation unit is used to propagate the dynamic GNSS measurement noise covariance matrix. An adaptive Kalman filter positioning module is used to fuse GNSS observations and inertial navigation predictions to output a high-precision estimation sequence of the buoy's three-dimensional position and velocity; The ocean parameter extraction and early warning module is used to perform tidal harmonic analysis, wave spectrum analysis, and abnormal event detection.

Images